diff options
Diffstat (limited to 'fs/ocfs2/dlmglue.c')
| -rw-r--r-- | fs/ocfs2/dlmglue.c | 1723 |
1 files changed, 1210 insertions, 513 deletions
diff --git a/fs/ocfs2/dlmglue.c b/fs/ocfs2/dlmglue.c index 1f1873bf41f..52cfe99ae05 100644 --- a/fs/ocfs2/dlmglue.c +++ b/fs/ocfs2/dlmglue.c @@ -27,17 +27,12 @@ #include <linux/slab.h> #include <linux/highmem.h> #include <linux/mm.h> -#include <linux/crc32.h> #include <linux/kthread.h> #include <linux/pagemap.h> #include <linux/debugfs.h> #include <linux/seq_file.h> - -#include <cluster/heartbeat.h> -#include <cluster/nodemanager.h> -#include <cluster/tcp.h> - -#include <dlm/dlmapi.h> +#include <linux/time.h> +#include <linux/quotaops.h> #define MLOG_MASK_PREFIX ML_DLM_GLUE #include <cluster/masklog.h> @@ -53,9 +48,12 @@ #include "heartbeat.h" #include "inode.h" #include "journal.h" +#include "stackglue.h" #include "slot_map.h" #include "super.h" #include "uptodate.h" +#include "quota.h" +#include "refcounttree.h" #include "buffer_head_io.h" @@ -65,11 +63,15 @@ struct ocfs2_mask_waiter { struct completion mw_complete; unsigned long mw_mask; unsigned long mw_goal; +#ifdef CONFIG_OCFS2_FS_STATS + ktime_t mw_lock_start; +#endif }; static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres); static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres); static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres); +static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres); /* * Return value from ->downconvert_worker functions. @@ -91,6 +93,9 @@ struct ocfs2_unblock_ctl { enum ocfs2_unblock_action unblock_action; }; +/* Lockdep class keys */ +struct lock_class_key lockdep_keys[OCFS2_NUM_LOCK_TYPES]; + static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres, int new_level); static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres); @@ -104,6 +109,12 @@ static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres, static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb, struct ocfs2_lock_res *lockres); +static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres); + +static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres, + int new_level); +static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres, + int blocking); #define mlog_meta_lvb(__level, __lockres) ocfs2_dump_meta_lvb_info(__level, __PRETTY_FUNCTION__, __LINE__, __lockres) @@ -113,7 +124,7 @@ static void ocfs2_dump_meta_lvb_info(u64 level, unsigned int line, struct ocfs2_lock_res *lockres) { - struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb; + struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb); mlog(level, "LVB information for %s (called from %s:%u):\n", lockres->l_name, function, line); @@ -242,6 +253,14 @@ static struct ocfs2_lock_res_ops ocfs2_rename_lops = { .flags = 0, }; +static struct ocfs2_lock_res_ops ocfs2_nfs_sync_lops = { + .flags = 0, +}; + +static struct ocfs2_lock_res_ops ocfs2_orphan_scan_lops = { + .flags = LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB, +}; + static struct ocfs2_lock_res_ops ocfs2_dentry_lops = { .get_osb = ocfs2_get_dentry_osb, .post_unlock = ocfs2_dentry_post_unlock, @@ -259,29 +278,16 @@ static struct ocfs2_lock_res_ops ocfs2_flock_lops = { .flags = 0, }; -/* - * This is the filesystem locking protocol version. - * - * Whenever the filesystem does new things with locks (adds or removes a - * lock, orders them differently, does different things underneath a lock), - * the version must be changed. The protocol is negotiated when joining - * the dlm domain. A node may join the domain if its major version is - * identical to all other nodes and its minor version is greater than - * or equal to all other nodes. When its minor version is greater than - * the other nodes, it will run at the minor version specified by the - * other nodes. - * - * If a locking change is made that will not be compatible with older - * versions, the major number must be increased and the minor version set - * to zero. If a change merely adds a behavior that can be disabled when - * speaking to older versions, the minor version must be increased. If a - * change adds a fully backwards compatible change (eg, LVB changes that - * are just ignored by older versions), the version does not need to be - * updated. - */ -const struct dlm_protocol_version ocfs2_locking_protocol = { - .pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR, - .pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR, +static struct ocfs2_lock_res_ops ocfs2_qinfo_lops = { + .set_lvb = ocfs2_set_qinfo_lvb, + .get_osb = ocfs2_get_qinfo_osb, + .flags = LOCK_TYPE_REQUIRES_REFRESH | LOCK_TYPE_USES_LVB, +}; + +static struct ocfs2_lock_res_ops ocfs2_refcount_block_lops = { + .check_downconvert = ocfs2_check_refcount_downconvert, + .downconvert_worker = ocfs2_refcount_convert_worker, + .flags = 0, }; static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres) @@ -291,6 +297,11 @@ static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres) lockres->l_type == OCFS2_LOCK_TYPE_OPEN; } +static inline struct ocfs2_lock_res *ocfs2_lksb_to_lock_res(struct ocfs2_dlm_lksb *lksb) +{ + return container_of(lksb, struct ocfs2_lock_res, l_lksb); +} + static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres) { BUG_ON(!ocfs2_is_inode_lock(lockres)); @@ -305,6 +316,19 @@ static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res return (struct ocfs2_dentry_lock *)lockres->l_priv; } +static inline struct ocfs2_mem_dqinfo *ocfs2_lock_res_qinfo(struct ocfs2_lock_res *lockres) +{ + BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_QINFO); + + return (struct ocfs2_mem_dqinfo *)lockres->l_priv; +} + +static inline struct ocfs2_refcount_tree * +ocfs2_lock_res_refcount_tree(struct ocfs2_lock_res *res) +{ + return container_of(res, struct ocfs2_refcount_tree, rf_lockres); +} + static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres) { if (lockres->l_ops->get_osb) @@ -316,12 +340,19 @@ static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *l static int ocfs2_lock_create(struct ocfs2_super *osb, struct ocfs2_lock_res *lockres, int level, - int dlm_flags); + u32 dlm_flags); static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres, int wanted); -static void ocfs2_cluster_unlock(struct ocfs2_super *osb, - struct ocfs2_lock_res *lockres, - int level); +static void __ocfs2_cluster_unlock(struct ocfs2_super *osb, + struct ocfs2_lock_res *lockres, + int level, unsigned long caller_ip); +static inline void ocfs2_cluster_unlock(struct ocfs2_super *osb, + struct ocfs2_lock_res *lockres, + int level) +{ + __ocfs2_cluster_unlock(osb, lockres, level, _RET_IP_); +} + static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres); static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres); static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres); @@ -330,10 +361,14 @@ static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb, struct ocfs2_lock_res *lockres); static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres, int convert); -#define ocfs2_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)); \ +#define ocfs2_log_dlm_error(_func, _err, _lockres) do { \ + if ((_lockres)->l_type != OCFS2_LOCK_TYPE_DENTRY) \ + mlog(ML_ERROR, "DLM error %d while calling %s on resource %s\n", \ + _err, _func, _lockres->l_name); \ + else \ + mlog(ML_ERROR, "DLM error %d while calling %s on resource %.*s%08x\n", \ + _err, _func, OCFS2_DENTRY_LOCK_INO_START - 1, (_lockres)->l_name, \ + (unsigned int)ocfs2_get_dentry_lock_ino(_lockres)); \ } while (0) static int ocfs2_downconvert_thread(void *arg); static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb, @@ -342,12 +377,13 @@ static int ocfs2_inode_lock_update(struct inode *inode, struct buffer_head **bh); static void ocfs2_drop_osb_locks(struct ocfs2_super *osb); static inline int ocfs2_highest_compat_lock_level(int level); -static void ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres, - int new_level); +static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres, + int new_level); static int ocfs2_downconvert_lock(struct ocfs2_super *osb, struct ocfs2_lock_res *lockres, int new_level, - int lvb); + int lvb, + unsigned int generation); static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb, struct ocfs2_lock_res *lockres); static int ocfs2_cancel_convert(struct ocfs2_super *osb, @@ -361,8 +397,6 @@ static void ocfs2_build_lock_name(enum ocfs2_lock_type type, { int len; - mlog_entry_void(); - BUG_ON(type >= OCFS2_NUM_LOCK_TYPES); len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x", @@ -372,8 +406,6 @@ static void ocfs2_build_lock_name(enum ocfs2_lock_type type, BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1)); mlog(0, "built lock resource with name: %s\n", name); - - mlog_exit_void(); } static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock); @@ -396,6 +428,71 @@ static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res) spin_unlock(&ocfs2_dlm_tracking_lock); } +#ifdef CONFIG_OCFS2_FS_STATS +static void ocfs2_init_lock_stats(struct ocfs2_lock_res *res) +{ + res->l_lock_refresh = 0; + memset(&res->l_lock_prmode, 0, sizeof(struct ocfs2_lock_stats)); + memset(&res->l_lock_exmode, 0, sizeof(struct ocfs2_lock_stats)); +} + +static void ocfs2_update_lock_stats(struct ocfs2_lock_res *res, int level, + struct ocfs2_mask_waiter *mw, int ret) +{ + u32 usec; + ktime_t kt; + struct ocfs2_lock_stats *stats; + + if (level == LKM_PRMODE) + stats = &res->l_lock_prmode; + else if (level == LKM_EXMODE) + stats = &res->l_lock_exmode; + else + return; + + kt = ktime_sub(ktime_get(), mw->mw_lock_start); + usec = ktime_to_us(kt); + + stats->ls_gets++; + stats->ls_total += ktime_to_ns(kt); + /* overflow */ + if (unlikely(stats->ls_gets == 0)) { + stats->ls_gets++; + stats->ls_total = ktime_to_ns(kt); + } + + if (stats->ls_max < usec) + stats->ls_max = usec; + + if (ret) + stats->ls_fail++; +} + +static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres) +{ + lockres->l_lock_refresh++; +} + +static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw) +{ + mw->mw_lock_start = ktime_get(); +} +#else +static inline void ocfs2_init_lock_stats(struct ocfs2_lock_res *res) +{ +} +static inline void ocfs2_update_lock_stats(struct ocfs2_lock_res *res, + int level, struct ocfs2_mask_waiter *mw, int ret) +{ +} +static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres) +{ +} +static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw) +{ +} +#endif + static void ocfs2_lock_res_init_common(struct ocfs2_super *osb, struct ocfs2_lock_res *res, enum ocfs2_lock_type type, @@ -406,15 +503,24 @@ static void ocfs2_lock_res_init_common(struct ocfs2_super *osb, res->l_ops = ops; res->l_priv = priv; - res->l_level = LKM_IVMODE; - res->l_requested = LKM_IVMODE; - res->l_blocking = LKM_IVMODE; + res->l_level = DLM_LOCK_IV; + res->l_requested = DLM_LOCK_IV; + res->l_blocking = DLM_LOCK_IV; res->l_action = OCFS2_AST_INVALID; res->l_unlock_action = OCFS2_UNLOCK_INVALID; res->l_flags = OCFS2_LOCK_INITIALIZED; ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug); + + ocfs2_init_lock_stats(res); +#ifdef CONFIG_DEBUG_LOCK_ALLOC + if (type != OCFS2_LOCK_TYPE_OPEN) + lockdep_init_map(&res->l_lockdep_map, ocfs2_lock_type_strings[type], + &lockdep_keys[type], 0); + else + res->l_lockdep_map.key = NULL; +#endif } void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res) @@ -462,6 +568,13 @@ static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres) return OCFS2_SB(inode->i_sb); } +static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres) +{ + struct ocfs2_mem_dqinfo *info = lockres->l_priv; + + return