ext4: Fix potential quota deadlock
Dmitry Monakhov [Thu, 10 Dec 2009 03:31:45 +0000 (03:31 +0000)]
We have to delay vfs_dq_claim_space() until allocation context destruction.
Currently we have following call-trace:
ext4_mb_new_blocks()
  /* task is already holding ac->alloc_semp */
 ->ext4_mb_mark_diskspace_used
    ->vfs_dq_claim_space()  /*  acquire dqptr_sem here. Possible deadlock */
 ->ext4_mb_release_context() /* drop ac->alloc_semp here */

Let's move quota claiming to ext4_da_update_reserve_space()

 =======================================================
 [ INFO: possible circular locking dependency detected ]
 2.6.32-rc7 #18
 -------------------------------------------------------
 write-truncate-/3465 is trying to acquire lock:
  (&s->s_dquot.dqptr_sem){++++..}, at: [<c025e73b>] dquot_claim_space+0x3b/0x1b0

 but task is already holding lock:
  (&meta_group_info[i]->alloc_sem){++++..}, at: [<c02ce962>] ext4_mb_load_buddy+0xb2/0x370

 which lock already depends on the new lock.

 the existing dependency chain (in reverse order) is:

 -> #3 (&meta_group_info[i]->alloc_sem){++++..}:
        [<c017d04b>] __lock_acquire+0xd7b/0x1260
        [<c017d5ea>] lock_acquire+0xba/0xd0
        [<c0527191>] down_read+0x51/0x90
        [<c02ce962>] ext4_mb_load_buddy+0xb2/0x370
        [<c02d0c1c>] ext4_mb_free_blocks+0x46c/0x870
        [<c029c9d3>] ext4_free_blocks+0x73/0x130
        [<c02c8cfc>] ext4_ext_truncate+0x76c/0x8d0
        [<c02a8087>] ext4_truncate+0x187/0x5e0
        [<c01e0f7b>] vmtruncate+0x6b/0x70
        [<c022ec02>] inode_setattr+0x62/0x190
        [<c02a2d7a>] ext4_setattr+0x25a/0x370
        [<c022ee81>] notify_change+0x151/0x340
        [<c021349d>] do_truncate+0x6d/0xa0
        [<c0221034>] may_open+0x1d4/0x200
        [<c022412b>] do_filp_open+0x1eb/0x910
        [<c021244d>] do_sys_open+0x6d/0x140
        [<c021258e>] sys_open+0x2e/0x40
        [<c0103100>] sysenter_do_call+0x12/0x32

 -> #2 (&ei->i_data_sem){++++..}:
        [<c017d04b>] __lock_acquire+0xd7b/0x1260
        [<c017d5ea>] lock_acquire+0xba/0xd0
        [<c0527191>] down_read+0x51/0x90
        [<c02a5787>] ext4_get_blocks+0x47/0x450
        [<c02a74c1>] ext4_getblk+0x61/0x1d0
        [<c02a7a7f>] ext4_bread+0x1f/0xa0
        [<c02bcddc>] ext4_quota_write+0x12c/0x310
        [<c0262d23>] qtree_write_dquot+0x93/0x120
        [<c0261708>] v2_write_dquot+0x28/0x30
        [<c025d3fb>] dquot_commit+0xab/0xf0
        [<c02be977>] ext4_write_dquot+0x77/0x90
        [<c02be9bf>] ext4_mark_dquot_dirty+0x2f/0x50
        [<c025e321>] dquot_alloc_inode+0x101/0x180
        [<c029fec2>] ext4_new_inode+0x602/0xf00
        [<c02ad789>] ext4_create+0x89/0x150
        [<c0221ff2>] vfs_create+0xa2/0xc0
        [<c02246e7>] do_filp_open+0x7a7/0x910
        [<c021244d>] do_sys_open+0x6d/0x140
        [<c021258e>] sys_open+0x2e/0x40
        [<c0103100>] sysenter_do_call+0x12/0x32

 -> #1 (&sb->s_type->i_mutex_key#7/4){+.+...}:
        [<c017d04b>] __lock_acquire+0xd7b/0x1260
        [<c017d5ea>] lock_acquire+0xba/0xd0
        [<c0526505>] mutex_lock_nested+0x65/0x2d0
        [<c0260c9d>] vfs_load_quota_inode+0x4bd/0x5a0
        [<c02610af>] vfs_quota_on_path+0x5f/0x70
        [<c02bc812>] ext4_quota_on+0x112/0x190
        [<c026345a>] sys_quotactl+0x44a/0x8a0
        [<c0103100>] sysenter_do_call+0x12/0x32

