diff options
Diffstat (limited to 'fs/ocfs2')
-rw-r--r-- | fs/ocfs2/Makefile | 1 | ||||
-rw-r--r-- | fs/ocfs2/stack_o2cb.c | 395 | ||||
-rw-r--r-- | fs/ocfs2/stackglue.c | 385 | ||||
-rw-r--r-- | fs/ocfs2/stackglue.h | 123 |
4 files changed, 532 insertions, 372 deletions
diff --git a/fs/ocfs2/Makefile b/fs/ocfs2/Makefile index 3ba64af2695..8e86195fedb 100644 --- a/fs/ocfs2/Makefile +++ b/fs/ocfs2/Makefile @@ -25,6 +25,7 @@ ocfs2-objs := \ resize.o \ slot_map.o \ stackglue.o \ + stack_o2cb.o \ suballoc.o \ super.o \ symlink.o \ diff --git a/fs/ocfs2/stack_o2cb.c b/fs/ocfs2/stack_o2cb.c new file mode 100644 index 00000000000..c9bc3541a33 --- /dev/null +++ b/fs/ocfs2/stack_o2cb.c @@ -0,0 +1,395 @@ +/* -*- mode: c; c-basic-offset: 8; -*- + * vim: noexpandtab sw=8 ts=8 sts=0: + * + * stack_o2cb.c + * + * Code which interfaces ocfs2 with the o2cb stack. + * + * Copyright (C) 2007 Oracle. All rights reserved. + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public + * License as published by the Free Software Foundation, version 2. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License for more details. + */ + +#include <linux/crc32.h> +#include <linux/kmod.h> + +/* Needed for AOP_TRUNCATED_PAGE in mlog_errno() */ +#include <linux/fs.h> + +#include "cluster/masklog.h" +#include "cluster/nodemanager.h" +#include "cluster/heartbeat.h" + +#include "stackglue.h" + +struct o2dlm_private { + struct dlm_eviction_cb op_eviction_cb; +}; + +/* These should be identical */ +#if (DLM_LOCK_IV != LKM_IVMODE) +# error Lock modes do not match +#endif +#if (DLM_LOCK_NL != LKM_NLMODE) +# error Lock modes do not match +#endif +#if (DLM_LOCK_CR != LKM_CRMODE) +# error Lock modes do not match +#endif +#if (DLM_LOCK_CW != LKM_CWMODE) +# error Lock modes do not match +#endif +#if (DLM_LOCK_PR != LKM_PRMODE) +# error Lock modes do not match +#endif +#if (DLM_LOCK_PW != LKM_PWMODE) +# error Lock modes do not match +#endif +#if (DLM_LOCK_EX != LKM_EXMODE) +# error Lock modes do not match +#endif +static inline int mode_to_o2dlm(int mode) +{ + BUG_ON(mode > LKM_MAXMODE); + + return mode; +} + +#define map_flag(_generic, _o2dlm) \ + if (flags & (_generic)) { \ + flags &= ~(_generic); \ + o2dlm_flags |= (_o2dlm); \ + } +static int flags_to_o2dlm(u32 flags) +{ + int o2dlm_flags = 0; + + map_flag(DLM_LKF_NOQUEUE, LKM_NOQUEUE); + map_flag(DLM_LKF_CANCEL, LKM_CANCEL); + map_flag(DLM_LKF_CONVERT, LKM_CONVERT); + map_flag(DLM_LKF_VALBLK, LKM_VALBLK); + map_flag(DLM_LKF_IVVALBLK, LKM_INVVALBLK); + map_flag(DLM_LKF_ORPHAN, LKM_ORPHAN); + map_flag(DLM_LKF_FORCEUNLOCK, LKM_FORCE); + map_flag(DLM_LKF_TIMEOUT, LKM_TIMEOUT); + map_flag(DLM_LKF_LOCAL, LKM_LOCAL); + + /* map_flag() should have cleared every flag passed in */ + BUG_ON(flags != 0); + + return o2dlm_flags; +} +#undef map_flag + +/* + * Map an o2dlm status to standard errno values. + * + * o2dlm only uses a handful of these, and returns even fewer to the + * caller. Still, we try to assign sane values to each error. + * + * The following value pairs have special meanings to dlmglue, thus + * the right hand side needs to stay unique - never duplicate the + * mapping elsewhere in the table! + * + * DLM_NORMAL: 0 + * DLM_NOTQUEUED: -EAGAIN + * DLM_CANCELGRANT: -EBUSY + * DLM_CANCEL: -DLM_ECANCEL + */ +/* Keep in sync with dlmapi.h */ +static int status_map[] = { + [DLM_NORMAL] = 0, /* Success */ + [DLM_GRANTED] = -EINVAL, + [DLM_DENIED] = -EACCES, + [DLM_DENIED_NOLOCKS] = -EACCES, + [DLM_WORKING] = -EACCES, + [DLM_BLOCKED] = -EINVAL, + [DLM_BLOCKED_ORPHAN] = -EINVAL, + [DLM_DENIED_GRACE_PERIOD] = -EACCES, + [DLM_SYSERR] = -ENOMEM, /* It