OCFS2_SB(info->dqi_gi.dqi_sb); +} + static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres) { struct ocfs2_file_private *fp = lockres->l_priv; @@ -549,6 +662,26 @@ static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res, &ocfs2_rename_lops, osb); } +static void ocfs2_nfs_sync_lock_res_init(struct ocfs2_lock_res *res, + struct ocfs2_super *osb) +{ + /* nfs_sync lockres doesn't come from a slab so we call init + * once on it manually. */ + ocfs2_lock_res_init_once(res); + ocfs2_build_lock_name(OCFS2_LOCK_TYPE_NFS_SYNC, 0, 0, res->l_name); + ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_NFS_SYNC, + &ocfs2_nfs_sync_lops, osb); +} + +static void ocfs2_orphan_scan_lock_res_init(struct ocfs2_lock_res *res, + struct ocfs2_super *osb) +{ + ocfs2_lock_res_init_once(res); + ocfs2_build_lock_name(OCFS2_LOCK_TYPE_ORPHAN_SCAN, 0, 0, res->l_name); + ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_ORPHAN_SCAN, + &ocfs2_orphan_scan_lops, osb); +} + void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres, struct ocfs2_file_private *fp) { @@ -564,10 +697,30 @@ void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres, lockres->l_flags |= OCFS2_LOCK_NOCACHE; } -void ocfs2_lock_res_free(struct ocfs2_lock_res *res) +void ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res *lockres, + struct ocfs2_mem_dqinfo *info) { - mlog_entry_void(); + ocfs2_lock_res_init_once(lockres); + ocfs2_build_lock_name(OCFS2_LOCK_TYPE_QINFO, info->dqi_gi.dqi_type, + 0, lockres->l_name); + ocfs2_lock_res_init_common(OCFS2_SB(info->dqi_gi.dqi_sb), lockres, + OCFS2_LOCK_TYPE_QINFO, &ocfs2_qinfo_lops, + info); +} +void ocfs2_refcount_lock_res_init(struct ocfs2_lock_res *lockres, + struct ocfs2_super *osb, u64 ref_blkno, + unsigned int generation) +{ + ocfs2_lock_res_init_once(lockres); + ocfs2_build_lock_name(OCFS2_LOCK_TYPE_REFCOUNT, ref_blkno, + generation, lockres->l_name); + ocfs2_lock_res_init_common(osb, lockres, OCFS2_LOCK_TYPE_REFCOUNT, + &ocfs2_refcount_block_lops, osb); +} + +void ocfs2_lock_res_free(struct ocfs2_lock_res *res) +{ if (!(res->l_flags & OCFS2_LOCK_INITIALIZED)) return; @@ -593,50 +746,42 @@ void ocfs2_lock_res_free(struct ocfs2_lock_res *res) memset(&res->l_lksb, 0, sizeof(res->l_lksb)); res->l_flags = 0UL; - mlog_exit_void(); } static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres, int level) { - mlog_entry_void(); - BUG_ON(!lockres); switch(level) { - case LKM_EXMODE: + case DLM_LOCK_EX: lockres->l_ex_holders++; break; - case LKM_PRMODE: + case DLM_LOCK_PR: lockres->l_ro_holders++; break; default: BUG(); } - - mlog_exit_void(); } static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres, int level) { - mlog_entry_void(); - BUG_ON(!lockres); switch(level) { - case LKM_EXMODE: + case DLM_LOCK_EX: BUG_ON(!lockres->l_ex_holders); lockres->l_ex_holders--; break; - case LKM_PRMODE: + case DLM_LOCK_PR: BUG_ON(!lockres->l_ro_holders); lockres->l_ro_holders--; break; default: BUG(); } - mlog_exit_void(); } /* WARNING: This function lives in a world where the only three lock @@ -644,12 +789,12 @@ static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres, * lock types are added. */ static inline int ocfs2_highest_compat_lock_level(int level) { - int new_level = LKM_EXMODE; + int new_level = DLM_LOCK_EX; - if (level == LKM_EXMODE) - new_level = LKM_NLMODE; - else if (level == LKM_PRMODE) - new_level = LKM_PRMODE; + if (level == DLM_LOCK_EX) + new_level = DLM_LOCK_NL; + else if (level == DLM_LOCK_PR) + new_level = DLM_LOCK_PR; return new_level; } @@ -683,28 +828,22 @@ static void lockres_clear_flags(struct ocfs2_lock_res *lockres, static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres) { - mlog_entry_void(); - BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY)); BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED)); BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); - BUG_ON(lockres->l_blocking <= LKM_NLMODE); + BUG_ON(lockres->l_blocking <= DLM_LOCK_NL); lockres->l_level = lockres->l_requested; if (lockres->l_level <= ocfs2_highest_compat_lock_level(lockres->l_blocking)) { - lockres->l_blocking = LKM_NLMODE; + lockres->l_blocking = DLM_LOCK_NL; lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED); } lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); - - mlog_exit_void(); } static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres) { - mlog_entry_void(); - BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY)); BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED)); @@ -712,24 +851,28 @@ static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lo * information is already up to data. Convert from NL to * *anything* however should mark ourselves as needing an * update */ - if (lockres->l_level == LKM_NLMODE && + if (lockres->l_level == DLM_LOCK_NL && lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH) lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); lockres->l_level = lockres->l_requested; - lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); - mlog_exit_void(); + /* + * We set the OCFS2_LOCK_UPCONVERT_FINISHING flag before clearing + * the OCFS2_LOCK_BUSY flag to prevent the dc thread from + * downconverting the lock before the upconvert has fully completed. + */ + lockres_or_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING); + + lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); } static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres) { - mlog_entry_void(); - BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY))); BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED); - if (lockres->l_requested > LKM_NLMODE && + if (lockres->l_requested > DLM_LOCK_NL && !(lockres->l_flags & OCFS2_LOCK_LOCAL) && lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH) lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH); @@ -737,20 +880,15 @@ static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *loc lockres->l_level = lockres->l_requested; lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED); lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); - - mlog_exit_void(); } static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level) { int needs_downconvert = 0; - mlog_entry_void(); assert_spin_locked(&lockres->l_lock); - lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED); - if (level > lockres->l_blocking) { /* only schedule a downconvert if we haven't already scheduled * one that goes low enough to satisfy the level we're @@ -763,21 +901,133 @@ static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, lockres->l_blocking = level; } - mlog_exit(needs_downconvert); + mlog(ML_BASTS, "lockres %s, block %d, level %d, l_block %d, dwn %d\n", + lockres->l_name, level, lockres->l_level, lockres->l_blocking, + needs_downconvert); + + if (needs_downconvert) + lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED); + mlog(0, "needs_downconvert = %d\n", needs_downconvert); return needs_downconvert; } -static void ocfs2_blocking_ast(void *opaque, int level) +/* + * OCFS2_LOCK_PENDING and l_pending_gen. + * + * Why does OCFS2_LOCK_PENDING exist? To close a race between setting + * OCFS2_LOCK_BUSY and calling ocfs2_dlm_lock(). See ocfs2_unblock_lock() + * for more details on the race. + * + * OCFS2_LOCK_PENDING closes the race quite nicely. However, it introduces + * a race on itself. In o2dlm, we can get the ast before ocfs2_dlm_lock() + * returns. The ast clears OCFS2_LOCK_BUSY, and must therefore clear + * OCFS2_LOCK_PENDING at the same time. When ocfs2_dlm_lock() returns, + * the caller is going to try to clear PENDING again. If nothing else is + * happening, __lockres_clear_pending() sees PENDING is unset and does + * nothing. + * + * But what if another path (eg downconvert thread) has just started a + * new locking action? The other path has re-set PENDING. Our path + * cannot clear PENDING, because that will re-open the original race + * window. + * + * [Example] + * + * ocfs2_meta_lock() + * ocfs2_cluster_lock() + * set BUSY + * set PENDING + * drop l_lock + * ocfs2_dlm_lock() + * ocfs2_locking_ast() ocfs2_downconvert_thread() + * clear PENDING ocfs2_unblock_lock() + * take_l_lock + * !BUSY + * ocfs2_prepare_downconvert() + * set BUSY + * set PENDING + * drop l_lock + * take l_lock + * clear PENDING + * drop l_lock + * <window> + * ocfs2_dlm_lock() + * + * So as you can see, we now have a window where l_lock is not held, + * PENDING is not set, and ocfs2_dlm_lock() has not been called. + * + * The core problem is that ocfs2_cluster_lock() has cleared the PENDING + * set by ocfs2_prepare_downconvert(). That wasn't nice. + * + * To solve this we introduce l_pending_gen. A call to + * lockres_clear_pending() will only do so when it is passed a generation + * number that matches the lockres. lockres_set_pending() will return the + * current generation number. When ocfs2_cluster_lock() goes to clear + * PENDING, it passes the generation it got from set_pending(). In our + * example above, the generation numbers will *not* match. Thus, + * ocfs2_cluster_lock() will not clear the PENDING set by + * ocfs2_prepare_downconvert(). + */ + +/* Unlocked version for ocfs2_locking_ast() */ +static void __lockres_clear_pending(struct ocfs2_lock_res *lockres, + unsigned int generation, + struct ocfs2_super *osb) { - struct ocfs2_lock_res *lockres = opaque; + assert_spin_locked(&lockres->l_lock); + + /* + * The ast and locking functions can race us here. The winner + * will clear pending, the loser will not. + */ + if (!(lockres->l_flags & OCFS2_LOCK_PENDING) || + (lockres->l_pending_gen != generation)) + return; + + lockres_clear_flags(lockres, OCFS2_LOCK_PENDING); + lockres->l_pending_gen++; + + /* + * The downconvert thread may have skipped us because we + * were PENDING. Wake it up. + */ + if (lockres->l_flags & OCFS2_LOCK_BLOCKED) + ocfs2_wake_downconvert_thread(osb); +} + +/* Locked version for callers of ocfs2_dlm_lock() */ +static void lockres_clear_pending(struct ocfs2_lock_res *lockres, + unsigned int generation, + struct ocfs2_super *osb) +{ + unsigned long flags; + + spin_lock_irqsave(&lockres->l_lock, flags); + __lockres_clear_pending(lockres, generation, osb); + spin_unlock_irqrestore(&lockres->l_lock, flags); +} + +static unsigned int lockres_set_pending(struct ocfs2_lock_res *lockres) +{ + assert_spin_locked(&lockres->l_lock); + BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY)); + + lockres_or_flags(lockres, OCFS2_LOCK_PENDING); + + return lockres->l_pending_gen; +} + +static void ocfs2_blocking_ast(struct ocfs2_dlm_lksb *lksb, int level) +{ + struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb); struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres); int needs_downconvert; unsigned long flags; - BUG_ON(level <= LKM_NLMODE); + BUG_ON(level <= DLM_LOCK_NL); - mlog(0, "BAST fired for lockres %s, blocking %d, level %d type %s\n", - lockres->l_name, level, lockres->l_level, + mlog(ML_BASTS, "BAST fired for lockres %s, blocking %d, level %d, " + "type %s\n", lockres->l_name, level, lockres->l_level, ocfs2_lock_type_string(lockres->l_type)); /* @@ -798,21 +1048,33 @@ static void ocfs2_blocking_ast(void *opaque, int level) ocfs2_wake_downconvert_thread(osb); } -static void ocfs2_locking_ast(void *opaque) +static void ocfs2_locking_ast(struct ocfs2_dlm_lksb *lksb) { - struct ocfs2_lock_res *lockres = opaque; - struct dlm_lockstatus *lksb = &lockres->l_lksb; + struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb); + struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres); unsigned long flags; + int status; spin_lock_irqsave(&lockres->l_lock, flags); - if (lksb->status != DLM_NORMAL) { - mlog(ML_ERROR, "lockres %s: lksb status value of %u!