 -> #0 (&s->s_dquot.dqptr_sem){++++..}:
        [<c017d361>] __lock_acquire+0x1091/0x1260
        [<c017d5ea>] lock_acquire+0xba/0xd0
        [<c0527191>] down_read+0x51/0x90
        [<c025e73b>] dquot_claim_space+0x3b/0x1b0
        [<c02cb95f>] ext4_mb_mark_diskspace_used+0x36f/0x380
        [<c02d210a>] ext4_mb_new_blocks+0x34a/0x530
        [<c02c83fb>] ext4_ext_get_blocks+0x122b/0x13c0
        [<c02a5966>] ext4_get_blocks+0x226/0x450
        [<c02a5ff3>] mpage_da_map_blocks+0xc3/0xaa0
        [<c02a6ed6>] ext4_da_writepages+0x506/0x790
        [<c01de272>] do_writepages+0x22/0x50
        [<c01d766d>] __filemap_fdatawrite_range+0x6d/0x80
        [<c01d7b9b>] filemap_flush+0x2b/0x30
        [<c02a40ac>] ext4_alloc_da_blocks+0x5c/0x60
        [<c029e595>] ext4_release_file+0x75/0xb0
        [<c0216b59>] __fput+0xf9/0x210
        [<c0216c97>] fput+0x27/0x30
        [<c02122dc>] filp_close+0x4c/0x80
        [<c014510e>] put_files_struct+0x6e/0xd0
        [<c01451b7>] exit_files+0x47/0x60
        [<c0146a24>] do_exit+0x144/0x710
        [<c0147028>] do_group_exit+0x38/0xa0
        [<c0159abc>] get_signal_to_deliver+0x2ac/0x410
        [<c0102849>] do_notify_resume+0xb9/0x890
        [<c01032d2>] work_notifysig+0x13/0x21

 other info that might help us debug this:

 3 locks held by write-truncate-/3465:
  #0:  (jbd2_handle){+.+...}, at: [<c02e1f8f>] start_this_handle+0x38f/0x5c0
  #1:  (&ei->i_data_sem){++++..}, at: [<c02a57f6>] ext4_get_blocks+0xb6/0x450
  #2:  (&meta_group_info[i]->alloc_sem){++++..}, at: [<c02ce962>] ext4_mb_load_buddy+0xb2/0x370

 stack backtrace:
 Pid: 3465, comm: write-truncate- Not tainted 2.6.32-rc7 #18
 Call Trace:
  [<c0524cb3>] ? printk+0x1d/0x22
  [<c017ac9a>] print_circular_bug+0xca/0xd0
  [<c017d361>] __lock_acquire+0x1091/0x1260
  [<c016bca2>] ? sched_clock_local+0xd2/0x170
  [<c0178fd0>] ? trace_hardirqs_off_caller+0x20/0xd0
  [<c017d5ea>] lock_acquire+0xba/0xd0
  [<c025e73b>] ? dquot_claim_space+0x3b/0x1b0
  [<c0527191>] down_read+0x51/0x90
  [<c025e73b>] ? dquot_claim_space+0x3b/0x1b0
  [<c025e73b>] dquot_claim_space+0x3b/0x1b0
  [<c02cb95f>] ext4_mb_mark_diskspace_used+0x36f/0x380
  [<c02d210a>] ext4_mb_new_blocks+0x34a/0x530
  [<c02c601d>] ? ext4_ext_find_extent+0x25d/0x280
  [<c02c83fb>] ext4_ext_get_blocks+0x122b/0x13c0
  [<c016bca2>] ? sched_clock_local+0xd2/0x170
  [<c016be60>] ? sched_clock_cpu+0x120/0x160
  [<c016beef>] ? cpu_clock+0x4f/0x60
  [<c0178fd0>] ? trace_hardirqs_off_caller+0x20/0xd0
  [<c052712c>] ? down_write+0x8c/0xa0
  [<c02a5966>] ext4_get_blocks+0x226/0x450
  [<c016be60>] ? sched_clock_cpu+0x120/0x160
  [<c016beef>] ? cpu_clock+0x4f/0x60
  [<c017908b>] ? trace_hardirqs_off+0xb/0x10
  [<c02a5ff3>] mpage_da_map_blocks+0xc3/0xaa0
  [<c01d69cc>] ? find_get_pages_tag+0x16c/0x180
  [<c01d6860>] ? find_get_pages_tag+0x0/0x180
  [<c02a73bd>] ? __mpage_da_writepage+0x16d/0x1a0
  [<c01dfc4e>] ? pagevec_lookup_tag+0x2e/0x40
  [<c01ddf1b>] ? write_cache_pages+0xdb/0x3d0
  [<c02a7250>] ? __mpage_da_writepage+0x0/0x1a0
  [<c02a6ed6>] ext4_da_writepages+0x506/0x790
  [<c016beef>] ? cpu_clock+0x4f/0x60
  [<c016bca2>] ? sched_clock_local+0xd2/0x170
  [<c016be60>] ? sched_clock_cpu+0x120/0x160
  [<c016be60>] ? sched_clock_cpu+0x120/0x160
  [<c02a69d0>] ? ext4_da_writepages+0x0/0x790
  [<c01de272>] do_writepages+0x22/0x50
  [<c01d766d>] __filemap_fdatawrite_range+0x6d/0x80
  [<c01d7b9b>] filemap_flush+0x2b/0x30
  [<c02a40ac>] ext4_alloc_da_blocks+0x5c/0x60
  [<c029e595>] ext4_release_file+0x75/0xb0
  [<c0216b59>] __fput+0xf9/0x210
  [<c0216c97>] fput+0x27/0x30
  [<c02122dc>] filp_close+0x4c/0x80
  [<c014510e>] put_files_struct+0x6e/0xd0
  [<c01451b7>] exit_files+0x47/0x60
  [<c0146a24>] do_exit+0x144/0x710
  [<c017b163>] ? lock_release_holdtime+0x33/0x210
  [<c0528137>] ? _spin_unlock_irq+0x27/0x30
  [<c0147028>] do_group_exit+0x38/0xa0
  [<c017babb>] ? trace_hardirqs_on+0xb/0x10
  [<c0159abc>] get_signal_to_deliver+0x2ac/0x410
  [<c0102849>] do_notify_resume+0xb9/0x890
  [<c0178fd0>] ? trace_hardirqs_off_caller+0x20/0xd0
  [<c017b163>] ? lock_release_holdtime+0x33/0x210
  [<c0165b50>] ? autoremove_wake_function+0x0/0x50
  [<c017ba54>] ? trace_hardirqs_on_caller+0x134/0x190
  [<c017babb>] ? trace_hardirqs_on+0xb/0x10
  [<c0300ba4>] ? security_file_permission+0x14/0x20
  [<c0215761>] ? vfs_write+0x131/0x190
  [<c0214f50>] ? do_sync_write+0x0/0x120
  [<c0103115>] ? sysenter_do_call+0x27/0x32
  [<c01032d2>] work_notifysig+0x13/0x21

