btrfs: drop logs when we've aborted a transaction
authorJosef Bacik <josef@toxicpanda.com>
Tue, 24 Mar 2020 14:47:52 +0000 (10:47 -0400)
committerDavid Sterba <dsterba@suse.com>
Thu, 23 Apr 2020 15:23:11 +0000 (17:23 +0200)
Dave reported a problem where we were panicing with generic/475 with
misc-5.7.  This is because we were doing IO after we had stopped all of
the worker threads, because we do the log tree cleanup on roots at drop
time.  Cleaning up the log tree will always need to do reads if we
happened to have evicted the blocks from memory.

Because of this simply add a helper to btrfs_cleanup_transaction() that
will go through and drop all of the log roots.  This gets run before we
do the close_ctree() work, and thus we are allowed to do any reads that
we would need.  I ran this through many iterations of generic/475 with
constrained memory and I did not see the issue.

  general protection fault, probably for non-canonical address 0x6b6b6b6b6b6b6b6b: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC PTI
  CPU: 2 PID: 12359 Comm: umount Tainted: G        W 5.6.0-rc7-btrfs-next-58 #1
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.0-59-gc9ba5276e321-prebuilt.qemu.org 04/01/2014
  RIP: 0010:btrfs_queue_work+0x33/0x1c0 [btrfs]
  RSP: 0018:ffff9cfb015937d8 EFLAGS: 00010246
  RAX: 0000000000000000 RBX: ffff8eb5e339ed80 RCX: 0000000000000000
  RDX: 0000000000000001 RSI: ffff8eb5eb33b770 RDI: ffff8eb5e37a0460
  RBP: ffff8eb5eb33b770 R08: 000000000000020c R09: ffffffff9fc09ac0
  R10: 0000000000000007 R11: 0000000000000000 R12: 6b6b6b6b6b6b6b6b
  R13: ffff9cfb00229040 R14: 0000000000000008 R15: ffff8eb5d3868000
  FS:  00007f167ea022c0(0000) GS:ffff8eb5fae00000(0000) knlGS:0000000000000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 00007f167e5e0cb1 CR3: 0000000138c18004 CR4: 00000000003606e0
  DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
  DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
  Call Trace:
   btrfs_end_bio+0x81/0x130 [btrfs]
   __split_and_process_bio+0xaf/0x4e0 [dm_mod]
   ? percpu_counter_add_batch+0xa3/0x120
   dm_process_bio+0x98/0x290 [dm_mod]
   ? generic_make_request+0xfb/0x410
   dm_make_request+0x4d/0x120 [dm_mod]
   ? generic_make_request+0xfb/0x410
   generic_make_request+0x12a/0x410
   ? submit_bio+0x38/0x160
   submit_bio+0x38/0x160
   ? percpu_counter_add_batch+0xa3/0x120
   btrfs_map_bio+0x289/0x570 [btrfs]
   ? kmem_cache_alloc+0x24d/0x300
   btree_submit_bio_hook+0x79/0xc0 [btrfs]
   submit_one_bio+0x31/0x50 [btrfs]
   read_extent_buffer_pages+0x2fe/0x450 [btrfs]
   btree_read_extent_buffer_pages+0x7e/0x170 [btrfs]
   walk_down_log_tree+0x343/0x690 [btrfs]
   ? walk_log_tree+0x3d/0x380 [btrfs]
   walk_log_tree+0xf7/0x380 [btrfs]
   ? plist_requeue+0xf0/0xf0
   ? delete_node+0x4b/0x230
   free_log_tree+0x4c/0x130 [btrfs]
   ? wait_log_commit+0x140/0x140 [btrfs]
   btrfs_free_log+0x17/0x30 [btrfs]
   btrfs_drop_and_free_fs_root+0xb0/0xd0 [btrfs]
   btrfs_free_fs_roots+0x10c/0x190 [btrfs]
   ? do_raw_spin_unlock+0x49/0xc0
   ? _raw_spin_unlock+0x29/0x40
   ? release_extent_buffer+0x121/0x170 [btrfs]
   close_ctree+0x289/0x2e6 [btrfs]
   generic_shutdown_super+0x6c/0x110
   kill_anon_super+0xe/0x30
   btrfs_kill_super+0x12/0x20 [btrfs]
   deactivate_locked_super+0x3a/0x70

Reported-by: David Sterba <dsterba@suse.com>
Fixes: 8c38938c7bb096 ("btrfs: move the root freeing stuff into btrfs_put_root")
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: David Sterba <dsterba@suse.com>
fs/btrfs/disk-io.c

index a6cb5cb..d10c7be 100644 (file)
@@ -2036,9 +2036,6 @@ void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
                for (i = 0; i < ret; i++)
                        btrfs_drop_and_free_fs_root(fs_info, gang[i]);
        }
-
-       if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
-               btrfs_free_log_root_tree(NULL, fs_info);
 }
 
 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
@@ -3888,7 +3885,7 @@ void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
        spin_unlock(&fs_info->fs_roots_radix_lock);
 
        if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
-               btrfs_free_log(NULL, root);
+               ASSERT(root->log_root == NULL);
                if (root->reloc_root) {
                        btrfs_put_root(root->reloc_root);
                        root->reloc_root = NULL;
@@ -4211,6 +4208,36 @@ static void btrfs_error_commit_super(struct btrfs_fs_info *fs_info)
        up_write(&fs_info->cleanup_work_sem);
 }
 
+static void btrfs_drop_all_logs(struct btrfs_fs_info *fs_info)
+{
+       struct btrfs_root *gang[8];
+       u64 root_objectid = 0;
+       int ret;
+
+       spin_lock(&fs_info->fs_roots_radix_lock);
+       while ((ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
+                                            (void **)gang, root_objectid,
+                                            ARRAY_SIZE(gang))) != 0) {
+               int i;
+
+               for (i = 0; i < ret; i++)
+                       gang[i] = btrfs_grab_root(gang[i]);
+               spin_unlock(&fs_info->fs_roots_radix_lock);
+
+               for (i = 0; i < ret; i++) {
+                       if (!gang[i])
+                               continue;
+                       root_objectid = gang[i]->root_key.objectid;
+                       btrfs_free_log(NULL, gang[i]);
+                       btrfs_put_root(gang[i]);
+               }
+               root_objectid++;
+               spin_lock(&fs_info->fs_roots_radix_lock);
+       }
+       spin_unlock(&fs_info->fs_roots_radix_lock);
+       btrfs_free_log_root_tree(NULL, fs_info);
+}
+
 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
 {
        struct btrfs_ordered_extent *ordered;
@@ -4603,6 +4630,7 @@ static int btrfs_cleanup_transaction(struct btrfs_fs_info *fs_info)
        btrfs_destroy_delayed_inodes(fs_info);
        btrfs_assert_delayed_root_empty(fs_info);
        btrfs_destroy_all_delalloc_inodes(fs_info);
+       btrfs_drop_all_logs(fs_info);
        mutex_unlock(&fs_info->transaction_kthread_mutex);
 
        return 0;