projects / GitHub / moto-9609 / android_kernel_motorola_exynos9610.git / commitdiff
? search:
summary | shortlog | log | commit | commitdiff | tree
raw | patch | inline | side by side (parent: fedc004)
Btrfs: fix deadlock running delayed iputs at transaction commit time
authorFilipe Manana <fdmanana@suse.com>
Fri, 15 Jan 2016 11:05:12 +0000 (11:05 +0000)
committerChris Mason <clm@fb.com>
Wed, 20 Jan 2016 02:21:41 +0000 (18:21 -0800)
While running a stress test I ran into a deadlock when running the delayed
iputs at transaction time, which produced the following report and trace:

[  886.399989] =============================================
[  886.400871] [ INFO: possible recursive locking detected ]
[  886.401663] 4.4.0-rc6-btrfs-next-18+ #1 Not tainted
[  886.402384] ---------------------------------------------
[  886.403182] fio/8277 is trying to acquire lock:
[  886.403568]  (&fs_info->delayed_iput_sem){++++..}, at: [<ffffffffa0538823>] btrfs_run_delayed_iputs+0x36/0xbf [btrfs]
[  886.403568]
[  886.403568] but task is already holding lock:
[  886.403568]  (&fs_info->delayed_iput_sem){++++..}, at: [<ffffffffa0538823>] btrfs_run_delayed_iputs+0x36/0xbf [btrfs]
[  886.403568]
[  886.403568] other info that might help us debug this:
[  886.403568]  Possible unsafe locking scenario:
[  886.403568]
[  886.403568]        CPU0
[  886.403568]        ----
[  886.403568]   lock(&fs_info->delayed_iput_sem);
[  886.403568]   lock(&fs_info->delayed_iput_sem);
[  886.403568]
[  886.403568]  *** DEADLOCK ***
[  886.403568]
[  886.403568]  May be due to missing lock nesting notation
[  886.403568]
[  886.403568] 3 locks held by fio/8277:
[  886.403568]  #0:  (sb_writers#11){.+.+.+}, at: [<ffffffff81174c4c>] __sb_start_write+0x5f/0xb0
[  886.403568]  #1:  (&sb->s_type->i_mutex_key#15){+.+.+.}, at: [<ffffffffa054620d>] btrfs_file_write_iter+0x73/0x408 [btrfs]
[  886.403568]  #2:  (&fs_info->delayed_iput_sem){++++..}, at: [<ffffffffa0538823>] btrfs_run_delayed_iputs+0x36/0xbf [btrfs]
[  886.403568]
[  886.403568] stack backtrace:
[  886.403568] CPU: 6 PID: 8277 Comm: fio Not tainted 4.4.0-rc6-btrfs-next-18+ #1
[  886.403568] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS by qemu-project.org 04/01/2014
[  886.403568]  0000000000000000 ffff88009f80f770 ffffffff8125d4fd ffffffff82af1fc0
[  886.403568]  ffff88009f80f830 ffffffff8108e5f9 0000000200000000 ffff88009fd92290
[  886.403568]  0000000000000000 ffffffff82af1fc0 ffffffff829cfb01 00042b216d008804
[  886.403568] Call Trace:
[  886.403568]  [<ffffffff8125d4fd>] dump_stack+0x4e/0x79
[  886.403568]  [<ffffffff8108e5f9>] __lock_acquire+0xd42/0xf0b
[  886.403568]  [<ffffffff810c22db>] ? __module_address+0xdf/0x108
[  886.403568]  [<ffffffff8108eb77>] lock_acquire+0x10d/0x194
[  886.403568]  [<ffffffff8108eb77>] ? lock_acquire+0x10d/0x194
[  886.403568]  [<ffffffffa0538823>] ? btrfs_run_delayed_iputs+0x36/0xbf [btrfs]
