diff options
Diffstat (limited to 'fs/ocfs2/dlm')
| -rw-r--r-- | fs/ocfs2/dlm/Makefile | 7 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/dlmapi.h | 10 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/dlmast.c | 164 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/dlmcommon.h | 436 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/dlmconvert.c | 84 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/dlmdebug.c | 917 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/dlmdebug.h | 57 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/dlmdomain.c | 1301 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/dlmfs.c | 640 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/dlmfsver.c | 42 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/dlmfsver.h | 31 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/dlmlock.c | 179 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/dlmmaster.c | 2149 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/dlmrecovery.c | 1426 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/dlmthread.c | 342 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/dlmunlock.c | 126 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/dlmver.c | 42 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/dlmver.h | 31 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/userdlm.c | 658 | ||||
| -rw-r--r-- | fs/ocfs2/dlm/userdlm.h | 111 |
20 files changed, 5506 insertions, 3247 deletions
diff --git a/fs/ocfs2/dlm/Makefile b/fs/ocfs2/dlm/Makefile index ce3f7c29d27..bd1aab1f49a 100644 --- a/fs/ocfs2/dlm/Makefile +++ b/fs/ocfs2/dlm/Makefile @@ -1,8 +1,7 @@ -EXTRA_CFLAGS += -Ifs/ocfs2 +ccflags-y := -Ifs/ocfs2 -obj-$(CONFIG_OCFS2_FS) += ocfs2_dlm.o ocfs2_dlmfs.o +obj-$(CONFIG_OCFS2_FS_O2CB) += ocfs2_dlm.o ocfs2_dlm-objs := dlmdomain.o dlmdebug.o dlmthread.o dlmrecovery.o \ - dlmmaster.o dlmast.o dlmconvert.o dlmlock.o dlmunlock.o dlmver.o + dlmmaster.o dlmast.o dlmconvert.o dlmlock.o dlmunlock.o -ocfs2_dlmfs-objs := userdlm.o dlmfs.o dlmfsver.o diff --git a/fs/ocfs2/dlm/dlmapi.h b/fs/ocfs2/dlm/dlmapi.h index 53652f51c0e..3cfa114aa39 100644 --- a/fs/ocfs2/dlm/dlmapi.h +++ b/fs/ocfs2/dlm/dlmapi.h @@ -95,7 +95,7 @@ const char *dlm_errname(enum dlm_status err); mlog(ML_ERROR, "dlm status = %s\n", dlm_errname((st))); \ } while (0) -#define DLM_LKSB_UNUSED1 0x01 +#define DLM_LKSB_UNUSED1 0x01 #define DLM_LKSB_PUT_LVB 0x02 #define DLM_LKSB_GET_LVB 0x04 #define DLM_LKSB_UNUSED2 0x08 @@ -182,6 +182,7 @@ enum dlm_status dlmlock(struct dlm_ctxt *dlm, struct dlm_lockstatus *lksb, int flags, const char *name, + int namelen, dlm_astlockfunc_t *ast, void *data, dlm_bastlockfunc_t *bast); @@ -192,7 +193,12 @@ enum dlm_status dlmunlock(struct dlm_ctxt *dlm, dlm_astunlockfunc_t *unlockast, void *data); -struct dlm_ctxt * dlm_register_domain(const char *domain, u32 key); +struct dlm_protocol_version { + u8 pv_major; + u8 pv_minor; +}; +struct dlm_ctxt * dlm_register_domain(const char *domain, u32 key, + struct dlm_protocol_version *fs_proto); void dlm_unregister_domain(struct dlm_ctxt *dlm); diff --git a/fs/ocfs2/dlm/dlmast.c b/fs/ocfs2/dlm/dlmast.c index 8d17d28ef91..b46278f9ae4 100644 --- a/fs/ocfs2/dlm/dlmast.c +++ b/fs/ocfs2/dlm/dlmast.c @@ -28,9 +28,7 @@ #include <linux/module.h> #include <linux/fs.h> #include <linux/types.h> -#include <linux/slab.h> #include <linux/highmem.h> -#include <linux/utsname.h> #include <linux/init.h> #include <linux/sysctl.h> #include <linux/random.h> @@ -43,7 +41,6 @@ #include "cluster/heartbeat.h" #include "cluster/nodemanager.h" #include "cluster/tcp.h" -#include "cluster/endian.h" #include "dlmapi.h" #include "dlmcommon.h" @@ -91,22 +88,31 @@ static int dlm_should_cancel_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock) return 0; } -static void __dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock) +void __dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock) { - mlog_entry_void(); + struct dlm_lock_resource *res; BUG_ON(!dlm); BUG_ON(!lock); + res = lock->lockres; + assert_spin_locked(&dlm->ast_lock); + if (!list_empty(&lock->ast_list)) { - mlog(ML_ERROR, "ast list not empty!! pending=%d, newlevel=%d\n", + mlog(ML_ERROR, "%s: res %.*s, lock %u:%llu, " + "AST list not empty, pending %d, newlevel %d\n", + dlm->name, res->lockname.len, res->lockname.name, + dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)), lock->ast_pending, lock->ml.type); BUG(); } - BUG_ON(!list_empty(&lock->ast_list)); if (lock->ast_pending) - mlog(0, "lock has an ast getting flushed right now\n"); + mlog(0, "%s: res %.*s, lock %u:%llu, AST getting flushed\n", + dlm->name, res->lockname.len, res->lockname.name, + dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie))); /* putting lock on list, add a ref */ dlm_lock_get(lock); @@ -114,9 +120,10 @@ static void __dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock) /* check to see if this ast obsoletes the bast */ if (dlm_should_cancel_bast(dlm, lock)) { - struct dlm_lock_resource *res = lock->lockres; - mlog(0, "%s: cancelling bast for %.*s\n", - dlm->name, res->lockname.len, res->lockname.name); + mlog(0, "%s: res %.*s, lock %u:%llu, Cancelling BAST\n", + dlm->name, res->lockname.len, res->lockname.name, + dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie))); lock->bast_pending = 0; list_del_init(&lock->bast_list); lock->ml.highest_blocked = LKM_IVMODE; @@ -126,7 +133,7 @@ static void __dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock) dlm_lock_put(lock); /* free up the reserved bast that we are cancelling. * guaranteed that this will not be the last reserved - * ast because *both* an ast and a bast were reserved + * ast because *both* an ast and a bast were reserved * to get to this point. the res->spinlock will not be * taken here */ dlm_lockres_release_ast(dlm, res); @@ -138,8 +145,6 @@ static void __dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock) void dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock) { - mlog_entry_void(); - BUG_ON(!dlm); BUG_ON(!lock); @@ -149,17 +154,23 @@ void dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock) } -static void __dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock) +void __dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock) { - mlog_entry_void(); + struct dlm_lock_resource *res; BUG_ON(!dlm); BUG_ON(!lock); + assert_spin_locked(&dlm->ast_lock); + res = lock->lockres; + BUG_ON(!list_empty(&lock->bast_list)); if (lock->bast_pending) - mlog(0, "lock has a bast getting flushed right now\n"); + mlog(0, "%s: res %.*s, lock %u:%llu, BAST getting flushed\n", + dlm->name, res->lockname.len, res->lockname.name, + dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie))); /* putting lock on list, add a ref */ dlm_lock_get(lock); @@ -171,8 +182,6 @@ static void __dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock) void dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock) { - mlog_entry_void(); - BUG_ON(!dlm); BUG_ON(!lock); @@ -188,23 +197,24 @@ static void dlm_update_lvb(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, BUG_ON(!lksb); /* only updates if this node masters the lockres */ + spin_lock(&res->spinlock); if (res->owner == dlm->node_num) { - - spin_lock(&res->spinlock); /* check the lksb flags for the direction */ if (lksb->flags & DLM_LKSB_GET_LVB) { mlog(0, "getting lvb from lockres for %s node\n", lock->ml.node == dlm->node_num ? "master" : "remote"); memcpy(lksb->lvb, res->lvb, DLM_LVB_LEN); - } else if (lksb->flags & DLM_LKSB_PUT_LVB) { - mlog(0, "setting lvb from lockres for %s node\n", - lock->ml.node == dlm->node_num ? "master" : - "remote"); - memcpy(res->lvb, lksb->lvb, DLM_LVB_LEN); } - spin_unlock(&res->spinlock); + /* Do nothing for lvb put requests - they should be done in + * place when the lock is downconverted - otherwise we risk + * racing gets and puts which could result in old lvb data + * being propagated. We leave the put flag set and clear it + * here. In the future we might want to clear it at the time + * the put is actually done. + */ } + spin_unlock(&res->spinlock); /* reset any lvb flags on the lksb */ lksb->flags &= ~(DLM_LKSB_PUT_LVB|DLM_LKSB_GET_LVB); @@ -216,7 +226,10 @@ void dlm_do_local_ast(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, dlm_astlockfunc_t *fn; struct dlm_lockstatus *lksb; - mlog_entry_void(); + mlog(0, "%s: res %.*s, lock %u:%llu, Local AST\n", dlm->name, + res->lockname.len, res->lockname.name, + dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie))); lksb = lock->lksb; fn = lock->ast; @@ -234,7 +247,10 @@ int dlm_do_remote_ast(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, struct dlm_lockstatus *lksb; int lksbflags; - mlog_entry_void(); + mlog(0, "%s: res %.*s, lock %u:%llu, Remote AST\n", dlm->name, + res->lockname.len, res->lockname.name, + dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie))); lksb = lock->lksb; BUG_ON(lock->ml.node == dlm->node_num); @@ -253,15 +269,21 @@ void dlm_do_local_bast(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, { dlm_bastlockfunc_t *fn = lock->bast; - mlog_entry_void(); BUG_ON(lock->ml.node != dlm->node_num); + mlog(0, "%s: res %.*s, lock %u:%llu, Local BAST, blocked %d\n", + dlm->name, res->lockname.len, res->lockname.name, + dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)), + blocked_type); + (*fn)(lock->astdata, blocked_type); } -int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data) +int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { int ret; unsigned int locklen; @@ -270,9 +292,10 @@ int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data) struct dlm_lock *lock = NULL; struct dlm_proxy_ast *past = (struct dlm_proxy_ast *) msg->buf; char *name; - struct list_head *iter, *head=NULL; - u64 cookie; + struct list_head *head = NULL; + __be64 cookie; u32 flags; + u8 node; if (!dlm_grab(dlm)) { dlm_error(DLM_REJECTED); @@ -284,18 +307,21 @@ int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data) name = past->name; locklen = past->namelen; - cookie = be64_to_cpu(past->cookie); + cookie = past->cookie; flags = be32_to_cpu(past->flags); + node = past->node_idx; if (locklen > DLM_LOCKID_NAME_MAX) { ret = DLM_IVBUFLEN; - mlog(ML_ERROR, "Invalid name length in proxy ast handler!\n"); + mlog(ML_ERROR, "Invalid name length (%d) in proxy ast " + "handler!\n", locklen); goto leave; } if ((flags & (LKM_PUT_LVB|LKM_GET_LVB)) == (LKM_PUT_LVB|LKM_GET_LVB)) { - mlog(ML_ERROR, "both PUT and GET lvb specified\n"); + mlog(ML_ERROR, "Both PUT and GET lvb specified, (0x%x)\n", + flags); ret = DLM_BADARGS; goto leave; } @@ -307,18 +333,22 @@ int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data) if (past->type != DLM_AST && past->type != DLM_BAST) { - mlog(ML_ERROR, "Unknown ast type! %d, cookie=%"MLFu64", " - "name=%.*s\n", past->type, cookie, locklen, name); + mlog(ML_ERROR, "Unknown ast type! %d, cookie=%u:%llu" + "name=%.*s, node=%u\n", past->type, + dlm_get_lock_cookie_node(be64_to_cpu(cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(cookie)), + locklen, name, node); ret = DLM_IVLOCKID; goto leave; } res = dlm_lookup_lockres(dlm, name, locklen); if (!res) { - mlog(ML_ERROR, "got %sast for unknown lockres! " - "cookie=%"MLFu64", name=%.*s, namelen=%u\n", - past->type == DLM_AST ? "" : "b", - cookie, locklen, name, locklen); + mlog(0, "Got %sast for unknown lockres! cookie=%u:%llu, " + "name=%.*s, node=%u\n", (past->type == DLM_AST ? "" : "b"), + dlm_get_lock_cookie_node(be64_to_cpu(cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(cookie)), + locklen, name, node); ret = DLM_IVLOCKID; goto leave; } @@ -326,25 +356,25 @@ int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data) /* cannot get a proxy ast message if this node owns it */ BUG_ON(res->owner == dlm->node_num); - mlog(0, "lockres %.*s\n", res->lockname.len, res->lockname.name); + mlog(0, "%s: res %.*s\n", dlm->name, res->lockname.len, + res->lockname.name); spin_lock(&res->spinlock); if (res->state & DLM_LOCK_RES_RECOVERING) { - mlog(0, "responding with DLM_RECOVERING!\n"); + mlog(0, "Responding with DLM_RECOVERING!\n"); ret = DLM_RECOVERING; goto unlock_out; } if (res->state & DLM_LOCK_RES_MIGRATING) { - mlog(0, "responding with DLM_MIGRATING!\n"); + mlog(0, "Responding with DLM_MIGRATING!\n"); ret = DLM_MIGRATING; goto unlock_out; } /* try convert queue for both ast/bast */ head = &res->converting; lock = NULL; - list_for_each(iter, head) { - lock = list_entry (iter, struct dlm_lock, list); - if (be64_to_cpu(lock->ml.cookie) == cookie) + list_for_each_entry(lock, head, list) { + if (lock->ml.cookie == cookie) goto do_ast; } @@ -354,15 +384,16 @@ int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data) else head = &res->granted; - list_for_each(iter, head) { - lock = list_entry (iter, struct dlm_lock, list); - if (be64_to_cpu(lock->ml.cookie) == cookie) + list_for_each_entry(lock, head, list) { + if (lock->ml.cookie == cookie) goto do_ast; } - mlog(ML_ERROR, "got %sast for unknown lock! cookie=%"MLFu64", " - "name=%.*s, namelen=%u\n", - past->type == DLM_AST ? "" : "b", cookie, locklen, name, locklen); + mlog(0, "Got %sast for unknown lock! cookie=%u:%llu, name=%.*s, " + "node=%u\n", past->type == DLM_AST ? "" : "b", + dlm_get_lock_cookie_node(be64_to_cpu(cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(cookie)), + locklen, name, node); ret = DLM_NORMAL; unlock_out: @@ -373,10 +404,13 @@ do_ast: ret = DLM_NORMAL; if (past->type == DLM_AST) { /* do not alter lock refcount. switching lists. */ - list_del_init(&lock->list); - list_add_tail(&lock->list, &res->granted); - mlog(0, "ast: adding to granted list... type=%d, " - "convert_type=%d\n", lock->ml.type, lock->ml.convert_type); + list_move_tail(&lock->list, &res->granted); + mlog(0, "%s: res %.*s, lock %u:%llu, Granted type %d => %d\n", + dlm->name, res->lockname.len, res->lockname.name, + dlm_get_lock_cookie_node(be64_to_cpu(cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(cookie)), + lock->ml.type, lock->ml.convert_type); + if (lock->ml.convert_type != LKM_IVMODE) { lock->ml.type = lock->ml.convert_type; lock->ml.convert_type = LKM_IVMODE; @@ -400,7 +434,6 @@ do_ast: dlm_do_local_bast(dlm, res, lock, past->blocked_type); leave: - if (res) dlm_lockres_put(res); @@ -420,9 +453,9 @@ int dlm_send_proxy_ast_msg(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, size_t veclen = 1; int status; - mlog_entry("res %.*s, to=%u, type=%d, blocked_type=%d\n", - res->lockname.len, res->lockname.name, lock->ml.node, - msg_type, blocked_type); + mlog(0, "%s: res %.*s, to %u, type %d, blocked_type %d\n", dlm->name, + res->lockname.len, res->lockname.name, lock->ml.node, msg_type, + blocked_type); memset(&past, 0, sizeof(struct dlm_proxy_ast)); past.node_idx = dlm->node_num; @@ -435,7 +468,6 @@ int dlm_send_proxy_ast_msg(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, vec[0].iov_len = sizeof(struct dlm_proxy_ast); vec[0].iov_base = &past; if (flags & DLM_LKSB_GET_LVB) { - mlog(0, "returning requested LVB data\n"); be32_add_cpu(&past.flags, LKM_GET_LVB); vec[1].iov_len = DLM_LVB_LEN; vec[1].iov_base = lock->lksb->lvb; @@ -445,7 +477,9 @@ int dlm_send_proxy_ast_msg(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, ret = o2net_send_message_vec(DLM_PROXY_AST_MSG, dlm->key, vec, veclen, lock->ml.node, &status); if (ret < 0) - mlog_errno(ret); + mlog(ML_ERROR, "%s: res %.*s, error %d send AST to node %u\n", + dlm->name, res->lockname.len, res->lockname.name, ret, + lock->ml.node); else { if (status == DLM_RECOVERING) { mlog(ML_ERROR, "sent AST to node %u, it thinks this " @@ -455,7 +489,7 @@ int dlm_send_proxy_ast_msg(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, mlog(ML_ERROR, "sent AST to node %u, it returned " "DLM_MIGRATING!\n", lock->ml.node); BUG(); - } else if (status != DLM_NORMAL) { + } else if (status != DLM_NORMAL && status != DLM_IVLOCKID) { mlog(ML_ERROR, "AST to node %u returned %d!\n", lock->ml.node, status); /* ignore it */ diff --git a/fs/ocfs2/dlm/dlmcommon.h b/fs/ocfs2/dlm/dlmcommon.h index 3fecba0a602..fae17c640df 100644 --- a/fs/ocfs2/dlm/dlmcommon.h +++ b/fs/ocfs2/dlm/dlmcommon.h @@ -37,14 +37,53 @@ #define DLM_THREAD_SHUFFLE_INTERVAL 5 // flush everything every 5 passes #define DLM_THREAD_MS 200 // flush at least every 200 ms -#define DLM_HASH_BITS 7 -#define DLM_HASH_SIZE (1 << DLM_HASH_BITS) -#define DLM_HASH_MASK (DLM_HASH_SIZE - 1) +#define DLM_HASH_SIZE_DEFAULT (1 << 17) +#if DLM_HASH_SIZE_DEFAULT < PAGE_SIZE +# define DLM_HASH_PAGES 1 +#else +# define DLM_HASH_PAGES (DLM_HASH_SIZE_DEFAULT / PAGE_SIZE) +#endif +#define DLM_BUCKETS_PER_PAGE (PAGE_SIZE / sizeof(struct hlist_head)) +#define DLM_HASH_BUCKETS (DLM_HASH_PAGES * DLM_BUCKETS_PER_PAGE) + +/* Intended to make it easier for us to switch out hash functions */ +#define dlm_lockid_hash(_n, _l) full_name_hash(_n, _l) + +enum dlm_mle_type { + DLM_MLE_BLOCK = 0, + DLM_MLE_MASTER = 1, + DLM_MLE_MIGRATION = 2, + DLM_MLE_NUM_TYPES = 3, +}; + +struct dlm_master_list_entry { + struct hlist_node master_hash_node; + struct list_head hb_events; + struct dlm_ctxt *dlm; + spinlock_t spinlock; + wait_queue_head_t wq; + atomic_t woken; + struct kref mle_refs; + int inuse; + unsigned long maybe_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; + unsigned long vote_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; + unsigned long response_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; + unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; + u8 master; + u8 new_master; + enum dlm_mle_type type; + struct o2hb_callback_func mle_hb_up; + struct o2hb_callback_func mle_hb_down; + struct dlm_lock_resource *mleres; + unsigned char mname[DLM_LOCKID_NAME_MAX]; + unsigned int mnamelen; + unsigned int mnamehash; +}; enum dlm_ast_type { DLM_AST = 0, - DLM_BAST, - DLM_ASTUNLOCK + DLM_BAST = 1, + DLM_ASTUNLOCK = 2, }; @@ -63,12 +102,12 @@ static inline int dlm_is_recovery_lock(const char *lock_name, int name_len) return 0; } -#define DLM_RECO_STATE_ACTIVE 0x0001 +#define DLM_RECO_STATE_ACTIVE 0x0001 +#define DLM_RECO_STATE_FINALIZE 0x0002 struct dlm_recovery_ctxt { struct list_head resources; - struct list_head received; struct list_head node_data; u8 new_master; u8 dead_node; @@ -79,22 +118,24 @@ struct dlm_recovery_ctxt enum dlm_ctxt_state { DLM_CTXT_NEW = 0, - DLM_CTXT_JOINED, - DLM_CTXT_IN_SHUTDOWN, - DLM_CTXT_LEAVING, + DLM_CTXT_JOINED = 1, + DLM_CTXT_IN_SHUTDOWN = 2, + DLM_CTXT_LEAVING = 3, }; struct dlm_ctxt { struct list_head list; - struct list_head *resources; + struct hlist_head **lockres_hash; struct list_head dirty_list; struct list_head purge_list; struct list_head pending_asts; struct list_head pending_basts; + struct list_head tracking_list; unsigned int purge_count; spinlock_t spinlock; spinlock_t ast_lock; + spinlock_t track_lock; char *name; u8 node_num; u32 key; @@ -102,16 +143,21 @@ struct dlm_ctxt wait_queue_head_t dlm_join_events; unsigned long live_nodes_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; unsigned long domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; + unsigned long exit_domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; unsigned long recovery_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; struct dlm_recovery_ctxt reco; spinlock_t master_lock; - struct list_head master_list; + struct hlist_head **master_hash; struct list_head mle_hb_events; /* these give a really vague idea of the system load */ - atomic_t local_resources; - atomic_t remote_resources; - atomic_t unknown_resources; + atomic_t mle_tot_count[DLM_MLE_NUM_TYPES]; + atomic_t mle_cur_count[DLM_MLE_NUM_TYPES]; + atomic_t res_tot_count; + atomic_t res_cur_count; + + struct dlm_debug_ctxt *dlm_debug_ctxt; + struct dentry *dlm_debugfs_subroot; /* NOTE: Next three are protected by dlm_domain_lock */ struct kref dlm_refs; @@ -122,6 +168,7 @@ struct dlm_ctxt struct o2hb_callback_func dlm_hb_down; struct task_struct *dlm_thread_task; struct task_struct *dlm_reco_thread_task; + struct workqueue_struct *dlm_worker; wait_queue_head_t dlm_thread_wq; wait_queue_head_t dlm_reco_thread_wq; wait_queue_head_t ast_wq; @@ -132,13 +179,31 @@ struct dlm_ctxt spinlock_t work_lock; struct list_head dlm_domain_handlers; struct list_head dlm_eviction_callbacks; + + /* The filesystem specifies this at domain registration. We + * cache it here to know what to tell other nodes. */ + struct dlm_protocol_version fs_locking_proto; + /* This is the inter-dlm communication version */ + struct dlm_protocol_version dlm_locking_proto; }; +static inline struct hlist_head *dlm_lockres_hash(struct dlm_ctxt *dlm, unsigned i) +{ + return dlm->lockres_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] + (i % DLM_BUCKETS_PER_PAGE); +} + +static inline struct hlist_head *dlm_master_hash(struct dlm_ctxt *dlm, + unsigned i) +{ + return dlm->master_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] + + (i % DLM_BUCKETS_PER_PAGE); +} + /* these keventd work queue items are for less-frequently * called functions that cannot be directly called from the * net message handlers for some reason, usually because * they need to send net messages of their own. */ -void dlm_dispatch_work(void *data); +void dlm_dispatch_work(struct work_struct *work); struct dlm_lock_resource; struct dlm_work_item; @@ -155,6 +220,7 @@ struct dlm_mig_lockres_priv { struct dlm_lock_resource *lockres; u8 real_master; + u8 extra_ref; }; struct dlm_assert_master_priv @@ -165,6 +231,11 @@ struct dlm_assert_master_priv unsigned ignore_higher:1; }; +struct dlm_deref_lockres_priv +{ + struct dlm_lock_resource *deref_res; + u8 deref_node; +}; struct dlm_work_item { @@ -176,6 +247,7 @@ struct dlm_work_item struct dlm_request_all_locks_priv ral; struct dlm_mig_lockres_priv ml; struct dlm_assert_master_priv am; + struct dlm_deref_lockres_priv dl; } u; }; @@ -207,6 +279,12 @@ static inline void __dlm_set_joining_node(struct dlm_ctxt *dlm, #define DLM_LOCK_RES_DIRTY 0x00000008 #define DLM_LOCK_RES_IN_PROGRESS 0x00000010 #define DLM_LOCK_RES_MIGRATING 0x00000020 +#define DLM_LOCK_RES_DROPPING_REF 0x00000040 +#define DLM_LOCK_RES_BLOCK_DIRTY 0x00001000 +#define DLM_LOCK_RES_SETREF_INPROG 0x00002000 + +/* max milliseconds to wait to sync up a network failure with a node death */ +#define DLM_NODE_DEATH_WAIT_MAX (5 * 1000) #define DLM_PURGE_INTERVAL_MS (8 * 1000) @@ -214,31 +292,47 @@ struct dlm_lock_resource { /* WARNING: Please see the comment in dlm_init_lockres before * adding fields here. */ - struct list_head list; + struct hlist_node hash_node; + struct qstr lockname; struct kref refs; - /* please keep these next 3 in this order - * some funcs want to iterate over all lists */ + /* + * Please keep granted, converting, and blocked in this order, + * as some funcs want to iterate over all lists. + * + * All four lists are protected by the hash's reference. + */ struct list_head granted; struct list_head converting; struct list_head blocked; + struct list_head purge; + /* + * These two lists require you to hold an additional reference + * while they are on the list. + */ struct list_head dirty; struct list_head recovering; // dlm_recovery_ctxt.resources list + /* Added during init and removed during release */ + struct list_head tracking; /* dlm->tracking_list */ + /* unused lock resources have their last_used stamped and are * put on a list for the dlm thread to run. */ - struct list_head purge; unsigned long last_used; + struct dlm_ctxt *dlm; + unsigned migration_pending:1; atomic_t asts_reserved; spinlock_t spinlock; wait_queue_head_t wq; u8 owner; //node which owns the lock resource, or unknown u16 state; - struct qstr lockname; char lvb[DLM_LVB_LEN]; + unsigned int inflight_locks; + unsigned int inflight_assert_workers; + unsigned long refmap[BITS_TO_LONGS(O2NM_MAX_NODES)]; }; struct dlm_migratable_lock @@ -295,10 +389,31 @@ struct dlm_lock enum dlm_lockres_list { DLM_GRANTED_LIST = 0, - DLM_CONVERTING_LIST, - DLM_BLOCKED_LIST + DLM_CONVERTING_LIST = 1, + DLM_BLOCKED_LIST = 2, }; +static inline int dlm_lvb_is_empty(char *lvb) +{ + int i; + for (i=0; i<DLM_LVB_LEN; i++) + if (lvb[i]) + return 0; + return 1; +} + +static inline char *dlm_list_in_text(enum dlm_lockres_list idx) +{ + if (idx == DLM_GRANTED_LIST) + return "granted"; + else if (idx == DLM_CONVERTING_LIST) + return "converting"; + else if (idx == DLM_BLOCKED_LIST) + return "blocked"; + else + return "unknown"; +} + static inline struct list_head * dlm_list_idx_to_ptr(struct dlm_lock_resource *res, enum dlm_lockres_list idx) { @@ -325,25 +440,28 @@ struct dlm_node_iter enum { - DLM_MASTER_REQUEST_MSG = 500, - DLM_UNUSED_MSG1, /* 501 */ - DLM_ASSERT_MASTER_MSG, /* 502 */ - DLM_CREATE_LOCK_MSG, /* 503 */ - DLM_CONVERT_LOCK_MSG, /* 504 */ - DLM_PROXY_AST_MSG, /* 505 */ - DLM_UNLOCK_LOCK_MSG, /* 506 */ - DLM_UNUSED_MSG2, /* 507 */ - DLM_MIGRATE_REQUEST_MSG, /* 508 */ - DLM_MIG_LOCKRES_MSG, /* 509 */ - DLM_QUERY_JOIN_MSG, /* 510 */ - DLM_ASSERT_JOINED_MSG, /* 511 */ - DLM_CANCEL_JOIN_MSG, /* 512 */ - DLM_EXIT_DOMAIN_MSG, /* 513 */ - DLM_MASTER_REQUERY_MSG, /* 514 */ - DLM_LOCK_REQUEST_MSG, /* 515 */ - DLM_RECO_DATA_DONE_MSG, /* 516 */ - DLM_BEGIN_RECO_MSG, /* 517 */ - DLM_FINALIZE_RECO_MSG /* 518 */ + DLM_MASTER_REQUEST_MSG = 500, + DLM_UNUSED_MSG1 = 501, + DLM_ASSERT_MASTER_MSG = 502, + DLM_CREATE_LOCK_MSG = 503, + DLM_CONVERT_LOCK_MSG = 504, + DLM_PROXY_AST_MSG = 505, + DLM_UNLOCK_LOCK_MSG = 506, + DLM_DEREF_LOCKRES_MSG = 507, + DLM_MIGRATE_REQUEST_MSG = 508, + DLM_MIG_LOCKRES_MSG = 509, + DLM_QUERY_JOIN_MSG = 510, + DLM_ASSERT_JOINED_MSG = 511, + DLM_CANCEL_JOIN_MSG = 512, + DLM_EXIT_DOMAIN_MSG = 513, + DLM_MASTER_REQUERY_MSG = 514, + DLM_LOCK_REQUEST_MSG = 515, + DLM_RECO_DATA_DONE_MSG = 516, + DLM_BEGIN_RECO_MSG = 517, + DLM_FINALIZE_RECO_MSG = 518, + DLM_QUERY_REGION = 519, + DLM_QUERY_NODEINFO = 520, + DLM_BEGIN_EXIT_DOMAIN_MSG = 521, }; struct dlm_reco_node_data @@ -356,19 +474,19 @@ struct dlm_reco_node_data enum { DLM_RECO_NODE_DATA_DEAD = -1, DLM_RECO_NODE_DATA_INIT = 0, - DLM_RECO_NODE_DATA_REQUESTING, - DLM_RECO_NODE_DATA_REQUESTED, - DLM_RECO_NODE_DATA_RECEIVING, - DLM_RECO_NODE_DATA_DONE, - DLM_RECO_NODE_DATA_FINALIZE_SENT, + DLM_RECO_NODE_DATA_REQUESTING = 1, + DLM_RECO_NODE_DATA_REQUESTED = 2, + DLM_RECO_NODE_DATA_RECEIVING = 3, + DLM_RECO_NODE_DATA_DONE = 4, + DLM_RECO_NODE_DATA_FINALIZE_SENT = 5, }; enum { DLM_MASTER_RESP_NO = 0, - DLM_MASTER_RESP_YES, - DLM_MASTER_RESP_MAYBE, - DLM_MASTER_RESP_ERROR + DLM_MASTER_RESP_YES = 1, + DLM_MASTER_RESP_MAYBE = 2, + DLM_MASTER_RESP_ERROR = 3, }; @@ -382,6 +500,9 @@ struct dlm_master_request u8 name[O2NM_MAX_NAME_LEN]; }; +#define DLM_ASSERT_RESPONSE_REASSERT 0x00000001 +#define DLM_ASSERT_RESPONSE_MASTERY_REF 0x00000002 + #define DLM_ASSERT_MASTER_MLE_CLEANUP 0x00000001 #define DLM_ASSERT_MASTER_REQUERY 0x00000002 #define DLM_ASSERT_MASTER_FINISH_MIGRATION 0x00000004 @@ -395,6 +516,8 @@ struct dlm_assert_master u8 name[O2NM_MAX_NAME_LEN]; }; +#define DLM_MIGRATE_RESPONSE_MASTERY_REF 0x00000001 + struct dlm_migrate_request { u8 master; @@ -538,10 +661,26 @@ struct dlm_proxy_ast #define DLM_PROXY_AST_MAX_LEN (sizeof(struct dlm_proxy_ast)+DLM_LVB_LEN) #define DLM_MOD_KEY (0x666c6172) -enum dlm_query_join_response { +enum dlm_query_join_response_code { JOIN_DISALLOW = 0, - JOIN_OK, - JOIN_OK_NO_MAP, + JOIN_OK = 1, + JOIN_OK_NO_MAP = 2, + JOIN_PROTOCOL_MISMATCH = 3, +}; + +struct dlm_query_join_packet { + u8 code; /* Response code. dlm_minor and fs_minor + are only valid if this is JOIN_OK */ + u8 dlm_minor; /* The minor version of the protocol the + dlm is speaking. */ + u8 fs_minor; /* The minor version of the protocol the + filesystem is speaking. */ + u8 reserved; +}; + +union dlm_query_join_response { + __be32 intval; + struct dlm_query_join_packet packet; }; struct dlm_lock_request @@ -574,12 +713,18 @@ struct dlm_begin_reco }; +#define BITS_PER_BYTE 8 +#define BITS_TO_BYTES(bits) (((bits)+BITS_PER_BYTE-1)/BITS_PER_BYTE) + struct dlm_query_join_request { u8 node_idx; u8 pad1[2]; u8 name_len; + struct dlm_protocol_version dlm_proto; + struct dlm_protocol_version fs_proto; u8 domain[O2NM_MAX_NAME_LEN]; + u8 node_map[BITS_TO_BYTES(O2NM_MAX_NODES)]; }; struct dlm_assert_joined @@ -598,6 +743,31 @@ struct dlm_cancel_join u8 domain[O2NM_MAX_NAME_LEN]; }; +struct dlm_query_region { + u8 qr_node; + u8 qr_numregions; + u8 qr_namelen; + u8 pad1; + u8 qr_domain[O2NM_MAX_NAME_LEN]; + u8 qr_regions[O2HB_MAX_REGION_NAME_LEN * O2NM_MAX_REGIONS]; +}; + +struct dlm_node_info { + u8 ni_nodenum; + u8 pad1; + __be16 ni_ipv4_port; + __be32 ni_ipv4_address; +}; + +struct dlm_query_nodeinfo { + u8 qn_nodenum; + u8 qn_numnodes; + u8 qn_namelen; + u8 pad1; + u8 qn_domain[O2NM_MAX_NAME_LEN]; + struct dlm_node_info qn_nodes[O2NM_MAX_NODES]; +}; + struct dlm_exit_domain { u8 node_idx; @@ -608,10 +778,21 @@ struct dlm_finalize_reco { u8 node_idx; u8 dead_node; - __be16 pad1; + u8 flags; + u8 pad1; __be32 pad2; }; +struct dlm_deref_lockres +{ + u32 pad1; + u16 pad2; + u8 node_idx; + u8 namelen; + + u8 name[O2NM_MAX_NAME_LEN]; +}; + static inline enum dlm_status __dlm_lockres_state_to_status(struct dlm_lock_resource *res) { @@ -629,6 +810,21 @@ __dlm_lockres_state_to_status(struct dlm_lock_resource *res) return status; } +static inline u8 dlm_get_lock_cookie_node(u64 cookie) +{ + u8 ret; + cookie >>= 56; + ret = (u8)(cookie & 0xffULL); + return ret; +} + +static inline unsigned long long dlm_get_lock_cookie_seq(u64 cookie) +{ + unsigned long long ret; + ret = ((unsigned long long)cookie) & 0x00ffffffffffffffULL; + return ret; +} + struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie, struct dlm_lockstatus *lksb); void dlm_lock_get(struct dlm_lock *lock); @@ -637,16 +833,20 @@ void dlm_lock_put(struct dlm_lock *lock); void dlm_lock_attach_lockres(struct dlm_lock *lock, struct dlm_lock_resource *res); -int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data); -int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data); -int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data); +int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); +int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); +int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); void dlm_revert_pending_convert(struct dlm_lock_resource *res, struct dlm_lock *lock); void dlm_revert_pending_lock(struct dlm_lock_resource *res, struct dlm_lock *lock); -int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data); +int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); void dlm_commit_pending_cancel(struct dlm_lock_resource *res, struct dlm_lock *lock); void dlm_commit_pending_unlock(struct dlm_lock_resource *res, @@ -657,6 +857,10 @@ void dlm_complete_thread(struct dlm_ctxt *dlm); int dlm_launch_recovery_thread(struct dlm_ctxt *dlm); void dlm_complete_recovery_thread(struct dlm_ctxt *dlm); void dlm_wait_for_recovery(struct dlm_ctxt *dlm); +void dlm_kick_recovery_thread(struct dlm_ctxt *dlm); +int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node); +void dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout); +void dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout); void dlm_put(struct dlm_ctxt *dlm); struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm); @@ -666,33 +870,54 @@ void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); void dlm_lockres_calc_usage(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); -void dlm_purge_lockres(struct dlm_ctxt *dlm, - struct dlm_lock_resource *lockres); -void dlm_lockres_get(struct dlm_lock_resource *res); +static inline void dlm_lockres_get(struct dlm_lock_resource *res) +{ + /* This is called on every lookup, so it might be worth + * inlining. */ + kref_get(&res->refs); +} void dlm_lockres_put(struct dlm_lock_resource *res); -void __dlm_unhash_lockres(struct dlm_lock_resource *res); -void __dlm_insert_lockres(struct dlm_ctxt *dlm, - struct dlm_lock_resource *res); +void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); +void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); +struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm, + const char *name, + unsigned int len, + unsigned int hash); struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm, const char *name, - unsigned int len); + unsigned int len, + unsigned int hash); struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm, const char *name, unsigned int len); int dlm_is_host_down(int errno); -void dlm_change_lockres_owner(struct dlm_ctxt *dlm, - struct dlm_lock_resource *res, - u8 owner); + struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm, const char *lockid, + int namelen, int flags); struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm, const char *name, unsigned int namelen); +void dlm_lockres_set_refmap_bit(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res, int bit); +void dlm_lockres_clear_refmap_bit(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res, int bit); + +void dlm_lockres_drop_inflight_ref(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res); +void dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res); + +void __dlm_lockres_grab_inflight_worker(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res); + void dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock); void dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock); +void __dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock); +void __dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock); void dlm_do_local_ast(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, struct dlm_lock *lock); @@ -739,10 +964,7 @@ int dlm_heartbeat_init(struct dlm_ctxt *dlm); void dlm_hb_node_down_cb(struct o2nm_node *node, int idx, void *data); void dlm_hb_node_up_cb(struct o2nm_node *node, int idx, void *data); -int dlm_lockres_is_dirty(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); -int dlm_migrate_lockres(struct dlm_ctxt *dlm, - struct dlm_lock_resource *res, - u8 target); +int dlm_empty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); int dlm_finish_migration(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, u8 old_master); @@ -750,15 +972,30 @@ void dlm_lockres_release_ast(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); void __dlm_lockres_reserve_ast(struct dlm_lock_resource *res); -int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data); -int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data); -int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data); -int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data); -int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data); -int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data); -int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data); -int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data); -int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data); +int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); +int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); +void dlm_assert_master_post_handler(int status, void *data, void *ret_data); +int dlm_deref_lockres_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); +int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); +int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); +int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); +int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); +int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); +int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); +int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); +int dlm_do_master_requery(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, + u8 nodenum, u8 *real_master); + int dlm_dispatch_assert_master(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, @@ -787,14 +1024,28 @@ static inline void __dlm_wait_on_lockres(struct dlm_lock_resource *res) DLM_LOCK_RES_MIGRATING)); } +void __dlm_unlink_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle); +void __dlm_insert_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle); + +/* create/destroy slab caches */ +int dlm_init_master_caches(void); +void dlm_destroy_master_caches(void); + +int dlm_init_lock_cache(void); +void dlm_destroy_lock_cache(void); int dlm_init_mle_cache(void); void dlm_destroy_mle_cache(void); + void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up); +int dlm_drop_lockres_ref(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res); void dlm_clean_master_list(struct dlm_ctxt *dlm, u8 dead_node); +void dlm_force_free_mles(struct dlm_ctxt *dlm); int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock); - +int __dlm_lockres_has_locks(struct dlm_lock_resource *res); +int __dlm_lockres_unused(struct dlm_lock_resource *res); static inline const char * dlm_lock_mode_name(int mode) { @@ -831,11 +1082,9 @@ static inline int dlm_lock_compatible(int existing, int request) static inline int dlm_lock_on_list(struct list_head *head, struct dlm_lock *lock) { - struct list_head *iter; struct dlm_lock *tmplock; - list_for_each(iter, head) { - tmplock = list_entry(iter, struct dlm_lock, list); + list_for_each_entry(tmplock, head, list) { if (tmplock == lock) return 1; } @@ -879,6 +1128,23 @@ static inline int dlm_node_iter_next(struct dlm_node_iter *iter) return bit; } +static inline void dlm_set_lockres_owner(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res, + u8 owner) +{ + assert_spin_locked(&res->spinlock); + res->owner = owner; +} + +static inline void dlm_change_lockres_owner(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res, + u8 owner) +{ + assert_spin_locked(&res->spinlock); + + if (owner != res->owner) + dlm_set_lockres_owner(dlm, res, owner); +} #endif /* DLMCOMMON_H */ diff --git a/fs/ocfs2/dlm/dlmconvert.c b/fs/ocfs2/dlm/dlmconvert.c index 6001b22a997..e36d63ff178 100644 --- a/fs/ocfs2/dlm/dlmconvert.c +++ b/fs/ocfs2/dlm/dlmconvert.c @@ -28,9 +28,7 @@ #include <linux/module.h> #include <linux/fs.h> #include <linux/types.h> -#include <linux/slab.h> #include <linux/highmem.h> -#include <linux/utsname.h> #include <linux/init.h> #include <linux/sysctl.h> #include <linux/random.h> @@ -125,13 +123,12 @@ static enum dlm_status __dlmconvert_master(struct dlm_ctxt *dlm, int *kick_thread) { enum dlm_status status = DLM_NORMAL; - struct list_head *iter; struct dlm_lock *tmplock=NULL; assert_spin_locked(&res->spinlock); - mlog_entry("type=%d, convert_type=%d, new convert_type=%d\n", - lock->ml.type, lock->ml.convert_type, type); + mlog(0, "type=%d, convert_type=%d, new convert_type=%d\n", + lock->ml.type, lock->ml.convert_type, type); spin_lock(&lock->spinlock); @@ -187,16 +184,14 @@ static enum dlm_status __dlmconvert_master(struct dlm_ctxt *dlm, /* upconvert from here on */ status = DLM_NORMAL; - list_for_each(iter, &res->granted) { - tmplock = list_entry(iter, struct dlm_lock, list); + list_for_each_entry(tmplock, &res->granted, list) { if (tmplock == lock) continue; if (!dlm_lock_compatible(tmplock->ml.type, type)) goto switch_queues; } - list_for_each(iter, &res->converting) { - tmplock = list_entry(iter, struct dlm_lock, list); + list_for_each_entry(tmplock, &res->converting, list) { if (!dlm_lock_compatible(tmplock->ml.type, type)) goto switch_queues; /* existing conversion requests take precedence */ @@ -214,6 +209,9 @@ grant: if (lock->ml.node == dlm->node_num) mlog(0, "doing in-place convert for nonlocal lock\n"); lock->ml.type = type; + if (lock->lksb->flags & DLM_LKSB_PUT_LVB) + memcpy(res->lvb, lock->lksb->lvb, DLM_LVB_LEN); + status = DLM_NORMAL; *call_ast = 1; goto unlock_exit; @@ -231,8 +229,7 @@ switch_queues: lock->ml.convert_type = type; /* do not alter lock refcount. switching lists. */ - list_del_init(&lock->list); - list_add_tail(&lock->list, &res->converting); + list_move_tail(&lock->list, &res->converting); unlock_exit: spin_unlock(&lock->spinlock); @@ -248,8 +245,7 @@ void dlm_revert_pending_convert(struct dlm_lock_resource *res, struct dlm_lock *lock) { /* do not alter lock refcount. switching lists. */ - list_del_init(&lock->list); - list_add_tail(&lock->list, &res->granted); + list_move_tail(&lock->list, &res->granted); lock->ml.convert_type = LKM_IVMODE; lock->lksb->flags &= ~(DLM_LKSB_GET_LVB|DLM_LKSB_PUT_LVB); } @@ -284,16 +280,17 @@ enum dlm_status dlmconvert_remote(struct dlm_ctxt *dlm, if (lock->ml.convert_type != LKM_IVMODE) { __dlm_print_one_lock_resource(res); mlog(ML_ERROR, "converting a remote lock that is already " - "converting! (cookie=%"MLFu64", conv=%d)\n", - lock->ml.cookie, lock->ml.convert_type); + "converting! (cookie=%u:%llu, conv=%d)\n", + dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)), + lock->ml.convert_type); status = DLM_DENIED; goto bail; } res->state |= DLM_LOCK_RES_IN_PROGRESS; /* move lock to local convert queue */ /* do not alter lock refcount. switching lists. */ - list_del_init(&lock->list); - list_add_tail(&lock->list, &res->converting); + list_move_tail(&lock->list, &res->converting); lock->convert_pending = 1; lock->ml.convert_type = type; @@ -353,7 +350,7 @@ static enum dlm_status dlm_send_remote_convert_request(struct dlm_ctxt *dlm, struct kvec vec[2]; size_t veclen = 1; - mlog_entry("%.*s\n", res->lockname.len, res->lockname.name); + mlog(0, "%.*s\n", res->lockname.len, res->lockname.name); memset(&convert, 0, sizeof(struct dlm_convert_lock)); convert.node_idx = dlm->node_num; @@ -390,8 +387,15 @@ static enum dlm_status dlm_send_remote_convert_request(struct dlm_ctxt *dlm, } else if (ret != DLM_NORMAL && ret != DLM_NOTQUEUED) dlm_error(ret); } else { - mlog_errno(tmpret); + mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to " + "node %u\n", tmpret, DLM_CONVERT_LOCK_MSG, dlm->key, + res->owner); if (dlm_is_host_down(tmpret)) { + /* instead of logging the same network error over + * and over, sleep here and wait for the heartbeat + * to notice the node is dead. times out after 5s. */ + dlm_wait_for_node_death(dlm, res->owner, + DLM_NODE_DEATH_WAIT_MAX); ret = DLM_RECOVERING; mlog(0, "node %u died so returning DLM_RECOVERING " "from convert message!\n", res->owner); @@ -411,17 +415,18 @@ static enum dlm_status dlm_send_remote_convert_request(struct dlm_ctxt *dlm, * returns: DLM_NORMAL, DLM_IVLOCKID, DLM_BADARGS, * status from __dlmconvert_master */ -int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data) +int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { struct dlm_ctxt *dlm = data; struct dlm_convert_lock *cnv = (struct dlm_convert_lock *)msg->buf; struct dlm_lock_resource *res = NULL; - struct list_head *iter; struct dlm_lock *lock = NULL; + struct dlm_lock *tmp_lock; struct dlm_lockstatus *lksb; enum dlm_status status = DLM_NORMAL; u32 flags; - int call_ast = 0, kick_thread = 0; + int call_ast = 0, kick_thread = 0, ast_reserved = 0, wake = 0; if (!dlm_grab(dlm)) { dlm_error(DLM_REJECTED); @@ -457,19 +462,28 @@ int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data) } spin_lock(&res->spinlock); - list_for_each(iter, &res->granted) { - lock = list_entry(iter, struct dlm_lock, list); - if (lock->ml.cookie == cnv->cookie && - lock->ml.node == cnv->node_idx) { + status = __dlm_lockres_state_to_status(res); + if (status != DLM_NORMAL) { + spin_unlock(&res->spinlock); + dlm_error(status); + goto leave; + } + list_for_each_entry(tmp_lock, &res->granted, list) { + if (tmp_lock->ml.cookie == cnv->cookie && + tmp_lock->ml.node == cnv->node_idx) { + lock = tmp_lock; dlm_lock_get(lock); break; } - lock = NULL; } spin_unlock(&res->spinlock); if (!lock) { status = DLM_IVLOCKID; - dlm_error(status); + mlog(ML_ERROR, "did not find lock to convert on grant queue! " + "cookie=%u:%llu\n", + dlm_get_lock_cookie_node(be64_to_cpu(cnv->cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(cnv->cookie))); + dlm_print_one_lock_resource(res); goto leave; } @@ -490,13 +504,17 @@ int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data) status = __dlm_lockres_state_to_status(res); if (status == DLM_NORMAL) { __dlm_lockres_reserve_ast(res); + ast_reserved = 1; res->state |= DLM_LOCK_RES_IN_PROGRESS; status = __dlmconvert_master(dlm, res, lock, flags, cnv->requested_type, &call_ast, &kick_thread); res->state &= ~DLM_LOCK_RES_IN_PROGRESS; + wake = 1; } spin_unlock(&res->spinlock); + if (wake) + wake_up(&res->wq); if (status != DLM_NORMAL) { if (status != DLM_NOTQUEUED) @@ -505,17 +523,13 @@ int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data) } leave: - if (!lock) - mlog(ML_ERROR, "did not find lock to convert on grant queue! " - "cookie=%"MLFu64"\n", - cnv->cookie); - else + if (lock) dlm_lock_put(lock); - /* either queue the ast or release it */ + /* either queue the ast or release it, if reserved */ if (call_ast) dlm_queue_ast(dlm, lock); - else + else if (ast_reserved) dlm_lockres_release_ast(dlm, res); if (kick_thread) diff --git a/fs/ocfs2/dlm/dlmdebug.c b/fs/ocfs2/dlm/dlmdebug.c index f339fe27975..18f13c2e4a1 100644 --- a/fs/ocfs2/dlm/dlmdebug.c +++ b/fs/ocfs2/dlm/dlmdebug.c @@ -5,7 +5,7 @@ * * debug functionality for the dlm * - * Copyright (C) 2004 Oracle. All rights reserved. + * Copyright (C) 2004, 2008 Oracle. All rights reserved. * * This program is free software; you can redistribute it and/or * modify it under the terms of the GNU General Public @@ -27,9 +27,10 @@ #include <linux/types.h> #include <linux/slab.h> #include <linux/highmem.h> -#include <linux/utsname.h> #include <linux/sysctl.h> #include <linux/spinlock.h> +#include <linux/debugfs.h> +#include <linux/export.h> #include "cluster/heartbeat.h" #include "cluster/nodemanager.h" @@ -37,107 +38,103 @@ #include "dlmapi.h" #include "dlmcommon.h" -#include "dlmdebug.h" - #include "dlmdomain.h" #include "dlmdebug.h" #define MLOG_MASK_PREFIX ML_DLM #include "cluster/masklog.h" +static int stringify_lockname(const char *lockname, int locklen, char *buf, + int len); + void dlm_print_one_lock_resource(struct dlm_lock_resource *res) { - mlog(ML_NOTICE, "lockres: %.*s, owner=%u, state=%u\n", - res->lockname.len, res->lockname.name, - res->owner, res->state); spin_lock(&res->spinlock); __dlm_print_one_lock_resource(res); spin_unlock(&res->spinlock); } -void __dlm_print_one_lock_resource(struct dlm_lock_resource *res) +static void dlm_print_lockres_refmap(struct dlm_lock_resource *res) { - struct list_head *iter2; - struct dlm_lock *lock; - + int bit; assert_spin_locked(&res->spinlock); - mlog(ML_NOTICE, "lockres: %.*s, owner=%u, state=%u\n", - res->lockname.len, res->lockname.name, - res->owner, res->state); - mlog(ML_NOTICE, " last used: %lu, on purge list: %s\n", - res->last_used, list_empty(&res->purge) ? "no" : "yes"); - mlog(ML_NOTICE, " granted queue: \n"); - list_for_each(iter2, &res->granted) { - lock = list_entry(iter2, struct dlm_lock, list); - spin_lock(&lock->spinlock); - mlog(ML_NOTICE, " type=%d, conv=%d, node=%u, " - "cookie=%"MLFu64", ast=(empty=%c,pend=%c), bast=(empty=%c,pend=%c)\n", - lock->ml.type, lock->ml.convert_type, lock->ml.node, lock->ml.cookie, - list_empty(&lock->ast_list) ? 