reiserfs: Relax reiserfs lock while freeing the journal
authorFrederic Weisbecker <fweisbec@gmail.com>
Wed, 30 Dec 2009 04:56:08 +0000 (05:56 +0100)
committerFrederic Weisbecker <fweisbec@gmail.com>
Sat, 2 Jan 2010 00:56:54 +0000 (01:56 +0100)
Keeping the reiserfs lock while freeing the journal on
umount path triggers a lock inversion between bdev->bd_mutex
and the reiserfs lock.

We don't need the reiserfs lock at this stage. The filesystem
is not usable anymore, and there are no more pending commits,
everything got flushed (even this operation was done in parallel
and didn't required the reiserfs lock from the current process).

This fixes the following lockdep report:

=======================================================
[ INFO: possible circular locking dependency detected ]
2.6.32-atom #172
-------------------------------------------------------
umount/3904 is trying to acquire lock:
 (&bdev->bd_mutex){+.+.+.}, at: [<c10de2c2>] __blkdev_put+0x22/0x160

but task is already holding lock:
 (&REISERFS_SB(s)->lock){+.+.+.}, at: [<c1143279>] reiserfs_write_lock+0x29/0x40

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #3 (&REISERFS_SB(s)->lock){+.+.+.}:
       [<c105ea7f>] __lock_acquire+0x11ff/0x19e0
       [<c105f2c8>] lock_acquire+0x68/0x90
       [<c140199b>] mutex_lock_nested+0x5b/0x340
       [<c1143229>] reiserfs_write_lock_once+0x29/0x50
       [<c111c485>] reiserfs_get_block+0x85/0x1620
       [<c10e1040>] do_mpage_readpage+0x1f0/0x6d0
       [<c10e1640>] mpage_readpages+0xc0/0x100
       [<c1119b89>] reiserfs_readpages+0x19/0x20
       [<c108f1ec>] __do_page_cache_readahead+0x1bc/0x260
       [<c108f2b8>] ra_submit+0x28/0x40
       [<c1087e3e>] filemap_fault+0x40e/0x420
       [<c109b5fd>] __do_fault+0x3d/0x430
       [<c109d47e>] handle_mm_fault+0x12e/0x790
       [<c1022a65>] do_page_fault+0x135/0x330
       [<c1403663>] error_code+0x6b/0x70
       [<c10ef9ca>] load_elf_binary+0x82a/0x1a10
       [<c10ba130>] search_binary_handler+0x90/0x1d0
       [<c10bb70f>] do_execve+0x1df/0x250
       [<c1001746>] sys_execve+0x46/0x70
       [<c1002fa5>] syscall_call+0x7/0xb

-> #2 (&mm->mmap_sem){++++++}:
       [<c105ea7f>] __lock_acquire+0x11ff/0x19e0
       [<c105f2c8>] lock_acquire+0x68/0x90
       [<c109b1ab>] might_fault+0x8b/0xb0
       [<c11b8f52>] copy_to_user+0x32/0x70
       [<c10c3b94>] filldir64+0xa4/0xf0
       [<c1109116>] sysfs_readdir+0x116/0x210
       [<c10c3e1d>] vfs_readdir+0x8d/0xb0
       [<c10c3ea9>] sys_getdents64+0x69/0xb0
       [<c1002ec4>] sysenter_do_call+0x12/0x32

-> #1 (sysfs_mutex){+.+.+.}:
       [<c105ea7f>] __lock_acquire+0x11ff/0x19e0
       [<c105f2c8>] lock_acquire+0x68/0x90
       [<c140199b>] mutex_lock_nested+0x5b/0x340
       [<c110951c>] sysfs_addrm_start+0x2c/0xb0
       [<c1109aa0>] create_dir+0x40/0x90
       [<c1109b1b>] sysfs_create_dir+0x2b/0x50
       [<c11b2352>] kobject_add_internal+0xc2/0x1b0
       [<c11b2531>] kobject_add_varg+0x31/0x50
       [<c11b25ac>] kobject_add+0x2c/0x60
       [<c1258294>] device_add+0x94/0x560
       [<c11036ea>] add_partition+0x18a/0x2a0
       [<c110418a>] rescan_partitions+0x33a/0x450
       [<c10de5bf>] __blkdev_get+0x12f/0x2d0
       [<c10de76a>] blkdev_get+0xa/0x10
       [<c11034b8>] register_disk+0x108/0x130
       [<c11a87a9>] add_disk+0xd9/0x130
       [<c12998e5>] sd_probe_async+0x105/0x1d0
       [<c10528af>] async_thread+0xcf/0x230
       [<c104bfd4>] kthread+0x74/0x80
       [<c1003aab>] kernel_thread_helper+0x7/0x3c