[  886.489542]  [<ffffffff8148556b>] down_read+0x3e/0x4d
[  886.489542]  [<ffffffffa0538823>] ? btrfs_run_delayed_iputs+0x36/0xbf [btrfs]
[  886.489542]  [<ffffffffa0538823>] btrfs_run_delayed_iputs+0x36/0xbf [btrfs]
[  886.489542]  [<ffffffffa0533953>] btrfs_commit_transaction+0x8f5/0x96e [btrfs]
[  886.489542]  [<ffffffffa0521d7a>] flush_space+0x435/0x44a [btrfs]
[  886.489542]  [<ffffffffa052218b>] ? reserve_metadata_bytes+0x26a/0x384 [btrfs]
[  886.489542]  [<ffffffffa05221ae>] reserve_metadata_bytes+0x28d/0x384 [btrfs]
[  886.489542]  [<ffffffffa052256c>] ? btrfs_block_rsv_refill+0x58/0x96 [btrfs]
[  886.489542]  [<ffffffffa0522584>] btrfs_block_rsv_refill+0x70/0x96 [btrfs]
[  886.489542]  [<ffffffffa053d747>] btrfs_evict_inode+0x394/0x55a [btrfs]
[  886.489542]  [<ffffffff81188e31>] evict+0xa7/0x15c
[  886.489542]  [<ffffffff81189878>] iput+0x1d3/0x266
[  886.489542]  [<ffffffffa053887c>] btrfs_run_delayed_iputs+0x8f/0xbf [btrfs]
[  886.489542]  [<ffffffffa0533953>] btrfs_commit_transaction+0x8f5/0x96e [btrfs]
[  886.489542]  [<ffffffff81085096>] ? signal_pending_state+0x31/0x31
[  886.489542]  [<ffffffffa0521191>] btrfs_alloc_data_chunk_ondemand+0x1d7/0x288 [btrfs]
[  886.489542]  [<ffffffffa0521282>] btrfs_check_data_free_space+0x40/0x59 [btrfs]
[  886.489542]  [<ffffffffa05228f5>] btrfs_delalloc_reserve_space+0x1e/0x4e [btrfs]
[  886.489542]  [<ffffffffa053620a>] btrfs_direct_IO+0x10c/0x27e [btrfs]
[  886.489542]  [<ffffffff8111d9a1>] generic_file_direct_write+0xb3/0x128
[  886.489542]  [<ffffffffa05463c3>] btrfs_file_write_iter+0x229/0x408 [btrfs]
[  886.489542]  [<ffffffff8108ae38>] ? __lock_is_held+0x38/0x50
[  886.489542]  [<ffffffff8117279e>] __vfs_write+0x7c/0xa5
[  886.489542]  [<ffffffff81172cda>] vfs_write+0xa0/0xe4
[  886.489542]  [<ffffffff811734cc>] SyS_write+0x50/0x7e
[  886.489542]  [<ffffffff814872d7>] entry_SYSCALL_64_fastpath+0x12/0x6f
[ 1081.852335] INFO: task fio:8244 blocked for more than 120 seconds.
[ 1081.854348]       Not tainted 4.4.0-rc6-btrfs-next-18+ #1
[ 1081.857560] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[ 1081.863227] fio        D ffff880213f9bb28     0  8244   8240 0x00000000
[ 1081.868719]  ffff880213f9bb28 00ffffff810fc6b0 ffffffff0000000a ffff88023ed55240
[ 1081.872499]  ffff880206b5d400 ffff880213f9c000 ffff88020a4d5318 ffff880206b5d400
[ 1081.876834]  ffffffff00000001 ffff880206b5d400 ffff880213f9bb40 ffffffff81482ba4
[ 1081.880782] Call Trace:
[ 1081.881793]  [<ffffffff81482ba4>] schedule+0x7f/0x97
[ 1081.883340]  [<ffffffff81485eb5>] rwsem_down_write_failed+0x2d5/0x325
[ 1081.895525]  [<ffffffff8108d48d>] ? trace_hardirqs_on_caller+0x16/0x1ab
[ 1081.897419]  [<ffffffff81269723>] call_rwsem_down_write_failed+0x13/0x20
[ 1081.899251]  [<ffffffff81269723>] ? call_rwsem_down_write_failed+0x13/0x20
[ 1081.901063]  [<ffffffff81089fae>] ? __down_write_nested.isra.0+0x1f/0x21
[ 1081.902365]  [<ffffffff814855bd>] down_write+0x43/0x57