'y' : 'n', - lock->ast_pending ? 'y' : 'n', - list_empty(&lock->bast_list) ? 'y' : 'n', - lock->bast_pending ? 'y' : 'n'); - spin_unlock(&lock->spinlock); - } - mlog(ML_NOTICE, " converting queue: \n"); - list_for_each(iter2, &res->converting) { - lock = list_entry(iter2, struct dlm_lock, list); - spin_lock(&lock->spinlock); - mlog(ML_NOTICE, " type=%d, conv=%d, node=%u, " - "cookie=%"MLFu64", ast=(empty=%c,pend=%c), bast=(empty=%c,pend=%c)\n", - lock->ml.type, lock->ml.convert_type, lock->ml.node, lock->ml.cookie, - list_empty(&lock->ast_list) ? 'y' : 'n', - lock->ast_pending ? 'y' : 'n', - list_empty(&lock->bast_list) ? 'y' : 'n', - lock->bast_pending ? 'y' : 'n'); - spin_unlock(&lock->spinlock); - } - mlog(ML_NOTICE, " blocked queue: \n"); - list_for_each(iter2, &res->blocked) { - lock = list_entry(iter2, struct dlm_lock, list); - spin_lock(&lock->spinlock); - mlog(ML_NOTICE, " type=%d, conv=%d, node=%u, " - "cookie=%"MLFu64", ast=(empty=%c,pend=%c), bast=(empty=%c,pend=%c)\n", - lock->ml.type, lock->ml.convert_type, lock->ml.node, lock->ml.cookie, - list_empty(&lock->ast_list) ? 'y' : 'n', - lock->ast_pending ? 'y' : 'n', - list_empty(&lock->bast_list) ? 'y' : 'n', - lock->bast_pending ? 'y' : 'n'); - spin_unlock(&lock->spinlock); + printk(" refmap nodes: [ "); + bit = 0; + while (1) { + bit = find_next_bit(res->refmap, O2NM_MAX_NODES, bit); + if (bit >= O2NM_MAX_NODES) + break; + printk("%u ", bit); + bit++; } + printk("], inflight=%u\n", res->inflight_locks); } -void dlm_print_one_lock(struct dlm_lock *lockid) +static void __dlm_print_lock(struct dlm_lock *lock) { - dlm_print_one_lock_resource(lockid->lockres); + spin_lock(&lock->spinlock); + + printk(" type=%d, conv=%d, node=%u, cookie=%u:%llu, " + "ref=%u, ast=(empty=%c,pend=%c), bast=(empty=%c,pend=%c), " + "pending=(conv=%c,lock=%c,cancel=%c,unlock=%c)\n", + lock->ml.type, lock->ml.convert_type, lock->ml.node, + dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)), + atomic_read(&lock->lock_refs.refcount), + (list_empty(&lock->ast_list) ? 'y' : 'n'), + (lock->ast_pending ? 'y' : 'n'), + (list_empty(&lock->bast_list) ? 'y' : 'n'), + (lock->bast_pending ? 'y' : 'n'), + (lock->convert_pending ? 'y' : 'n'), + (lock->lock_pending ? 'y' : 'n'), + (lock->cancel_pending ? 'y' : 'n'), + (lock->unlock_pending ? 'y' : 'n')); + + spin_unlock(&lock->spinlock); } -EXPORT_SYMBOL_GPL(dlm_print_one_lock); -void dlm_dump_lock_resources(struct dlm_ctxt *dlm) +void __dlm_print_one_lock_resource(struct dlm_lock_resource *res) { - struct dlm_lock_resource *res; - struct list_head *iter; - struct list_head *bucket; - int i; + struct dlm_lock *lock; + char buf[DLM_LOCKID_NAME_MAX]; - mlog(ML_NOTICE, "struct dlm_ctxt: %s, node=%u, key=%u\n", - dlm->name, dlm->node_num, dlm->key); - if (!dlm || !dlm->name) { - mlog(ML_ERROR, "dlm=%p\n", dlm); - return; - } + assert_spin_locked(&res->spinlock); - spin_lock(&dlm->spinlock); - for (i=0; i<DLM_HASH_SIZE; i++) { - bucket = &(dlm->resources[i]); - list_for_each(iter, bucket) { - res = list_entry(iter, struct dlm_lock_resource, list); - dlm_print_one_lock_resource(res); - } + stringify_lockname(res->lockname.name, res->lockname.len, + buf, sizeof(buf)); + printk("lockres: %s, owner=%u, state=%u\n", + buf, res->owner, res->state); + printk(" last used: %lu, refcnt: %u, on purge list: %s\n", + res->last_used, atomic_read(&res->refs.refcount), + list_empty(&res->purge) ? "no" : "yes"); + printk(" on dirty list: %s, on reco list: %s, " + "migrating pending: %s\n", + list_empty(&res->dirty) ? "no" : "yes", + list_empty(&res->recovering) ? "no" : "yes", + res->migration_pending ? "yes" : "no"); + printk(" inflight locks: %d, asts reserved: %d\n", + res->inflight_locks, atomic_read(&res->asts_reserved)); + dlm_print_lockres_refmap(res); + printk(" granted queue:\n"); + list_for_each_entry(lock, &res->granted, list) { + __dlm_print_lock(lock); } - spin_unlock(&dlm->spinlock); + printk(" converting queue:\n"); + list_for_each_entry(lock, &res->converting, list) { + __dlm_print_lock(lock); + } + printk(" blocked queue:\n"); + list_for_each_entry(lock, &res->blocked, list) { + __dlm_print_lock(lock); + } +} + +void dlm_print_one_lock(struct dlm_lock *lockid) +{ + dlm_print_one_lock_resource(lockid->lockres); } +EXPORT_SYMBOL_GPL(dlm_print_one_lock); static const char *dlm_errnames[] = { [DLM_NORMAL] = "DLM_NORMAL", @@ -244,3 +241,771 @@ const char *dlm_errname(enum dlm_status err) return dlm_errnames[err]; } EXPORT_SYMBOL_GPL(dlm_errname); + +/* NOTE: This function converts a lockname into a string. It uses knowledge + * of the format of the lockname that should be outside the purview of the dlm. + * We are adding only to make dlm debugging slightly easier. + * + * For more on lockname formats, please refer to dlmglue.c and ocfs2_lockid.h. + */ +static int stringify_lockname(const char *lockname, int locklen, char *buf, + int len) +{ + int out = 0; + __be64 inode_blkno_be; + +#define OCFS2_DENTRY_LOCK_INO_START 18 + if (*lockname == 'N') { + memcpy((__be64 *)&inode_blkno_be, + (char *)&lockname[OCFS2_DENTRY_LOCK_INO_START], + sizeof(__be64)); + out += snprintf(buf + out, len - out, "%.*s%08x", + OCFS2_DENTRY_LOCK_INO_START - 1, lockname, + (unsigned int)be64_to_cpu(inode_blkno_be)); + } else + out += snprintf(buf + out, len - out, "%.*s", + locklen, lockname); + return out; +} + +static int stringify_nodemap(unsigned long *nodemap, int maxnodes, + char *buf, int len) +{ + int out = 0; + int i = -1; + + while ((i = find_next_bit(nodemap, maxnodes, i + 1)) < maxnodes) + out += snprintf(buf + out, len - out, "%d ", i); + + return out; +} + +static int dump_mle(struct dlm_master_list_entry *mle, char *buf, int len) +{ + int out = 0; + char *mle_type; + + if (mle->type == DLM_MLE_BLOCK) + mle_type = "BLK"; + else if (mle->type == DLM_MLE_MASTER) + mle_type = "MAS"; + else + mle_type = "MIG"; + + out += stringify_lockname(mle->mname, mle->mnamelen, buf + out, len - out); + out += snprintf(buf + out, len - out, + "\t%3s\tmas=%3u\tnew=%3u\tevt=%1d\tuse=%1d\tref=%3d\n", + mle_type, mle->master, mle->new_master, + !list_empty(&mle->hb_events), + !!mle->inuse, + atomic_read(&mle->mle_refs.refcount)); + + out += snprintf(buf + out, len - out, "Maybe="); + out += stringify_nodemap(mle->maybe_map, O2NM_MAX_NODES, + buf + out, len - out); + out += snprintf(buf + out, len - out, "\n"); + + out += snprintf(buf + out, len - out, "Vote="); + out += stringify_nodemap(mle->vote_map, O2NM_MAX_NODES, + buf + out, len - out); + out += snprintf(buf + out, len - out, "\n"); + + out += snprintf(buf + out, len - out, "Response="); + out += stringify_nodemap(mle->response_map, O2NM_MAX_NODES, + buf + out, len - out); + out += snprintf(buf + out, len - out, "\n"); + + out += snprintf(buf + out, len - out, "Node="); + out += stringify_nodemap(mle->node_map, O2NM_MAX_NODES, + buf + out, len - out); + out += snprintf(buf + out, len - out, "\n"); + + out += snprintf(buf + out, len - out, "\n"); + + return out; +} + +void dlm_print_one_mle(struct dlm_master_list_entry *mle) +{ + char *buf; + + buf = (char *) get_zeroed_page(GFP_NOFS); + if (buf) { + dump_mle(mle, buf, PAGE_SIZE - 1); + free_page((unsigned long)buf); + } +} + +#ifdef CONFIG_DEBUG_FS + +static struct dentry *dlm_debugfs_root; + +#define DLM_DEBUGFS_DIR "o2dlm" +#define DLM_DEBUGFS_DLM_STATE "dlm_state" +#define DLM_DEBUGFS_LOCKING_STATE "locking_state" +#define DLM_DEBUGFS_MLE_STATE "mle_state" +#define DLM_DEBUGFS_PURGE_LIST "purge_list" + +/* begin - utils funcs */ +static void dlm_debug_free(struct kref *kref) +{ + struct dlm_debug_ctxt *dc; + + dc = container_of(kref, struct dlm_debug_ctxt, debug_refcnt); + + kfree(dc); +} + +static void dlm_debug_put(struct dlm_debug_ctxt *dc) +{ + if (dc) + kref_put(&dc->debug_refcnt, dlm_debug_free); +} + +static void dlm_debug_get(struct dlm_debug_ctxt *dc) +{ + kref_get(&dc->debug_refcnt); +} + +static int debug_release(struct inode *inode, struct file *file) +{ + free_page((unsigned long)file->private_data); + return 0; +} + +static ssize_t debug_read(struct file *file, char __user *buf, + size_t nbytes, loff_t *ppos) +{ + return simple_read_from_buffer(buf, nbytes, ppos, file->private_data, + i_size_read(file->f_mapping->host)); +} +/* end - util funcs */ + +/* begin - purge list funcs */ +static int debug_purgelist_print(struct dlm_ctxt *dlm, char *buf, int len) +{ + struct dlm_lock_resource *res; + int out = 0; + unsigned long total = 0; + + out += snprintf(buf + out, len - out, + "Dumping Purgelist for Domain: %s\n", dlm->name); + + spin_lock(&dlm->spinlock); + list_for_each_entry(res, &dlm->purge_list, purge) { + ++total; + if (len - out < 100) + continue; + spin_lock(&res->spinlock); + out += stringify_lockname(res->lockname.name, + res->lockname.len, + buf + out, len - out); + out += snprintf(buf + out, len - out, "\t%ld\n", + (jiffies - res->last_used)/HZ); + spin_unlock(&res->spinlock); + } + spin_unlock(&dlm->spinlock); + + out += snprintf(buf + out, len - out, "Total on list: %ld\n", total); + + return out; +} + +static int debug_purgelist_open(struct inode *inode, struct file *file) +{ + struct dlm_ctxt *dlm = inode->i_private; + char *buf = NULL; + + buf = (char *) get_zeroed_page(GFP_NOFS); + if (!buf) + goto bail; + + i_size_write(inode, debug_purgelist_print(dlm, buf, PAGE_SIZE - 1)); + + file->private_data = buf; + + return 0; +bail: + return -ENOMEM; +} + +static const struct file_operations debug_purgelist_fops = { + .open = debug_purgelist_open, + .release = debug_release, + .read = debug_read, + .llseek = generic_file_llseek, +}; +/* end - purge list funcs */ + +/* begin - debug mle funcs */ +static int debug_mle_print(struct dlm_ctxt *dlm, char *buf, int len) +{ + struct dlm_master_list_entry *mle; + struct hlist_head *bucket; + int i, out = 0; + unsigned long total = 0, longest = 0, bucket_count = 0; + + out += snprintf(buf + out, len - out, + "Dumping MLEs for Domain: %s\n", dlm->name); + + spin_lock(&dlm->master_lock); + for (i = 0; i < DLM_HASH_BUCKETS; i++) { + bucket = dlm_master_hash(dlm, i); + hlist_for_each_entry(mle, bucket, master_hash_node) { + ++total; + ++bucket_count; + if (len - out < 200) + continue; + out += dump_mle(mle, buf + out, len - out); + } + longest = max(longest, bucket_count); + bucket_count = 0; + } + spin_unlock(&dlm->master_lock); + + out += snprintf(buf + out, len - out, + "Total: %ld, Longest: %ld\n", total, longest); + return out; +} + +static int debug_mle_open(struct inode *inode, struct file *file) +{ + struct dlm_ctxt *dlm = inode->i_private; + char *buf = NULL; + + buf = (char *) get_zeroed_page(GFP_NOFS); + if (!buf) + goto bail; + + i_size_write(inode, debug_mle_print(dlm, buf, PAGE_SIZE - 1)); + + file->private_data = buf; + + return 0; +bail: + return -ENOMEM; +} + +static const struct file_operations debug_mle_fops = { + .open = debug_mle_open, + .release = debug_release, + .read = debug_read, + .llseek = generic_file_llseek, +}; + +/* end - debug mle funcs */ + +/* begin - debug lockres funcs */ +static int dump_lock(struct dlm_lock *lock, int list_type, char *buf, int len) +{ + int out; + +#define DEBUG_LOCK_VERSION 1 + spin_lock(&lock->spinlock); + out = snprintf(buf, len, "LOCK:%d,%d,%d,%d,%d,%d:%lld,%d,%d,%d,%d,%d," + "%d,%d,%d,%d\n", + DEBUG_LOCK_VERSION, + list_type, lock->ml.type, lock->ml.convert_type, + lock->ml.node, + dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)), + !list_empty(&lock->ast_list), + !list_empty(&lock->bast_list), + lock->ast_pending, lock->bast_pending, + lock->convert_pending, lock->lock_pending, + lock->cancel_pending, lock->unlock_pending, + atomic_read(&lock->lock_refs.refcount)); + spin_unlock(&lock->spinlock); + + return out; +} + +static int dump_lockres(struct dlm_lock_resource *res, char *buf, int len) +{ + struct dlm_lock *lock; + int i; + int out = 0; + + out += snprintf(buf + out, len - out, "NAME:"); + out += stringify_lockname(res->lockname.name, res->lockname.len, + buf + out, len - out); + out += snprintf(buf + out, len - out, "\n"); + +#define DEBUG_LRES_VERSION 1 + out += snprintf(buf + out, len - out, + "LRES:%d,%d,%d,%ld,%d,%d,%d,%d,%d,%d,%d\n", + DEBUG_LRES_VERSION, + res->owner, res->state, res->last_used, + !list_empty(&res->purge), + !list_empty(&res->dirty), + !list_empty(&res->recovering), + res->inflight_locks, res->migration_pending, + atomic_read(&res->asts_reserved), + atomic_read(&res->refs.refcount)); + + /* refmap */ + out += snprintf(buf + out, len - out, "RMAP:"); + out += stringify_nodemap(res->refmap, O2NM_MAX_NODES, + buf + out, len - out); + out += snprintf(buf + out, len - out, "\n"); + + /* lvb */ + out += snprintf(buf + out, len - out, "LVBX:"); + for (i = 0; i < DLM_LVB_LEN; i++) + out += snprintf(buf + out, len - out, + "%02x", (unsigned char)res->lvb[i]); + out += snprintf(buf + out, len - out, "\n"); + + /* granted */ + list_for_each_entry(lock, &res->granted, list) + out += dump_lock(lock, 0, buf + out, len - out); + + /* converting */ + list_for_each_entry(lock, &res->converting, list) + out += dump_lock(lock, 1, buf + out, len - out); + + /* blocked */ + list_for_each_entry(lock, &res->blocked, list) + out += dump_lock(lock, 2, buf + out, len - out); + + out += snprintf(buf + out, len - out, "\n"); + + return out; +} + +static void *lockres_seq_start(struct seq_file *m, loff_t *pos) +{ + struct debug_lockres *dl = m->private; + struct dlm_ctxt *dlm = dl->dl_ctxt; + struct dlm_lock_resource *oldres = dl->dl_res; + struct dlm_lock_resource *res = NULL; + struct list_head *track_list; + + spin_lock(&dlm->track_lock); + if (oldres) + track_list = &oldres->tracking; + else { + track_list = &dlm->tracking_list; + if (list_empty(track_list)) { + dl = NULL; + spin_unlock(&dlm->track_lock); + goto bail; + } + } + + list_for_each_entry(res, track_list, tracking) { + if (&res->tracking == &dlm->tracking_list) + res = NULL; + else + dlm_lockres_get(res); + break; + } + spin_unlock(&dlm->track_lock); + + if (oldres) + dlm_lockres_put(oldres); + + dl->dl_res = res; + + if (res) { + spin_lock(&res->spinlock); + dump_lockres(res, dl->dl_buf, dl->dl_len - 1); + spin_unlock(&res->spinlock); + } else + dl = NULL; + +bail: + /* passed to seq_show */ + return dl; +} + +static void lockres_seq_stop(struct seq_file *m, void *v) +{ +} + +static void *lockres_seq_next(struct seq_file *m, void *v, loff_t *pos) +{ + return NULL; +} + +static int lockres_seq_show(struct seq_file *s, void *v) +{ + struct debug_lockres *dl = (struct debug_lockres *)v; + + seq_printf(s, "%s", dl->dl_buf); + + return 0; +} + +static const struct seq_operations debug_lockres_ops = { + .start = lockres_seq_start, + .stop = lockres_seq_stop, + .next = lockres_seq_next, + .show = lockres_seq_show, +}; + +static int debug_lockres_open(struct inode *inode, struct file *file) +{ + struct dlm_ctxt *dlm = inode->i_private; + int ret = -ENOMEM; + struct seq_file *seq; + struct debug_lockres *dl = NULL; + + dl = kzalloc(sizeof(struct debug_lockres), GFP_KERNEL); + if (!dl) { + mlog_errno(ret); + goto bail; + } + + dl->dl_len = PAGE_SIZE; + dl->dl_buf = kmalloc(dl->dl_len, GFP_KERNEL); + if (!dl->dl_buf) { + mlog_errno(ret); + goto bail; + } + + ret = seq_open(file, &debug_lockres_ops); + if (ret) { + mlog_errno(ret); + goto bail; + } + + seq = file->private_data; + seq->private = dl; + + dlm_grab(dlm); + dl->dl_ctxt = dlm; + + return 0; +bail: + if (dl) + kfree(dl->dl_buf); + kfree(dl); + return ret; +} + +static int debug_lockres_release(struct inode *inode, struct file *file) +{ + struct seq_file *seq = file->private_data; + struct debug_lockres *dl = (struct debug_lockres *)seq->private; + + if (dl->dl_res) + dlm_lockres_put(dl->dl_res); + dlm_put(dl->dl_ctxt); + kfree(dl->dl_buf); + return seq_release_private(inode, file); +} + +static const struct file_operations debug_lockres_fops = { + .open = debug_lockres_open, + .release = debug_lockres_release, + .read = seq_read, + .llseek = seq_lseek, +}; +/* end - debug lockres funcs */ + +/* begin - debug state funcs */ +static int debug_state_print(struct dlm_ctxt *dlm, char *buf, int len) +{ + int out = 0; + struct dlm_reco_node_data *node; + char *state; + int cur_mles = 0, tot_mles = 0; + int i; + + spin_lock(&dlm->spinlock); + + switch (dlm->dlm_state) { + case DLM_CTXT_NEW: + state = "NEW"; break; + case DLM_CTXT_JOINED: + state = "JOINED"; break; + case DLM_CTXT_IN_SHUTDOWN: + state = "SHUTDOWN"; break; + case DLM_CTXT_LEAVING: + state = "LEAVING"; break; + default: + state = "UNKNOWN"; break; + } + + /* Domain: xxxxxxxxxx Key: 0xdfbac769 */ + out += snprintf(buf + out, len - out, + "Domain: %s Key: 0x%08x Protocol: %d.%d\n", + dlm->name, dlm->key, dlm->dlm_locking_proto.pv_major, + dlm->dlm_locking_proto.pv_minor); + + /* Thread Pid: xxx Node: xxx State: xxxxx */ + out += snprintf(buf + out, len - out, + "Thread Pid: %d Node: %d State: %s\n", + task_pid_nr(dlm->dlm_thread_task), dlm->node_num, state); + + /* Number of Joins: xxx Joining Node: xxx */ + out += snprintf(buf + out, len - out, + "Number of Joins: %d Joining Node: %d\n", + dlm->num_joins, dlm->joining_node); + + /* Domain Map: xx xx xx */ + out += snprintf(buf + out, len - out, "Domain Map: "); + out += stringify_nodemap(dlm->domain_map, O2NM_MAX_NODES, + buf + out, len - out); + out += snprintf(buf + out, len - out, "\n"); + + /* Exit Domain Map: xx xx xx */ + out += snprintf(buf + out, len - out, "Exit Domain Map: "); + out += stringify_nodemap(dlm->exit_domain_map, O2NM_MAX_NODES, + buf + out, len - out); + out += snprintf(buf + out, len - out, "\n"); + + /* Live Map: xx xx xx */ + out += snprintf(buf + out, len - out, "Live Map: "); + out += stringify_nodemap(dlm->live_nodes_map, O2NM_MAX_NODES, + buf + out, len - out); + out += snprintf(buf + out, len - out, "\n"); + + /* Lock Resources: xxx (xxx) */ + out += snprintf(buf + out, len - out, + "Lock Resources: %d (%d)\n", + atomic_read(&dlm->res_cur_count), + atomic_read(&dlm->res_tot_count)); + + for (i = 0; i < DLM_MLE_NUM_TYPES; ++i) + tot_mles += atomic_read(&dlm->mle_tot_count[i]); + + for (i = 0; i < DLM_MLE_NUM_TYPES; ++i) + cur_mles += atomic_read(&dlm->mle_cur_count[i]); + + /* MLEs: xxx (xxx) */ + out += snprintf(buf + out, len - out, + "MLEs: %d (%d)\n", cur_mles, tot_mles); + + /* Blocking: xxx (xxx) */ + out += snprintf(buf + out, len - out, + " Blocking: %d (%d)\n", + atomic_read(&dlm->mle_cur_count[DLM_MLE_BLOCK]), + atomic_read(&dlm->mle_tot_count[DLM_MLE_BLOCK])); + + /* Mastery: xxx (xxx) */ + out += snprintf(buf + out, len - out, + " Mastery: %d (%d)\n", + atomic_read(&dlm->mle_cur_count[DLM_MLE_MASTER]), + atomic_read(&dlm->mle_tot_count[DLM_MLE_MASTER])); + + /* Migration: xxx (xxx) */ + out += snprintf(buf + out, len - out, + " Migration: %d (%d)\n", + atomic_read(&dlm->mle_cur_count[DLM_MLE_MIGRATION]), + atomic_read(&dlm->mle_tot_count[DLM_MLE_MIGRATION])); + + /* Lists: Dirty=Empty Purge=InUse PendingASTs=Empty ... */ + out += snprintf(buf + out, len - out, + "Lists: Dirty=%s Purge=%s PendingASTs=%s " + "PendingBASTs=%s\n", + (list_empty(&dlm->dirty_list) ? "Empty" : "InUse"), + (list_empty(&dlm->purge_list) ? "Empty" : "InUse"), + (list_empty(&dlm->pending_asts) ? "Empty" : "InUse"), + (list_empty(&dlm->pending_basts) ? "Empty" : "InUse")); + + /* Purge Count: xxx Refs: xxx */ + out += snprintf(buf + out, len - out, + "Purge Count: %d Refs: %d\n", dlm->purge_count, + atomic_read(&dlm->dlm_refs.refcount)); + + /* Dead Node: xxx */ + out += snprintf(buf + out, len - out, + "Dead Node: %d\n", dlm->reco.dead_node); + + /* What about DLM_RECO_STATE_FINALIZE? */ + if (dlm->reco.state == DLM_RECO_STATE_ACTIVE) + state = "ACTIVE"; + else + state = "INACTIVE"; + + /* Recovery Pid: xxxx Master: xxx State: xxxx */ + out += snprintf(buf + out, len - out, + "Recovery Pid: %d Master: %d State: %s\n", + task_pid_nr(dlm->dlm_reco_thread_task), + dlm->reco.new_master, state); + + /* Recovery Map: xx xx */ + out += snprintf(buf + out, len - out, "Recovery Map: "); + out += stringify_nodemap(dlm->recovery_map, O2NM_MAX_NODES, + buf + out, len - out); + out += snprintf(buf + out, len - out, "\n"); + + /* Recovery Node State: */ + out += snprintf(buf + out, len - out, "Recovery Node State:\n"); + list_for_each_entry(node, &dlm->reco.node_data, list) { + switch (node->state) { + case DLM_RECO_NODE_DATA_INIT: + state = "INIT"; + break; + case DLM_RECO_NODE_DATA_REQUESTING: + state = "REQUESTING"; + break; + case DLM_RECO_NODE_DATA_DEAD: + state = "DEAD"; + break; + case DLM_RECO_NODE_DATA_RECEIVING: + state = "RECEIVING"; + break; + case DLM_RECO_NODE_DATA_REQUESTED: + state = "REQUESTED"; + break; + case DLM_RECO_NODE_DATA_DONE: + state = "DONE"; + break; + case DLM_RECO_NODE_DATA_FINALIZE_SENT: + state = "FINALIZE-SENT"; + break; + default: + state = "BAD"; + break; + } + out += snprintf(buf + out, len - out, "\t%u - %s\n", + node->node_num, state); + } + + spin_unlock(&dlm->spinlock); + + return out; +} + +static int debug_state_open(struct inode *inode, struct file *file) +{ + struct dlm_ctxt *dlm = inode->i_private; + char *buf = NULL; + + buf = (char *) get_zeroed_page(GFP_NOFS); + if (!buf) + goto bail; + + i_size_write(inode, debug_state_print(dlm, buf, PAGE_SIZE - 1)); + + file->private_data = buf; + + return 0; +bail: + return -ENOMEM; +} + +static const struct file_operations debug_state_fops = { + .open = debug_state_open, + .release = debug_release, + .read = debug_read, + .llseek = generic_file_llseek, +}; +/* end - debug state funcs */ + +/* files in subroot */ +int dlm_debug_init(struct dlm_ctxt *dlm) +{ + struct dlm_debug_ctxt *dc = dlm->dlm_debug_ctxt; + + /* for dumping dlm_ctxt */ + dc->debug_state_dentry = debugfs_create_file(DLM_DEBUGFS_DLM_STATE, + S_IFREG|S_IRUSR, + dlm->dlm_debugfs_subroot, + dlm, &debug_state_fops); + if (!dc->debug_state_dentry) { + mlog_errno(-ENOMEM); + goto bail; + } + + /* for dumping lockres */ + dc->debug_lockres_dentry = + debugfs_create_file(DLM_DEBUGFS_LOCKING_STATE, + S_IFREG|S_IRUSR, + dlm->dlm_debugfs_subroot, + dlm, &debug_lockres_fops); + if (!dc->debug_lockres_dentry) { + mlog_errno(-ENOMEM); + goto bail; + } + + /* for dumping mles */ + dc->debug_mle_dentry = debugfs_create_file(DLM_DEBUGFS_MLE_STATE, + S_IFREG|S_IRUSR, + dlm->dlm_debugfs_subroot, + dlm, &debug_mle_fops); + if (!dc->debug_mle_dentry) { + mlog_errno(-ENOMEM); + goto bail; + } + + /* for dumping lockres on the purge list */ + dc->debug_purgelist_dentry = + debugfs_create_file(DLM_DEBUGFS_PURGE_LIST, + S_IFREG|S_IRUSR, + dlm->dlm_debugfs_subroot, + dlm, &debug_purgelist_fops); + if (!dc->debug_purgelist_dentry) { + mlog_errno(-ENOMEM); + goto bail; + } + + dlm_debug_get(dc); + return 0; + +bail: + dlm_debug_shutdown(dlm); + return -ENOMEM; +} + +void dlm_debug_shutdown(struct dlm_ctxt *dlm) +{ + struct dlm_debug_ctxt *dc = dlm->dlm_debug_ctxt; + + if (dc) { + debugfs_remove(dc->debug_purgelist_dentry); + debugfs_remove(dc->debug_mle_dentry); + debugfs_remove(dc->debug_lockres_dentry); + debugfs_remove(dc->debug_state_dentry); + dlm_debug_put(dc); + } +} + +/* subroot - domain dir */ +int dlm_create_debugfs_subroot(struct dlm_ctxt *dlm) +{ + dlm->dlm_debugfs_subroot = debugfs_create_dir(dlm->name, + dlm_debugfs_root); + if (!dlm->dlm_debugfs_subroot) { + mlog_errno(-ENOMEM); + goto bail; + } + + dlm->dlm_debug_ctxt = kzalloc(sizeof(struct dlm_debug_ctxt), + GFP_KERNEL); + if (!dlm->dlm_debug_ctxt) { + mlog_errno(-ENOMEM); + goto bail; + } + kref_init(&dlm->dlm_debug_ctxt->debug_refcnt); + + return 0; +bail: + dlm_destroy_debugfs_subroot(dlm); + return -ENOMEM; +} + +void dlm_destroy_debugfs_subroot(struct dlm_ctxt *dlm) +{ + debugfs_remove(dlm->dlm_debugfs_subroot); +} + +/* debugfs root */ +int dlm_create_debugfs_root(void) +{ + dlm_debugfs_root = debugfs_create_dir(DLM_DEBUGFS_DIR, NULL); + if (!dlm_debugfs_root) { + mlog_errno(-ENOMEM); + return -ENOMEM; + } + return 0; +} + +void dlm_destroy_debugfs_root(void) +{ + debugfs_remove(dlm_debugfs_root); +} +#endif /* CONFIG_DEBUG_FS */ diff --git a/fs/ocfs2/dlm/dlmdebug.h b/fs/ocfs2/dlm/dlmdebug.h index 6858510c3cc..1f27c4812d1 100644 --- a/fs/ocfs2/dlm/dlmdebug.h +++ b/fs/ocfs2/dlm/dlmdebug.h @@ -3,7 +3,7 @@ * * dlmdebug.h * - * Copyright (C) 2004 Oracle. All rights reserved. + * Copyright (C) 2008 Oracle. All rights reserved. * * This program is free software; you can redistribute it and/or * modify it under the terms of the GNU General Public @@ -25,6 +25,57 @@ #ifndef DLMDEBUG_H #define DLMDEBUG_H -void dlm_dump_lock_resources(struct dlm_ctxt *dlm); +void dlm_print_one_mle(struct dlm_master_list_entry *mle); -#endif +#ifdef CONFIG_DEBUG_FS + +struct dlm_debug_ctxt { + struct kref debug_refcnt; + struct dentry *debug_state_dentry; + struct dentry *debug_lockres_dentry; + struct dentry *debug_mle_dentry; + struct dentry *debug_purgelist_dentry; +}; + +struct debug_lockres { + int dl_len; + char *dl_buf; + struct dlm_ctxt *dl_ctxt; + struct dlm_lock_resource *dl_res; +}; + +int dlm_debug_init(struct dlm_ctxt *dlm); +void dlm_debug_shutdown(struct dlm_ctxt *dlm); + +int dlm_create_debugfs_subroot(struct dlm_ctxt *dlm); +void dlm_destroy_debugfs_subroot(struct dlm_ctxt *dlm); + +int dlm_create_debugfs_root(void); +void dlm_destroy_debugfs_root(void); + +#else + +static inline int dlm_debug_init(struct dlm_ctxt *dlm) +{ + return 0; +} +static inline void dlm_debug_shutdown(struct dlm_ctxt *dlm) +{ +} +static inline int dlm_create_debugfs_subroot(struct dlm_ctxt *dlm) +{ + return 0; +} +static inline void dlm_destroy_debugfs_subroot(struct dlm_ctxt *dlm) +{ +} +static inline int dlm_create_debugfs_root(void) +{ + return 0; +} +static inline void dlm_destroy_debugfs_root(void) +{ +} + +#endif /* CONFIG_DEBUG_FS */ +#endif /* DLMDEBUG_H */ diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c index da3c22045f8..39efc5057a3 100644 --- a/fs/ocfs2/dlm/dlmdomain.c +++ b/fs/ocfs2/dlm/dlmdomain.c @@ -28,11 +28,11 @@ #include <linux/types.h> #include <linux/slab.h> #include <linux/highmem.h> -#include <linux/utsname.h> #include <linux/init.h> #include <linux/spinlock.h> #include <linux/delay.h> #include <linux/err.h> +#include <linux/debugfs.h> #include "cluster/heartbeat.h" #include "cluster/nodemanager.h" @@ -40,16 +40,71 @@ #include "dlmapi.h" #include "dlmcommon.h" - -#include "dlmdebug.h" #include "dlmdomain.h" - -#include "dlmver.h" +#include "dlmdebug.h" #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_DOMAIN) #include "cluster/masklog.h" /* + * ocfs2 node maps are array of long int, which limits to send them freely + * across the wire due to endianness issues. To workaround this, we convert + * long ints to byte arrays. Following 3 routines are helper functions to + * set/test/copy bits within those array of bytes + */ +static inline void byte_set_bit(u8 nr, u8 map[]) +{ + map[nr >> 3] |= (1UL << (nr & 7)); +} + +static inline int byte_test_bit(u8 nr, u8 map[]) +{ + return ((1UL << (nr & 7)) & (map[nr >> 3])) != 0; +} + +static inline void byte_copymap(u8 dmap[], unsigned long smap[], + unsigned int sz) +{ + unsigned int nn; + + if (!sz) + return; + + memset(dmap, 0, ((sz + 7) >> 3)); + for (nn = 0 ; nn < sz; nn++) + if (test_bit(nn, smap)) + byte_set_bit(nn, dmap); +} + +static void dlm_free_pagevec(void **vec, int pages) +{ + while (pages--) + free_page((unsigned long)vec[pages]); + kfree(vec); +} + +static void **dlm_alloc_pagevec(int pages) +{ + void **vec = kmalloc(pages * sizeof(void *), GFP_KERNEL); + int i; + + if (!vec) + return NULL; + + for (i = 0; i < pages; i++) + if (!(vec[i] = (void *)__get_free_page(GFP_KERNEL))) + goto out_free; + + mlog(0, "Allocated DLM hash pagevec; %d pages (%lu expected), %lu buckets per page\n", + pages, (unsigned long)DLM_HASH_PAGES, + (unsigned long)DLM_BUCKETS_PER_PAGE); + return vec; +out_free: + dlm_free_pagevec(vec, i); + return NULL; +} + +/* * * spinlock lock ordering: if multiple locks are needed, obey this ordering: * dlm_domain_lock @@ -62,72 +117,130 @@ * */ -spinlock_t dlm_domain_lock = SPIN_LOCK_UNLOCKED; +DEFINE_SPINLOCK(dlm_domain_lock); LIST_HEAD(dlm_domains); static DECLARE_WAIT_QUEUE_HEAD(dlm_domain_events); +/* + * The supported protocol version for DLM communication. Running domains + * will have a negotiated version with the same major number and a minor + * number equal or smaller. The dlm_ctxt->dlm_locking_proto field should + * be used to determine what a running domain is actually using. + * + * New in version 1.1: + * - Message DLM_QUERY_REGION added to support global heartbeat + * - Message DLM_QUERY_NODEINFO added to allow online node removes + * New in version 1.2: + * - Message DLM_BEGIN_EXIT_DOMAIN_MSG added to mark start of exit domain + */ +static const struct dlm_protocol_version dlm_protocol = { + .pv_major = 1, + .pv_minor = 2, +}; + #define DLM_DOMAIN_BACKOFF_MS 200 -static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data); -static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data); -static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data); -static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data); +static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); +static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); +static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); +static int dlm_query_region_handler(struct o2net_msg *msg, u32 len, + void *data, void **ret_data); +static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data); +static int dlm_protocol_compare(struct dlm_protocol_version *existing, + struct dlm_protocol_version *request); static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm); -void __dlm_unhash_lockres(struct dlm_lock_resource *lockres) +void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) { - list_del_init(&lockres->list); - dlm_lockres_put(lockres); + if (hlist_unhashed(&res->hash_node)) + return; + + mlog(0, "%s: Unhash res %.*s\n", dlm->name, res->lockname.len, + res->lockname.name); + hlist_del_init(&res->hash_node); + dlm_lockres_put(res); } -void __dlm_insert_lockres(struct dlm_ctxt *dlm, - struct dlm_lock_resource *res) +void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) { - struct list_head *bucket; + struct hlist_head *bucket; struct qstr *q; assert_spin_locked(&dlm->spinlock); q = &res->lockname; - q->hash = full_name_hash(q->name, q->len); - bucket = &(dlm->resources[q->hash & DLM_HASH_MASK]); + bucket = dlm_lockres_hash(dlm, q->hash); /* get a reference for our hashtable */ dlm_lockres_get(res); - list_add_tail(&res->list, bucket); + hlist_add_head(&res->hash_node, bucket); + + mlog(0, "%s: Hash res %.*s\n", dlm->name, res->lockname.len, + res->lockname.name); } -struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm, - const char *name, - unsigned int len) +struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm, + const char *name, + unsigned int len, + unsigned int hash) { - unsigned int hash; - struct list_head *iter; - struct dlm_lock_resource *tmpres=NULL; - struct list_head *bucket; + struct hlist_head *bucket; + struct dlm_lock_resource *res; - mlog_entry("%.*s\n", len, name); + mlog(0, "%.*s\n", len, name); assert_spin_locked(&dlm->spinlock); - hash = full_name_hash(name, len); + bucket = dlm_lockres_hash(dlm, hash); + + hlist_for_each_entry(res, bucket, hash_node) { + if (res->lockname.name[0] != name[0]) + continue; + if (unlikely(res->lockname.len != len)) + continue; + if (memcmp(res->lockname.name + 1, name + 1, len - 1)) + continue; + dlm_lockres_get(res); + return res; + } + return NULL; +} + +/* intended to be called by functions which do not care about lock + * resources which are being purged (most net _handler functions). + * this will return NULL for any lock resource which is found but + * currently in the process of dropping its mastery reference. + * use __dlm_lookup_lockres_full when you need the lock resource + * regardless (e.g. dlm_get_lock_resource) */ +struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm, + const char *name, + unsigned int len, + unsigned int hash) +{ + struct dlm_lock_resource *res = NULL; - bucket = &(dlm->resources[hash & DLM_HASH_MASK]); + mlog(0, "%.*s\n", len, name); - /* check for pre-existing lock */ - list_for_each(iter, bucket) { - tmpres = list_entry(iter, struct dlm_lock_resource, list); - if (tmpres->lockname.len == len && - memcmp(tmpres->lockname.name, name, len) == 0) { - dlm_lockres_get(tmpres); - break; - } + assert_spin_locked(&dlm->spinlock); - tmpres = NULL; + res = __dlm_lookup_lockres_full(dlm, name, len, hash); + if (res) { + spin_lock(&res->spinlock); + if (res->state & DLM_LOCK_RES_DROPPING_REF) { + spin_unlock(&res->spinlock); + dlm_lockres_put(res); + return NULL; + } + spin_unlock(&res->spinlock); } - return tmpres; + + return res; } struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm, @@ -135,31 +248,29 @@ struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm, unsigned int len) { struct dlm_lock_resource *res; + unsigned int hash = dlm_lockid_hash(name, len); spin_lock(&dlm->spinlock); - res = __dlm_lookup_lockres(dlm, name, len); + res = __dlm_lookup_lockres(dlm, name, len, hash); spin_unlock(&dlm->spinlock); return res; } static struct dlm_ctxt * __dlm_lookup_domain_full(const char *domain, int len) { - struct dlm_ctxt *tmp = NULL; - struct list_head *iter; + struct dlm_ctxt *tmp; assert_spin_locked(&dlm_domain_lock); /* tmp->name here is always NULL terminated, * but domain may not be! */ - list_for_each(iter, &dlm_domains) { - tmp = list_entry (iter, struct dlm_ctxt, list); + list_for_each_entry(tmp, &dlm_domains, list) { if (strlen(tmp->name) == len && memcmp(tmp->name, domain, len)==0) - break; - tmp = NULL; + return tmp; } - return tmp; + return NULL; } /* For null terminated domain strings ONLY */ @@ -193,12 +304,15 @@ static int dlm_wait_on_domain_helper(const char *domain) static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm) { - if (dlm->resources) - free_page((unsigned long) dlm->resources); + dlm_destroy_debugfs_subroot(dlm); - if (dlm->name) - kfree(dlm->name); + if (dlm->lockres_hash) + dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES); + + if (dlm->master_hash) + dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES); + kfree(dlm->name); kfree(dlm); } @@ -245,25 +359,22 @@ static void __dlm_get(struct dlm_ctxt *dlm) * you shouldn't trust your pointer. */ struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm) { - struct list_head *iter; - struct dlm_ctxt *target = NULL; + struct dlm_ctxt *target; + struct dlm_ctxt *ret = NULL; spin_lock(&dlm_domain_lock); - list_for_each(iter, &dlm_domains) { - target = list_entry (iter, struct dlm_ctxt, list); - + list_for_each_entry(target, &dlm_domains, list) { if (target == dlm) { __dlm_get(target); + ret = target; break; } - - target = NULL; } spin_unlock(&dlm_domain_lock); - return target; + return ret; } int dlm_domain_fully_joined(struct dlm_ctxt *dlm) @@ -278,11 +389,22 @@ int dlm_domain_fully_joined(struct dlm_ctxt *dlm) return ret; } +static void dlm_destroy_dlm_worker(struct dlm_ctxt *dlm) +{ + if (dlm->dlm_worker) { + flush_workqueue(dlm->dlm_worker); + destroy_workqueue(dlm->dlm_worker); + dlm->dlm_worker = NULL; + } +} + static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm) { dlm_unregister_domain_handlers(dlm); + dlm_debug_shutdown(dlm); dlm_complete_thread(dlm); dlm_complete_recovery_thread(dlm); + dlm_destroy_dlm_worker(dlm); /* We've left the domain. Now we can take ourselves out of the * list and allow the kref stuff to help us free the @@ -295,43 +417,61 @@ static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm) wake_up(&dlm_domain_events); } -static void dlm_migrate_all_locks(struct dlm_ctxt *dlm) +static int dlm_migrate_all_locks(struct dlm_ctxt *dlm) { - int i; + int i, num, n, ret = 0; struct dlm_lock_resource *res; + struct hlist_node *iter; + struct hlist_head *bucket; + int dropped; mlog(0, "Migrating locks from domain %s\n", dlm->name); -restart: + + num = 0; spin_lock(&dlm->spinlock); - for (i=0; i<DLM_HASH_SIZE; i++) { - while (!list_empty(&dlm->resources[i])) { - res = list_entry(dlm->resources[i].next, - struct dlm_lock_resource, list); - /* need reference when manually grabbing lockres */ + for (i = 0; i < DLM_HASH_BUCKETS; i++) { +redo_bucket: + n = 0; + bucket = dlm_lockres_hash(dlm, i); + iter = bucket->first; + while (iter) { + n++; + res = hlist_entry(iter, struct dlm_lock_resource, + hash_node); dlm_lockres_get(res); - /* this should unhash the lockres - * and exit with dlm->spinlock */ - mlog(0, "purging res=%p\n", res); - if (dlm_lockres_is_dirty(dlm, res)) { - /* HACK! this should absolutely go. - * need to figure out why some empty - * lockreses are still marked dirty */ - mlog(ML_ERROR, "lockres %.*s dirty!\n", - res->lockname.len, res->lockname.name); - - spin_unlock(&dlm->spinlock); - dlm_kick_thread(dlm, res); - wait_event(dlm->ast_wq, !dlm_lockres_is_dirty(dlm, res)); - dlm_lockres_put(res); - goto restart; - } - dlm_purge_lockres(dlm, res); + /* migrate, if necessary. this will drop the dlm + * spinlock and retake it if it does migration. */ + dropped = dlm_empty_lockres(dlm, res); + + spin_lock(&res->spinlock); + if (dropped) + __dlm_lockres_calc_usage(dlm, res); + else + iter = res->hash_node.next; + spin_unlock(&res->spinlock); + dlm_lockres_put(res); + + if (dropped) { + cond_resched_lock(&dlm->spinlock); + goto redo_bucket; + } } + cond_resched_lock(&dlm->spinlock); + num += n; } spin_unlock(&dlm->spinlock); - + wake_up(&dlm->dlm_thread_wq); + + /* let the dlm thread take care of purging, keep scanning until + * nothing remains in the hash */ + if (num) { + mlog(0, "%s: %d lock resources in hash last pass\n", + dlm->name, num); + ret = -EAGAIN; + } mlog(0, "DONE Migrating locks from domain %s\n", dlm->name); + return ret; } static int dlm_no_joining_node(struct dlm_ctxt *dlm) @@ -345,6 +485,28 @@ static int dlm_no_joining_node(struct dlm_ctxt *dlm) return ret; } +static int dlm_begin_exit_domain_handler(struct o2net_msg *msg, u32 len, + void *data, void **ret_data) +{ + struct dlm_ctxt *dlm = data; + unsigned int node; + struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf; + + if (!dlm_grab(dlm)) + return 0; + + node = exit_msg->node_idx; + mlog(0, "%s: Node %u sent a begin exit domain message\n", dlm->name, node); + + spin_lock(&dlm->spinlock); + set_bit(node, dlm->exit_domain_map); + spin_unlock(&dlm->spinlock); + + dlm_put(dlm); + + return 0; +} + static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm) { /* Yikes, a double spinlock! I need domain_lock for the dlm @@ -370,35 +532,37 @@ again: static void __dlm_print_nodes(struct dlm_ctxt *dlm) { - int node = -1; + int node = -1, num = 0; assert_spin_locked(&dlm->spinlock); - mlog(ML_NOTICE, "Nodes in my domain (\"%s\"):\n", dlm->name); - + printk("( "); while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES, node + 1)) < O2NM_MAX_NODES) { - mlog(ML_NOTICE, " node %d\n", node); + printk("%d ", node); + ++num; } + printk(") %u nodes\n", num); } -static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data) +static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { struct dlm_ctxt *dlm = data; unsigned int node; struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf; - mlog_entry("%p %u %p", msg, len, data); + mlog(0, "%p %u %p", msg, len, data); if (!dlm_grab(dlm)) return 0; node = exit_msg->node_idx; - mlog(0, "Node %u leaves domain %s\n", node, dlm->name); - spin_lock(&dlm->spinlock); clear_bit(node, dlm->domain_map); + clear_bit(node, dlm->exit_domain_map); + printk(KERN_NOTICE "o2dlm: Node %u leaves domain %s ", node, dlm->name); __dlm_print_nodes(dlm); /* notify anything attached to the heartbeat events */ @@ -411,27 +575,56 @@ static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data) return 0; } -static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm, +static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm, u32 msg_type, unsigned int node) { int status; struct dlm_exit_domain leave_msg; - mlog(0, "Asking node %u if we can leave the domain %s me = %u\n", - node, dlm->name, dlm->node_num); + mlog(0, "%s: Sending domain exit message %u to node %u\n", dlm->name, + msg_type, node); memset(&leave_msg, 0, sizeof(leave_msg)); leave_msg.node_idx = dlm->node_num; - status = o2net_send_message(DLM_EXIT_DOMAIN_MSG, dlm->key, - &leave_msg, sizeof(leave_msg), node, - NULL); - - mlog(0, "status return %d from o2net_send_message\n", status); + status = o2net_send_message(msg_type, dlm->key, &leave_msg, + sizeof(leave_msg), node, NULL); + if (status < 0) + mlog(ML_ERROR, "Error %d sending domain exit message %u " + "to node %u on domain %s\n", status, msg_type, node, + dlm->name); return status; } +static void dlm_begin_exit_domain(struct dlm_ctxt *dlm) +{ + int node = -1; + + /* Support for begin exit domain was added in 1.2 */ + if (dlm->dlm_locking_proto.pv_major == 1 && + dlm->dlm_locking_proto.pv_minor < 2) + return; + + /* + * Unlike DLM_EXIT_DOMAIN_MSG, DLM_BEGIN_EXIT_DOMAIN_MSG is purely + * informational. Meaning if a node does not receive the message, + * so be it. + */ + spin_lock(&dlm->spinlock); + while (1) { + node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES, node + 1); + if (node >= O2NM_MAX_NODES) + break; + if (node == dlm->node_num) + continue; + + spin_unlock(&dlm->spinlock); + dlm_send_one_domain_exit(dlm, DLM_BEGIN_EXIT_DOMAIN_MSG, node); + spin_lock(&dlm->spinlock); + } + spin_unlock(&dlm->spinlock); +} static void dlm_leave_domain(struct dlm_ctxt *dlm) { @@ -457,7 +650,8 @@ static void dlm_leave_domain(struct dlm_ctxt *dlm) clear_node = 1; - status = dlm_send_one_domain_exit(dlm, node); + status = dlm_send_one_domain_exit(dlm, DLM_EXIT_DOMAIN_MSG, + node); if (status < 0 && status != -ENOPROTOOPT && status != -ENOTCONN) { @@ -511,6 +705,7 @@ int dlm_shutting_down(struct dlm_ctxt *dlm) void dlm_unregister_domain(struct dlm_ctxt *dlm) { int leave = 0; + struct dlm_lock_resource *res; spin_lock(&dlm_domain_lock); BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED); @@ -531,24 +726,108 @@ void dlm_unregister_domain(struct dlm_ctxt *dlm) if (leave) { mlog(0, "shutting down domain %s\n", dlm->name); + dlm_begin_exit_domain(dlm); /* We changed dlm state, notify the thread */ dlm_kick_thread(dlm, NULL); - dlm_migrate_all_locks(dlm); + while (dlm_migrate_all_locks(dlm)) { + /* Give dlm_thread time to purge the lockres' */ + msleep(500); + mlog(0, "%s: more migration to do\n", dlm->name); + } + + /* This list should be empty. If not, print remaining lockres */ + if (!list_empty(&dlm->tracking_list)) { + mlog(ML_ERROR, "Following lockres' are still on the " + "tracking list:\n"); + list_for_each_entry(res, &dlm->tracking_list, tracking) + dlm_print_one_lock_resource(res); + } + dlm_mark_domain_leaving(dlm); dlm_leave_domain(dlm); + printk(KERN_NOTICE "o2dlm: Leaving domain %s\n", dlm->name); + dlm_force_free_mles(dlm); dlm_complete_dlm_shutdown(dlm); } dlm_put(dlm); } EXPORT_SYMBOL_GPL(dlm_unregister_domain); -static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data) +static int dlm_query_join_proto_check(char *proto_type, int node, + struct dlm_protocol_version *ours, + struct dlm_protocol_version *request) +{ + int rc; + struct dlm_protocol_version proto = *request; + + if (!dlm_protocol_compare(ours, &proto)) { + mlog(0, + "node %u wanted to join with %s locking protocol " + "%u.%u, we respond with %u.%u\n", + node, proto_type, + request->pv_major, + request->pv_minor, + proto.pv_major, proto.pv_minor); + request->pv_minor = proto.pv_minor; + rc = 0; + } else { + mlog(ML_NOTICE, + "Node %u wanted to join with %s locking " + "protocol %u.%u, but we have %u.%u, disallowing\n", + node, proto_type, + request->pv_major, + request->pv_minor, + ours->pv_major, + ours->pv_minor); + rc = 1; + } + + return rc; +} + +/* + * struct dlm_query_join_packet is made up of four one-byte fields. They + * are effectively in big-endian order already. However, little-endian + * machines swap them before putting the packet on the wire (because + * query_join's response is a status, and that status is treated as a u32 + * on the wire). Thus, a big-endian and little-endian machines will treat + * this structure differently. + * + * The solution is to have little-endian machines swap the structure when + * converting from the structure to the u32 representation. This will + * result in the structure having the correct format on the wire no matter + * the host endian format. + */ +static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet, + u32 *wire) +{ + union dlm_query_join_response response; + + response.packet = *packet; + *wire = be32_to_cpu(response.intval); +} + +static void dlm_query_join_wire_to_packet(u32 wire, + struct dlm_query_join_packet *packet) +{ + union dlm_query_join_response response; + + response.intval = cpu_to_be32(wire); + *packet = response.packet; +} + +static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { struct dlm_query_join_request *query; - enum dlm_query_join_response response; + struct dlm_query_join_packet packet = { + .code = JOIN_DISALLOW, + }; struct dlm_ctxt *dlm = NULL; + u32 response; + u8 nodenum; query = (struct dlm_query_join_request *) msg->buf; @@ -564,17 +843,42 @@ static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data) mlog(0, "node %u is not in our live map yet\n", query->node_idx); - response = JOIN_DISALLOW; + packet.code = JOIN_DISALLOW; goto respond; } - response = JOIN_OK_NO_MAP; + packet.code = JOIN_OK_NO_MAP; spin_lock(&dlm_domain_lock); dlm = __dlm_lookup_domain_full(query->domain, query->name_len); + if (!dlm) + goto unlock_respond; + + /* + * There is a small window where the joining node may not see the + * node(s) that just left but still part of the cluster. DISALLOW + * join request if joining node has different node map. + */ + nodenum=0; + while (nodenum < O2NM_MAX_NODES) { + if (test_bit(nodenum, dlm->domain_map)) { + if (!byte_test_bit(nodenum, query->node_map)) { + mlog(0, "disallow join as node %u does not " + "have node %u in its nodemap\n", + query->node_idx, nodenum); + packet.code = JOIN_DISALLOW; + goto unlock_respond; + } + } + nodenum++; + } + /* Once the dlm ctxt is marked as leaving then we don't want - * to be put in someone's domain map. */ + * to be put in someone's domain map. + * Also, explicitly disallow joining at certain troublesome + * times (ie. during recovery). */ if (dlm && dlm->dlm_state != DLM_CTXT_LEAVING) { + int bit = query->node_idx; spin_lock(&dlm->spinlock); if (dlm->dlm_state == DLM_CTXT_NEW && @@ -582,30 +886,60 @@ static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data) /*If this is a brand new context and we * haven't started our join process yet, then * the other node won the race. */ - response = JOIN_OK_NO_MAP; + packet.code = JOIN_OK_NO_MAP; } else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) { /* Disallow parallel joins. */ - response = JOIN_DISALLOW; + packet.code = JOIN_DISALLOW; + } else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) { + mlog(0, "node %u trying to join, but recovery " + "is ongoing.