unsigned long mw_goal;
};
-static void ocfs2_inode_ast_func(void *opaque);
static void ocfs2_inode_bast_func(void *opaque,
int level);
-static void ocfs2_dentry_ast_func(void *opaque);
static void ocfs2_dentry_bast_func(void *opaque,
int level);
-static void ocfs2_super_ast_func(void *opaque);
static void ocfs2_super_bast_func(void *opaque,
int level);
-static void ocfs2_rename_ast_func(void *opaque);
static void ocfs2_rename_bast_func(void *opaque,
int level);
enum ocfs2_unblock_action unblock_action;
};
-/* so far, all locks have gotten along with the same unlock ast */
-static void ocfs2_unlock_ast_func(void *opaque,
- enum dlm_status status);
static int ocfs2_unblock_meta(struct ocfs2_lock_res *lockres,
struct ocfs2_unblock_ctl *ctl);
static int ocfs2_unblock_data(struct ocfs2_lock_res *lockres,
static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
struct ocfs2_lock_res *lockres);
+/*
+ * OCFS2 Lock Resource Operations
+ *
+ * These fine tune the behavior of the generic dlmglue locking infrastructure.
+ */
struct ocfs2_lock_res_ops {
- void (*ast)(void *);
void (*bast)(void *, int);
- void (*unlock_ast)(void *, enum dlm_status);
int (*unblock)(struct ocfs2_lock_res *, struct ocfs2_unblock_ctl *);
void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *);
+
+ /*
+ * LOCK_TYPE_* flags which describe the specific requirements
+ * of a lock type. Descriptions of each individual flag follow.
+ */
+ int flags;
};
+/*
+ * Some locks want to "refresh" potentially stale data when a
+ * meaningful (PRMODE or EXMODE) lock level is first obtained. If this
+ * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the
+ * individual lockres l_flags member from the ast function. It is
+ * expected that the locking wrapper will clear the
+ * OCFS2_LOCK_NEEDS_REFRESH flag when done.
+ */
+#define LOCK_TYPE_REQUIRES_REFRESH 0x1
+
typedef int (ocfs2_convert_worker_t)(struct ocfs2_lock_res *, int);
static int ocfs2_generic_unblock_lock(struct ocfs2_super *osb,
struct ocfs2_lock_res *lockres,
ocfs2_convert_worker_t *worker);
static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
- .ast = ocfs2_inode_ast_func,
.bast = ocfs2_inode_bast_func,
- .unlock_ast = ocfs2_unlock_ast_func,
.unblock = ocfs2_unblock_inode_lock,
+ .flags = 0,
};
static struct ocfs2_lock_res_ops ocfs2_inode_meta_lops = {
- .ast = ocfs2_inode_ast_func,
.bast = ocfs2_inode_bast_func,
- .unlock_ast = ocfs2_unlock_ast_func,
.unblock = ocfs2_unblock_meta,
+ .flags = LOCK_TYPE_REQUIRES_REFRESH,
};
static struct ocfs2_lock_res_ops ocfs2_inode_data_lops = {
- .ast = ocfs2_inode_ast_func,
.bast = ocfs2_inode_bast_func,
- .unlock_ast = ocfs2_unlock_ast_func,
.unblock = ocfs2_unblock_data,
+ .flags = 0,
};
static struct ocfs2_lock_res_ops ocfs2_super_lops = {
- .ast = ocfs2_super_ast_func,
.bast = ocfs2_super_bast_func,
- .unlock_ast = ocfs2_unlock_ast_func,
.unblock = ocfs2_unblock_osb_lock,
+ .flags = LOCK_TYPE_REQUIRES_REFRESH,
};
static struct ocfs2_lock_res_ops ocfs2_rename_lops = {
- .ast = ocfs2_rename_ast_func,
.bast = ocfs2_rename_bast_func,
- .unlock_ast = ocfs2_unlock_ast_func,
.unblock = ocfs2_unblock_osb_lock,
+ .flags = 0,
};
static struct ocfs2_lock_res_ops ocfs2_dentry_lops = {
- .ast = ocfs2_dentry_ast_func,
.bast = ocfs2_dentry_bast_func,
- .unlock_ast = ocfs2_unlock_ast_func,
.unblock = ocfs2_unblock_dentry_lock,
.post_unlock = ocfs2_dentry_post_unlock,
+ .flags = 0,
};
static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
enum ocfs2_lock_type type,
+ unsigned int generation,
struct inode *inode)
{
struct ocfs2_lock_res_ops *ops;
};
ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno,
- inode->i_generation, res->l_name);
+ generation, res->l_name);
ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode);
}
* 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 == LKM_NLMODE &&
+ lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
lockres->l_level = lockres->l_requested;
BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
if (lockres->l_requested > LKM_NLMODE &&
- !(lockres->l_flags & OCFS2_LOCK_LOCAL))
+ !(lockres->l_flags & OCFS2_LOCK_LOCAL) &&
+ lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
lockres->l_level = lockres->l_requested;