CC: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: Jan Kara <jack@suse.cz>

fs/ext4/inode.c
fs/ext4/mballoc.c

index a31980d..ddd0a9c 100644 (file)
@@ -1046,7 +1046,7 @@ static int ext4_calc_metadata_amount(struct inode *inode, int blocks)
 static void ext4_da_update_reserve_space(struct inode *inode, int used)
 {
        struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
-       int total, mdb, mdb_free;
+       int total, mdb, mdb_free, mdb_claim = 0;
 
        spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
        /* recalculate the number of metablocks still need to be reserved */
@@ -1059,7 +1059,9 @@ static void ext4_da_update_reserve_space(struct inode *inode, int used)
 
        if (mdb_free) {
                /* Account for allocated meta_blocks */
-               mdb_free -= EXT4_I(inode)->i_allocated_meta_blocks;
+               mdb_claim = EXT4_I(inode)->i_allocated_meta_blocks;
+               BUG_ON(mdb_free < mdb_claim);
+               mdb_free -= mdb_claim;
 
                /* update fs dirty blocks counter */
                percpu_counter_sub(&sbi->s_dirtyblocks_counter, mdb_free);
@@ -1070,8 +1072,11 @@ static void ext4_da_update_reserve_space(struct inode *inode, int used)
        /* update per-inode reservations */
        BUG_ON(used  > EXT4_I(inode)->i_reserved_data_blocks);
        EXT4_I(inode)->i_reserved_data_blocks -= used;
+       percpu_counter_sub(&sbi->s_dirtyblocks_counter, used + mdb_claim);
        spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
 
+       vfs_dq_claim_block(inode, used + mdb_claim);
+
        /*
         * free those over-booking quota for metadata blocks
         */
index b1fd3da..d34afad 100644 (file)
@@ -2755,12 +2755,6 @@ ext4_mb_mark_diskspace_used(struct ext4_allocation_context *ac,
        if (!(ac->ac_flags & EXT4_MB_DELALLOC_RESERVED))
                /* release all the reserved blocks if non delalloc */
                percpu_counter_sub(&sbi->s_dirtyblocks_counter, reserv_blks);
-       else {
-               percpu_counter_sub(&sbi->s_dirtyblocks_counter,
-                                               ac->ac_b_ex.fe_len);
-               /* convert reserved quota blocks to real quota blocks */
-               vfs_dq_claim_block(ac->ac_inode, ac->ac_b_ex.fe_len);
-       }
 
        if (sbi->s_log_groups_per_flex) {
                ext4_group_t flex_group = ext4_flex_group(sbi,