\n", bit); + packet.code = JOIN_DISALLOW; + } else if (test_bit(bit, dlm->recovery_map)) { + mlog(0, "node %u trying to join, but it " + "still needs recovery.\n", bit); + packet.code = JOIN_DISALLOW; + } else if (test_bit(bit, dlm->domain_map)) { + mlog(0, "node %u trying to join, but it " + "is still in the domain! needs recovery?\n", + bit); + packet.code = JOIN_DISALLOW; } else { /* Alright we're fully a part of this domain * so we keep some state as to who's joining * and indicate to him that needs to be fixed * up. */ - response = JOIN_OK; - __dlm_set_joining_node(dlm, query->node_idx); + + /* Make sure we speak compatible locking protocols. */ + if (dlm_query_join_proto_check("DLM", bit, + &dlm->dlm_locking_proto, + &query->dlm_proto)) { + packet.code = JOIN_PROTOCOL_MISMATCH; + } else if (dlm_query_join_proto_check("fs", bit, + &dlm->fs_locking_proto, + &query->fs_proto)) { + packet.code = JOIN_PROTOCOL_MISMATCH; + } else { + packet.dlm_minor = query->dlm_proto.pv_minor; + packet.fs_minor = query->fs_proto.pv_minor; + packet.code = JOIN_OK; + __dlm_set_joining_node(dlm, query->node_idx); + } } spin_unlock(&dlm->spinlock); } +unlock_respond: spin_unlock(&dlm_domain_lock); respond: - mlog(0, "We respond with %u\n", response); + mlog(0, "We respond with %u\n", packet.code); + dlm_query_join_packet_to_wire(&packet, &response); return response; } -static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data) +static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { struct dlm_assert_joined *assert; struct dlm_ctxt *dlm = NULL; @@ -625,9 +959,20 @@ static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data) * domain. Set him in the map and clean up our * leftover join state. */ BUG_ON(dlm->joining_node != assert->node_idx); + + if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) { + mlog(0, "dlm recovery is ongoing, disallow join\n"); + spin_unlock(&dlm->spinlock); + spin_unlock(&dlm_domain_lock); + return -EAGAIN; + } + set_bit(assert->node_idx, dlm->domain_map); + clear_bit(assert->node_idx, dlm->exit_domain_map); __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN); + printk(KERN_NOTICE "o2dlm: Node %u joins domain %s ", + assert->node_idx, dlm->name); __dlm_print_nodes(dlm); /* notify anything attached to the heartbeat events */ @@ -640,7 +985,373 @@ static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data) return 0; } -static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data) +static int dlm_match_regions(struct dlm_ctxt *dlm, + struct dlm_query_region *qr, + char *local, int locallen) +{ + char *remote = qr->qr_regions; + char *l, *r; + int localnr, i, j, foundit; + int status = 0; + + if (!o2hb_global_heartbeat_active()) { + if (qr->qr_numregions) { + mlog(ML_ERROR, "Domain %s: Joining node %d has global " + "heartbeat enabled but local node %d does not\n", + qr->qr_domain, qr->qr_node, dlm->node_num); + status = -EINVAL; + } + goto bail; + } + + if (o2hb_global_heartbeat_active() && !qr->qr_numregions) { + mlog(ML_ERROR, "Domain %s: Local node %d has global " + "heartbeat enabled but joining node %d does not\n", + qr->qr_domain, dlm->node_num, qr->qr_node); + status = -EINVAL; + goto bail; + } + + r = remote; + for (i = 0; i < qr->qr_numregions; ++i) { + mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, r); + r += O2HB_MAX_REGION_NAME_LEN; + } + + localnr = min(O2NM_MAX_REGIONS, locallen/O2HB_MAX_REGION_NAME_LEN); + localnr = o2hb_get_all_regions(local, (u8)localnr); + + /* compare local regions with remote */ + l = local; + for (i = 0; i < localnr; ++i) { + foundit = 0; + r = remote; + for (j = 0; j <= qr->qr_numregions; ++j) { + if (!memcmp(l, r, O2HB_MAX_REGION_NAME_LEN)) { + foundit = 1; + break; + } + r += O2HB_MAX_REGION_NAME_LEN; + } + if (!foundit) { + status = -EINVAL; + mlog(ML_ERROR, "Domain %s: Region '%.*s' registered " + "in local node %d but not in joining node %d\n", + qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, l, + dlm->node_num, qr->qr_node); + goto bail; + } + l += O2HB_MAX_REGION_NAME_LEN; + } + + /* compare remote with local regions */ + r = remote; + for (i = 0; i < qr->qr_numregions; ++i) { + foundit = 0; + l = local; + for (j = 0; j < localnr; ++j) { + if (!memcmp(r, l, O2HB_MAX_REGION_NAME_LEN)) { + foundit = 1; + break; + } + l += O2HB_MAX_REGION_NAME_LEN; + } + if (!foundit) { + status = -EINVAL; + mlog(ML_ERROR, "Domain %s: Region '%.*s' registered " + "in joining node %d but not in local node %d\n", + qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, r, + qr->qr_node, dlm->node_num); + goto bail; + } + r += O2HB_MAX_REGION_NAME_LEN; + } + +bail: + return status; +} + +static int dlm_send_regions(struct dlm_ctxt *dlm, unsigned long *node_map) +{ + struct dlm_query_region *qr = NULL; + int status, ret = 0, i; + char *p; + + if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES) + goto bail; + + qr = kzalloc(sizeof(struct dlm_query_region), GFP_KERNEL); + if (!qr) { + ret = -ENOMEM; + mlog_errno(ret); + goto bail; + } + + qr->qr_node = dlm->node_num; + qr->qr_namelen = strlen(dlm->name); + memcpy(qr->qr_domain, dlm->name, qr->qr_namelen); + /* if local hb, the numregions will be zero */ + if (o2hb_global_heartbeat_active()) + qr->qr_numregions = o2hb_get_all_regions(qr->qr_regions, + O2NM_MAX_REGIONS); + + p = qr->qr_regions; + for (i = 0; i < qr->qr_numregions; ++i, p += O2HB_MAX_REGION_NAME_LEN) + mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, p); + + i = -1; + while ((i = find_next_bit(node_map, O2NM_MAX_NODES, + i + 1)) < O2NM_MAX_NODES) { + if (i == dlm->node_num) + continue; + + mlog(0, "Sending regions to node %d\n", i); + + ret = o2net_send_message(DLM_QUERY_REGION, DLM_MOD_KEY, qr, + sizeof(struct dlm_query_region), + i, &status); + if (ret >= 0) + ret = status; + if (ret) { + mlog(ML_ERROR, "Region mismatch %d, node %d\n", + ret, i); + break; + } + } + +bail: + kfree(qr); + return ret; +} + +static int dlm_query_region_handler(struct o2net_msg *msg, u32 len, + void *data, void **ret_data) +{ + struct dlm_query_region *qr; + struct dlm_ctxt *dlm = NULL; + char *local = NULL; + int status = 0; + + qr = (struct dlm_query_region *) msg->buf; + + mlog(0, "Node %u queries hb regions on domain %s\n", qr->qr_node, + qr->qr_domain); + + /* buffer used in dlm_mast_regions() */ + local = kmalloc(sizeof(qr->qr_regions), GFP_KERNEL); + if (!local) + return -ENOMEM; + + status = -EINVAL; + + spin_lock(&dlm_domain_lock); + dlm = __dlm_lookup_domain_full(qr->qr_domain, qr->qr_namelen); + if (!dlm) { + mlog(ML_ERROR, "Node %d queried hb regions on domain %s " + "before join domain\n", qr->qr_node, qr->qr_domain); + goto out_domain_lock; + } + + spin_lock(&dlm->spinlock); + if (dlm->joining_node != qr->qr_node) { + mlog(ML_ERROR, "Node %d queried hb regions on domain %s " + "but joining node is %d\n", qr->qr_node, qr->qr_domain, + dlm->joining_node); + goto out_dlm_lock; + } + + /* Support for global heartbeat was added in 1.1 */ + if (dlm->dlm_locking_proto.pv_major == 1 && + dlm->dlm_locking_proto.pv_minor == 0) { + mlog(ML_ERROR, "Node %d queried hb regions on domain %s " + "but active dlm protocol is %d.%d\n", qr->qr_node, + qr->qr_domain, dlm->dlm_locking_proto.pv_major, + dlm->dlm_locking_proto.pv_minor); + goto out_dlm_lock; + } + + status = dlm_match_regions(dlm, qr, local, sizeof(qr->qr_regions)); + +out_dlm_lock: + spin_unlock(&dlm->spinlock); + +out_domain_lock: + spin_unlock(&dlm_domain_lock); + + kfree(local); + + return status; +} + +static int dlm_match_nodes(struct dlm_ctxt *dlm, struct dlm_query_nodeinfo *qn) +{ + struct o2nm_node *local; + struct dlm_node_info *remote; + int i, j; + int status = 0; + + for (j = 0; j < qn->qn_numnodes; ++j) + mlog(0, "Node %3d, %pI4:%u\n", qn->qn_nodes[j].ni_nodenum, + &(qn->qn_nodes[j].ni_ipv4_address), + ntohs(qn->qn_nodes[j].ni_ipv4_port)); + + for (i = 0; i < O2NM_MAX_NODES && !status; ++i) { + local = o2nm_get_node_by_num(i); + remote = NULL; + for (j = 0; j < qn->qn_numnodes; ++j) { + if (qn->qn_nodes[j].ni_nodenum == i) { + remote = &(qn->qn_nodes[j]); + break; + } + } + + if (!local && !remote) + continue; + + if ((local && !remote) || (!local && remote)) + status = -EINVAL; + + if (!status && + ((remote->ni_nodenum != local->nd_num) || + (remote->ni_ipv4_port != local->nd_ipv4_port) || + (remote->ni_ipv4_address != local->nd_ipv4_address))) + status = -EINVAL; + + if (status) { + if (remote && !local) + mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) " + "registered in joining node %d but not in " + "local node %d\n", qn->qn_domain, + remote->ni_nodenum, + &(remote->ni_ipv4_address), + ntohs(remote->ni_ipv4_port), + qn->qn_nodenum, dlm->node_num); + if (local && !remote) + mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) " + "registered in local node %d but not in " + "joining node %d\n", qn->qn_domain, + local->nd_num, &(local->nd_ipv4_address), + ntohs(local->nd_ipv4_port), + dlm->node_num, qn->qn_nodenum); + BUG_ON((!local && !remote)); + } + + if (local) + o2nm_node_put(local); + } + + return status; +} + +static int dlm_send_nodeinfo(struct dlm_ctxt *dlm, unsigned long *node_map) +{ + struct dlm_query_nodeinfo *qn = NULL; + struct o2nm_node *node; + int ret = 0, status, count, i; + + if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES) + goto bail; + + qn = kzalloc(sizeof(struct dlm_query_nodeinfo), GFP_KERNEL); + if (!qn) { + ret = -ENOMEM; + mlog_errno(ret); + goto bail; + } + + for (i = 0, count = 0; i < O2NM_MAX_NODES; ++i) { + node = o2nm_get_node_by_num(i); + if (!node) + continue; + qn->qn_nodes[count].ni_nodenum = node->nd_num; + qn->qn_nodes[count].ni_ipv4_port = node->nd_ipv4_port; + qn->qn_nodes[count].ni_ipv4_address = node->nd_ipv4_address; + mlog(0, "Node %3d, %pI4:%u\n", node->nd_num, + &(node->nd_ipv4_address), ntohs(node->nd_ipv4_port)); + ++count; + o2nm_node_put(node); + } + + qn->qn_nodenum = dlm->node_num; + qn->qn_numnodes = count; + qn->qn_namelen = strlen(dlm->name); + memcpy(qn->qn_domain, dlm->name, qn->qn_namelen); + + i = -1; + while ((i = find_next_bit(node_map, O2NM_MAX_NODES, + i + 1)) < O2NM_MAX_NODES) { + if (i == dlm->node_num) + continue; + + mlog(0, "Sending nodeinfo to node %d\n", i); + + ret = o2net_send_message(DLM_QUERY_NODEINFO, DLM_MOD_KEY, + qn, sizeof(struct dlm_query_nodeinfo), + i, &status); + if (ret >= 0) + ret = status; + if (ret) { + mlog(ML_ERROR, "node mismatch %d, node %d\n", ret, i); + break; + } + } + +bail: + kfree(qn); + return ret; +} + +static int dlm_query_nodeinfo_handler(struct o2net_msg *msg, u32 len, + void *data, void **ret_data) +{ + struct dlm_query_nodeinfo *qn; + struct dlm_ctxt *dlm = NULL; + int locked = 0, status = -EINVAL; + + qn = (struct dlm_query_nodeinfo *) msg->buf; + + mlog(0, "Node %u queries nodes on domain %s\n", qn->qn_nodenum, + qn->qn_domain); + + spin_lock(&dlm_domain_lock); + dlm = __dlm_lookup_domain_full(qn->qn_domain, qn->qn_namelen); + if (!dlm) { + mlog(ML_ERROR, "Node %d queried nodes on domain %s before " + "join domain\n", qn->qn_nodenum, qn->qn_domain); + goto bail; + } + + spin_lock(&dlm->spinlock); + locked = 1; + if (dlm->joining_node != qn->qn_nodenum) { + mlog(ML_ERROR, "Node %d queried nodes on domain %s but " + "joining node is %d\n", qn->qn_nodenum, qn->qn_domain, + dlm->joining_node); + goto bail; + } + + /* Support for node query was added in 1.1 */ + if (dlm->dlm_locking_proto.pv_major == 1 && + dlm->dlm_locking_proto.pv_minor == 0) { + mlog(ML_ERROR, "Node %d queried nodes on domain %s " + "but active dlm protocol is %d.%d\n", qn->qn_nodenum, + qn->qn_domain, dlm->dlm_locking_proto.pv_major, + dlm->dlm_locking_proto.pv_minor); + goto bail; + } + + status = dlm_match_nodes(dlm, qn); + +bail: + if (locked) + spin_unlock(&dlm->spinlock); + spin_unlock(&dlm_domain_lock); + + return status; +} + +static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { struct dlm_cancel_join *cancel; struct dlm_ctxt *dlm = NULL; @@ -683,7 +1394,9 @@ static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm, &cancel_msg, sizeof(cancel_msg), node, NULL); if (status < 0) { - mlog_errno(status); + mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to " + "node %u\n", status, DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY, + node); goto bail; } @@ -703,7 +1416,7 @@ static int dlm_send_join_cancels(struct dlm_ctxt *dlm, sizeof(unsigned long))) { mlog(ML_ERROR, "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n", - map_size, BITS_TO_LONGS(O2NM_MAX_NODES)); + map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES)); return -EINVAL; } @@ -730,10 +1443,12 @@ static int dlm_send_join_cancels(struct dlm_ctxt *dlm, static int dlm_request_join(struct dlm_ctxt *dlm, int node, - enum dlm_query_join_response *response) + enum dlm_query_join_response_code *response) { - int status, retval; + int status; struct dlm_query_join_request join_msg; + struct dlm_query_join_packet packet; + u32 join_resp; mlog(0, "querying node %d\n", node); @@ -741,13 +1456,21 @@ static int dlm_request_join(struct dlm_ctxt *dlm, join_msg.node_idx = dlm->node_num; join_msg.name_len = strlen(dlm->name); memcpy(join_msg.domain, dlm->name, join_msg.name_len); + join_msg.dlm_proto = dlm->dlm_locking_proto; + join_msg.fs_proto = dlm->fs_locking_proto; + + /* copy live node map to join message */ + byte_copymap(join_msg.node_map, dlm->live_nodes_map, O2NM_MAX_NODES); status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg, - sizeof(join_msg), node, &retval); + sizeof(join_msg), node, &join_resp); if (status < 0 && status != -ENOPROTOOPT) { - mlog_errno(status); + mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to " + "node %u\n", status, DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, + node); goto bail; } + dlm_query_join_wire_to_packet(join_resp, &packet); /* -ENOPROTOOPT from the net code means the other side isn't listening for our message type -- that's fine, it means @@ -756,18 +1479,43 @@ static int dlm_request_join(struct dlm_ctxt *dlm, if (status == -ENOPROTOOPT) { status = 0; *response = JOIN_OK_NO_MAP; - } else if (retval == JOIN_DISALLOW || - retval == JOIN_OK || - retval == JOIN_OK_NO_MAP) { - *response = retval; + } else if (packet.code == JOIN_DISALLOW || + packet.code == JOIN_OK_NO_MAP) { + *response = packet.code; + } else if (packet.code == JOIN_PROTOCOL_MISMATCH) { + mlog(ML_NOTICE, + "This node requested DLM locking protocol %u.%u and " + "filesystem locking protocol %u.%u. At least one of " + "the protocol versions on node %d is not compatible, " + "disconnecting\n", + dlm->dlm_locking_proto.pv_major, + dlm->dlm_locking_proto.pv_minor, + dlm->fs_locking_proto.pv_major, + dlm->fs_locking_proto.pv_minor, + node); + status = -EPROTO; + *response = packet.code; + } else if (packet.code == JOIN_OK) { + *response = packet.code; + /* Use the same locking protocol as the remote node */ + dlm->dlm_locking_proto.pv_minor = packet.dlm_minor; + dlm->fs_locking_proto.pv_minor = packet.fs_minor; + mlog(0, + "Node %d responds JOIN_OK with DLM locking protocol " + "%u.%u and fs locking protocol %u.%u\n", + node, + dlm->dlm_locking_proto.pv_major, + dlm->dlm_locking_proto.pv_minor, + dlm->fs_locking_proto.pv_major, + dlm->fs_locking_proto.pv_minor); } else { status = -EINVAL; - mlog(ML_ERROR, "invalid response %d from node %u\n", retval, - node); + mlog(ML_ERROR, "invalid response %d from node %u\n", + packet.code, node); } mlog(0, "status %d, node %d response is %d\n", status, node, - *response); + *response); bail: return status; @@ -777,6 +1525,7 @@ static int dlm_send_one_join_assert(struct dlm_ctxt *dlm, unsigned int node) { int status; + int ret; struct dlm_assert_joined assert_msg; mlog(0, "Sending join assert to node %u\n", node); @@ -788,9 +1537,13 @@ static int dlm_send_one_join_assert(struct dlm_ctxt *dlm, status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY, &assert_msg, sizeof(assert_msg), node, - NULL); + &ret); if (status < 0) - mlog_errno(status); + mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to " + "node %u\n", status, DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY, + node); + else + status = ret; return status; } @@ -836,7 +1589,7 @@ struct domain_join_ctxt { static int dlm_should_restart_join(struct dlm_ctxt *dlm, struct domain_join_ctxt *ctxt, - enum dlm_query_join_response response) + enum dlm_query_join_response_code response) { int ret; @@ -862,11 +1615,11 @@ static int dlm_try_to_join_domain(struct dlm_ctxt *dlm) { int status = 0, tmpstat, node; struct domain_join_ctxt *ctxt; - enum dlm_query_join_response response; + enum dlm_query_join_response_code response = JOIN_DISALLOW; - mlog_entry("%p", dlm); + mlog(0, "%p", dlm); - ctxt = kcalloc(1, sizeof(*ctxt), GFP_KERNEL); + ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL); if (!ctxt) { status = -ENOMEM; mlog_errno(status); @@ -920,6 +1673,21 @@ static int dlm_try_to_join_domain(struct dlm_ctxt *dlm) set_bit(dlm->node_num, dlm->domain_map); spin_unlock(&dlm->spinlock); + /* Support for global heartbeat and node info was added in 1.1 */ + if (dlm->dlm_locking_proto.pv_major > 1 || + dlm->dlm_locking_proto.pv_minor > 0) { + status = dlm_send_nodeinfo(dlm, ctxt->yes_resp_map); + if (status) { + mlog_errno(status); + goto bail; + } + status = dlm_send_regions(dlm, ctxt->yes_resp_map); + if (status) { + mlog_errno(status); + goto bail; + } + } + dlm_send_join_asserts(dlm, ctxt->yes_resp_map); /* Joined state *must* be set before the joining node @@ -934,8 +1702,10 @@ static int dlm_try_to_join_domain(struct dlm_ctxt *dlm) bail: spin_lock(&dlm->spinlock); __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN); - if (!status) + if (!status) { + printk(KERN_NOTICE "o2dlm: Joining domain %s ", dlm->name); __dlm_print_nodes(dlm); + } spin_unlock(&dlm->spinlock); if (ctxt) { @@ -956,8 +1726,8 @@ bail: static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm) { - o2hb_unregister_callback(&dlm->dlm_hb_up); - o2hb_unregister_callback(&dlm->dlm_hb_down); + o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_up); + o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_down); o2net_unregister_handler_list(&dlm->dlm_domain_handlers); } @@ -969,111 +1739,126 @@ static int dlm_register_domain_handlers(struct dlm_ctxt *dlm) o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB, dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI); - status = o2hb_register_callback(&dlm->dlm_hb_down); + status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_down); if (status) goto bail; o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB, dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI); - status = o2hb_register_callback(&dlm->dlm_hb_up); + status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_up); if (status) goto bail; status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key, sizeof(struct dlm_master_request), dlm_master_request_handler, - dlm, &dlm->dlm_domain_handlers); + dlm, NULL, &dlm->dlm_domain_handlers); if (status) goto bail; status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key, sizeof(struct dlm_assert_master), dlm_assert_master_handler, - dlm, &dlm->dlm_domain_handlers); + dlm, dlm_assert_master_post_handler, + &dlm->dlm_domain_handlers); if (status) goto bail; status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key, sizeof(struct dlm_create_lock), dlm_create_lock_handler, - dlm, &dlm->dlm_domain_handlers); + dlm, NULL, &dlm->dlm_domain_handlers); if (status) goto bail; status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key, DLM_CONVERT_LOCK_MAX_LEN, dlm_convert_lock_handler, - dlm, &dlm->dlm_domain_handlers); + dlm, NULL, &dlm->dlm_domain_handlers); if (status) goto bail; status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key, DLM_UNLOCK_LOCK_MAX_LEN, dlm_unlock_lock_handler, - dlm, &dlm->dlm_domain_handlers); + dlm, NULL, &dlm->dlm_domain_handlers); if (status) goto bail; status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key, DLM_PROXY_AST_MAX_LEN, dlm_proxy_ast_handler, - dlm, &dlm->dlm_domain_handlers); + dlm, NULL, &dlm->dlm_domain_handlers); if (status) goto bail; status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key, sizeof(struct dlm_exit_domain), dlm_exit_domain_handler, - dlm, &dlm->dlm_domain_handlers); + dlm, NULL, &dlm->dlm_domain_handlers); + if (status) + goto bail; + + status = o2net_register_handler(DLM_DEREF_LOCKRES_MSG, dlm->key, + sizeof(struct dlm_deref_lockres), + dlm_deref_lockres_handler, + dlm, NULL, &dlm->dlm_domain_handlers); if (status) goto bail; status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key, sizeof(struct dlm_migrate_request), dlm_migrate_request_handler, - dlm, &dlm->dlm_domain_handlers); + dlm, NULL, &dlm->dlm_domain_handlers); if (status) goto bail; status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key, DLM_MIG_LOCKRES_MAX_LEN, dlm_mig_lockres_handler, - dlm, &dlm->dlm_domain_handlers); + dlm, NULL, &dlm->dlm_domain_handlers); if (status) goto bail; status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key, sizeof(struct dlm_master_requery), dlm_master_requery_handler, - dlm, &dlm->dlm_domain_handlers); + dlm, NULL, &dlm->dlm_domain_handlers); if (status) goto bail; status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key, sizeof(struct dlm_lock_request), dlm_request_all_locks_handler, - dlm, &dlm->dlm_domain_handlers); + dlm, NULL, &dlm->dlm_domain_handlers); if (status) goto bail; status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key, sizeof(struct dlm_reco_data_done), dlm_reco_data_done_handler, - dlm, &dlm->dlm_domain_handlers); + dlm, NULL, &dlm->dlm_domain_handlers); if (status) goto bail; status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key, sizeof(struct dlm_begin_reco), dlm_begin_reco_handler, - dlm, &dlm->dlm_domain_handlers); + dlm, NULL, &dlm->dlm_domain_handlers); if (status) goto bail; status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key, sizeof(struct dlm_finalize_reco), dlm_finalize_reco_handler, - dlm, &dlm->dlm_domain_handlers); + dlm, NULL, &dlm->dlm_domain_handlers); + if (status) + goto bail; + + status = o2net_register_handler(DLM_BEGIN_EXIT_DOMAIN_MSG, dlm->key, + sizeof(struct dlm_exit_domain), + dlm_begin_exit_domain_handler, + dlm, NULL, &dlm->dlm_domain_handlers); if (status) goto bail; @@ -1087,6 +1872,8 @@ bail: static int dlm_join_domain(struct dlm_ctxt *dlm) { int status; + unsigned int backoff; + unsigned int total_backoff = 0; BUG_ON(!dlm); @@ -1110,19 +1897,41 @@ static int dlm_join_domain(struct dlm_ctxt *dlm) goto bail; } + status = dlm_debug_init(dlm); + if (status < 0) { + mlog_errno(status); + goto bail; + } + + dlm->dlm_worker = create_singlethread_workqueue("dlm_wq"); + if (!dlm->dlm_worker) { + status = -ENOMEM; + mlog_errno(status); + goto bail; + } + do { - unsigned int backoff; status = dlm_try_to_join_domain(dlm); /* If we're racing another node to the join, then we * need to back off temporarily and let them * complete. */ +#define DLM_JOIN_TIMEOUT_MSECS 90000 if (status == -EAGAIN) { if (signal_pending(current)) { status = -ERESTARTSYS; goto bail; } + if (total_backoff > + msecs_to_jiffies(DLM_JOIN_TIMEOUT_MSECS)) { + status = -ERESTARTSYS; + mlog(ML_NOTICE, "Timed out joining dlm domain " + "%s after %u msecs\n", dlm->name, + jiffies_to_msecs(total_backoff)); + goto bail; + } + /* * <chip> After you! * <dale> No, after you! @@ -1132,6 +1941,7 @@ static int dlm_join_domain(struct dlm_ctxt *dlm) */ backoff = (unsigned int)(jiffies & 0x3); backoff *= DLM_DOMAIN_BACKOFF_MS; + total_backoff += backoff; mlog(0, "backoff %d\n", backoff); msleep(backoff); } @@ -1148,8 +1958,10 @@ bail: if (status) { dlm_unregister_domain_handlers(dlm); + dlm_debug_shutdown(dlm); dlm_complete_thread(dlm); dlm_complete_recovery_thread(dlm); + dlm_destroy_dlm_worker(dlm); } return status; @@ -1159,15 +1971,16 @@ static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain, u32 key) { int i; + int ret; struct dlm_ctxt *dlm = NULL; - dlm = kcalloc(1, sizeof(*dlm), GFP_KERNEL); + dlm = kzalloc(sizeof(*dlm), GFP_KERNEL); if (!dlm) { mlog_errno(-ENOMEM); goto leave; } - dlm->name = kmalloc(strlen(domain) + 1, GFP_KERNEL); + dlm->name = kstrdup(domain, GFP_KERNEL); if (dlm->name == NULL) { mlog_errno(-ENOMEM); kfree(dlm); @@ -1175,33 +1988,56 @@ static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain, goto leave; } - dlm->resources = (struct list_head *) __get_free_page(GFP_KERNEL); - if (!dlm->resources) { + dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES); + if (!dlm->lockres_hash) { mlog_errno(-ENOMEM); kfree(dlm->name); kfree(dlm); dlm = NULL; goto leave; } - memset(dlm->resources, 0, PAGE_SIZE); - for (i=0; i<DLM_HASH_SIZE; i++) - INIT_LIST_HEAD(&dlm->resources[i]); + for (i = 0; i < DLM_HASH_BUCKETS; i++) + INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i)); + + dlm->master_hash = (struct hlist_head **) + dlm_alloc_pagevec(DLM_HASH_PAGES); + if (!dlm->master_hash) { + mlog_errno(-ENOMEM); + dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES); + kfree(dlm->name); + kfree(dlm); + dlm = NULL; + goto leave; + } + + for (i = 0; i < DLM_HASH_BUCKETS; i++) + INIT_HLIST_HEAD(dlm_master_hash(dlm, i)); - strcpy(dlm->name, domain); dlm->key = key; dlm->node_num = o2nm_this_node(); + ret = dlm_create_debugfs_subroot(dlm); + if (ret < 0) { + dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES); + dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES); + kfree(dlm->name); + kfree(dlm); + dlm = NULL; + goto leave; + } + spin_lock_init(&dlm->spinlock); spin_lock_init(&dlm->master_lock); spin_lock_init(&dlm->ast_lock); + spin_lock_init(&dlm->track_lock); INIT_LIST_HEAD(&dlm->list); INIT_LIST_HEAD(&dlm->dirty_list); INIT_LIST_HEAD(&dlm->reco.resources); - INIT_LIST_HEAD(&dlm->reco.received); INIT_LIST_HEAD(&dlm->reco.node_data); INIT_LIST_HEAD(&dlm->purge_list); INIT_LIST_HEAD(&dlm->dlm_domain_handlers); + INIT_LIST_HEAD(&dlm->tracking_list); dlm->reco.state = 0; INIT_LIST_HEAD(&dlm->pending_asts); @@ -1216,12 +2052,12 @@ static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain, dlm->dlm_thread_task = NULL; dlm->dlm_reco_thread_task = NULL; + dlm->dlm_worker = NULL; init_waitqueue_head(&dlm->dlm_thread_wq); init_waitqueue_head(&dlm->dlm_reco_thread_wq); init_waitqueue_head(&dlm->reco.event); init_waitqueue_head(&dlm->ast_wq); init_waitqueue_head(&dlm->migration_wq); - INIT_LIST_HEAD(&dlm->master_list); INIT_LIST_HEAD(&dlm->mle_hb_events); dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN; @@ -1229,13 +2065,17 @@ static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain, dlm->reco.new_master = O2NM_INVALID_NODE_NUM; dlm->reco.dead_node = O2NM_INVALID_NODE_NUM; - atomic_set(&dlm->local_resources, 0); - atomic_set(&dlm->remote_resources, 0); - atomic_set(&dlm->unknown_resources, 0); + + atomic_set(&dlm->res_tot_count, 0); + atomic_set(&dlm->res_cur_count, 0); + for (i = 0; i < DLM_MLE_NUM_TYPES; ++i) { + atomic_set(&dlm->mle_tot_count[i], 0); + atomic_set(&dlm->mle_cur_count[i], 0); + } spin_lock_init(&dlm->work_lock); INIT_LIST_HEAD(&dlm->work_list); - INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work, dlm); + INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work); kref_init(&dlm->dlm_refs); dlm->dlm_state = DLM_CTXT_NEW; @@ -1250,28 +2090,49 @@ leave: } /* - * dlm_register_domain: one-time setup per "domain" + * Compare a requested locking protocol version against the current one. + * + * If the major numbers are different, they are incompatible. + * If the current minor is greater than the request, they are incompatible. + * If the current minor is less than or equal to the request, they are + * compatible, and the requester should run at the current minor version. + */ +static int dlm_protocol_compare(struct dlm_protocol_version *existing, + struct dlm_protocol_version *request) +{ + if (existing->pv_major != request->pv_major) + return 1; + + if (existing->pv_minor > request->pv_minor) + return 1; + + if (existing->pv_minor < request->pv_minor) + request->pv_minor = existing->pv_minor; + + return 0; +} + +/* + * dlm_register_domain: one-time setup per "domain". + * + * The filesystem passes in the requested locking version via proto. + * If registration was successful, proto will contain the negotiated + * locking protocol. */ struct dlm_ctxt * dlm_register_domain(const char *domain, - u32 key) + u32 key, + struct dlm_protocol_version *fs_proto) { int ret; struct dlm_ctxt *dlm = NULL; struct dlm_ctxt *new_ctxt = NULL; - if (strlen(domain) > O2NM_MAX_NAME_LEN) { + if (strlen(domain) >= O2NM_MAX_NAME_LEN) { ret = -ENAMETOOLONG; mlog(ML_ERROR, "domain name length too long\n"); goto leave; } - if (!o2hb_check_local_node_heartbeating()) { - mlog(ML_ERROR, "the local node has not been configured, or is " - "not heartbeating\n"); - ret = -EPROTO; - goto leave; - } - mlog(0, "register called for domain \"%s\"\n", domain); retry: @@ -1296,6 +2157,16 @@ retry: goto retry; } + if (dlm_protocol_compare(&dlm->fs_locking_proto, fs_proto)) { + spin_unlock(&dlm_domain_lock); + mlog(ML_ERROR, + "Requested locking protocol version is not " + "compatible with already registered domain " + "\"%s\"\n", domain); + ret = -EPROTO; + goto leave; + } + __dlm_get(dlm); dlm->num_joins++; @@ -1326,6 +2197,13 @@ retry: list_add_tail(&dlm->list, &dlm_domains); spin_unlock(&dlm_domain_lock); + /* + * Pass the locking protocol version into the join. If the join + * succeeds, it will have the negotiated protocol set. + */ + dlm->dlm_locking_proto = dlm_protocol; + dlm->fs_locking_proto = *fs_proto; + ret = dlm_join_domain(dlm); if (ret) { mlog_errno(ret); @@ -1333,6 +2211,9 @@ retry: goto leave; } + /* Tell the caller what locking protocol we negotiated */ + *fs_proto = dlm->fs_locking_proto; + ret = 0; leave: if (new_ctxt) @@ -1359,22 +2240,36 @@ static int dlm_register_net_handlers(void) status = o2net_register_handler(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, sizeof(struct dlm_query_join_request), dlm_query_join_handler, - NULL, &dlm_join_handlers); + NULL, NULL, &dlm_join_handlers); if (status) goto bail; status = o2net_register_handler(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY, sizeof(struct dlm_assert_joined), dlm_assert_joined_handler, - NULL, &dlm_join_handlers); + NULL, NULL, &dlm_join_handlers); if (status) goto bail; status = o2net_register_handler(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY, sizeof(struct dlm_cancel_join), dlm_cancel_join_handler, - NULL, &dlm_join_handlers); + NULL, NULL, &dlm_join_handlers); + if (status) + goto bail; + + status = o2net_register_handler(DLM_QUERY_REGION, DLM_MOD_KEY, + sizeof(struct dlm_query_region), + dlm_query_region_handler, + NULL, NULL, &dlm_join_handlers); + if (status) + goto bail; + + status = o2net_register_handler(DLM_QUERY_NODEINFO, DLM_MOD_KEY, + sizeof(struct dlm_query_nodeinfo), + dlm_query_nodeinfo_handler, + NULL, NULL, &dlm_join_handlers); bail: if (status < 0) dlm_unregister_net_handlers(); @@ -1398,13 +2293,10 @@ static DECLARE_RWSEM(dlm_callback_sem); void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm, int node_num) { - struct list_head *iter; struct dlm_eviction_cb *cb; down_read(&dlm_callback_sem); - list_for_each(iter, &dlm->dlm_eviction_callbacks) { - cb = list_entry(iter, struct dlm_eviction_cb, ec_item); - + list_for_each_entry(cb, &dlm->dlm_eviction_callbacks, ec_item) { cb->ec_func(node_num, cb->ec_data); } up_read(&dlm_callback_sem); @@ -1441,29 +2333,56 @@ static int __init dlm_init(void) { int status; - dlm_print_version(); - status = dlm_init_mle_cache(); - if (status) - return -1; + if (status) { + mlog(ML_ERROR, "Could not create o2dlm_mle slabcache\n"); + goto error; + } + + status = dlm_init_master_caches(); + if (status) { + mlog(ML_ERROR, "Could not create o2dlm_lockres and " + "o2dlm_lockname slabcaches\n"); + goto error; + } + + status = dlm_init_lock_cache(); + if (status) { + mlog(ML_ERROR, "Count not create o2dlm_lock slabcache\n"); + goto error; + } status = dlm_register_net_handlers(); if (status) { - dlm_destroy_mle_cache(); - return -1; + mlog(ML_ERROR, "Unable to register network handlers\n"); + goto error; } + status = dlm_create_debugfs_root(); + if (status) + goto error; + return 0; +error: + dlm_unregister_net_handlers(); + dlm_destroy_lock_cache(); + dlm_destroy_master_caches(); + dlm_destroy_mle_cache(); + return -1; } static void __exit dlm_exit (void) { + dlm_destroy_debugfs_root(); dlm_unregister_net_handlers(); + dlm_destroy_lock_cache(); + dlm_destroy_master_caches(); dlm_destroy_mle_cache(); } MODULE_AUTHOR("Oracle"); MODULE_LICENSE("GPL"); +MODULE_DESCRIPTION("OCFS2 Distributed Lock Management"); module_init(dlm_init); module_exit(dlm_exit); diff --git a/fs/ocfs2/dlm/dlmfs.c b/fs/ocfs2/dlm/dlmfs.c deleted file mode 100644 index dd2d24dc25e..00000000000 --- a/fs/ocfs2/dlm/dlmfs.c +++ /dev/null @@ -1,640 +0,0 @@ -/* -*- mode: c; c-basic-offset: 8; -*- - * vim: noexpandtab sw=8 ts=8 sts=0: - * - * dlmfs.c - * - * Code which implements the kernel side of a minimal userspace - * interface to our DLM. This file handles the virtual file system - * used for communication with userspace. Credit should go to ramfs, - * which was a template for the fs side of this module. - * - * Copyright (C) 2003, 2004 Oracle. All rights reserved. - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public - * License as published by the Free Software Foundation; either - * version 2 of the License, or (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * General Public License for more details. - * - * You should have received a copy of the GNU General Public - * License along with this program; if not, write to the - * Free Software Foundation, Inc., 59 Temple Place - Suite 330, - * Boston, MA 021110-1307, USA. - */ - -/* Simple VFS hooks based on: */ -/* - * Resizable simple ram filesystem for Linux. - * - * Copyright (C) 2000 Linus Torvalds. - * 2000 Transmeta Corp. - */ - -#include <linux/module.h> -#include <linux/fs.h> -#include <linux/pagemap.h> -#include <linux/types.h> -#include <linux/slab.h> -#include <linux/highmem.h> -#include <linux/init.h> -#include <linux/string.h> -#include <linux/smp_lock.h> -#include <linux/backing-dev.h> - -#include <asm/uaccess.h> - - -#include "cluster/nodemanager.h" -#include "cluster/heartbeat.h" -#include "cluster/tcp.h" - -#include "dlmapi.h" - -#include "userdlm.h" - -#include "dlmfsver.h" - -#define MLOG_MASK_PREFIX ML_DLMFS -#include "cluster/masklog.h" - -static struct super_operations dlmfs_ops; -static struct file_operations dlmfs_file_operations; -static struct inode_operations dlmfs_dir_inode_operations; -static struct inode_operations dlmfs_root_inode_operations; -static struct inode_operations dlmfs_file_inode_operations; -static kmem_cache_t *dlmfs_inode_cache; - -struct workqueue_struct *user_dlm_worker; - -/* - * decodes a set of open flags into a valid lock level and a set of flags. - * returns < 0 if we have invalid flags - * flags which mean something to us: - * O_RDONLY -> PRMODE level - * O_WRONLY -> EXMODE level - * - * O_NONBLOCK -> LKM_NOQUEUE - */ -static int dlmfs_decode_open_flags(int open_flags, - int *level, - int *flags) -{ - if (open_flags & (O_WRONLY|O_RDWR)) - *level = LKM_EXMODE; - else - *level = LKM_PRMODE; - - *flags = 0; - if (open_flags & O_NONBLOCK) - *flags |= LKM_NOQUEUE; - - return 0; -} - -static int dlmfs_file_open(struct inode *inode, - struct file *file) -{ - int status, level, flags; - struct dlmfs_filp_private *fp = NULL; - struct dlmfs_inode_private *ip; - - if (S_ISDIR(inode->i_mode)) - BUG(); - - mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino, - file->f_flags); - - status = dlmfs_decode_open_flags(file->f_flags, &level, &flags); - if (status < 0) - goto bail; - - /* We don't want to honor O_APPEND at read/write time as it - * doesn't make sense for LVB writes. */ - file->f_flags &= ~O_APPEND; - - fp = kmalloc(sizeof(*fp), GFP_KERNEL); - if (!fp) { - status = -ENOMEM; - goto bail; - } - fp->fp_lock_level = level; - - ip = DLMFS_I(inode); - - status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags); - if (status < 0) { - /* this is a strange error to return here but I want - * to be able userspace to be able to distinguish a - * valid lock request from one that simply couldn't be - * granted. */ - if (flags & LKM_NOQUEUE && status == -EAGAIN) - status = -ETXTBSY; - kfree(fp); - goto bail; - } - - file->private_data = fp; -bail: - return status; -} - -static int dlmfs_file_release(struct inode *inode, - struct file *file) -{ - int level, status; - struct dlmfs_inode_private *ip = DLMFS_I(inode); - struct dlmfs_filp_private *fp = - (struct dlmfs_filp_private *) file->private_data; - - if (S_ISDIR(inode->i_mode)) - BUG(); - - mlog(0, "close called on inode %lu\n", inode->i_ino); - - status = 0; - if (fp) { - level = fp->fp_lock_level; - if (level != LKM_IVMODE) - user_dlm_cluster_unlock(&ip->ip_lockres, level); - - kfree(fp); - file->private_data = NULL; - } - - return 0; -} - -static ssize_t dlmfs_file_read(struct file *filp, - char __user *buf, - size_t count, - loff_t *ppos) -{ - int bytes_left; - ssize_t readlen; - char *lvb_buf; - struct inode *inode = filp->f_dentry->d_inode; - - mlog(0, "inode %lu, count = %zu, *ppos = %llu\n", - inode->i_ino, count, *ppos); - - if (*ppos >= i_size_read(inode)) - return 0; - - if (!count) - return 0; - - if (!access_ok(VERIFY_WRITE, buf, count)) - return -EFAULT; - - /* don't read past the lvb */ - if ((count + *ppos) > i_size_read(inode)) - readlen = i_size_read(inode) - *ppos; - else - readlen = count - *ppos; - - lvb_buf = kmalloc(readlen, GFP_KERNEL); - if (!lvb_buf) - return -ENOMEM; - - user_dlm_read_lvb(inode, lvb_buf, readlen); - bytes_left = __copy_to_user(buf, lvb_buf, readlen); - readlen -= bytes_left; - - kfree(lvb_buf); - - *ppos = *ppos + readlen; - - mlog(0, "read %zd bytes\n", readlen); - return readlen; -} - -static ssize_t dlmfs_file_write(struct file *filp, - const char __user *buf, - size_t count, - loff_t *ppos) -{ - int bytes_left; - ssize_t writelen; - char *lvb_buf; - struct inode *inode = filp->f_dentry->d_inode; - - mlog(0, "inode %lu, count = %zu, *ppos = %llu\n", - inode->i_ino, count, *ppos); - - if (*ppos >= i_size_read(inode)) - return -ENOSPC; - - if (!count) - return 0; - - if (!access_ok(VERIFY_READ, buf, count)) - return -EFAULT; - - /* don't write past the lvb */ - if ((count + *ppos) > i_size_read(inode)) - writelen = i_size_read(inode) - *ppos; - else - writelen = count - *ppos; - - lvb_buf = kmalloc(writelen, GFP_KERNEL); - if (!lvb_buf) - return -ENOMEM; - - bytes_left = copy_from_user(lvb_buf, buf, writelen); - writelen -= bytes_left; - if (writelen) - user_dlm_write_lvb(inode, lvb_buf, writelen); - - kfree(lvb_buf); - - *ppos = *ppos + writelen; - mlog(0, "wrote %zd bytes\n", writelen); - return writelen; -} - -static void dlmfs_init_once(void *foo, - kmem_cache_t *cachep, - unsigned long flags) -{ - struct dlmfs_inode_private *ip = - (struct dlmfs_inode_private *) foo; - - if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) == - SLAB_CTOR_CONSTRUCTOR) { - ip->ip_dlm = NULL; - ip->ip_parent = NULL; - - inode_init_once(&ip->ip_vfs_inode); - } -} - -static struct inode *dlmfs_alloc_inode(struct super_block *sb) -{ - struct dlmfs_inode_private *ip; - - ip = kmem_cache_alloc(dlmfs_inode_cache, SLAB_NOFS); - if (!ip) - return NULL; - - return &ip->ip_vfs_inode; -} - -static void dlmfs_destroy_inode(struct inode *inode) -{ - kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode)); -} - -static void dlmfs_clear_inode(struct inode *inode) -{ - int status; - struct dlmfs_inode_private *ip; - - if (!inode) - return; - - mlog(0, "inode %lu\n", inode->i_ino); - - ip = DLMFS_I(inode); - - if (S_ISREG(inode->i_mode)) { - status = user_dlm_destroy_lock(&ip->ip_lockres); - if (status < 0) - mlog_errno(status); - iput(ip->ip_parent); - goto clear_fields; - } - - mlog(0, "we're a directory, ip->ip_dlm = 0x%p\n", ip->ip_dlm); - /* we must be a directory. If required, lets unregister the - * dlm context now. */ - if (ip->ip_dlm) - user_dlm_unregister_context(ip->ip_dlm); -clear_fields: - ip->ip_parent = NULL; - ip->ip_dlm = NULL; -} - -static struct backing_dev_info dlmfs_backing_dev_info = { - .ra_pages = 0, /* No readahead */ - .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, -}; - -static struct inode *dlmfs_get_root_inode(struct super_block *sb) -{ - struct inode *inode = new_inode(sb); - int mode = S_IFDIR | 0755; - struct dlmfs_inode_private *ip; - - if (inode) { - ip = DLMFS_I(inode); - - inode->i_mode = mode; - inode->i_uid = current->fsuid; - inode->i_gid = current->fsgid; - inode->i_blksize = PAGE_CACHE_SIZE; - inode->i_blocks = 0; - inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info; - inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; - inode->i_nlink++; - - inode->i_fop = &simple_dir_operations; - inode->i_op = &dlmfs_root_inode_operations; - } - - return inode; -} - -static struct inode *dlmfs_get_inode(struct inode *parent, - struct dentry *dentry, - int mode) -{ - struct super_block *sb = parent->i_sb; - struct inode * inode = new_inode(sb); - struct dlmfs_inode_private *ip; - - if (!inode) - return NULL; - - inode->i_mode = mode; - inode->i_uid = current->fsuid; - inode->i_gid = current->fsgid; - inode->i_blksize = PAGE_CACHE_SIZE; - inode->i_blocks = 0; - inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info; - inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; - - ip = DLMFS_I(inode); - ip->ip_dlm = DLMFS_I(parent)->ip_dlm; - - switch (mode & S_IFMT) { - default: - /* for now we don't support anything other than - * directories and regular files. */ - BUG(); - break; - case S_IFREG: - inode->i_op = &dlmfs_file_inode_operations; - inode->i_fop = &dlmfs_file_operations; - - i_size_write(inode, DLM_LVB_LEN); - - user_dlm_lock_res_init(&ip->ip_lockres, dentry); - - /* released at clear_inode time, this insures that we - * get to drop the dlm reference on each lock *before* - * we call the unregister code for releasing parent - * directories. */ - ip->ip_parent = igrab(parent); - BUG_ON(!ip->ip_parent); - break; - case S_IFDIR: - inode->i_op = &dlmfs_dir_inode_operations; - inode->i_fop = &simple_dir_operations; - - /* directory inodes start off with i_nlink == - * 2 (for "." entry) */ - inode->i_nlink++; - break; - } - - if (parent->i_mode & S_ISGID) { - inode->i_gid = parent->i_gid; - if (S_ISDIR(mode)) - inode->i_mode |= S_ISGID; - } - - return inode; -} - -/* - * File creation. Allocate an inode, and we're done.. - */ -/* SMP-safe */ -static int dlmfs_mkdir(struct inode * dir, - struct dentry * dentry, - int mode) -{ - int status; - struct inode *inode = NULL; - struct qstr *domain = &dentry->d_name; - struct dlmfs_inode_private *ip; - struct dlm_ctxt *dlm; - - mlog(0, "mkdir %.*s\n", domain->len, domain->name); - - /* verify that we have a proper domain */ - if (domain->len >= O2NM_MAX_NAME_LEN) { - status = -EINVAL; - mlog(ML_ERROR, "invalid domain name for directory.\n"); - goto bail; - } - - inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR); - if (!inode) { - status = -ENOMEM; - mlog_errno(status); - goto bail; - } - - ip = DLMFS_I(inode); - - dlm = user_dlm_register_context(domain); - if (IS_ERR(dlm)) { - status = PTR_ERR(dlm); - mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n", - status, domain->len, domain->name); - goto bail; - } - ip->ip_dlm = dlm; - - dir->i_nlink++; - d_instantiate(dentry, inode); - dget(dentry); /* Extra count - pin the dentry in core */ - - status = 0; -bail: - if (status < 0) - iput(inode); - return status; -} - -static int dlmfs_create(struct inode *dir, - struct dentry *dentry, - int mode, - struct nameidata *nd) -{ - int status = 0; - struct inode *inode; - struct qstr *name = &dentry->d_name; - - mlog(0, "create %.*s\n", name->len, name->name); - - /* verify name is valid and doesn't contain any dlm reserved - * characters */ - if (name->len >= USER_DLM_LOCK_ID_MAX_LEN || - name->name[0] == '$') { - status = -EINVAL; - mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len, - name->name); - goto bail; - } - - inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG); - if (!inode) { - status = -ENOMEM; - mlog_errno(status); - goto bail; - } - - d_instantiate(dentry, inode); - dget(dentry); /* Extra count - pin the dentry in core */ -bail: - return status; -} - -static int dlmfs_unlink(struct inode *dir, - struct dentry *dentry) -{ - int status; - struct inode *inode = dentry->d_inode; - - mlog(0, "unlink inode %lu\n", inode->i_ino); - - /* if there are no current holders, or none that are waiting - * to acquire a lock, this basically destroys our lockres. */ - status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres); - if (status < 0) { - mlog(ML_ERROR, "unlink %.