summaryrefslogtreecommitdiffstats
path: root/fs/ocfs2/journal.c
diff options
context:
space:
mode:
authorSunil Mushran <sunil.mushran@oracle.com>2008-07-14 17:31:10 -0700
committerMark Fasheh <mfasheh@suse.com>2008-07-31 16:21:14 -0700
commit539d8264093560b917ee3afe4c7f74e5da09d6a5 (patch)
tree1fce83387272c0b2d61bd945769f4984aa5e79ce /fs/ocfs2/journal.c
parentc69991aac71a8beb57c11d651c7fd4b24c32aa8b (diff)
downloadop-kernel-dev-539d8264093560b917ee3afe4c7f74e5da09d6a5.zip
op-kernel-dev-539d8264093560b917ee3afe4c7f74e5da09d6a5.tar.gz
[PATCH 2/2] ocfs2: Fix race between mount and recovery
As the fs recovery is asynchronous, there is a small chance that another node can mount (and thus recover) the slot before the recovery thread gets to it. If this happens, the recovery thread will block indefinitely on the journal/slot lock as that lock will be held for the duration of the mount (by design) by the node assigned to that slot. The solution implemented is to keep track of the journal replays using a recovery generation in the journal inode, which will be incremented by the thread replaying that journal. The recovery thread, before attempting the blocking lock on the journal/slot lock, will compare the generation on disk with what it has cached and skip recovery if it does not match. This bug appears to have been inadvertently introduced during the mount/umount vote removal by mainline commit 34d024f84345807bf44163fac84e921513dde323. In the mount voting scheme, the messaging would indirectly indicate that the slot was being recovered. Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com> Signed-off-by: Mark Fasheh <mfasheh@suse.com>
Diffstat (limited to 'fs/ocfs2/journal.c')
-rw-r--r--fs/ocfs2/journal.c173
1 files changed, 133 insertions, 40 deletions
diff --git a/fs/ocfs2/journal.c b/fs/ocfs2/journal.c
index a8c19cb..7a37240 100644
--- a/fs/ocfs2/journal.c
+++ b/fs/ocfs2/journal.c
@@ -57,7 +57,7 @@ static int __ocfs2_recovery_thread(void *arg);
static int ocfs2_commit_cache(struct ocfs2_super *osb);
static int ocfs2_wait_on_mount(struct ocfs2_super *osb);
static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
- int dirty);
+ int dirty, int replayed);
static int ocfs2_trylock_journal(struct ocfs2_super *osb,
int slot_num);
static int ocfs2_recover_orphans(struct ocfs2_super *osb,
@@ -562,8 +562,18 @@ done:
return status;
}
+static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
+{
+ le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
+}
+
+static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
+{
+ return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
+}
+
static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
- int dirty)
+ int dirty, int replayed)
{
int status;
unsigned int flags;
@@ -593,6 +603,9 @@ static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
flags &= ~OCFS2_JOURNAL_DIRTY_FL;
fe->id1.journal1.ij_flags = cpu_to_le32(flags);
+ if (replayed)
+ ocfs2_bump_recovery_generation(fe);
+
status = ocfs2_write_block(osb, bh, journal->j_inode);
if (status < 0)
mlog_errno(status);
@@ -667,7 +680,7 @@ void ocfs2_journal_shutdown(struct ocfs2_super *osb)
* Do not toggle if flush was unsuccessful otherwise
* will leave dirty metadata in a "clean" journal
