Skip to contentSkip to conversation

memleaks in btrfs-devel/misc-next

DISCUSSION Johannes Thumshirn 3 participants 5 messages Jan 7, 2020 – Jan 8, 2020 Btrfs file system

lore
Share & tools
lore
#0 lore raw

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

#1 lore raw
Quoted 86 lines · Johannes Thumshirn

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

Quoted 7 lines

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

Thanks for reporting this. I'm looking into it now. Do you know what the refcount of the object is and other state related to it?

Thanks,
Dennis

#2 lore raw
In reply to Johannes Thumshirn
Quoted 86 lines · Johannes Thumshirn

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

Quoted 7 lines

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 <[email protected]>
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().

---
 fs/btrfs/volumes.c | 1 +
 1 file changed, 1 insertion(+)
fs/btrfs/volumes.c +1 -0
index 65e78e59d5c4..eb55df0d4038 100644
@@ -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);
Signature · 1 line
2.17.1
author Tested-by
#3 lore raw

Am 08.01.20 um 01:31 schrieb Dennis Zhou:
[...]

Quoted 3 lines

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 for the quick turn around.

@David can you fold this into
63c3d72cf65e ("btrfs: add the beginning of async discard, discard workqueue")

Quoted 27 lines

--
From: Dennis Zhou <[email protected]>
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().

---
 fs/btrfs/volumes.c | 1 +
 1 file changed, 1 insertion(+)
fs/btrfs/volumes.c +1 -0
index 65e78e59d5c4..eb55df0d4038 100644
@@ -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);
#4 lore raw
Quoted 7 lines · Johannes Thumshirn

Am 08.01.20 um 01:31 schrieb Dennis Zhou:
[...]

Quoted 3 lines

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?

Quoted 6 lines

Thanks for the quick turn around.

@David can you fold this into
63c3d72cf65e ("btrfs: add the beginning of async discard, discard workqueue")

Folded and pushed, thanks for catching it.