*s, error %d from destroy\n", - dentry->d_name.len, dentry->d_name.name, status); - goto bail; - } - status = simple_unlink(dir, dentry); -bail: - return status; -} - -static int dlmfs_fill_super(struct super_block * sb, - void * data, - int silent) -{ - struct inode * inode; - struct dentry * root; - - sb->s_maxbytes = MAX_LFS_FILESIZE; - sb->s_blocksize = PAGE_CACHE_SIZE; - sb->s_blocksize_bits = PAGE_CACHE_SHIFT; - sb->s_magic = DLMFS_MAGIC; - sb->s_op = &dlmfs_ops; - inode = dlmfs_get_root_inode(sb); - if (!inode) - return -ENOMEM; - - root = d_alloc_root(inode); - if (!root) { - iput(inode); - return -ENOMEM; - } - sb->s_root = root; - return 0; -} - -static struct file_operations dlmfs_file_operations = { - .open = dlmfs_file_open, - .release = dlmfs_file_release, - .read = dlmfs_file_read, - .write = dlmfs_file_write, -}; - -static struct inode_operations dlmfs_dir_inode_operations = { - .create = dlmfs_create, - .lookup = simple_lookup, - .unlink = dlmfs_unlink, -}; - -/* this way we can restrict mkdir to only the toplevel of the fs. */ -static struct inode_operations dlmfs_root_inode_operations = { - .lookup = simple_lookup, - .mkdir = dlmfs_mkdir, - .rmdir = simple_rmdir, -}; - -static struct super_operations dlmfs_ops = { - .statfs = simple_statfs, - .alloc_inode = dlmfs_alloc_inode, - .destroy_inode = dlmfs_destroy_inode, - .clear_inode = dlmfs_clear_inode, - .drop_inode = generic_delete_inode, -}; - -static struct inode_operations dlmfs_file_inode_operations = { - .getattr = simple_getattr, -}; - -static struct super_block *dlmfs_get_sb(struct file_system_type *fs_type, - int flags, const char *dev_name, void *data) -{ - return get_sb_nodev(fs_type, flags, data, dlmfs_fill_super); -} - -static struct file_system_type dlmfs_fs_type = { - .owner = THIS_MODULE, - .name = "ocfs2_dlmfs", - .get_sb = dlmfs_get_sb, - .kill_sb = kill_litter_super, -}; - -static int __init init_dlmfs_fs(void) -{ - int status; - int cleanup_inode = 0, cleanup_worker = 0; - - dlmfs_print_version(); - - dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache", - sizeof(struct dlmfs_inode_private), - 0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT, - dlmfs_init_once, NULL); - if (!dlmfs_inode_cache) - return -ENOMEM; - cleanup_inode = 1; - - user_dlm_worker = create_singlethread_workqueue("user_dlm"); - if (!user_dlm_worker) { - status = -ENOMEM; - goto bail; - } - cleanup_worker = 1; - - status = register_filesystem(&dlmfs_fs_type); -bail: - if (status) { - if (cleanup_inode) - kmem_cache_destroy(dlmfs_inode_cache); - if (cleanup_worker) - destroy_workqueue(user_dlm_worker); - } else - printk("OCFS2 User DLM kernel interface loaded\n"); - return status; -} - -static void __exit exit_dlmfs_fs(void) -{ - unregister_filesystem(&dlmfs_fs_type); - - flush_workqueue(user_dlm_worker); - destroy_workqueue(user_dlm_worker); - - if (kmem_cache_destroy(dlmfs_inode_cache)) - printk(KERN_INFO "dlmfs_inode_cache: not all structures " - "were freed\n"); -} - -MODULE_AUTHOR("Oracle"); -MODULE_LICENSE("GPL"); - -module_init(init_dlmfs_fs) -module_exit(exit_dlmfs_fs) diff --git a/fs/ocfs2/dlm/dlmfsver.c b/fs/ocfs2/dlm/dlmfsver.c deleted file mode 100644 index d2be3ad841f..00000000000 --- a/fs/ocfs2/dlm/dlmfsver.c +++ /dev/null @@ -1,42 +0,0 @@ -/* -*- mode: c; c-basic-offset: 8; -*- - * vim: noexpandtab sw=8 ts=8 sts=0: - * - * dlmfsver.c - * - * version string - * - * Copyright (C) 2002, 2005 Oracle. All rights reserved. - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public - * License as published by the Free Software Foundation; either - * version 2 of the License, or (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * General Public License for more details. - * - * You should have received a copy of the GNU General Public - * License along with this program; if not, write to the - * Free Software Foundation, Inc., 59 Temple Place - Suite 330, - * Boston, MA 021110-1307, USA. - */ - -#include <linux/module.h> -#include <linux/kernel.h> - -#include "dlmfsver.h" - -#define DLM_BUILD_VERSION "1.3.3" - -#define VERSION_STR "OCFS2 DLMFS " DLM_BUILD_VERSION - -void dlmfs_print_version(void) -{ - printk(KERN_INFO "%s\n", VERSION_STR); -} - -MODULE_DESCRIPTION(VERSION_STR); - -MODULE_VERSION(DLM_BUILD_VERSION); diff --git a/fs/ocfs2/dlm/dlmfsver.h b/fs/ocfs2/dlm/dlmfsver.h deleted file mode 100644 index f35eadbed25..00000000000 --- a/fs/ocfs2/dlm/dlmfsver.h +++ /dev/null @@ -1,31 +0,0 @@ -/* -*- mode: c; c-basic-offset: 8; -*- - * vim: noexpandtab sw=8 ts=8 sts=0: - * - * dlmver.h - * - * Function prototypes - * - * Copyright (C) 2005 Oracle. All rights reserved. - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public - * License as published by the Free Software Foundation; either - * version 2 of the License, or (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * General Public License for more details. - * - * You should have received a copy of the GNU General Public - * License along with this program; if not, write to the - * Free Software Foundation, Inc., 59 Temple Place - Suite 330, - * Boston, MA 021110-1307, USA. - */ - -#ifndef DLMFS_VER_H -#define DLMFS_VER_H - -void dlmfs_print_version(void); - -#endif /* DLMFS_VER_H */ diff --git a/fs/ocfs2/dlm/dlmlock.c b/fs/ocfs2/dlm/dlmlock.c index d1a0038557a..66c2a491f68 100644 --- a/fs/ocfs2/dlm/dlmlock.c +++ b/fs/ocfs2/dlm/dlmlock.c @@ -30,7 +30,6 @@ #include <linux/types.h> #include <linux/slab.h> #include <linux/highmem.h> -#include <linux/utsname.h> #include <linux/init.h> #include <linux/sysctl.h> #include <linux/random.h> @@ -53,7 +52,9 @@ #define MLOG_MASK_PREFIX ML_DLM #include "cluster/masklog.h" -static spinlock_t dlm_cookie_lock = SPIN_LOCK_UNLOCKED; +static struct kmem_cache *dlm_lock_cache; + +static DEFINE_SPINLOCK(dlm_cookie_lock); static u64 dlm_next_cookie = 1; static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm, @@ -64,6 +65,22 @@ static void dlm_init_lock(struct dlm_lock *newlock, int type, static void dlm_lock_release(struct kref *kref); static void dlm_lock_detach_lockres(struct dlm_lock *lock); +int dlm_init_lock_cache(void) +{ + dlm_lock_cache = kmem_cache_create("o2dlm_lock", + sizeof(struct dlm_lock), + 0, SLAB_HWCACHE_ALIGN, NULL); + if (dlm_lock_cache == NULL) + return -ENOMEM; + return 0; +} + +void dlm_destroy_lock_cache(void) +{ + if (dlm_lock_cache) + kmem_cache_destroy(dlm_lock_cache); +} + /* Tell us whether we can grant a new lock request. * locking: * caller needs: res->spinlock @@ -74,21 +91,19 @@ static void dlm_lock_detach_lockres(struct dlm_lock *lock); static int dlm_can_grant_new_lock(struct dlm_lock_resource *res, struct dlm_lock *lock) { - struct list_head *iter; struct dlm_lock *tmplock; - list_for_each(iter, &res->granted) { - tmplock = list_entry(iter, struct dlm_lock, list); - + list_for_each_entry(tmplock, &res->granted, list) { if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type)) return 0; } - list_for_each(iter, &res->converting) { - tmplock = list_entry(iter, struct dlm_lock, list); - + list_for_each_entry(tmplock, &res->converting, list) { if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type)) return 0; + if (!dlm_lock_compatible(tmplock->ml.convert_type, + lock->ml.type)) + return 0; } return 1; @@ -108,7 +123,7 @@ static enum dlm_status dlmlock_master(struct dlm_ctxt *dlm, int call_ast = 0, kick_thread = 0; enum dlm_status status = DLM_NORMAL; - mlog_entry("type=%d\n", lock->ml.type); + mlog(0, "type=%d\n", lock->ml.type); spin_lock(&res->spinlock); /* if called from dlm_create_lock_handler, need to @@ -141,13 +156,24 @@ static enum dlm_status dlmlock_master(struct dlm_ctxt *dlm, res->lockname.len)) { kick_thread = 1; call_ast = 1; + } else { + mlog(0, "%s: returning DLM_NORMAL to " + "node %u for reco lock\n", dlm->name, + lock->ml.node); } } else { /* for NOQUEUE request, unless we get the * lock right away, return DLM_NOTQUEUED */ - if (flags & LKM_NOQUEUE) + if (flags & LKM_NOQUEUE) { status = DLM_NOTQUEUED; - else { + if (dlm_is_recovery_lock(res->lockname.name, + res->lockname.len)) { + mlog(0, "%s: returning NOTQUEUED to " + "node %u for reco lock\n", dlm->name, + lock->ml.node); + } + } else { + status = DLM_NORMAL; dlm_lock_get(lock); list_add_tail(&lock->list, &res->blocked); kick_thread = 1; @@ -191,15 +217,22 @@ static enum dlm_status dlmlock_remote(struct dlm_ctxt *dlm, struct dlm_lock *lock, int flags) { enum dlm_status status = DLM_DENIED; + int lockres_changed = 1; - mlog_entry("type=%d\n", lock->ml.type); - mlog(0, "lockres %.*s, flags = 0x%x\n", res->lockname.len, + mlog(0, "type=%d, lockres %.*s, flags = 0x%x\n", + lock->ml.type, res->lockname.len, res->lockname.name, flags); + /* + * Wait if resource is getting recovered, remastered, etc. + * If the resource was remastered and new owner is self, then exit. + */ spin_lock(&res->spinlock); - - /* will exit this call with spinlock held */ __dlm_wait_on_lockres(res); + if (res->owner == dlm->node_num) { + spin_unlock(&res->spinlock); + return DLM_RECOVERING; + } res->state |= DLM_LOCK_RES_IN_PROGRESS; /* add lock to local (secondary) queue */ @@ -216,14 +249,42 @@ static enum dlm_status dlmlock_remote(struct dlm_ctxt *dlm, res->state &= ~DLM_LOCK_RES_IN_PROGRESS; lock->lock_pending = 0; if (status != DLM_NORMAL) { - if (status != DLM_NOTQUEUED) + if (status == DLM_RECOVERING && + dlm_is_recovery_lock(res->lockname.name, + res->lockname.len)) { + /* recovery lock was mastered by dead node. + * we need to have calc_usage shoot down this + * lockres and completely remaster it. */ + mlog(0, "%s: recovery lock was owned by " + "dead node %u, remaster it now.\n", + dlm->name, res->owner); + } else if (status != DLM_NOTQUEUED) { + /* + * DO NOT call calc_usage, as this would unhash + * the remote lockres before we ever get to use + * it. treat as if we never made any change to + * the lockres. + */ + lockres_changed = 0; dlm_error(status); + } dlm_revert_pending_lock(res, lock); dlm_lock_put(lock); + } else if (dlm_is_recovery_lock(res->lockname.name, + res->lockname.len)) { + /* special case for the $RECOVERY lock. + * there will never be an AST delivered to put + * this lock on the proper secondary queue + * (granted), so do it manually. */ + mlog(0, "%s: $RECOVERY lock for this node (%u) is " + "mastered by %u; got lock, manually granting (no ast)\n", + dlm->name, dlm->node_num, res->owner); + list_move_tail(&lock->list, &res->granted); } spin_unlock(&res->spinlock); - dlm_lockres_calc_usage(dlm, res); + if (lockres_changed) + dlm_lockres_calc_usage(dlm, res); wake_up(&res->wq); return status; @@ -245,8 +306,6 @@ static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm, int tmpret, status = 0; enum dlm_status ret; - mlog_entry_void(); - memset(&create, 0, sizeof(create)); create.node_idx = dlm->node_num; create.requested_type = lock->ml.type; @@ -258,17 +317,23 @@ static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm, tmpret = o2net_send_message(DLM_CREATE_LOCK_MSG, dlm->key, &create, sizeof(create), res->owner, &status); if (tmpret >= 0) { - // successfully sent and received - ret = status; // this is already a dlm_status + ret = status; + if (ret == DLM_REJECTED) { + mlog(ML_ERROR, "%s: res %.*s, Stale lockres no longer " + "owned by node %u. That node is coming back up " + "currently.\n", dlm->name, create.namelen, + create.name, res->owner); + dlm_print_one_lock_resource(res); + BUG(); + } } else { - mlog_errno(tmpret); - if (dlm_is_host_down(tmpret)) { + mlog(ML_ERROR, "%s: res %.*s, Error %d send CREATE LOCK to " + "node %u\n", dlm->name, create.namelen, create.name, + tmpret, res->owner); + if (dlm_is_host_down(tmpret)) ret = DLM_RECOVERING; - mlog(0, "node %u died so returning DLM_RECOVERING " - "from lock message!\n", res->owner); - } else { + else ret = dlm_err_to_dlm_status(tmpret); - } } return ret; @@ -302,7 +367,7 @@ static void dlm_lock_release(struct kref *kref) mlog(0, "freeing kernel-allocated lksb\n"); kfree(lock->lksb); } - kfree(lock); + kmem_cache_free(dlm_lock_cache, lock); } /* associate a lock with it's lockres, getting a ref on the lockres */ @@ -361,15 +426,15 @@ struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie, struct dlm_lock *lock; int kernel_allocated = 0; - lock = kcalloc(1, sizeof(*lock), GFP_KERNEL); + lock = kmem_cache_zalloc(dlm_lock_cache, GFP_NOFS); if (!lock) return NULL; if (!lksb) { /* zero memory only if kernel-allocated */ - lksb = kcalloc(1, sizeof(*lksb), GFP_KERNEL); + lksb = kzalloc(sizeof(*lksb), GFP_NOFS); if (!lksb) { - kfree(lock); + kmem_cache_free(dlm_lock_cache, lock); return NULL; } kernel_allocated = 1; @@ -390,7 +455,8 @@ struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie, * held on exit: none * returns: DLM_NORMAL, DLM_SYSERR, DLM_IVLOCKID, DLM_NOTQUEUED */ -int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data) +int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { struct dlm_ctxt *dlm = data; struct dlm_create_lock *create = (struct dlm_create_lock *)msg->buf; @@ -403,16 +469,19 @@ int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data) BUG_ON(!dlm); - mlog_entry_void(); - if (!dlm_grab(dlm)) return DLM_REJECTED; - mlog_bug_on_msg(!dlm_domain_fully_joined(dlm), - "Domain %s not fully joined!\n", dlm->name); - name = create->name; namelen = create->namelen; + status = DLM_REJECTED; + if (!dlm_domain_fully_joined(dlm)) { + mlog(ML_ERROR, "Domain %s not fully joined, but node %u is " + "sending a create_lock message for lock %.*s!\n", + dlm->name, create->node_idx, namelen, name); + dlm_error(status); + goto leave; + } status = DLM_IVBUFLEN; if (namelen > DLM_LOCKID_NAME_MAX) { @@ -488,8 +557,8 @@ static inline void dlm_get_next_cookie(u8 node_num, u64 *cookie) enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode, struct dlm_lockstatus *lksb, int flags, - const char *name, dlm_astlockfunc_t *ast, void *data, - dlm_bastlockfunc_t *bast) + const char *name, int namelen, dlm_astlockfunc_t *ast, + void *data, dlm_bastlockfunc_t *bast) { enum dlm_status status; struct dlm_lock_resource *res = NULL; @@ -519,7 +588,7 @@ enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode, recovery = (flags & LKM_RECOVERY); if (recovery && - (!dlm_is_recovery_lock(name, strlen(name)) || convert) ) { + (!dlm_is_recovery_lock(name, namelen) || convert) ) { dlm_error(status); goto error; } @@ -591,7 +660,7 @@ retry_convert: } status = DLM_IVBUFLEN; - if (strlen(name) > DLM_LOCKID_NAME_MAX || strlen(name) < 1) { + if (namelen > DLM_LOCKID_NAME_MAX || namelen < 1) { dlm_error(status); goto error; } @@ -607,7 +676,7 @@ retry_convert: dlm_wait_for_recovery(dlm); /* find or create the lock resource */ - res = dlm_get_lock_resource(dlm, name, flags); + res = dlm_get_lock_resource(dlm, name, namelen, flags); if (!res) { status = DLM_IVLOCKID; dlm_error(status); @@ -643,13 +712,29 @@ retry_lock: if (status == DLM_RECOVERING || status == DLM_MIGRATING || status == DLM_FORWARD) { - mlog(0, "retrying lock with migration/" - "recovery/in progress\n"); msleep(100); - dlm_wait_for_recovery(dlm); - goto retry_lock; + if (recovery) { + if (status != DLM_RECOVERING) + goto retry_lock; + /* wait to see the node go down, then + * drop down and allow the lockres to + * get cleaned up. need to remaster. */ + dlm_wait_for_node_death(dlm, res->owner, + DLM_NODE_DEATH_WAIT_MAX); + } else { + dlm_wait_for_recovery(dlm); + goto retry_lock; + } } + /* Inflight taken in dlm_get_lock_resource() is dropped here */ + spin_lock(&res->spinlock); + dlm_lockres_drop_inflight_ref(dlm, res); + spin_unlock(&res->spinlock); + + dlm_lockres_calc_usage(dlm, res); + dlm_kick_thread(dlm, res); + if (status != DLM_NORMAL) { lock->lksb->flags &= ~DLM_LKSB_GET_LVB; if (status != DLM_NOTQUEUED) diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c index 27e984f7e4c..82abf0cc9a1 100644 --- a/fs/ocfs2/dlm/dlmmaster.c +++ b/fs/ocfs2/dlm/dlmmaster.c @@ -30,7 +30,6 @@ #include <linux/types.h> #include <linux/slab.h> #include <linux/highmem.h> -#include <linux/utsname.h> #include <linux/init.h> #include <linux/sysctl.h> #include <linux/random.h> @@ -47,48 +46,12 @@ #include "dlmapi.h" #include "dlmcommon.h" -#include "dlmdebug.h" #include "dlmdomain.h" +#include "dlmdebug.h" #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_MASTER) #include "cluster/masklog.h" -enum dlm_mle_type { - DLM_MLE_BLOCK, - DLM_MLE_MASTER, - DLM_MLE_MIGRATION -}; - -struct dlm_lock_name -{ - u8 len; - u8 name[DLM_LOCKID_NAME_MAX]; -}; - -struct dlm_master_list_entry -{ - struct list_head list; - struct list_head hb_events; - struct dlm_ctxt *dlm; - spinlock_t spinlock; - wait_queue_head_t wq; - atomic_t woken; - struct kref mle_refs; - unsigned long maybe_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; - unsigned long vote_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; - unsigned long response_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; - unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)]; - u8 master; - u8 new_master; - enum dlm_mle_type type; - struct o2hb_callback_func mle_hb_up; - struct o2hb_callback_func mle_hb_down; - union { - struct dlm_lock_resource *res; - struct dlm_lock_name name; - } u; -}; - static void dlm_mle_node_down(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle, struct o2nm_node *node, @@ -99,107 +62,29 @@ static void dlm_mle_node_up(struct dlm_ctxt *dlm, int idx); static void dlm_assert_master_worker(struct dlm_work_item *item, void *data); -static int dlm_do_assert_master(struct dlm_ctxt *dlm, const char *lockname, - unsigned int namelen, void *nodemap, - u32 flags); +static int dlm_do_assert_master(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res, + void *nodemap, u32 flags); +static void dlm_deref_lockres_worker(struct dlm_work_item *item, void *data); static inline int dlm_mle_equal(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle, const char *name, unsigned int namelen) { - struct dlm_lock_resource *res; - if (dlm != mle->dlm) return 0; - if (mle->type == DLM_MLE_BLOCK || - mle->type == DLM_MLE_MIGRATION) { - if (namelen != mle->u.name.len || - memcmp(name, mle->u.name.name, namelen)!=0) - return 0; - } else { - res = mle->u.res; - if (namelen != res->lockname.len || - memcmp(res->lockname.name, name, namelen) != 0) - return 0; - } - return 1; -} - -#if 0 -/* Code here is included but defined out as it aids debugging */ - -void dlm_print_one_mle(struct dlm_master_list_entry *mle) -{ - int i = 0, refs; - char *type; - char attached; - u8 master; - unsigned int namelen; - const char *name; - struct kref *k; - - k = &mle->mle_refs; - if (mle->type == DLM_MLE_BLOCK) - type = "BLK"; - else if (mle->type == DLM_MLE_MASTER) - type = "MAS"; - else - type = "MIG"; - refs = atomic_read(&k->refcount); - master = mle->master; - attached = (list_empty(&mle->hb_events) ? 'N' : 'Y'); - - if (mle->type != DLM_MLE_MASTER) { - namelen = mle->u.name.len; - name = mle->u.name.name; - } else { - namelen = mle->u.res->lockname.len; - name = mle->u.res->lockname.name; - } - - mlog(ML_NOTICE, " #%3d: %3s %3d %3u %3u %c (%d)%.*s\n", - i, type, refs, master, mle->new_master, attached, - namelen, namelen, name); -} - -static void dlm_dump_mles(struct dlm_ctxt *dlm) -{ - struct dlm_master_list_entry *mle; - struct list_head *iter; - - mlog(ML_NOTICE, "dumping all mles for domain %s:\n", dlm->name); - mlog(ML_NOTICE, " ####: type refs owner new events? lockname nodemap votemap respmap maybemap\n"); - spin_lock(&dlm->master_lock); - list_for_each(iter, &dlm->master_list) { - mle = list_entry(iter, struct dlm_master_list_entry, list); - dlm_print_one_mle(mle); - } - spin_unlock(&dlm->master_lock); -} - -int dlm_dump_all_mles(const char __user *data, unsigned int len) -{ - struct list_head *iter; - struct dlm_ctxt *dlm; + if (namelen != mle->mnamelen || + memcmp(name, mle->mname, namelen) != 0) + return 0; - spin_lock(&dlm_domain_lock); - list_for_each(iter, &dlm_domains) { - dlm = list_entry (iter, struct dlm_ctxt, list); - mlog(ML_NOTICE, "found dlm: %p, name=%s\n", dlm, dlm->name); - dlm_dump_mles(dlm); - } - spin_unlock(&dlm_domain_lock); - return len; + return 1; } -EXPORT_SYMBOL_GPL(dlm_dump_all_mles); - -#endif /* 0 */ - - -static kmem_cache_t *dlm_mle_cache = NULL; +static struct kmem_cache *dlm_lockres_cache; +static struct kmem_cache *dlm_lockname_cache; +static struct kmem_cache *dlm_mle_cache; static void dlm_mle_release(struct kref *kref); static void dlm_init_mle(struct dlm_master_list_entry *mle, @@ -214,7 +99,8 @@ static int dlm_find_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry **mle, char *name, unsigned int namelen); -static int dlm_do_master_request(struct dlm_master_list_entry *mle, int to); +static int dlm_do_master_request(struct dlm_lock_resource *res, + struct dlm_master_list_entry *mle, int to); static int dlm_wait_for_lock_mastery(struct dlm_ctxt *dlm, @@ -239,6 +125,8 @@ static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm, static int dlm_mark_lockres_migrating(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, u8 target); +static int dlm_pre_master_reco_lockres(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res); int dlm_is_host_down(int errno) @@ -312,6 +200,31 @@ static inline void dlm_mle_detach_hb_events(struct dlm_ctxt *dlm, spin_unlock(&dlm->spinlock); } +static void dlm_get_mle_inuse(struct dlm_master_list_entry *mle) +{ + struct dlm_ctxt *dlm; + dlm = mle->dlm; + + assert_spin_locked(&dlm->spinlock); + assert_spin_locked(&dlm->master_lock); + mle->inuse++; + kref_get(&mle->mle_refs); +} + +static void dlm_put_mle_inuse(struct dlm_master_list_entry *mle) +{ + struct dlm_ctxt *dlm; + dlm = mle->dlm; + + spin_lock(&dlm->spinlock); + spin_lock(&dlm->master_lock); + mle->inuse--; + __dlm_put_mle(mle); + spin_unlock(&dlm->master_lock); + spin_unlock(&dlm->spinlock); + +} + /* remove from list and free */ static void __dlm_put_mle(struct dlm_master_list_entry *mle) { @@ -320,9 +233,14 @@ static void __dlm_put_mle(struct dlm_master_list_entry *mle) assert_spin_locked(&dlm->spinlock); assert_spin_locked(&dlm->master_lock); - BUG_ON(!atomic_read(&mle->mle_refs.refcount)); - - kref_put(&mle->mle_refs, dlm_mle_release); + if (!atomic_read(&mle->mle_refs.refcount)) { + /* this may or may not crash, but who cares. + * it's a BUG. */ + mlog(ML_ERROR, "bad mle: %p\n", mle); + dlm_print_one_mle(mle); + BUG(); + } else + kref_put(&mle->mle_refs, dlm_mle_release); } @@ -355,7 +273,7 @@ static void dlm_init_mle(struct dlm_master_list_entry *mle, mle->dlm = dlm; mle->type = type; - INIT_LIST_HEAD(&mle->list); + INIT_HLIST_NODE(&mle->master_hash_node); INIT_LIST_HEAD(&mle->hb_events); memset(mle->maybe_map, 0, sizeof(mle->maybe_map)); spin_lock_init(&mle->spinlock); @@ -365,20 +283,29 @@ static void dlm_init_mle(struct dlm_master_list_entry *mle, memset(mle->response_map, 0, sizeof(mle->response_map)); mle->master = O2NM_MAX_NODES; mle->new_master = O2NM_MAX_NODES; + mle->inuse = 0; + + BUG_ON(mle->type != DLM_MLE_BLOCK && + mle->type != DLM_MLE_MASTER && + mle->type != DLM_MLE_MIGRATION); if (mle->type == DLM_MLE_MASTER) { BUG_ON(!res); - mle->u.res = res; - } else if (mle->type == DLM_MLE_BLOCK) { - BUG_ON(!name); - memcpy(mle->u.name.name, name, namelen); - mle->u.name.len = namelen; - } else /* DLM_MLE_MIGRATION */ { + mle->mleres = res; + memcpy(mle->mname, res->lockname.name, res->lockname.len); + mle->mnamelen = res->lockname.len; + mle->mnamehash = res->lockname.hash; + } else { BUG_ON(!name); - memcpy(mle->u.name.name, name, namelen); - mle->u.name.len = namelen; + mle->mleres = NULL; + memcpy(mle->mname, name, namelen); + mle->mnamelen = namelen; + mle->mnamehash = dlm_lockid_hash(name, namelen); } + atomic_inc(&dlm->mle_tot_count[mle->type]); + atomic_inc(&dlm->mle_cur_count[mle->type]); + /* copy off the node_map and register hb callbacks on our copy */ memcpy(mle->node_map, dlm->domain_map, sizeof(mle->node_map)); memcpy(mle->vote_map, dlm->domain_map, sizeof(mle->vote_map)); @@ -389,6 +316,24 @@ static void dlm_init_mle(struct dlm_master_list_entry *mle, __dlm_mle_attach_hb_events(dlm, mle); } +void __dlm_unlink_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle) +{ + assert_spin_locked(&dlm->spinlock); + assert_spin_locked(&dlm->master_lock); + + if (!hlist_unhashed(&mle->master_hash_node)) + hlist_del_init(&mle->master_hash_node); +} + +void __dlm_insert_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle) +{ + struct hlist_head *bucket; + + assert_spin_locked(&dlm->master_lock); + + bucket = dlm_master_hash(dlm, mle->mnamehash); + hlist_add_head(&mle->master_hash_node, bucket); +} /* returns 1 if found, 0 if not */ static int dlm_find_mle(struct dlm_ctxt *dlm, @@ -396,12 +341,14 @@ static int dlm_find_mle(struct dlm_ctxt *dlm, char *name, unsigned int namelen) { struct dlm_master_list_entry *tmpmle; - struct list_head *iter; + struct hlist_head *bucket; + unsigned int hash; assert_spin_locked(&dlm->master_lock); - list_for_each(iter, &dlm->master_list) { - tmpmle = list_entry(iter, struct dlm_master_list_entry, list); + hash = dlm_lockid_hash(name, namelen); + bucket = dlm_master_hash(dlm, hash); + hlist_for_each_entry(tmpmle, bucket, master_hash_node) { if (!dlm_mle_equal(dlm, tmpmle, name, namelen)) continue; dlm_get_mle(tmpmle); @@ -414,13 +361,10 @@ static int dlm_find_mle(struct dlm_ctxt *dlm, void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up) { struct dlm_master_list_entry *mle; - struct list_head *iter; assert_spin_locked(&dlm->spinlock); - - list_for_each(iter, &dlm->mle_hb_events) { - mle = list_entry(iter, struct dlm_master_list_entry, - hb_events); + + list_for_each_entry(mle, &dlm->mle_hb_events, hb_events) { if (node_up) dlm_mle_node_up(dlm, mle, NULL, idx); else @@ -459,10 +403,10 @@ static void dlm_mle_node_up(struct dlm_ctxt *dlm, int dlm_init_mle_cache(void) { - dlm_mle_cache = kmem_cache_create("dlm_mle_cache", + dlm_mle_cache = kmem_cache_create("o2dlm_mle", sizeof(struct dlm_master_list_entry), 0, SLAB_HWCACHE_ALIGN, - NULL, NULL); + NULL); if (dlm_mle_cache == NULL) return -ENOMEM; return 0; @@ -479,29 +423,23 @@ static void dlm_mle_release(struct kref *kref) struct dlm_master_list_entry *mle; struct dlm_ctxt *dlm; - mlog_entry_void(); - mle = container_of(kref, struct dlm_master_list_entry, mle_refs); dlm = mle->dlm; - if (mle->type != DLM_MLE_MASTER) { - mlog(0, "calling mle_release for %.*s, type %d\n", - mle->u.name.len, mle->u.name.name, mle->type); - } else { - mlog(0, "calling mle_release for %.*s, type %d\n", - mle->u.res->lockname.len, - mle->u.res->lockname.name, mle->type); - } assert_spin_locked(&dlm->spinlock); assert_spin_locked(&dlm->master_lock); + mlog(0, "Releasing mle for %.*s, type %d\n", mle->mnamelen, mle->mname, + mle->type); + /* remove from list if not already */ - if (!list_empty(&mle->list)) - list_del_init(&mle->list); + __dlm_unlink_mle(dlm, mle); /* detach the mle from the domain node up/down events */ __dlm_mle_detach_hb_events(dlm, mle); + atomic_dec(&dlm->mle_cur_count[mle->type]); + /* NOTE: kfree under spinlock here. * if this is bad, we can move this to a freelist. */ kmem_cache_free(dlm_mle_cache, mle); @@ -512,48 +450,46 @@ static void dlm_mle_release(struct kref *kref) * LOCK RESOURCE FUNCTIONS */ -static void dlm_set_lockres_owner(struct dlm_ctxt *dlm, - struct dlm_lock_resource *res, - u8 owner) +int dlm_init_master_caches(void) { - assert_spin_locked(&res->spinlock); - - mlog_entry("%.*s, %u\n", res->lockname.len, res->lockname.name, owner); - - if (owner == dlm->node_num) - atomic_inc(&dlm->local_resources); - else if (owner == DLM_LOCK_RES_OWNER_UNKNOWN) - atomic_inc(&dlm->unknown_resources); - else - atomic_inc(&dlm->remote_resources); + dlm_lockres_cache = kmem_cache_create("o2dlm_lockres", + sizeof(struct dlm_lock_resource), + 0, SLAB_HWCACHE_ALIGN, NULL); + if (!dlm_lockres_cache) + goto bail; + + dlm_lockname_cache = kmem_cache_create("o2dlm_lockname", + DLM_LOCKID_NAME_MAX, 0, + SLAB_HWCACHE_ALIGN, NULL); + if (!dlm_lockname_cache) + goto bail; - res->owner = owner; + return 0; +bail: + dlm_destroy_master_caches(); + return -ENOMEM; } -void dlm_change_lockres_owner(struct dlm_ctxt *dlm, - struct dlm_lock_resource *res, u8 owner) +void dlm_destroy_master_caches(void) { - assert_spin_locked(&res->spinlock); - - if (owner == res->owner) - return; - - if (res->owner == dlm->node_num) - atomic_dec(&dlm->local_resources); - else if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN) - atomic_dec(&dlm->unknown_resources); - else - atomic_dec(&dlm->remote_resources); + if (dlm_lockname_cache) { + kmem_cache_destroy(dlm_lockname_cache); + dlm_lockname_cache = NULL; + } - dlm_set_lockres_owner(dlm, res, owner); + if (dlm_lockres_cache) { + kmem_cache_destroy(dlm_lockres_cache); + dlm_lockres_cache = NULL; + } } - static void dlm_lockres_release(struct kref *kref) { struct dlm_lock_resource *res; + struct dlm_ctxt *dlm; res = container_of(kref, struct dlm_lock_resource, refs); + dlm = res->dlm; /* This should not happen -- all lockres' have a name * associated with them at init time. */ @@ -562,9 +498,43 @@ static void dlm_lockres_release(struct kref *kref) mlog(0, "destroying lockres %.*s\n", res->lockname.len, res->lockname.name); + spin_lock(&dlm->track_lock); + if (!list_empty(&res->tracking)) + list_del_init(&res->tracking); + else { + mlog(ML_ERROR, "Resource %.*s not on the Tracking list\n", + res->lockname.len, res->lockname.name); + dlm_print_one_lock_resource(res); + } + spin_unlock(&dlm->track_lock); + + atomic_dec(&dlm->res_cur_count); + + if (!hlist_unhashed(&res->hash_node) || + !list_empty(&res->granted) || + !list_empty(&res->converting) || + !list_empty(&res->blocked) || + !list_empty(&res->dirty) || + !list_empty(&res->recovering) || + !list_empty(&res->purge)) { + mlog(ML_ERROR, + "Going to BUG for resource %.*s." + " We're on a list! [%c%c%c%c%c%c%c]\n", + res->lockname.len, res->lockname.name, + !hlist_unhashed(&res->hash_node) ? 'H' : ' ', + !list_empty(&res->granted) ? 'G' : ' ', + !list_empty(&res->converting) ? 'C' : ' ', + !list_empty(&res->blocked) ? 'B' : ' ', + !list_empty(&res->dirty) ? 'D' : ' ', + !list_empty(&res->recovering) ? 'R' : ' ', + !list_empty(&res->purge) ? 'P' : ' '); + + dlm_print_one_lock_resource(res); + } + /* By the time we're ready to blow this guy away, we shouldn't * be on any lists. */ - BUG_ON(!list_empty(&res->list)); + BUG_ON(!hlist_unhashed(&res->hash_node)); BUG_ON(!list_empty(&res->granted)); BUG_ON(!list_empty(&res->converting)); BUG_ON(!list_empty(&res->blocked)); @@ -572,14 +542,9 @@ static void dlm_lockres_release(struct kref *kref) BUG_ON(!list_empty(&res->recovering)); BUG_ON(!list_empty(&res->purge)); - kfree(res->lockname.name); + kmem_cache_free(dlm_lockname_cache, (void *)res->lockname.name); - kfree(res); -} - -void dlm_lockres_get(struct dlm_lock_resource *res) -{ - kref_get(&res->refs); + kmem_cache_free(dlm_lockres_cache, res); } void dlm_lockres_put(struct dlm_lock_resource *res) @@ -601,22 +566,30 @@ static void dlm_init_lockres(struct dlm_ctxt *dlm, memcpy(qname, name, namelen); res->lockname.len = namelen; - res->lockname.hash = full_name_hash(name, namelen); + res->lockname.hash = dlm_lockid_hash(name, namelen); init_waitqueue_head(&res->wq); spin_lock_init(&res->spinlock); - INIT_LIST_HEAD(&res->list); + INIT_HLIST_NODE(&res->hash_node); INIT_LIST_HEAD(&res->granted); INIT_LIST_HEAD(&res->converting); INIT_LIST_HEAD(&res->blocked); INIT_LIST_HEAD(&res->dirty); INIT_LIST_HEAD(&res->recovering); INIT_LIST_HEAD(&res->purge); + INIT_LIST_HEAD(&res->tracking); atomic_set(&res->asts_reserved, 0); res->migration_pending = 0; + res->inflight_locks = 0; + res->inflight_assert_workers = 0; + + res->dlm = dlm; kref_init(&res->refs); + atomic_inc(&dlm->res_tot_count); + atomic_inc(&dlm->res_cur_count); + /* just for consistency */ spin_lock(&res->spinlock); dlm_set_lockres_owner(dlm, res, DLM_LOCK_RES_OWNER_UNKNOWN); @@ -626,27 +599,126 @@ static void dlm_init_lockres(struct dlm_ctxt *dlm, res->last_used = 0; + spin_lock(&dlm->spinlock); + list_add_tail(&res->tracking, &dlm->tracking_list); + spin_unlock(&dlm->spinlock); + memset(res->lvb, 0, DLM_LVB_LEN); + memset(res->refmap, 0, sizeof(res->refmap)); } struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm, const char *name, unsigned int namelen) { - struct dlm_lock_resource *res; + struct dlm_lock_resource *res = NULL; - res = kmalloc(sizeof(struct dlm_lock_resource), GFP_KERNEL); + res = kmem_cache_zalloc(dlm_lockres_cache, GFP_NOFS); if (!res) - return NULL; + goto error; - res->lockname.name = kmalloc(namelen, GFP_KERNEL); - if (!res->lockname.name) { - kfree(res); - return NULL; - } + res->lockname.name = kmem_cache_zalloc(dlm_lockname_cache, GFP_NOFS); + if (!res->lockname.name) + goto error; dlm_init_lockres(dlm, res, name, namelen); return res; + +error: + if (res && res->lockname.name) + kmem_cache_free(dlm_lockname_cache, (void *)res->lockname.name); + + if (res) + kmem_cache_free(dlm_lockres_cache, res); + return NULL; +} + +void dlm_lockres_set_refmap_bit(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res, int bit) +{ + assert_spin_locked(&res->spinlock); + + mlog(0, "res %.*s, set node %u, %ps()\n", res->lockname.len, + res->lockname.name, bit, __builtin_return_address(0)); + + set_bit(bit, res->refmap); +} + +void dlm_lockres_clear_refmap_bit(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res, int bit) +{ + assert_spin_locked(&res->spinlock); + + mlog(0, "res %.*s, clr node %u, %ps()\n", res->lockname.len, + res->lockname.name, bit, __builtin_return_address(0)); + + clear_bit(bit, res->refmap); +} + + +void dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res) +{ + assert_spin_locked(&res->spinlock); + + res->inflight_locks++; + + mlog(0, "%s: res %.*s, inflight++: now %u, %ps()\n", dlm->name, + res->lockname.len, res->lockname.name, res->inflight_locks, + __builtin_return_address(0)); +} + +void dlm_lockres_drop_inflight_ref(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res) +{ + assert_spin_locked(&res->spinlock); + + BUG_ON(res->inflight_locks == 0); + + res->inflight_locks--; + + mlog(0, "%s: res %.*s, inflight--: now %u, %ps()\n", dlm->name, + res->lockname.len, res->lockname.name, res->inflight_locks, + __builtin_return_address(0)); + + wake_up(&res->wq); +} + +void __dlm_lockres_grab_inflight_worker(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res) +{ + assert_spin_locked(&res->spinlock); + res->inflight_assert_workers++; + mlog(0, "%s:%.*s: inflight assert worker++: now %u\n", + dlm->name, res->lockname.len, res->lockname.name, + res->inflight_assert_workers); +} + +static void dlm_lockres_grab_inflight_worker(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res) +{ + spin_lock(&res->spinlock); + __dlm_lockres_grab_inflight_worker(dlm, res); + spin_unlock(&res->spinlock); +} + +static void __dlm_lockres_drop_inflight_worker(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res) +{ + assert_spin_locked(&res->spinlock); + BUG_ON(res->inflight_assert_workers == 0); + res->inflight_assert_workers--; + mlog(0, "%s:%.*s: inflight assert worker--: now %u\n", + dlm->name, res->lockname.len, res->lockname.name, + res->inflight_assert_workers); +} + +static void dlm_lockres_drop_inflight_worker(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res) +{ + spin_lock(&res->spinlock); + __dlm_lockres_drop_inflight_worker(dlm, res); + spin_unlock(&res->spinlock); } /* @@ -667,6 +739,7 @@ struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm, */ struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm, const char *lockid, + int namelen, int flags) { struct dlm_lock_resource *tmpres=NULL, *res=NULL; @@ -675,21 +748,47 @@ struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm, int blocked = 0; int ret, nodenum; struct dlm_node_iter iter; - unsigned int namelen; + unsigned int hash; int tries = 0; + int bit, wait_on_recovery = 0; BUG_ON(!lockid); - namelen = strlen(lockid); + hash = dlm_lockid_hash(lockid, namelen); mlog(0, "get lockres %s (len %d)\n", lockid, namelen); lookup: spin_lock(&dlm->spinlock); - tmpres = __dlm_lookup_lockres(dlm, lockid, namelen); + tmpres = __dlm_lookup_lockres_full(dlm, lockid, namelen, hash); if (tmpres) { spin_unlock(&dlm->spinlock); - mlog(0, "found in hash!\n"); + spin_lock(&tmpres->spinlock); + /* Wait on the thread that is mastering the resource */ + if (tmpres->owner == DLM_LOCK_RES_OWNER_UNKNOWN) { + __dlm_wait_on_lockres(tmpres); + BUG_ON(tmpres->owner == DLM_LOCK_RES_OWNER_UNKNOWN); + spin_unlock(&tmpres->spinlock); + dlm_lockres_put(tmpres); + tmpres = NULL; + goto lookup; + } + + /* Wait on the resource purge to complete before continuing */ + if (tmpres->state & DLM_LOCK_RES_DROPPING_REF) { + BUG_ON(tmpres->owner == dlm->node_num); + __dlm_wait_on_lockres_flags(tmpres, + DLM_LOCK_RES_DROPPING_REF); + spin_unlock(&tmpres->spinlock); + dlm_lockres_put(tmpres); + tmpres = NULL; + goto lookup; + } + + /* Grab inflight ref to pin the resource */ + dlm_lockres_grab_inflight_ref(dlm, tmpres); + + spin_unlock(&tmpres->spinlock); if (res) dlm_lockres_put(res); res = tmpres; @@ -700,8 +799,7 @@ lookup: spin_unlock(&dlm->spinlock); mlog(0, "allocating a new resource\n"); /* nothing found and we need to allocate one. */ - alloc_mle = (struct dlm_master_list_entry *) - kmem_cache_alloc(dlm_mle_cache, GFP_KERNEL); + alloc_mle = kmem_cache_alloc(dlm_mle_cache, GFP_NOFS); if (!alloc_mle) goto leave; res = dlm_new_lockres(dlm, lockid, namelen); @@ -718,6 +816,7 @@ lookup: spin_lock(&res->spinlock); dlm_change_lockres_owner(dlm, res, dlm->node_num); __dlm_insert_lockres(dlm, res); + dlm_lockres_grab_inflight_ref(dlm, res); spin_unlock(&res->spinlock); spin_unlock(&dlm->spinlock); /* lockres still marked IN_PROGRESS */ @@ -730,29 +829,40 @@ lookup: /* if we found a block, wait for lock to be mastered by another node */ blocked = dlm_find_mle(dlm, &mle, (char *)lockid, namelen); if (blocked) { + int mig; if (mle->type == DLM_MLE_MASTER) { mlog(ML_ERROR, "master entry for nonexistent lock!\n"); BUG(); - } else if (mle->type == DLM_MLE_MIGRATION) { - /* migration is in progress! */ - /* the good news is that we now know the - * "current" master (mle->master). */ - + } + mig = (mle->type == DLM_MLE_MIGRATION); + /* if there is a migration in progress, let the migration + * finish before continuing. we can wait for the absence + * of the MIGRATION mle: either the migrate finished or + * one of the nodes died and the mle was cleaned up. + * if there is a BLOCK here, but it already has a master + * set, we are too late. the master does not have a ref + * for us in the refmap. detach the mle and drop it. + * either way, go back to the top and start over. */ + if (mig || mle->master != O2NM_MAX_NODES) { + BUG_ON(mig && mle->master == dlm->node_num); + /* we arrived too late. the master does not + * have a ref for us. retry. */ + mlog(0, "%s:%.*s: late on %s\n", + dlm->name, namelen, lockid, + mig ? "MIGRATION" : "BLOCK"); spin_unlock(&dlm->master_lock); - assert_spin_locked(&dlm->spinlock); - - /* set the lockres owner and hash it */ - spin_lock(&res->spinlock); - dlm_set_lockres_owner(dlm, res, mle->master); - __dlm_insert_lockres(dlm, res); - spin_unlock(&res->spinlock); spin_unlock(&dlm->spinlock); /* master is known, detach */ - dlm_mle_detach_hb_events(dlm, mle); + if (!mig) + dlm_mle_detach_hb_events(dlm, mle); dlm_put_mle(mle); mle = NULL; - goto wake_waiters; + /* this is lame, but we can't wait on either + * the mle or lockres waitqueue here */ + if (mig) + msleep(100); + goto lookup; } } else { /* go ahead and try to master lock on this node */ @@ -761,7 +871,19 @@ lookup: alloc_mle = NULL; dlm_init_mle(mle, DLM_MLE_MASTER, dlm, res, NULL, 0); set_bit(dlm->node_num, mle->maybe_map); - list_add(&mle->list, &dlm->master_list); + __dlm_insert_mle(dlm, mle); + + /* still holding the dlm spinlock, check the recovery map + * to see if there are any nodes that still need to be + * considered. these will not appear in the mle nodemap + * but they might own this lockres. wait on them. */ + bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0); + if (bit < O2NM_MAX_NODES) { + mlog(0, "%s: res %.*s, At least one node (%d) " + "to recover before lock mastery can begin\n", + dlm->name, namelen, (char *)lockid, bit); + wait_on_recovery = 1; + } } /* at this point there is either a DLM_MLE_BLOCK or a @@ -771,28 +893,78 @@ lookup: /* finally add the lockres to its hash bucket */ __dlm_insert_lockres(dlm, res); + + /* Grab inflight ref to pin the resource */ + spin_lock(&res->spinlock); + dlm_lockres_grab_inflight_ref(dlm, res); + spin_unlock(&res->spinlock); + /* get an extra ref on the mle in case this is a BLOCK * if so, the creator of the BLOCK may try to put the last * ref at this time in the assert master handler, so we * need an extra one to keep from a bad ptr deref. */ - dlm_get_mle(mle); + dlm_get_mle_inuse(mle); spin_unlock(&dlm->master_lock); spin_unlock(&dlm->spinlock); +redo_request: + while (wait_on_recovery) { + /* any cluster changes that occurred after dropping the + * dlm spinlock would be detectable be a change on the mle, + * so we only need to clear out the recovery map once. */ + if (dlm_is_recovery_lock(lockid, namelen)) { + mlog(0, "%s: Recovery map is not empty, but must " + "master $RECOVERY lock now\n", dlm->name); + if (!dlm_pre_master_reco_lockres(dlm, res)) + wait_on_recovery = 0; + else { + mlog(0, "%s: waiting 500ms for heartbeat state " + "change\n", dlm->name); + msleep(500); + } + continue; + } + + dlm_kick_recovery_thread(dlm); + msleep(1000); + dlm_wait_for_recovery(dlm); + + spin_lock(&dlm->spinlock); + bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0); + if (bit < O2NM_MAX_NODES) { + mlog(0, "%s: res %.*s, At least one node (%d) " + "to recover before lock mastery can begin\n", + dlm->name, namelen, (char *)lockid, bit); + wait_on_recovery = 1; + } else + wait_on_recovery = 0; + spin_unlock(&dlm->spinlock); + + if (wait_on_recovery) + dlm_wait_for_node_recovery(dlm, bit, 10000); + } + /* must wait for lock to be mastered elsewhere */ if (blocked) goto wait; -redo_request: ret = -EINVAL; dlm_node_iter_init(mle->vote_map, &iter); while ((nodenum = dlm_node_iter_next(&iter)) >= 0) { - ret = dlm_do_master_request(mle, nodenum); + ret = dlm_do_master_request(res, mle, nodenum); if (ret < 0) mlog_errno(ret); if (mle->master != O2NM_MAX_NODES) { /* found a master ! */ - break; + if (mle->master <= nodenum) + break; + /* if our master request has not reached the master + * yet, keep going until it does. this is how the + * master will know that asserts are needed back to + * the lower nodes. */ + mlog(0, "%s: res %.*s, Requests only up to %u but " + "master is %u, keep going\n", dlm->name, namelen, + lockid, nodenum, mle->master); } } @@ -800,23 +972,24 @@ wait: /* keep going until the response map includes all nodes */ ret = dlm_wait_for_lock_mastery(dlm, res, mle, &blocked); if (ret < 0) { - mlog(0, "%s:%.*s: node map changed, redo the " - "master request now, blocked=%d\n", - dlm->name, res->lockname.len, + wait_on_recovery = 1; + mlog(0, "%s: res %.*s, Node map changed, redo the master " + "request now, blocked=%d\n", dlm->name, res->lockname.len, res->lockname.name, blocked); if (++tries > 20) { - mlog(ML_ERROR, "%s:%.*s: spinning on " - "dlm_wait_for_lock_mastery, blocked=%d\n", - dlm->name, res->lockname.len, + mlog(ML_ERROR, "%s: res %.*s, Spinning on " + "dlm_wait_for_lock_mastery, blocked = %d\n", + dlm->name, res->lockname.len, res->lockname.name, blocked); dlm_print_one_lock_resource(res); - /* dlm_print_one_mle(mle); */ + dlm_print_one_mle(mle); tries = 0; } goto redo_request; } - mlog(0, "lockres mastered by %u\n", res->owner); + mlog(0, "%s: res %.*s, Mastered by %u\n", dlm->name, res->lockname.len, + res->lockname.name, res->owner); /* make sure we never continue without this */ BUG_ON(res->owner == O2NM_MAX_NODES); @@ -824,7 +997,7 @@ wait: dlm_mle_detach_hb_events(dlm, mle); dlm_put_mle(mle); /* put the extra ref */ - dlm_put_mle(mle); + dlm_put_mle_inuse(mle); wake_waiters: spin_lock(&res->spinlock); @@ -860,7 +1033,21 @@ recheck: /* check if another node has already become the owner */ spin_lock(&res->spinlock); if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) { + mlog(0, "%s:%.*s: owner is suddenly %u\n", dlm->name, + res->lockname.len, res->lockname.name, res->owner); spin_unlock(&res->spinlock); + /* this will cause the master to re-assert across + * the whole cluster, freeing up mles */ + if (res->owner != dlm->node_num) { + ret = dlm_do_master_request(res, mle, res->owner); + if (ret < 0) { + /* give recovery a chance to run */ + mlog(ML_ERROR, "link to %u went down?: %d\n", res->owner, ret); + msleep(500); + goto recheck; + } + } + ret = 0; goto leave; } spin_unlock(&res->spinlock); @@ -880,7 +1067,7 @@ recheck: ret = dlm_restart_lock_mastery(dlm, res, mle, *blocked); b = (mle->type == DLM_MLE_BLOCK); if ((*blocked && !b) || (!*blocked && b)) { - mlog(0, "%s:%.*s: status change: old=%d new=%d\n", + mlog(0, "%s:%.*s: status change: old=%d new=%d\n", dlm->name, res->lockname.len, res->lockname.name, *blocked, b); *blocked = b; @@ -894,6 +1081,12 @@ recheck: "rechecking now\n", dlm->name, res->lockname.len, res->lockname.name); goto recheck; + } else { + if (!voting_done) { + mlog(0, "map not changed and voting not done " + "for %s:%.*s\n", dlm->name, res->lockname.len, + res->lockname.name); + } } if (m != O2NM_MAX_NODES) { @@ -910,6 +1103,8 @@ recheck: * now tell other nodes that I am * mastering this. */ mle->master = dlm->node_num; + /* ref was grabbed in get_lock_resource + * will be dropped in dlmlock_master */ assert = 1; sleep = 0; } @@ -935,7 +1130,8 @@ recheck: (atomic_read(&mle->woken) == 1), timeo); if (res->owner == O2NM_MAX_NODES) { - mlog(0, "waiting again\n"); + mlog(0, "%s:%.*s: waiting again\n", dlm->name, + res->lockname.len, res->lockname.name); goto recheck; } mlog(0, "done waiting, master is %u\n", res->owner); @@ -948,8 +1144,7 @@ recheck: m = dlm->node_num; mlog(0, "about to master %.*s here, this=%u\n", res->lockname.len, res->lockname.name, m); - ret = dlm_do_assert_master(dlm, res->lockname.name, - res->lockname.len, mle->vote_map, 0); + ret = dlm_do_assert_master(dlm, res, mle->vote_map, 0); if (ret) { /* This is a failure in the network path, * not in the response to the assert_master @@ -965,6 +1160,8 @@ recheck: /* set the lockres owner */ spin_lock(&res->spinlock); + /* mastery reference obtained either during + * assert_master_handler or in get_lock_resource */ dlm_change_lockres_owner(dlm, res, m); spin_unlock(&res->spinlock); @@ -1050,17 +1247,10 @@ static int dlm_restart_lock_mastery(struct dlm_ctxt *dlm, node = dlm_bitmap_diff_iter_next(&bdi, &sc); while (node >= 0) { if (sc == NODE_UP) { - /* a node came up. easy. might not even need - * to talk to it if its node number is higher - * or if we are already blocked. */ - mlog(0, "node up! %d\n", node); - if (blocked) - goto next; - - if (node > dlm->node_num) { - mlog(0, "node > this node. skipping.\n"); - goto next; - } + /* a node came up. clear any old vote from + * the response map and set it in the vote map + * then restart the mastery. */ + mlog(ML_NOTICE, "node %d up while restarting\n", node); /* redo the master request, but only for the new node */ mlog(0, "sending request to new node\n"); @@ -1068,18 +1258,6 @@ static int dlm_restart_lock_mastery(struct dlm_ctxt *dlm, set_bit(node, mle->vote_map); } else { mlog(ML_ERROR, "node down! %d\n", node); - - /* if the node wasn't involved in mastery skip it, - * but clear it out from the maps so that it will - * not affect mastery of this lockres */ - clear_bit(node, mle->response_map); - clear_bit(node, mle->vote_map); - if (!test_bit(node, mle->maybe_map)) - goto next; - - /* if we're already blocked on lock mastery, and the - * dead node wasn't the expected master, or there is - * another node in the maybe_map, keep waiting */ if (blocked) { int lowest = find_next_bit(mle->maybe_map, O2NM_MAX_NODES, 0); @@ -1087,54 +1265,53 @@ static int dlm_restart_lock_mastery(struct dlm_ctxt *dlm, /* act like it was never there */ clear_bit(node, mle->maybe_map); - if (node != lowest) - goto next; - - mlog(ML_ERROR, "expected master %u died while " - "this node was blocked waiting on it!