mlog_exit_void();
}
-static void ocfs2_inode_ast_func(void *opaque)
-{
- struct ocfs2_lock_res *lockres = opaque;
- struct inode *inode;
- struct dlm_lockstatus *lksb;
- unsigned long flags;
-
- mlog_entry_void();
-
- inode = ocfs2_lock_res_inode(lockres);
-
- mlog(0, "AST fired for inode %llu, l_action = %u, type = %s\n",
- (unsigned long long)OCFS2_I(inode)->ip_blkno, lockres->l_action,
- ocfs2_lock_type_string(lockres->l_type));
-
- BUG_ON(!ocfs2_is_inode_lock(lockres));
-
- spin_lock_irqsave(&lockres->l_lock, flags);
-
- lksb = &(lockres->l_lksb);
- if (lksb->status != DLM_NORMAL) {
- mlog(ML_ERROR, "ocfs2_inode_ast_func: lksb status value of %u "
- "on inode %llu\n", lksb->status,
- (unsigned long long)OCFS2_I(inode)->ip_blkno);
- spin_unlock_irqrestore(&lockres->l_lock, flags);
- mlog_exit_void();
- return;
- }
-
- switch(lockres->l_action) {
- case OCFS2_AST_ATTACH:
- ocfs2_generic_handle_attach_action(lockres);
- lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
- break;
- case OCFS2_AST_CONVERT:
- ocfs2_generic_handle_convert_action(lockres);
- break;
- case OCFS2_AST_DOWNCONVERT:
- 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",
- lockres->l_name, lockres->l_action, lockres->l_flags,
- lockres->l_unlock_action);
-
- BUG();
- }
-
- /* data and rw locking ignores refresh flag for now. */
- if (lockres->l_type != OCFS2_LOCK_TYPE_META)
- lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
-
- /* set it to something invalid so if we get called again we
- * can catch it. */
- lockres->l_action = OCFS2_AST_INVALID;
- spin_unlock_irqrestore(&lockres->l_lock, flags);
- wake_up(&lockres->l_event);
-
- mlog_exit_void();
-}
-
static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
int level)
{
mlog_exit_void();
}
-static void ocfs2_generic_ast_func(struct ocfs2_lock_res *lockres,
- int ignore_refresh)
+static void ocfs2_locking_ast(void *opaque)
{
+ struct ocfs2_lock_res *lockres = opaque;
struct dlm_lockstatus *lksb = &lockres->l_lksb;
unsigned long flags;
switch(lockres->l_action) {
case OCFS2_AST_ATTACH:
ocfs2_generic_handle_attach_action(lockres);
+ lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
break;
case OCFS2_AST_CONVERT:
ocfs2_generic_handle_convert_action(lockres);
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",
+ lockres->l_name, lockres->l_action, lockres->l_flags,
+ lockres->l_unlock_action);
BUG();
}
- if (ignore_refresh)
- lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
-
/* set it to something invalid so if we get called again we
* can catch it. */
lockres->l_action = OCFS2_AST_INVALID;
spin_unlock_irqrestore(&lockres->l_lock, flags);
}
-static void ocfs2_super_ast_func(void *opaque)
-{
- struct ocfs2_lock_res *lockres = opaque;
-
- mlog_entry_void();
- mlog(0, "Superblock AST fired\n");
-
- BUG_ON(!ocfs2_is_super_lock(lockres));
- ocfs2_generic_ast_func(lockres, 0);
-
- mlog_exit_void();
-}
-
static void ocfs2_super_bast_func(void *opaque,
int level)
{
mlog_exit_void();
}
-static void ocfs2_rename_ast_func(void *opaque)
-{
- struct ocfs2_lock_res *lockres = opaque;
-
- mlog_entry_void();
-
- mlog(0, "Rename AST fired\n");
-
- BUG_ON(!ocfs2_is_rename_lock(lockres));
-
- ocfs2_generic_ast_func(lockres, 1);
-
- mlog_exit_void();
-}
-
static void ocfs2_rename_bast_func(void *opaque,
int level)
{
mlog_exit_void();
}
-static void ocfs2_dentry_ast_func(void *opaque)
-{
- struct ocfs2_lock_res *lockres = opaque;
-
- BUG_ON(!lockres);
-
- ocfs2_generic_ast_func(lockres, 1);
-}
-
static void ocfs2_dentry_bast_func(void *opaque, int level)
{
struct ocfs2_lock_res *lockres = opaque;
dlm_flags,
lockres->l_name,
OCFS2_LOCK_ID_MAX_LEN - 1,
- lockres->l_ops->ast,
+ ocfs2_locking_ast,
lockres,
lockres->l_ops->bast);
if (status != DLM_NORMAL) {
lkm_flags|LKM_CONVERT|LKM_VALBLK,
lockres->l_name,
OCFS2_LOCK_ID_MAX_LEN - 1,
- lockres->l_ops->ast,
+ ocfs2_locking_ast,
lockres,
lockres->l_ops->bast);
if (status != DLM_NORMAL) {
int ocfs2_create_new_lock(struct ocfs2_super *osb,
struct ocfs2_lock_res *lockres,
- int ex)
+ int ex,
+ int local)
{
int level = ex ? LKM_EXMODE : LKM_PRMODE;
unsigned long flags;
+ int lkm_flags = local ? LKM_LOCAL : 0;
spin_lock_irqsave(&lockres->l_lock, flags);
BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
spin_unlock_irqrestore(&lockres->l_lock, flags);
- return ocfs2_lock_create(osb, lockres, level, LKM_LOCAL);
+ return ocfs2_lock_create(osb, lockres, level, lkm_flags);
}
/* Grants us an EX lock on the data and metadata resources, skipping
* on a resource which has an invalid one -- we'll set it
* valid when we release the EX. */
- ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1);
+ ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1);
if (ret) {
mlog_errno(ret);
goto bail;
}
- ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_meta_lockres, 1);
+ /*
+ * We don't want to use LKM_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_meta_lockres, 1, 0);
if (ret) {
mlog_errno(ret);
goto bail;
}
- ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_data_lockres, 1);
+ ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_data_lockres, 1, 1);
if (ret) {
mlog_errno(ret);
goto bail;
lvb = (struct ocfs2_meta_lvb *) lockres->l_lksb.lvb;
- lvb->lvb_version = cpu_to_be32(OCFS2_LVB_VERSION);
+ /*
+ * Invalidate the LVB of a deleted inode - this way other
+ * nodes are forced to go to disk and discover the new inode
+ * status.
+ */
+ if (oi->ip_flags & OCFS2_INODE_DELETED) {
+ lvb->lvb_version = 0;
+ goto out;
+ }
+
+ 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_imtime_packed =
cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime));
lvb->lvb_iattr = cpu_to_be32(oi->ip_attr);
+ lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
+out:
mlog_meta_lvb(0, lockres);
mlog_exit_void();
mlog_exit_void();
}
-static inline int ocfs2_meta_lvb_is_trustable(struct ocfs2_lock_res *lockres)
+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;
- if (be32_to_cpu(lvb->lvb_version) == OCFS2_LVB_VERSION)
+ if (lvb->lvb_version == OCFS2_LVB_VERSION
+ && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
return 1;
return 0;
}
* map (directories, bitmap files, etc) */
ocfs2_extent_map_trunc(inode, 0);
- if (ocfs2_meta_lvb_is_trustable(lockres)) {
+ if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
mlog(0, "Trusting LVB on inode %llu\n",
(unsigned long long)oi->ip_blkno);
ocfs2_refresh_inode_from_lvb(inode);
wait_event(osb->recovery_event,
ocfs2_node_map_is_empty(osb, &osb->recovery_map));
+ /*
+ * We only see this flag if we're being called from
+ * ocfs2_read_locked_inode(). It means we're locking an inode
+ * which hasn't been populated yet, so clear the refresh flag
+ * and let the caller handle it.
+ */
+ if (inode->i_state & I_NEW) {
+ status = 0;
+ ocfs2_complete_lock_res_refresh(lockres, 0);
+ goto bail;
+ }
+
/* This is fun. The caller may want a bh back, or it may
* not. ocfs2_meta_lock_update definitely wants one in, but
* may or may not read one, depending on what's in the
mlog_exit_void();
}
-static void ocfs2_unlock_ast_func(void *opaque, enum dlm_status status)
+static void ocfs2_unlock_ast(void *opaque, enum dlm_status status)
{
struct ocfs2_lock_res *lockres = opaque;
unsigned long flags;
mlog(0, "lock %s\n", lockres->l_name);
status = dlmunlock(osb->dlm, &lockres->l_lksb, LKM_VALBLK,
- lockres->l_ops->unlock_ast, lockres);
+ ocfs2_unlock_ast, lockres);
if (status != DLM_NORMAL) {
ocfs2_log_dlm_error("dlmunlock", status, lockres);
mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
dlm_flags,
lockres->l_name,
OCFS2_LOCK_ID_MAX_LEN - 1,
- lockres->l_ops->ast,
+ ocfs2_locking_ast,
lockres,
lockres->l_ops->bast);
if (status != DLM_NORMAL) {
status = dlmunlock(osb->dlm,
&lockres->l_lksb,
LKM_CANCEL,
- lockres->l_ops->unlock_ast,
+ ocfs2_unlock_ast,
lockres);
if (status != DLM_NORMAL) {
ocfs2_log_dlm_error("dlmunlock", status, lockres);
mlog(level, "LVB information for %s (called from %s:%u):\n",
lockres->l_name, function, line);
- mlog(level, "version: %u, clusters: %u\n",
- be32_to_cpu(lvb->lvb_version), be32_to_cpu(lvb->lvb_iclusters));
+ mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
+ lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
+ be32_to_cpu(lvb->lvb_igeneration));
mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
(unsigned long long)be64_to_cpu(lvb->lvb_isize),
be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),