\n", - lockres->l_name, lksb->status); + status = ocfs2_dlm_lock_status(&lockres->l_lksb); + + if (status == -EAGAIN) { + lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); + goto out; + } + + if (status) { + mlog(ML_ERROR, "lockres %s: lksb status value of %d!\n", + lockres->l_name, status); spin_unlock_irqrestore(&lockres->l_lock, flags); return; } + mlog(ML_BASTS, "AST fired for lockres %s, action %d, unlock %d, " + "level %d => %d\n", lockres->l_name, lockres->l_action, + lockres->l_unlock_action, lockres->l_level, lockres->l_requested); + switch(lockres->l_action) { case OCFS2_AST_ATTACH: ocfs2_generic_handle_attach_action(lockres); @@ -825,29 +1087,118 @@ static void ocfs2_locking_ast(void *opaque) ocfs2_generic_handle_downconvert_action(lockres); break; default: - mlog(ML_ERROR, "lockres %s: ast fired with invalid action: %u " - "lockres flags = 0x%lx, unlock action: %u\n", + mlog(ML_ERROR, "lockres %s: AST fired with invalid action: %u, " + "flags 0x%lx, unlock: %u\n", lockres->l_name, lockres->l_action, lockres->l_flags, lockres->l_unlock_action); BUG(); } - +out: /* set it to something invalid so if we get called again we * can catch it. */ lockres->l_action = OCFS2_AST_INVALID; + /* Did we try to cancel this lock? Clear that state */ + if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) + lockres->l_unlock_action = OCFS2_UNLOCK_INVALID; + + /* + * We may have beaten the locking functions here. We certainly + * know that dlm_lock() has been called :-) + * Because we can't have two lock calls in flight at once, we + * can use lockres->l_pending_gen. + */ + __lockres_clear_pending(lockres, lockres->l_pending_gen, osb); + + wake_up(&lockres->l_event); + spin_unlock_irqrestore(&lockres->l_lock, flags); +} + +static void ocfs2_unlock_ast(struct ocfs2_dlm_lksb *lksb, int error) +{ + struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb); + unsigned long flags; + + mlog(ML_BASTS, "UNLOCK AST fired for lockres %s, action = %d\n", + lockres->l_name, lockres->l_unlock_action); + + spin_lock_irqsave(&lockres->l_lock, flags); + if (error) { + mlog(ML_ERROR, "Dlm passes error %d for lock %s, " + "unlock_action %d\n", error, lockres->l_name, + lockres->l_unlock_action); + spin_unlock_irqrestore(&lockres->l_lock, flags); + return; + } + + switch(lockres->l_unlock_action) { + case OCFS2_UNLOCK_CANCEL_CONVERT: + mlog(0, "Cancel convert success for %s\n", lockres->l_name); + lockres->l_action = OCFS2_AST_INVALID; + /* Downconvert thread may have requeued this lock, we + * need to wake it. */ + if (lockres->l_flags & OCFS2_LOCK_BLOCKED) + ocfs2_wake_downconvert_thread(ocfs2_get_lockres_osb(lockres)); + break; + case OCFS2_UNLOCK_DROP_LOCK: + lockres->l_level = DLM_LOCK_IV; + break; + default: + BUG(); + } + + lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); + lockres->l_unlock_action = OCFS2_UNLOCK_INVALID; wake_up(&lockres->l_event); spin_unlock_irqrestore(&lockres->l_lock, flags); } +/* + * This is the filesystem locking protocol. It provides the lock handling + * hooks for the underlying DLM. It has a maximum version number. + * The version number allows interoperability with systems running at + * the same major number and an equal or smaller minor number. + * + * Whenever the filesystem does new things with locks (adds or removes a + * lock, orders them differently, does different things underneath a lock), + * the version must be changed. The protocol is negotiated when joining + * the dlm domain. A node may join the domain if its major version is + * identical to all other nodes and its minor version is greater than + * or equal to all other nodes. When its minor version is greater than + * the other nodes, it will run at the minor version specified by the + * other nodes. + * + * If a locking change is made that will not be compatible with older + * versions, the major number must be increased and the minor version set + * to zero. If a change merely adds a behavior that can be disabled when + * speaking to older versions, the minor version must be increased. If a + * change adds a fully backwards compatible change (eg, LVB changes that + * are just ignored by older versions), the version does not need to be + * updated. + */ +static struct ocfs2_locking_protocol lproto = { + .lp_max_version = { + .pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR, + .pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR, + }, + .lp_lock_ast = ocfs2_locking_ast, + .lp_blocking_ast = ocfs2_blocking_ast, + .lp_unlock_ast = ocfs2_unlock_ast, +}; + +void ocfs2_set_locking_protocol(void) +{ + ocfs2_stack_glue_set_max_proto_version(&lproto.lp_max_version); +} + static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres, int convert) { unsigned long flags; - mlog_entry_void(); spin_lock_irqsave(&lockres->l_lock, flags); lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); + lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING); if (convert) lockres->l_action = OCFS2_AST_INVALID; else @@ -855,7 +1206,6 @@ static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres, spin_unlock_irqrestore(&lockres->l_lock, flags); wake_up(&lockres->l_event); - mlog_exit_void(); } /* Note: If we detect another process working on the lock (i.e., @@ -865,15 +1215,13 @@ static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres, static int ocfs2_lock_create(struct ocfs2_super *osb, struct ocfs2_lock_res *lockres, int level, - int dlm_flags) + u32 dlm_flags) { int ret = 0; - enum dlm_status status = DLM_NORMAL; unsigned long flags; + unsigned int gen; - mlog_entry_void(); - - mlog(0, "lock %s, level = %d, flags = %d\n", lockres->l_name, level, + mlog(0, "lock %s, level = %d, flags = %u\n", lockres->l_name, level, dlm_flags); spin_lock_irqsave(&lockres->l_lock, flags); @@ -886,27 +1234,24 @@ static int ocfs2_lock_create(struct ocfs2_super *osb, lockres->l_action = OCFS2_AST_ATTACH; lockres->l_requested = level; lockres_or_flags(lockres, OCFS2_LOCK_BUSY); + gen = lockres_set_pending(lockres); spin_unlock_irqrestore(&lockres->l_lock, flags); - status = dlmlock(osb->dlm, - level, - &lockres->l_lksb, - dlm_flags, - lockres->l_name, - OCFS2_LOCK_ID_MAX_LEN - 1, - ocfs2_locking_ast, - lockres, - ocfs2_blocking_ast); - if (status != DLM_NORMAL) { - ocfs2_log_dlm_error("dlmlock", status, lockres); - ret = -EINVAL; + ret = ocfs2_dlm_lock(osb->cconn, + level, + &lockres->l_lksb, + dlm_flags, + lockres->l_name, + OCFS2_LOCK_ID_MAX_LEN - 1); + lockres_clear_pending(lockres, gen, osb); + if (ret) { + ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres); ocfs2_recover_from_dlm_error(lockres, 1); } - mlog(0, "lock %s, successfull return from dlmlock\n", lockres->l_name); + mlog(0, "lock %s, return from ocfs2_dlm_lock\n", lockres->l_name); bail: - mlog_exit(ret); return ret; } @@ -952,13 +1297,14 @@ static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw) { INIT_LIST_HEAD(&mw->mw_item); init_completion(&mw->mw_complete); + ocfs2_init_start_time(mw); } static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw) { wait_for_completion(&mw->mw_complete); /* Re-arm the completion in case we want to wait on it again */ - INIT_COMPLETION(mw->mw_complete); + reinit_completion(&mw->mw_complete); return mw->mw_status; } @@ -1009,39 +1355,40 @@ static int ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter *mw, else ret = mw->mw_status; /* Re-arm the completion in case we want to wait on it again */ - INIT_COMPLETION(mw->mw_complete); + reinit_completion(&mw->mw_complete); return ret; } -static int ocfs2_cluster_lock(struct ocfs2_super *osb, - struct ocfs2_lock_res *lockres, - int level, - int lkm_flags, - int arg_flags) +static int __ocfs2_cluster_lock(struct ocfs2_super *osb, + struct ocfs2_lock_res *lockres, + int level, + u32 lkm_flags, + int arg_flags, + int l_subclass, + unsigned long caller_ip) { struct ocfs2_mask_waiter mw; - enum dlm_status status; int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR); int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */ unsigned long flags; - - mlog_entry_void(); + unsigned int gen; + int noqueue_attempted = 0; ocfs2_init_mask_waiter(&mw); if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) - lkm_flags |= LKM_VALBLK; + lkm_flags |= DLM_LKF_VALBLK; again: wait = 0; + spin_lock_irqsave(&lockres->l_lock, flags); + if (catch_signals && signal_pending(current)) { ret = -ERESTARTSYS; - goto out; + goto unlock; } - spin_lock_irqsave(&lockres->l_lock, flags); - mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING, "Cluster lock called on freeing lockres %s! flags " "0x%lx\n", lockres->l_name, lockres->l_flags); @@ -1058,6 +1405,25 @@ again: goto unlock; } + if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING) { + /* + * We've upconverted. If the lock now has a level we can + * work with, we take it. If, however, the lock is not at the + * required level, we go thru the full cycle. One way this could + * happen is if a process requesting an upconvert to PR is + * closely followed by another requesting upconvert to an EX. + * If the process requesting EX lands here, we want it to + * continue attempting to upconvert and let the process + * requesting PR take the lock. + * If multiple processes request upconvert to PR, the first one + * here will take the lock. The others will have to go thru the + * OCFS2_LOCK_BLOCKED check to ensure that there is no pending + * downconvert request. + */ + if (level <= lockres->l_level) + goto update_holders; + } + if (lockres->l_flags & OCFS2_LOCK_BLOCKED && !ocfs2_may_continue_on_blocked_lock(lockres, level)) { /* is the lock is currently blocked on behalf of @@ -1068,52 +1434,55 @@ again: } if (level > lockres->l_level) { + if (noqueue_attempted > 0) { + ret = -EAGAIN; + goto unlock; + } + if (lkm_flags & DLM_LKF_NOQUEUE) + noqueue_attempted = 1; + if (lockres->l_action != OCFS2_AST_INVALID) mlog(ML_ERROR, "lockres %s has action %u pending\n", lockres->l_name, lockres->l_action); if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) { lockres->l_action = OCFS2_AST_ATTACH; - lkm_flags &= ~LKM_CONVERT; + lkm_flags &= ~DLM_LKF_CONVERT; } else { lockres->l_action = OCFS2_AST_CONVERT; - lkm_flags |= LKM_CONVERT; + lkm_flags |= DLM_LKF_CONVERT; } lockres->l_requested = level; lockres_or_flags(lockres, OCFS2_LOCK_BUSY); + gen = lockres_set_pending(lockres); spin_unlock_irqrestore(&lockres->l_lock, flags); - BUG_ON(level == LKM_IVMODE); - BUG_ON(level == LKM_NLMODE); + BUG_ON(level == DLM_LOCK_IV); + BUG_ON(level == DLM_LOCK_NL); - mlog(0, "lock %s, convert from %d to level = %d\n", + mlog(ML_BASTS, "lockres %s, convert from %d to %d\n", lockres->l_name, lockres->l_level, level); /* call dlm_lock to upgrade lock now */ - status = dlmlock(osb->dlm, - level, - &lockres->l_lksb, - lkm_flags, - lockres->l_name, - OCFS2_LOCK_ID_MAX_LEN - 1, - ocfs2_locking_ast, - lockres, - ocfs2_blocking_ast); - if (status != DLM_NORMAL) { - if ((lkm_flags & LKM_NOQUEUE) && - (status == DLM_NOTQUEUED)) - ret = -EAGAIN; - else { - ocfs2_log_dlm_error("dlmlock", status, - lockres); - ret = -EINVAL; + ret = ocfs2_dlm_lock(osb->cconn, + level, + &lockres->l_lksb, + lkm_flags, + lockres->l_name, + OCFS2_LOCK_ID_MAX_LEN - 1); + lockres_clear_pending(lockres, gen, osb); + if (ret) { + if (!