\n", - node); - lowest = find_next_bit(mle->maybe_map, - O2NM_MAX_NODES, - lowest+1); - if (lowest < O2NM_MAX_NODES) { - mlog(0, "still blocked. waiting " - "on %u now\n", lowest); - goto next; + if (node == lowest) { + mlog(0, "expected master %u died" + " while this node was blocked " + "waiting on it!\n", node); + lowest = find_next_bit(mle->maybe_map, + O2NM_MAX_NODES, + lowest+1); + if (lowest < O2NM_MAX_NODES) { + mlog(0, "%s:%.*s:still " + "blocked. waiting on %u " + "now\n", dlm->name, + res->lockname.len, + res->lockname.name, + lowest); + } else { + /* mle is an MLE_BLOCK, but + * there is now nothing left to + * block on. we need to return + * all the way back out and try + * again with an MLE_MASTER. + * dlm_do_local_recovery_cleanup + * has already run, so the mle + * refcount is ok */ + mlog(0, "%s:%.*s: no " + "longer blocking. try to " + "master this here\n", + dlm->name, + res->lockname.len, + res->lockname.name); + mle->type = DLM_MLE_MASTER; + mle->mleres = res; + } } - - /* mle is an MLE_BLOCK, but there is now - * nothing left to block on. we need to return - * all the way back out and try again with - * an MLE_MASTER. dlm_do_local_recovery_cleanup - * has already run, so the mle refcount is ok */ - mlog(0, "no longer blocking. we can " - "try to master this here\n"); - mle->type = DLM_MLE_MASTER; - memset(mle->maybe_map, 0, - sizeof(mle->maybe_map)); - memset(mle->response_map, 0, - sizeof(mle->maybe_map)); - memcpy(mle->vote_map, mle->node_map, - sizeof(mle->node_map)); - mle->u.res = res; - set_bit(dlm->node_num, mle->maybe_map); - - ret = -EAGAIN; - goto next; } - clear_bit(node, mle->maybe_map); - if (node > dlm->node_num) - goto next; - - mlog(0, "dead node in map!\n"); - /* yuck. go back and re-contact all nodes - * in the vote_map, removing this node. */ - memset(mle->response_map, 0, - sizeof(mle->response_map)); + /* now blank out everything, as if we had never + * contacted anyone */ + memset(mle->maybe_map, 0, sizeof(mle->maybe_map)); + memset(mle->response_map, 0, sizeof(mle->response_map)); + /* reset the vote_map to the current node_map */ + memcpy(mle->vote_map, mle->node_map, + sizeof(mle->node_map)); + /* put myself into the maybe map */ + if (mle->type != DLM_MLE_BLOCK) + set_bit(dlm->node_num, mle->maybe_map); } ret = -EAGAIN; -next: node = dlm_bitmap_diff_iter_next(&bdi, &sc); } return ret; @@ -1151,7 +1328,8 @@ next: * */ -static int dlm_do_master_request(struct dlm_master_list_entry *mle, int to) +static int dlm_do_master_request(struct dlm_lock_resource *res, + struct dlm_master_list_entry *mle, int to) { struct dlm_ctxt *dlm = mle->dlm; struct dlm_master_request request; @@ -1162,14 +1340,8 @@ static int dlm_do_master_request(struct dlm_master_list_entry *mle, int to) BUG_ON(mle->type == DLM_MLE_MIGRATION); - if (mle->type != DLM_MLE_MASTER) { - request.namelen = mle->u.name.len; - memcpy(request.name, mle->u.name.name, request.namelen); - } else { - request.namelen = mle->u.res->lockname.len; - memcpy(request.name, mle->u.res->lockname.name, - request.namelen); - } + request.namelen = (u8)mle->mnamelen; + memcpy(request.name, mle->mname, request.namelen); again: ret = o2net_send_message(DLM_MASTER_REQUEST_MSG, dlm->key, &request, @@ -1207,6 +1379,9 @@ again: case DLM_MASTER_RESP_YES: set_bit(to, mle->response_map); mlog(0, "node %u is the master, response=YES\n", to); + mlog(0, "%s:%.*s: master node %u now knows I have a " + "reference\n", dlm->name, res->lockname.len, + res->lockname.name, to); mle->master = to; break; case DLM_MASTER_RESP_NO: @@ -1247,17 +1422,19 @@ out: * * if possible, TRIM THIS DOWN!!! */ -int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data) +int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { u8 response = DLM_MASTER_RESP_MAYBE; struct dlm_ctxt *dlm = data; - struct dlm_lock_resource *res; + struct dlm_lock_resource *res = NULL; struct dlm_master_request *request = (struct dlm_master_request *) msg->buf; struct dlm_master_list_entry *mle = NULL, *tmpmle = NULL; char *name; - unsigned int namelen; + unsigned int namelen, hash; int found, ret; int set_maybe; + int dispatch_assert = 0; if (!dlm_grab(dlm)) return DLM_MASTER_RESP_NO; @@ -1269,6 +1446,7 @@ int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data) name = request->name; namelen = request->namelen; + hash = dlm_lockid_hash(name, namelen); if (namelen > DLM_LOCKID_NAME_MAX) { response = DLM_IVBUFLEN; @@ -1277,16 +1455,17 @@ int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data) way_up_top: spin_lock(&dlm->spinlock); - res = __dlm_lookup_lockres(dlm, name, namelen); + res = __dlm_lookup_lockres(dlm, name, namelen, hash); if (res) { spin_unlock(&dlm->spinlock); /* take care of the easy cases up front */ spin_lock(&res->spinlock); - if (res->state & DLM_LOCK_RES_RECOVERING) { + if (res->state & (DLM_LOCK_RES_RECOVERING| + DLM_LOCK_RES_MIGRATING)) { spin_unlock(&res->spinlock); mlog(0, "returning DLM_MASTER_RESP_ERROR since res is " - "being recovered\n"); + "being recovered/migrated\n"); response = DLM_MASTER_RESP_ERROR; if (mle) kmem_cache_free(dlm_mle_cache, mle); @@ -1294,9 +1473,8 @@ way_up_top: } if (res->owner == dlm->node_num) { - u32 flags = DLM_ASSERT_MASTER_MLE_CLEANUP; + dlm_lockres_set_refmap_bit(dlm, res, request->node_idx); spin_unlock(&res->spinlock); - // mlog(0, "this node is the master\n"); response = DLM_MASTER_RESP_YES; if (mle) kmem_cache_free(dlm_mle_cache, mle); @@ -1307,16 +1485,7 @@ way_up_top: * caused all nodes up to this one to * create mles. this node now needs to * go back and clean those up. */ - mlog(0, "%u is the owner of %.*s, cleaning everyone else\n", - dlm->node_num, res->lockname.len, res->lockname.name); - ret = dlm_dispatch_assert_master(dlm, res, 1, - request->node_idx, - flags); - if (ret < 0) { - mlog(ML_ERROR, "failed to dispatch assert " - "master work\n"); - response = DLM_MASTER_RESP_ERROR; - } + dispatch_assert = 1; goto send_response; } else if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) { spin_unlock(&res->spinlock); @@ -1353,7 +1522,6 @@ way_up_top: mlog(0, "node %u is master, but trying to migrate to " "node %u.\n", tmpmle->master, tmpmle->new_master); if (tmpmle->master == dlm->node_num) { - response = DLM_MASTER_RESP_YES; mlog(ML_ERROR, "no owner on lockres, but this " "node is trying to migrate it to %u?!\n", tmpmle->new_master); @@ -1364,9 +1532,15 @@ way_up_top: } } else if (tmpmle->master != DLM_LOCK_RES_OWNER_UNKNOWN) { set_maybe = 0; - if (tmpmle->master == dlm->node_num) + if (tmpmle->master == dlm->node_num) { response = DLM_MASTER_RESP_YES; - else + /* this node will be the owner. + * go back and clean the mles on any + * other nodes */ + dispatch_assert = 1; + dlm_lockres_set_refmap_bit(dlm, res, + request->node_idx); + } else response = DLM_MASTER_RESP_NO; } else { // mlog(0, "this node is attempting to " @@ -1402,48 +1576,38 @@ way_up_top: spin_unlock(&dlm->master_lock); spin_unlock(&dlm->spinlock); - mle = (struct dlm_master_list_entry *) - kmem_cache_alloc(dlm_mle_cache, GFP_KERNEL); + mle = kmem_cache_alloc(dlm_mle_cache, GFP_NOFS); if (!mle) { - // bad bad bad... this sucks. response = DLM_MASTER_RESP_ERROR; + mlog_errno(-ENOMEM); goto send_response; } - spin_lock(&dlm->spinlock); - dlm_init_mle(mle, DLM_MLE_BLOCK, dlm, NULL, - name, namelen); - spin_unlock(&dlm->spinlock); goto way_up_top; } // mlog(0, "this is second time thru, already allocated, " // "add the block.\n"); + dlm_init_mle(mle, DLM_MLE_BLOCK, dlm, NULL, name, namelen); set_bit(request->node_idx, mle->maybe_map); - list_add(&mle->list, &dlm->master_list); + __dlm_insert_mle(dlm, mle); response = DLM_MASTER_RESP_NO; } else { // mlog(0, "mle was found\n"); set_maybe = 1; spin_lock(&tmpmle->spinlock); + if (tmpmle->master == dlm->node_num) { + mlog(ML_ERROR, "no lockres, but an mle with this node as master!\n"); + BUG(); + } if (tmpmle->type == DLM_MLE_BLOCK) response = DLM_MASTER_RESP_NO; else if (tmpmle->type == DLM_MLE_MIGRATION) { mlog(0, "migration mle was found (%u->%u)\n", tmpmle->master, tmpmle->new_master); - if (tmpmle->master == dlm->node_num) { - mlog(ML_ERROR, "no lockres, but migration mle " - "says that this node is master!\n"); - BUG(); - } /* real master can respond on its own */ response = DLM_MASTER_RESP_NO; - } else { - if (tmpmle->master == dlm->node_num) { - response = DLM_MASTER_RESP_YES; - set_maybe = 0; - } else - response = DLM_MASTER_RESP_MAYBE; - } + } else + response = DLM_MASTER_RESP_MAYBE; if (set_maybe) set_bit(request->node_idx, tmpmle->maybe_map); spin_unlock(&tmpmle->spinlock); @@ -1456,6 +1620,34 @@ way_up_top: dlm_put_mle(tmpmle); } send_response: + /* + * __dlm_lookup_lockres() grabbed a reference to this lockres. + * The reference is released by dlm_assert_master_worker() under + * the call to dlm_dispatch_assert_master(). If + * dlm_assert_master_worker() isn't called, we drop it here. + */ + if (dispatch_assert) { + if (response != DLM_MASTER_RESP_YES) + mlog(ML_ERROR, "invalid response %d\n", response); + if (!res) { + mlog(ML_ERROR, "bad lockres while trying to assert!\n"); + BUG(); + } + mlog(0, "%u is the owner of %.*s, cleaning everyone else\n", + dlm->node_num, res->lockname.len, res->lockname.name); + ret = dlm_dispatch_assert_master(dlm, res, 0, request->node_idx, + DLM_ASSERT_MASTER_MLE_CLEANUP); + if (ret < 0) { + mlog(ML_ERROR, "failed to dispatch assert master work\n"); + response = DLM_MASTER_RESP_ERROR; + dlm_lockres_put(res); + } else + dlm_lockres_grab_inflight_worker(dlm, res); + } else { + if (res) + dlm_lockres_put(res); + } + dlm_put(dlm); return response; } @@ -1470,21 +1662,33 @@ send_response: * can periodically run all locks owned by this node * and re-assert across the cluster... */ -static int dlm_do_assert_master(struct dlm_ctxt *dlm, const char *lockname, - unsigned int namelen, void *nodemap, - u32 flags) +static int dlm_do_assert_master(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res, + void *nodemap, u32 flags) { struct dlm_assert_master assert; int to, tmpret; struct dlm_node_iter iter; int ret = 0; + int reassert; + const char *lockname = res->lockname.name; + unsigned int namelen = res->lockname.len; BUG_ON(namelen > O2NM_MAX_NAME_LEN); + spin_lock(&res->spinlock); + res->state |= DLM_LOCK_RES_SETREF_INPROG; + spin_unlock(&res->spinlock); + +again: + reassert = 0; + /* note that if this nodemap is empty, it returns 0 */ dlm_node_iter_init(nodemap, &iter); while ((to = dlm_node_iter_next(&iter)) >= 0) { int r = 0; + struct dlm_master_list_entry *mle = NULL; + mlog(0, "sending assert master to %d (%.*s)\n", to, namelen, lockname); memset(&assert, 0, sizeof(assert)); @@ -1496,24 +1700,65 @@ static int dlm_do_assert_master(struct dlm_ctxt *dlm, const char *lockname, tmpret = o2net_send_message(DLM_ASSERT_MASTER_MSG, dlm->key, &assert, sizeof(assert), to, &r); if (tmpret < 0) { - mlog(ML_ERROR, "assert_master returned %d!\n", tmpret); + mlog(ML_ERROR, "Error %d when sending message %u (key " + "0x%x) to node %u\n", tmpret, + DLM_ASSERT_MASTER_MSG, dlm->key, to); if (!dlm_is_host_down(tmpret)) { - mlog(ML_ERROR, "unhandled error!\n"); + mlog(ML_ERROR, "unhandled error=%d!\n", tmpret); BUG(); } /* a node died. finish out the rest of the nodes. */ - mlog(ML_ERROR, "link to %d went down!\n", to); + mlog(0, "link to %d went down!\n", to); /* any nonzero status return will do */ ret = tmpret; + r = 0; } else if (r < 0) { /* ok, something horribly messed. kill thyself. */ mlog(ML_ERROR,"during assert master of %.*s to %u, " "got %d.\n", namelen, lockname, to, r); - dlm_dump_lock_resources(dlm); + spin_lock(&dlm->spinlock); + spin_lock(&dlm->master_lock); + if (dlm_find_mle(dlm, &mle, (char *)lockname, + namelen)) { + dlm_print_one_mle(mle); + __dlm_put_mle(mle); + } + spin_unlock(&dlm->master_lock); + spin_unlock(&dlm->spinlock); BUG(); } + + if (r & DLM_ASSERT_RESPONSE_REASSERT && + !(r & DLM_ASSERT_RESPONSE_MASTERY_REF)) { + mlog(ML_ERROR, "%.*s: very strange, " + "master MLE but no lockres on %u\n", + namelen, lockname, to); + } + + if (r & DLM_ASSERT_RESPONSE_REASSERT) { + mlog(0, "%.*s: node %u create mles on other " + "nodes and requests a re-assert\n", + namelen, lockname, to); + reassert = 1; + } + if (r & DLM_ASSERT_RESPONSE_MASTERY_REF) { + mlog(0, "%.*s: node %u has a reference to this " + "lockres, set the bit in the refmap\n", + namelen, lockname, to); + spin_lock(&res->spinlock); + dlm_lockres_set_refmap_bit(dlm, res, to); + spin_unlock(&res->spinlock); + } } + if (reassert) + goto again; + + spin_lock(&res->spinlock); + res->state &= ~DLM_LOCK_RES_SETREF_INPROG; + spin_unlock(&res->spinlock); + wake_up(&res->wq); + return ret; } @@ -1526,21 +1771,25 @@ static int dlm_do_assert_master(struct dlm_ctxt *dlm, const char *lockname, * * if possible, TRIM THIS DOWN!!! */ -int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data) +int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { struct dlm_ctxt *dlm = data; struct dlm_master_list_entry *mle = NULL; struct dlm_assert_master *assert = (struct dlm_assert_master *)msg->buf; struct dlm_lock_resource *res = NULL; char *name; - unsigned int namelen; + unsigned int namelen, hash; u32 flags; + int master_request = 0, have_lockres_ref = 0; + int ret = 0; if (!dlm_grab(dlm)) return 0; name = assert->name; namelen = assert->namelen; + hash = dlm_lockid_hash(name, namelen); flags = be32_to_cpu(assert->flags); if (namelen > DLM_LOCKID_NAME_MAX) { @@ -1565,7 +1814,7 @@ int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data) if (bit >= O2NM_MAX_NODES) { /* not necessarily an error, though less likely. * could be master just re-asserting. */ - mlog(ML_ERROR, "no bits set in the maybe_map, but %u " + mlog(0, "no bits set in the maybe_map, but %u " "is asserting! (%.*s)\n", assert->node_idx, namelen, name); } else if (bit != assert->node_idx) { @@ -1577,19 +1826,36 @@ int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data) * number winning the mastery will respond * YES to mastery requests, but this node * had no way of knowing. let it pass. */ - mlog(ML_ERROR, "%u is the lowest node, " + mlog(0, "%u is the lowest node, " "%u is asserting. (%.*s) %u must " "have begun after %u won.\n", bit, assert->node_idx, namelen, name, bit, assert->node_idx); } } + if (mle->type == DLM_MLE_MIGRATION) { + if (flags & DLM_ASSERT_MASTER_MLE_CLEANUP) { + mlog(0, "%s:%.*s: got cleanup assert" + " from %u for migration\n", + dlm->name, namelen, name, + assert->node_idx); + } else if (!(flags & DLM_ASSERT_MASTER_FINISH_MIGRATION)) { + mlog(0, "%s:%.*s: got unrelated assert" + " from %u for migration, ignoring\n", + dlm->name, namelen, name, + assert->node_idx); + __dlm_put_mle(mle); + spin_unlock(&dlm->master_lock); + spin_unlock(&dlm->spinlock); + goto done; + } + } } spin_unlock(&dlm->master_lock); /* ok everything checks out with the MLE * now check to see if there is a lockres */ - res = __dlm_lookup_lockres(dlm, name, namelen); + res = __dlm_lookup_lockres(dlm, name, namelen, hash); if (res) { spin_lock(&res->spinlock); if (res->state & DLM_LOCK_RES_RECOVERING) { @@ -1598,13 +1864,14 @@ int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data) goto kill; } if (!mle) { - if (res->owner != assert->node_idx) { - mlog(ML_ERROR, "assert_master from " - "%u, but current owner is " - "%u! (%.*s)\n", - assert->node_idx, res->owner, - namelen, name); - goto kill; + if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN && + res->owner != assert->node_idx) { + mlog(ML_ERROR, "DIE! Mastery assert from %u, " + "but current owner is %u! (%.*s)\n", + assert->node_idx, res->owner, namelen, + name); + __dlm_print_one_lock_resource(res); + BUG(); } } else if (mle->type != DLM_MLE_MIGRATION) { if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) { @@ -1644,69 +1911,158 @@ int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data) ok: spin_unlock(&res->spinlock); } - spin_unlock(&dlm->spinlock); // mlog(0, "woo! got an assert_master from node %u!\n", // assert->node_idx); if (mle) { - int extra_ref; - + int extra_ref = 0; + int nn = -1; + int rr, err = 0; + spin_lock(&mle->spinlock); - extra_ref = !!(mle->type == DLM_MLE_BLOCK - || mle->type == DLM_MLE_MIGRATION); + if (mle->type == DLM_MLE_BLOCK || mle->type == DLM_MLE_MIGRATION) + extra_ref = 1; + else { + /* MASTER mle: if any bits set in the response map + * then the calling node needs to re-assert to clear + * up nodes that this node contacted */ + while ((nn = find_next_bit (mle->response_map, O2NM_MAX_NODES, + nn+1)) < O2NM_MAX_NODES) { + if (nn != dlm->node_num && nn != assert->node_idx) { + master_request = 1; + break; + } + } + } mle->master = assert->node_idx; atomic_set(&mle->woken, 1); wake_up(&mle->wq); spin_unlock(&mle->spinlock); - if (mle->type == DLM_MLE_MIGRATION && res) { - mlog(0, "finishing off migration of lockres %.*s, " - "from %u to %u\n", - res->lockname.len, res->lockname.name, - dlm->node_num, mle->new_master); + if (res) { + int wake = 0; spin_lock(&res->spinlock); - res->state &= ~DLM_LOCK_RES_MIGRATING; - dlm_change_lockres_owner(dlm, res, mle->new_master); - BUG_ON(res->state & DLM_LOCK_RES_DIRTY); + if (mle->type == DLM_MLE_MIGRATION) { + mlog(0, "finishing off migration of lockres %.*s, " + "from %u to %u\n", + res->lockname.len, res->lockname.name, + dlm->node_num, mle->new_master); + res->state &= ~DLM_LOCK_RES_MIGRATING; + wake = 1; + dlm_change_lockres_owner(dlm, res, mle->new_master); + BUG_ON(res->state & DLM_LOCK_RES_DIRTY); + } else { + dlm_change_lockres_owner(dlm, res, mle->master); + } spin_unlock(&res->spinlock); + have_lockres_ref = 1; + if (wake) + wake_up(&res->wq); } - /* master is known, detach if not already detached */ - dlm_mle_detach_hb_events(dlm, mle); - dlm_put_mle(mle); - + + /* master is known, detach if not already detached. + * ensures that only one assert_master call will happen + * on this mle. */ + spin_lock(&dlm->master_lock); + + rr = atomic_read(&mle->mle_refs.refcount); + if (mle->inuse > 0) { + if (extra_ref && rr < 3) + err = 1; + else if (!extra_ref && rr < 2) + err = 1; + } else { + if (extra_ref && rr < 2) + err = 1; + else if (!extra_ref && rr < 1) + err = 1; + } + if (err) { + mlog(ML_ERROR, "%s:%.*s: got assert master from %u " + "that will mess up this node, refs=%d, extra=%d, " + "inuse=%d\n", dlm->name, namelen, name, + assert->node_idx, rr, extra_ref, mle->inuse); + dlm_print_one_mle(mle); + } + __dlm_unlink_mle(dlm, mle); + __dlm_mle_detach_hb_events(dlm, mle); + __dlm_put_mle(mle); if (extra_ref) { /* the assert master message now balances the extra * ref given by the master / migration request message. * if this is the last put, it will be removed * from the list. */ - dlm_put_mle(mle); + __dlm_put_mle(mle); + } + spin_unlock(&dlm->master_lock); + } else if (res) { + if (res->owner != assert->node_idx) { + mlog(0, "assert_master from %u, but current " + "owner is %u (%.*s), no mle\n", assert->node_idx, + res->owner, namelen, name); } } + spin_unlock(&dlm->spinlock); done: - if (res) - dlm_lockres_put(res); + ret = 0; + if (res) { + spin_lock(&res->spinlock); + res->state |= DLM_LOCK_RES_SETREF_INPROG; + spin_unlock(&res->spinlock); + *ret_data = (void *)res; + } dlm_put(dlm); - return 0; + if (master_request) { + mlog(0, "need to tell master to reassert\n"); + /* positive. negative would shoot down the node. */ + ret |= DLM_ASSERT_RESPONSE_REASSERT; + if (!have_lockres_ref) { + mlog(ML_ERROR, "strange, got assert from %u, MASTER " + "mle present here for %s:%.*s, but no lockres!\n", + assert->node_idx, dlm->name, namelen, name); + } + } + if (have_lockres_ref) { + /* let the master know we have a reference to the lockres */ + ret |= DLM_ASSERT_RESPONSE_MASTERY_REF; + mlog(0, "%s:%.*s: got assert from %u, need a ref\n", + dlm->name, namelen, name, assert->node_idx); + } + return ret; kill: /* kill the caller! */ - spin_unlock(&res->spinlock); - spin_unlock(&dlm->spinlock); - dlm_lockres_put(res); mlog(ML_ERROR, "Bad message received from another node. Dumping state " "and killing the other node now! This node is OK and can continue.\n"); - dlm_dump_lock_resources(dlm); + __dlm_print_one_lock_resource(res); + spin_unlock(&res->spinlock); + spin_unlock(&dlm->spinlock); + *ret_data = (void *)res; dlm_put(dlm); return -EINVAL; } +void dlm_assert_master_post_handler(int status, void *data, void *ret_data) +{ + struct dlm_lock_resource *res = (struct dlm_lock_resource *)ret_data; + + if (ret_data) { + spin_lock(&res->spinlock); + res->state &= ~DLM_LOCK_RES_SETREF_INPROG; + spin_unlock(&res->spinlock); + wake_up(&res->wq); + dlm_lockres_put(res); + } + return; +} + int dlm_dispatch_assert_master(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, int ignore_higher, u8 request_from, u32 flags) { struct dlm_work_item *item; - item = kcalloc(1, sizeof(*item), GFP_KERNEL); + item = kzalloc(sizeof(*item), GFP_ATOMIC); if (!item) return -ENOMEM; @@ -1720,11 +2076,15 @@ int dlm_dispatch_assert_master(struct dlm_ctxt *dlm, item->u.am.request_from = request_from; item->u.am.flags = flags; + if (ignore_higher) + mlog(0, "IGNORE HIGHER: %.*s\n", res->lockname.len, + res->lockname.name); + spin_lock(&dlm->work_lock); list_add_tail(&item->list, &dlm->work_list); spin_unlock(&dlm->work_lock); - schedule_work(&dlm->dispatched_work); + queue_work(dlm->dlm_worker, &dlm->dispatched_work); return 0; } @@ -1765,109 +2125,356 @@ static void dlm_assert_master_worker(struct dlm_work_item *item, void *data) } } + /* + * If we're migrating this lock to someone else, we are no + * longer allowed to assert out own mastery. OTOH, we need to + * prevent migration from starting while we're still asserting + * our dominance. The reserved ast delays migration. + */ + spin_lock(&res->spinlock); + if (res->state & DLM_LOCK_RES_MIGRATING) { + mlog(0, "Someone asked us to assert mastery, but we're " + "in the middle of migration. Skipping assert, " + "the new master will handle that.\n"); + spin_unlock(&res->spinlock); + goto put; + } else + __dlm_lockres_reserve_ast(res); + spin_unlock(&res->spinlock); + /* this call now finishes out the nodemap * even if one or more nodes die */ mlog(0, "worker about to master %.*s here, this=%u\n", res->lockname.len, res->lockname.name, dlm->node_num); - ret = dlm_do_assert_master(dlm, res->lockname.name, - res->lockname.len, - nodemap, flags); + ret = dlm_do_assert_master(dlm, res, nodemap, flags); if (ret < 0) { /* no need to restart, we are done */ - mlog_errno(ret); + if (!dlm_is_host_down(ret)) + mlog_errno(ret); } + /* Ok, we've asserted ourselves. Let's let migration start. */ + dlm_lockres_release_ast(dlm, res); + +put: + dlm_lockres_drop_inflight_worker(dlm, res); + dlm_lockres_put(res); mlog(0, "finished with dlm_assert_master_worker\n"); } +/* SPECIAL CASE for the $RECOVERY lock used by the recovery thread. + * We cannot wait for node recovery to complete to begin mastering this + * lockres because this lockres is used to kick off recovery! ;-) + * So, do a pre-check on all living nodes to see if any of those nodes + * think that $RECOVERY is currently mastered by a dead node. If so, + * we wait a short time to allow that node to get notified by its own + * heartbeat stack, then check again. All $RECOVERY lock resources + * mastered by dead nodes are purged when the hearbeat callback is + * fired, so we can know for sure that it is safe to continue once + * the node returns a live node or no node. */ +static int dlm_pre_master_reco_lockres(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res) +{ + struct dlm_node_iter iter; + int nodenum; + int ret = 0; + u8 master = DLM_LOCK_RES_OWNER_UNKNOWN; + + spin_lock(&dlm->spinlock); + dlm_node_iter_init(dlm->domain_map, &iter); + spin_unlock(&dlm->spinlock); + + while ((nodenum = dlm_node_iter_next(&iter)) >= 0) { + /* do not send to self */ + if (nodenum == dlm->node_num) + continue; + ret = dlm_do_master_requery(dlm, res, nodenum, &master); + if (ret < 0) { + mlog_errno(ret); + if (!dlm_is_host_down(ret)) + BUG(); + /* host is down, so answer for that node would be + * DLM_LOCK_RES_OWNER_UNKNOWN. continue. */ + ret = 0; + } + + if (master != DLM_LOCK_RES_OWNER_UNKNOWN) { + /* check to see if this master is in the recovery map */ + spin_lock(&dlm->spinlock); + if (test_bit(master, dlm->recovery_map)) { + mlog(ML_NOTICE, "%s: node %u has not seen " + "node %u go down yet, and thinks the " + "dead node is mastering the recovery " + "lock. must wait.\n", dlm->name, + nodenum, master); + ret = -EAGAIN; + } + spin_unlock(&dlm->spinlock); + mlog(0, "%s: reco lock master is %u\n", dlm->name, + master); + break; + } + } + return ret; +} /* - * DLM_MIGRATE_LOCKRES + * DLM_DEREF_LOCKRES_MSG */ +int dlm_drop_lockres_ref(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) +{ + struct dlm_deref_lockres deref; + int ret = 0, r; + const char *lockname; + unsigned int namelen; -int dlm_migrate_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, - u8 target) + lockname = res->lockname.name; + namelen = res->lockname.len; + BUG_ON(namelen > O2NM_MAX_NAME_LEN); + + memset(&deref, 0, sizeof(deref)); + deref.node_idx = dlm->node_num; + deref.namelen = namelen; + memcpy(deref.name, lockname, namelen); + + ret = o2net_send_message(DLM_DEREF_LOCKRES_MSG, dlm->key, + &deref, sizeof(deref), res->owner, &r); + if (ret < 0) + mlog(ML_ERROR, "%s: res %.*s, error %d send DEREF to node %u\n", + dlm->name, namelen, lockname, ret, res->owner); + else if (r < 0) { + /* BAD. other node says I did not have a ref. */ + mlog(ML_ERROR, "%s: res %.*s, DEREF to node %u got %d\n", + dlm->name, namelen, lockname, res->owner, r); + dlm_print_one_lock_resource(res); + BUG(); + } + return ret; +} + +int dlm_deref_lockres_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { - struct dlm_master_list_entry *mle = NULL; - struct dlm_master_list_entry *oldmle = NULL; - struct dlm_migratable_lockres *mres = NULL; - int ret = -EINVAL; - const char *name; + struct dlm_ctxt *dlm = data; + struct dlm_deref_lockres *deref = (struct dlm_deref_lockres *)msg->buf; + struct dlm_lock_resource *res = NULL; + char *name; unsigned int namelen; - int mle_added = 0; - struct list_head *queue, *iter; - int i; - struct dlm_lock *lock; - int empty = 1; + int ret = -EINVAL; + u8 node; + unsigned int hash; + struct dlm_work_item *item; + int cleared = 0; + int dispatch = 0; if (!dlm_grab(dlm)) - return -EINVAL; + return 0; - name = res->lockname.name; - namelen = res->lockname.len; + name = deref->name; + namelen = deref->namelen; + node = deref->node_idx; + + if (namelen > DLM_LOCKID_NAME_MAX) { + mlog(ML_ERROR, "Invalid name length!"); + goto done; + } + if (deref->node_idx >= O2NM_MAX_NODES) { + mlog(ML_ERROR, "Invalid node number: %u\n", node); + goto done; + } - mlog(0, "migrating %.*s to %u\n", namelen, name, target); + hash = dlm_lockid_hash(name, namelen); + + spin_lock(&dlm->spinlock); + res = __dlm_lookup_lockres_full(dlm, name, namelen, hash); + if (!res) { + spin_unlock(&dlm->spinlock); + mlog(ML_ERROR, "%s:%.*s: bad lockres name\n", + dlm->name, namelen, name); + goto done; + } + spin_unlock(&dlm->spinlock); - /* - * ensure this lockres is a proper candidate for migration - */ spin_lock(&res->spinlock); - if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN) { - mlog(0, "cannot migrate lockres with unknown owner!\n"); - spin_unlock(&res->spinlock); - goto leave; + if (res->state & DLM_LOCK_RES_SETREF_INPROG) + dispatch = 1; + else { + BUG_ON(res->state & DLM_LOCK_RES_DROPPING_REF); + if (test_bit(node, res->refmap)) { + dlm_lockres_clear_refmap_bit(dlm, res, node); + cleared = 1; + } } - if (res->owner != dlm->node_num) { - mlog(0, "cannot migrate lockres this node doesn't own!\n"); - spin_unlock(&res->spinlock); - goto leave; + spin_unlock(&res->spinlock); + + if (!dispatch) { + if (cleared) + dlm_lockres_calc_usage(dlm, res); + else { + mlog(ML_ERROR, "%s:%.*s: node %u trying to drop ref " + "but it is already dropped!\n", dlm->name, + res->lockname.len, res->lockname.name, node); + dlm_print_one_lock_resource(res); + } + ret = 0; + goto done; } - mlog(0, "checking queues...\n"); - queue = &res->granted; - for (i=0; i<3; i++) { - list_for_each(iter, queue) { - lock = list_entry (iter, struct dlm_lock, list); - empty = 0; - if (lock->ml.node == dlm->node_num) { - mlog(0, "found a lock owned by this node " - "still on the %s queue! will not " - "migrate this lockres\n", - i==0 ? "granted" : - (i==1 ? "converting" : "blocked")); - spin_unlock(&res->spinlock); - ret = -ENOTEMPTY; - goto leave; + + item = kzalloc(sizeof(*item), GFP_NOFS); + if (!item) { + ret = -ENOMEM; + mlog_errno(ret); + goto done; + } + + dlm_init_work_item(dlm, item, dlm_deref_lockres_worker, NULL); + item->u.dl.deref_res = res; + item->u.dl.deref_node = node; + + spin_lock(&dlm->work_lock); + list_add_tail(&item->list, &dlm->work_list); + spin_unlock(&dlm->work_lock); + + queue_work(dlm->dlm_worker, &dlm->dispatched_work); + return 0; + +done: + if (res) + dlm_lockres_put(res); + dlm_put(dlm); + + return ret; +} + +static void dlm_deref_lockres_worker(struct dlm_work_item *item, void *data) +{ + struct dlm_ctxt *dlm; + struct dlm_lock_resource *res; + u8 node; + u8 cleared = 0; + + dlm = item->dlm; + res = item->u.dl.deref_res; + node = item->u.dl.deref_node; + + spin_lock(&res->spinlock); + BUG_ON(res->state & DLM_LOCK_RES_DROPPING_REF); + if (test_bit(node, res->refmap)) { + __dlm_wait_on_lockres_flags(res, DLM_LOCK_RES_SETREF_INPROG); + dlm_lockres_clear_refmap_bit(dlm, res, node); + cleared = 1; + } + spin_unlock(&res->spinlock); + + if (cleared) { + mlog(0, "%s:%.*s node %u ref dropped in dispatch\n", + dlm->name, res->lockname.len, res->lockname.name, node); + dlm_lockres_calc_usage(dlm, res); + } else { + mlog(ML_ERROR, "%s:%.*s: node %u trying to drop ref " + "but it is already dropped!\n", dlm->name, + res->lockname.len, res->lockname.name, node); + dlm_print_one_lock_resource(res); + } + + dlm_lockres_put(res); +} + +/* + * A migrateable resource is one that is : + * 1. locally mastered, and, + * 2. zero local locks, and, + * 3. one or more non-local locks, or, one or more references + * Returns 1 if yes, 0 if not. + */ +static int dlm_is_lockres_migrateable(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res) +{ + enum dlm_lockres_list idx; + int nonlocal = 0, node_ref; + struct list_head *queue; + struct dlm_lock *lock; + u64 cookie; + + assert_spin_locked(&res->spinlock); + + /* delay migration when the lockres is in MIGRATING state */ + if (res->state & DLM_LOCK_RES_MIGRATING) + return 0; + + if (res->owner != dlm->node_num) + return 0; + + for (idx = DLM_GRANTED_LIST; idx <= DLM_BLOCKED_LIST; idx++) { + queue = dlm_list_idx_to_ptr(res, idx); + list_for_each_entry(lock, queue, list) { + if (lock->ml.node != dlm->node_num) { + nonlocal++; + continue; } + cookie = be64_to_cpu(lock->ml.cookie); + mlog(0, "%s: Not migrateable res %.*s, lock %u:%llu on " + "%s list\n", dlm->name, res->lockname.len, + res->lockname.name, + dlm_get_lock_cookie_node(cookie), + dlm_get_lock_cookie_seq(cookie), + dlm_list_in_text(idx)); + return 0; } - queue++; } - mlog(0, "all locks on this lockres are nonlocal. continuing\n"); - spin_unlock(&res->spinlock); - /* no work to do */ - if (empty) { - mlog(0, "no locks were found on this lockres! done!\n"); - ret = 0; - goto leave; + if (!nonlocal) { + node_ref = find_next_bit(res->refmap, O2NM_MAX_NODES, 0); + if (node_ref >= O2NM_MAX_NODES) + return 0; } - /* - * preallocate up front - * if this fails, abort - */ + mlog(0, "%s: res %.*s, Migrateable\n", dlm->name, res->lockname.len, + res->lockname.name); + + return 1; +} + +/* + * DLM_MIGRATE_LOCKRES + */ + + +static int dlm_migrate_lockres(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res, u8 target) +{ + struct dlm_master_list_entry *mle = NULL; + struct dlm_master_list_entry *oldmle = NULL; + struct dlm_migratable_lockres *mres = NULL; + int ret = 0; + const char *name; + unsigned int namelen; + int mle_added = 0; + int wake = 0; + + if (!dlm_grab(dlm)) + return -EINVAL; + + BUG_ON(target == O2NM_MAX_NODES); + name = res->lockname.name; + namelen = res->lockname.len; + + mlog(0, "%s: Migrating %.*s to node %u\n", dlm->name, namelen, name, + target); + + /* preallocate up front. if this fails, abort */ ret = -ENOMEM; - mres = (struct dlm_migratable_lockres *) __get_free_page(GFP_KERNEL); + mres = (struct dlm_migratable_lockres *) __get_free_page(GFP_NOFS); if (!mres) { mlog_errno(ret); goto leave; } - mle = (struct dlm_master_list_entry *) kmem_cache_alloc(dlm_mle_cache, - GFP_KERNEL); + mle = kmem_cache_alloc(dlm_mle_cache, GFP_NOFS); if (!mle) { mlog_errno(ret); goto leave; @@ -1875,35 +2482,10 @@ int dlm_migrate_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, ret = 0; /* - * find a node to migrate the lockres to - */ - - mlog(0, "picking a migration node\n"); - spin_lock(&dlm->spinlock); - /* pick a new node */ - if (!test_bit(target, dlm->domain_map) || - target >= O2NM_MAX_NODES) { - target = dlm_pick_migration_target(dlm, res); - } - mlog(0, "node %u chosen for migration\n", target); - - if (target >= O2NM_MAX_NODES || - !test_bit(target, dlm->domain_map)) { - /* target chosen is not alive */ - ret = -EINVAL; - } - - if (ret) { - spin_unlock(&dlm->spinlock); - goto fail; - } - - mlog(0, "continuing with target = %u\n", target); - - /* * clear any existing master requests and * add the migration mle to the list */ + spin_lock(&dlm->spinlock); spin_lock(&dlm->master_lock); ret = dlm_add_migration_mle(dlm, res, mle, &oldmle, name, namelen, target, dlm->node_num); @@ -1926,6 +2508,7 @@ int dlm_migrate_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, res->lockname.name, target); spin_lock(&res->spinlock); res->state &= ~DLM_LOCK_RES_MIGRATING; + wake = 1; spin_unlock(&res->spinlock); ret = -EINVAL; } @@ -1943,6 +2526,7 @@ fail: dlm_put_mle(mle); } else if (mle) { kmem_cache_free(dlm_mle_cache, mle); + mle = NULL; } goto leave; } @@ -1953,6 +2537,9 @@ fail: * the lockres */ + /* now that remote nodes are spinning on the MIGRATING flag, + * ensure that all assert_master work is flushed. */ + flush_workqueue(dlm->dlm_worker); /* get an extra reference on the mle. * otherwise the assert_master from the new @@ -1961,7 +2548,7 @@ fail: * take both dlm->spinlock and dlm->master_lock */ spin_lock(&dlm->spinlock); spin_lock(&dlm->master_lock); - dlm_get_mle(mle); + dlm_get_mle_inuse(mle); spin_unlock(&dlm->master_lock); spin_unlock(&dlm->spinlock); @@ -1978,7 +2565,14 @@ fail: /* migration failed, detach and clean up mle */ dlm_mle_detach_hb_events(dlm, mle); dlm_put_mle(mle); - dlm_put_mle(mle); + dlm_put_mle_inuse(mle); + spin_lock(&res->spinlock); + res->state &= ~DLM_LOCK_RES_MIGRATING; + wake = 1; + spin_unlock(&res->spinlock); + if (dlm_is_host_down(ret)) + dlm_wait_for_node_death(dlm, target, + DLM_NODE_DEATH_WAIT_MAX); goto leave; } @@ -1990,7 +2584,7 @@ fail: * is complete everywhere. if the target dies while this is * going on, some nodes could potentially see the target as the * master, so it is important that my recovery finds the migration - * mle and sets the master to UNKNONWN. */ + * mle and sets the master to UNKNOWN. */ /* wait for new node to assert master */ @@ -2004,16 +2598,29 @@ fail: res->owner == target) break; - mlog(0, "timed out during migration\n"); - } - if (ret == -ERESTARTSYS) { - /* migration failed, detach and clean up mle */ - dlm_mle_detach_hb_events(dlm, mle); - dlm_put_mle(mle); - dlm_put_mle(mle); - goto leave; - } - /* TODO: if node died: stop, clean up, return error */ + mlog(0, "%s:%.*s: timed out during migration\n", + dlm->name, res->lockname.len, res->lockname.name); + /* avoid hang during shutdown when migrating lockres + * to a node which also goes down */ + if (dlm_is_node_dead(dlm, target)) { + mlog(0, "%s:%.*s: expected migration " + "target %u is no longer up, restarting\n", + dlm->name, res->lockname.len, + res->lockname.name, target); + ret = -EINVAL; + /* migration failed, detach and clean up mle */ + dlm_mle_detach_hb_events(dlm, mle); + dlm_put_mle(mle); + dlm_put_mle_inuse(mle); + spin_lock(&res->spinlock); + res->state &= ~DLM_LOCK_RES_MIGRATING; + wake = 1; + spin_unlock(&res->spinlock); + goto leave; + } + } else + mlog(0, "%s:%.*s: caught signal during migration\n", + dlm->name, res->lockname.len, res->lockname.name); } /* all done, set the owner, clear the flag */ @@ -2026,7 +2633,7 @@ fail: /* master is known, detach if not already detached */ dlm_mle_detach_hb_events(dlm, mle); - dlm_put_mle(mle); + dlm_put_mle_inuse(mle); ret = 0; dlm_lockres_calc_usage(dlm, res); @@ -2036,16 +2643,61 @@ leave: if (ret < 0) dlm_kick_thread(dlm, res); - /* TODO: cleanup */ + /* wake up waiters if the MIGRATING flag got set + * but migration failed */ + if (wake) + wake_up(&res->wq); + if (mres) free_page((unsigned long)mres); dlm_put(dlm); - mlog(0, "returning %d\n", ret); + mlog(0, "%s: Migrating %.*s to %u, returns %d\n", dlm->name, namelen, + name, target, ret); return ret; } -EXPORT_SYMBOL_GPL(dlm_migrate_lockres); + +#define DLM_MIGRATION_RETRY_MS 100 + +/* + * Should be called only after beginning the domain leave process. + * There should not be any remaining locks on nonlocal lock resources, + * and there should be no local locks left on locally mastered resources. + * + * Called with the dlm spinlock held, may drop it to do migration, but + * will re-acquire before exit. + * + * Returns: 1 if dlm->spinlock was dropped/retaken, 0 if never dropped + */ +int dlm_empty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) +{ + int ret; + int lock_dropped = 0; + u8 target = O2NM_MAX_NODES; + + assert_spin_locked(&dlm->spinlock); + + spin_lock(&res->spinlock); + if (dlm_is_lockres_migrateable(dlm, res)) + target = dlm_pick_migration_target(dlm, res); + spin_unlock(&res->spinlock); + + if (target == O2NM_MAX_NODES) + goto leave; + + /* Wheee! Migrate lockres here! Will sleep so drop spinlock. */ + spin_unlock(&dlm->spinlock); + lock_dropped = 1; + ret = dlm_migrate_lockres(dlm, res, target); + if (ret) + mlog(0, "%s: res %.*s, Migrate to node %u failed with %d\n", + dlm->name, res->lockname.len, res->lockname.name, + target, ret); + spin_lock(&dlm->spinlock); +leave: + return lock_dropped; +} int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock) { @@ -2067,7 +2719,7 @@ static int dlm_migration_can_proceed(struct dlm_ctxt *dlm, can_proceed = !!(res->state & DLM_LOCK_RES_MIGRATING); spin_unlock(&res->spinlock); - /* target has died, so make the caller break out of the + /* target has died, so make the caller break out of the * wait_event, but caller must recheck the domain_map */ spin_lock(&dlm->spinlock); if (!test_bit(mig_target, dlm->domain_map)) @@ -2076,7 +2728,8 @@ static int dlm_migration_can_proceed(struct dlm_ctxt *dlm, return can_proceed; } -int dlm_lockres_is_dirty(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) +static int dlm_lockres_is_dirty(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res) { int ret; spin_lock(&res->spinlock); @@ -2105,8 +2758,15 @@ static int dlm_mark_lockres_migrating(struct dlm_ctxt *dlm, __dlm_lockres_reserve_ast(res); spin_unlock(&res->spinlock); - /* now flush all the pending asts.. hang out for a bit */ + /* now flush all the pending asts */ dlm_kick_thread(dlm, res); + /* before waiting on DIRTY, block processes which may + * try to dirty the lockres before MIGRATING is set */ + spin_lock(&res->spinlock); + BUG_ON(res->state & DLM_LOCK_RES_BLOCK_DIRTY); + res->state |= DLM_LOCK_RES_BLOCK_DIRTY; + spin_unlock(&res->spinlock); + /* now wait on any pending asts and the DIRTY state */ wait_event(dlm->ast_wq, !dlm_lockres_is_dirty(dlm, res)); dlm_lockres_release_ast(dlm, res); @@ -2133,6 +2793,7 @@ again: goto again; } + ret = 0; /* did the target go down or die? */ spin_lock(&dlm->spinlock); if (!test_bit(target, dlm->domain_map)) { @@ -2143,9 +2804,21 @@ again: spin_unlock(&dlm->spinlock); /* + * if target is down, we need to clear DLM_LOCK_RES_BLOCK_DIRTY for + * another try; otherwise, we are sure the MIGRATING state is there, + * drop the unneded state which blocked threads trying to DIRTY + */ + spin_lock(&res->spinlock); + BUG_ON(!(res->state & DLM_LOCK_RES_BLOCK_DIRTY)); + res->state &= ~DLM_LOCK_RES_BLOCK_DIRTY; + if (!ret) + BUG_ON(!(res->state & DLM_LOCK_RES_MIGRATING)); + spin_unlock(&res->spinlock); + + /* * at this point: * - * o the DLM_LOCK_RES_MIGRATING flag is set + * o the DLM_LOCK_RES_MIGRATING flag is set if target not down * o there are no pending asts on this lockres * o all processes trying to reserve an ast on this * lockres must wait for the MIGRATING flag to clear @@ -2159,18 +2832,16 @@ again: static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) { - struct list_head *iter, *iter2; struct list_head *queue = &res->granted; - int i; - struct dlm_lock *lock; + int i, bit; + struct dlm_lock *lock, *next; assert_spin_locked(&res->spinlock); BUG_ON(res->owner == dlm->node_num); for (i=0; i<3; i++) { - list_for_each_safe(iter, iter2, queue) { - lock = list_entry (iter, struct dlm_lock, list); + list_for_each_entry_safe(lock, next, queue, list) { if (lock->ml.node != dlm->node_num) { mlog(0, "putting lock for node %u\n", lock->ml.node); @@ -2179,65 +2850,83 @@ static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm, BUG_ON(!list_empty(&lock->bast_list)); BUG_ON(lock->ast_pending); BUG_ON(lock->bast_pending); + dlm_lockres_clear_refmap_bit(dlm, res, + lock->ml.node); list_del_init(&lock->list); dlm_lock_put(lock); + /* In a normal unlock, we would have added a + * DLM_UNLOCK_FREE_LOCK action. Force it. */ + dlm_lock_put(lock); } } queue++; } + bit = 0; + while (1) { + bit = find_next_bit(res->refmap, O2NM_MAX_NODES, bit); + if (bit >= O2NM_MAX_NODES) + break; + /* do not clear the local node reference, if there is a + * process holding this, let it drop the ref itself */ + if (bit != dlm->node_num) { + mlog(0, "%s:%.*s: node %u had a ref to this " + "migrating lockres, clearing\n", dlm->name, + res->lockname.len, res->lockname.name, bit); + dlm_lockres_clear_refmap_bit(dlm, res, bit); + } + bit++; + } } -/* for now this is not too intelligent. we will - * need stats to make this do the right thing. - * this just finds the first lock on one of the - * queues and uses that node as the target. */ +/* + * Pick a node to migrate the lock resource to. This function selects a + * potential target based first on the locks and then on refmap. It skips + * nodes that are in the process of exiting the domain. + */ static u8 dlm_pick_migration_target(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) { - int i; + enum dlm_lockres_list idx; struct list_head *queue = &res->granted; - struct list_head *iter; struct dlm_lock *lock; - int nodenum; + int noderef; + u8 nodenum = O2NM_MAX_NODES; assert_spin_locked(&dlm->spinlock); + assert_spin_locked(&res->spinlock); - spin_lock(&res->spinlock); - for (i=0; i<3; i++) { - list_for_each(iter, queue) { - /* up to the caller to make sure this node - * is alive */ - lock = list_entry (iter, struct dlm_lock, list); - if (lock->ml.node != dlm->node_num) { - spin_unlock(&res->spinlock); - return lock->ml.node; - } + /* Go through all the locks */ + for (idx = DLM_GRANTED_LIST; idx <= DLM_BLOCKED_LIST; idx++) { + queue = dlm_list_idx_to_ptr(res, idx); + list_for_each_entry(lock, queue, list) { + if (lock->ml.node == dlm->node_num) + continue; + if (test_bit(lock->ml.node, dlm->exit_domain_map)) + continue; + nodenum = lock->ml.node; + goto bail; } - queue++; } - spin_unlock(&res->spinlock); - mlog(0, "have not found a suitable target yet! checking domain map\n"); - /* ok now we're getting desperate. pick anyone alive. */ - nodenum = -1; + /* Go thru the refmap */ + noderef = -1; while (1) { - nodenum = find_next_bit(dlm->domain_map, - O2NM_MAX_NODES, nodenum+1); - mlog(0, "found %d in domain map\n", nodenum); - if (nodenum >= O2NM_MAX_NODES) + noderef = find_next_bit(res->refmap, O2NM_MAX_NODES, + noderef + 1); + if (noderef >= O2NM_MAX_NODES) break; - if (nodenum != dlm->node_num) { - mlog(0, "picking %d\n", nodenum); - return nodenum; - } + if (noderef == dlm->node_num) + continue; + if (test_bit(noderef, dlm->exit_domain_map)) + continue; + nodenum = noderef; + goto bail; } - mlog(0, "giving up. no master to migrate to\n"); - return DLM_LOCK_RES_OWNER_UNKNOWN; +bail: + return nodenum; } - - /* this is called by the new master once all lockres * data has been received */ static int dlm_do_migrate_request(struct dlm_ctxt *dlm, @@ -2246,7 +2935,7 @@ static int dlm_do_migrate_request(struct dlm_ctxt *dlm, struct dlm_node_iter *iter) { struct dlm_migrate_request migrate; - int ret, status = 0; + int ret, skip, status = 0; int nodenum; memset(&migrate, 0, sizeof(migrate)); @@ -2263,15 +2952,42 @@ static int dlm_do_migrate_request(struct dlm_ctxt *dlm, nodenum == new_master) continue; + /* We could race exit domain. If exited, skip. */ + spin_lock(&dlm->spinlock); + skip = (!test_bit(nodenum, dlm->domain_map)); + spin_unlock(&dlm->spinlock); + if (skip) { + clear_bit(nodenum, iter->node_map); + continue; + } + ret = o2net_send_message(DLM_MIGRATE_REQUEST_MSG, dlm->key, &migrate, sizeof(migrate), nodenum, &status); - if (ret < 0) - mlog_errno(ret); - else if (status < 0) { + if (ret < 0) { + mlog(ML_ERROR, "%s: res %.*s, Error %d send " + "MIGRATE_REQUEST to node %u\n", dlm->name, + migrate.namelen, migrate.name, ret, nodenum); + if (!dlm_is_host_down(ret)) { + mlog(ML_ERROR, "unhandled error=%d!\n", ret); + BUG(); + } + clear_bit(nodenum, iter->node_map); + ret = 0; + } else if (status < 0) { mlog(0, "migrate request (node %u) returned %d!\n", nodenum, status); ret = status; + } else if (status == DLM_MIGRATE_RESPONSE_MASTERY_REF) { + /* during the migration request we short-circuited + * the mastery of the lockres. make sure we have + * a mastery ref for nodenum */ + mlog(0, "%s:%.