On Tue, Jan 07, 2020 at 12:43:30PM +0100, Johannes Thumshirn wrote:
> Hi Dennis,
>
> When I'm running btrfs/003 and btrfs/004 in my test setup I get the
> following kmemleak report:
>
> rapido1:/home/johannes/src/xfstests-dev# cat results/btrfs/004.kmemleak
> EXPERIMENTAL kmemleak reported some memory leaks! Due to the way kmemleak
> works, the leak might be from an earlier test, or something totally
> unrelated.
> unreferenced object 0xffff88821eee3c00 (size 1024):
> comm "mount", pid 7247, jiffies 4294900263 (age 26.468s)
> hex dump (first 32 bytes):
> 00 b0 ca 2a 82 88 ff ff 00 00 00 00 00 00 00 00 ...*............
> 00 00 00 00 00 00 00 00 00 00 10 00 00 00 00 00 ................
> backtrace:
> [<00000000f993bddc>] btrfs_create_block_group_cache+0x20/0x1a0 [btrfs]
> [<00000000e0df8aa6>] btrfs_read_block_groups+0x42f/0x780 [btrfs]
> [<00000000792a6ecd>] open_ctree+0x17a8/0x1e93 [btrfs]
> [<0000000042fa9fb8>] btrfs_mount_root+0x4e1/0x5a0 [btrfs]
> [<00000000586791fa>] legacy_get_tree+0x22/0x40
> [<00000000cb72e180>] vfs_get_tree+0x1b/0x90
> [<0000000059613ba1>] fc_mount+0x9/0x30
> [<00000000a752e56e>] vfs_kern_mount.part.40+0x6a/0x80
> [<000000000524dce6>] btrfs_mount+0x138/0x860 [btrfs]
> [<00000000586791fa>] legacy_get_tree+0x22/0x40
> [<00000000cb72e180>] vfs_get_tree+0x1b/0x90
> [<000000005f5112f8>] do_mount+0x674/0x900
> [<000000007ead6809>] __x64_sys_mount+0x81/0xd0
> [<0000000028386837>] do_syscall_64+0x43/0x140
> [<00000000e8583d73>] entry_SYSCALL_64_after_hwframe+0x44/0xa9
> unreferenced object 0xffff88821cbdfb40 (size 192):
> comm "mount", pid 7247, jiffies 4294900263 (age 26.468s)
> hex dump (first 32 bytes):
> 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
> 00 00 00 00 00 00 00 00 cc 00 00 00 00 00 00 00 ................
> backtrace:
> [<0000000042643477>] btrfs_create_block_group_cache+0x3d/0x1a0 [btrfs]
> [<00000000e0df8aa6>] btrfs_read_block_groups+0x42f/0x780 [btrfs]
> [<00000000792a6ecd>] open_ctree+0x17a8/0x1e93 [btrfs]
> [<0000000042fa9fb8>] btrfs_mount_root+0x4e1/0x5a0 [btrfs]
> [<00000000586791fa>] legacy_get_tree+0x22/0x40
> [<00000000cb72e180>] vfs_get_tree+0x1b/0x90
> [<0000000059613ba1>] fc_mount+0x9/0x30
> [<00000000a752e56e>] vfs_kern_mount.part.40+0x6a/0x80
> [<000000000524dce6>] btrfs_mount+0x138/0x860 [btrfs]
> [<00000000586791fa>] legacy_get_tree+0x22/0x40
> [<00000000cb72e180>] vfs_get_tree+0x1b/0x90
> [<000000005f5112f8>] do_mount+0x674/0x900
> [<000000007ead6809>] __x64_sys_mount+0x81/0xd0
> [<0000000028386837>] do_syscall_64+0x43/0x140
> [<00000000e8583d73>] entry_SYSCALL_64_after_hwframe+0x44/0xa9
>
>
> I've bisected it down to:
> 63c3d72cf65e ("btrfs: add the beginning of async discard, discard
> workqueue")
>
> The backtrace points to this:
> (gdb) l *(btrfs_create_block_group_cache+0x20)
> 0xa8060 is in btrfs_create_block_group_cache (fs/btrfs/block-group.c:1641).
> 1636 struct btrfs_fs_info *fs_info, u64 start, u64 size)
> 1637 {
> 1638 struct btrfs_block_group *cache;
> 1639
> 1640 cache = kzalloc(sizeof(*cache), GFP_NOFS);
> 1641 if (!cache)
> 1642 return NULL;
> 1643
> 1644 cache->free_space_ctl =
> kzalloc(sizeof(*cache->free_space_ctl),
> 1645 GFP_NOFS);
> (gdb) l *(btrfs_create_block_group_cache+0x3d)
> 0xa807d is in btrfs_create_block_group_cache (fs/btrfs/block-group.c:1646).
> 1641 if (!cache)
> 1642 return NULL;
> 1643
> 1644 cache->free_space_ctl =
> kzalloc(sizeof(*cache->free_space_ctl),
> 1645 GFP_NOFS);
> 1646 if (!cache->free_space_ctl) {
> 1647 kfree(cache);
> 1648 return NULL;
> 1649 }
> 1650
>
>
> so we're leaking both cache and cache->free_space_ctl.
>
> I'm looking into it right now but maybe you're faster seeing what's
> missing here.
>
> Byte,
> Johannes
I believe it's because I forgot to put a reference in the relocation
path. The below seems to fix it in my tests, but would you mind
verifying?
Thanks,
Dennis
--
From: Dennis Zhou <dennis@xxxxxxxxxx>
Date: Tue, 7 Jan 2020 14:14:04 -0800
Subject: [PATCH] btrfs: put lookup reference in btrfs_relocate_chunk()
Async discard requires looking up the block_group in the relocation path
to cancel any work items against it. However, I forgot to put the
reference from btrfs_lookup_block_group().
Reported-by: Johannes Thumshirn <jth@xxxxxxxxxx>
Signed-off-by: Dennis Zhou <dennis@xxxxxxxxxx>
---
fs/btrfs/volumes.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 65e78e59d5c4..eb55df0d4038 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -2898,6 +2898,7 @@ static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
if (!block_group)
return -ENOENT;
btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
+ btrfs_put_block_group(block_group);
trans = btrfs_start_trans_remove_block_group(root->fs_info,
chunk_offset);
--
2.17.1
