git.ipfire.org Git - thirdparty/kernel/stable-queue.git/blob

   1 From dfd65c34f343bfc44a0a219b9160c66d7d5046b8 Mon Sep 17 00:00:00 2001
   2 From: Sasha Levin <sashal@kernel.org>
   3 Date: Tue, 1 Sep 2020 08:09:01 -0400
   4 Subject: btrfs: sysfs: init devices outside of the chunk_mutex
   5
   6 From: Josef Bacik <josef@toxicpanda.com>
   7
   8 [ Upstream commit ca10845a56856fff4de3804c85e6424d0f6d0cde ]
   9
  10 While running btrfs/061, btrfs/073, btrfs/078, or btrfs/178 we hit the
  11 following lockdep splat:
  12
  13   ======================================================
  14   WARNING: possible circular locking dependency detected
  15   5.9.0-rc3+ #4 Not tainted
  16   ------------------------------------------------------
  17   kswapd0/100 is trying to acquire lock:
  18   ffff96ecc22ef4a0 (&delayed_node->mutex){+.+.}-{3:3}, at: __btrfs_release_delayed_node.part.0+0x3f/0x330
  19
  20   but task is already holding lock:
  21   ffffffff8dd74700 (fs_reclaim){+.+.}-{0:0}, at: __fs_reclaim_acquire+0x5/0x30
  22
  23   which lock already depends on the new lock.
  24
  25   the existing dependency chain (in reverse order) is:
  26
  27   -> #3 (fs_reclaim){+.+.}-{0:0}:
  28          fs_reclaim_acquire+0x65/0x80
  29          slab_pre_alloc_hook.constprop.0+0x20/0x200
  30          kmem_cache_alloc+0x37/0x270
  31          alloc_inode+0x82/0xb0
  32          iget_locked+0x10d/0x2c0
  33          kernfs_get_inode+0x1b/0x130
  34          kernfs_get_tree+0x136/0x240
  35          sysfs_get_tree+0x16/0x40
  36          vfs_get_tree+0x28/0xc0
  37          path_mount+0x434/0xc00
  38          __x64_sys_mount+0xe3/0x120
  39          do_syscall_64+0x33/0x40
  40          entry_SYSCALL_64_after_hwframe+0x44/0xa9
  41
  42   -> #2 (kernfs_mutex){+.+.}-{3:3}:
  43          __mutex_lock+0x7e/0x7e0
  44          kernfs_add_one+0x23/0x150
  45          kernfs_create_link+0x63/0xa0
  46          sysfs_do_create_link_sd+0x5e/0xd0
  47          btrfs_sysfs_add_devices_dir+0x81/0x130
  48          btrfs_init_new_device+0x67f/0x1250
  49          btrfs_ioctl+0x1ef/0x2e20
  50          __x64_sys_ioctl+0x83/0xb0
  51          do_syscall_64+0x33/0x40
  52          entry_SYSCALL_64_after_hwframe+0x44/0xa9
  53
  54   -> #1 (&fs_info->chunk_mutex){+.+.}-{3:3}:
  55          __mutex_lock+0x7e/0x7e0
  56          btrfs_chunk_alloc+0x125/0x3a0
  57          find_free_extent+0xdf6/0x1210
  58          btrfs_reserve_extent+0xb3/0x1b0
  59          btrfs_alloc_tree_block+0xb0/0x310
  60          alloc_tree_block_no_bg_flush+0x4a/0x60
  61          __btrfs_cow_block+0x11a/0x530
  62          btrfs_cow_block+0x104/0x220
  63          btrfs_search_slot+0x52e/0x9d0
  64          btrfs_insert_empty_items+0x64/0xb0
  65          btrfs_insert_delayed_items+0x90/0x4f0
  66          btrfs_commit_inode_delayed_items+0x93/0x140
  67          btrfs_log_inode+0x5de/0x2020
  68          btrfs_log_inode_parent+0x429/0xc90
  69          btrfs_log_new_name+0x95/0x9b
  70          btrfs_rename2+0xbb9/0x1800
  71          vfs_rename+0x64f/0x9f0
  72          do_renameat2+0x320/0x4e0
  73          __x64_sys_rename+0x1f/0x30
  74          do_syscall_64+0x33/0x40
  75          entry_SYSCALL_64_after_hwframe+0x44/0xa9
  76
  77   -> #0 (&delayed_node->mutex){+.+.}-{3:3}:
  78          __lock_acquire+0x119c/0x1fc0
  79          lock_acquire+0xa7/0x3d0
  80          __mutex_lock+0x7e/0x7e0
  81          __btrfs_release_delayed_node.part.0+0x3f/0x330
  82          btrfs_evict_inode+0x24c/0x500
  83          evict+0xcf/0x1f0
  84          dispose_list+0x48/0x70
  85          prune_icache_sb+0x44/0x50
  86          super_cache_scan+0x161/0x1e0
  87          do_shrink_slab+0x178/0x3c0
  88          shrink_slab+0x17c/0x290
  89          shrink_node+0x2b2/0x6d0
  90          balance_pgdat+0x30a/0x670
  91          kswapd+0x213/0x4c0
  92          kthread+0x138/0x160
  93          ret_from_fork+0x1f/0x30
  94