*s: need ref for node %u\n", + dlm->name, res->lockname.len, res->lockname.name, + nodenum); + spin_lock(&res->spinlock); + dlm_lockres_set_refmap_bit(dlm, res, nodenum); + spin_unlock(&res->spinlock); } } @@ -2290,14 +3006,15 @@ static int dlm_do_migrate_request(struct dlm_ctxt *dlm, * we will have no mle in the list to start with. now we can add an mle for * the migration and this should be the only one found for those scanning the * list. */ -int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data) +int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { struct dlm_ctxt *dlm = data; struct dlm_lock_resource *res = NULL; struct dlm_migrate_request *migrate = (struct dlm_migrate_request *) msg->buf; struct dlm_master_list_entry *mle = NULL, *oldmle = NULL; const char *name; - unsigned int namelen; + unsigned int namelen, hash; int ret = 0; if (!dlm_grab(dlm)) @@ -2305,10 +3022,10 @@ int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data) name = migrate->name; namelen = migrate->namelen; + hash = dlm_lockid_hash(name, namelen); /* preallocate.. if this fails, abort */ - mle = (struct dlm_master_list_entry *) kmem_cache_alloc(dlm_mle_cache, - GFP_KERNEL); + mle = kmem_cache_alloc(dlm_mle_cache, GFP_NOFS); if (!mle) { ret = -ENOMEM; @@ -2317,9 +3034,7 @@ int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data) /* check for pre-existing lock */ spin_lock(&dlm->spinlock); - res = __dlm_lookup_lockres(dlm, name, namelen); - spin_lock(&dlm->master_lock); - + res = __dlm_lookup_lockres(dlm, name, namelen, hash); if (res) { spin_lock(&res->spinlock); if (res->state & DLM_LOCK_RES_RECOVERING) { @@ -2337,14 +3052,15 @@ int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data) spin_unlock(&res->spinlock); } + spin_lock(&dlm->master_lock); /* ignore status. only nonzero status would BUG. */ ret = dlm_add_migration_mle(dlm, res, mle, &oldmle, name, namelen, migrate->new_master, migrate->master); -unlock: spin_unlock(&dlm->master_lock); +unlock: spin_unlock(&dlm->spinlock); if (oldmle) { @@ -2379,8 +3095,6 @@ static int dlm_add_migration_mle(struct dlm_ctxt *dlm, *oldmle = NULL; - mlog_entry_void(); - assert_spin_locked(&dlm->spinlock); assert_spin_locked(&dlm->master_lock); @@ -2412,9 +3126,18 @@ static int dlm_add_migration_mle(struct dlm_ctxt *dlm, tmp->master = master; atomic_set(&tmp->woken, 1); wake_up(&tmp->wq); - /* remove it from the list so that only one - * mle will be found */ - list_del_init(&tmp->list); + /* remove it so that only one mle will be found */ + __dlm_unlink_mle(dlm, tmp); + __dlm_mle_detach_hb_events(dlm, tmp); + if (tmp->type == DLM_MLE_MASTER) { + ret = DLM_MIGRATE_RESPONSE_MASTERY_REF; + mlog(0, "%s:%.*s: master=%u, newmaster=%u, " + "telling master to get ref " + "for cleared out mle during " + "migration\n", dlm->name, + namelen, name, master, + new_master); + } } spin_unlock(&tmp->spinlock); } @@ -2422,135 +3145,166 @@ static int dlm_add_migration_mle(struct dlm_ctxt *dlm, /* now add a migration mle to the tail of the list */ dlm_init_mle(mle, DLM_MLE_MIGRATION, dlm, res, name, namelen); mle->new_master = new_master; + /* the new master will be sending an assert master for this. + * at that point we will get the refmap reference */ mle->master = master; /* do this for consistency with other mle types */ set_bit(new_master, mle->maybe_map); - list_add(&mle->list, &dlm->master_list); + __dlm_insert_mle(dlm, mle); return ret; } - -void dlm_clean_master_list(struct dlm_ctxt *dlm, u8 dead_node) +/* + * Sets the owner of the lockres, associated to the mle, to UNKNOWN + */ +static struct dlm_lock_resource *dlm_reset_mleres_owner(struct dlm_ctxt *dlm, + struct dlm_master_list_entry *mle) { - struct list_head *iter, *iter2; - struct dlm_master_list_entry *mle; struct dlm_lock_resource *res; - mlog_entry("dlm=%s, dead node=%u\n", dlm->name, dead_node); -top: - assert_spin_locked(&dlm->spinlock); + /* Find the lockres associated to the mle and set its owner to UNK */ + res = __dlm_lookup_lockres(dlm, mle->mname, mle->mnamelen, + mle->mnamehash); + if (res) { + spin_unlock(&dlm->master_lock); - /* clean the master list */ - spin_lock(&dlm->master_lock); - list_for_each_safe(iter, iter2, &dlm->master_list) { - mle = list_entry(iter, struct dlm_master_list_entry, list); + /* move lockres onto recovery list */ + spin_lock(&res->spinlock); + dlm_set_lockres_owner(dlm, res, DLM_LOCK_RES_OWNER_UNKNOWN); + dlm_move_lockres_to_recovery_list(dlm, res); + spin_unlock(&res->spinlock); + dlm_lockres_put(res); - BUG_ON(mle->type != DLM_MLE_BLOCK && - mle->type != DLM_MLE_MASTER && - mle->type != DLM_MLE_MIGRATION); + /* about to get rid of mle, detach from heartbeat */ + __dlm_mle_detach_hb_events(dlm, mle); - /* MASTER mles are initiated locally. the waiting - * process will notice the node map change - * shortly. let that happen as normal. */ - if (mle->type == DLM_MLE_MASTER) - continue; + /* dump the mle */ + spin_lock(&dlm->master_lock); + __dlm_put_mle(mle); + spin_unlock(&dlm->master_lock); + } + return res; +} - /* BLOCK mles are initiated by other nodes. - * need to clean up if the dead node would have - * been the master. */ - if (mle->type == DLM_MLE_BLOCK) { - int bit; +static void dlm_clean_migration_mle(struct dlm_ctxt *dlm, + struct dlm_master_list_entry *mle) +{ + __dlm_mle_detach_hb_events(dlm, mle); - spin_lock(&mle->spinlock); - bit = find_next_bit(mle->maybe_map, O2NM_MAX_NODES, 0); - if (bit != dead_node) { - mlog(0, "mle found, but dead node %u would " - "not have been master\n", dead_node); - spin_unlock(&mle->spinlock); - } else { - /* must drop the refcount by one since the - * assert_master will never arrive. this - * may result in the mle being unlinked and - * freed, but there may still be a process - * waiting in the dlmlock path which is fine. */ - mlog(ML_ERROR, "node %u was expected master\n", - dead_node); - atomic_set(&mle->woken, 1); - spin_unlock(&mle->spinlock); - wake_up(&mle->wq); - /* final put will take care of list removal */ - __dlm_put_mle(mle); - } - continue; - } + spin_lock(&mle->spinlock); + __dlm_unlink_mle(dlm, mle); + atomic_set(&mle->woken, 1); + spin_unlock(&mle->spinlock); - /* everything else is a MIGRATION mle */ - - /* the rule for MIGRATION mles is that the master - * becomes UNKNOWN if *either* the original or - * the new master dies. all UNKNOWN lockreses - * are sent to whichever node becomes the recovery - * master. the new master is responsible for - * determining if there is still a master for - * this lockres, or if he needs to take over - * mastery. either way, this node should expect - * another message to resolve this. */ - if (mle->master != dead_node && - mle->new_master != dead_node) - continue; + wake_up(&mle->wq); +} + +static void dlm_clean_block_mle(struct dlm_ctxt *dlm, + struct dlm_master_list_entry *mle, u8 dead_node) +{ + int bit; - /* if we have reached this point, this mle needs to - * be removed from the list and freed. */ + BUG_ON(mle->type != DLM_MLE_BLOCK); - /* remove from the list early. NOTE: unlinking - * list_head while in list_for_each_safe */ - spin_lock(&mle->spinlock); - list_del_init(&mle->list); + spin_lock(&mle->spinlock); + bit = find_next_bit(mle->maybe_map, O2NM_MAX_NODES, 0); + if (bit != dead_node) { + mlog(0, "mle found, but dead node %u would not have been " + "master\n", dead_node); + spin_unlock(&mle->spinlock); + } else { + /* Must drop the refcount by one since the assert_master will + * never arrive. This may result in the mle being unlinked and + * freed, but there may still be a process waiting in the + * dlmlock path which is fine. */ + mlog(0, "node %u was expected master\n", dead_node); atomic_set(&mle->woken, 1); spin_unlock(&mle->spinlock); wake_up(&mle->wq); - mlog(0, "node %u died during migration from " - "%u to %u!\n", dead_node, - mle->master, mle->new_master); - /* if there is a lockres associated with this - * mle, find it and set its owner to UNKNOWN */ - res = __dlm_lookup_lockres(dlm, mle->u.name.name, - mle->u.name.len); - if (res) { - /* unfortunately if we hit this rare case, our - * lock ordering is messed. we need to drop - * the master lock so that we can take the - * lockres lock, meaning that we will have to - * restart from the head of list. */ - spin_unlock(&dlm->master_lock); + /* Do not need events any longer, so detach from heartbeat */ + __dlm_mle_detach_hb_events(dlm, mle); + __dlm_put_mle(mle); + } +} - /* move lockres onto recovery list */ - spin_lock(&res->spinlock); - dlm_set_lockres_owner(dlm, res, - DLM_LOCK_RES_OWNER_UNKNOWN); - dlm_move_lockres_to_recovery_list(dlm, res); - spin_unlock(&res->spinlock); - dlm_lockres_put(res); +void dlm_clean_master_list(struct dlm_ctxt *dlm, u8 dead_node) +{ + struct dlm_master_list_entry *mle; + struct dlm_lock_resource *res; + struct hlist_head *bucket; + struct hlist_node *tmp; + unsigned int i; - /* dump the mle */ - spin_lock(&dlm->master_lock); - __dlm_put_mle(mle); - spin_unlock(&dlm->master_lock); + mlog(0, "dlm=%s, dead node=%u\n", dlm->name, dead_node); +top: + assert_spin_locked(&dlm->spinlock); - /* restart */ - goto top; - } + /* clean the master list */ + spin_lock(&dlm->master_lock); + for (i = 0; i < DLM_HASH_BUCKETS; i++) { + bucket = dlm_master_hash(dlm, i); + hlist_for_each_entry_safe(mle, tmp, bucket, master_hash_node) { + BUG_ON(mle->type != DLM_MLE_BLOCK && + mle->type != DLM_MLE_MASTER && + mle->type != DLM_MLE_MIGRATION); + + /* MASTER mles are initiated locally. The waiting + * process will notice the node map change shortly. + * Let that happen as normal. */ + if (mle->type == DLM_MLE_MASTER) + continue; + + /* BLOCK mles are initiated by other nodes. Need to + * clean up if the dead node would have been the + * master. */ + if (mle->type == DLM_MLE_BLOCK) { + dlm_clean_block_mle(dlm, mle, dead_node); + continue; + } - /* this may be the last reference */ - __dlm_put_mle(mle); + /* Everything else is a MIGRATION mle */ + + /* The rule for MIGRATION mles is that the master + * becomes UNKNOWN if *either* the original or the new + * master dies. All UNKNOWN lockres' are sent to + * whichever node becomes the recovery master. The new + * master is responsible for determining if there is + * still a master for this lockres, or if he needs to + * take over mastery. Either way, this node should + * expect another message to resolve this. */ + + if (mle->master != dead_node && + mle->new_master != dead_node) + continue; + + /* If we have reached this point, this mle needs to be + * removed from the list and freed. */ + dlm_clean_migration_mle(dlm, mle); + + mlog(0, "%s: node %u died during migration from " + "%u to %u!\n", dlm->name, dead_node, mle->master, + mle->new_master); + + /* If we find a lockres associated with the mle, we've + * hit this rare case that messes up our lock ordering. + * If so, we need to drop the master lock so that we can + * take the lockres lock, meaning that we will have to + * restart from the head of list. */ + res = dlm_reset_mleres_owner(dlm, mle); + if (res) + /* restart */ + goto top; + + /* This may be the last reference */ + __dlm_put_mle(mle); + } } spin_unlock(&dlm->master_lock); } - int dlm_finish_migration(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, u8 old_master) { @@ -2563,6 +3317,13 @@ int dlm_finish_migration(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, clear_bit(dlm->node_num, iter.node_map); spin_unlock(&dlm->spinlock); + /* ownership of the lockres is changing. account for the + * mastery reference here since old_master will briefly have + * a reference after the migration completes */ + spin_lock(&res->spinlock); + dlm_lockres_set_refmap_bit(dlm, res, old_master); + spin_unlock(&res->spinlock); + mlog(0, "now time to do a migrate request to other nodes\n"); ret = dlm_do_migrate_request(dlm, res, old_master, dlm->node_num, &iter); @@ -2575,8 +3336,7 @@ int dlm_finish_migration(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, res->lockname.len, res->lockname.name); /* this call now finishes out the nodemap * even if one or more nodes die */ - ret = dlm_do_assert_master(dlm, res->lockname.name, - res->lockname.len, iter.node_map, + ret = dlm_do_assert_master(dlm, res, iter.node_map, DLM_ASSERT_MASTER_FINISH_MIGRATION); if (ret < 0) { /* no longer need to retry. all living nodes contacted. */ @@ -2588,8 +3348,7 @@ int dlm_finish_migration(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, set_bit(old_master, iter.node_map); mlog(0, "doing assert master of %.*s back to %u\n", res->lockname.len, res->lockname.name, old_master); - ret = dlm_do_assert_master(dlm, res->lockname.name, - res->lockname.len, iter.node_map, + ret = dlm_do_assert_master(dlm, res, iter.node_map, DLM_ASSERT_MASTER_FINISH_MIGRATION); if (ret < 0) { mlog(0, "assert master to original master failed " @@ -2662,3 +3421,41 @@ void dlm_lockres_release_ast(struct dlm_ctxt *dlm, wake_up(&res->wq); wake_up(&dlm->migration_wq); } + +void dlm_force_free_mles(struct dlm_ctxt *dlm) +{ + int i; + struct hlist_head *bucket; + struct dlm_master_list_entry *mle; + struct hlist_node *tmp; + + /* + * We notified all other nodes that we are exiting the domain and + * marked the dlm state to DLM_CTXT_LEAVING. If any mles are still + * around we force free them and wake any processes that are waiting + * on the mles + */ + spin_lock(&dlm->spinlock); + spin_lock(&dlm->master_lock); + + BUG_ON(dlm->dlm_state != DLM_CTXT_LEAVING); + BUG_ON((find_next_bit(dlm->domain_map, O2NM_MAX_NODES, 0) < O2NM_MAX_NODES)); + + for (i = 0; i < DLM_HASH_BUCKETS; i++) { + bucket = dlm_master_hash(dlm, i); + hlist_for_each_entry_safe(mle, tmp, bucket, master_hash_node) { + if (mle->type != DLM_MLE_BLOCK) { + mlog(ML_ERROR, "bad mle: %p\n", mle); + dlm_print_one_mle(mle); + } + atomic_set(&mle->woken, 1); + wake_up(&mle->wq); + + __dlm_unlink_mle(dlm, mle); + __dlm_mle_detach_hb_events(dlm, mle); + __dlm_put_mle(mle); + } + } + spin_unlock(&dlm->master_lock); + spin_unlock(&dlm->spinlock); +} diff --git a/fs/ocfs2/dlm/dlmrecovery.c b/fs/ocfs2/dlm/dlmrecovery.c index 0c8eb1093f0..45067faf569 100644 --- a/fs/ocfs2/dlm/dlmrecovery.c +++ b/fs/ocfs2/dlm/dlmrecovery.c @@ -30,7 +30,6 @@ #include <linux/types.h> #include <linux/slab.h> #include <linux/highmem.h> -#include <linux/utsname.h> #include <linux/init.h> #include <linux/sysctl.h> #include <linux/random.h> @@ -39,6 +38,7 @@ #include <linux/inet.h> #include <linux/timer.h> #include <linux/kthread.h> +#include <linux/delay.h> #include "cluster/heartbeat.h" @@ -55,9 +55,6 @@ static void dlm_do_local_recovery_cleanup(struct dlm_ctxt *dlm, u8 dead_node); static int dlm_recovery_thread(void *data); -void dlm_complete_recovery_thread(struct dlm_ctxt *dlm); -int dlm_launch_recovery_thread(struct dlm_ctxt *dlm); -static void dlm_kick_recovery_thread(struct dlm_ctxt *dlm); static int dlm_do_recovery(struct dlm_ctxt *dlm); static int dlm_pick_recovery_master(struct dlm_ctxt *dlm); @@ -77,15 +74,9 @@ static int dlm_send_mig_lockres_msg(struct dlm_ctxt *dlm, u8 send_to, struct dlm_lock_resource *res, int total_locks); -static int dlm_lockres_master_requery(struct dlm_ctxt *dlm, - struct dlm_lock_resource *res, - u8 *real_master); static int dlm_process_recovery_data(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, struct dlm_migratable_lockres *mres); -static int dlm_do_master_requery(struct dlm_ctxt *dlm, - struct dlm_lock_resource *res, - u8 nodenum, u8 *real_master); static int dlm_send_finalize_reco_message(struct dlm_ctxt *dlm); static int dlm_send_all_done_msg(struct dlm_ctxt *dlm, u8 dead_node, u8 send_to); @@ -100,11 +91,14 @@ static void dlm_reco_unlock_ast(void *astdata, enum dlm_status st); static void dlm_request_all_locks_worker(struct dlm_work_item *item, void *data); static void dlm_mig_lockres_worker(struct dlm_work_item *item, void *data); +static int dlm_lockres_master_requery(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res, + u8 *real_master); static u64 dlm_get_next_mig_cookie(void); -static spinlock_t dlm_reco_state_lock = SPIN_LOCK_UNLOCKED; -static spinlock_t dlm_mig_cookie_lock = SPIN_LOCK_UNLOCKED; +static DEFINE_SPINLOCK(dlm_reco_state_lock); +static DEFINE_SPINLOCK(dlm_mig_cookie_lock); static u64 dlm_mig_cookie = 1; static u64 dlm_get_next_mig_cookie(void) @@ -120,30 +114,60 @@ static u64 dlm_get_next_mig_cookie(void) return c; } +static inline void dlm_set_reco_dead_node(struct dlm_ctxt *dlm, + u8 dead_node) +{ + assert_spin_locked(&dlm->spinlock); + if (dlm->reco.dead_node != dead_node) + mlog(0, "%s: changing dead_node from %u to %u\n", + dlm->name, dlm->reco.dead_node, dead_node); + dlm->reco.dead_node = dead_node; +} + +static inline void dlm_set_reco_master(struct dlm_ctxt *dlm, + u8 master) +{ + assert_spin_locked(&dlm->spinlock); + mlog(0, "%s: changing new_master from %u to %u\n", + dlm->name, dlm->reco.new_master, master); + dlm->reco.new_master = master; +} + +static inline void __dlm_reset_recovery(struct dlm_ctxt *dlm) +{ + assert_spin_locked(&dlm->spinlock); + clear_bit(dlm->reco.dead_node, dlm->recovery_map); + dlm_set_reco_dead_node(dlm, O2NM_INVALID_NODE_NUM); + dlm_set_reco_master(dlm, O2NM_INVALID_NODE_NUM); +} + static inline void dlm_reset_recovery(struct dlm_ctxt *dlm) { spin_lock(&dlm->spinlock); - clear_bit(dlm->reco.dead_node, dlm->recovery_map); - dlm->reco.dead_node = O2NM_INVALID_NODE_NUM; - dlm->reco.new_master = O2NM_INVALID_NODE_NUM; + __dlm_reset_recovery(dlm); spin_unlock(&dlm->spinlock); } /* Worker function used during recovery. */ -void dlm_dispatch_work(void *data) +void dlm_dispatch_work(struct work_struct *work) { - struct dlm_ctxt *dlm = (struct dlm_ctxt *)data; + struct dlm_ctxt *dlm = + container_of(work, struct dlm_ctxt, dispatched_work); LIST_HEAD(tmp_list); - struct list_head *iter, *iter2; - struct dlm_work_item *item; + struct dlm_work_item *item, *next; dlm_workfunc_t *workfunc; + int tot=0; spin_lock(&dlm->work_lock); list_splice_init(&dlm->work_list, &tmp_list); spin_unlock(&dlm->work_lock); - list_for_each_safe(iter, iter2, &tmp_list) { - item = list_entry(iter, struct dlm_work_item, list); + list_for_each_entry(item, &tmp_list, list) { + tot++; + } + mlog(0, "%s: work thread has %d work items\n", dlm->name, tot); + + list_for_each_entry_safe(item, next, &tmp_list, list) { workfunc = item->func; list_del_init(&item->list); @@ -164,7 +188,7 @@ void dlm_dispatch_work(void *data) * RECOVERY THREAD */ -static void dlm_kick_recovery_thread(struct dlm_ctxt *dlm) +void dlm_kick_recovery_thread(struct dlm_ctxt *dlm) { /* wake the recovery thread * this will wake the reco thread in one of three places @@ -225,6 +249,52 @@ void dlm_complete_recovery_thread(struct dlm_ctxt *dlm) * */ +static void dlm_print_reco_node_status(struct dlm_ctxt *dlm) +{ + struct dlm_reco_node_data *ndata; + struct dlm_lock_resource *res; + + mlog(ML_NOTICE, "%s(%d): recovery info, state=%s, dead=%u, master=%u\n", + dlm->name, task_pid_nr(dlm->dlm_reco_thread_task), + dlm->reco.state & DLM_RECO_STATE_ACTIVE ? "ACTIVE" : "inactive", + dlm->reco.dead_node, dlm->reco.new_master); + + list_for_each_entry(ndata, &dlm->reco.node_data, list) { + char *st = "unknown"; + switch (ndata->state) { + case DLM_RECO_NODE_DATA_INIT: + st = "init"; + break; + case DLM_RECO_NODE_DATA_REQUESTING: + st = "requesting"; + break; + case DLM_RECO_NODE_DATA_DEAD: + st = "dead"; + break; + case DLM_RECO_NODE_DATA_RECEIVING: + st = "receiving"; + break; + case DLM_RECO_NODE_DATA_REQUESTED: + st = "requested"; + break; + case DLM_RECO_NODE_DATA_DONE: + st = "done"; + break; + case DLM_RECO_NODE_DATA_FINALIZE_SENT: + st = "finalize-sent"; + break; + default: + st = "bad"; + break; + } + mlog(ML_NOTICE, "%s: reco state, node %u, state=%s\n", + dlm->name, ndata->node_num, st); + } + list_for_each_entry(res, &dlm->reco.resources, recovering) { + mlog(ML_NOTICE, "%s: lockres %.*s on recovering list\n", + dlm->name, res->lockname.len, res->lockname.name); + } +} #define DLM_RECO_THREAD_TIMEOUT_MS (5 * 1000) @@ -237,7 +307,7 @@ static int dlm_recovery_thread(void *data) mlog(0, "dlm thread running for %s...\n", dlm->name); while (!kthread_should_stop()) { - if (dlm_joined(dlm)) { + if (dlm_domain_fully_joined(dlm)) { status = dlm_do_recovery(dlm); if (status == -EAGAIN) { /* do not sleep, recheck immediately. */ @@ -256,6 +326,73 @@ static int dlm_recovery_thread(void *data) return 0; } +/* returns true when the recovery master has contacted us */ +static int dlm_reco_master_ready(struct dlm_ctxt *dlm) +{ + int ready; + spin_lock(&dlm->spinlock); + ready = (dlm->reco.new_master != O2NM_INVALID_NODE_NUM); + spin_unlock(&dlm->spinlock); + return ready; +} + +/* returns true if node is no longer in the domain + * could be dead or just not joined */ +int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node) +{ + int dead; + spin_lock(&dlm->spinlock); + dead = !test_bit(node, dlm->domain_map); + spin_unlock(&dlm->spinlock); + return dead; +} + +/* returns true if node is no longer in the domain + * could be dead or just not joined */ +static int dlm_is_node_recovered(struct dlm_ctxt *dlm, u8 node) +{ + int recovered; + spin_lock(&dlm->spinlock); + recovered = !test_bit(node, dlm->recovery_map); + spin_unlock(&dlm->spinlock); + return recovered; +} + + +void dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout) +{ + if (dlm_is_node_dead(dlm, node)) + return; + + printk(KERN_NOTICE "o2dlm: Waiting on the death of node %u in " + "domain %s\n", node, dlm->name); + + if (timeout) + wait_event_timeout(dlm->dlm_reco_thread_wq, + dlm_is_node_dead(dlm, node), + msecs_to_jiffies(timeout)); + else + wait_event(dlm->dlm_reco_thread_wq, + dlm_is_node_dead(dlm, node)); +} + +void dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout) +{ + if (dlm_is_node_recovered(dlm, node)) + return; + + printk(KERN_NOTICE "o2dlm: Waiting on the recovery of node %u in " + "domain %s\n", node, dlm->name); + + if (timeout) + wait_event_timeout(dlm->dlm_reco_thread_wq, + dlm_is_node_recovered(dlm, node), + msecs_to_jiffies(timeout)); + else + wait_event(dlm->dlm_reco_thread_wq, + dlm_is_node_recovered(dlm, node)); +} + /* callers of the top-level api calls (dlmlock/dlmunlock) should * block on the dlm->reco.event when recovery is in progress. * the dlm recovery thread will set this state when it begins @@ -274,6 +411,13 @@ static int dlm_in_recovery(struct dlm_ctxt *dlm) void dlm_wait_for_recovery(struct dlm_ctxt *dlm) { + if (dlm_in_recovery(dlm)) { + mlog(0, "%s: reco thread %d in recovery: " + "state=%d, master=%u, dead=%u\n", + dlm->name, task_pid_nr(dlm->dlm_reco_thread_task), + dlm->reco.state, dlm->reco.new_master, + dlm->reco.dead_node); + } wait_event(dlm->reco.event, !dlm_in_recovery(dlm)); } @@ -281,6 +425,8 @@ static void dlm_begin_recovery(struct dlm_ctxt *dlm) { spin_lock(&dlm->spinlock); BUG_ON(dlm->reco.state & DLM_RECO_STATE_ACTIVE); + printk(KERN_NOTICE "o2dlm: Begin recovery on domain %s for node %u\n", + dlm->name, dlm->reco.dead_node); dlm->reco.state |= DLM_RECO_STATE_ACTIVE; spin_unlock(&dlm->spinlock); } @@ -291,12 +437,22 @@ static void dlm_end_recovery(struct dlm_ctxt *dlm) BUG_ON(!(dlm->reco.state & DLM_RECO_STATE_ACTIVE)); dlm->reco.state &= ~DLM_RECO_STATE_ACTIVE; spin_unlock(&dlm->spinlock); + printk(KERN_NOTICE "o2dlm: End recovery on domain %s\n", dlm->name); wake_up(&dlm->reco.event); } +static void dlm_print_recovery_master(struct dlm_ctxt *dlm) +{ + printk(KERN_NOTICE "o2dlm: Node %u (%s) is the Recovery Master for the " + "dead node %u in domain %s\n", dlm->reco.new_master, + (dlm->node_num == dlm->reco.new_master ? "me" : "he"), + dlm->reco.dead_node, dlm->name); +} + static int dlm_do_recovery(struct dlm_ctxt *dlm) { int status = 0; + int ret; spin_lock(&dlm->spinlock); @@ -306,23 +462,23 @@ static int dlm_do_recovery(struct dlm_ctxt *dlm) mlog(0, "new master %u died while recovering %u!\n", dlm->reco.new_master, dlm->reco.dead_node); /* unset the new_master, leave dead_node */ - dlm->reco.new_master = O2NM_INVALID_NODE_NUM; + dlm_set_reco_master(dlm, O2NM_INVALID_NODE_NUM); } /* select a target to recover */ if (dlm->reco.dead_node == O2NM_INVALID_NODE_NUM) { int bit; - bit = find_next_bit (dlm->recovery_map, O2NM_MAX_NODES+1, 0); + bit = find_next_bit (dlm->recovery_map, O2NM_MAX_NODES, 0); if (bit >= O2NM_MAX_NODES || bit < 0) - dlm->reco.dead_node = O2NM_INVALID_NODE_NUM; + dlm_set_reco_dead_node(dlm, O2NM_INVALID_NODE_NUM); else - dlm->reco.dead_node = bit; + dlm_set_reco_dead_node(dlm, bit); } else if (!test_bit(dlm->reco.dead_node, dlm->recovery_map)) { /* BUG? */ mlog(ML_ERROR, "dead_node %u no longer in recovery map!\n", dlm->reco.dead_node); - dlm->reco.dead_node = O2NM_INVALID_NODE_NUM; + dlm_set_reco_dead_node(dlm, O2NM_INVALID_NODE_NUM); } if (dlm->reco.dead_node == O2NM_INVALID_NODE_NUM) { @@ -331,7 +487,8 @@ static int dlm_do_recovery(struct dlm_ctxt *dlm) /* return to main thread loop and sleep. */ return 0; } - mlog(0, "recovery thread found node %u in the recovery map!\n", + mlog(0, "%s(%d):recovery thread found node %u in the recovery map!\n", + dlm->name, task_pid_nr(dlm->dlm_reco_thread_task), dlm->reco.dead_node); spin_unlock(&dlm->spinlock); @@ -343,17 +500,19 @@ static int dlm_do_recovery(struct dlm_ctxt *dlm) goto master_here; if (dlm->reco.new_master == O2NM_INVALID_NODE_NUM) { - /* choose a new master */ - if (!dlm_pick_recovery_master(dlm)) { + /* choose a new master, returns 0 if this node + * is the master, -EEXIST if it's another node. + * this does not return until a new master is chosen + * or recovery completes entirely. */ + ret = dlm_pick_recovery_master(dlm); + if (!ret) { /* already notified everyone. go. */ - dlm->reco.new_master = dlm->node_num; goto master_here; } mlog(0, "another node will master this recovery session.\n"); } - mlog(0, "dlm=%s, new_master=%u, this node=%u, dead_node=%u\n", - dlm->name, dlm->reco.new_master, - dlm->node_num, dlm->reco.dead_node); + + dlm_print_recovery_master(dlm); /* it is safe to start everything back up here * because all of the dead node's lock resources @@ -364,16 +523,24 @@ static int dlm_do_recovery(struct dlm_ctxt *dlm) return 0; master_here: - mlog(0, "mastering recovery of %s:%u here(this=%u)!\n", - dlm->name, dlm->reco.dead_node, dlm->node_num); + dlm_print_recovery_master(dlm); status = dlm_remaster_locks(dlm, dlm->reco.dead_node); if (status < 0) { - mlog(ML_ERROR, "error %d remastering locks for node %u, " - "retrying.\n", status, dlm->reco.dead_node); + /* we should never hit this anymore */ + mlog(ML_ERROR, "%s: Error %d remastering locks for node %u, " + "retrying.\n", dlm->name, status, dlm->reco.dead_node); + /* yield a bit to allow any final network messages + * to get handled on remaining nodes */ + msleep(100); } else { /* success! see if any other nodes need recovery */ - dlm_reset_recovery(dlm); + mlog(0, "DONE mastering recovery of %s:%u here(this=%u)!\n", + dlm->name, dlm->reco.dead_node, dlm->node_num); + spin_lock(&dlm->spinlock); + __dlm_reset_recovery(dlm); + dlm->reco.state &= ~DLM_RECO_STATE_FINALIZE; + spin_unlock(&dlm->spinlock); } dlm_end_recovery(dlm); @@ -385,23 +552,28 @@ static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node) { int status = 0; struct dlm_reco_node_data *ndata; - struct list_head *iter; int all_nodes_done; int destroy = 0; int pass = 0; - status = dlm_init_recovery_area(dlm, dead_node); - if (status < 0) - goto leave; + do { + /* we have become recovery master. there is no escaping + * this, so just keep trying until we get it. */ + status = dlm_init_recovery_area(dlm, dead_node); + if (status < 0) { + mlog(ML_ERROR, "%s: failed to alloc recovery area, " + "retrying\n", dlm->name); + msleep(1000); + } + } while (status != 0); /* safe to access the node data list without a lock, since this * process is the only one to change the list */ - list_for_each(iter, &dlm->reco.node_data) { - ndata = list_entry (iter, struct dlm_reco_node_data, list); + list_for_each_entry(ndata, &dlm->reco.node_data, list) { BUG_ON(ndata->state != DLM_RECO_NODE_DATA_INIT); ndata->state = DLM_RECO_NODE_DATA_REQUESTING; - mlog(0, "requesting lock info from node %u\n", + mlog(0, "%s: Requesting lock info from node %u\n", dlm->name, ndata->node_num); if (ndata->node_num == dlm->node_num) { @@ -409,17 +581,38 @@ static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node) continue; } - status = dlm_request_all_locks(dlm, ndata->node_num, dead_node); - if (status < 0) { - mlog_errno(status); - if (dlm_is_host_down(status)) - ndata->state = DLM_RECO_NODE_DATA_DEAD; - else { - destroy = 1; - goto leave; + do { + status = dlm_request_all_locks(dlm, ndata->node_num, + dead_node); + if (status < 0) { + mlog_errno(status); + if (dlm_is_host_down(status)) { + /* node died, ignore it for recovery */ + status = 0; + ndata->state = DLM_RECO_NODE_DATA_DEAD; + /* wait for the domain map to catch up + * with the network state. */ + wait_event_timeout(dlm->dlm_reco_thread_wq, + dlm_is_node_dead(dlm, + ndata->node_num), + msecs_to_jiffies(1000)); + mlog(0, "waited 1 sec for %u, " + "dead? %s\n", ndata->node_num, + dlm_is_node_dead(dlm, ndata->node_num) ? + "yes" : "no"); + } else { + /* -ENOMEM on the other node */ + mlog(0, "%s: node %u returned " + "%d during recovery, retrying " + "after a short wait\n", + dlm->name, ndata->node_num, + status); + msleep(100); + } } - } + } while (status != 0); + spin_lock(&dlm_reco_state_lock); switch (ndata->state) { case DLM_RECO_NODE_DATA_INIT: case DLM_RECO_NODE_DATA_FINALIZE_SENT: @@ -430,10 +623,9 @@ static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node) mlog(0, "node %u died after requesting " "recovery info for node %u\n", ndata->node_num, dead_node); - // start all over - destroy = 1; - status = -EAGAIN; - goto leave; + /* fine. don't need this node's info. + * continue without it. */ + break; case DLM_RECO_NODE_DATA_REQUESTING: ndata->state = DLM_RECO_NODE_DATA_REQUESTED; mlog(0, "now receiving recovery data from " @@ -451,9 +643,10 @@ static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node) ndata->node_num, dead_node); break; } + spin_unlock(&dlm_reco_state_lock); } - mlog(0, "done requesting all lock info\n"); + mlog(0, "%s: Done requesting all lock info\n", dlm->name); /* nodes should be sending reco data now * just need to wait */ @@ -463,9 +656,7 @@ static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node) * done, or if anyone died */ all_nodes_done = 1; spin_lock(&dlm_reco_state_lock); - list_for_each(iter, &dlm->reco.node_data) { - ndata = list_entry (iter, struct dlm_reco_node_data, list); - + list_for_each_entry(ndata, &dlm->reco.node_data, list) { mlog(0, "checking recovery state of node %u\n", ndata->node_num); switch (ndata->state) { @@ -481,18 +672,22 @@ static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node) "requesting recovery info for " "node %u\n", ndata->node_num, dead_node); - spin_unlock(&dlm_reco_state_lock); - // start all over - destroy = 1; - status = -EAGAIN; - goto leave; + break; case DLM_RECO_NODE_DATA_RECEIVING: case DLM_RECO_NODE_DATA_REQUESTED: + mlog(0, "%s: node %u still in state %s\n", + dlm->name, ndata->node_num, + ndata->state==DLM_RECO_NODE_DATA_RECEIVING ? + "receiving" : "requested"); all_nodes_done = 0; break; case DLM_RECO_NODE_DATA_DONE: + mlog(0, "%s: node %u state is done\n", + dlm->name, ndata->node_num); break; case DLM_RECO_NODE_DATA_FINALIZE_SENT: + mlog(0, "%s: node %u state is finalize\n", + dlm->name, ndata->node_num); break; } } @@ -503,6 +698,14 @@ static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node) if (all_nodes_done) { int ret; + /* Set this flag on recovery master to avoid + * a new recovery for another dead node start + * before the recovery is not done. That may + * cause recovery hung.*/ + spin_lock(&dlm->spinlock); + dlm->reco.state |= DLM_RECO_STATE_FINALIZE; + spin_unlock(&dlm->spinlock); + /* all nodes are now in DLM_RECO_NODE_DATA_DONE state * just send a finalize message to everyone and * clean up */ @@ -522,7 +725,7 @@ static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node) jiffies, dlm->reco.dead_node, dlm->node_num, dlm->reco.new_master); destroy = 1; - status = ret; + status = 0; /* rescan everything marked dirty along the way */ dlm_kick_thread(dlm, NULL); break; @@ -535,11 +738,9 @@ static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node) } -leave: if (destroy) dlm_destroy_recovery_area(dlm, dead_node); - mlog_exit(status); return status; } @@ -561,7 +762,7 @@ static int dlm_init_recovery_area(struct dlm_ctxt *dlm, u8 dead_node) } BUG_ON(num == dead_node); - ndata = kcalloc(1, sizeof(*ndata), GFP_KERNEL); + ndata = kzalloc(sizeof(*ndata), GFP_NOFS); if (!ndata) { dlm_destroy_recovery_area(dlm, dead_node); return -ENOMEM; @@ -579,16 +780,14 @@ static int dlm_init_recovery_area(struct dlm_ctxt *dlm, u8 dead_node) static void dlm_destroy_recovery_area(struct dlm_ctxt *dlm, u8 dead_node) { - struct list_head *iter, *iter2; - struct dlm_reco_node_data *ndata; + struct dlm_reco_node_data *ndata, *next; LIST_HEAD(tmplist); spin_lock(&dlm_reco_state_lock); list_splice_init(&dlm->reco.node_data, &tmplist); spin_unlock(&dlm_reco_state_lock); - list_for_each_safe(iter, iter2, &tmplist) { - ndata = list_entry (iter, struct dlm_reco_node_data, list); + list_for_each_entry_safe(ndata, next, &tmplist, list) { list_del_init(&ndata->list); kfree(ndata); } @@ -598,7 +797,8 @@ static int dlm_request_all_locks(struct dlm_ctxt *dlm, u8 request_from, u8 dead_node) { struct dlm_lock_request lr; - enum dlm_status ret; + int ret; + int status; mlog(0, "\n"); @@ -611,21 +811,24 @@ static int dlm_request_all_locks(struct dlm_ctxt *dlm, u8 request_from, lr.dead_node = dead_node; // send message - ret = DLM_NOLOCKMGR; ret = o2net_send_message(DLM_LOCK_REQUEST_MSG, dlm->key, - &lr, sizeof(lr), request_from, NULL); + &lr, sizeof(lr), request_from, &status); /* negative status is handled by caller */ if (ret < 0) - mlog_errno(ret); - + mlog(ML_ERROR, "%s: Error %d send LOCK_REQUEST to node %u " + "to recover dead node %u\n", dlm->name, ret, + request_from, dead_node); + else + ret = status; // return from here, then // sleep until all received or error return ret; } -int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data) +int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { struct dlm_ctxt *dlm = data; struct dlm_lock_request *lr = (struct dlm_lock_request *)msg->buf; @@ -635,16 +838,25 @@ int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data) if (!dlm_grab(dlm)) return -EINVAL; + if (lr->dead_node != dlm->reco.dead_node) { + mlog(ML_ERROR, "%s: node %u sent dead_node=%u, but local " + "dead_node is %u\n", dlm->name, lr->node_idx, + lr->dead_node, dlm->reco.dead_node); + dlm_print_reco_node_status(dlm); + /* this is a hack */ + dlm_put(dlm); + return -ENOMEM; + } BUG_ON(lr->dead_node != dlm->reco.dead_node); - item = kcalloc(1, sizeof(*item), GFP_KERNEL); + item = kzalloc(sizeof(*item), GFP_NOFS); if (!item) { dlm_put(dlm); return -ENOMEM; } /* this will get freed by dlm_request_all_locks_worker */ - buf = (char *) __get_free_page(GFP_KERNEL); + buf = (char *) __get_free_page(GFP_NOFS); if (!buf) { kfree(item); dlm_put(dlm); @@ -659,7 +871,7 @@ int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data) spin_lock(&dlm->work_lock); list_add_tail(&item->list, &dlm->work_list); spin_unlock(&dlm->work_lock); - schedule_work(&dlm->dispatched_work); + queue_work(dlm->dlm_worker, &dlm->dispatched_work); dlm_put(dlm); return 0; @@ -671,18 +883,37 @@ static void dlm_request_all_locks_worker(struct dlm_work_item *item, void *data) struct dlm_lock_resource *res; struct dlm_ctxt *dlm; LIST_HEAD(resources); - struct list_head *iter; int ret; u8 dead_node, reco_master; + int skip_all_done = 0; dlm = item->dlm; dead_node = item->u.ral.dead_node; reco_master = item->u.ral.reco_master; - BUG_ON(dead_node != dlm->reco.dead_node); - BUG_ON(reco_master != dlm->reco.new_master); - mres = (struct dlm_migratable_lockres *)data; + mlog(0, "%s: recovery worker started, dead=%u, master=%u\n", + dlm->name, dead_node, reco_master); + + if (dead_node != dlm->reco.dead_node || + reco_master != dlm->reco.new_master) { + /* worker could have been created before the recovery master + * died. if so, do not continue, but do not error. */ + if (dlm->reco.new_master == O2NM_INVALID_NODE_NUM) { + mlog(ML_NOTICE, "%s: will not send recovery state, " + "recovery master %u died, thread=(dead=%u,mas=%u)" + " current=(dead=%u,mas=%u)\n", dlm->name, + reco_master, dead_node, reco_master, + dlm->reco.dead_node, dlm->reco.new_master); + } else { + mlog(ML_NOTICE, "%s: reco state invalid: reco(dead=%u, " + "master=%u), request(dead=%u, master=%u)\n", + dlm->name, dlm->reco.dead_node, + dlm->reco.new_master, dead_node, reco_master); + } + goto leave; + } + /* lock resources should have already been moved to the * dlm->reco.resources list. now move items from that list * to a temp list if the dead owner matches. note that the @@ -692,12 +923,19 @@ static void dlm_request_all_locks_worker(struct dlm_work_item *item, void *data) dlm_move_reco_locks_to_list(dlm, &resources, dead_node); /* now we can begin blasting lockreses without the dlm lock */ - list_for_each(iter, &resources) { - res = list_entry (iter, struct dlm_lock_resource, recovering); + + /* any errors returned will be due to the new_master dying, + * the dlm_reco_thread should detect this */ + list_for_each_entry(res, &resources, recovering) { ret = dlm_send_one_lockres(dlm, res, mres, reco_master, DLM_MRES_RECOVERY); - if (ret < 0) - mlog_errno(ret); + if (ret < 0) { + mlog(ML_ERROR, "%s: node %u went down while sending " + "recovery state for dead node %u, ret=%d\n", dlm->name, + reco_master, dead_node, ret); + skip_all_done = 1; + break; + } } /* move the resources back to the list */ @@ -705,10 +943,15 @@ static void dlm_request_all_locks_worker(struct dlm_work_item *item, void *data) list_splice_init(&resources, &dlm->reco.resources); spin_unlock(&dlm->spinlock); - ret = dlm_send_all_done_msg(dlm, dead_node, reco_master); - if (ret < 0) - mlog_errno(ret); - + if (!skip_all_done) { + ret = dlm_send_all_done_msg(dlm, dead_node, reco_master); + if (ret < 0) { + mlog(ML_ERROR, "%s: node %u went down while sending " + "recovery all-done for dead node %u, ret=%d\n", + dlm->name, reco_master, dead_node, ret); + } + } +leave: free_page((unsigned long)data); } @@ -727,18 +970,24 @@ static int dlm_send_all_done_msg(struct dlm_ctxt *dlm, u8 dead_node, u8 send_to) ret = o2net_send_message(DLM_RECO_DATA_DONE_MSG, dlm->key, &done_msg, sizeof(done_msg), send_to, &tmpret); - /* negative status is ignored by the caller */ - if (ret >= 0) + if (ret < 0) { + mlog(ML_ERROR, "%s: Error %d send RECO_DATA_DONE to node %u " + "to recover dead node %u\n", dlm->name, ret, send_to, + dead_node); + if (!dlm_is_host_down(ret)) { + BUG(); + } + } else ret = tmpret; return ret; } -int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data) +int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { struct dlm_ctxt *dlm = data; struct dlm_reco_data_done *done = (struct dlm_reco_data_done *)msg->buf; - struct list_head *iter; struct dlm_reco_node_data *ndata = NULL; int ret = -EINVAL; @@ -748,24 +997,30 @@ int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data) mlog(0, "got DATA DONE: dead_node=%u, reco.dead_node=%u, " "node_idx=%u, this node=%u\n", done->dead_node, dlm->reco.dead_node, done->node_idx, dlm->node_num); - BUG_ON(done->dead_node != dlm->reco.dead_node); + + mlog_bug_on_msg((done->dead_node != dlm->reco.dead_node), + "Got DATA DONE: dead_node=%u, reco.dead_node=%u, " + "node_idx=%u, this node=%u\n", done->dead_node, + dlm->reco.dead_node, done->node_idx, dlm->node_num); spin_lock(&dlm_reco_state_lock); - list_for_each(iter, &dlm->reco.node_data) { - ndata = list_entry (iter, struct dlm_reco_node_data, list); + list_for_each_entry(ndata, &dlm->reco.node_data, list) { if (ndata->node_num != done->node_idx) continue; switch (ndata->state) { + /* should have moved beyond INIT but not to FINALIZE yet */ case DLM_RECO_NODE_DATA_INIT: case DLM_RECO_NODE_DATA_DEAD: - case DLM_RECO_NODE_DATA_DONE: case DLM_RECO_NODE_DATA_FINALIZE_SENT: mlog(ML_ERROR, "bad ndata state for node %u:" " state=%d\n", ndata->node_num, ndata->state); BUG(); break; + /* these states are possible at this point, anywhere along + * the line of recovery */ + case DLM_RECO_NODE_DATA_DONE: case DLM_RECO_NODE_DATA_RECEIVING: case DLM_RECO_NODE_DATA_REQUESTED: case DLM_RECO_NODE_DATA_REQUESTING: @@ -797,26 +1052,40 @@ static void dlm_move_reco_locks_to_list(struct dlm_ctxt *dlm, struct list_head *list, u8 dead_node) { - struct dlm_lock_resource *res; - struct list_head *iter, *iter2; + struct dlm_lock_resource *res, *next; + struct dlm_lock *lock; spin_lock(&dlm->spinlock); - list_for_each_safe(iter, iter2, &dlm->reco.resources) { - res = list_entry (iter, struct dlm_lock_resource, recovering); + list_for_each_entry_safe(res, next, &dlm->reco.resources, recovering) { + /* always prune any $RECOVERY entries for dead nodes, + * otherwise hangs can occur during later recovery */ if (dlm_is_recovery_lock(res->lockname.name, - res->lockname.len)) + res->lockname.len)) { + spin_lock(&res->spinlock); + list_for_each_entry(lock, &res->granted, list) { + if (lock->ml.node == dead_node) { + mlog(0, "AHA! there was " + "a $RECOVERY lock for dead " + "node %u (%s)!\n", + dead_node, dlm->name); + list_del_init(&lock->list); + dlm_lock_put(lock); + break; + } + } + spin_unlock(&res->spinlock); continue; + } + if (res->owner == dead_node) { mlog(0, "found lockres owned by dead node while " "doing recovery for node %u. sending it.\n", dead_node); - list_del_init(&res->recovering); - list_add_tail(&res->recovering, list); + list_move_tail(&res->recovering, list); } else if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN) { mlog(0, "found UNKNOWN owner while doing recovery " "for node %u. sending it.\n", dead_node); - list_del_init(&res->recovering); - list_add_tail(&res->recovering, list); + list_move_tail(&res->recovering, list); } } spin_unlock(&dlm->spinlock); @@ -862,13 +1131,22 @@ static int dlm_send_mig_lockres_msg(struct dlm_ctxt *dlm, if (total_locks == mres_total_locks) mres->flags |= DLM_MRES_ALL_DONE; + mlog(0, "%s:%.*s: sending mig lockres (%s) to %u\n", + dlm->name, res->lockname.len, res->lockname.name, + orig_flags & DLM_MRES_MIGRATION ? "migration" : "recovery", + send_to); + /* send it */ ret = o2net_send_message(DLM_MIG_LOCKRES_MSG, dlm->key, mres, sz, send_to, &status); if (ret < 0) { /* XXX: negative status is not handled. * this will end up killing this node. */ - mlog_errno(ret); + mlog(ML_ERROR, "%s: res %.*s, Error %d send MIG_LOCKRES to " + "node %u (%s)\n", dlm->name, mres->lockname_len, + mres->lockname, ret, send_to, + (orig_flags & DLM_MRES_MIGRATION ? + "migration" : "recovery")); } else { /* might get an -ENOMEM back here */ ret = status; @@ -896,7 +1174,7 @@ static void dlm_init_migratable_lockres(struct dlm_migratable_lockres *mres, u8 flags, u8 master) { /* mres here is one full page */ - memset(mres, 0, PAGE_SIZE); + clear_page(mres); mres->lockname_len = namelen; memcpy(mres->lockname, lockname, namelen); mres->num_locks = 0; @@ -906,6 +1184,39 @@ static void dlm_init_migratable_lockres(struct dlm_migratable_lockres *mres, mres->master = master; } +static void dlm_prepare_lvb_for_migration(struct dlm_lock *lock, + struct dlm_migratable_lockres *mres, + int queue) +{ + if (!lock->lksb) + return; + + /* Ignore lvb in all locks in the blocked list */ + if (queue == DLM_BLOCKED_LIST) + return; + + /* Only consider lvbs in locks with granted EX or PR lock levels */ + if (lock->ml.type != LKM_EXMODE && lock->ml.type != LKM_PRMODE) + return; + + if (dlm_lvb_is_empty(mres->lvb)) { + memcpy(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN); + return; + } + + /* Ensure the lvb copied for migration matches in other valid locks */ + if (!memcmp(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN)) + return; + + mlog(ML_ERROR, "Mismatched lvb in lock cookie=%u:%llu, name=%.*s, " + "node=%u\n", + dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)), + lock->lockres->lockname.len, lock->lockres->lockname.name, + lock->ml.node); + dlm_print_one_lock_resource(lock->lockres); + BUG(); +} /* returns 1 if this lock fills the network structure, * 0 otherwise */ @@ -923,19 +1234,7 @@ static int dlm_add_lock_to_array(struct dlm_lock *lock, ml->list = queue; if (lock->lksb) { ml->flags = lock->lksb->flags; - /* send our current lvb */ - if (ml->type == LKM_EXMODE || - ml->type == LKM_PRMODE) { - /* if it is already set, this had better be a PR - * and it has to match */ - if (mres->lvb[0] && (ml->type == LKM_EXMODE || - memcmp(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN))) { - mlog(ML_ERROR, "mismatched lvbs!\n"); - __dlm_print_one_lock_resource(lock->lockres); - BUG(); - } - memcpy(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN); - } + dlm_prepare_lvb_for_migration(lock, mres, queue); } ml->node = lock->ml.node; mres->num_locks++; @@ -945,12 +1244,40 @@ static int dlm_add_lock_to_array(struct dlm_lock *lock, return 0; } +static void dlm_add_dummy_lock(struct dlm_ctxt *dlm, + struct dlm_migratable_lockres *mres) +{ + struct dlm_lock dummy; + memset(&dummy, 0, sizeof(dummy)); + dummy.ml.cookie = 0; + dummy.ml.type = LKM_IVMODE; + dummy.ml.convert_type = LKM_IVMODE; + dummy.ml.highest_blocked = LKM_IVMODE; + dummy.lksb = NULL; + dummy.ml.node = dlm->node_num; + dlm_add_lock_to_array(&dummy, mres, DLM_BLOCKED_LIST); +} + +static inline int dlm_is_dummy_lock(struct dlm_ctxt *dlm, + struct dlm_migratable_lock *ml, + u8 *nodenum) +{ + if (unlikely(ml->cookie == 0 && + ml->type == LKM_IVMODE && + ml->convert_type == LKM_IVMODE && + ml->highest_blocked == LKM_IVMODE && + ml->list == DLM_BLOCKED_LIST)) { + *nodenum = ml->node; + return 1; + } + return 0; +} int dlm_send_one_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, struct dlm_migratable_lockres *mres, u8 send_to, u8 flags) { - struct list_head *queue, *iter; + struct list_head *queue; int total_locks, i; u64 mig_cookie = 0; struct dlm_lock *lock; @@ -976,9 +1303,7 @@ int dlm_send_one_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, total_locks = 0; for (i=DLM_GRANTED_LIST; i<=DLM_BLOCKED_LIST; i++) { queue = dlm_list_idx_to_ptr(res, i); - list_for_each(iter, queue) { - lock = list_entry (iter, struct dlm_lock, list); - + list_for_each_entry(lock, queue, list) { /* add another lock. */ total_locks++; if (!dlm_add_lock_to_array(lock, mres, i)) @@ -988,22 +1313,33 @@ int dlm_send_one_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, * we must send it immediately. */ ret = dlm_send_mig_lockres_msg(dlm, mres, send_to, res, total_locks); - if (ret < 0) { - // TODO - mlog(ML_ERROR, "dlm_send_mig_lockres_msg " - "returned %d, TODO\n", ret); - BUG(); - } + if (ret < 0) + goto error; } } + if (total_locks == 0) { + /* send a dummy lock to indicate a mastery reference only */ + mlog(0, "%s:%.*s: sending dummy lock to %u, %s\n", + dlm->name, res->lockname.len, res->lockname.name, + send_to, flags & DLM_MRES_RECOVERY ? "recovery" : + "migration"); + dlm_add_dummy_lock(dlm, mres); + } /* flush any remaining locks */ ret = dlm_send_mig_lockres_msg(dlm, mres, send_to, res, total_locks); - if (ret < 0) { - // TODO - mlog(ML_ERROR, "dlm_send_mig_lockres_msg returned %d, " - "TODO\n", ret); + if (ret < 0) + goto error; + return ret; + +error: + mlog(ML_ERROR, "%s: dlm_send_mig_lockres_msg returned %d\n", + dlm->name, ret); + if (!dlm_is_host_down(ret)) BUG(); - } + mlog(0, "%s: node %u went down while sending %s " + "lockres %.