is what it is */ + [DLM_NOSUPPORT] = -EPROTO, + [DLM_CANCELGRANT] = -EBUSY, /* Cancel after grant */ + [DLM_IVLOCKID] = -EINVAL, + [DLM_SYNC] = -EINVAL, + [DLM_BADTYPE] = -EINVAL, + [DLM_BADRESOURCE] = -EINVAL, + [DLM_MAXHANDLES] = -ENOMEM, + [DLM_NOCLINFO] = -EINVAL, + [DLM_NOLOCKMGR] = -EINVAL, + [DLM_NOPURGED] = -EINVAL, + [DLM_BADARGS] = -EINVAL, + [DLM_VOID] = -EINVAL, + [DLM_NOTQUEUED] = -EAGAIN, /* Trylock failed */ + [DLM_IVBUFLEN] = -EINVAL, + [DLM_CVTUNGRANT] = -EPERM, + [DLM_BADPARAM] = -EINVAL, + [DLM_VALNOTVALID] = -EINVAL, + [DLM_REJECTED] = -EPERM, + [DLM_ABORT] = -EINVAL, + [DLM_CANCEL] = -DLM_ECANCEL, /* Successful cancel */ + [DLM_IVRESHANDLE] = -EINVAL, + [DLM_DEADLOCK] = -EDEADLK, + [DLM_DENIED_NOASTS] = -EINVAL, + [DLM_FORWARD] = -EINVAL, + [DLM_TIMEOUT] = -ETIMEDOUT, + [DLM_IVGROUPID] = -EINVAL, + [DLM_VERS_CONFLICT] = -EOPNOTSUPP, + [DLM_BAD_DEVICE_PATH] = -ENOENT, + [DLM_NO_DEVICE_PERMISSION] = -EPERM, + [DLM_NO_CONTROL_DEVICE] = -ENOENT, + [DLM_RECOVERING] = -ENOTCONN, + [DLM_MIGRATING] = -ERESTART, + [DLM_MAXSTATS] = -EINVAL, +}; + +static int dlm_status_to_errno(enum dlm_status status) +{ + BUG_ON(status > (sizeof(status_map) / sizeof(status_map[0]))); + + return status_map[status]; +} + +static void o2dlm_lock_ast_wrapper(void *astarg) +{ + BUG_ON(stack_glue_lproto == NULL); + + stack_glue_lproto->lp_lock_ast(astarg); +} + +static void o2dlm_blocking_ast_wrapper(void *astarg, int level) +{ + BUG_ON(stack_glue_lproto == NULL); + + stack_glue_lproto->lp_blocking_ast(astarg, level); +} + +static void o2dlm_unlock_ast_wrapper(void *astarg, enum dlm_status status) +{ + int error = dlm_status_to_errno(status); + + BUG_ON(stack_glue_lproto == NULL); + + /* + * In o2dlm, you can get both the lock_ast() for the lock being + * granted and the unlock_ast() for the CANCEL failing. A + * successful cancel sends DLM_NORMAL here. If the + * lock grant happened before the cancel arrived, you get + * DLM_CANCELGRANT. + * + * There's no need for the double-ast. If we see DLM_CANCELGRANT, + * we just ignore it. We expect the lock_ast() to handle the + * granted lock. + */ + if (status == DLM_CANCELGRANT) + return; + + stack_glue_lproto->lp_unlock_ast(astarg, error); +} + +static int o2cb_dlm_lock(struct ocfs2_cluster_connection *conn, + int mode, + union ocfs2_dlm_lksb *lksb, + u32 flags, + void *name, + unsigned int namelen, + void *astarg) +{ + enum dlm_status status; + int o2dlm_mode = mode_to_o2dlm(mode); + int o2dlm_flags = flags_to_o2dlm(flags); + int ret; + + status = dlmlock(conn->cc_lockspace, o2dlm_mode, &lksb->lksb_o2dlm, + o2dlm_flags, name, namelen, + o2dlm_lock_ast_wrapper, astarg, + o2dlm_blocking_ast_wrapper); + ret = dlm_status_to_errno(status); + return ret; +} + +static int o2cb_dlm_unlock(struct ocfs2_cluster_connection *conn, + union ocfs2_dlm_lksb *lksb, + u32 flags, + void *astarg) +{ + enum dlm_status status; + int o2dlm_flags = flags_to_o2dlm(flags); + int ret; + + status = dlmunlock(conn->cc_lockspace, &lksb->lksb_o2dlm, + o2dlm_flags, o2dlm_unlock_ast_wrapper, astarg); + ret = dlm_status_to_errno(status); + return ret; +} + +static int o2cb_dlm_lock_status(union ocfs2_dlm_lksb *lksb) +{ + return dlm_status_to_errno(lksb->lksb_o2dlm.status); +} + +static void *o2cb_dlm_lvb(union ocfs2_dlm_lksb *lksb) +{ + return (void *)(lksb->lksb_o2dlm.lvb); +} + +static void o2cb_dump_lksb(union ocfs2_dlm_lksb *lksb) +{ + dlm_print_one_lock(lksb->lksb_o2dlm.lockid); +} + +/* + * Called from the dlm when it's about to evict a node. This is how the + * classic stack signals node death. + */ +static void o2dlm_eviction_cb(int node_num, void *data) +{ + struct ocfs2_cluster_connection *conn = data; + + mlog(ML_NOTICE, "o2dlm has evicted node %d from group %.*s\n", + node_num, conn->cc_namelen, conn->cc_name); + + conn->cc_recovery_handler(node_num, conn->cc_recovery_data); +} + +static int o2cb_cluster_connect(struct ocfs2_cluster_connection *conn) +{ + int rc = 0; + u32 dlm_key; + struct dlm_ctxt *dlm; + struct o2dlm_private *priv; + struct dlm_protocol_version dlm_version; + + BUG_ON(conn == NULL); + + /* for now we only have one cluster/node, make sure we see it + * in the heartbeat universe */ + if (!o2hb_check_local_node_heartbeating()) { + rc = -EINVAL; + goto out; + } + + priv = kzalloc(sizeof(struct o2dlm_private), GFP_KERNEL); + if (!priv) { + rc = -ENOMEM; + goto out_free; + } + + /* This just fills the structure in. It is safe to pass conn. */ + dlm_setup_eviction_cb(&priv->op_eviction_cb, o2dlm_eviction_cb, + conn); + + conn->cc_private = priv; + + /* used by the dlm code to make message headers unique, each + * node in this domain must agree on this. */ + dlm_key = crc32_le(0, conn->cc_name, conn->cc_namelen); + dlm_version.pv_major = conn->cc_version.pv_major; + dlm_version.pv_minor = conn->cc_version.pv_minor; + + dlm = dlm_register_domain(conn->cc_name, dlm_key, &dlm_version); + if (IS_ERR(dlm)) { + rc = PTR_ERR(dlm); + mlog_errno(rc); + goto out_free; + } + + conn->cc_version.pv_major = dlm_version.pv_major; + conn->cc_version.pv_minor = dlm_version.pv_minor; + conn->cc_lockspace = dlm; + + dlm_register_eviction_cb(dlm, &priv->op_eviction_cb); + +out_free: + if (rc && conn->cc_private) + kfree(conn->cc_private); + +out: + return rc; +} + +static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection *conn) +{ + struct dlm_ctxt *dlm = conn->cc_lockspace; + struct o2dlm_private *priv = conn->cc_private; + + dlm_unregister_eviction_cb(&priv->op_eviction_cb); + conn->cc_private = NULL; + kfree(priv); + + dlm_unregister_domain(dlm); + conn->cc_lockspace = NULL; + + return 0; +} + +static void o2hb_stop(const char *group) +{ + int ret; + char *argv[5], *envp[3]; + + argv[0] = (char *)o2nm_get_hb_ctl_path(); + argv[1] = "-K"; + argv[2] = "-u"; + argv[3] = (char *)group; + argv[4] = NULL; + + mlog(0, "Run: %s %s %s %s\n", argv[0], argv[1], argv[2], argv[3]); + + /* minimal command environment taken from cpu_run_sbin_hotplug */ + envp[0] = "HOME=/"; + envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; + envp[2] = NULL; + + ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC); + if (ret < 0) + mlog_errno(ret); +} + +/* + * Hangup is a hack for tools compatibility. Older ocfs2-tools software + * expects the filesystem to call "ocfs2_hb_ctl" during unmount. This + * happens regardless of whether the DLM got started, so we can't do it + * in ocfs2_cluster_disconnect(). We bring the o2hb_stop() function into + * the glue and provide a "hangup" API for super.c to call. + * + * Other stacks will eventually provide a NULL ->hangup() pointer. + */ +static void o2cb_cluster_hangup(const char *group, int grouplen) +{ + o2hb_stop(group); +} + +static int o2cb_cluster_this_node(unsigned int *node) +{ + int node_num; + + node_num = o2nm_this_node(); + if (node_num == O2NM_INVALID_NODE_NUM) + return -ENOENT; + + if (node_num >= O2NM_MAX_NODES) + return -EOVERFLOW; + + *node = node_num; + return 0; +} + +struct ocfs2_stack_operations o2cb_stack_ops = { + .connect = o2cb_cluster_connect, + .disconnect = o2cb_cluster_disconnect, + .hangup = o2cb_cluster_hangup, + .this_node = o2cb_cluster_this_node, + .dlm_lock = o2cb_dlm_lock, + .dlm_unlock = o2cb_dlm_unlock, + .lock_status = o2cb_dlm_lock_status, + .lock_lvb = o2cb_dlm_lvb, + .dump_lksb = o2cb_dump_lksb, +}; + diff --git a/fs/ocfs2/stackglue.c b/fs/ocfs2/stackglue.c index e35dde6217f..e197367b6bd 100644 --- a/fs/ocfs2/stackglue.c +++ b/fs/ocfs2/stackglue.c @@ -19,204 +19,17 @@ */ #include <linux/slab.h> -#include <linux/crc32.h> #include <linux/kmod.h> /* Needed for AOP_TRUNCATED_PAGE in mlog_errno() */ #include <linux/fs.h> #include "cluster/masklog.h" -#include "cluster/nodemanager.h" -#include "cluster/heartbeat.h" #include "stackglue.h" -static struct ocfs2_locking_protocol *lproto; - -struct o2dlm_private { - struct dlm_eviction_cb op_eviction_cb; -}; - -/* These should be identical */ -#if (DLM_LOCK_IV != LKM_IVMODE) -# error Lock modes do not match -#endif -#if (DLM_LOCK_NL != LKM_NLMODE) -# error Lock modes do not match -#endif -#if (DLM_LOCK_CR != LKM_CRMODE) -# error Lock modes do not match -#endif -#if (DLM_LOCK_CW != LKM_CWMODE) -# error Lock modes do not match -#endif -#if (DLM_LOCK_PR != LKM_PRMODE) -# error Lock modes do not match -#endif -#if (DLM_LOCK_PW != LKM_PWMODE) -# error Lock modes do not match -#endif -#if (DLM_LOCK_EX != LKM_EXMODE) -# error Lock modes do not match -#endif -static inline int mode_to_o2dlm(int mode) -{ - BUG_ON(mode > LKM_MAXMODE); - - return mode; -} - -#define map_flag(_generic, _o2dlm) \ - if (flags & (_generic)) { \ - flags &= ~(_generic); \ - o2dlm_flags |= (_o2dlm); \ - } -static int flags_to_o2dlm(u32 flags) -{ - int o2dlm_flags = 0; - - map_flag(DLM_LKF_NOQUEUE, LKM_NOQUEUE); - map_flag(DLM_LKF_CANCEL, LKM_CANCEL); - map_flag(DLM_LKF_CONVERT, LKM_CONVERT); - map_flag(DLM_LKF_VALBLK, LKM_VALBLK); - map_flag(DLM_LKF_IVVALBLK, LKM_INVVALBLK); - map_flag(DLM_LKF_ORPHAN, LKM_ORPHAN); - map_flag(DLM_LKF_FORCEUNLOCK, LKM_FORCE); - map_flag(DLM_LKF_TIMEOUT, LKM_TIMEOUT); - map_flag(DLM_LKF_LOCAL, LKM_LOCAL); - - /* map_flag() should have cleared every flag passed in */ - BUG_ON(flags != 0); - - return o2dlm_flags; -} -#undef map_flag - -/* - * Map an o2dlm status to standard errno values. - * - * o2dlm only uses a handful of these, and returns even fewer to the - * caller. Still, we try to assign sane values to each error. - * - * The following value pairs have special meanings to dlmglue, thus - * the right hand side needs to stay unique - never duplicate the - * mapping elsewhere in the table! - * - * DLM_NORMAL: 0 - * DLM_NOTQUEUED: -EAGAIN - * DLM_CANCELGRANT: -EBUSY - * DLM_CANCEL: -DLM_ECANCEL - */ -/* Keep in sync with dlmapi.h */ -static int status_map[] = { - [DLM_NORMAL] = 0, /* Success */ - [DLM_GRANTED] = -EINVAL, - [DLM_DENIED] = -EACCES, - [DLM_DENIED_NOLOCKS] = -EACCES, - [DLM_WORKING] = -EACCES, - [DLM_BLOCKED] = -EINVAL, - [DLM_BLOCKED_ORPHAN] = -EINVAL, - [DLM_DENIED_GRACE_PERIOD] = -EACCES, - [DLM_SYSERR] = -ENOMEM, /* It is what it is */ - [DLM_NOSUPPORT] = -EPROTO, - [DLM_CANCELGRANT] = -EBUSY, /* Cancel after grant */ - [DLM_IVLOCKID] = -EINVAL, - [DLM_SYNC] = -EINVAL, - [DLM_BADTYPE] = -EINVAL, - [DLM_BADRESOURCE] = -EINVAL, - [DLM_MAXHANDLES] = -ENOMEM, - [DLM_NOCLINFO] = -EINVAL, - [DLM_NOLOCKMGR] = -EINVAL, - [DLM_NOPURGED] = -EINVAL, - [DLM_BADARGS] = -EINVAL, - [DLM_VOID] = -EINVAL, - [DLM_NOTQUEUED] = -EAGAIN, /* Trylock failed */ - [DLM_IVBUFLEN] = -EINVAL, - [DLM_CVTUNGRANT] = -EPERM, - [DLM_BADPARAM] = -EINVAL, - [DLM_VALNOTVALID] = -EINVAL, - [DLM_REJECTED] = -EPERM, - [DLM_ABORT] = -EINVAL, - [DLM_CANCEL] = -DLM_ECANCEL, /* Successful