Btrfs: fix block group ->space_info null pointer dereference

When we create a block group we add it to the rbtree of block groups
before setting its ->space_info field (while it's NULL). This is
problematic since other tasks can access the block group from the
rbtree and attempt to use its ->space_info before it is set by
btrfs_make_block_group().

This can happen for example when a concurrent fitrim ioctl operation
is ongoing, which produces a trace like the following when
CONFIG_DEBUG_PAGEALLOC is set.

[11509.604369] BUG: unable to handle kernel NULL pointer dereference at 0000000000000018
[11509.606373] IP: [<ffffffff8107d675>] __lock_acquire+0xb4/0xf02
[11509.608179] PGD 2296a8067 PUD 22f4a2067 PMD 0
[11509.608179] Oops: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
[11509.608179] Modules linked in: btrfs crc32c_generic xor raid6_pq nfsd auth_rpcgss oid_registry nfs_acl nfs lockd grace fscache sunrpc loop fuse acpi_cpufreq processor i2c_piix4 psmou
[11509.608179] CPU: 10 PID: 8538 Comm: fstrim Tainted: G        W       4.0.0-rc5-btrfs-next-9+ #2
[11509.608179] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.7.5-0-ge51488c-20140602_164612-nilsson.home.kraxel.org 04/01/2014
[11509.608179] task: ffff88009f5c46d0 ti: ffff8801b3edc000 task.ti: ffff8801b3edc000
[11509.608179] RIP: 0010:[<ffffffff8107d675>]  [<ffffffff8107d675>] __lock_acquire+0xb4/0xf02
[11509.608179] RSP: 0018:ffff8801b3edf9e8  EFLAGS: 00010002
[11509.608179] RAX: 0000000000000046 RBX: 0000000000000000 RCX: 0000000000000000
[11509.608179] RDX: 0000000000000000 RSI: 0000000000000000 RDI: 0000000000000018
[11509.608179] RBP: ffff8801b3edfaa8 R08: 0000000000000001 R09: 0000000000000000
[11509.608179] R10: 0000000000000000 R11: ffff88009f5c4f98 R12: 0000000000000000
[11509.608179] R13: 0000000000000000 R14: 0000000000000018 R15: ffff88009f5c46d0
[11509.608179] FS:  00007f280a10e840(0000) GS:ffff88023ed40000(0000) knlGS:0000000000000000
[11509.608179] CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005003b
[11509.608179] CR2: 0000000000000018 CR3: 00000002119bc000 CR4: 00000000000006e0
[11509.608179] Stack:
[11509.608179]  0000000000000000 0000000000000000 0000000000000004 0000000000000000
[11509.608179]  ffff880100000000 ffffffff00000000 0000000000000001 ffffffff00000000
[11509.608179]  0000000000000001 0000000000000000 ffff880100000000 00000000000006c4
[11509.608179] Call Trace:
[11509.608179]  [<ffffffff8107dc57>] ? __lock_acquire+0x696/0xf02
[11509.608179]  [<ffffffff8107e806>] lock_acquire+0xa5/0x116
[11509.608179]  [<ffffffffa04cc876>] ? do_trimming+0x51/0x145 [btrfs]
[11509.608179]  [<ffffffff81434f37>] _raw_spin_lock+0x34/0x44
[11509.608179]  [<ffffffffa04cc876>] ? do_trimming+0x51/0x145 [btrfs]
[11509.608179]  [<ffffffffa04cc876>] do_trimming+0x51/0x145 [btrfs]
[11509.608179]  [<ffffffffa04cde7d>] btrfs_trim_block_group+0x201/0x491 [btrfs]
[11509.608179]  [<ffffffffa04849e2>] btrfs_trim_fs+0xe0/0x129 [btrfs]
[11509.608179]  [<ffffffffa04bb80a>] btrfs_ioctl_fitrim+0x138/0x167 [btrfs]
[11509.608179]  [<ffffffffa04c002f>] btrfs_ioctl+0x50d/0x21e8 [btrfs]
[11509.608179]  [<ffffffff81123bda>] ? might_fault+0x58/0xb5
[11509.608179]  [<ffffffff81123bda>] ? might_fault+0x58/0xb5
[11509.608179]  [<ffffffff81123bda>] ? might_fault+0x58/0xb5
[11509.608179]  [<ffffffff81158050>] ? cp_new_stat+0x147/0x15e
[11509.608179]  [<ffffffff81163041>] do_vfs_ioctl+0x3c6/0x479
[11509.608179]  [<ffffffff81158116>] ? SYSC_newfstat+0x25/0x2e
[11509.608179]  [<ffffffff81435b54>] ? ret_from_sys_call+0x1d/0x58
[11509.608179]  [<ffffffff8116b915>] ? __fget_light+0x2d/0x4f
[11509.608179]  [<ffffffff8116314e>] SyS_ioctl+0x5a/0x7f
[11509.608179]  [<ffffffff81435b32>] system_call_fastpath+0x12/0x17
[11509.608179] Code: f4 01 00 0f 85 c0 00 00 00 48 c7 c1 f3 1f 7d 81 48 c7 c2 aa cb 7c 81 be fc 0b 00 00 eb 70 83 3d 61 eb 9c 00 00 0f 84 a5 00 00 00 <49> 81 3e 40 a3 2b 82 b8 00 00 00
[11509.608179] RIP  [<ffffffff8107d675>] __lock_acquire+0xb4/0xf02
[11509.608179]  RSP <ffff8801b3edf9e8>
[11509.608179] CR2: 0000000000000018
[11509.608179] ---[ end trace 570a5c6769f0e49a ]---