*s\n", dlm->name, send_to, + flags & DLM_MRES_RECOVERY ? "recovery" : "migration", + res->lockname.len, res->lockname.name); return ret; } @@ -1022,13 +1358,15 @@ int dlm_send_one_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, * do we spin? returning an error only delays the problem really */ -int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data) +int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { struct dlm_ctxt *dlm = data; struct dlm_migratable_lockres *mres = (struct dlm_migratable_lockres *)msg->buf; int ret = 0; u8 real_master; + u8 extra_refs = 0; char *buf = NULL; struct dlm_work_item *item = NULL; struct dlm_lock_resource *res = NULL; @@ -1051,8 +1389,8 @@ int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data) mlog(0, "all done flag. all lockres data received!\n"); ret = -ENOMEM; - buf = kmalloc(be16_to_cpu(msg->data_len), GFP_KERNEL); - item = kcalloc(1, sizeof(*item), GFP_KERNEL); + buf = kmalloc(be16_to_cpu(msg->data_len), GFP_NOFS); + item = kzalloc(sizeof(*item), GFP_NOFS); if (!buf || !item) goto leave; @@ -1080,6 +1418,7 @@ int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data) mres->lockname_len, mres->lockname); ret = -EFAULT; spin_unlock(&res->spinlock); + dlm_lockres_put(res); goto leave; } res->state |= DLM_LOCK_RES_MIGRATING; @@ -1106,22 +1445,38 @@ int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data) __dlm_insert_lockres(dlm, res); spin_unlock(&dlm->spinlock); + /* Add an extra ref for this lock-less lockres lest the + * dlm_thread purges it before we get the chance to add + * locks to it */ + dlm_lockres_get(res); + + /* There are three refs that need to be put. + * 1. Taken above. + * 2. kref_init in dlm_new_lockres()->dlm_init_lockres(). + * 3. dlm_lookup_lockres() + * The first one is handled at the end of this function. The + * other two are handled in the worker thread after locks have + * been attached. Yes, we don't wait for purge time to match + * kref_init. The lockres will still have atleast one ref + * added because it is in the hash __dlm_insert_lockres() */ + extra_refs++; + /* now that the new lockres is inserted, * make it usable by other processes */ spin_lock(&res->spinlock); res->state &= ~DLM_LOCK_RES_IN_PROGRESS; spin_unlock(&res->spinlock); - - /* add an extra ref for just-allocated lockres - * otherwise the lockres will be purged immediately */ - dlm_lockres_get(res); - + wake_up(&res->wq); } /* at this point we have allocated everything we need, * and we have a hashed lockres with an extra ref and * the proper res->state flags. */ ret = 0; + spin_lock(&res->spinlock); + /* drop this either when master requery finds a different master + * or when a lock is added by the recovery worker */ + dlm_lockres_grab_inflight_ref(dlm, res); if (mres->master == DLM_LOCK_RES_OWNER_UNKNOWN) { /* migration cannot have an unknown master */ BUG_ON(!(mres->flags & DLM_MRES_RECOVERY)); @@ -1129,10 +1484,11 @@ int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data) "unknown owner.. will need to requery: " "%.*s\n", mres->lockname_len, mres->lockname); } else { - spin_lock(&res->spinlock); + /* take a reference now to pin the lockres, drop it + * when locks are added in the worker */ dlm_change_lockres_owner(dlm, res, dlm->node_num); - spin_unlock(&res->spinlock); } + spin_unlock(&res->spinlock); /* queue up work for dlm_mig_lockres_worker */ dlm_grab(dlm); /* get an extra ref for the work item */ @@ -1140,38 +1496,43 @@ int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data) dlm_init_work_item(dlm, item, dlm_mig_lockres_worker, buf); item->u.ml.lockres = res; /* already have a ref */ item->u.ml.real_master = real_master; + item->u.ml.extra_ref = extra_refs; spin_lock(&dlm->work_lock); list_add_tail(&item->list, &dlm->work_list); spin_unlock(&dlm->work_lock); - schedule_work(&dlm->dispatched_work); + queue_work(dlm->dlm_worker, &dlm->dispatched_work); leave: + /* One extra ref taken needs to be put here */ + if (extra_refs) + dlm_lockres_put(res); + dlm_put(dlm); if (ret < 0) { - if (buf) - kfree(buf); - if (item) - kfree(item); + kfree(buf); + kfree(item); + mlog_errno(ret); } - mlog_exit(ret); return ret; } static void dlm_mig_lockres_worker(struct dlm_work_item *item, void *data) { - struct dlm_ctxt *dlm = data; + struct dlm_ctxt *dlm; struct dlm_migratable_lockres *mres; int ret = 0; struct dlm_lock_resource *res; u8 real_master; + u8 extra_ref; dlm = item->dlm; mres = (struct dlm_migratable_lockres *)data; res = item->u.ml.lockres; real_master = item->u.ml.real_master; + extra_ref = item->u.ml.extra_ref; if (real_master == DLM_LOCK_RES_OWNER_UNKNOWN) { /* this case is super-rare. only occurs if @@ -1179,7 +1540,7 @@ static void dlm_mig_lockres_worker(struct dlm_work_item *item, void *data) again: ret = dlm_lockres_master_requery(dlm, res, &real_master); if (ret < 0) { - mlog(0, "dlm_lockres_master_requery failure: %d\n", + mlog(0, "dlm_lockres_master_requery ret=%d\n", ret); goto again; } @@ -1188,6 +1549,9 @@ again: "this node will take it.\n", res->lockname.len, res->lockname.name); } else { + spin_lock(&res->spinlock); + dlm_lockres_drop_inflight_ref(dlm, res); + spin_unlock(&res->spinlock); mlog(0, "master needs to respond to sender " "that node %u still owns %.*s\n", real_master, res->lockname.len, @@ -1211,8 +1575,13 @@ again: } leave: + /* See comment in dlm_mig_lockres_handler() */ + if (res) { + if (extra_ref) + dlm_lockres_put(res); + dlm_lockres_put(res); + } kfree(data); - mlog_exit(ret); } @@ -1261,8 +1630,10 @@ static int dlm_lockres_master_requery(struct dlm_ctxt *dlm, ret = dlm_do_master_requery(dlm, res, nodenum, real_master); if (ret < 0) { mlog_errno(ret); - BUG(); - /* TODO: need to figure a way to restart this */ + if (!dlm_is_host_down(ret)) + BUG(); + /* host is down, so answer for that node would be + * DLM_LOCK_RES_OWNER_UNKNOWN. continue. */ } if (*real_master != DLM_LOCK_RES_OWNER_UNKNOWN) { mlog(0, "lock master is %u\n", *real_master); @@ -1273,9 +1644,8 @@ static int dlm_lockres_master_requery(struct dlm_ctxt *dlm, } -static int dlm_do_master_requery(struct dlm_ctxt *dlm, - struct dlm_lock_resource *res, - u8 nodenum, u8 *real_master) +int dlm_do_master_requery(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, + u8 nodenum, u8 *real_master) { int ret = -EINVAL; struct dlm_master_requery req; @@ -1290,7 +1660,9 @@ static int dlm_do_master_requery(struct dlm_ctxt *dlm, &req, sizeof(req), nodenum, &status); /* XXX: negative status not handled properly here. */ if (ret < 0) - mlog_errno(ret); + mlog(ML_ERROR, "Error %d when sending message %u (key " + "0x%x) to node %u\n", ret, DLM_MASTER_REQUERY_MSG, + dlm->key, nodenum); else { BUG_ON(status < 0); BUG_ON(status > DLM_LOCK_RES_OWNER_UNKNOWN); @@ -1306,11 +1678,13 @@ static int dlm_do_master_requery(struct dlm_ctxt *dlm, /* this function cannot error, so unless the sending * or receiving of the message failed, the owner can * be trusted */ -int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data) +int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { struct dlm_ctxt *dlm = data; struct dlm_master_requery *req = (struct dlm_master_requery *)msg->buf; struct dlm_lock_resource *res = NULL; + unsigned int hash; int master = DLM_LOCK_RES_OWNER_UNKNOWN; u32 flags = DLM_ASSERT_MASTER_REQUERY; @@ -1320,8 +1694,10 @@ int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data) return master; } + hash = dlm_lockid_hash(req->name, req->namelen); + spin_lock(&dlm->spinlock); - res = __dlm_lookup_lockres(dlm, req->name, req->namelen); + res = __dlm_lookup_lockres(dlm, req->name, req->namelen, hash); if (res) { spin_lock(&res->spinlock); master = res->owner; @@ -1332,8 +1708,10 @@ int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data) mlog_errno(-ENOMEM); /* retry!? */ BUG(); - } - } + } else + __dlm_lockres_grab_inflight_worker(dlm, res); + } else /* put.. incase we are not the master */ + dlm_lockres_put(res); spin_unlock(&res->spinlock); } spin_unlock(&dlm->spinlock); @@ -1384,22 +1762,39 @@ static int dlm_process_recovery_data(struct dlm_ctxt *dlm, struct dlm_migratable_lockres *mres) { struct dlm_migratable_lock *ml; - struct list_head *queue; + struct list_head *queue, *iter; + struct list_head *tmpq = NULL; struct dlm_lock *newlock = NULL; struct dlm_lockstatus *lksb = NULL; int ret = 0; - int i; - struct list_head *iter; - struct dlm_lock *lock = NULL; + int i, j, bad; + struct dlm_lock *lock; + u8 from = O2NM_MAX_NODES; + unsigned int added = 0; + __be64 c; mlog(0, "running %d locks for this lockres\n", mres->num_locks); for (i=0; i<mres->num_locks; i++) { ml = &(mres->ml[i]); + + if (dlm_is_dummy_lock(dlm, ml, &from)) { + /* placeholder, just need to set the refmap bit */ + BUG_ON(mres->num_locks != 1); + mlog(0, "%s:%.*s: dummy lock for %u\n", + dlm->name, mres->lockname_len, mres->lockname, + from); + spin_lock(&res->spinlock); + dlm_lockres_set_refmap_bit(dlm, res, from); + spin_unlock(&res->spinlock); + added++; + break; + } BUG_ON(ml->highest_blocked != LKM_IVMODE); newlock = NULL; lksb = NULL; queue = dlm_list_num_to_pointer(res, ml->list); + tmpq = NULL; /* if the lock is for the local node it needs to * be moved to the proper location within the queue. @@ -1408,33 +1803,78 @@ static int dlm_process_recovery_data(struct dlm_ctxt *dlm, /* MIGRATION ONLY! */ BUG_ON(!(mres->flags & DLM_MRES_MIGRATION)); + lock = NULL; spin_lock(&res->spinlock); - list_for_each(iter, queue) { - lock = list_entry (iter, struct dlm_lock, list); - if (lock->ml.cookie != ml->cookie) + for (j = DLM_GRANTED_LIST; j <= DLM_BLOCKED_LIST; j++) { + tmpq = dlm_list_idx_to_ptr(res, j); + list_for_each(iter, tmpq) { + lock = list_entry(iter, + struct dlm_lock, list); + if (lock->ml.cookie == ml->cookie) + break; lock = NULL; - else + } + if (lock) break; } /* lock is always created locally first, and * destroyed locally last. it must be on the list */ if (!lock) { - mlog(ML_ERROR, "could not find local lock " - "with cookie %"MLFu64"!\n", - ml->cookie); + c = ml->cookie; + mlog(ML_ERROR, "Could not find local lock " + "with cookie %u:%llu, node %u, " + "list %u, flags 0x%x, type %d, " + "conv %d, highest blocked %d\n", + dlm_get_lock_cookie_node(be64_to_cpu(c)), + dlm_get_lock_cookie_seq(be64_to_cpu(c)), + ml->node, ml->list, ml->flags, ml->type, + ml->convert_type, ml->highest_blocked); + __dlm_print_one_lock_resource(res); + BUG(); + } + + if (lock->ml.node != ml->node) { + c = lock->ml.cookie; + mlog(ML_ERROR, "Mismatched node# in lock " + "cookie %u:%llu, name %.*s, node %u\n", + dlm_get_lock_cookie_node(be64_to_cpu(c)), + dlm_get_lock_cookie_seq(be64_to_cpu(c)), + res->lockname.len, res->lockname.name, + lock->ml.node); + c = ml->cookie; + mlog(ML_ERROR, "Migrate lock cookie %u:%llu, " + "node %u, list %u, flags 0x%x, type %d, " + "conv %d, highest blocked %d\n", + dlm_get_lock_cookie_node(be64_to_cpu(c)), + dlm_get_lock_cookie_seq(be64_to_cpu(c)), + ml->node, ml->list, ml->flags, ml->type, + ml->convert_type, ml->highest_blocked); + __dlm_print_one_lock_resource(res); BUG(); } - BUG_ON(lock->ml.node != ml->node); + + if (tmpq != queue) { + c = ml->cookie; + mlog(0, "Lock cookie %u:%llu was on list %u " + "instead of list %u for %.*s\n", + dlm_get_lock_cookie_node(be64_to_cpu(c)), + dlm_get_lock_cookie_seq(be64_to_cpu(c)), + j, ml->list, res->lockname.len, + res->lockname.name); + __dlm_print_one_lock_resource(res); + spin_unlock(&res->spinlock); + continue; + } /* see NOTE above about why we do not update * to match the master here */ /* move the lock to its proper place */ /* do not alter lock refcount. switching lists. */ - list_del_init(&lock->list); - list_add_tail(&lock->list, queue); + list_move_tail(&lock->list, queue); spin_unlock(&res->spinlock); + added++; mlog(0, "just reordered a local lock!\n"); continue; @@ -1456,28 +1896,55 @@ static int dlm_process_recovery_data(struct dlm_ctxt *dlm, } lksb->flags |= (ml->flags & (DLM_LKSB_PUT_LVB|DLM_LKSB_GET_LVB)); - - if (mres->lvb[0]) { + + if (ml->type == LKM_NLMODE) + goto skip_lvb; + + /* + * If the lock is in the blocked list it can't have a valid lvb, + * so skip it + */ + if (ml->list == DLM_BLOCKED_LIST) + goto skip_lvb; + + if (!dlm_lvb_is_empty(mres->lvb)) { if (lksb->flags & DLM_LKSB_PUT_LVB) { /* other node was trying to update * lvb when node died. recreate the * lksb with the updated lvb. */ memcpy(lksb->lvb, mres->lvb, DLM_LVB_LEN); + /* the lock resource lvb update must happen + * NOW, before the spinlock is dropped. + * we no longer wait for the AST to update + * the lvb. */ + memcpy(res->lvb, mres->lvb, DLM_LVB_LEN); } else { - /* otherwise, the node is sending its + /* otherwise, the node is sending its * most recent valid lvb info */ BUG_ON(ml->type != LKM_EXMODE && ml->type != LKM_PRMODE); - if (res->lvb[0] && (ml->type == LKM_EXMODE || - memcmp(res->lvb, mres->lvb, DLM_LVB_LEN))) { - mlog(ML_ERROR, "received bad lvb!\n"); - __dlm_print_one_lock_resource(res); - BUG(); + if (!dlm_lvb_is_empty(res->lvb) && + (ml->type == LKM_EXMODE || + memcmp(res->lvb, mres->lvb, DLM_LVB_LEN))) { + int i; + mlog(ML_ERROR, "%s:%.*s: received bad " + "lvb! type=%d\n", dlm->name, + res->lockname.len, + res->lockname.name, ml->type); + printk("lockres lvb=["); + for (i=0; i<DLM_LVB_LEN; i++) + printk("%02x", res->lvb[i]); + printk("]\nmigrated lvb=["); + for (i=0; i<DLM_LVB_LEN; i++) + printk("%02x", mres->lvb[i]); + printk("]\n"); + dlm_print_one_lock_resource(res); + BUG(); } memcpy(res->lvb, mres->lvb, DLM_LVB_LEN); } } - +skip_lvb: /* NOTE: * wrt lock queue ordering and recovery: @@ -1495,21 +1962,62 @@ static int dlm_process_recovery_data(struct dlm_ctxt *dlm, * relative to each other, but clearly *not* * preserved relative to locks from other nodes. */ + bad = 0; spin_lock(&res->spinlock); - dlm_lock_get(newlock); - list_add_tail(&newlock->list, queue); + list_for_each_entry(lock, queue, list) { + if (lock->ml.cookie == ml->cookie) { + c = lock->ml.cookie; + mlog(ML_ERROR, "%s:%.*s: %u:%llu: lock already " + "exists on this lockres!\n", dlm->name, + res->lockname.len, res->lockname.name, + dlm_get_lock_cookie_node(be64_to_cpu(c)), + dlm_get_lock_cookie_seq(be64_to_cpu(c))); + + mlog(ML_NOTICE, "sent lock: type=%d, conv=%d, " + "node=%u, cookie=%u:%llu, queue=%d\n", + ml->type, ml->convert_type, ml->node, + dlm_get_lock_cookie_node(be64_to_cpu(ml->cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(ml->cookie)), + ml->list); + + __dlm_print_one_lock_resource(res); + bad = 1; + break; + } + } + if (!bad) { + dlm_lock_get(newlock); + if (mres->flags & DLM_MRES_RECOVERY && + ml->list == DLM_CONVERTING_LIST && + newlock->ml.type > + newlock->ml.convert_type) { + /* newlock is doing downconvert, add it to the + * head of converting list */ + list_add(&newlock->list, queue); + } else + list_add_tail(&newlock->list, queue); + mlog(0, "%s:%.*s: added lock for node %u, " + "setting refmap bit\n", dlm->name, + res->lockname.len, res->lockname.name, ml->node); + dlm_lockres_set_refmap_bit(dlm, res, ml->node); + added++; + } spin_unlock(&res->spinlock); } mlog(0, "done running all the locks\n"); leave: + /* balance the ref taken when the work was queued */ + spin_lock(&res->spinlock); + dlm_lockres_drop_inflight_ref(dlm, res); + spin_unlock(&res->spinlock); + if (ret < 0) { mlog_errno(ret); if (newlock) dlm_lock_put(newlock); } - mlog_exit(ret); return ret; } @@ -1517,19 +2025,27 @@ void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) { int i; - struct list_head *queue, *iter, *iter2; - struct dlm_lock *lock; + struct list_head *queue; + struct dlm_lock *lock, *next; + assert_spin_locked(&dlm->spinlock); + assert_spin_locked(&res->spinlock); res->state |= DLM_LOCK_RES_RECOVERING; - if (!list_empty(&res->recovering)) + if (!list_empty(&res->recovering)) { + mlog(0, + "Recovering res %s:%.*s, is already on recovery list!\n", + dlm->name, res->lockname.len, res->lockname.name); list_del_init(&res->recovering); + dlm_lockres_put(res); + } + /* We need to hold a reference while on the recovery list */ + dlm_lockres_get(res); list_add_tail(&res->recovering, &dlm->reco.resources); /* find any pending locks and put them back on proper list */ for (i=DLM_BLOCKED_LIST; i>=DLM_GRANTED_LIST; i--) { queue = dlm_list_idx_to_ptr(res, i); - list_for_each_safe(iter, iter2, queue) { - lock = list_entry (iter, struct dlm_lock, list); + list_for_each_entry_safe(lock, next, queue, list) { dlm_lock_get(lock); if (lock->convert_pending) { /* move converting lock back to granted */ @@ -1594,23 +2110,27 @@ static void dlm_finish_local_lockres_recovery(struct dlm_ctxt *dlm, u8 dead_node, u8 new_master) { int i; - struct list_head *iter, *iter2, *bucket; - struct dlm_lock_resource *res; - - mlog_entry_void(); + struct hlist_head *bucket; + struct dlm_lock_resource *res, *next; assert_spin_locked(&dlm->spinlock); - list_for_each_safe(iter, iter2, &dlm->reco.resources) { - res = list_entry (iter, struct dlm_lock_resource, recovering); + list_for_each_entry_safe(res, next, &dlm->reco.resources, recovering) { if (res->owner == dead_node) { + mlog(0, "%s: res %.*s, Changing owner from %u to %u\n", + dlm->name, res->lockname.len, res->lockname.name, + res->owner, new_master); list_del_init(&res->recovering); spin_lock(&res->spinlock); + /* new_master has our reference from + * the lock state sent during recovery */ dlm_change_lockres_owner(dlm, res, new_master); res->state &= ~DLM_LOCK_RES_RECOVERING; - __dlm_dirty_lockres(dlm, res); + if (__dlm_lockres_has_locks(res)) + __dlm_dirty_lockres(dlm, res); spin_unlock(&res->spinlock); wake_up(&res->wq); + dlm_lockres_put(res); } } @@ -1618,33 +2138,33 @@ static void dlm_finish_local_lockres_recovery(struct dlm_ctxt *dlm, * for now we need to run the whole hash, clear * the RECOVERING state and set the owner * if necessary */ - for (i=0; i<DLM_HASH_SIZE; i++) { - bucket = &(dlm->resources[i]); - list_for_each(iter, bucket) { - res = list_entry (iter, struct dlm_lock_resource, list); - if (res->state & DLM_LOCK_RES_RECOVERING) { - if (res->owner == dead_node) { - mlog(0, "(this=%u) res %.*s owner=%u " - "was not on recovering list, but " - "clearing state anyway\n", - dlm->node_num, res->lockname.len, - res->lockname.name, new_master); - } else if (res->owner == dlm->node_num) { - mlog(0, "(this=%u) res %.*s owner=%u " - "was not on recovering list, " - "owner is THIS node, clearing\n", - dlm->node_num, res->lockname.len, - res->lockname.name, new_master); - } else - continue; + for (i = 0; i < DLM_HASH_BUCKETS; i++) { + bucket = dlm_lockres_hash(dlm, i); + hlist_for_each_entry(res, bucket, hash_node) { + if (!(res->state & DLM_LOCK_RES_RECOVERING)) + continue; - spin_lock(&res->spinlock); - dlm_change_lockres_owner(dlm, res, new_master); - res->state &= ~DLM_LOCK_RES_RECOVERING; - __dlm_dirty_lockres(dlm, res); - spin_unlock(&res->spinlock); - wake_up(&res->wq); + if (res->owner != dead_node && + res->owner != dlm->node_num) + continue; + + if (!list_empty(&res->recovering)) { + list_del_init(&res->recovering); + dlm_lockres_put(res); } + + /* new_master has our reference from + * the lock state sent during recovery */ + mlog(0, "%s: res %.*s, Changing owner from %u to %u\n", + dlm->name, res->lockname.len, res->lockname.name, + res->owner, new_master); + spin_lock(&res->spinlock); + dlm_change_lockres_owner(dlm, res, new_master); + res->state &= ~DLM_LOCK_RES_RECOVERING; + if (__dlm_lockres_has_locks(res)) + __dlm_dirty_lockres(dlm, res); + spin_unlock(&res->spinlock); + wake_up(&res->wq); } } } @@ -1663,7 +2183,7 @@ static inline int dlm_lvb_needs_invalidation(struct dlm_lock *lock, int local) static void dlm_revalidate_lvb(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, u8 dead_node) { - struct list_head *iter, *queue; + struct list_head *queue; struct dlm_lock *lock; int blank_lvb = 0, local = 0; int i; @@ -1673,7 +2193,7 @@ static void dlm_revalidate_lvb(struct dlm_ctxt *dlm, assert_spin_locked(&res->spinlock); if (res->owner == dlm->node_num) - /* if this node owned the lockres, and if the dead node + /* if this node owned the lockres, and if the dead node * had an EX when he died, blank out the lvb */ search_node = dead_node; else { @@ -1685,8 +2205,7 @@ static void dlm_revalidate_lvb(struct dlm_ctxt *dlm, for (i=DLM_GRANTED_LIST; i<=DLM_CONVERTING_LIST; i++) { queue = dlm_list_idx_to_ptr(res, i); - list_for_each(iter, queue) { - lock = list_entry (iter, struct dlm_lock, list); + list_for_each_entry(lock, queue, list) { if (lock->ml.node == search_node) { if (dlm_lvb_needs_invalidation(lock, local)) { /* zero the lksb lvb and lockres lvb */ @@ -1707,39 +2226,66 @@ static void dlm_revalidate_lvb(struct dlm_ctxt *dlm, static void dlm_free_dead_locks(struct dlm_ctxt *dlm, struct dlm_lock_resource *res, u8 dead_node) { - struct list_head *iter, *tmpiter; - struct dlm_lock *lock; + struct dlm_lock *lock, *next; + unsigned int freed = 0; /* this node is the lockres master: * 1) remove any stale locks for the dead node - * 2) if the dead node had an EX when he died, blank out the lvb + * 2) if the dead node had an EX when he died, blank out the lvb */ assert_spin_locked(&dlm->spinlock); assert_spin_locked(&res->spinlock); + /* We do two dlm_lock_put(). One for removing from list and the other is + * to force the DLM_UNLOCK_FREE_LOCK action so as to free the locks */ + /* TODO: check pending_asts, pending_basts here */ - list_for_each_safe(iter, tmpiter, &res->granted) { - lock = list_entry (iter, struct dlm_lock, list); + list_for_each_entry_safe(lock, next, &res->granted, list) { if (lock->ml.node == dead_node) { list_del_init(&lock->list); dlm_lock_put(lock); + /* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */ + dlm_lock_put(lock); + freed++; } } - list_for_each_safe(iter, tmpiter, &res->converting) { - lock = list_entry (iter, struct dlm_lock, list); + list_for_each_entry_safe(lock, next, &res->converting, list) { if (lock->ml.node == dead_node) { list_del_init(&lock->list); dlm_lock_put(lock); + /* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */ + dlm_lock_put(lock); + freed++; } } - list_for_each_safe(iter, tmpiter, &res->blocked) { - lock = list_entry (iter, struct dlm_lock, list); + list_for_each_entry_safe(lock, next, &res->blocked, list) { if (lock->ml.node == dead_node) { list_del_init(&lock->list); dlm_lock_put(lock); + /* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */ + dlm_lock_put(lock); + freed++; } } + if (freed) { + mlog(0, "%s:%.*s: freed %u locks for dead node %u, " + "dropping ref from lockres\n", dlm->name, + res->lockname.len, res->lockname.name, freed, dead_node); + if(!test_bit(dead_node, res->refmap)) { + mlog(ML_ERROR, "%s:%.*s: freed %u locks for dead node %u, " + "but ref was not set\n", dlm->name, + res->lockname.len, res->lockname.name, freed, dead_node); + __dlm_print_one_lock_resource(res); + } + dlm_lockres_clear_refmap_bit(dlm, res, dead_node); + } else if (test_bit(dead_node, res->refmap)) { + mlog(0, "%s:%.*s: dead node %u had a ref, but had " + "no locks and had not purged before dying\n", dlm->name, + res->lockname.len, res->lockname.name, dead_node); + dlm_lockres_clear_refmap_bit(dlm, res, dead_node); + } + /* do not kick thread yet */ __dlm_dirty_lockres(dlm, res); } @@ -1753,10 +2299,10 @@ static void dlm_free_dead_locks(struct dlm_ctxt *dlm, static void dlm_do_local_recovery_cleanup(struct dlm_ctxt *dlm, u8 dead_node) { - struct list_head *iter; struct dlm_lock_resource *res; int i; - struct list_head *bucket; + struct hlist_head *bucket; + struct dlm_lock *lock; /* purge any stale mles */ @@ -1776,22 +2322,52 @@ static void dlm_do_local_recovery_cleanup(struct dlm_ctxt *dlm, u8 dead_node) * can be kicked again to see if any ASTs or BASTs * need to be fired as a result. */ - for (i=0; i<DLM_HASH_SIZE; i++) { - bucket = &(dlm->resources[i]); - list_for_each(iter, bucket) { - res = list_entry (iter, struct dlm_lock_resource, list); + for (i = 0; i < DLM_HASH_BUCKETS; i++) { + bucket = dlm_lockres_hash(dlm, i); + hlist_for_each_entry(res, bucket, hash_node) { + /* always prune any $RECOVERY entries for dead nodes, + * otherwise hangs can occur during later recovery */ if (dlm_is_recovery_lock(res->lockname.name, - res->lockname.len)) + res->lockname.len)) { + spin_lock(&res->spinlock); + list_for_each_entry(lock, &res->granted, list) { + if (lock->ml.node == dead_node) { + mlog(0, "AHA! there was " + "a $RECOVERY lock for dead " + "node %u (%s)!\n", + dead_node, dlm->name); + list_del_init(&lock->list); + dlm_lock_put(lock); + break; + } + } + spin_unlock(&res->spinlock); continue; - + } spin_lock(&res->spinlock); /* zero the lvb if necessary */ dlm_revalidate_lvb(dlm, res, dead_node); - if (res->owner == dead_node) - dlm_move_lockres_to_recovery_list(dlm, res); - else if (res->owner == dlm->node_num) { + if (res->owner == dead_node) { + if (res->state & DLM_LOCK_RES_DROPPING_REF) { + mlog(ML_NOTICE, "%s: res %.*s, Skip " + "recovery as it is being freed\n", + dlm->name, res->lockname.len, + res->lockname.name); + } else + dlm_move_lockres_to_recovery_list(dlm, + res); + + } else if (res->owner == dlm->node_num) { dlm_free_dead_locks(dlm, res, dead_node); __dlm_lockres_calc_usage(dlm, res); + } else if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN) { + if (test_bit(dead_node, res->refmap)) { + mlog(0, "%s:%.*s: dead node %u had a ref, but had " + "no locks and had not purged before dying\n", + dlm->name, res->lockname.len, + res->lockname.name, dead_node); + dlm_lockres_clear_refmap_bit(dlm, res, dead_node); + } } spin_unlock(&res->spinlock); } @@ -1803,6 +2379,26 @@ static void __dlm_hb_node_down(struct dlm_ctxt *dlm, int idx) { assert_spin_locked(&dlm->spinlock); + if (dlm->reco.new_master == idx) { + mlog(0, "%s: recovery master %d just died\n", + dlm->name, idx); + if (dlm->reco.state & DLM_RECO_STATE_FINALIZE) { + /* finalize1 was reached, so it is safe to clear + * the new_master and dead_node. that recovery + * is complete. */ + mlog(0, "%s: dead master %d had reached " + "finalize1 state, clearing\n", dlm->name, idx); + dlm->reco.state &= ~DLM_RECO_STATE_FINALIZE; + __dlm_reset_recovery(dlm); + } + } + + /* Clean up join state on node death. */ + if (dlm->joining_node == idx) { + mlog(0, "Clearing join state for node %u\n", idx); + __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN); + } + /* check to see if the node is already considered dead */ if (!test_bit(idx, dlm->live_nodes_map)) { mlog(0, "for domain %s, node %d is already dead. " @@ -1821,12 +2417,6 @@ static void __dlm_hb_node_down(struct dlm_ctxt *dlm, int idx) clear_bit(idx, dlm->live_nodes_map); - /* Clean up join state on node death. */ - if (dlm->joining_node == idx) { - mlog(0, "Clearing join state for node %u\n", idx); - __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN); - } - /* make sure local cleanup occurs before the heartbeat events */ if (!test_bit(idx, dlm->recovery_map)) dlm_do_local_recovery_cleanup(dlm, idx); @@ -1836,6 +2426,7 @@ static void __dlm_hb_node_down(struct dlm_ctxt *dlm, int idx) mlog(0, "node %u being removed from domain map!\n", idx); clear_bit(idx, dlm->domain_map); + clear_bit(idx, dlm->exit_domain_map); /* wake up migration waiters if a node goes down. * perhaps later we can genericize this for other waiters. */ wake_up(&dlm->migration_wq); @@ -1854,6 +2445,13 @@ void dlm_hb_node_down_cb(struct o2nm_node *node, int idx, void *data) if (!dlm_grab(dlm)) return; + /* + * This will notify any dlm users that a node in our domain + * went away without notifying us first. + */ + if (test_bit(idx, dlm->domain_map)) + dlm_fire_domain_eviction_callbacks(dlm, idx); + spin_lock(&dlm->spinlock); __dlm_hb_node_down(dlm, idx); spin_unlock(&dlm->spinlock); @@ -1869,12 +2467,9 @@ void dlm_hb_node_up_cb(struct o2nm_node *node, int idx, void *data) return; spin_lock(&dlm->spinlock); - set_bit(idx, dlm->live_nodes_map); - - /* notify any mles attached to the heartbeat events */ - dlm_hb_event_notify_attached(dlm, idx, 1); - + /* do NOT notify mle attached to the heartbeat events. + * new nodes are not interesting in mastery until joined. */ spin_unlock(&dlm->spinlock); dlm_put(dlm); @@ -1897,7 +2492,18 @@ static void dlm_reco_unlock_ast(void *astdata, enum dlm_status st) mlog(0, "unlockast for recovery lock fired!\n"); } - +/* + * dlm_pick_recovery_master will continually attempt to use + * dlmlock() on the special "$RECOVERY" lockres with the + * LKM_NOQUEUE flag to get an EX. every thread that enters + * this function on each node racing to become the recovery + * master will not stop attempting this until either: + * a) this node gets the EX (and becomes the recovery master), + * or b) dlm->reco.new_master gets set to some nodenum + * != O2NM_INVALID_NODE_NUM (another node will do the reco). + * so each time a recovery master is needed, the entire cluster + * will sync at this point. if the new master dies, that will + * be detected in dlm_do_recovery */ static int dlm_pick_recovery_master(struct dlm_ctxt *dlm) { enum dlm_status ret; @@ -1906,23 +2512,70 @@ static int dlm_pick_recovery_master(struct dlm_ctxt *dlm) mlog(0, "starting recovery of %s at %lu, dead=%u, this=%u\n", dlm->name, jiffies, dlm->reco.dead_node, dlm->node_num); -retry: +again: memset(&lksb, 0, sizeof(lksb)); ret = dlmlock(dlm, LKM_EXMODE, &lksb, LKM_NOQUEUE|LKM_RECOVERY, - DLM_RECOVERY_LOCK_NAME, dlm_reco_ast, dlm, dlm_reco_bast); + DLM_RECOVERY_LOCK_NAME, DLM_RECOVERY_LOCK_NAME_LEN, + dlm_reco_ast, dlm, dlm_reco_bast); + + mlog(0, "%s: dlmlock($RECOVERY) returned %d, lksb=%d\n", + dlm->name, ret, lksb.status); if (ret == DLM_NORMAL) { mlog(0, "dlm=%s dlmlock says I got it (this=%u)\n", dlm->name, dlm->node_num); - /* I am master, send message to all nodes saying - * that I am beginning a recovery session */ - status = dlm_send_begin_reco_message(dlm, - dlm->reco.dead_node); + + /* got the EX lock. check to see if another node + * just became the reco master */ + if (dlm_reco_master_ready(dlm)) { + mlog(0, "%s: got reco EX lock, but %u will " + "do the recovery\n", dlm->name, + dlm->reco.new_master); + status = -EEXIST; + } else { + status = 0; + + /* see if recovery was already finished elsewhere */ + spin_lock(&dlm->spinlock); + if (dlm->reco.dead_node == O2NM_INVALID_NODE_NUM) { + status = -EINVAL; + mlog(0, "%s: got reco EX lock, but " + "node got recovered already\n", dlm->name); + if (dlm->reco.new_master != O2NM_INVALID_NODE_NUM) { + mlog(ML_ERROR, "%s: new master is %u " + "but no dead node!\n", + dlm->name, dlm->reco.new_master); + BUG(); + } + } + spin_unlock(&dlm->spinlock); + } + + /* if this node has actually become the recovery master, + * set the master and send the messages to begin recovery */ + if (!status) { + mlog(0, "%s: dead=%u, this=%u, sending " + "begin_reco now\n", dlm->name, + dlm->reco.dead_node, dlm->node_num); + status = dlm_send_begin_reco_message(dlm, + dlm->reco.dead_node); + /* this always succeeds */ + BUG_ON(status); + + /* set the new_master to this node */ + spin_lock(&dlm->spinlock); + dlm_set_reco_master(dlm, dlm->node_num); + spin_unlock(&dlm->spinlock); + } /* recovery lock is a special case. ast will not get fired, * so just go ahead and unlock it. */ ret = dlmunlock(dlm, &lksb, 0, dlm_reco_unlock_ast, dlm); + if (ret == DLM_DENIED) { + mlog(0, "got DLM_DENIED, trying LKM_CANCEL\n"); + ret = dlmunlock(dlm, &lksb, LKM_CANCEL, dlm_reco_unlock_ast, dlm); + } if (ret != DLM_NORMAL) { /* this would really suck. this could only happen * if there was a network error during the unlock @@ -1930,20 +2583,46 @@ retry: * is actually "done" and the lock structure is * even freed. we can continue, but only * because this specific lock name is special. */ - mlog(0, "dlmunlock returned %d\n", ret); - } - - if (status < 0) { - mlog(0, "failed to send recovery message. " - "must retry with new node map.\n"); - goto retry; + mlog(ML_ERROR, "dlmunlock returned %d\n", ret); } } else if (ret == DLM_NOTQUEUED) { mlog(0, "dlm=%s dlmlock says another node got it (this=%u)\n", dlm->name, dlm->node_num); /* another node is master. wait on - * reco.new_master != O2NM_INVALID_NODE_NUM */ + * reco.new_master != O2NM_INVALID_NODE_NUM + * for at most one second */ + wait_event_timeout(dlm->dlm_reco_thread_wq, + dlm_reco_master_ready(dlm), + msecs_to_jiffies(1000)); + if (!dlm_reco_master_ready(dlm)) { + mlog(0, "%s: reco master taking awhile\n", + dlm->name); + goto again; + } + /* another node has informed this one that it is reco master */ + mlog(0, "%s: reco master %u is ready to recover %u\n", + dlm->name, dlm->reco.new_master, dlm->reco.dead_node); status = -EEXIST; + } else if (ret == DLM_RECOVERING) { + mlog(0, "dlm=%s dlmlock says master node died (this=%u)\n", + dlm->name, dlm->node_num); + goto again; + } else { + struct dlm_lock_resource *res; + + /* dlmlock returned something other than NOTQUEUED or NORMAL */ + mlog(ML_ERROR, "%s: got %s from dlmlock($RECOVERY), " + "lksb.status=%s\n", dlm->name, dlm_errname(ret), + dlm_errname(lksb.status)); + res = dlm_lookup_lockres(dlm, DLM_RECOVERY_LOCK_NAME, + DLM_RECOVERY_LOCK_NAME_LEN); + if (res) { + dlm_print_one_lock_resource(res); + dlm_lockres_put(res); + } else { + mlog(ML_ERROR, "recovery lock not found\n"); + } + BUG(); } return status; @@ -1957,9 +2636,7 @@ static int dlm_send_begin_reco_message(struct dlm_ctxt *dlm, u8 dead_node) int nodenum; int status; - mlog_entry("%u\n", dead_node); - - mlog(0, "dead node is %u\n", dead_node); + mlog(0, "%s: dead node is %u\n", dlm->name, dead_node); spin_lock(&dlm->spinlock); dlm_node_iter_init(dlm->domain_map, &iter); @@ -1982,7 +2659,7 @@ static int dlm_send_begin_reco_message(struct dlm_ctxt *dlm, u8 dead_node) mlog(0, "not sending begin reco to self\n"); continue; } - +retry: ret = -EINVAL; mlog(0, "attempting to send begin reco msg to %d\n", nodenum); @@ -1991,11 +2668,35 @@ static int dlm_send_begin_reco_message(struct dlm_ctxt *dlm, u8 dead_node) /* negative status is handled ok by caller here */ if (ret >= 0) ret = status; + if (dlm_is_host_down(ret)) { + /* node is down. not involved in recovery + * so just keep going */ + mlog(ML_NOTICE, "%s: node %u was down when sending " + "begin reco msg (%d)\n", dlm->name, nodenum, ret); + ret = 0; + } + + /* + * Prior to commit aad1b15310b9bcd59fa81ab8f2b1513b59553ea8, + * dlm_begin_reco_handler() returned EAGAIN and not -EAGAIN. + * We are handling both for compatibility reasons. + */ + if (ret == -EAGAIN || ret == EAGAIN) { + mlog(0, "%s: trying to start recovery of node " + "%u, but node %u is waiting for last recovery " + "to complete, backoff for a bit\n", dlm->name, + dead_node, nodenum); + msleep(100); + goto retry; + } if (ret < 0) { struct dlm_lock_resource *res; + + /* this is now a serious problem, possibly ENOMEM + * in the network stack. must retry */ mlog_errno(ret); mlog(ML_ERROR, "begin reco of dlm %s to node %u " - " returned %d\n", dlm->name, nodenum, ret); + "returned %d\n", dlm->name, nodenum, ret); res = dlm_lookup_lockres(dlm, DLM_RECOVERY_LOCK_NAME, DLM_RECOVERY_LOCK_NAME_LEN); if (res) { @@ -2004,14 +2705,18 @@ static int dlm_send_begin_reco_message(struct dlm_ctxt *dlm, u8 dead_node) } else { mlog(ML_ERROR, "recovery lock not found\n"); } - break; + /* sleep for a bit in hopes that we can avoid + * another ENOMEM */ + msleep(100); + goto retry; } } return ret; } -int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data) +int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { struct dlm_ctxt *dlm = data; struct dlm_begin_reco *br = (struct dlm_begin_reco *)msg->buf; @@ -2020,35 +2725,72 @@ int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data) if (!dlm_grab(dlm)) return 0; - mlog(0, "node %u wants to recover node %u\n", - br->node_idx, br->dead_node); + spin_lock(&dlm->spinlock); + if (dlm->reco.state & DLM_RECO_STATE_FINALIZE) { + mlog(0, "%s: node %u wants to recover node %u (%u:%u) " + "but this node is in finalize state, waiting on finalize2\n", + dlm->name, br->node_idx, br->dead_node, + dlm->reco.dead_node, dlm->reco.new_master); + spin_unlock(&dlm->spinlock); + dlm_put(dlm); + return -EAGAIN; + } + spin_unlock(&dlm->spinlock); + + mlog(0, "%s: node %u wants to recover node %u (%u:%u)\n", + dlm->name, br->node_idx, br->dead_node, + dlm->reco.dead_node, dlm->reco.new_master); dlm_fire_domain_eviction_callbacks(dlm, br->dead_node); spin_lock(&dlm->spinlock); if (dlm->reco.new_master != O2NM_INVALID_NODE_NUM) { - mlog(0, "new_master already set to %u!\n", - dlm->reco.new_master); + if (test_bit(dlm->reco.new_master, dlm->recovery_map)) { + mlog(0, "%s: new_master %u died, changing " + "to %u\n", dlm->name, dlm->reco.new_master, + br->node_idx); + } else { + mlog(0, "%s: new_master %u NOT DEAD, changing " + "to %u\n", dlm->name, dlm->reco.new_master, + br->node_idx); + /* may not have seen the new master as dead yet */ + } } if (dlm->reco.dead_node != O2NM_INVALID_NODE_NUM) { - mlog(0, "dead_node already set to %u!\n", - dlm->reco.dead_node); + mlog(ML_NOTICE, "%s: dead_node previously set to %u, " + "node %u changing it to %u\n", dlm->name, + dlm->reco.dead_node, br->node_idx, br->dead_node); } - dlm->reco.new_master = br->node_idx; - dlm->reco.dead_node = br->dead_node; + dlm_set_reco_master(dlm, br->node_idx); + dlm_set_reco_dead_node(dlm, br->dead_node); if (!test_bit(br->dead_node, dlm->recovery_map)) { - mlog(ML_ERROR, "recovery master %u sees %u as dead, but this " + mlog(0, "recovery master %u sees %u as dead, but this " "node has not yet. marking %u as dead\n", br->node_idx, br->dead_node, br->dead_node); + if (!test_bit(br->dead_node, dlm->domain_map) || + !test_bit(br->dead_node, dlm->live_nodes_map)) + mlog(0, "%u not in domain/live_nodes map " + "so setting it in reco map manually\n", + br->dead_node); + /* force the recovery cleanup in __dlm_hb_node_down + * both of these will be cleared in a moment */ + set_bit(br->dead_node, dlm->domain_map); + set_bit(br->dead_node, dlm->live_nodes_map); __dlm_hb_node_down(dlm, br->dead_node); } spin_unlock(&dlm->spinlock); dlm_kick_recovery_thread(dlm); + + mlog(0, "%s: recovery started by node %u, for %u (%u:%u)\n", + dlm->name, br->node_idx, br->dead_node, + dlm->reco.dead_node, dlm->reco.new_master); + dlm_put(dlm); return 0; } +#define DLM_FINALIZE_STAGE2 0x01 static int dlm_send_finalize_reco_message(struct dlm_ctxt *dlm) { int ret = 0; @@ -2056,54 +2798,72 @@ static int dlm_send_finalize_reco_message(struct dlm_ctxt *dlm) struct dlm_node_iter iter; int nodenum; int status; + int stage = 1; - mlog(0, "finishing recovery for node %s:%u\n", - dlm->name, dlm->reco.dead_node); + mlog(0, "finishing recovery for node %s:%u, " + "stage %d\n", dlm->name, dlm->reco.dead_node, stage); spin_lock(&dlm->spinlock); dlm_node_iter_init(dlm->domain_map, &iter); spin_unlock(&dlm->spinlock); +stage2: memset(&fr, 0, sizeof(fr)); fr.node_idx = dlm->node_num; fr.dead_node = dlm->reco.dead_node; + if (stage == 2) + fr.flags |= DLM_FINALIZE_STAGE2; while ((nodenum = dlm_node_iter_next(&iter)) >= 0) { if (nodenum == dlm->node_num) continue; ret = o2net_send_message(DLM_FINALIZE_RECO_MSG, dlm->key, &fr, sizeof(fr), nodenum, &status); - if (ret >= 0) { + if (ret >= 0) ret = status; + if (ret < 0) { + mlog(ML_ERROR, "Error %d when sending message %u (key " + "0x%x) to node %u\n", ret, DLM_FINALIZE_RECO_MSG, + dlm->key, nodenum); if (dlm_is_host_down(ret)) { - /* this has no effect on this recovery - * session, so set the status to zero to + /* this has no effect on this recovery + * session, so set the status to zero to * finish out the last recovery */ mlog(ML_ERROR, "node %u went down after this " "node finished recovery.\n", nodenum); ret = 0; + continue; } - } - if (ret < 0) { - mlog_errno(ret); break; } } + if (stage == 1) { + /* reset the node_iter back to the top and send finalize2 */ + iter.curnode = -1; + stage = 2; + goto stage2; + } return ret; } -int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data) +int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { struct dlm_ctxt *dlm = data; struct dlm_finalize_reco *fr = (struct dlm_finalize_reco *)msg->buf; + int stage = 1; /* ok to return 0, domain has gone away */ if (!dlm_grab(dlm)) return 0; - mlog(0, "node %u finalizing recovery of node %u\n", - fr->node_idx, fr->dead_node); + if (fr->flags & DLM_FINALIZE_STAGE2) + stage = 2; + + mlog(0, "%s: node %u finalizing recovery stage%d of " + "node %u (%u:%u)\n", dlm->name, fr->node_idx, stage, + fr->dead_node, dlm->reco.dead_node, dlm->reco.new_master); spin_lock(&dlm->spinlock); @@ -2120,13 +2880,41 @@ int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data) BUG(); } - dlm_finish_local_lockres_recovery(dlm, fr->dead_node, fr->node_idx); - - spin_unlock(&dlm->spinlock); + switch (stage) { + case 1: + dlm_finish_local_lockres_recovery(dlm, fr->dead_node, fr->node_idx); + if (dlm->reco.state & DLM_RECO_STATE_FINALIZE) { + mlog(ML_ERROR, "%s: received finalize1 from " + "new master %u for dead node %u, but " + "this node has already received it!\n", + dlm->name, fr->node_idx, fr->dead_node); + dlm_print_reco_node_status(dlm); + BUG(); + } + dlm->reco.state |= DLM_RECO_STATE_FINALIZE; + spin_unlock(&dlm->spinlock); + break; + case 2: + if (!(dlm->reco.state & DLM_RECO_STATE_FINALIZE)) { + mlog(ML_ERROR, "%s: received finalize2 from " + "new master %u for dead node %u, but " + "this node did not have finalize1!\n", + dlm->name, fr->node_idx, fr->dead_node); + dlm_print_reco_node_status(dlm); + BUG(); + } + dlm->reco.state &= ~DLM_RECO_STATE_FINALIZE; + __dlm_reset_recovery(dlm); + spin_unlock(&dlm->spinlock); + dlm_kick_recovery_thread(dlm); + break; + default: + BUG(); + } - dlm_reset_recovery(dlm); + mlog(0, "%s: recovery done, reco master was %u, dead now %u, master now %u\n", + dlm->name, fr->node_idx, dlm->reco.dead_node, dlm->reco.new_master); - dlm_kick_recovery_thread(dlm); dlm_put(dlm); return 0; } diff --git a/fs/ocfs2/dlm/dlmthread.c b/fs/ocfs2/dlm/dlmthread.c index 5be9d14f12c..69aac6f088a 100644 --- a/fs/ocfs2/dlm/dlmthread.c +++ b/fs/ocfs2/dlm/dlmthread.c @@ -28,9 +28,7 @@ #include <linux/module.h> #include <linux/fs.h> #include <linux/types.h> -#include <linux/slab.h> #include <linux/highmem.h> -#include <linux/utsname.h> #include <linux/init.h> #include <linux/sysctl.h> #include <linux/random.h> @@ -39,6 +37,7 @@ #include <linux/inet.h> #include <linux/timer.h> #include <linux/kthread.h> +#include <linux/delay.h> #include "cluster/heartbeat.h" @@ -53,7 +52,6 @@ #include "cluster/masklog.h" static int dlm_thread(void *data); - static void dlm_flush_asts(struct dlm_ctxt *dlm); #define dlm_lock_is_remote(dlm, lock) ((lock)->ml.node != (dlm)->node_num) @@ -76,18 +74,47 @@ repeat: goto repeat; } remove_wait_queue(&res->wq, &wait); - current->state = TASK_RUNNING; + __set_current_state(TASK_RUNNING); } - -static int __dlm_lockres_unused(struct dlm_lock_resource *res) +int __dlm_lockres_has_locks(struct dlm_lock_resource *res) { if (list_empty(&res->granted) && list_empty(&res->converting) && - list_empty(&res->blocked) && - list_empty(&res->dirty)) - return 1; - return 0; + list_empty(&res->blocked)) + return 0; + return 1; +} + +/* "unused": the lockres has no locks, is not on the dirty list, + * has no inflight locks (in the gap between mastery and acquiring + * the first lock), and has no bits in its refmap. + * truly ready to be freed. */ +int __dlm_lockres_unused(struct dlm_lock_resource *res) +{ + int bit; + + assert_spin_locked(&res->spinlock); + + if (__dlm_lockres_has_locks(res)) + return 0; + + /* Locks are in the process of being created */ + if (res->inflight_locks) + return 0; + + if (!list_empty(&res->dirty) || res->state & DLM_LOCK_RES_DIRTY) + return 0; + + if (res->state & DLM_LOCK_RES_RECOVERING) + return 0; + + /* Another node has this resource with this node as the master */ + bit = find_next_bit(res->refmap, O2NM_MAX_NODES, 0); + if (bit < O2NM_MAX_NODES) + return 0; + + return 1; } @@ -97,25 +124,25 @@ static int __dlm_lockres_unused(struct dlm_lock_resource *res) void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) { - mlog_entry("%.*s\n", res->lockname.len, res->lockname.name); - assert_spin_locked(&dlm->spinlock); assert_spin_locked(&res->spinlock); if (__dlm_lockres_unused(res)){ if (list_empty(&res->purge)) { - mlog(0, "putting lockres %.*s from purge list\n", - res->lockname.len, res->lockname.name); + mlog(0, "%s: Adding res %.*s to purge list\n", + dlm->name, res->lockname.len, res->lockname.name); res->last_used = jiffies; + dlm_lockres_get(res); list_add_tail(&res->purge, &dlm->purge_list); dlm->purge_count++; } } else if (!list_empty(&res->purge)) { - mlog(0, "removing lockres %.*s from purge list\n", - res->lockname.len, res->lockname.name); + mlog(0, "%s: Removing res %.*s from purge list\n", + dlm->name, res->lockname.len, res->lockname.name); list_del_init(&res->purge); + dlm_lockres_put(res); dlm->purge_count--; } } @@ -123,7 +150,6 @@ void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm, void dlm_lockres_calc_usage(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) { - mlog_entry("%.*s\n", res->lockname.len, res->lockname.name); spin_lock(&dlm->spinlock); spin_lock(&res->spinlock); @@ -133,49 +159,66 @@ void dlm_lockres_calc_usage(struct dlm_ctxt *dlm, spin_unlock(&dlm->spinlock); } -/* TODO: Eventual API: Called with the dlm spinlock held, may drop it - * to do migration, but will re-acquire before exit. */ -void dlm_purge_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *lockres) +static void dlm_purge_lockres(struct dlm_ctxt *dlm, + struct dlm_lock_resource *res) { int master; - int ret; + int ret = 0; - spin_lock(&lockres->spinlock); - master = lockres->owner == dlm->node_num; - spin_unlock(&lockres->spinlock); + assert_spin_locked(&dlm->spinlock); + assert_spin_locked(&res->spinlock); - mlog(0, "purging lockres %.*s, master = %d\n", lockres->lockname.len, - lockres->lockname.name, master); + master = (res->owner == dlm->node_num); - /* Non master is the easy case -- no migration required, just - * quit. */ - if (!master) - goto finish; + mlog(0, "%s: Purging res %.*s, master %d\n", dlm->name, + res->lockname.len, res->lockname.name, master); - /* Wheee! Migrate lockres here! */ - spin_unlock(&dlm->spinlock); -again: + if (!master) { + res->state |= DLM_LOCK_RES_DROPPING_REF; + /* drop spinlock... retake below */ + spin_unlock(&res->spinlock); + spin_unlock(&dlm->spinlock); - ret = dlm_migrate_lockres(dlm, lockres, O2NM_MAX_NODES); - if (ret == -ENOTEMPTY) { - mlog(ML_ERROR, "lockres %.*s still has local locks!\n", - lockres->lockname.len, lockres->lockname.name); + spin_lock(&res->spinlock); + /* This ensures that clear refmap is sent after the set */ + __dlm_wait_on_lockres_flags(res, DLM_LOCK_RES_SETREF_INPROG); + spin_unlock(&res->spinlock); - BUG(); - } else if (ret < 0) { - mlog(ML_NOTICE, "lockres %.