-> #0 (&bdev->bd_mutex){+.+.+.}:
       [<c105f176>] __lock_acquire+0x18f6/0x19e0
       [<c105f2c8>] lock_acquire+0x68/0x90
       [<c140199b>] mutex_lock_nested+0x5b/0x340
       [<c10de2c2>] __blkdev_put+0x22/0x160
       [<c10de40a>] blkdev_put+0xa/0x10
       [<c113ce22>] free_journal_ram+0xd2/0x130
       [<c113ea18>] do_journal_release+0x98/0x190
       [<c113eb2a>] journal_release+0xa/0x10
       [<c1128eb6>] reiserfs_put_super+0x36/0x130
       [<c10b776f>] generic_shutdown_super+0x4f/0xe0
       [<c10b7825>] kill_block_super+0x25/0x40
       [<c11255df>] reiserfs_kill_sb+0x7f/0x90
       [<c10b7f4a>] deactivate_super+0x7a/0x90
       [<c10cccd8>] mntput_no_expire+0x98/0xd0
       [<c10ccfcc>] sys_umount+0x4c/0x310
       [<c10cd2a9>] sys_oldumount+0x19/0x20
       [<c1002ec4>] sysenter_do_call+0x12/0x32

other info that might help us debug this:

2 locks held by umount/3904:
 #0:  (&type->s_umount_key#30){+++++.}, at: [<c10b7f45>] deactivate_super+0x75/0x90
 #1:  (&REISERFS_SB(s)->lock){+.+.+.}, at: [<c1143279>] reiserfs_write_lock+0x29/0x40

stack backtrace:
Pid: 3904, comm: umount Not tainted 2.6.32-atom #172
Call Trace:
 [<c13ff903>] ? printk+0x18/0x1a
 [<c105d33a>] print_circular_bug+0xca/0xd0
 [<c105f176>] __lock_acquire+0x18f6/0x19e0
 [<c108b66f>] ? free_pcppages_bulk+0x1f/0x250
 [<c105f2c8>] lock_acquire+0x68/0x90
 [<c10de2c2>] ? __blkdev_put+0x22/0x160
 [<c10de2c2>] ? __blkdev_put+0x22/0x160
 [<c140199b>] mutex_lock_nested+0x5b/0x340
 [<c10de2c2>] ? __blkdev_put+0x22/0x160
 [<c105c932>] ? mark_held_locks+0x62/0x80
 [<c10afe12>] ? kfree+0x92/0xd0
 [<c10de2c2>] __blkdev_put+0x22/0x160
 [<c105cc3b>] ? trace_hardirqs_on+0xb/0x10
 [<c10de40a>] blkdev_put+0xa/0x10
 [<c113ce22>] free_journal_ram+0xd2/0x130
 [<c113ea18>] do_journal_release+0x98/0x190
 [<c113eb2a>] journal_release+0xa/0x10
 [<c1128eb6>] reiserfs_put_super+0x36/0x130
 [<c1050596>] ? up_write+0x16/0x30
 [<c10b776f>] generic_shutdown_super+0x4f/0xe0
 [<c10b7825>] kill_block_super+0x25/0x40
 [<c10f41e0>] ? vfs_quota_off+0x0/0x20
 [<c11255df>] reiserfs_kill_sb+0x7f/0x90
 [<c10b7f4a>] deactivate_super+0x7a/0x90
 [<c10cccd8>] mntput_no_expire+0x98/0xd0
 [<c10ccfcc>] sys_umount+0x4c/0x310
 [<c10cd2a9>] sys_oldumount+0x19/0x20
 [<c1002ec4>] sysenter_do_call+0x12/0x32

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Alexander Beregalov <a.beregalov@gmail.com>
Cc: Chris Mason <chris.mason@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
fs/reiserfs/journal.c

index a059879667b1a7fdfd1399541ead50baf18e1af4..83ac4d3b3cb0f019f128b30c81f6e2e94e8b5a98 100644 (file)
@@ -2009,10 +2009,11 @@ static int do_journal_release(struct reiserfs_transaction_handle *th,
                destroy_workqueue(commit_wq);
                commit_wq = NULL;
        }
-       reiserfs_write_lock(sb);
 
        free_journal_ram(sb);
 
+       reiserfs_write_lock(sb);
+
        return 0;
 }