[ 1081.903846]  [<ffffffffa05211b0>] ? btrfs_alloc_data_chunk_ondemand+0x1f6/0x288 [btrfs]
[ 1081.906078]  [<ffffffffa05211b0>] btrfs_alloc_data_chunk_ondemand+0x1f6/0x288 [btrfs]
[ 1081.908846]  [<ffffffff8108d461>] ? mark_held_locks+0x56/0x6c
[ 1081.910409]  [<ffffffffa0521282>] btrfs_check_data_free_space+0x40/0x59 [btrfs]
[ 1081.912482]  [<ffffffffa05228f5>] btrfs_delalloc_reserve_space+0x1e/0x4e [btrfs]
[ 1081.914597]  [<ffffffffa053620a>] btrfs_direct_IO+0x10c/0x27e [btrfs]
[ 1081.919037]  [<ffffffff8111d9a1>] generic_file_direct_write+0xb3/0x128
[ 1081.920754]  [<ffffffffa05463c3>] btrfs_file_write_iter+0x229/0x408 [btrfs]
[ 1081.922496]  [<ffffffff8108ae38>] ? __lock_is_held+0x38/0x50
[ 1081.923922]  [<ffffffff8117279e>] __vfs_write+0x7c/0xa5
[ 1081.925275]  [<ffffffff81172cda>] vfs_write+0xa0/0xe4
[ 1081.926584]  [<ffffffff811734cc>] SyS_write+0x50/0x7e
[ 1081.927968]  [<ffffffff814872d7>] entry_SYSCALL_64_fastpath+0x12/0x6f
[ 1081.985293] INFO: lockdep is turned off.
[ 1081.986132] INFO: task fio:8249 blocked for more than 120 seconds.
[ 1081.987434]       Not tainted 4.4.0-rc6-btrfs-next-18+ #1
[ 1081.988534] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[ 1081.990147] fio        D ffff880218febbb8     0  8249   8240 0x00000000
[ 1081.991626]  ffff880218febbb8 00ffffff81486b8e ffff88020000000b ffff88023ed75240
[ 1081.993258]  ffff8802120a9a00 ffff880218fec000 ffff88020a4d5318 ffff8802120a9a00
[ 1081.994850]  ffffffff00000001 ffff8802120a9a00 ffff880218febbd0 ffffffff81482ba4
[ 1081.996485] Call Trace:
[ 1081.997037]  [<ffffffff81482ba4>] schedule+0x7f/0x97
[ 1081.998017]  [<ffffffff81485eb5>] rwsem_down_write_failed+0x2d5/0x325
[ 1081.999241]  [<ffffffff810852a5>] ? finish_wait+0x6d/0x76
[ 1082.000306]  [<ffffffff81269723>] call_rwsem_down_write_failed+0x13/0x20
[ 1082.001533]  [<ffffffff81269723>] ? call_rwsem_down_write_failed+0x13/0x20
[ 1082.002776]  [<ffffffff81089fae>] ? __down_write_nested.isra.0+0x1f/0x21
[ 1082.003995]  [<ffffffff814855bd>] down_write+0x43/0x57
[ 1082.005000]  [<ffffffffa05211b0>] ? btrfs_alloc_data_chunk_ondemand+0x1f6/0x288 [btrfs]
[ 1082.007403]  [<ffffffffa05211b0>] btrfs_alloc_data_chunk_ondemand+0x1f6/0x288 [btrfs]
[ 1082.008988]  [<ffffffffa0545064>] btrfs_fallocate+0x7c1/0xc2f [btrfs]
[ 1082.010193]  [<ffffffff8108a1ba>] ? percpu_down_read+0x4e/0x77
[ 1082.011280]  [<ffffffff81174c4c>] ? __sb_start_write+0x5f/0xb0
[ 1082.012265]  [<ffffffff81174c4c>] ? __sb_start_write+0x5f/0xb0
[ 1082.013021]  [<ffffffff811712e4>] vfs_fallocate+0x170/0x1ff
[ 1082.013738]  [<ffffffff81181ebb>] ioctl_preallocate+0x89/0x9b
[ 1082.014778]  [<ffffffff811822d7>] do_vfs_ioctl+0x40a/0x4ea
[ 1082.015778]  [<ffffffff81176ea7>] ? SYSC_newfstat+0x25/0x2e
[ 1082.016806]  [<ffffffff8118b4de>] ? __fget_light+0x4d/0x71
[ 1082.017789]  [<ffffffff8118240e>] SyS_ioctl+0x57/0x79
[ 1082.018706]  [<ffffffff814872d7>] entry_SYSCALL_64_fastpath+0x12/0x6f