*s: migrate failed, retrying\n", - lockres->lockname.len, lockres->lockname.name); - goto again; + /* clear our bit from the master's refmap, ignore errors */ + ret = dlm_drop_lockres_ref(dlm, res); + if (ret < 0) { + if (!dlm_is_host_down(ret)) + BUG(); + } + spin_lock(&dlm->spinlock); + spin_lock(&res->spinlock); } - spin_lock(&dlm->spinlock); - -finish: - if (!list_empty(&lockres->purge)) { - list_del_init(&lockres->purge); + if (!list_empty(&res->purge)) { + mlog(0, "%s: Removing res %.*s from purgelist, master %d\n", + dlm->name, res->lockname.len, res->lockname.name, master); + list_del_init(&res->purge); + dlm_lockres_put(res); dlm->purge_count--; } - __dlm_unhash_lockres(lockres); + + if (!__dlm_lockres_unused(res)) { + mlog(ML_ERROR, "%s: res %.*s in use after deref\n", + dlm->name, res->lockname.len, res->lockname.name); + __dlm_print_one_lock_resource(res); + BUG(); + } + + __dlm_unhash_lockres(dlm, res); + + /* lockres is not in the hash now. drop the flag and wake up + * any processes waiting in dlm_get_lock_resource. */ + if (!master) { + res->state &= ~DLM_LOCK_RES_DROPPING_REF; + spin_unlock(&res->spinlock); + wake_up(&res->wq); + } else + spin_unlock(&res->spinlock); } static void dlm_run_purge_list(struct dlm_ctxt *dlm, @@ -194,17 +237,7 @@ static void dlm_run_purge_list(struct dlm_ctxt *dlm, lockres = list_entry(dlm->purge_list.next, struct dlm_lock_resource, purge); - /* Status of the lockres *might* change so double - * check. If the lockres is unused, holding the dlm - * spinlock will prevent people from getting and more - * refs on it -- there's no need to keep the lockres - * spinlock. */ spin_lock(&lockres->spinlock); - unused = __dlm_lockres_unused(lockres); - spin_unlock(&lockres->spinlock); - - if (!unused) - continue; purge_jiffies = lockres->last_used + msecs_to_jiffies(DLM_PURGE_INTERVAL_MS); @@ -216,17 +249,34 @@ static void dlm_run_purge_list(struct dlm_ctxt *dlm, * in tail order, we can stop at the first * unpurgable resource -- anyone added after * him will have a greater last_used value */ + spin_unlock(&lockres->spinlock); break; } - list_del_init(&lockres->purge); - dlm->purge_count--; + /* Status of the lockres *might* change so double + * check. If the lockres is unused, holding the dlm + * spinlock will prevent people from getting and more + * refs on it. */ + unused = __dlm_lockres_unused(lockres); + if (!unused || + (lockres->state & DLM_LOCK_RES_MIGRATING) || + (lockres->inflight_assert_workers != 0)) { + mlog(0, "%s: res %.*s is in use or being remastered, " + "used %d, state %d, assert master workers %u\n", + dlm->name, lockres->lockname.len, + lockres->lockname.name, + !unused, lockres->state, + lockres->inflight_assert_workers); + list_move_tail(&lockres->purge, &dlm->purge_list); + spin_unlock(&lockres->spinlock); + continue; + } + + dlm_lockres_get(lockres); - /* This may drop and reacquire the dlm spinlock if it - * has to do migration. */ - mlog(0, "calling dlm_purge_lockres!\n"); dlm_purge_lockres(dlm, lockres); - mlog(0, "DONE calling dlm_purge_lockres!\n"); + + dlm_lockres_put(lockres); /* Avoid adding any scheduling latencies */ cond_resched_lock(&dlm->spinlock); @@ -239,19 +289,15 @@ static void dlm_shuffle_lists(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) { struct dlm_lock *lock, *target; - struct list_head *iter; - struct list_head *head; int can_grant = 1; - //mlog(0, "res->lockname.len=%d\n", res->lockname.len); - //mlog(0, "res->lockname.name=%p\n", res->lockname.name); - //mlog(0, "shuffle res %.*s\n", res->lockname.len, - // res->lockname.name); - - /* because this function is called with the lockres + /* + * Because this function is called with the lockres * spinlock, and because we know that it is not migrating/ * recovering/in-progress, it is fine to reserve asts and - * basts right before queueing them all throughout */ + * basts right before queueing them all throughout + */ + assert_spin_locked(&dlm->ast_lock); assert_spin_locked(&res->spinlock); BUG_ON((res->state & (DLM_LOCK_RES_MIGRATING| DLM_LOCK_RES_RECOVERING| @@ -260,18 +306,16 @@ static void dlm_shuffle_lists(struct dlm_ctxt *dlm, converting: if (list_empty(&res->converting)) goto blocked; - mlog(0, "res %.*s has locks on a convert queue\n", res->lockname.len, - res->lockname.name); + mlog(0, "%s: res %.*s has locks on the convert queue\n", dlm->name, + res->lockname.len, res->lockname.name); target = list_entry(res->converting.next, struct dlm_lock, list); if (target->ml.convert_type == LKM_IVMODE) { - mlog(ML_ERROR, "%.*s: converting a lock with no " - "convert_type!\n", res->lockname.len, res->lockname.name); + mlog(ML_ERROR, "%s: res %.*s converting lock to invalid mode\n", + dlm->name, res->lockname.len, res->lockname.name); BUG(); } - head = &res->granted; - list_for_each(iter, head) { - lock = list_entry(iter, struct dlm_lock, list); + list_for_each_entry(lock, &res->granted, list) { if (lock==target) continue; if (!dlm_lock_compatible(lock->ml.type, @@ -280,7 +324,7 @@ converting: /* queue the BAST if not already */ if (lock->ml.highest_blocked == LKM_IVMODE) { __dlm_lockres_reserve_ast(res); - dlm_queue_bast(dlm, lock); + __dlm_queue_bast(dlm, lock); } /* update the highest_blocked if needed */ if (lock->ml.highest_blocked < target->ml.convert_type) @@ -288,9 +332,8 @@ converting: target->ml.convert_type; } } - head = &res->converting; - list_for_each(iter, head) { - lock = list_entry(iter, struct dlm_lock, list); + + list_for_each_entry(lock, &res->converting, list) { if (lock==target) continue; if (!dlm_lock_compatible(lock->ml.type, @@ -298,7 +341,7 @@ converting: can_grant = 0; if (lock->ml.highest_blocked == LKM_IVMODE) { __dlm_lockres_reserve_ast(res); - dlm_queue_bast(dlm, lock); + __dlm_queue_bast(dlm, lock); } if (lock->ml.highest_blocked < target->ml.convert_type) lock->ml.highest_blocked = @@ -311,15 +354,17 @@ converting: spin_lock(&target->spinlock); BUG_ON(target->ml.highest_blocked != LKM_IVMODE); - mlog(0, "calling ast for converting lock: %.*s, have: %d, " - "granting: %d, node: %u\n", res->lockname.len, - res->lockname.name, target->ml.type, + mlog(0, "%s: res %.*s, AST for Converting lock %u:%llu, type " + "%d => %d, node %u\n", dlm->name, res->lockname.len, + res->lockname.name, + dlm_get_lock_cookie_node(be64_to_cpu(target->ml.cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(target->ml.cookie)), + target->ml.type, target->ml.convert_type, target->ml.node); target->ml.type = target->ml.convert_type; target->ml.convert_type = LKM_IVMODE; - list_del_init(&target->list); - list_add_tail(&target->list, &res->granted); + list_move_tail(&target->list, &res->granted); BUG_ON(!target->lksb); target->lksb->status = DLM_NORMAL; @@ -327,7 +372,7 @@ converting: spin_unlock(&target->spinlock); __dlm_lockres_reserve_ast(res); - dlm_queue_ast(dlm, target); + __dlm_queue_ast(dlm, target); /* go back and check for more */ goto converting; } @@ -337,32 +382,28 @@ blocked: goto leave; target = list_entry(res->blocked.next, struct dlm_lock, list); - head = &res->granted; - list_for_each(iter, head) { - lock = list_entry(iter, struct dlm_lock, list); + list_for_each_entry(lock, &res->granted, list) { if (lock==target) continue; if (!dlm_lock_compatible(lock->ml.type, target->ml.type)) { can_grant = 0; if (lock->ml.highest_blocked == LKM_IVMODE) { __dlm_lockres_reserve_ast(res); - dlm_queue_bast(dlm, lock); + __dlm_queue_bast(dlm, lock); } if (lock->ml.highest_blocked < target->ml.type) lock->ml.highest_blocked = target->ml.type; } } - head = &res->converting; - list_for_each(iter, head) { - lock = list_entry(iter, struct dlm_lock, list); + list_for_each_entry(lock, &res->converting, list) { if (lock==target) continue; if (!dlm_lock_compatible(lock->ml.type, target->ml.type)) { can_grant = 0; if (lock->ml.highest_blocked == LKM_IVMODE) { __dlm_lockres_reserve_ast(res); - dlm_queue_bast(dlm, lock); + __dlm_queue_bast(dlm, lock); } if (lock->ml.highest_blocked < target->ml.type) lock->ml.highest_blocked = target->ml.type; @@ -375,13 +416,15 @@ blocked: spin_lock(&target->spinlock); BUG_ON(target->ml.highest_blocked != LKM_IVMODE); - mlog(0, "calling ast for blocked lock: %.*s, granting: %d, " - "node: %u\n", res->lockname.len, res->lockname.name, + mlog(0, "%s: res %.*s, AST for Blocked lock %u:%llu, type %d, " + "node %u\n", dlm->name, res->lockname.len, + res->lockname.name, + dlm_get_lock_cookie_node(be64_to_cpu(target->ml.cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(target->ml.cookie)), target->ml.type, target->ml.node); - // target->ml.type is already correct - list_del_init(&target->list); - list_add_tail(&target->list, &res->granted); + /* target->ml.type is already correct */ + list_move_tail(&target->list, &res->granted); BUG_ON(!target->lksb); target->lksb->status = DLM_NORMAL; @@ -389,7 +432,7 @@ blocked: spin_unlock(&target->spinlock); __dlm_lockres_reserve_ast(res); - dlm_queue_ast(dlm, target); + __dlm_queue_ast(dlm, target); /* go back and check for more */ goto converting; } @@ -401,7 +444,6 @@ leave: /* must have NO locks when calling this with res !=NULL * */ void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) { - mlog_entry("dlm=%p, res=%p\n", dlm, res); if (res) { spin_lock(&dlm->spinlock); spin_lock(&res->spinlock); @@ -414,24 +456,32 @@ void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) { - mlog_entry("dlm=%p, res=%p\n", dlm, res); - assert_spin_locked(&dlm->spinlock); assert_spin_locked(&res->spinlock); /* don't shuffle secondary queues */ - if ((res->owner == dlm->node_num) && - !(res->state & DLM_LOCK_RES_DIRTY)) { - list_add_tail(&res->dirty, &dlm->dirty_list); - res->state |= DLM_LOCK_RES_DIRTY; + if ((res->owner == dlm->node_num)) { + if (res->state & (DLM_LOCK_RES_MIGRATING | + DLM_LOCK_RES_BLOCK_DIRTY)) + return; + + if (list_empty(&res->dirty)) { + /* ref for dirty_list */ + dlm_lockres_get(res); + list_add_tail(&res->dirty, &dlm->dirty_list); + res->state |= DLM_LOCK_RES_DIRTY; + } } + + mlog(0, "%s: res %.*s\n", dlm->name, res->lockname.len, + res->lockname.name); } /* Launch the NM thread for the mounted volume */ int dlm_launch_thread(struct dlm_ctxt *dlm) { - mlog(0, "starting dlm thread...\n"); + mlog(0, "Starting dlm_thread...\n"); dlm->dlm_thread_task = kthread_run(dlm_thread, dlm, "dlm_thread"); if (IS_ERR(dlm->dlm_thread_task)) { @@ -446,7 +496,7 @@ int dlm_launch_thread(struct dlm_ctxt *dlm) void dlm_complete_thread(struct dlm_ctxt *dlm) { if (dlm->dlm_thread_task) { - mlog(ML_KTHREAD, "waiting for dlm thread to exit\n"); + mlog(ML_KTHREAD, "Waiting for dlm thread to exit\n"); kthread_stop(dlm->dlm_thread_task); dlm->dlm_thread_task = NULL; } @@ -477,7 +527,12 @@ static void dlm_flush_asts(struct dlm_ctxt *dlm) /* get an extra ref on lock */ dlm_lock_get(lock); res = lock->lockres; - mlog(0, "delivering an ast for this lockres\n"); + mlog(0, "%s: res %.*s, Flush AST for lock %u:%llu, type %d, " + "node %u\n", dlm->name, res->lockname.len, + res->lockname.name, + dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)), + lock->ml.type, lock->ml.node); BUG_ON(!lock->ast_pending); @@ -498,9 +553,9 @@ static void dlm_flush_asts(struct dlm_ctxt *dlm) /* possible that another ast was queued while * we were delivering the last one */ if (!list_empty(&lock->ast_list)) { - mlog(0, "aha another ast got queued while " - "we were finishing the last one. will " - "keep the ast_pending flag set.\n"); + mlog(0, "%s: res %.*s, AST queued while flushing last " + "one\n", dlm->name, res->lockname.len, + res->lockname.name); } else lock->ast_pending = 0; @@ -531,8 +586,12 @@ static void dlm_flush_asts(struct dlm_ctxt *dlm) dlm_lock_put(lock); spin_unlock(&dlm->ast_lock); - mlog(0, "delivering a bast for this lockres " - "(blocked = %d\n", hi); + mlog(0, "%s: res %.*s, Flush BAST for lock %u:%llu, " + "blocked %d, node %u\n", + dlm->name, res->lockname.len, res->lockname.name, + dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)), + hi, lock->ml.node); if (lock->ml.node != dlm->node_num) { ret = dlm_send_proxy_bast(dlm, res, lock, hi); @@ -546,9 +605,9 @@ static void dlm_flush_asts(struct dlm_ctxt *dlm) /* possible that another bast was queued while * we were delivering the last one */ if (!list_empty(&lock->bast_list)) { - mlog(0, "aha another bast got queued while " - "we were finishing the last one. will " - "keep the bast_pending flag set.\n"); + mlog(0, "%s: res %.*s, BAST queued while flushing last " + "one\n", dlm->name, res->lockname.len, + res->lockname.name); } else lock->bast_pending = 0; @@ -602,35 +661,41 @@ static int dlm_thread(void *data) dlm_lockres_get(res); spin_lock(&res->spinlock); - res->state &= ~DLM_LOCK_RES_DIRTY; + /* We clear the DLM_LOCK_RES_DIRTY state once we shuffle lists below */ list_del_init(&res->dirty); spin_unlock(&res->spinlock); spin_unlock(&dlm->spinlock); + /* Drop dirty_list ref */ + dlm_lockres_put(res); /* lockres can be re-dirtied/re-added to the * dirty_list in this gap, but that is ok */ + spin_lock(&dlm->ast_lock); spin_lock(&res->spinlock); if (res->owner != dlm->node_num) { __dlm_print_one_lock_resource(res); - mlog(ML_ERROR, "inprog:%s, mig:%s, reco:%s, dirty:%s\n", - res->state & DLM_LOCK_RES_IN_PROGRESS ? "yes" : "no", - res->state & DLM_LOCK_RES_MIGRATING ? "yes" : "no", - res->state & DLM_LOCK_RES_RECOVERING ? "yes" : "no", - res->state & DLM_LOCK_RES_DIRTY ? "yes" : "no"); + mlog(ML_ERROR, "%s: inprog %d, mig %d, reco %d," + " dirty %d\n", dlm->name, + !!(res->state & DLM_LOCK_RES_IN_PROGRESS), + !!(res->state & DLM_LOCK_RES_MIGRATING), + !!(res->state & DLM_LOCK_RES_RECOVERING), + !!(res->state & DLM_LOCK_RES_DIRTY)); } BUG_ON(res->owner != dlm->node_num); /* it is now ok to move lockreses in these states * to the dirty list, assuming that they will only be * dirty for a short while. */ + BUG_ON(res->state & DLM_LOCK_RES_MIGRATING); if (res->state & (DLM_LOCK_RES_IN_PROGRESS | - DLM_LOCK_RES_MIGRATING | DLM_LOCK_RES_RECOVERING)) { /* move it to the tail and keep going */ + res->state &= ~DLM_LOCK_RES_DIRTY; spin_unlock(&res->spinlock); - mlog(0, "delaying list shuffling for in-" - "progress lockres %.*s, state=%d\n", + spin_unlock(&dlm->ast_lock); + mlog(0, "%s: res %.*s, inprogress, delay list " + "shuffle, state %d\n", dlm->name, res->lockname.len, res->lockname.name, res->state); delay = 1; @@ -642,12 +707,11 @@ static int dlm_thread(void *data) * spinlock and do NOT have the dlm lock. * safe to reserve/queue asts and run the lists. */ - mlog(0, "calling dlm_shuffle_lists with dlm=%p, " - "res=%p\n", dlm, res); - /* called while holding lockres lock */ dlm_shuffle_lists(dlm, res); + res->state &= ~DLM_LOCK_RES_DIRTY; spin_unlock(&res->spinlock); + spin_unlock(&dlm->ast_lock); dlm_lockres_calc_usage(dlm, res); @@ -658,8 +722,7 @@ in_progress: * it on the back of the list */ if (delay) { spin_lock(&res->spinlock); - list_add_tail(&res->dirty, &dlm->dirty_list); - res->state |= DLM_LOCK_RES_DIRTY; + __dlm_dirty_lockres(dlm, res); spin_unlock(&res->spinlock); } dlm_lockres_put(res); @@ -667,7 +730,8 @@ in_progress: /* unlikely, but we may need to give time to * other tasks */ if (!--n) { - mlog(0, "throttling dlm_thread\n"); + mlog(0, "%s: Throttling dlm thread\n", + dlm->name); break; } } @@ -677,7 +741,7 @@ in_progress: /* yield and continue right away if there is more work to do */ if (!n) { - yield(); + cond_resched(); continue; } diff --git a/fs/ocfs2/dlm/dlmunlock.c b/fs/ocfs2/dlm/dlmunlock.c index cec2ce1cd31..2e3c9dbab68 100644 --- a/fs/ocfs2/dlm/dlmunlock.c +++ b/fs/ocfs2/dlm/dlmunlock.c @@ -28,9 +28,7 @@ #include <linux/module.h> #include <linux/fs.h> #include <linux/types.h> -#include <linux/slab.h> #include <linux/highmem.h> -#include <linux/utsname.h> #include <linux/init.h> #include <linux/sysctl.h> #include <linux/random.h> @@ -117,12 +115,12 @@ static enum dlm_status dlmunlock_common(struct dlm_ctxt *dlm, else BUG_ON(res->owner == dlm->node_num); - spin_lock(&dlm->spinlock); + spin_lock(&dlm->ast_lock); /* We want to be sure that we're not freeing a lock * that still has AST's pending... */ in_use = !list_empty(&lock->ast_list); - spin_unlock(&dlm->spinlock); - if (in_use) { + spin_unlock(&dlm->ast_lock); + if (in_use && !(flags & LKM_CANCEL)) { mlog(ML_ERROR, "lockres %.*s: Someone is calling dlmunlock " "while waiting for an ast!", res->lockname.len, res->lockname.name); @@ -131,7 +129,7 @@ static enum dlm_status dlmunlock_common(struct dlm_ctxt *dlm, spin_lock(&res->spinlock); if (res->state & DLM_LOCK_RES_IN_PROGRESS) { - if (master_node) { + if (master_node && !(flags & LKM_CANCEL)) { mlog(ML_ERROR, "lockres in progress!\n"); spin_unlock(&res->spinlock); return DLM_FORWARD; @@ -147,6 +145,10 @@ static enum dlm_status dlmunlock_common(struct dlm_ctxt *dlm, goto leave; } + if (res->state & DLM_LOCK_RES_MIGRATING) { + status = DLM_MIGRATING; + goto leave; + } /* see above for what the spec says about * LKM_CANCEL and the lock queue state */ @@ -155,7 +157,7 @@ static enum dlm_status dlmunlock_common(struct dlm_ctxt *dlm, else status = dlm_get_unlock_actions(dlm, res, lock, lksb, &actions); - if (status != DLM_NORMAL) + if (status != DLM_NORMAL && (status != DLM_CANCELGRANT || !master_node)) goto leave; /* By now this has been masked out of cancel requests. */ @@ -183,11 +185,26 @@ static enum dlm_status dlmunlock_common(struct dlm_ctxt *dlm, spin_lock(&lock->spinlock); /* if the master told us the lock was already granted, * let the ast handle all of these actions */ - if (status == DLM_NORMAL && - lksb->status == DLM_CANCELGRANT) { + if (status == DLM_CANCELGRANT) { actions &= ~(DLM_UNLOCK_REMOVE_LOCK| DLM_UNLOCK_REGRANT_LOCK| DLM_UNLOCK_CLEAR_CONVERT_TYPE); + } else if (status == DLM_RECOVERING || + status == DLM_MIGRATING || + status == DLM_FORWARD || + status == DLM_NOLOCKMGR + ) { + /* must clear the actions because this unlock + * is about to be retried. cannot free or do + * any list manipulation. */ + mlog(0, "%s:%.*s: clearing actions, %s\n", + dlm->name, res->lockname.len, + res->lockname.name, + status==DLM_RECOVERING?"recovering": + (status==DLM_MIGRATING?"migrating": + (status == DLM_FORWARD ? "forward" : + "nolockmanager"))); + actions = 0; } if (flags & LKM_CANCEL) lock->cancel_pending = 0; @@ -231,8 +248,10 @@ leave: if (actions & DLM_UNLOCK_FREE_LOCK) { /* this should always be coupled with list removal */ BUG_ON(!(actions & DLM_UNLOCK_REMOVE_LOCK)); - mlog(0, "lock %"MLFu64" should be gone now! refs=%d\n", - lock->ml.cookie, atomic_read(&lock->lock_refs.refcount)-1); + mlog(0, "lock %u:%llu should be gone now! refs=%d\n", + dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)), + atomic_read(&lock->lock_refs.refcount)-1); dlm_lock_put(lock); } if (actions & DLM_UNLOCK_CALL_AST) @@ -256,8 +275,7 @@ void dlm_commit_pending_unlock(struct dlm_lock_resource *res, void dlm_commit_pending_cancel(struct dlm_lock_resource *res, struct dlm_lock *lock) { - list_del_init(&lock->list); - list_add_tail(&lock->list, &res->granted); + list_move_tail(&lock->list, &res->granted); lock->ml.convert_type = LKM_IVMODE; } @@ -302,7 +320,17 @@ static enum dlm_status dlm_send_remote_unlock_request(struct dlm_ctxt *dlm, struct kvec vec[2]; size_t veclen = 1; - mlog_entry("%.*s\n", res->lockname.len, res->lockname.name); + mlog(0, "%.*s\n", res->lockname.len, res->lockname.name); + + if (owner == dlm->node_num) { + /* ended up trying to contact ourself. this means + * that the lockres had been remote but became local + * via a migration. just retry it, now as local */ + mlog(0, "%s:%.*s: this node became the master due to a " + "migration, re-evaluate now\n", dlm->name, + res->lockname.len, res->lockname.name); + return DLM_FORWARD; + } memset(&unlock, 0, sizeof(unlock)); unlock.node_idx = dlm->node_num; @@ -325,16 +353,12 @@ static enum dlm_status dlm_send_remote_unlock_request(struct dlm_ctxt *dlm, vec, veclen, owner, &status); if (tmpret >= 0) { // successfully sent and received - if (status == DLM_CANCELGRANT) - ret = DLM_NORMAL; - else if (status == DLM_FORWARD) { + if (status == DLM_FORWARD) mlog(0, "master was in-progress. retry\n"); - ret = DLM_FORWARD; - } else - ret = status; - lksb->status = status; + ret = status; } else { - mlog_errno(tmpret); + mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to " + "node %u\n", tmpret, DLM_UNLOCK_LOCK_MSG, dlm->key, owner); if (dlm_is_host_down(tmpret)) { /* NOTE: this seems strange, but it is what we want. * when the master goes down during a cancel or @@ -343,12 +367,14 @@ static enum dlm_status dlm_send_remote_unlock_request(struct dlm_ctxt *dlm, * updated state to the recovery master. this thread * just needs to finish out the operation and call * the unlockast. */ - ret = DLM_NORMAL; + if (dlm_is_node_dead(dlm, owner)) + ret = DLM_NORMAL; + else + ret = DLM_NOLOCKMGR; } else { /* something bad. this will BUG in ocfs2 */ ret = dlm_err_to_dlm_status(tmpret); } - lksb->status = ret; } return ret; @@ -362,12 +388,12 @@ static enum dlm_status dlm_send_remote_unlock_request(struct dlm_ctxt *dlm, * returns: DLM_NORMAL, DLM_BADARGS, DLM_IVLOCKID, * return value from dlmunlock_master */ -int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data) +int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data, + void **ret_data) { struct dlm_ctxt *dlm = data; struct dlm_unlock_lock *unlock = (struct dlm_unlock_lock *)msg->buf; struct dlm_lock_resource *res = NULL; - struct list_head *iter; struct dlm_lock *lock = NULL; enum dlm_status status = DLM_NORMAL; int found = 0, i; @@ -437,8 +463,7 @@ int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data) } for (i=0; i<3; i++) { - list_for_each(iter, queue) { - lock = list_entry(iter, struct dlm_lock, list); + list_for_each_entry(lock, queue, list) { if (lock->ml.cookie == unlock->cookie && lock->ml.node == unlock->node_idx) { dlm_lock_get(lock); @@ -459,6 +484,10 @@ int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data) /* lock was found on queue */ lksb = lock->lksb; + if (flags & (LKM_VALBLK|LKM_PUT_LVB) && + lock->ml.type != LKM_EXMODE) + flags &= ~(LKM_VALBLK|LKM_PUT_LVB); + /* unlockast only called on originating node */ if (flags & LKM_PUT_LVB) { lksb->flags |= DLM_LKSB_PUT_LVB; @@ -480,13 +509,11 @@ int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data) not_found: if (!found) mlog(ML_ERROR, "failed to find lock to unlock! " - "cookie=%"MLFu64"\n", - unlock->cookie); - else { - /* send the lksb->status back to the other node */ - status = lksb->status; + "cookie=%u:%llu\n", + dlm_get_lock_cookie_node(be64_to_cpu(unlock->cookie)), + dlm_get_lock_cookie_seq(be64_to_cpu(unlock->cookie))); + else dlm_lock_put(lock); - } leave: if (res) @@ -508,26 +535,22 @@ static enum dlm_status dlm_get_cancel_actions(struct dlm_ctxt *dlm, if (dlm_lock_on_list(&res->blocked, lock)) { /* cancel this outright */ - lksb->status = DLM_NORMAL; status = DLM_NORMAL; *actions = (DLM_UNLOCK_CALL_AST | DLM_UNLOCK_REMOVE_LOCK); } else if (dlm_lock_on_list(&res->converting, lock)) { /* cancel the request, put back on granted */ - lksb->status = DLM_NORMAL; status = DLM_NORMAL; *actions = (DLM_UNLOCK_CALL_AST | DLM_UNLOCK_REMOVE_LOCK | DLM_UNLOCK_REGRANT_LOCK | DLM_UNLOCK_CLEAR_CONVERT_TYPE); } else if (dlm_lock_on_list(&res->granted, lock)) { - /* too late, already granted. DLM_CANCELGRANT */ - lksb->status = DLM_CANCELGRANT; - status = DLM_NORMAL; + /* too late, already granted. */ + status = DLM_CANCELGRANT; *actions = DLM_UNLOCK_CALL_AST; } else { mlog(ML_ERROR, "lock to cancel is not on any list!\n"); - lksb->status = DLM_IVLOCKID; status = DLM_IVLOCKID; *actions = 0; } @@ -544,13 +567,11 @@ static enum dlm_status dlm_get_unlock_actions(struct dlm_ctxt *dlm, /* unlock request */ if (!dlm_lock_on_list(&res->granted, lock)) { - lksb->status = DLM_DENIED; status = DLM_DENIED; dlm_error(status); *actions = 0; } else { /* unlock granted lock */ - lksb->status = DLM_NORMAL; status = DLM_NORMAL; *actions = (DLM_UNLOCK_FREE_LOCK | DLM_UNLOCK_CALL_AST | @@ -571,8 +592,6 @@ enum dlm_status dlmunlock(struct dlm_ctxt *dlm, struct dlm_lockstatus *lksb, struct dlm_lock *lock = NULL; int call_ast, is_master; - mlog_entry_void(); - if (!lksb) { dlm_error(DLM_BADARGS); return DLM_BADARGS; @@ -607,6 +626,8 @@ retry: spin_lock(&res->spinlock); is_master = (res->owner == dlm->node_num); + if (flags & LKM_VALBLK && lock->ml.type != LKM_EXMODE) + flags &= ~LKM_VALBLK; spin_unlock(&res->spinlock); if (is_master) { @@ -623,7 +644,9 @@ retry: if (status == DLM_RECOVERING || status == DLM_MIGRATING || - status == DLM_FORWARD) { + status == DLM_FORWARD || + status == DLM_NOLOCKMGR) { + /* We want to go away for a tiny bit to allow recovery * / migration to complete on this resource. I don't * know of any wait queue we could sleep on as this @@ -635,26 +658,29 @@ retry: msleep(50); mlog(0, "retrying unlock due to pending recovery/" - "migration/in-progress\n"); + "migration/in-progress/reconnect\n"); goto retry; } if (call_ast) { - mlog(0, "calling unlockast(%p, %d)\n", data, lksb->status); + mlog(0, "calling unlockast(%p, %d)\n", data, status); if (is_master) { - /* it is possible that there is one last bast + /* it is possible that there is one last bast * pending. make sure it is flushed, then * call the unlockast. * not an issue if this is a mastered remotely, * since this lock has been removed from the * lockres queues and cannot be found. */ dlm_kick_thread(dlm, NULL); - wait_event(dlm->ast_wq, + wait_event(dlm->ast_wq, dlm_lock_basts_flushed(dlm, lock)); } - (*unlockast)(data, lksb->status); + (*unlockast)(data, status); } + if (status == DLM_CANCELGRANT) + status = DLM_NORMAL; + if (status == DLM_NORMAL) { mlog(0, "kicking the thread\n"); dlm_kick_thread(dlm, res); diff --git a/fs/ocfs2/dlm/dlmver.c b/fs/ocfs2/dlm/dlmver.c deleted file mode 100644 index 7ef2653f8f4..00000000000 --- a/fs/ocfs2/dlm/dlmver.c +++ /dev/null @@ -1,42 +0,0 @@ -/* -*- mode: c; c-basic-offset: 8; -*- - * vim: noexpandtab sw=8 ts=8 sts=0: - * - * dlmver.c - * - * version string - * - * Copyright (C) 2002, 2005 Oracle. All rights reserved. - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public - * License as published by the Free Software Foundation; either - * version 2 of the License, or (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * General Public License for more details. - * - * You should have received a copy of the GNU General Public - * License along with this program; if not, write to the - * Free Software Foundation, Inc., 59 Temple Place - Suite 330, - * Boston, MA 021110-1307, USA. - */ - -#include <linux/module.h> -#include <linux/kernel.h> - -#include "dlmver.h" - -#define DLM_BUILD_VERSION "1.3.3" - -#define VERSION_STR "OCFS2 DLM " DLM_BUILD_VERSION - -void dlm_print_version(void) -{ - printk(KERN_INFO "%s\n", VERSION_STR); -} - -MODULE_DESCRIPTION(VERSION_STR); - -MODULE_VERSION(DLM_BUILD_VERSION); diff --git a/fs/ocfs2/dlm/dlmver.h b/fs/ocfs2/dlm/dlmver.h deleted file mode 100644 index f674aee77a1..00000000000 --- a/fs/ocfs2/dlm/dlmver.h +++ /dev/null @@ -1,31 +0,0 @@ -/* -*- mode: c; c-basic-offset: 8; -*- - * vim: noexpandtab sw=8 ts=8 sts=0: - * - * dlmfsver.h - * - * Function prototypes - * - * Copyright (C) 2005 Oracle. All rights reserved. - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public - * License as published by the Free Software Foundation; either - * version 2 of the License, or (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * General Public License for more details. - * - * You should have received a copy of the GNU General Public - * License along with this program; if not, write to the - * Free Software Foundation, Inc., 59 Temple Place - Suite 330, - * Boston, MA 021110-1307, USA. - */ - -#ifndef DLM_VER_H -#define DLM_VER_H - -void dlm_print_version(void); - -#endif /* DLM_VER_H */ diff --git a/fs/ocfs2/dlm/userdlm.c b/fs/ocfs2/dlm/userdlm.c deleted file mode 100644 index e1fdd288796..00000000000 --- a/fs/ocfs2/dlm/userdlm.c +++ /dev/null @@ -1,658 +0,0 @@ -/* -*- mode: c; c-basic-offset: 8; -*- - * vim: noexpandtab sw=8 ts=8 sts=0: - * - * userdlm.c - * - * Code which implements the kernel side of a minimal userspace - * interface to our DLM. - * - * Many of the functions here are pared down versions of dlmglue.c - * functions. - * - * Copyright (C) 2003, 2004 Oracle. All rights reserved. - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public - * License as published by the Free Software Foundation; either - * version 2 of the License, or (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * General Public License for more details. - * - * You should have received a copy of the GNU General Public - * License along with this program; if not, write to the - * Free Software Foundation, Inc., 59 Temple Place - Suite 330, - * Boston, MA 021110-1307, USA. - */ - -#include <asm/signal.h> - -#include <linux/module.h> -#include <linux/fs.h> -#include <linux/types.h> -#include <linux/crc32.h> - - -#include "cluster/nodemanager.h" -#include "cluster/heartbeat.h" -#include "cluster/tcp.h" - -#include "dlmapi.h" - -#include "userdlm.h" - -#define MLOG_MASK_PREFIX ML_DLMFS -#include "cluster/masklog.h" - -static inline int user_check_wait_flag(struct user_lock_res *lockres, - int flag) -{ - int ret; - - spin_lock(&lockres->l_lock); - ret = lockres->l_flags & flag; - spin_unlock(&lockres->l_lock); - - return ret; -} - -static inline void user_wait_on_busy_lock(struct user_lock_res *lockres) - -{ - wait_event(lockres->l_event, - !user_check_wait_flag(lockres, USER_LOCK_BUSY)); -} - -static inline void user_wait_on_blocked_lock(struct user_lock_res *lockres) - -{ - wait_event(lockres->l_event, - !user_check_wait_flag(lockres, USER_LOCK_BLOCKED)); -} - -/* I heart container_of... */ -static inline struct dlm_ctxt * -dlm_ctxt_from_user_lockres(struct user_lock_res *lockres) -{ - struct dlmfs_inode_private *ip; - - ip = container_of(lockres, - struct dlmfs_inode_private, - ip_lockres); - return ip->ip_dlm; -} - -static struct inode * -user_dlm_inode_from_user_lockres(struct user_lock_res *lockres) -{ - struct dlmfs_inode_private *ip; - - ip = container_of(lockres, - struct dlmfs_inode_private, - ip_lockres); - return &ip->ip_vfs_inode; -} - -static inline void user_recover_from_dlm_error(struct user_lock_res *lockres) -{ - spin_lock(&lockres->l_lock); - lockres->l_flags &= ~USER_LOCK_BUSY; - spin_unlock(&lockres->l_lock); -} - -#define user_log_dlm_error(_func, _stat, _lockres) do { \ - mlog(ML_ERROR, "Dlm error \"%s\" while calling %s on " \ - "resource %s: %s\n", dlm_errname(_stat), _func, \ - _lockres->l_name, dlm_errmsg(_stat)); \ -} while (0) - -/* WARNING: This function lives in a world where the only three lock - * levels are EX, PR, and NL. It *will* have to be adjusted when more - * lock types are added. */ -static inline int user_highest_compat_lock_level(int level) -{ - int new_level = LKM_EXMODE; - - if (level == LKM_EXMODE) - new_level = LKM_NLMODE; - else if (level == LKM_PRMODE) - new_level = LKM_PRMODE; - return new_level; -} - -static void user_ast(void *opaque) -{ - struct user_lock_res *lockres = opaque; - struct dlm_lockstatus *lksb; - - mlog(0, "AST fired for lockres %s\n", lockres->l_name); - - spin_lock(&lockres->l_lock); - - lksb = &(lockres->l_lksb); - if (lksb->status != DLM_NORMAL) { - mlog(ML_ERROR, "lksb status value of %u on lockres %s\n", - lksb->status, lockres->l_name); - spin_unlock(&lockres->l_lock); - return; - } - - /* we're downconverting. */ - if (lockres->l_requested < lockres->l_level) { - if (lockres->l_requested <= - user_highest_compat_lock_level(lockres->l_blocking)) { - lockres->l_blocking = LKM_NLMODE; - lockres->l_flags &= ~USER_LOCK_BLOCKED; - } - } - - lockres->l_level = lockres->l_requested; - lockres->l_requested = LKM_IVMODE; - lockres->l_flags |= USER_LOCK_ATTACHED; - lockres->l_flags &= ~USER_LOCK_BUSY; - - spin_unlock(&lockres->l_lock); - - wake_up(&lockres->l_event); -} - -static inline void user_dlm_grab_inode_ref(struct user_lock_res *lockres) -{ - struct inode *inode; - inode = user_dlm_inode_from_user_lockres(lockres); - if (!igrab(inode)) - BUG(); -} - -static void user_dlm_unblock_lock(void *opaque); - -static void __user_dlm_queue_lockres(struct user_lock_res *lockres) -{ - if (!(lockres->l_flags & USER_LOCK_QUEUED)) { - user_dlm_grab_inode_ref(lockres); - - INIT_WORK(&lockres->l_work, user_dlm_unblock_lock, - lockres); - - queue_work(user_dlm_worker, &lockres->l_work); - lockres->l_flags |= USER_LOCK_QUEUED; - } -} - -static void __user_dlm_cond_queue_lockres(struct user_lock_res *lockres) -{ - int queue = 0; - - if (!(lockres->l_flags & USER_LOCK_BLOCKED)) - return; - - switch (lockres->l_blocking) { - case LKM_EXMODE: - if (!lockres->l_ex_holders && !lockres->l_ro_holders) - queue = 1; - break; - case LKM_PRMODE: - if (!lockres->l_ex_holders) - queue = 1; - break; - default: - BUG(); - } - - if (queue) - __user_dlm_queue_lockres(lockres); -} - -static void user_bast(void *opaque, int level) -{ - struct user_lock_res *lockres = opaque; - - mlog(0, "Blocking AST fired for lockres %s. Blocking level %d\n", - lockres->l_name, level); - - spin_lock(&lockres->l_lock); - lockres->l_flags |= USER_LOCK_BLOCKED; - if (level > lockres->l_blocking) - lockres->l_blocking = level; - - __user_dlm_queue_lockres(lockres); - spin_unlock(&lockres->l_lock); - - wake_up(&lockres->l_event); -} - -static void user_unlock_ast(void *opaque, enum dlm_status status) -{ - struct user_lock_res *lockres = opaque; - - mlog(0, "UNLOCK AST called on lock %s\n", lockres->l_name); - - if (status != DLM_NORMAL) - mlog(ML_ERROR, "Dlm returns status %d\n", status); - - spin_lock(&lockres->l_lock); - if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) - lockres->l_level = LKM_IVMODE; - else { - lockres->l_requested = LKM_IVMODE; /* cancel an - * upconvert - * request. */ - lockres->l_flags &= ~USER_LOCK_IN_CANCEL; - /* we want the unblock thread to look at it again - * now. */ - __user_dlm_queue_lockres(lockres); - } - - lockres->l_flags &= ~USER_LOCK_BUSY; - spin_unlock(&lockres->l_lock); - - wake_up(&lockres->l_event); -} - -static inline void user_dlm_drop_inode_ref(struct user_lock_res *lockres) -{ - struct inode *inode; - inode = user_dlm_inode_from_user_lockres(lockres); - iput(inode); -} - -static void user_dlm_unblock_lock(void *opaque) -{ - int new_level, status; - struct user_lock_res *lockres = (struct user_lock_res *) opaque; - struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres); - - mlog(0, "processing lockres %s\n", lockres->l_name); - - spin_lock(&lockres->l_lock); - - BUG_ON(!(lockres->l_flags & USER_LOCK_BLOCKED)); - BUG_ON(!(lockres->l_flags & USER_LOCK_QUEUED)); - - /* notice that we don't clear USER_LOCK_BLOCKED here. That's - * for user_ast to do. */ - lockres->l_flags &= ~USER_LOCK_QUEUED; - - if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) { - mlog(0, "lock is in teardown so we do nothing\n"); - spin_unlock(&lockres->l_lock); - goto drop_ref; - } - - if (lockres->l_flags & USER_LOCK_BUSY) { - mlog(0, "BUSY flag detected...\n"); - if (lockres->l_flags & USER_LOCK_IN_CANCEL) { - spin_unlock(&lockres->l_lock); - goto drop_ref; - } - - lockres->l_flags |= USER_LOCK_IN_CANCEL; - spin_unlock(&lockres->l_lock); - - status = dlmunlock(dlm, - &lockres->l_lksb, - LKM_CANCEL, - user_unlock_ast, - lockres); - if (status == DLM_CANCELGRANT) { - /* If we got this, then the ast was fired - * before we could cancel. We cleanup our - * state, and restart the function. */ - spin_lock(&lockres->l_lock); - lockres->l_flags &= ~USER_LOCK_IN_CANCEL; - spin_unlock(&lockres->l_lock); - } else if (status != DLM_NORMAL) - user_log_dlm_error("dlmunlock", status, lockres); - goto drop_ref; - } - - /* If there are still incompat holders, we can exit safely - * without worrying about re-queueing this lock as that will - * happen on the last call to user_cluster_unlock. */ - if ((lockres->l_blocking == LKM_EXMODE) - && (lockres->l_ex_holders || lockres->l_ro_holders)) { - spin_unlock(&lockres->l_lock); - mlog(0, "can't downconvert for ex: ro = %u, ex = %u\n", - lockres->l_ro_holders, lockres->l_ex_holders); - goto drop_ref; - } - - if ((lockres->l_blocking == LKM_PRMODE) - && lockres->l_ex_holders) { - spin_unlock(&lockres->l_lock); - mlog(0, "can't downconvert for pr: ex = %u\n", - lockres->l_ex_holders); - goto drop_ref; - } - - /* yay, we can downconvert now. */ - new_level = user_highest_compat_lock_level(lockres->l_blocking); - lockres->l_requested = new_level; - lockres->l_flags |= USER_LOCK_BUSY; - mlog(0, "Downconvert lock from %d to %d\n", - lockres->l_level, new_level); - spin_unlock(&lockres->l_lock); - - /* need lock downconvert request now... */ - status = dlmlock(dlm, - new_level, - &lockres->l_lksb, - LKM_CONVERT|LKM_VALBLK, - lockres->l_name, - user_ast, - lockres, - user_bast); - if (status != DLM_NORMAL) { - user_log_dlm_error("dlmlock", status, lockres); - user_recover_from_dlm_error(lockres); - } - -drop_ref: - user_dlm_drop_inode_ref(lockres); -} - -static inline void user_dlm_inc_holders(struct user_lock_res *lockres, - int level) -{ - switch(level) { - case LKM_EXMODE: - lockres->l_ex_holders++; - break; - case LKM_PRMODE: - lockres->l_ro_holders++; - break; - default: - BUG(); - } -} - -/* predict what lock level we'll be dropping down to on behalf - * of another node, and return true if the currently wanted - * level will be compatible with it. */ -static inline int -user_may_continue_on_blocked_lock(struct user_lock_res *lockres, - int wanted) -{ - BUG_ON(!(lockres->l_flags & USER_LOCK_BLOCKED)); - - return wanted <= user_highest_compat_lock_level(lockres->l_blocking); -} - -int user_dlm_cluster_lock(struct user_lock_res *lockres, - int level, - int lkm_flags) -{ - int status, local_flags; - struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres); - - if (level != LKM_EXMODE && - level != LKM_PRMODE) { - mlog(ML_ERROR, "lockres %s: invalid request!\n", - lockres->l_name); - status = -EINVAL; - goto bail; - } - - mlog(0, "lockres %s: asking for %s lock, passed flags = 0x%x\n", - lockres->l_name, - (level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE", - lkm_flags); - -again: - if (signal_pending(current)) { - status = -ERESTARTSYS; - goto bail; - } - - spin_lock(&lockres->l_lock); - - /* We only compare against the currently granted level - * here. If the lock is blocked waiting on a downconvert, - * we'll get caught below. */ - if ((lockres->l_flags & USER_LOCK_BUSY) && - (level > lockres->l_level)) { - /* is someone sitting in dlm_lock? If so, wait on - * them. */ - spin_unlock(&lockres->l_lock); - - user_wait_on_busy_lock(lockres); - goto again; - } - - if ((lockres->l_flags & USER_LOCK_BLOCKED) && - (!user_may_continue_on_blocked_lock(lockres, level))) { - /* is the lock is currently blocked on behalf of - * another node */ - spin_unlock(&lockres->l_lock); - - user_wait_on_blocked_lock(lockres); - goto again; - } - - if (level > lockres->l_level) { - local_flags = lkm_flags | LKM_VALBLK; - if (lockres->l_level != LKM_IVMODE) - local_flags |= LKM_CONVERT; - - lockres->l_requested = level; - lockres->l_flags |= USER_LOCK_BUSY; - spin_unlock(&lockres->l_lock); - - BUG_ON(level == LKM_IVMODE); - BUG_ON(level == LKM_NLMODE); - - mlog(0, "lock %s, get lock from %d to level = %d\n", - lockres->l_name, lockres->l_level, level); - - /* call dlm_lock to upgrade lock now */ - status = dlmlock(dlm, - level, - &lockres->l_lksb, - local_flags, - lockres->l_name, - user_ast, - lockres, - user_bast); - if (status != DLM_NORMAL) { - if ((lkm_flags & LKM_NOQUEUE) && - (status == DLM_NOTQUEUED)) - status = -EAGAIN; - else { - user_log_dlm_error("dlmlock", status, lockres); - status = -EINVAL; - } - user_recover_from_dlm_error(lockres); - goto bail; - } - - mlog(0, "lock %s, successfull return from dlmlock\n", - lockres->l_name); - - user_wait_on_busy_lock(lockres); - goto again; - } - - user_dlm_inc_holders(lockres, level); - spin_unlock(&lockres->l_lock); - - mlog(0, "lockres %s: Got %s lock!\n", lockres->l_name, - (level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE"); - - status = 0; -bail: - return status; -} - -static inline void user_dlm_dec_holders(struct user_lock_res *lockres, - int level) -{ - switch(level) { - case LKM_EXMODE: - BUG_ON(!lockres->l_ex_holders); - lockres->l_ex_holders--; - break; - case LKM_PRMODE: - BUG_ON(!lockres->l_ro_holders); - lockres->l_ro_holders--; - break; - default: - BUG(); - } -} - -void user_dlm_cluster_unlock(struct user_lock_res *lockres, - int level) -{ - if (level != LKM_EXMODE && - level != LKM_PRMODE) { - mlog(ML_ERROR, "lockres %s: invalid request!\n", lockres->l_name); - return; - } - - mlog(0, "lockres %s: dropping %s lock\n", lockres->l_name, - (level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE"); - - spin_lock(&lockres->l_lock); - user_dlm_dec_holders(lockres, level); - __user_dlm_cond_queue_lockres(lockres); - spin_unlock(&lockres->l_lock); -} - -void user_dlm_write_lvb(struct inode *inode, - const char *val, - unsigned int len) -{ - struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres; - char *lvb = lockres->l_lksb.lvb; - - BUG_ON(len > DLM_LVB_LEN); - - spin_lock(&lockres->l_lock); - - BUG_ON(lockres->l_level < LKM_EXMODE); - memcpy(lvb, val, len); - - spin_unlock(&lockres->l_lock); -} - -void user_dlm_read_lvb(struct inode *inode, - char *val, - unsigned int len) -{ - struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres; - char *lvb = lockres->l_lksb.lvb; - - BUG_ON(len > DLM_LVB_LEN); - - spin_lock(&lockres->l_lock); - - BUG_ON(lockres->l_level < LKM_PRMODE); - memcpy(val, lvb, len); - - spin_unlock(&lockres->l_lock); -} - -void user_dlm_lock_res_init(struct user_lock_res *lockres, - struct dentry *dentry) -{ - memset(lockres, 0, sizeof(*lockres)); - - spin_lock_init(&lockres->l_lock); - init_waitqueue_head(&lockres->l_event); - lockres->l_level = LKM_IVMODE; - lockres->l_requested = LKM_IVMODE; - lockres->l_blocking = LKM_IVMODE; - - /* should have been checked before getting here. */ - BUG_ON(dentry->d_name.len >= USER_DLM_LOCK_ID_MAX_LEN); - - memcpy(lockres->l_name, - dentry->d_name.name, - dentry->d_name.len); -} - -int user_dlm_destroy_lock(struct user_lock_res *lockres) -{ - int status = -EBUSY; - struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres); - - mlog(0, "asked to destroy %s\n", lockres->l_name); - - spin_lock(&lockres->l_lock); - while (lockres->l_flags & USER_LOCK_BUSY) { - spin_unlock(&lockres->l_lock); - - mlog(0, "lock %s is busy\n", lockres->l_name); - - user_wait_on_busy_lock(lockres); - - spin_lock(&lockres->l_lock); - } - - if (lockres->l_ro_holders || lockres->l_ex_holders) { - spin_unlock(&lockres->l_lock); - mlog(0, "lock %s has holders\n", lockres->l_name); - goto bail; - } - - status = 0; - if (!(lockres->l_flags & USER_LOCK_ATTACHED)) { - spin_unlock(&lockres->l_lock); - mlog(0, "lock %s is not attached\n", lockres->l_name); - goto bail; - } - - lockres->l_flags &= ~USER_LOCK_ATTACHED; - lockres->l_flags |= USER_LOCK_BUSY; - lockres->l_flags |= USER_LOCK_IN_TEARDOWN; - spin_unlock(&lockres->l_lock); - - mlog(0, "unlocking lockres %s\n", lockres->l_name); - status = dlmunlock(dlm, - &lockres->l_lksb, - LKM_VALBLK, - user_unlock_ast, - lockres); - if (status != DLM_NORMAL) { - user_log_dlm_error("dlmunlock", status, lockres); - status = -EINVAL; - goto bail; - } - - user_wait_on_busy_lock(lockres); - - status = 0; -bail: - return status; -} - -struct dlm_ctxt *user_dlm_register_context(struct qstr *name) -{ - struct dlm_ctxt *dlm; - u32 dlm_key; - char *domain; - - domain = kmalloc(name->len + 1, GFP_KERNEL); - if (!domain) { - mlog_errno(-ENOMEM); - return ERR_PTR(-ENOMEM); - } - - dlm_key = crc32_le(0, name->name, name->len); - - snprintf(domain, name->len + 1, "%.*s", name->len, name->name); - - dlm = dlm_register_domain(domain, dlm_key); - if (IS_ERR(dlm)) - mlog_errno(PTR_ERR(dlm)); - - kfree(domain); - return dlm; -} - -void user_dlm_unregister_context(struct dlm_ctxt *dlm) -{ - dlm_unregister_domain(dlm); -} diff --git a/fs/ocfs2/dlm/userdlm.h b/fs/ocfs2/dlm/userdlm.h deleted file mode 100644 index 04178bc40b7..00000000000 --- a/fs/ocfs2/dlm/userdlm.h +++ /dev/null @@ -1,111 +0,0 @@ -/* -*- mode: c; c-basic-offset: 8; -*- - * vim: noexpandtab sw=8 ts=8 sts=0: - * - * userdlm.h - * - * Userspace dlm defines - * - * Copyright (C) 2002, 2004 Oracle. All rights reserved. - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public - * License as published by the Free Software Foundation; either - * version 2 of the License, or (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * General Public License for more details. - * - * You should have received a copy of the GNU General Public - * License along with this program; if not, write to the - * Free Software Foundation, Inc., 59 Temple Place - Suite 330, - * Boston, MA 021110-1307, USA. - */ - - -#ifndef USERDLM_H -#define USERDLM_H - -#include <linux/module.h> -#include <linux/fs.h> -#include <linux/types.h> -#include <linux/workqueue.h> - -/* user_lock_res->l_flags flags. */ -#define USER_LOCK_ATTACHED (0x00000001) /* have we initialized - * the lvb */ -#define USER_LOCK_BUSY (0x00000002) /* we are currently in - * dlm_lock */ -#define USER_LOCK_BLOCKED (0x00000004) /* blocked waiting to - * downconvert*/ -#define USER_LOCK_IN_TEARDOWN (0x00000008) /* we're currently - * destroying this - * lock. */ -#define USER_LOCK_QUEUED (0x00000010) /* lock is on the - * workqueue */ -#define USER_LOCK_IN_CANCEL (0x00000020) - -struct user_lock_res { - spinlock_t l_lock; - - int l_flags; - -#define USER_DLM_LOCK_ID_MAX_LEN 32 - char l_name[USER_DLM_LOCK_ID_MAX_LEN]; - int l_level; - unsigned int l_ro_holders; - unsigned int l_ex_holders; - struct dlm_lockstatus l_lksb; - - int l_requested; - int l_blocking; - - wait_queue_head_t l_event; - - struct work_struct l_work; -}; - -extern struct workqueue_struct *user_dlm_worker; - -void user_dlm_lock_res_init(struct user_lock_res *lockres, - struct dentry *dentry); -int user_dlm_destroy_lock(struct user_lock_res *lockres); -int user_dlm_cluster_lock(struct user_lock_res *lockres, - int level, - int lkm_flags); -void user_dlm_cluster_unlock(struct user_lock_res *lockres, - int level); -void user_dlm_write_lvb(struct inode *inode, - const char *val, - unsigned int len); -void user_dlm_read_lvb(struct inode *inode, - char *val, - unsigned int len); -struct dlm_ctxt *user_dlm_register_context(struct qstr *name); -void user_dlm_unregister_context(struct dlm_ctxt *dlm); - -struct dlmfs_inode_private { - struct dlm_ctxt *ip_dlm; - - struct user_lock_res ip_lockres; /* unused for directories. */ - struct inode *ip_parent; - - struct inode ip_vfs_inode; -}; - -static inline struct dlmfs_inode_private * -DLMFS_I(struct inode *inode) -{ - return container_of(inode, - struct dlmfs_inode_private, - ip_vfs_inode); -} - -struct dlmfs_filp_private { - int fp_lock_level; -}; - -#define DLMFS_MAGIC 0x76a9f425 - -#endif /* USERDLM_H */ |