*/
- status = ocfs2_journal_toggle_dirty(osb, 0);
+ status = ocfs2_journal_toggle_dirty(osb, 0, 0);
if (status < 0)
mlog_errno(status);
}
@@ -710,7 +723,7 @@ static void ocfs2_clear_journal_error(struct super_block *sb,
}
}
-int ocfs2_journal_load(struct ocfs2_journal *journal, int local)
+int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
{
int status = 0;
struct ocfs2_super *osb;
@@ -729,7 +742,7 @@ int ocfs2_journal_load(struct ocfs2_journal *journal, int local)
ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
- status = ocfs2_journal_toggle_dirty(osb, 1);
+ status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
if (status < 0) {
mlog_errno(status);
goto done;
@@ -771,7 +784,7 @@ int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
goto bail;
}
- status = ocfs2_journal_toggle_dirty(journal->j_osb, 0);
+ status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
if (status < 0)
mlog_errno(status);
@@ -1034,6 +1047,12 @@ restart:
spin_unlock(&osb->osb_lock);
mlog(0, "All nodes recovered\n");
+ /* Refresh all journal recovery generations from disk */
+ status = ocfs2_check_journals_nolocks(osb);
+ status = (status == -EROFS) ? 0 : status;
+ if (status < 0)
+ mlog_errno(status);
+
ocfs2_super_unlock(osb, 1);
/* We always run recovery on our own orphan dir - the dead
@@ -1096,6 +1115,42 @@ out:
mlog_exit_void();
}
+static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
+ int slot_num,
+ struct buffer_head **bh,
+ struct inode **ret_inode)
+{
+ int status = -EACCES;
+ struct inode *inode = NULL;
+
+ BUG_ON(slot_num >= osb->max_slots);
+
+ inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
+ slot_num);
+ if (!inode || is_bad_inode(inode)) {
+ mlog_errno(status);
+ goto bail;
+ }
+ SET_INODE_JOURNAL(inode);
+
+ status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, bh, 0, inode);
+ if (status < 0) {
+ mlog_errno(status);
+ goto bail;
+ }
+
+ status = 0;
+
+bail:
+ if (inode) {
+ if (status || !ret_inode)
+ iput(inode);
+ else
+ *ret_inode = inode;
+ }
+ return status;
+}
+
/* Does the actual journal replay and marks the journal inode as
* clean. Will only replay if the journal inode is marked dirty. */
static int ocfs2_replay_journal(struct ocfs2_super *osb,
@@ -1109,22 +1164,36 @@ static int ocfs2_replay_journal(struct ocfs2_super *osb,
struct ocfs2_dinode *fe;
journal_t *journal = NULL;
struct buffer_head *bh = NULL;
+ u32 slot_reco_gen;
- inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
- slot_num);
- if (inode == NULL) {
- status = -EACCES;
+ status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
+ if (status) {
mlog_errno(status);
goto done;
}
- if (is_bad_inode(inode)) {
- status = -EACCES;
- iput(inode);
- inode = NULL;
- mlog_errno(status);
+
+ fe = (struct ocfs2_dinode *)bh->b_data;
+ slot_reco_gen = ocfs2_get_recovery_generation(fe);
+ brelse(bh);
+ bh = NULL;
+
+ /*
+ * As the fs recovery is asynchronous, there is a small chance that
+ * another node mounted (and recovered) the slot before the recovery
+ * thread could get the lock. To handle that, we dirty read the journal
+ * inode for that slot to get the recovery generation. If it is
+ * different than what we expected, the slot has been recovered.
+ * If not, it needs recovery.
+ */
+ if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
+ mlog(0, "Slot %u already recovered (old/new=%u/%u)\n", slot_num,
+ osb->slot_recovery_generations[slot_num], slot_reco_gen);
+ osb->slot_recovery_generations[slot_num] = slot_reco_gen;
+ status = -EBUSY;
goto done;
}
- SET_INODE_JOURNAL(inode);
+
+ /* Continue with recovery as the journal has not yet been recovered */
status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
if (status < 0) {
@@ -1138,9 +1207,12 @@ static int ocfs2_replay_journal(struct ocfs2_super *osb,
fe = (struct ocfs2_dinode *) bh->b_data;
flags = le32_to_cpu(fe->id1.journal1.ij_flags);
+ slot_reco_gen = ocfs2_get_recovery_generation(fe);
if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
mlog(0, "No recovery required for node %d\n", node_num);
+ /* Refresh recovery generation for the slot */
+ osb->slot_recovery_generations[slot_num] = slot_reco_gen;
goto done;
}
@@ -1188,6 +1260,11 @@ static int ocfs2_replay_journal(struct ocfs2_super *osb,
flags &= ~OCFS2_JOURNAL_DIRTY_FL;
fe->id1.journal1.ij_flags = cpu_to_le32(flags);
+ /* Increment recovery generation to indicate successful recovery */
+ ocfs2_bump_recovery_generation(fe);
+ osb->slot_recovery_generations[slot_num] =
+ ocfs2_get_recovery_generation(fe);
+
status = ocfs2_write_block(osb, bh, inode);
if (status < 0)
mlog_errno(status);
@@ -1252,6 +1329,13 @@ static int ocfs2_recover_node(struct ocfs2_super *osb,
status = ocfs2_replay_journal(osb, node_num, slot_num);
if (status < 0) {
+ if (status == -EBUSY) {
+ mlog(0, "Skipping recovery for slot %u (node %u) "
+ "as another node has recovered it\n", slot_num,
+ node_num);
+ status = 0;
+ goto done;
+ }
mlog_errno(status);
goto done;
}
@@ -1334,12 +1418,29 @@ int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
{
unsigned int node_num;
int status, i;
+ struct buffer_head *bh = NULL;
+ struct ocfs2_dinode *di;
/* This is called with the super block cluster lock, so we
* know that the slot map can't change underneath us. */
spin_lock(&osb->osb_lock);
for (i = 0; i < osb->max_slots; i++) {
+ /* Read journal inode to get the recovery generation */
+ status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
+ if (status) {
+ mlog_errno(status);
+ goto bail;
+ }
+ di = (struct ocfs2_dinode *)bh->b_data;
+ osb->slot_recovery_generations[i] =
+ ocfs2_get_recovery_generation(di);
+ brelse(bh);
+ bh = NULL;
+
+ mlog(0, "Slot %u recovery generation is %u\n", i,
+ osb->slot_recovery_generations[i]);
+
if (i == osb->slot_num)
continue;
@@ -1603,49 +1704,41 @@ static int ocfs2_commit_thread(void *arg)
return 0;
}
-/* Look for a dirty journal without taking any cluster locks. Used for
- * hard readonly access to determine whether the file system journals
- * require recovery. */
+/* Reads all the journal inodes without taking any cluster locks. Used
+ * for hard readonly access to determine whether any journal requires
+ * recovery. Also used to refresh the recovery generation numbers after
+ * a journal has been recovered by another node.
+ */
int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
{
int ret = 0;
unsigned int slot;
- struct buffer_head *di_bh;
+ struct buffer_head *di_bh = NULL;
struct ocfs2_dinode *di;
- struct inode *journal = NULL;
+ int journal_dirty = 0;
for(slot = 0; slot < osb->max_slots; slot++) {
- journal = ocfs2_get_system_file_inode(osb,
- JOURNAL_SYSTEM_INODE,
- slot);
- if (!journal || is_bad_inode(journal)) {
- ret = -EACCES;
- mlog_errno(ret);
- goto out;
- }
-
- di_bh = NULL;
- ret = ocfs2_read_block(osb, OCFS2_I(journal)->ip_blkno, &di_bh,
- 0, journal);
- if (ret < 0) {
+ ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
+ if (ret) {
mlog_errno(ret);
goto out;
}
di = (struct ocfs2_dinode *) di_bh->b_data;
+ osb->slot_recovery_generations[slot] =
+ ocfs2_get_recovery_generation(di);
+
if (le32_to_cpu(di->id1.journal1.ij_flags) &
OCFS2_JOURNAL_DIRTY_FL)
- ret = -EROFS;
+ journal_dirty = 1;
brelse(di_bh);
- if (ret)
- break;
+ di_bh = NULL;
}
out:
- if (journal)
- iput(journal);
-
+ if (journal_dirty)
+ ret = -EROFS;
return ret;
}
OpenPOWER on IntegriCloud