* [PATCH] ocfs2: unregister the JBD2 shrinker before journal shutdown
@ 2026-11-22 17:39 Andrew Morton
0 siblings, 0 replies; only message in thread
From: Andrew Morton @ 2026-11-22 17:39 UTC (permalink / raw)
To: Joseph Qi
Cc: ocfs2-devel, Andrew Morton, syzbot+38fdbb551940270021f1,
Cen Zhang, Mark Fasheh, Joel Becker, Heming Zhao,
Theodore Ts'o, Jan Kara, stable
From: Andrew Morton <akpm@linux-foundation.org>
Subject: ocfs2: unregister the JBD2 shrinker before journal shutdown
Date: Sat, 21 Nov 2026 19:34:56 -0800
sysbot reports a lockdep cycle through shrinker_mutex, filesystem reclaim
and osb->journal_shutdown_sem. Unmount holds the superblock's s_umount
exclusively, while the OCFS2 reclaim path must first get its read side
with super_trylock_shared(). That excludes the two paths for the same
superblock; the report does not establish a runtime deadlock. Still, the
lock ordering is visible to lockdep and produces the warning.
The write side of journal_shutdown_sem must cover all of
jbd2_journal_destroy(): ocfs2_clear_inode() can otherwise release a JBD2
inode through a journal that is being destroyed. Move only the JBD2
shrinker unregister operation before taking the semaphore.
shrinker_free() waits for active shrinker users, so the journal and its
checkpoint counter remain alive while they finish. A successfully
initialized JBD2 journal always has a checkpoint counter. Destroy it
after unregistering the shrinker, whether unregistering happened here or
before taking the OCFS2 semaphore. The semaphore continues to protect
JBD2 destruction, removal of osb->journal and freeing the OCFS2 journal.
The userspace-visible result is removal of this lockdep warning during
OCFS2 unmount; the report does not demonstrate a hung unmount.
Fixes: f277230d002d ("ocfs2: serialize inode journal access with shutdown")
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reported-by: <syzbot+38fdbb551940270021f1@syzkaller.appspotmail.com>
Closes: https://syzkaller.appspot.com/bug?extid=38fdbb551940270021f1
Link: https://lore.kernel.org/6b0257ca.6ccd1e40.868ea.001a.GAE@google.com
Tested-by: Andrew Morton <akpm@linux-foundation.org>
Link: https://lore.kernel.org/20261122093400.b726ecddf154f493dec1c892@linux-foundation.org
Cc: Cen Zhang <zzzccc427@gmail.com>
Cc: Mark Fasheh <mark@fasheh.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Joseph Qi <joseph.qi@linux.alibaba.com>
Cc: Heming Zhao <heming.zhao@suse.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Jan Kara <jack@suse.cz>
Cc: <stable@vger.kernel.org>
---
Tested with CONFIG_PROVE_LOCKING=y in x86-64 QEMU guests. A omitted
this patch; B added it alone. OCFS2 and JBD2 were built in.
Syzbot's C reproducer mounted a local OCFS2 image, accessed 200 files,
exercised reclaim and unmounted it. A reproduced the reported circular
locking warning: shrinker_free() ran from jbd2_journal_destroy() while
ocfs2_journal_shutdown() held journal_shutdown_sem. The unmount
completed; this test did not show a runtime deadlock.
B completed the same sequence without a warning. Three further
mount/reclaim/unmount cycles on B also completed without warnings.
An ext4 read/write/unmount control passed on both kernels.
Both builds needed the same temporary correction for an unrelated
mm/util.c compile error in another queued patch. The correction was
removed after testing. The test did not force an active JBD2 shrinker
callback to overlap unmount or exercise failed-mount cleanup.
fs/jbd2/journal.c | 17 +++++++++++++----
fs/ocfs2/journal.c | 1 +
include/linux/jbd2.h | 1 +
3 files changed, 15 insertions(+), 4 deletions(-)
--- a/fs/jbd2/journal.c~ocfs2-unregister-the-jbd2-shrinker-before-journal-shutdown
+++ a/fs/jbd2/journal.c
@@ -75,6 +75,7 @@ EXPORT_SYMBOL(jbd2_journal_check_availab
EXPORT_SYMBOL(jbd2_journal_set_features);
EXPORT_SYMBOL(jbd2_journal_load);
EXPORT_SYMBOL(jbd2_journal_destroy);
+EXPORT_SYMBOL(jbd2_journal_unregister_shrinker);
EXPORT_SYMBOL(jbd2_journal_abort);
EXPORT_SYMBOL(jbd2_journal_errno);
EXPORT_SYMBOL(jbd2_journal_ack_err);
@@ -2100,6 +2101,16 @@ int jbd2_journal_load(journal_t *journal
return 0;
}
+void jbd2_journal_unregister_shrinker(journal_t *journal)
+{
+ struct shrinker *shrinker = journal->j_shrinker;
+
+ if (!shrinker)
+ return;
+ journal->j_shrinker = NULL;
+ shrinker_free(shrinker);
+}
+
/**
* jbd2_journal_destroy() - Release a journal_t structure.
* @journal: Journal to act on.
@@ -2173,10 +2184,8 @@ int jbd2_journal_destroy(journal_t *jour
brelse(journal->j_sb_buffer);
}
- if (journal->j_shrinker) {
- percpu_counter_destroy(&journal->j_checkpoint_jh_count);
- shrinker_free(journal->j_shrinker);
- }
+ jbd2_journal_unregister_shrinker(journal);
+ percpu_counter_destroy(&journal->j_checkpoint_jh_count);
if (journal->j_proc_entry)
jbd2_stats_proc_exit(journal);
iput(journal->j_inode);
--- a/fs/ocfs2/journal.c~ocfs2-unregister-the-jbd2-shrinker-before-journal-shutdown
+++ a/fs/ocfs2/journal.c
@@ -1119,6 +1119,7 @@ void ocfs2_journal_shutdown(struct ocfs2
* Exclude the final JBD2 inode cleanup in ocfs2_clear_inode() while
* the journal is destroyed and freed.
*/
+ jbd2_journal_unregister_shrinker(journal->j_journal);
down_write(&osb->journal_shutdown_sem);
/* Shutdown the kernel journal system */
--- a/include/linux/jbd2.h~ocfs2-unregister-the-jbd2-shrinker-before-journal-shutdown
+++ a/include/linux/jbd2.h
@@ -1566,6 +1566,7 @@ extern void jbd2_journal_clear_featur
(journal_t *, unsigned long, unsigned long, unsigned long);
extern int jbd2_journal_load (journal_t *journal);
extern int jbd2_journal_destroy (journal_t *);
+extern void jbd2_journal_unregister_shrinker(journal_t *);
extern int jbd2_journal_recover (journal_t *journal);
extern int jbd2_journal_wipe (journal_t *, int);
extern int jbd2_journal_skip_recovery (journal_t *);
_
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2026-11-22 17:39 UTC | newest]
Thread overview: (only message) (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-22 17:39 [PATCH] ocfs2: unregister the JBD2 shrinker before journal shutdown Andrew Morton
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox