btrfs: release path before initializing extent tree in btrfs_read_locked_inode()

In btrfs_read_locked_inode() we are calling btrfs_init_file_extent_tree()
while holding a path with a read locked leaf from a subvolume tree, and
btrfs_init_file_extent_tree() may do a GFP_KERNEL allocation, which can
trigger reclaim.

This can create a circular lock dependency which lockdep warns about with
the following splat:

   [27386.164433] ======================================================
   [27386.164574] WARNING: possible circular locking dependency detected
   [27386.164583] 6.18.0+ #4 Tainted: G     U
   [27386.164591] ------------------------------------------------------
   [27386.164599] kswapd0/117 is trying to acquire lock:
   [27386.164606] ffff8d9b6333c5b8 (&delayed_node->mutex){+.+.}-{3:3}, at:
   __btrfs_release_delayed_node.part.0+0x39/0x2f0
   [27386.164625]
                  but task is already holding lock:
   [27386.164633] ffffffffa4ab8ce0 (fs_reclaim){+.+.}-{0:0}, at:
   balance_pgdat+0x195/0xc60
   [27386.164646]
                  which lock already depends on the new lock.

   [27386.164657]
                  the existing dependency chain (in reverse order) is:
   [27386.164667]
                  -> #2 (fs_reclaim){+.+.}-{0:0}:
   [27386.164677]        fs_reclaim_acquire+0x9d/0xd0
   [27386.164685]        __kmalloc_cache_noprof+0x59/0x750
   [27386.164694]        btrfs_init_file_extent_tree+0x90/0x100
   [27386.164702]        btrfs_read_locked_inode+0xc3/0x6b0
   [27386.164710]        btrfs_iget+0xbb/0xf0
   [27386.164716]        btrfs_lookup_dentry+0x3c5/0x8e0
   [27386.164724]        btrfs_lookup+0x12/0x30
   [27386.164731]        lookup_open.isra.0+0x1aa/0x6a0
   [27386.164739]        path_openat+0x5f7/0xc60
   [27386.164746]        do_filp_open+0xd6/0x180
   [27386.164753]        do_sys_openat2+0x8b/0xe0
   [27386.164760]        __x64_sys_openat+0x54/0xa0
   [27386.164768]        do_syscall_64+0x97/0x3e0
   [27386.164776]        entry_SYSCALL_64_after_hwframe+0x76/0x7e
   [27386.164784]
                  -> #1 (btrfs-tree-00){++++}-{3:3}:
   [27386.164794]        lock_release+0x127/0x2a0
   [27386.164801]        up_read+0x1b/0x30
   [27386.164808]        btrfs_search_slot+0x8e0/0xff0
   [27386.164817]        btrfs_lookup_inode+0x52/0xd0
   [27386.164825]        __btrfs_update_delayed_inode+0x73/0x520
   [27386.164833]        btrfs_commit_inode_delayed_inode+0x11a/0x120
   [27386.164842]        btrfs_log_inode+0x608/0x1aa0
   [27386.164849]        btrfs_log_inode_parent+0x249/0xf80
   [27386.164857]        btrfs_log_dentry_safe+0x3e/0x60
   [27386.164865]        btrfs_sync_file+0x431/0x690
   [27386.164872]        do_fsync+0x39/0x80
   [27386.164879]        __x64_sys_fsync+0x13/0x20
   [27386.164887]        do_syscall_64+0x97/0x3e0
   [27386.164894]        entry_SYSCALL_64_after_hwframe+0x76/0x7e
   [27386.164903]
                  -> #0 (&delayed_node->mutex){+.+.}-{3:3}:
   [27386.164913]        __lock_acquire+0x15e9/0x2820
   [27386.164920]        lock_acquire+0xc9/0x2d0
   [27386.164927]        __mutex_lock+0xcc/0x10a0
   [27386.164934]        __btrfs_release_delayed_node.part.0+0x39/0x2f0
   [27386.164944]        btrfs_evict_inode+0x20b/0x4b0
   [27386.164952]        evict+0x15a/0x2f0
   [27386.164958]        prune_icache_sb+0x91/0xd0
   [27386.164966]        super_cache_scan+0x150/0x1d0
   [27386.164974]        do_shrink_slab+0x155/0x6f0
   [27386.164981]        shrink_slab+0x48e/0x890
   [27386.164988]        shrink_one+0x11a/0x1f0
   [27386.164995]        shrink_node+0xbfd/0x1320
   [27386.165002]        balance_pgdat+0x67f/0xc60
   [27386.165321]        kswapd+0x1dc/0x3e0
   [27386.165643]        kthread+0xff/0x240
   [27386.165965]        ret_from_fork+0x223/0x280
   [27386.166287]        ret_from_fork_asm+0x1a/0x30
   [27386.166616]
                  other info that might help us debug this:

   [27386.167561] Chain exists of:
                    &delayed_node->mutex --> btrfs-tree-00 --> fs_reclaim

   [27386.168503]  Possible unsafe locking scenario:

   [27386.169110]        CPU0                    CPU1
   [27386.169411]        ----                    ----
   [27386.169707]   lock(fs_reclaim);
   [27386.169998]                                lock(btrfs-tree-00);
   [27386.170291]                                lock(fs_reclaim);
   [27386.170581]   lock(&delayed_node->mutex);
   [27386.170874]
                   *** DEADLOCK ***

   [27386.171716] 2 locks held by kswapd0/117:
   [27386.171999]  #0: ffffffffa4ab8ce0 (fs_reclaim){+.+.}-{0:0}, at:
   balance_pgdat+0x195/0xc60
   [27386.172294]  #1: ffff8d998344b0e0 (&type->s_umount_key#40){++++}-
   {3:3}, at: super_cache_scan+0x37/0x1d0
   [27386.172596]
                  stack backtrace:
   [27386.173183] CPU: 11 UID: 0 PID: 117 Comm: kswapd0 Tainted: G     U
   6.18.0+ #4 PREEMPT(lazy)
   [27386.173185] Tainted: [U]=USER
   [27386.173186] Hardware name: ASUS System Product Name/PRIME B560M-A
   AC, BIOS 2001 02/01/2023
   [27386.173187] Call Trace:
   [27386.173187]  <TASK>
   [27386.173189]  dump_stack_lvl+0x6e/0xa0
   [27386.173192]  print_circular_bug.cold+0x17a/0x1c0
   [27386.173194]  check_noncircular+0x175/0x190
   [27386.173197]  __lock_acquire+0x15e9/0x2820
   [27386.173200]  lock_acquire+0xc9/0x2d0
   [27386.173201]  ? __btrfs_release_delayed_node.part.0+0x39/0x2f0
   [27386.173204]  __mutex_lock+0xcc/0x10a0
   [27386.173206]  ? __btrfs_release_delayed_node.part.0+0x39/0x2f0
   [27386.173208]  ? __btrfs_release_delayed_node.part.0+0x39/0x2f0
   [27386.173211]  ? __btrfs_release_delayed_node.part.0+0x39/0x2f0
   [27386.173213]  __btrfs_release_delayed_node.part.0+0x39/0x2f0
   [27386.173215]  btrfs_evict_inode+0x20b/0x4b0
   [27386.173217]  ? lock_acquire+0xc9/0x2d0
   [27386.173220]  evict+0x15a/0x2f0
   [27386.173222]  prune_icache_sb+0x91/0xd0
   [27386.173224]  super_cache_scan+0x150/0x1d0
   [27386.173226]  do_shrink_slab+0x155/0x6f0
   [27386.173228]  shrink_slab+0x48e/0x890
   [27386.173229]  ? shrink_slab+0x2d2/0x890
   [27386.173231]  shrink_one+0x11a/0x1f0
   [27386.173234]  shrink_node+0xbfd/0x1320
   [27386.173236]  ? shrink_node+0xa2d/0x1320
   [27386.173236]  ? shrink_node+0xbd3/0x1320
   [27386.173239]  ? balance_pgdat+0x67f/0xc60
   [27386.173239]  balance_pgdat+0x67f/0xc60
   [27386.173241]  ? finish_task_switch.isra.0+0xc4/0x2a0
   [27386.173246]  kswapd+0x1dc/0x3e0
   [27386.173247]  ? __pfx_autoremove_wake_function+0x10/0x10
   [27386.173249]  ? __pfx_kswapd+0x10/0x10
   [27386.173250]  kthread+0xff/0x240
   [27386.173251]  ? __pfx_kthread+0x10/0x10
   [27386.173253]  ret_from_fork+0x223/0x280
   [27386.173255]  ? __pfx_kthread+0x10/0x10
   [27386.173257]  ret_from_fork_asm+0x1a/0x30
   [27386.173260]  </TASK>

This is because:

1) The fsync task is holding an inode's delayed node mutex (for a
   directory) while calling __btrfs_update_delayed_inode() and that needs
   to do a search on the subvolume's btree (therefore read lock some
   extent buffers);

2) The lookup task, at btrfs_lookup(), triggered reclaim with the
   GFP_KERNEL allocation done by btrfs_init_file_extent_tree() while
   holding a read lock on a subvolume leaf;

3) The reclaim triggered kswapd which is doing inode eviction for the
   directory inode the fsync task is using as an argument to
   btrfs_commit_inode_delayed_inode() - but in that call chain we are
   trying to read lock the same leaf that the lookup task is holding
   while calling btrfs_init_file_extent_tree() and doing the GFP_KERNEL
   allocation.

Fix this by calling btrfs_init_file_extent_tree() after we don't need the
path anymore and release it in btrfs_read_locked_inode().

Reported-by: Thomas Hellström <[email protected]>
Link: https://lore.kernel.org/linux-btrfs/[email protected]/
Fixes: 8679d26 ("btrfs: initialize inode::file_extent_tree after i_mode has been set")
Reviewed-by: Qu Wenruo <[email protected]>
Signed-off-by: Filipe Manana <[email protected]>
Signed-off-by: David Sterba <[email protected]>

Author: fdmanana

fs/btrfs/inode.c

@@ -4043,11 +4043,6 @@ static int btrfs_read_locked_inode(struct btrfs_inode *inode, struct btrfs_path
 	btrfs_set_inode_mapping_order(inode);
 
 cache_index:
-	ret = btrfs_init_file_extent_tree(inode);
-	if (ret)
-		goto out;
-	btrfs_inode_set_file_extent_range(inode, 0,
-			round_up(i_size_read(vfs_inode), fs_info->sectorsize));
 	/*
 	 * If we were modified in the current generation and evicted from memory
 	 * and then re-read we need to do a full sync since we don't have any
@@ -4134,6 +4129,20 @@ static int btrfs_read_locked_inode(struct btrfs_inode *inode, struct btrfs_path
 				  btrfs_ino(inode), btrfs_root_id(root), ret);
 	}
 
+	/*
+	 * We don't need the path anymore, so release it to avoid holding a read
+	 * lock on a leaf while calling btrfs_init_file_extent_tree(), which can
+	 * allocate memory that triggers reclaim (GFP_KERNEL) and cause a locking
+	 * dependency.
+	 */
+	btrfs_release_path(path);
+
+	ret = btrfs_init_file_extent_tree(inode);
+	if (ret)
+		goto out;
+	btrfs_inode_set_file_extent_range(inode, 0,
+			  round_up(i_size_read(vfs_inode), fs_info->sectorsize));
+
 	if (!maybe_acls)
 		cache_no_acl(vfs_inode);