Which corresponds to the following access in fs/btrfs/free-space-cache.c:

  static int do_trimming(struct btrfs_block_group_cache *block_group,
                         u64 *total_trimmed, u64 start, u64 bytes,
                         u64 reserved_start, u64 reserved_bytes,
                         struct btrfs_trim_range *trim_entry)
  {
       struct btrfs_space_info *space_info = block_group->space_info;
  (...)
       spin_lock(&space_info->lock);
       ^^^^^ - block_group->space_info is NULL...

Fix this by ensuring the block group's ->space_info is set before adding
the block group to the rbtree.

Signed-off-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>

diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 0ec3acd14cbf5e1273f331231f09165710d80d91..1cbc71d8cb96946beb166f5e59fd8525d58067fb 100644 (file)
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -3693,7 +3693,8 @@ static int update_space_info(struct btrfs_fs_info *info, u64 flags,
                found->disk_total += total_bytes * factor;
                found->bytes_used += bytes_used;
                found->disk_used += bytes_used * factor;
-               found->full = 0;
+               if (total_bytes > 0)
+                       found->full = 0;
                spin_unlock(&found->lock);
                *space_info = found;
                return 0;
@@ -3721,7 +3722,10 @@ static int update_space_info(struct btrfs_fs_info *info, u64 flags,
        found->bytes_reserved = 0;
        found->bytes_readonly = 0;
        found->bytes_may_use = 0;
-       found->full = 0;
+       if (total_bytes > 0)
+               found->full = 0;
+       else
+               found->full = 1;
        found->force_alloc = CHUNK_ALLOC_NO_FORCE;
        found->chunk_alloc = 0;
        found->flush = 0;
@@ -9562,6 +9566,19 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
 
        free_excluded_extents(root, cache);
 
+       /*
+        * Call to ensure the corresponding space_info object is created and
+        * assigned to our block group, but don't update its counters just yet.
+        * We want our bg to be added to the rbtree with its ->space_info set.
+        */
+       ret = update_space_info(root->fs_info, cache->flags, 0, 0,
+                               &cache->space_info);
+       if (ret) {
+               btrfs_remove_free_space_cache(cache);
+               btrfs_put_block_group(cache);
+               return ret;
+       }
+
        ret = btrfs_add_block_group_cache(root->fs_info, cache);
        if (ret) {
                btrfs_remove_free_space_cache(cache);
@@ -9569,6 +9586,10 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
                return ret;
        }
 
+       /*
+        * Now that our block group has its ->space_info set and is inserted in
+        * the rbtree, update the space info's counters.
+        */
        ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
                                &cache->space_info);
        if (ret) {