diff options
author | Mark Fasheh <mark.fasheh@oracle.com> | 2006-02-22 17:35:08 -0800 |
---|---|---|
committer | Mark Fasheh <mark.fasheh@oracle.com> | 2006-03-01 11:32:41 -0800 |
commit | b4df6ed8db0c387d38292e31f00adc4cd297ed5a (patch) | |
tree | d7dcaeecfa55b3fd9d6c4844d90e07759182f845 /fs/ocfs2/inode.c | |
parent | 895928b8380cc697ac56e9732cedf549c0a4f79c (diff) |
[PATCH] ocfs2: fix orphan recovery deadlock
Orphan dir recovery can deadlock with another process in
ocfs2_delete_inode() in some corner cases. Fix this by tracking recovery
state more closely and allowing it to handle inode wipes which might
deadlock.
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
Diffstat (limited to 'fs/ocfs2/inode.c')
-rw-r--r-- | fs/ocfs2/inode.c | 46 |
1 files changed, 45 insertions, 1 deletions
diff --git a/fs/ocfs2/inode.c b/fs/ocfs2/inode.c index 8122489c5762..315472a5c192 100644 --- a/fs/ocfs2/inode.c +++ b/fs/ocfs2/inode.c @@ -41,6 +41,7 @@ #include "dlmglue.h" #include "extent_map.h" #include "file.h" +#include "heartbeat.h" #include "inode.h" #include "journal.h" #include "namei.h" @@ -544,6 +545,42 @@ bail: return status; } +/* + * Serialize with orphan dir recovery. If the process doing + * recovery on this orphan dir does an iget() with the dir + * i_mutex held, we'll deadlock here. Instead we detect this + * and exit early - recovery will wipe this inode for us. + */ +static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb, + int slot) +{ + int ret = 0; + + spin_lock(&osb->osb_lock); + if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) { + mlog(0, "Recovery is happening on orphan dir %d, will skip " + "this inode\n", slot); + ret = -EDEADLK; + goto out; + } + /* This signals to the orphan recovery process that it should + * wait for us to handle the wipe. */ + osb->osb_orphan_wipes[slot]++; +out: + spin_unlock(&osb->osb_lock); + return ret; +} + +static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb, + int slot) +{ + spin_lock(&osb->osb_lock); + osb->osb_orphan_wipes[slot]--; + spin_unlock(&osb->osb_lock); + + wake_up(&osb->osb_wipe_event); +} + static int ocfs2_wipe_inode(struct inode *inode, struct buffer_head *di_bh) { @@ -555,6 +592,11 @@ static int ocfs2_wipe_inode(struct inode *inode, /* We've already voted on this so it should be readonly - no * spinlock needed. */ orphaned_slot = OCFS2_I(inode)->ip_orphaned_slot; + + status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot); + if (status) + return status; + orphan_dir_inode = ocfs2_get_system_file_inode(osb, ORPHAN_DIR_SYSTEM_INODE, orphaned_slot); @@ -597,6 +639,7 @@ bail_unlock_dir: brelse(orphan_dir_bh); bail: iput(orphan_dir_inode); + ocfs2_signal_wipe_completion(osb, orphaned_slot); return status; } @@ -822,7 +865,8 @@ void ocfs2_delete_inode(struct inode *inode) status = ocfs2_wipe_inode(inode, di_bh); if (status < 0) { - mlog_errno(status); + if (status != -EDEADLK) + mlog_errno(status); goto bail_unlock_inode; } |