cancel */ - [DLM_IVRESHANDLE] = -EINVAL, - [DLM_DEADLOCK] = -EDEADLK, - [DLM_DENIED_NOASTS] = -EINVAL, - [DLM_FORWARD] = -EINVAL, - [DLM_TIMEOUT] = -ETIMEDOUT, - [DLM_IVGROUPID] = -EINVAL, - [DLM_VERS_CONFLICT] = -EOPNOTSUPP, - [DLM_BAD_DEVICE_PATH] = -ENOENT, - [DLM_NO_DEVICE_PERMISSION] = -EPERM, - [DLM_NO_CONTROL_DEVICE] = -ENOENT, - [DLM_RECOVERING] = -ENOTCONN, - [DLM_MIGRATING] = -ERESTART, - [DLM_MAXSTATS] = -EINVAL, -}; - -static int dlm_status_to_errno(enum dlm_status status) -{ - BUG_ON(status > (sizeof(status_map) / sizeof(status_map[0]))); +struct ocfs2_locking_protocol *stack_glue_lproto; - return status_map[status]; -} - -static void o2dlm_lock_ast_wrapper(void *astarg) -{ - BUG_ON(lproto == NULL); - - lproto->lp_lock_ast(astarg); -} - -static void o2dlm_blocking_ast_wrapper(void *astarg, int level) -{ - BUG_ON(lproto == NULL); - - lproto->lp_blocking_ast(astarg, level); -} - -static void o2dlm_unlock_ast_wrapper(void *astarg, enum dlm_status status) -{ - int error = dlm_status_to_errno(status); - - BUG_ON(lproto == NULL); - - /* - * In o2dlm, you can get both the lock_ast() for the lock being - * granted and the unlock_ast() for the CANCEL failing. A - * successful cancel sends DLM_NORMAL here. If the - * lock grant happened before the cancel arrived, you get - * DLM_CANCELGRANT. - * - * There's no need for the double-ast. If we see DLM_CANCELGRANT, - * we just ignore it. We expect the lock_ast() to handle the - * granted lock. - */ - if (status == DLM_CANCELGRANT) - return; - - lproto->lp_unlock_ast(astarg, error); -} - -static int o2cb_dlm_lock(struct ocfs2_cluster_connection *conn, - int mode, - union ocfs2_dlm_lksb *lksb, - u32 flags, - void *name, - unsigned int namelen, - void *astarg) -{ - enum dlm_status status; - int o2dlm_mode = mode_to_o2dlm(mode); - int o2dlm_flags = flags_to_o2dlm(flags); - int ret; - - status = dlmlock(conn->cc_lockspace, o2dlm_mode, &lksb->lksb_o2dlm, - o2dlm_flags, name, namelen, - o2dlm_lock_ast_wrapper, astarg, - o2dlm_blocking_ast_wrapper); - ret = dlm_status_to_errno(status); - return ret; -} int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn, int mode, @@ -226,25 +39,10 @@ int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn, unsigned int namelen, void *astarg) { - BUG_ON(lproto == NULL); - - return o2cb_dlm_lock(conn, mode, lksb, flags, - name, namelen, astarg); -} - -static int o2cb_dlm_unlock(struct ocfs2_cluster_connection *conn, - union ocfs2_dlm_lksb *lksb, - u32 flags, - void *astarg) -{ - enum dlm_status status; - int o2dlm_flags = flags_to_o2dlm(flags); - int ret; + BUG_ON(stack_glue_lproto == NULL); - status = dlmunlock(conn->cc_lockspace, &lksb->lksb_o2dlm, - o2dlm_flags, o2dlm_unlock_ast_wrapper, astarg); - ret = dlm_status_to_errno(status); - return ret; + return o2cb_stack_ops.dlm_lock(conn, mode, lksb, flags, + name, namelen, astarg); } int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn, @@ -252,19 +50,14 @@ int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn, u32 flags, void *astarg) { - BUG_ON(lproto == NULL); + BUG_ON(stack_glue_lproto == NULL); - return o2cb_dlm_unlock(conn, lksb, flags, astarg); -} - -static int o2cb_dlm_lock_status(union ocfs2_dlm_lksb *lksb) -{ - return dlm_status_to_errno(lksb->lksb_o2dlm.status); + return o2cb_stack_ops.dlm_unlock(conn, lksb, flags, astarg); } int ocfs2_dlm_lock_status(union ocfs2_dlm_lksb *lksb) { - return o2cb_dlm_lock_status(lksb); + return o2cb_stack_ops.lock_status(lksb); } /* @@ -272,94 +65,14 @@ int ocfs2_dlm_lock_status(union ocfs2_dlm_lksb *lksb) * don't cast at the glue level. The real answer is that the header * ordering is nigh impossible. */ -static void *o2cb_dlm_lvb(union ocfs2_dlm_lksb *lksb) -{ - return (void *)(lksb->lksb_o2dlm.lvb); -} - void *ocfs2_dlm_lvb(union ocfs2_dlm_lksb *lksb) { - return o2cb_dlm_lvb(lksb); -} - -static void o2cb_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb) -{ - dlm_print_one_lock(lksb->lksb_o2dlm.lockid); + return o2cb_stack_ops.lock_lvb(lksb); } void ocfs2_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb) { - o2cb_dlm_dump_lksb(lksb); -} - -/* - * Called from the dlm when it's about to evict a node. This is how the - * classic stack signals node death. - */ -static void o2dlm_eviction_cb(int node_num, void *data) -{ - struct ocfs2_cluster_connection *conn = data; - - mlog(ML_NOTICE, "o2dlm has evicted node %d from group %.*s\n", - node_num, conn->cc_namelen, conn->cc_name); - - conn->cc_recovery_handler(node_num, conn->cc_recovery_data); -} - -static int o2cb_cluster_connect(struct ocfs2_cluster_connection *conn) -{ - int rc = 0; - u32 dlm_key; - struct dlm_ctxt *dlm; - struct o2dlm_private *priv; - struct dlm_protocol_version dlm_version; - - BUG_ON(conn == NULL); - - /* for now we only have one cluster/node, make sure we see it - * in the heartbeat universe */ - if (!o2hb_check_local_node_heartbeating()) { - rc = -EINVAL; - goto out; - } - - priv = kzalloc(sizeof(struct o2dlm_private), GFP_KERNEL); - if (!priv) { - rc = -ENOMEM; - goto out_free; - } - - /* This just fills the structure in. It is safe to pass conn. */ - dlm_setup_eviction_cb(&priv->op_eviction_cb, o2dlm_eviction_cb, - conn); - - conn->cc_private = priv; - - /* used by the dlm code to make message headers unique, each - * node in this domain must agree on this. */ - dlm_key = crc32_le(0, conn->cc_name, conn->cc_namelen); - dlm_version.pv_major = conn->cc_version.pv_major; - dlm_version.pv_minor = conn->cc_version.pv_minor; - - dlm = dlm_register_domain(conn->cc_name, dlm_key, &dlm_version); - if (IS_ERR(dlm)) { - rc = PTR_ERR(dlm); - mlog_errno(rc); - goto out_free; - } - - conn->cc_version.pv_major = dlm_version.pv_major; - conn->cc_version.pv_minor = dlm_version.pv_minor; - conn->cc_lockspace = dlm; - - dlm_register_eviction_cb(dlm, &priv->op_eviction_cb); - -out_free: - if (rc && conn->cc_private) - kfree(conn->cc_private); - -out: - return rc; + o2cb_stack_ops.dump_lksb(lksb); } int ocfs2_cluster_connect(const char *group, @@ -394,9 +107,9 @@ int ocfs2_cluster_connect(const char *group, new_conn->cc_recovery_data = recovery_data; /* Start the new connection at our maximum compatibility level */ - new_conn->cc_version = lproto->lp_max_version; + new_conn->cc_version = stack_glue_lproto->lp_max_version; - rc = o2cb_cluster_connect(new_conn); + rc = o2cb_stack_ops.connect(new_conn); if (rc) { mlog_errno(rc); goto out_free; @@ -412,29 +125,13 @@ out: return rc; } - -static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection *conn) -{ - struct dlm_ctxt *dlm = conn->cc_lockspace; - struct o2dlm_private *priv = conn->cc_private; - - dlm_unregister_eviction_cb(&priv->op_eviction_cb); - conn->cc_private = NULL; - kfree(priv); - - dlm_unregister_domain(dlm); - conn->cc_lockspace = NULL; - - return 0; -} - int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn) { int ret; BUG_ON(conn == NULL); - ret = o2cb_cluster_disconnect(conn); + ret = o2cb_stack_ops.disconnect(conn); /* XXX Should we free it anyway? */ if (!ret) @@ -443,75 +140,23 @@ int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn) return ret; } -static void