This happens because we can recursively acquire the semaphore
fs_info->delayed_iput_sem when attempting to allocate space to satisfy
a file write request as shown in the first trace above - when committing
a transaction we acquire (down_read) the semaphore before running the
delayed iputs, and when running a delayed iput() we can end up calling
an inode's eviction handler, which in turn commits another transaction
and attempts to acquire (down_read) again the semaphore to run more
delayed iput operations.
This results in a deadlock because if a task acquires multiple times a
semaphore it should invoke down_read_nested() with a different lockdep
class for each level of recursion.

Fix this by simplifying the implementation and use a mutex instead that
is acquired by the cleaner kthread before it runs the delayed iputs
instead of always acquiring a semaphore before delayed references are
run from anywhere.

Fixes: d7c151717a1e (btrfs: Fix NO_SPACE bug caused by delayed-iput)
Cc: stable@vger.kernel.org # 4.1+
Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
fs/btrfs/ctree.h patch | blob | blame | history
fs/btrfs/disk-io.c patch | blob | blame | history
fs/btrfs/extent-tree.c patch | blob | blame | history
fs/btrfs/inode.c patch | blob | blame | history

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index c5f40dc1f74fc06508ed794bde57e67b6fcd09c8..e9c2f8895eabc3028294ea0a378fd9ed67524569 100644 (file)
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -1614,7 +1614,7 @@ struct btrfs_fs_info {
 
        spinlock_t delayed_iput_lock;
        struct list_head delayed_iputs;
-       struct rw_semaphore delayed_iput_sem;
+       struct mutex cleaner_delayed_iput_mutex;
 
        /* this protects tree_mod_seq_list */
        spinlock_t tree_mod_seq_lock;
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 64f02c3d0dd0f2d1bdba9a891612469fcdf95b22..be03f93ca25718cfc6075b117a4120c98df5fa4d 100644 (file)
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -1801,7 +1801,10 @@ static int cleaner_kthread(void *arg)
                        goto sleep;
                }
 
+               mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
                btrfs_run_delayed_iputs(root);
+               mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
+
                again = btrfs_clean_one_deleted_snapshot(root);
                mutex_unlock(&root->fs_info->cleaner_mutex);
 
@@ -2571,8 +2574,8 @@ int open_ctree(struct super_block *sb,
        mutex_init(&fs_info->delete_unused_bgs_mutex);
        mutex_init(&fs_info->reloc_mutex);
        mutex_init(&fs_info->delalloc_root_mutex);
+       mutex_init(&fs_info->cleaner_delayed_iput_mutex);
        seqlock_init(&fs_info->profiles_lock);
-       init_rwsem(&fs_info->delayed_iput_sem);
 
        INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
        INIT_LIST_HEAD(&fs_info->space_info);
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 92843ab772a532fd510094a6de0a61781fbdee50..abcffa4b8231728febc91f2236f39482bd335c7f 100644 (file)
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -4153,11 +4153,12 @@ commit_trans:
                                if (ret)
                                        return ret;
                                /*
-                                * make sure that all running delayed iput are
-                                * done
+                                * The cleaner kthread might still be doing iput
+                                * operations. Wait for it to finish so that
+                                * more space is released.
                                 */
-                               down_write(&root->fs_info->delayed_iput_sem);
-                               up_write(&root->fs_info->delayed_iput_sem);
+                               mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
+                               mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
                                goto again;
                        } else {
                                btrfs_end_transaction(trans, root);
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 85afe66955cf395611892e31a31ca05245146f69..8ad9e2200442e2f365c531521d7a79ba2f5c4280 100644 (file)
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -3134,7 +3134,6 @@ void btrfs_run_delayed_iputs(struct btrfs_root *root)
 {
        struct btrfs_fs_info *fs_info = root->fs_info;
 
-       down_read(&fs_info->delayed_iput_sem);
        spin_lock(&fs_info->delayed_iput_lock);
        while (!list_empty(&fs_info->delayed_iputs)) {
                struct btrfs_inode *inode;
@@ -3153,7 +3152,6 @@ void btrfs_run_delayed_iputs(struct btrfs_root *root)
                spin_lock(&fs_info->delayed_iput_lock);
        }
        spin_unlock(&fs_info->delayed_iput_lock);
-       up_read(&root->fs_info->delayed_iput_sem);
 }
 
 /*