  95   other info that might help us debug this:
  96
  97   Chain exists of:
  98     &delayed_node->mutex --> kernfs_mutex --> fs_reclaim
  99
 100    Possible unsafe locking scenario:
 101
 102          CPU0                    CPU1
 103          ----                    ----
 104     lock(fs_reclaim);
 105                                  lock(kernfs_mutex);
 106                                  lock(fs_reclaim);
 107     lock(&delayed_node->mutex);
 108
 109    *** DEADLOCK ***
 110
 111   3 locks held by kswapd0/100:
 112    #0: ffffffff8dd74700 (fs_reclaim){+.+.}-{0:0}, at: __fs_reclaim_acquire+0x5/0x30
 113    #1: ffffffff8dd65c50 (shrinker_rwsem){++++}-{3:3}, at: shrink_slab+0x115/0x290
 114    #2: ffff96ed2ade30e0 (&type->s_umount_key#36){++++}-{3:3}, at: super_cache_scan+0x38/0x1e0
 115
 116   stack backtrace:
 117   CPU: 0 PID: 100 Comm: kswapd0 Not tainted 5.9.0-rc3+ #4
 118   Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.13.0-2.fc32 04/01/2014
 119   Call Trace:
 120    dump_stack+0x8b/0xb8
 121    check_noncircular+0x12d/0x150
 122    __lock_acquire+0x119c/0x1fc0
 123    lock_acquire+0xa7/0x3d0
 124    ? __btrfs_release_delayed_node.part.0+0x3f/0x330
 125    __mutex_lock+0x7e/0x7e0
 126    ? __btrfs_release_delayed_node.part.0+0x3f/0x330
 127    ? __btrfs_release_delayed_node.part.0+0x3f/0x330
 128    ? lock_acquire+0xa7/0x3d0
 129    ? find_held_lock+0x2b/0x80
 130    __btrfs_release_delayed_node.part.0+0x3f/0x330
 131    btrfs_evict_inode+0x24c/0x500
 132    evict+0xcf/0x1f0
 133    dispose_list+0x48/0x70
 134    prune_icache_sb+0x44/0x50
 135    super_cache_scan+0x161/0x1e0
 136    do_shrink_slab+0x178/0x3c0
 137    shrink_slab+0x17c/0x290
 138    shrink_node+0x2b2/0x6d0
 139    balance_pgdat+0x30a/0x670
 140    kswapd+0x213/0x4c0
 141    ? _raw_spin_unlock_irqrestore+0x41/0x50
 142    ? add_wait_queue_exclusive+0x70/0x70
 143    ? balance_pgdat+0x670/0x670
 144    kthread+0x138/0x160
 145    ? kthread_create_worker_on_cpu+0x40/0x40
 146    ret_from_fork+0x1f/0x30
 147
 148 This happens because we are holding the chunk_mutex at the time of
 149 adding in a new device.  However we only need to hold the
 150 device_list_mutex, as we're going to iterate over the fs_devices
 151 devices.  Move the sysfs init stuff outside of the chunk_mutex to get
 152 rid of this lockdep splat.
 153
 154 CC: stable@vger.kernel.org # 4.4.x: f3cd2c58110dad14e: btrfs: sysfs, rename device_link add/remove functions
 155 CC: stable@vger.kernel.org # 4.4.x
 156 Reported-by: David Sterba <dsterba@suse.com>
 157 Signed-off-by: Josef Bacik <josef@toxicpanda.com>
 158 Reviewed-by: David Sterba <dsterba@suse.com>
 159 Signed-off-by: David Sterba <dsterba@suse.com>
 160 Signed-off-by: Sasha Levin <sashal@kernel.org>
 161 ---
 162  fs/btrfs/volumes.c | 7 ++++---
 163  1 file changed, 4 insertions(+), 3 deletions(-)
 164
 165 diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
 166 index f3cb042a28d5c..3d92feff249a0 100644
 167 --- a/fs/btrfs/volumes.c
 168 +++ b/fs/btrfs/volumes.c
 169 @@ -2432,9 +2432,6 @@ int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path
 170         tmp = btrfs_super_num_devices(fs_info->super_copy);
 171         btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1);
 172
 173 -       /* add sysfs device entry */
 174 -       btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
 175 -
 176         /*
 177          * we've got more storage, clear any full flags on the space
 178          * infos
 179 @@ -2442,6 +2439,10 @@ int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path
 180         btrfs_clear_space_info_full(fs_info);
 181
 182         mutex_unlock(&fs_info->chunk_mutex);
 183 +
 184 +       /* Add sysfs device entry */
 185 +       btrfs_sysfs_add_device_link(fs_info->fs_devices, device);
 186 +
 187         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 188
 189         if (seeding_dev) {
 190 --
 191 2.27.0
 192