(lkm_flags & DLM_LKF_NOQUEUE) || + (ret != -EAGAIN)) { + ocfs2_log_dlm_error("ocfs2_dlm_lock", + ret, lockres); } ocfs2_recover_from_dlm_error(lockres, 1); goto out; } - mlog(0, "lock %s, successfull return from dlmlock\n", + mlog(0, "lock %s, successful return from ocfs2_dlm_lock\n", lockres->l_name); /* At this point we've gone inside the dlm and need to @@ -1124,11 +1493,14 @@ again: goto again; } +update_holders: /* Ok, if we get here then we're good to go. */ ocfs2_inc_holders(lockres, level); ret = 0; unlock: + lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING); + spin_unlock_irqrestore(&lockres->l_lock, flags); out: /* @@ -1153,23 +1525,49 @@ out: goto again; mlog_errno(ret); } + ocfs2_update_lock_stats(lockres, level, &mw, ret); - mlog_exit(ret); +#ifdef CONFIG_DEBUG_LOCK_ALLOC + if (!ret && lockres->l_lockdep_map.key != NULL) { + if (level == DLM_LOCK_PR) + rwsem_acquire_read(&lockres->l_lockdep_map, l_subclass, + !!(arg_flags & OCFS2_META_LOCK_NOQUEUE), + caller_ip); + else + rwsem_acquire(&lockres->l_lockdep_map, l_subclass, + !!(arg_flags & OCFS2_META_LOCK_NOQUEUE), + caller_ip); + } +#endif return ret; } -static void ocfs2_cluster_unlock(struct ocfs2_super *osb, - struct ocfs2_lock_res *lockres, - int level) +static inline int ocfs2_cluster_lock(struct ocfs2_super *osb, + struct ocfs2_lock_res *lockres, + int level, + u32 lkm_flags, + int arg_flags) +{ + return __ocfs2_cluster_lock(osb, lockres, level, lkm_flags, arg_flags, + 0, _RET_IP_); +} + + +static void __ocfs2_cluster_unlock(struct ocfs2_super *osb, + struct ocfs2_lock_res *lockres, + int level, + unsigned long caller_ip) { unsigned long flags; - mlog_entry_void(); spin_lock_irqsave(&lockres->l_lock, flags); ocfs2_dec_holders(lockres, level); ocfs2_downconvert_on_unlock(osb, lockres); spin_unlock_irqrestore(&lockres->l_lock, flags); - mlog_exit_void(); +#ifdef CONFIG_DEBUG_LOCK_ALLOC + if (lockres->l_lockdep_map.key != NULL) + rwsem_release(&lockres->l_lockdep_map, 1, caller_ip); +#endif } static int ocfs2_create_new_lock(struct ocfs2_super *osb, @@ -1177,9 +1575,9 @@ static int ocfs2_create_new_lock(struct ocfs2_super *osb, int ex, int local) { - int level = ex ? LKM_EXMODE : LKM_PRMODE; + int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; unsigned long flags; - int lkm_flags = local ? LKM_LOCAL : 0; + u32 lkm_flags = local ? DLM_LKF_LOCAL : 0; spin_lock_irqsave(&lockres->l_lock, flags); BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED); @@ -1203,8 +1601,6 @@ int ocfs2_create_new_inode_locks(struct inode *inode) BUG_ON(!inode); BUG_ON(!ocfs2_inode_is_new(inode)); - mlog_entry_void(); - mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno); /* NOTE: That we don't increment any of the holder counts, nor @@ -1222,7 +1618,7 @@ int ocfs2_create_new_inode_locks(struct inode *inode) } /* - * We don't want to use LKM_LOCAL on a meta data lock as they + * We don't want to use DLM_LKF_LOCAL on a meta data lock as they * don't use a generation in their lock names. */ ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_inode_lockres, 1, 0); @@ -1238,7 +1634,6 @@ int ocfs2_create_new_inode_locks(struct inode *inode) } bail: - mlog_exit(ret); return ret; } @@ -1250,8 +1645,6 @@ int ocfs2_rw_lock(struct inode *inode, int write) BUG_ON(!inode); - mlog_entry_void(); - mlog(0, "inode %llu take %s RW lock\n", (unsigned long long)OCFS2_I(inode)->ip_blkno, write ? "EXMODE" : "PRMODE"); @@ -1261,33 +1654,28 @@ int ocfs2_rw_lock(struct inode *inode, int write) lockres = &OCFS2_I(inode)->ip_rw_lockres; - level = write ? LKM_EXMODE : LKM_PRMODE; + level = write ? DLM_LOCK_EX : DLM_LOCK_PR; status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0, 0); if (status < 0) mlog_errno(status); - mlog_exit(status); return status; } void ocfs2_rw_unlock(struct inode *inode, int write) { - int level = write ? LKM_EXMODE : LKM_PRMODE; + int level = write ? DLM_LOCK_EX : DLM_LOCK_PR; struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres; struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); - mlog_entry_void(); - mlog(0, "inode %llu drop %s RW lock\n", (unsigned long long)OCFS2_I(inode)->ip_blkno, write ? "EXMODE" : "PRMODE"); if (!ocfs2_mount_local(osb)) ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level); - - mlog_exit_void(); } /* @@ -1301,23 +1689,20 @@ int ocfs2_open_lock(struct inode *inode) BUG_ON(!inode); - mlog_entry_void(); - mlog(0, "inode %llu take PRMODE open lock\n", (unsigned long long)OCFS2_I(inode)->ip_blkno); - if (ocfs2_mount_local(osb)) + if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb)) goto out; lockres = &OCFS2_I(inode)->ip_open_lockres; status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, - LKM_PRMODE, 0, 0); + DLM_LOCK_PR, 0, 0); if (status < 0) mlog_errno(status); out: - mlog_exit(status); return status; } @@ -1329,30 +1714,33 @@ int ocfs2_try_open_lock(struct inode *inode, int write) BUG_ON(!inode); - mlog_entry_void(); - mlog(0, "inode %llu try to take %s open lock\n", (unsigned long long)OCFS2_I(inode)->ip_blkno, write ? "EXMODE" : "PRMODE"); + if (ocfs2_is_hard_readonly(osb)) { + if (write) + status = -EROFS; + goto out; + } + if (ocfs2_mount_local(osb)) goto out; lockres = &OCFS2_I(inode)->ip_open_lockres; - level = write ? LKM_EXMODE : LKM_PRMODE; + level = write ? DLM_LOCK_EX : DLM_LOCK_PR; /* * The file system may already holding a PRMODE/EXMODE open lock. - * Since we pass LKM_NOQUEUE, the request won't block waiting on + * Since we pass DLM_LKF_NOQUEUE, the request won't block waiting on * other nodes and the -EAGAIN will indicate to the caller that * this inode is still in use. */ status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, - level, LKM_NOQUEUE, 0); + level, DLM_LKF_NOQUEUE, 0); out: - mlog_exit(status); return status; } @@ -1364,8 +1752,6 @@ void ocfs2_open_unlock(struct inode *inode) struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres; struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); - mlog_entry_void(); - mlog(0, "inode %llu drop open lock\n", (unsigned long long)OCFS2_I(inode)->ip_blkno); @@ -1374,13 +1760,13 @@ void ocfs2_open_unlock(struct inode *inode) if(lockres->l_ro_holders) ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, - LKM_PRMODE); + DLM_LOCK_PR); if(lockres->l_ex_holders) ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, - LKM_EXMODE); + DLM_LOCK_EX); out: - mlog_exit_void(); + return; } static int ocfs2_flock_handle_signal(struct ocfs2_lock_res *lockres, @@ -1434,7 +1820,7 @@ out: * ocfs2_file_lock() and ocfs2_file_unlock() map to a single pair of * flock() calls. The locking approach this requires is sufficiently * different from all other cluster lock types that we implement a - * seperate path to the "low-level" dlm calls. In particular: + * separate path to the "low-level" dlm calls. In particular: * * - No optimization of lock levels is done - we take at exactly * what's been requested. @@ -1453,8 +1839,8 @@ out: */ int ocfs2_file_lock(struct file *file, int ex, int trylock) { - int ret, level = ex ? LKM_EXMODE : LKM_PRMODE; - unsigned int lkm_flags = trylock ? LKM_NOQUEUE : 0; + int ret, level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; + unsigned int lkm_flags = trylock ? DLM_LKF_NOQUEUE : 0; unsigned long flags; struct ocfs2_file_private *fp = file->private_data; struct ocfs2_lock_res *lockres = &fp->fp_flock; @@ -1464,7 +1850,7 @@ int ocfs2_file_lock(struct file *file, int ex, int trylock) ocfs2_init_mask_waiter(&mw); if ((lockres->l_flags & OCFS2_LOCK_BUSY) || - (lockres->l_level > LKM_NLMODE)) { + (lockres->l_level > DLM_LOCK_NL)) { mlog(ML_ERROR, "File lock \"%s\" has busy or locked state: flags: 0x%lx, " "level: %u\n", lockres->l_name, lockres->l_flags, @@ -1481,7 +1867,7 @@ int ocfs2_file_lock(struct file *file, int ex, int trylock) * Get the lock at NLMODE to start - that way we * can cancel the upconvert request if need be. */ - ret = ocfs2_lock_create(osb, lockres, LKM_NLMODE, 0); + ret = ocfs2_lock_create(osb, lockres, DLM_LOCK_NL, 0); if (ret < 0) { mlog_errno(ret); goto out; @@ -1496,21 +1882,18 @@ int ocfs2_file_lock(struct file *file, int ex, int trylock) } lockres->l_action = OCFS2_AST_CONVERT; - lkm_flags |= LKM_CONVERT; + lkm_flags |= DLM_LKF_CONVERT; lockres->l_requested = level; lockres_or_flags(lockres, OCFS2_LOCK_BUSY); lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0); spin_unlock_irqrestore(&lockres->l_lock, flags); - ret = dlmlock(osb->dlm, level, &lockres->l_lksb, lkm_flags, - lockres->l_name, OCFS2_LOCK_ID_MAX_LEN - 1, - ocfs2_locking_ast, lockres, ocfs2_blocking_ast); - if (ret != DLM_NORMAL) { - if (trylock && ret == DLM_NOTQUEUED) - ret = -EAGAIN; - else { - ocfs2_log_dlm_error("dlmlock", ret, lockres); + ret = ocfs2_dlm_lock(osb->cconn, level, &lockres->l_lksb, lkm_flags, + lockres->l_name, OCFS2_LOCK_ID_MAX_LEN - 1); + if (ret) { + if (!trylock || (ret != -EAGAIN)) { + ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres); ret = -EINVAL; } @@ -1534,9 +1917,13 @@ int ocfs2_file_lock(struct file *file, int ex, int trylock) * outstanding lock request, so a cancel convert is * required. We intentionally overwrite 'ret' - if the * cancel fails and the lock was granted, it's easier - * to just bubble sucess back up to the user. + * to just bubble success back up to the user. */ ret = ocfs2_flock_handle_signal(lockres, level); + } else if (!ret && (level > lockres->l_level)) { + /* Trylock failed asynchronously */ + BUG_ON(!trylock); + ret = -EAGAIN; } out: @@ -1549,6 +1936,7 @@ out: void ocfs2_file_unlock(struct file *file) { int ret; + unsigned int gen; unsigned long flags; struct ocfs2_file_private *fp = file->private_data; struct ocfs2_lock_res *lockres = &fp->fp_flock; @@ -1560,7 +1948,7 @@ void ocfs2_file_unlock(struct file *file) if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) return; - if (lockres->l_level == LKM_NLMODE) + if (lockres->l_level == DLM_LOCK_NL) return; mlog(0, "Unlock: \"%s\" flags: 0x%lx, level: %d, act: %d\n", @@ -1572,13 +1960,13 @@ void ocfs2_file_unlock(struct file *file) * Fake a blocking ast for the downconvert code. */ lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED); - lockres->l_blocking = LKM_EXMODE; + lockres->l_blocking = DLM_LOCK_EX; - ocfs2_prepare_downconvert(lockres, LKM_NLMODE); + gen = ocfs2_prepare_downconvert(lockres, DLM_LOCK_NL); lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0); spin_unlock_irqrestore(&lockres->l_lock, flags); - ret = ocfs2_downconvert_lock(osb, lockres, LKM_NLMODE, 0); + ret = ocfs2_downconvert_lock(osb, lockres, DLM_LOCK_NL, 0, gen); if (ret) { mlog_errno(ret); return; @@ -1594,18 +1982,16 @@ static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb, { int kick = 0; - mlog_entry_void(); - /* If we know that another node is waiting on our lock, kick * the downconvert thread * pre-emptively when we reach a release * condition. */ if (lockres->l_flags & OCFS2_LOCK_BLOCKED) { switch(lockres->l_blocking) { - case LKM_EXMODE: + case DLM_LOCK_EX: if (!lockres->l_ex_holders && !lockres->l_ro_holders) kick = 1; break; - case LKM_PRMODE: + case DLM_LOCK_PR: if (!lockres->l_ex_holders) kick = 1; break; @@ -1616,8 +2002,6 @@ static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb, if (kick) ocfs2_wake_downconvert_thread(osb); - - mlog_exit_void(); } #define OCFS2_SEC_BITS 34 @@ -1646,9 +2030,7 @@ static void __ocfs2_stuff_meta_lvb(struct inode *inode) struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres; struct ocfs2_meta_lvb *lvb; - mlog_entry_void(); - - lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb; + lvb = ocfs2_dlm_lvb(&lockres->l_lksb); /* * Invalidate the LVB of a deleted inode - this way other @@ -1663,8 +2045,8 @@ static void __ocfs2_stuff_meta_lvb(struct inode *inode) lvb->lvb_version = OCFS2_LVB_VERSION; lvb->lvb_isize = cpu_to_be64(i_size_read(inode)); lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters); - lvb->lvb_iuid = cpu_to_be32(inode->i_uid); - lvb->lvb_igid = cpu_to_be32(inode->i_gid); + lvb->lvb_iuid = cpu_to_be32(i_uid_read(inode)); + lvb->lvb_igid = cpu_to_be32(i_gid_read(inode)); lvb->lvb_imode = cpu_to_be16(inode->i_mode); lvb->lvb_inlink = cpu_to_be16(inode->i_nlink); lvb->lvb_iatime_packed = @@ -1679,8 +2061,6 @@ static void __ocfs2_stuff_meta_lvb(struct inode *inode) out: mlog_meta_lvb(0, lockres); - - mlog_exit_void(); } static void ocfs2_unpack_timespec(struct timespec *spec, @@ -1696,11 +2076,9 @@ static void ocfs2_refresh_inode_from_lvb(struct inode *inode) struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres; struct ocfs2_meta_lvb *lvb; - mlog_entry_void(); - mlog_meta_lvb(0, lockres); - lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb; + lvb = ocfs2_dlm_lvb(&lockres->l_lksb); /* We're safe here without the lockres lock... */ spin_lock(&oi->ip_lock); @@ -1717,10 +2095,10 @@ static void ocfs2_refresh_inode_from_lvb(struct inode *inode) else inode->i_blocks = ocfs2_inode_sector_count(inode); - inode->i_uid = be32_to_cpu(lvb->lvb_iuid); - inode->i_gid = be32_to_cpu(lvb->lvb_igid); + i_uid_write(inode, be32_to_cpu(lvb->lvb_iuid)); + i_gid_write(inode, be32_to_cpu(lvb->lvb_igid)); inode->i_mode = be16_to_cpu(lvb->lvb_imode); - inode->i_nlink = be16_to_cpu(lvb->lvb_inlink); + set_nlink(inode, be16_to_cpu(lvb->lvb_inlink)); ocfs2_unpack_timespec(&inode->i_atime, be64_to_cpu(lvb->lvb_iatime_packed)); ocfs2_unpack_timespec(&inode->i_mtime, @@ -1728,16 +2106,15 @@ static void ocfs2_refresh_inode_from_lvb(struct inode *inode) ocfs2_unpack_timespec(&inode->i_ctime, be64_to_cpu(lvb->lvb_ictime_packed)); spin_unlock(&oi->ip_lock); - - mlog_exit_void(); } static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode, struct ocfs2_lock_res *lockres) { - struct ocfs2_meta_lvb *lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb; + struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb); - if (lvb->lvb_version == OCFS2_LVB_VERSION + if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) + && lvb->lvb_version == OCFS2_LVB_VERSION && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation) return 1; return 0; @@ -1755,8 +2132,6 @@ static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres) unsigned long flags; int status = 0; - mlog_entry_void(); - refresh_check: spin_lock_irqsave(&lockres->l_lock, flags); if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) { @@ -1777,7 +2152,7 @@ refresh_check: status = 1; bail: - mlog_exit(status); + mlog(0, "status %d\n", status); return status; } @@ -1787,7 +2162,6 @@ static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockre int status) { unsigned long flags; - mlog_entry_void(); spin_lock_irqsave(&lockres->l_lock, flags); lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING); @@ -1796,8 +2170,6 @@ static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockre spin_unlock_irqrestore(&lockres->l_lock, flags); wake_up(&lockres->l_event); - - mlog_exit_void(); } /* may or may not return a bh if it went to disk. */ @@ -1810,8 +2182,6 @@ static int ocfs2_inode_lock_update(struct inode *inode, struct ocfs2_dinode *fe; struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); - mlog_entry_void(); - if (ocfs2_mount_local(osb)) goto bail; @@ -1831,7 +2201,7 @@ static int ocfs2_inode_lock_update(struct inode *inode, /* This will discard any caching information we might have had * for the inode metadata. */ - ocfs2_metadata_cache_purge(inode); + ocfs2_metadata_cache_purge(INODE_CACHE(inode)); ocfs2_extent_map_trunc(inode, 0); @@ -1842,8 +2212,7 @@ static int ocfs2_inode_lock_update(struct inode *inode, } else { /* Boo, we have to go to disk. */ /* read bh, cast, ocfs2_refresh_inode */ - status = ocfs2_read_block(OCFS2_SB(inode->i_sb), oi->ip_blkno, - bh, OCFS2_BH_CACHED, inode); + status = ocfs2_read_inode_block(inode, bh); if (status < 0) { mlog_errno(status); goto bail_refresh; @@ -1851,18 +2220,14 @@ static int ocfs2_inode_lock_update(struct inode *inode, fe = (struct ocfs2_dinode *) (*bh)->b_data; /* This is a good chance to make sure we're not - * locking an invalid object. + * locking an invalid object. ocfs2_read_inode_block() + * already checked that the inode block is sane. * * We bug on a stale inode here because we checked * above whether it was wiped from disk. The wiping * node provides a guarantee that we receive that * message and can mark the inode before dropping any * locks associated with it. */ - if (!OCFS2_IS_VALID_DINODE(fe)) { - OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); - status = -EIO; - goto bail_refresh; - } mlog_bug_on_msg(inode->i_generation != le32_to_cpu(fe->i_generation), "Invalid dinode %llu disk generation: %u " @@ -1878,13 +2243,13 @@ static int ocfs2_inode_lock_update(struct inode *inode, le32_to_cpu(fe->i_flags)); ocfs2_refresh_inode(inode, fe); + ocfs2_track_lock_refresh(lockres); } status = 0; bail_refresh: ocfs2_complete_lock_res_refresh(lockres, status); bail: - mlog_exit(status); return status; } @@ -1903,11 +2268,7 @@ static int ocfs2_assign_bh(struct inode *inode, return 0; } - status = ocfs2_read_block(OCFS2_SB(inode->i_sb), - OCFS2_I(inode)->ip_blkno, - ret_bh, - OCFS2_BH_CACHED, - inode); + status = ocfs2_read_inode_block(inode, ret_bh); if (status < 0) mlog_errno(status); @@ -1918,20 +2279,20 @@ static int ocfs2_assign_bh(struct inode *inode, * returns < 0 error if the callback will never be called, otherwise * the result of the lock will be communicated via the callback. */ -int ocfs2_inode_lock_full(struct inode *inode, - struct buffer_head **ret_bh, - int ex, - int arg_flags) +int ocfs2_inode_lock_full_nested(struct inode *inode, + struct buffer_head **ret_bh, + int ex, + int arg_flags, + int subclass) { - int status, level, dlm_flags, acquired; + int status, level, acquired; + u32 dlm_flags; struct ocfs2_lock_res *lockres = NULL; struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); struct buffer_head *local_bh = NULL; BUG_ON(!inode); - mlog_entry_void(); - mlog(0, "inode %llu, take %s META lock\n", (unsigned long long)OCFS2_I(inode)->ip_blkno, ex ? "EXMODE" : "PRMODE"); @@ -1943,25 +2304,25 @@ int ocfs2_inode_lock_full(struct inode *inode, if (ocfs2_is_hard_readonly(osb)) { if (ex) status = -EROFS; - goto bail; + goto getbh; } if (ocfs2_mount_local(osb)) goto local; if (!(arg_flags & OCFS2_META_LOCK_RECOVERY)) - wait_event(osb->recovery_event, - ocfs2_node_map_is_empty(osb, &osb->recovery_map)); + ocfs2_wait_for_recovery(osb); lockres = &OCFS2_I(inode)->ip_inode_lockres; - level = ex ? LKM_EXMODE : LKM_PRMODE; + level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; dlm_flags = 0; if (arg_flags & OCFS2_META_LOCK_NOQUEUE) - dlm_flags |= LKM_NOQUEUE; + dlm_flags |= DLM_LKF_NOQUEUE; - status = ocfs2_cluster_lock(osb, lockres, level, dlm_flags, arg_flags); + status = __ocfs2_cluster_lock(osb, lockres, level, dlm_flags, + arg_flags, subclass, _RET_IP_); if (status < 0) { - if (status != -EAGAIN && status != -EIOCBRETRY) + if (status != -EAGAIN) mlog_errno(status); goto bail; } @@ -1974,8 +2335,7 @@ int ocfs2_inode_lock_full(struct inode *inode, * committed to owning this lock so we don't allow signals to * abort the operation. */ if (!(arg_flags & OCFS2_META_LOCK_RECOVERY)) - wait_event(osb->recovery_event, - ocfs2_node_map_is_empty(osb, &osb->recovery_map)); + ocfs2_wait_for_recovery(osb); local: /* @@ -2002,7 +2362,7 @@ local: mlog_errno(status); goto bail; } - +getbh: if (ret_bh) { status = ocfs2_assign_bh(inode, ret_bh, local_bh); if (status < 0) { @@ -2024,7 +2384,6 @@ bail: if (local_bh) brelse(local_bh); - mlog_exit(status); return status; } @@ -2074,7 +2433,6 @@ int ocfs2_inode_lock_atime(struct inode *inode, { int ret; - mlog_entry_void(); ret = ocfs2_inode_lock(inode, NULL, 0); if (ret < 0) { mlog_errno(ret); @@ -2102,19 +2460,16 @@ int ocfs2_inode_lock_atime(struct inode *inode, } else *level = 0; - mlog_exit(ret); return ret; } void ocfs2_inode_unlock(struct inode *inode, int ex) { - int level = ex ? LKM_EXMODE : LKM_PRMODE; + int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_inode_lockres; struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); - mlog_entry_void(); - mlog(0, "inode %llu drop %s META lock\n", (unsigned long long)OCFS2_I(inode)->ip_blkno, ex ? "EXMODE" : "PRMODE"); @@ -2122,20 +2477,55 @@ void ocfs2_inode_unlock(struct inode *inode, if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)) && !ocfs2_mount_local(osb)) ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level); +} + +int ocfs2_orphan_scan_lock(struct ocfs2_super *osb, u32 *seqno) +{ + struct ocfs2_lock_res *lockres; + struct ocfs2_orphan_scan_lvb *lvb; + int status = 0; + + if (ocfs2_is_hard_readonly(osb)) + return -EROFS; + + if (ocfs2_mount_local(osb)) + return 0; + + lockres = &osb->osb_orphan_scan.os_lockres; + status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0); + if (status < 0) + return status; + + lvb = ocfs2_dlm_lvb(&lockres->l_lksb); + if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) && + lvb->lvb_version == OCFS2_ORPHAN_LVB_VERSION) + *seqno = be32_to_cpu(lvb->lvb_os_seqno); + else + *seqno = osb->osb_orphan_scan.os_seqno + 1; + + return status; +} - mlog_exit_void(); +void ocfs2_orphan_scan_unlock(struct ocfs2_super *osb, u32 seqno) +{ + struct ocfs2_lock_res *lockres; + struct ocfs2_orphan_scan_lvb *lvb; + + if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb)) { + lockres = &osb->osb_orphan_scan.os_lockres; + lvb = ocfs2_dlm_lvb(&lockres->l_lksb); + lvb->lvb_version = OCFS2_ORPHAN_LVB_VERSION; + lvb->lvb_os_seqno = cpu_to_be32(seqno); + ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX); + } } int ocfs2_super_lock(struct ocfs2_super *osb, int ex) { int status = 0; - int level = ex ? LKM_EXMODE : LKM_PRMODE; + int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; struct ocfs2_lock_res *lockres = &osb->osb_super_lockres; - struct buffer_head *bh; - struct ocfs2_slot_info *si = osb->slot_info; - - mlog_entry_void(); if (ocfs2_is_hard_readonly(osb)) return -EROFS; @@ -2154,31 +2544,25 @@ int ocfs2_super_lock(struct ocfs2_super *osb, * refreshed, so we do it here. Of course, making sense of * everything is up to the caller :) */ status = ocfs2_should_refresh_lock_res(lockres); - if (status < 0) { - mlog_errno(status); - goto bail; - } if (status) { - bh = si->si_bh; - status = ocfs2_read_block(osb, bh->b_blocknr, &bh, 0, - si->si_inode); - if (status == 0) - ocfs2_update_slot_info(si); + status = ocfs2_refresh_slot_info(osb); ocfs2_complete_lock_res_refresh(lockres, status); - if (status < 0) + if (status < 0) { + ocfs2_cluster_unlock(osb, lockres, level); mlog_errno(status); + } + ocfs2_track_lock_refresh(lockres); } bail: - mlog_exit(status); return status; } void ocfs2_super_unlock(struct ocfs2_super *osb, int ex) { - int level = ex ? LKM_EXMODE : LKM_PRMODE; + int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; struct ocfs2_lock_res *lockres = &osb->osb_super_lockres; if (!ocfs2_mount_local(osb)) @@ -2196,7 +2580,7 @@ int ocfs2_rename_lock(struct ocfs2_super *osb) if (ocfs2_mount_local(osb)) return 0; - status = ocfs2_cluster_lock(osb, lockres, LKM_EXMODE, 0, 0); + status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0); if (status < 0) mlog_errno(status); @@ -2208,20 +2592,51 @@ void ocfs2_rename_unlock(struct ocfs2_super *osb) struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres; if (!ocfs2_mount_local(osb)) - ocfs2_cluster_unlock(osb, lockres, LKM_EXMODE); + ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX); +} + +int ocfs2_nfs_sync_lock(struct ocfs2_super *osb, int ex) +{ + int status; + struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres; + + if (ocfs2_is_hard_readonly(osb)) + return -EROFS; + + if (ocfs2_mount_local(osb)) + return 0; + + status = ocfs2_cluster_lock(osb, lockres, ex ? LKM_EXMODE : LKM_PRMODE, + 0, 0); + if (status < 0) + mlog(ML_ERROR, "lock on nfs sync lock failed %d\n", status); + + return status; +} + +void ocfs2_nfs_sync_unlock(struct ocfs2_super *osb, int ex) +{ + struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres; + + if (!ocfs2_mount_local(osb)) + ocfs2_cluster_unlock(osb, lockres, + ex ? LKM_EXMODE : LKM_PRMODE); } int ocfs2_dentry_lock(struct dentry *dentry, int ex) { int ret; - int level = ex ? LKM_EXMODE : LKM_PRMODE; + int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; struct ocfs2_dentry_lock *dl = dentry->d_fsdata; struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb); BUG_ON(!dl); - if (ocfs2_is_hard_readonly(osb)) - return -EROFS; + if (ocfs2_is_hard_readonly(osb)) { + if (ex) + return -EROFS; + return 0; + } if (ocfs2_mount_local(osb)) return 0; @@ -2235,11 +2650,11 @@ int ocfs2_dentry_lock(struct dentry *dentry, int ex) void ocfs2_dentry_unlock(struct dentry *dentry, int ex) { - int level = ex ? LKM_EXMODE : LKM_PRMODE; + int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; struct ocfs2_dentry_lock *dl = dentry->d_fsdata; struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb); - if (!ocfs2_mount_local(osb)) + if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb)) ocfs2_cluster_unlock(osb, &dl->dl_lockres, level); } @@ -2362,8 +2777,15 @@ static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos) return iter; } -/* So that debugfs.ocfs2 can determine which format is being used */ -#define OCFS2_DLM_DEBUG_STR_VERSION 1 +/* + * Version is used by debugfs.ocfs2 to determine the format being used + * + * New in version 2 + * - Lock stats printed + * New in version 3 + * - Max time in lock stats is in usecs (instead of nsecs) + */ +#define OCFS2_DLM_DEBUG_STR_VERSION 3 static int ocfs2_dlm_seq_show(struct seq_file *m, void *v) { int i; @@ -2400,10 +2822,51 @@ static int ocfs2_dlm_seq_show(struct seq_file *m, void *v) lockres->l_blocking); /* Dump the raw LVB */ - lvb = lockres->l_lksb.lvb; + lvb = ocfs2_dlm_lvb(&lockres->l_lksb); for(i = 0; i < DLM_LVB_LEN; i++) seq_printf(m, "0x%x\t", lvb[i]); +#ifdef CONFIG_OCFS2_FS_STATS +# define lock_num_prmode(_l) ((_l)->l_lock_prmode.ls_gets) +# define lock_num_exmode(_l) ((_l)->l_lock_exmode.ls_gets) +# define lock_num_prmode_failed(_l) ((_l)->l_lock_prmode.ls_fail) +# define lock_num_exmode_failed(_l) ((_l)->l_lock_exmode.ls_fail) +# define lock_total_prmode(_l) ((_l)->l_lock_prmode.ls_total) +# define lock_total_exmode(_l) ((_l)->l_lock_exmode.ls_total) +# define lock_max_prmode(_l) ((_l)->l_lock_prmode.ls_max) +# define lock_max_exmode(_l) ((_l)->l_lock_exmode.ls_max) +# define lock_refresh(_l) ((_l)->l_lock_refresh) +#else +# define lock_num_prmode(_l) (0) +# define lock_num_exmode(_l) (0) +# define lock_num_prmode_failed(_l) (0) +# define lock_num_exmode_failed(_l) (0) +# define lock_total_prmode(_l) (0ULL) +# define lock_total_exmode(_l) (0ULL) +# define lock_max_prmode(_l) (0) +# define lock_max_exmode(_l) (0) +# define lock_refresh(_l) (0) +#endif + /* The following seq_print was added in version 2 of this output */ + seq_printf(m, "%u\t" + "%u\t" + "%u\t" + "%u\t" + "%llu\t" + "%llu\t" + "%u\t" + "%u\t" + "%u\t", + lock_num_prmode(lockres), + lock_num_exmode(lockres), + lock_num_prmode_failed(lockres), + lock_num_exmode_failed(lockres), + lock_total_prmode(lockres), + lock_total_exmode(lockres), + lock_max_prmode(lockres), + lock_max_exmode(lockres), + lock_refresh(lockres)); + /* End the line */ seq_printf(m, "\n"); return 0; @@ -2418,7 +2881,7 @@ static const struct seq_operations ocfs2_dlm_seq_ops = { static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file) { - struct seq_file *seq = (struct seq_file *) file->private_data; + struct seq_file *seq = file->private_data; struct ocfs2_dlm_seq_priv *priv = seq->private; struct ocfs2_lock_res *res = &priv->p_iter_res; @@ -2452,7 +2915,7 @@ static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file) goto out; } - seq = (struct seq_file *) file->private_data; + seq = file->private_data; seq->private = priv; ocfs2_add_lockres_tracking(&priv->p_iter_res, @@ -2504,13 +2967,12 @@ static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb) int ocfs2_dlm_init(struct ocfs2_super *osb) { int status = 0; - u32 dlm_key; - struct dlm_ctxt *dlm = NULL; - - mlog_entry_void(); + struct ocfs2_cluster_connection *conn = NULL; - if (ocfs2_mount_local(osb)) + if (ocfs2_mount_local(osb)) { + osb->node_num = 0; goto local; + } status = ocfs2_dlm_init_debug(osb); if (status < 0) { @@ -2527,26 +2989,35 @@ int ocfs2_dlm_init(struct ocfs2_super *osb) goto bail; } - /* used by the dlm code to make message headers unique, each - * node in this domain must agree on this. */ - dlm_key = crc32_le(0, osb->uuid_str, strlen(osb->uuid_str)); - /* for now, uuid == domain */ - dlm = dlm_register_domain(osb->uuid_str, dlm_key, - &osb->osb_locking_proto); - if (IS_ERR(dlm)) { - status = PTR_ERR(dlm); + status = ocfs2_cluster_connect(osb->osb_cluster_stack, + osb->osb_cluster_name, + strlen(osb->osb_cluster_name), + osb->uuid_str, + strlen(osb->uuid_str), + &lproto, ocfs2_do_node_down, osb, + &conn); + if (status) { mlog_errno(status); goto bail; } - dlm_register_eviction_cb(dlm, &osb->osb_eviction_cb); + status = ocfs2_cluster_this_node(conn, &osb->node_num); + if (status < 0) { + mlog_errno(status); + mlog(ML_ERROR, + "could not find this host's node number\n"); + ocfs2_cluster_disconnect(conn, 0); + goto bail; + } local: ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb); ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb); + ocfs2_nfs_sync_lock_res_init(&osb->osb_nfs_sync_lockres, osb); + ocfs2_orphan_scan_lock_res_init(&osb->osb_orphan_scan.os_lockres, osb); - osb->dlm = dlm; + osb->cconn = conn; status = 0; bail: @@ -2556,18 +3027,20 @@ bail: kthread_stop(osb->dc_task); } - mlog_exit(status); return status; } -void ocfs2_dlm_shutdown(struct ocfs2_super *osb) +void ocfs2_dlm_shutdown(struct ocfs2_super *osb, + int hangup_pending) { - mlog_entry_void(); - - dlm_unregister_eviction_cb(&osb->osb_eviction_cb); - ocfs2_drop_osb_locks(osb); + /* + * Now that we have dropped all locks and ocfs2_dismount_volume() + * has disabled recovery, the DLM won't be talking to us. It's + * safe to tear things down before disconnecting the cluster. + */ + if (osb->dc_task) { kthread_stop(osb->dc_task); osb->dc_task = NULL; @@ -2575,84 +3048,28 @@ void ocfs2_dlm_shutdown(struct ocfs2_super *osb) ocfs2_lock_res_free(&osb->osb_super_lockres); ocfs2_lock_res_free(&osb->osb_rename_lockres); + ocfs2_lock_res_free(&osb->osb_nfs_sync_lockres); + ocfs2_lock_res_free(&osb->osb_orphan_scan.os_lockres); - dlm_unregister_domain(osb->dlm); - osb->dlm = NULL; + ocfs2_cluster_disconnect(osb->cconn, hangup_pending); + osb->cconn = NULL; ocfs2_dlm_shutdown_debug(osb); - - mlog_exit_void(); -} - -static void ocfs2_unlock_ast(void *opaque, enum dlm_status status) -{ - struct ocfs2_lock_res *lockres = opaque; - unsigned long flags; - - mlog_entry_void(); - - mlog(0, "UNLOCK AST called on lock %s, action = %d\n", lockres->l_name, - lockres->l_unlock_action); - - spin_lock_irqsave(&lockres->l_lock, flags); - /* We tried to cancel a convert request, but it was already - * granted. All we want to do here is clear our unlock - * state. The wake_up call done at the bottom is redundant - * (ocfs2_prepare_cancel_convert doesn't sleep on this) but doesn't - * hurt anything anyway */ - if (status == DLM_CANCELGRANT && - lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) { - mlog(0, "Got cancelgrant for %s\n", lockres->l_name); - - /* We don't clear the busy flag in this case as it - * should have been cleared by the ast which the dlm - * has called. */ - goto complete_unlock; - } - - if (status != DLM_NORMAL) { - mlog(ML_ERROR, "Dlm passes status %d for lock %s, " - "unlock_action %d\n", status, lockres->l_name, - lockres->l_unlock_action); - spin_unlock_irqrestore(&lockres->l_lock, flags); - return; - } - - switch(lockres->l_unlock_action) { - case OCFS2_UNLOCK_CANCEL_CONVERT: - mlog(0, "Cancel convert success for %s\n", lockres->l_name); - lockres->l_action = OCFS2_AST_INVALID; - break; - case OCFS2_UNLOCK_DROP_LOCK: - lockres->l_level = LKM_IVMODE; - break; - default: - BUG(); - } - - lockres_clear_flags(lockres, OCFS2_LOCK_BUSY); -complete_unlock: - lockres->l_unlock_action = OCFS2_UNLOCK_INVALID; - spin_unlock_irqrestore(&lockres->l_lock, flags); - - wake_up(&lockres->l_event); - - mlog_exit_void(); } static int ocfs2_drop_lock(struct ocfs2_super *osb, struct ocfs2_lock_res *lockres) { - enum dlm_status status; + int ret; unsigned long flags; - int lkm_flags = 0; + u32 lkm_flags = 0; /* We didn't get anywhere near actually using this lockres. */ if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED)) goto out; if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) - lkm_flags |= LKM_VALBLK; + lkm_flags |= DLM_LKF_VALBLK; spin_lock_irqsave(&lockres->l_lock, flags); @@ -2678,7 +3095,7 @@ static int ocfs2_drop_lock(struct ocfs2_super *osb, if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) { if (lockres->l_flags & OCFS2_LOCK_ATTACHED && - lockres->l_level == LKM_EXMODE && + lockres->l_level == DLM_LOCK_EX && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) lockres->l_ops->set_lvb(lockres); } @@ -2707,39 +3124,75 @@ static int ocfs2_drop_lock(struct ocfs2_super *osb, mlog(0, "lock %s\n", lockres->l_name); - status = dlmunlock(osb->dlm, &lockres->l_lksb, lkm_flags, - ocfs2_unlock_ast, lockres); - if (status != DLM_NORMAL) { - ocfs2_log_dlm_error("dlmunlock", status, lockres); + ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, lkm_flags); + if (ret) { + ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres); mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags); - dlm_print_one_lock(lockres->l_lksb.lockid); + ocfs2_dlm_dump_lksb(&lockres->l_lksb); BUG(); } - mlog(0, "lock %s, successfull return from dlmunlock\n", + mlog(0, "lock %s, successful return from ocfs2_dlm_unlock\n", lockres->l_name); ocfs2_wait_on_busy_lock(lockres); out: - mlog_exit(0); return 0; } +static void ocfs2_process_blocked_lock(struct ocfs2_super *osb, + struct ocfs2_lock_res *lockres); + /* Mark the lockres as being dropped. It will no longer be * queued if blocking, but we still may have to wait on it * being dequeued from the downconvert thread before we can consider - * it safe to drop. + * it safe to drop. * * You can *not* attempt to call cluster_lock on this lockres anymore. */ -void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres) +void ocfs2_mark_lockres_freeing(struct ocfs2_super *osb, + struct ocfs2_lock_res *lockres) { int status; struct ocfs2_mask_waiter mw; - unsigned long flags; + unsigned long flags, flags2; ocfs2_init_mask_waiter(&mw); spin_lock_irqsave(&lockres->l_lock, flags); lockres->l_flags |= OCFS2_LOCK_FREEING; + if (lockres->l_flags & OCFS2_LOCK_QUEUED && current == osb->dc_task) { + /* + * We know the downconvert is queued but not in progress + * because we are the downconvert thread and processing + * different lock. So we can just remove the lock from the + * queue. This is not only an optimization but also a way + * to avoid the following deadlock: + * ocfs2_dentry_post_unlock() + * ocfs2_dentry_lock_put() + * ocfs2_drop_dentry_lock() + * iput() + * ocfs2_evict_inode() + * ocfs2_clear_inode() + * ocfs2_mark_lockres_freeing() + * ... blocks waiting for OCFS2_LOCK_QUEUED + * since we are the downconvert thread which + * should