o2hb_stop(const char *group) -{ - int ret; - char *argv[5], *envp[3]; - - argv[0] = (char *)o2nm_get_hb_ctl_path(); - argv[1] = "-K"; - argv[2] = "-u"; - argv[3] = (char *)group; - argv[4] = NULL; - - mlog(0, "Run: %s %s %s %s\n", argv[0], argv[1], argv[2], argv[3]); - - /* minimal command environment taken from cpu_run_sbin_hotplug */ - envp[0] = "HOME=/"; - envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin"; - envp[2] = NULL; - - ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC); - if (ret < 0) - mlog_errno(ret); -} - -/* - * Hangup is a hack for tools compatibility. Older ocfs2-tools software - * expects the filesystem to call "ocfs2_hb_ctl" during unmount. This - * happens regardless of whether the DLM got started, so we can't do it - * in ocfs2_cluster_disconnect(). We bring the o2hb_stop() function into - * the glue and provide a "hangup" API for super.c to call. - * - * Other stacks will eventually provide a NULL ->hangup() pointer. - */ -static void o2cb_cluster_hangup(const char *group, int grouplen) -{ - o2hb_stop(group); -} - void ocfs2_cluster_hangup(const char *group, int grouplen) { BUG_ON(group == NULL); BUG_ON(group[grouplen] != '\0'); - o2cb_cluster_hangup(group, grouplen); -} - -static int o2cb_cluster_this_node(unsigned int *node) -{ - int node_num; - - node_num = o2nm_this_node(); - if (node_num == O2NM_INVALID_NODE_NUM) - return -ENOENT; - - if (node_num >= O2NM_MAX_NODES) - return -EOVERFLOW; - - *node = node_num; - return 0; + o2cb_stack_ops.hangup(group, grouplen); } int ocfs2_cluster_this_node(unsigned int *node) { - return o2cb_cluster_this_node(node); + return o2cb_stack_ops.this_node(node); } void ocfs2_stack_glue_set_locking_protocol(struct ocfs2_locking_protocol *proto) { BUG_ON(proto != NULL); - lproto = proto; + stack_glue_lproto = proto; } diff --git a/fs/ocfs2/stackglue.h b/fs/ocfs2/stackglue.h index decb147106f..083632215dc 100644 --- a/fs/ocfs2/stackglue.h +++ b/fs/ocfs2/stackglue.h @@ -25,6 +25,8 @@ #include <linux/list.h> #include <linux/dlmconstants.h> +#include "dlm/dlmapi.h" + /* * dlmconstants.h does not have a LOCAL flag. We hope to remove it * some day, but right now we need it. Let's fake it. This value is larger @@ -39,13 +41,18 @@ #define GROUP_NAME_MAX 64 -#include "dlm/dlmapi.h" - +/* + * ocfs2_protocol_version changes when ocfs2 does something different in + * its inter-node behavior. See dlmglue.c for more information. + */ struct ocfs2_protocol_version { u8 pv_major; u8 pv_minor; }; +/* + * The ocfs2_locking_protocol defines the handlers called on ocfs2's behalf. + */ struct ocfs2_locking_protocol { struct ocfs2_protocol_version lp_max_version; void (*lp_lock_ast)(void *astarg); @@ -53,10 +60,20 @@ struct ocfs2_locking_protocol { void (*lp_unlock_ast)(void *astarg, int error); }; +/* + * A union of all lock status structures. We define it here so that the + * size of the union is known. Lock status structures are embedded in + * ocfs2 inodes. + */ union ocfs2_dlm_lksb { struct dlm_lockstatus lksb_o2dlm; }; +/* + * A cluster connection. Mostly opaque to ocfs2, the connection holds + * state for the underlying stack. ocfs2 does use cc_version to determine + * locking compatibility. + */ struct ocfs2_cluster_connection { char cc_name[GROUP_NAME_MAX]; int cc_namelen; @@ -67,6 +84,106 @@ struct ocfs2_cluster_connection { void *cc_private; }; +/* + * Each cluster stack implements the stack operations structure. Not used + * in the ocfs2 code, the stackglue code translates generic cluster calls + * into stack operations. + */ +struct ocfs2_stack_operations { + /* + * The fs code calls ocfs2_cluster_connect() to attach a new + * filesystem to the cluster stack. The ->connect() op is passed + * an ocfs2_cluster_connection with the name and recovery field + * filled in. + * + * The stack must set up any notification mechanisms and create + * the filesystem lockspace in the DLM. The lockspace should be + * stored on cc_lockspace. Any other information can be stored on + * cc_private. + * + * ->connect() must not return until it is guaranteed that + * + * - Node down notifications for the filesystem will be recieved + * and passed to conn->cc_recovery_handler(). + * - Locking requests for the filesystem will be processed. + */ + int (*connect)(struct ocfs2_cluster_connection *conn); + + /* + * The fs code calls ocfs2_cluster_disconnect() when a filesystem + * no longer needs cluster services. All DLM locks have been + * dropped, and recovery notification is being ignored by the + * fs code. The stack must disengage from the DLM and discontinue + * recovery notification. + * + * Once ->disconnect() has returned, the connection structure will + * be freed. Thus, a stack must not return from ->disconnect() + * until it will no longer reference the conn pointer. + */ + int (*disconnect)(struct ocfs2_cluster_connection *conn); + + /* + * ocfs2_cluster_hangup() exists for compatibility with older + * ocfs2 tools. Only the classic stack really needs it. As such + * ->hangup() is not required of all stacks. See the comment by + * ocfs2_cluster_hangup() for more details. + */ + void (*hangup)(const char *group, int grouplen); + + /* + * ->this_node() returns the cluster's unique identifier for the + * local node. + */ + int (*this_node)(unsigned int *node); + + /* + * Call the underlying dlm lock function. The ->dlm_lock() + * callback should convert the flags and mode as appropriate. + * + * ast and bast functions are not part of the call because the + * stack will likely want to wrap ast and bast calls before passing + * them to stack->sp_proto. + */ + int (*dlm_lock)(struct ocfs2_cluster_connection *conn, + int mode, + union ocfs2_dlm_lksb *lksb, + u32 flags, + void *name, + unsigned int namelen, + void *astarg); + + /* + * Call the underlying dlm unlock function. The ->dlm_unlock() + * function should convert the flags as appropriate. + * + * The unlock ast is not passed, as the stack will want to wrap + * it before calling stack->sp_proto->lp_unlock_ast(). + */ + int (*dlm_unlock)(struct ocfs2_cluster_connection *conn, + union ocfs2_dlm_lksb *lksb, + u32 flags, + void *astarg); + + /* + * Return the status of the current lock status block. The fs + * code should never dereference the union. The ->lock_status() + * callback pulls out the stack-specific lksb, converts the status + * to a proper errno, and returns it. + */ + int (*lock_status)(union ocfs2_dlm_lksb *lksb); + + /* + * Pull the lvb pointer off of the stack-specific lksb. + */ + void *(*lock_lvb)(union ocfs2_dlm_lksb *lksb); + + /* + * This is an optoinal debugging hook. If provided, the + * stack can dump debugging information about this lock. + */ + void (*dump_lksb)(union ocfs2_dlm_lksb *lksb); +}; + int ocfs2_cluster_connect(const char *group, int grouplen, void (*recovery_handler)(int node_num, @@ -95,4 +212,6 @@ void ocfs2_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb); void ocfs2_stack_glue_set_locking_protocol(struct ocfs2_locking_protocol *proto); +extern struct ocfs2_locking_protocol *stack_glue_lproto; +extern struct ocfs2_stack_operations o2cb_stack_ops; #endif /* STACKGLUE_H */ |