clear the flag. + */ + spin_unlock_irqrestore(&lockres->l_lock, flags); + spin_lock_irqsave(&osb->dc_task_lock, flags2); + list_del_init(&lockres->l_blocked_list); + osb->blocked_lock_count--; + spin_unlock_irqrestore(&osb->dc_task_lock, flags2); + /* + * Warn if we recurse into another post_unlock call. Strictly + * speaking it isn't a problem but we need to be careful if + * that happens (stack overflow, deadlocks, ...) so warn if + * ocfs2 grows a path for which this can happen. + */ + WARN_ON_ONCE(lockres->l_ops->post_unlock); + /* Since the lock is freeing we don't do much in the fn below */ + ocfs2_process_blocked_lock(osb, lockres); + return; + } while (lockres->l_flags & OCFS2_LOCK_QUEUED) { lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0); spin_unlock_irqrestore(&lockres->l_lock, flags); @@ -2760,7 +3213,7 @@ void ocfs2_simple_drop_lockres(struct ocfs2_super *osb, { int ret; - ocfs2_mark_lockres_freeing(lockres); + ocfs2_mark_lockres_freeing(osb, lockres); ret = ocfs2_drop_lock(osb, lockres); if (ret) mlog_errno(ret); @@ -2770,14 +3223,14 @@ static void ocfs2_drop_osb_locks(struct ocfs2_super *osb) { ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres); ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres); + ocfs2_simple_drop_lockres(osb, &osb->osb_nfs_sync_lockres); + ocfs2_simple_drop_lockres(osb, &osb->osb_orphan_scan.os_lockres); } int ocfs2_drop_inode_locks(struct inode *inode) { int status, err; - mlog_entry_void(); - /* No need to call ocfs2_mark_lockres_freeing here - * ocfs2_clear_inode has done it for us. */ @@ -2802,81 +3255,82 @@ int ocfs2_drop_inode_locks(struct inode *inode) if (err < 0 && !status) status = err; - mlog_exit(status); return status; } -static void ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres, - int new_level) +static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres, + int new_level) { assert_spin_locked(&lockres->l_lock); - BUG_ON(lockres->l_blocking <= LKM_NLMODE); + BUG_ON(lockres->l_blocking <= DLM_LOCK_NL); if (lockres->l_level <= new_level) { - mlog(ML_ERROR, "lockres->l_level (%u) <= new_level (%u)\n", - lockres->l_level, new_level); + mlog(ML_ERROR, "lockres %s, lvl %d <= %d, blcklst %d, mask %d, " + "type %d, flags 0x%lx, hold %d %d, act %d %d, req %d, " + "block %d, pgen %d\n", lockres->l_name, lockres->l_level, + new_level, list_empty(&lockres->l_blocked_list), + list_empty(&lockres->l_mask_waiters), lockres->l_type, + lockres->l_flags, lockres->l_ro_holders, + lockres->l_ex_holders, lockres->l_action, + lockres->l_unlock_action, lockres->l_requested, + lockres->l_blocking, lockres->l_pending_gen); BUG(); } - mlog(0, "lock %s, new_level = %d, l_blocking = %d\n", - lockres->l_name, new_level, lockres->l_blocking); + mlog(ML_BASTS, "lockres %s, level %d => %d, blocking %d\n", + lockres->l_name, lockres->l_level, new_level, lockres->l_blocking); lockres->l_action = OCFS2_AST_DOWNCONVERT; lockres->l_requested = new_level; lockres_or_flags(lockres, OCFS2_LOCK_BUSY); + return lockres_set_pending(lockres); } static int ocfs2_downconvert_lock(struct ocfs2_super *osb, struct ocfs2_lock_res *lockres, int new_level, - int lvb) + int lvb, + unsigned int generation) { - int ret, dlm_flags = LKM_CONVERT; - enum dlm_status status; + int ret; + u32 dlm_flags = DLM_LKF_CONVERT; - mlog_entry_void(); + mlog(ML_BASTS, "lockres %s, level %d => %d\n", lockres->l_name, + lockres->l_level, new_level); if (lvb) - dlm_flags |= LKM_VALBLK; - - status = dlmlock(osb->dlm, - new_level, - &lockres->l_lksb, - dlm_flags, - lockres->l_name, - OCFS2_LOCK_ID_MAX_LEN - 1, - ocfs2_locking_ast, - lockres, - ocfs2_blocking_ast); - if (status != DLM_NORMAL) { - ocfs2_log_dlm_error("dlmlock", status, lockres); - ret = -EINVAL; + dlm_flags |= DLM_LKF_VALBLK; + + ret = ocfs2_dlm_lock(osb->cconn, + new_level, + &lockres->l_lksb, + dlm_flags, + lockres->l_name, + OCFS2_LOCK_ID_MAX_LEN - 1); + lockres_clear_pending(lockres, generation, osb); + if (ret) { + ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres); ocfs2_recover_from_dlm_error(lockres, 1); goto bail; } ret = 0; bail: - mlog_exit(ret); return ret; } -/* returns 1 when the caller should unlock and call dlmunlock */ +/* returns 1 when the caller should unlock and call ocfs2_dlm_unlock */ static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb, struct ocfs2_lock_res *lockres) { assert_spin_locked(&lockres->l_lock); - mlog_entry_void(); - mlog(0, "lock %s\n", lockres->l_name); - if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) { /* If we're already trying to cancel a lock conversion * then just drop the spinlock and allow the caller to * requeue this lock. */ - - mlog(0, "Lockres %s, skip convert\n", lockres->l_name); + mlog(ML_BASTS, "lockres %s, skip convert\n", lockres->l_name); return 0; } @@ -2891,6 +3345,8 @@ static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb, "lock %s, invalid flags: 0x%lx\n", lockres->l_name, lockres->l_flags); + mlog(ML_BASTS, "lockres %s\n", lockres->l_name); + return 1; } @@ -2898,26 +3354,16 @@ static int ocfs2_cancel_convert(struct ocfs2_super *osb, struct ocfs2_lock_res *lockres) { int ret; - enum dlm_status status; - - mlog_entry_void(); - mlog(0, "lock %s\n", lockres->l_name); - ret = 0; - status = dlmunlock(osb->dlm, - &lockres->l_lksb, - LKM_CANCEL, - ocfs2_unlock_ast, - lockres); - if (status != DLM_NORMAL) { - ocfs2_log_dlm_error("dlmunlock", status, lockres); - ret = -EINVAL; + ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, + DLM_LKF_CANCEL); + if (ret) { + ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres); ocfs2_recover_from_dlm_error(lockres, 0); } - mlog(0, "lock %s return from dlmunlock\n", lockres->l_name); + mlog(ML_BASTS, "lockres %s\n", lockres->l_name); - mlog_exit(ret); return ret; } @@ -2928,17 +3374,54 @@ static int ocfs2_unblock_lock(struct ocfs2_super *osb, unsigned long flags; int blocking; int new_level; + int level; int ret = 0; int set_lvb = 0; - - mlog_entry_void(); + unsigned int gen; spin_lock_irqsave(&lockres->l_lock, flags); - BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED)); - recheck: + /* + * Is it still blocking? If not, we have no more work to do. + */ + if (!(lockres->l_flags & OCFS2_LOCK_BLOCKED)) { + BUG_ON(lockres->l_blocking != DLM_LOCK_NL); + spin_unlock_irqrestore(&lockres->l_lock, flags); + ret = 0; + goto leave; + } + if (lockres->l_flags & OCFS2_LOCK_BUSY) { + /* XXX + * This is a *big* race. The OCFS2_LOCK_PENDING flag + * exists entirely for one reason - another thread has set + * OCFS2_LOCK_BUSY, but has *NOT* yet called dlm_lock(). + * + * If we do ocfs2_cancel_convert() before the other thread + * calls dlm_lock(), our cancel will do nothing. We will + * get no ast, and we will have no way of knowing the + * cancel failed. Meanwhile, the other thread will call + * into dlm_lock() and wait...forever. + * + * Why forever? Because another node has asked for the + * lock first; that's why we're here in unblock_lock(). + * + * The solution is OCFS2_LOCK_PENDING. When PENDING is + * set, we just requeue the unblock. Only when the other + * thread has called dlm_lock() and cleared PENDING will + * we then cancel their request. + * + * All callers of dlm_lock() must set OCFS2_DLM_PENDING + * at the same time they set OCFS2_DLM_BUSY. They must + * clear OCFS2_DLM_PENDING after dlm_lock() returns. + */ + if (lockres->l_flags & OCFS2_LOCK_PENDING) { + mlog(ML_BASTS, "lockres %s, ReQ: Pending\n", + lockres->l_name); + goto leave_requeue; + } + ctl->requeue = 1; ret = ocfs2_prepare_cancel_convert(osb, lockres); spin_unlock_irqrestore(&lockres->l_lock, flags); @@ -2950,31 +3433,70 @@ recheck: goto leave; } + /* + * This prevents livelocks. OCFS2_LOCK_UPCONVERT_FINISHING flag is + * set when the ast is received for an upconvert just before the + * OCFS2_LOCK_BUSY flag is cleared. Now if the fs received a bast + * on the heels of the ast, we want to delay the downconvert just + * enough to allow the up requestor to do its task. Because this + * lock is in the blocked queue, the lock will be downconverted + * as soon as the requestor is done with the lock. + */ + if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING) + goto leave_requeue; + + /* + * How can we block and yet be at NL? We were trying to upconvert + * from NL and got canceled. The code comes back here, and now + * we notice and clear BLOCKING. + */ + if (lockres->l_level == DLM_LOCK_NL) { + BUG_ON(lockres->l_ex_holders || lockres->l_ro_holders); + mlog(ML_BASTS, "lockres %s, Aborting dc\n", lockres->l_name); + lockres->l_blocking = DLM_LOCK_NL; + lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED); + spin_unlock_irqrestore(&lockres->l_lock, flags); + goto leave; + } + /* if we're blocking an exclusive and we have *any* holders, * then requeue. */ - if ((lockres->l_blocking == LKM_EXMODE) - && (lockres->l_ex_holders || lockres->l_ro_holders)) + if ((lockres->l_blocking == DLM_LOCK_EX) + && (lockres->l_ex_holders || lockres->l_ro_holders)) { + mlog(ML_BASTS, "lockres %s, ReQ: EX/PR Holders %u,%u\n", + lockres->l_name, lockres->l_ex_holders, + lockres->l_ro_holders); goto leave_requeue; + } /* If it's a PR we're blocking, then only * requeue if we've got any EX holders */ - if (lockres->l_blocking == LKM_PRMODE && - lockres->l_ex_holders) + if (lockres->l_blocking == DLM_LOCK_PR && + lockres->l_ex_holders) { + mlog(ML_BASTS, "lockres %s, ReQ: EX Holders %u\n", + lockres->l_name, lockres->l_ex_holders); goto leave_requeue; + } /* * Can we get a lock in this state if the holder counts are * zero? The meta data unblock code used to check this. */ if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH) - && (lockres->l_flags & OCFS2_LOCK_REFRESHING)) + && (lockres->l_flags & OCFS2_LOCK_REFRESHING)) { + mlog(ML_BASTS, "lockres %s, ReQ: Lock Refreshing\n", + lockres->l_name); goto leave_requeue; + } new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking); if (lockres->l_ops->check_downconvert - && !lockres->l_ops->check_downconvert(lockres, new_level)) + && !lockres->l_ops->check_downconvert(lockres, new_level)) { + mlog(ML_BASTS, "lockres %s, ReQ: Checkpointing\n", + lockres->l_name); goto leave_requeue; + } /* If we get here, then we know that there are no more * incompatible holders (and anyone asking for an incompatible @@ -2987,17 +3509,24 @@ recheck: * may sleep, so we save off a copy of what we're blocking as * it may change while we're not holding the spin lock. */ blocking = lockres->l_blocking; + level = lockres->l_level; spin_unlock_irqrestore(&lockres->l_lock, flags); ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking); - if (ctl->unblock_action == UNBLOCK_STOP_POST) + if (ctl->unblock_action == UNBLOCK_STOP_POST) { + mlog(ML_BASTS, "lockres %s, UNBLOCK_STOP_POST\n", + lockres->l_name); goto leave; + } spin_lock_irqsave(&lockres->l_lock, flags); - if (blocking != lockres->l_blocking) { + if ((blocking != lockres->l_blocking) || (level != lockres->l_level)) { /* If this changed underneath us, then we can't drop * it just yet. */ + mlog(ML_BASTS, "lockres %s, block=%d:%d, level=%d:%d, " + "Recheck\n", lockres->l_name, blocking, + lockres->l_blocking, level, lockres->l_level); goto recheck; } @@ -3005,7 +3534,7 @@ downconvert: ctl->requeue = 0; if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) { - if (lockres->l_level == LKM_EXMODE) + if (lockres->l_level == DLM_LOCK_EX) set_lvb = 1; /* @@ -3018,18 +3547,20 @@ downconvert: lockres->l_ops->set_lvb(lockres); } - ocfs2_prepare_downconvert(lockres, new_level); + gen = ocfs2_prepare_downconvert(lockres, new_level); spin_unlock_irqrestore(&lockres->l_lock, flags); - ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb); + ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb, + gen); + leave: - mlog_exit(ret); + if (ret) + mlog_errno(ret); return ret; leave_requeue: spin_unlock_irqrestore(&lockres->l_lock, flags); ctl->requeue = 1; - mlog_exit(0); return 0; } @@ -3038,10 +3569,18 @@ static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres, { struct inode *inode; struct address_space *mapping; + struct ocfs2_inode_info *oi; inode = ocfs2_lock_res_inode(lockres); mapping = inode->i_mapping; + if (S_ISDIR(inode->i_mode)) { + oi = OCFS2_I(inode); + oi->ip_dir_lock_gen++; + mlog(0, "generation: %u\n", oi->ip_dir_lock_gen); + goto out; + } + if (!S_ISREG(inode->i_mode)) goto out; @@ -3059,7 +3598,7 @@ static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres, (unsigned long long)OCFS2_I(inode)->ip_blkno); } sync_mapping_buffers(mapping); - if (blocking == LKM_EXMODE) { + if (blocking == DLM_LOCK_EX) { truncate_inode_pages(mapping, 0); } else { /* We only need to wait on the I/O if we're not also @@ -3074,22 +3613,30 @@ out: return UNBLOCK_CONTINUE; } -static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres, - int new_level) +static int ocfs2_ci_checkpointed(struct ocfs2_caching_info *ci, + struct ocfs2_lock_res *lockres, + int new_level) { - struct inode *inode = ocfs2_lock_res_inode(lockres); - int checkpointed = ocfs2_inode_fully_checkpointed(inode); + int checkpointed = ocfs2_ci_fully_checkpointed(ci); - BUG_ON(new_level != LKM_NLMODE && new_level != LKM_PRMODE); - BUG_ON(lockres->l_level != LKM_EXMODE && !checkpointed); + BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR); + BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed); if (checkpointed) return 1; - ocfs2_start_checkpoint(OCFS2_SB(inode->i_sb)); + ocfs2_start_checkpoint(OCFS2_SB(ocfs2_metadata_cache_get_super(ci))); return 0; } +static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres, + int new_level) +{ + struct inode *inode = ocfs2_lock_res_inode(lockres); + + return ocfs2_ci_checkpointed(INODE_CACHE(inode), lockres, new_level); +} + static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres) { struct inode *inode = ocfs2_lock_res_inode(lockres); @@ -3145,7 +3692,7 @@ static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres, * valid. The downconvert code will retain a PR for this node, * so there's no further work to do. */ - if (blocking == LKM_PRMODE) + if (blocking == DLM_LOCK_PR) return UNBLOCK_CONTINUE; /* @@ -3219,6 +3766,161 @@ static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres, return UNBLOCK_CONTINUE_POST; } +static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres, + int new_level) +{ + struct ocfs2_refcount_tree *tree = + ocfs2_lock_res_refcount_tree(lockres); + + return ocfs2_ci_checkpointed(&tree->rf_ci, lockres, new_level); +} + +static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres, + int blocking) +{ + struct ocfs2_refcount_tree *tree = + ocfs2_lock_res_refcount_tree(lockres); + + ocfs2_metadata_cache_purge(&tree->rf_ci); + + return UNBLOCK_CONTINUE; +} + +static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres) +{ + struct ocfs2_qinfo_lvb *lvb; + struct ocfs2_mem_dqinfo *oinfo = ocfs2_lock_res_qinfo(lockres); + struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb, + oinfo->dqi_gi.dqi_type); + + lvb = ocfs2_dlm_lvb(&lockres->l_lksb); + lvb->lvb_version = OCFS2_QINFO_LVB_VERSION; + lvb->lvb_bgrace = cpu_to_be32(info->dqi_bgrace); + lvb->lvb_igrace = cpu_to_be32(info->dqi_igrace); + lvb->lvb_syncms = cpu_to_be32(oinfo->dqi_syncms); + lvb->lvb_blocks = cpu_to_be32(oinfo->dqi_gi.dqi_blocks); + lvb->lvb_free_blk = cpu_to_be32(oinfo->dqi_gi.dqi_free_blk); + lvb->lvb_free_entry = cpu_to_be32(oinfo->dqi_gi.dqi_free_entry); +} + +void ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo *oinfo, int ex) +{ + struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock; + struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb); + int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; + + if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb)) + ocfs2_cluster_unlock(osb, lockres, level); +} + +static int ocfs2_refresh_qinfo(struct ocfs2_mem_dqinfo *oinfo) +{ + struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb, + oinfo->dqi_gi.dqi_type); + struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock; + struct ocfs2_qinfo_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb); + struct buffer_head *bh = NULL; + struct ocfs2_global_disk_dqinfo *gdinfo; + int status = 0; + + if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) && + lvb->lvb_version == OCFS2_QINFO_LVB_VERSION) { + info->dqi_bgrace = be32_to_cpu(lvb->lvb_bgrace); + info->dqi_igrace = be32_to_cpu(lvb->lvb_igrace); + oinfo->dqi_syncms = be32_to_cpu(lvb->lvb_syncms); + oinfo->dqi_gi.dqi_blocks = be32_to_cpu(lvb->lvb_blocks); + oinfo->dqi_gi.dqi_free_blk = be32_to_cpu(lvb->lvb_free_blk); + oinfo->dqi_gi.dqi_free_entry = + be32_to_cpu(lvb->lvb_free_entry); + } else { + status = ocfs2_read_quota_phys_block(oinfo->dqi_gqinode, + oinfo->dqi_giblk, &bh); + if (status) { + mlog_errno(status); + goto bail; + } + gdinfo = (struct ocfs2_global_disk_dqinfo *) + (bh->b_data + OCFS2_GLOBAL_INFO_OFF); + info->dqi_bgrace = le32_to_cpu(gdinfo->dqi_bgrace); + info->dqi_igrace = le32_to_cpu(gdinfo->dqi_igrace); + oinfo->dqi_syncms = le32_to_cpu(gdinfo->dqi_syncms); + oinfo->dqi_gi.dqi_blocks = le32_to_cpu(gdinfo->dqi_blocks); + oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(gdinfo->dqi_free_blk); + oinfo->dqi_gi.dqi_free_entry = + le32_to_cpu(gdinfo->dqi_free_entry); + brelse(bh); + ocfs2_track_lock_refresh(lockres); + } + +bail: + return status; +} + +/* Lock quota info, this function expects at least shared lock on the quota file + * so that we can safely refresh quota info from disk. */ +int ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo *oinfo, int ex) +{ + struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock; + struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb); + int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; + int status = 0; + + /* On RO devices, locking really isn't needed... */ + if (ocfs2_is_hard_readonly(osb)) { + if (ex) + status = -EROFS; + goto bail; + } + if (ocfs2_mount_local(osb)) + goto bail; + + status = ocfs2_cluster_lock(osb, lockres, level, 0, 0); + if (status < 0) { + mlog_errno(status); + goto bail; + } + if (!ocfs2_should_refresh_lock_res(lockres)) + goto bail; + /* OK, we have the lock but we need to refresh the quota info */ + status = ocfs2_refresh_qinfo(oinfo); + if (status) + ocfs2_qinfo_unlock(oinfo, ex); + ocfs2_complete_lock_res_refresh(lockres, status); +bail: + return status; +} + +int ocfs2_refcount_lock(struct ocfs2_refcount_tree *ref_tree, int ex) +{ + int status; + int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; + struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres; + struct ocfs2_super *osb = lockres->l_priv; + + + if (ocfs2_is_hard_readonly(osb)) + return -EROFS; + + if (ocfs2_mount_local(osb)) + return 0; + + status = ocfs2_cluster_lock(osb, lockres, level, 0, 0); + if (status < 0) + mlog_errno(status); + + return status; +} + +void ocfs2_refcount_unlock(struct ocfs2_refcount_tree *ref_tree, int ex) +{ + int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR; + struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres; + struct ocfs2_super *osb = lockres->l_priv; + + if (!ocfs2_mount_local(osb)) + ocfs2_cluster_unlock(osb, lockres, level); +} + static void ocfs2_process_blocked_lock(struct ocfs2_super *osb, struct ocfs2_lock_res *lockres) { @@ -3230,12 +3932,10 @@ static void ocfs2_process_blocked_lock(struct ocfs2_super *osb, * considered valid until we remove the OCFS2_LOCK_QUEUED * flag. */ - mlog_entry_void(); - BUG_ON(!lockres); BUG_ON(!lockres->l_ops); - mlog(0, "lockres %s blocked.\n", lockres->l_name); + mlog(ML_BASTS, "lockres %s blocked\n", lockres->l_name); /* Detect whether a lock has been marked as going away while * the downconvert thread was processing other things. A lock can @@ -3258,21 +3958,19 @@ unqueue: } else ocfs2_schedule_blocked_lock(osb, lockres); - mlog(0, "lockres %s, requeue = %s.\n", lockres->l_name, + mlog(ML_BASTS, "lockres %s, requeue = %s.\n", lockres->l_name, ctl.requeue ? "yes" : "no"); spin_unlock_irqrestore(&lockres->l_lock, flags); if (ctl.unblock_action != UNBLOCK_CONTINUE && lockres->l_ops->post_unlock) lockres->l_ops->post_unlock(osb, lockres); - - mlog_exit_void(); } static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb, struct ocfs2_lock_res *lockres) { - mlog_entry_void(); + unsigned long flags; assert_spin_locked(&lockres->l_lock); @@ -3280,32 +3978,29 @@ static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb, /* Do not schedule a lock for downconvert when it's on * the way to destruction - any nodes wanting access * to the resource will get it soon. */ - mlog(0, "Lockres %s won't be scheduled: flags 0x%lx\n", + mlog(ML_BASTS, "lockres %s won't be scheduled: flags 0x%lx\n", lockres->l_name, lockres->l_flags); return; } lockres_or_flags(lockres, OCFS2_LOCK_QUEUED); - spin_lock(&osb->dc_task_lock); + spin_lock_irqsave(&osb->dc_task_lock, flags); if (list_empty(&lockres->l_blocked_list)) { list_add_tail(&lockres->l_blocked_list, &osb->blocked_lock_list); osb->blocked_lock_count++; } - spin_unlock(&osb->dc_task_lock); - - mlog_exit_void(); + spin_unlock_irqrestore(&osb->dc_task_lock, flags); } static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb) { unsigned long processed; + unsigned long flags; struct ocfs2_lock_res *lockres; - mlog_entry_void(); - - spin_lock(&osb->dc_task_lock); + spin_lock_irqsave(&osb->dc_task_lock, flags); /* grab this early so we know to try again if a state change and * wake happens part-way through our work */ osb->dc_work_sequence = osb->dc_wake_sequence; @@ -3318,40 +4013,40 @@ static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb) struct ocfs2_lock_res, l_blocked_list); list_del_init(&lockres->l_blocked_list); osb->blocked_lock_count--; - spin_unlock(&osb->dc_task_lock); + spin_unlock_irqrestore(&osb->dc_task_lock, flags); BUG_ON(!processed); processed--; ocfs2_process_blocked_lock(osb, lockres); - spin_lock(&osb->dc_task_lock); + spin_lock_irqsave(&osb->dc_task_lock, flags); } - spin_unlock(&osb->dc_task_lock); - - mlog_exit_void(); + spin_unlock_irqrestore(&osb->dc_task_lock, flags); } static int ocfs2_downconvert_thread_lists_empty(struct ocfs2_super *osb) { int empty = 0; + unsigned long flags; - spin_lock(&osb->dc_task_lock); + spin_lock_irqsave(&osb->dc_task_lock, flags); if (list_empty(&osb->blocked_lock_list)) empty = 1; - spin_unlock(&osb->dc_task_lock); + spin_unlock_irqrestore(&osb->dc_task_lock, flags); return empty; } static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb) { int should_wake = 0; + unsigned long flags; - spin_lock(&osb->dc_task_lock); + spin_lock_irqsave(&osb->dc_task_lock, flags); if (osb->dc_work_sequence != osb->dc_wake_sequence) should_wake = 1; - spin_unlock(&osb->dc_task_lock); + spin_unlock_irqrestore(&osb->dc_task_lock, flags); return should_wake; } @@ -3381,10 +4076,12 @@ static int ocfs2_downconvert_thread(void *arg) void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb) { - spin_lock(&osb->dc_task_lock); + unsigned long flags; + + spin_lock_irqsave(&osb->dc_task_lock, flags); /* make sure the voting thread gets a swipe at whatever changes * the caller may have made to the voting state */ osb->dc_wake_sequence++; - spin_unlock(&osb->dc_task_lock); + spin_unlock_irqrestore(&osb->dc_task_lock, flags); wake_up(&osb->dc_event); } |