Hi Xiao,
On 10/8/22 5:06 PM, Xiao Ni wrote:
Hi all
Sometimes it will have a memory leak problem when raid10 takeover raid0.
And then panic.
[ 6973.704097] BUG bio-176 (Not tainted): Objects remaining in bio-176
on __kmem_cache_shutdown()
[ 6973.705627] -----------------------------------------------------------------------------
[ 6973.705627]
[ 6973.707331] Slab 0x0000000067cd8c6c objects=21 used=1
fp=0x000000005bf568c8
flags=0xfffffc0000100(slab|node=0|zone=1|lastcpupid=0x1fffff)
[ 6973.709498] CPU: 0 PID: 735535 Comm: mdadm Kdump: loaded Not
tainted 4.18.0-423.el8.x86_64 #1
It is an old kernel, does it also happen to latest mainline version? Not
sure mdadm test suite
covers this case.
[ 6973.711027] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011
[ 6973.712068] Call Trace:
[ 6973.712550] dump_stack+0x41/0x60
[ 6973.713185] slab_err.cold.117+0x53/0x67
[ 6973.713878] ? __unfreeze_partials+0x1a0/0x1a0
[ 6973.714659] ? cpumask_next+0x17/0x20
[ 6973.715309] __kmem_cache_shutdown+0x16e/0x300
[ 6973.716097] kmem_cache_destroy+0x4d/0x120
[ 6973.716843] bioset_exit+0xb0/0x100
[ 6973.717486] level_store+0x280/0x650
[ 6973.718147] ? security_capable+0x38/0x60
[ 6973.718856] md_attr_store+0x7c/0xc0
[ 6973.719489] kernfs_fop_write+0x11e/0x1a0
[ 6973.720200] vfs_write+0xa5/0x1b0
[ 6973.720785] ksys_write+0x4f/0xb0
[ 6973.721374] do_syscall_64+0x5b/0x1b0
[ 6973.722028] entry_SYSCALL_64_after_hwframe+0x61/0xc6
[ 6973.723005] RIP: 0033:0x7fb15d2b5bc8
[ 6973.723660] Code: 89 02 48 c7 c0 ff ff ff ff eb b3 0f 1f 80 00 00
00 00 f3 0f 1e fa 48 8d 05 55 4b 2a 00 8b 00 85 c0 75 17 b8 01 00 00
00 0f 05 <48> 3d 00 f0 ff ff 77 58 c3 0f 1f 80 00 00 00 00 41 54 49 89
d4 55
[ 6973.726926] RSP: 002b:00007ffed8ebc8a8 EFLAGS: 00000246 ORIG_RAX:
0000000000000001
[ 6973.728278] RAX: ffffffffffffffda RBX: 0000000000000004 RCX: 00007fb15d2b5bc8
[ 6973.729515] RDX: 0000000000000006 RSI: 0000560330046dd7 RDI: 0000000000000004
[ 6973.730748] RBP: 0000560330046dd7 R08: 000056033003b927 R09: 00007fb15d315d40
[ 6973.731986] R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000
[ 6973.733240] R13: 00007ffed8ebcc80 R14: 00007ffed8ebe8ee R15: 0000000000000000
[ 6973.734520] Disabling lock debugging due to kernel taint
[ 6973.735452] Object 0x000000000923869f @offset=3264
[ 6973.736317] kmem_cache_destroy bio-176: Slab cache still has objects
[ 6973.737479] CPU: 0 PID: 735535 Comm: mdadm Kdump: loaded Tainted: G
B --------- - - 4.18.0-423.el8.x86_64 #1
[ 6973.739438] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011
[ 6973.740437] Call Trace:
[ 6973.740877] dump_stack+0x41/0x60
[ 6973.741473] kmem_cache_destroy+0x116/0x120
[ 6973.742209] bioset_exit+0xb0/0x100
[ 6973.742824] level_store+0x280/0x650
[ 6973.743463] ? security_capable+0x38/0x60
[ 6973.744177] md_attr_store+0x7c/0xc0
[ 6973.744807] kernfs_fop_write+0x11e/0x1a0
[ 6973.745516] vfs_write+0xa5/0x1b0
[ 6973.746112] ksys_write+0x4f/0xb0
[ 6973.746695] do_syscall_64+0x5b/0x1b0
[ 6973.747373] entry_SYSCALL_64_after_hwframe+0x61/0xc6
[ 6973.748293] RIP: 0033:0x7fb15d2b5bc8
[ 6973.748929] Code: 89 02 48 c7 c0 ff ff ff ff eb b3 0f 1f 80 00 00
00 00 f3 0f 1e fa 48 8d 05 55 4b 2a 00 8b 00 85 c0 75 17 b8 01 00 00
00 0f 05 <48> 3d 00 f0 ff ff 77 58 c3 0f 1f 80 00 00 00 00 41 54 49 89
d4 55
[ 6973.752255] RSP: 002b:00007ffed8ebc8a8 EFLAGS: 00000246 ORIG_RAX:
0000000000000001
[ 6973.753628] RAX: ffffffffffffffda RBX: 0000000000000004 RCX: 00007fb15d2b5bc8
[ 6973.754900] RDX: 0000000000000006 RSI: 0000560330046dd7 RDI: 0000000000000004
[ 6973.756142] RBP: 0000560330046dd7 R08: 000056033003b927 R09: 00007fb15d315d40
[ 6973.757420] R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000
[ 6973.758702] R13: 00007ffed8ebcc80 R14: 00007ffed8ebe8ee R15: 0000000000000000
[ 6973.762429] BUG: unable to handle kernel NULL pointer dereference
at 0000000000000000
[ 6973.763818] PGD 0 P4D 0
[ 6973.764277] Oops: 0002 [#1] SMP PTI
[ 6973.764896] CPU: 0 PID: 12 Comm: ksoftirqd/0 Kdump: loaded Tainted:
G B --------- - - 4.18.0-423.el8.x86_64 #1
[ 69 it73.766944] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011
[ 6973.767999] RIP: 0010:mempool_free+0x52/0x80
[ 6973.768779] Code: e4 6f 96 00 e9 0f 72 96 00 48 89 f7 e8 a7 5e 75
00 48 63 53 08 3b 53 04 7d 30 48 8b 4b 10 8d 72 01 48 89 df 89 73 08
48 89 c6 <48> 89 2c d1 e8 f5 5e 75 00 48 8d 7b 30 31 c9 5b ba 01 00 00
00 be
[ 6973.772088] RSP: 0018:ffffbc6c4068fdf8 EFLAGS: 00010093
[ 6973.773026] RAX: 0000000000000293 RBX: ffff9b5844fed7d8 RCX: 0000000000000000
[ 6973.774318] RDX: 0000000000000000 RSI: 0000000000000293 RDI: ffff9b5844fed7d8
[ 6973.775560] RBP: ffff9b5759046cc0 R08: 000000000000003d R09: ffff9b57420e62a0
[ 6973.776794] R10: 0000000000000008 R11: ffff9b574a602a00 R12: ffff9b5845004ed0
[ 6973.778096] R13: 0000000000001000 R14: 0000000000000000 R15: 0000000000000000
[ 6973.779357] FS: 0000000000000000(0000) GS:ffff9b5870000000(0000)
knlGS:0000000000000000
[ 6973.780757] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 6973.781806] CR2: 0000000000000000 CR3: 0000000043410002 CR4: 00000000007706f0
[ 6973.783076] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 6973.784313] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 6973.785593] PKRU: 55555554
[ 6973.786098] Call Trace:
[ 6973.786549] md_end_io_acct+0x31/0x40
[ 6973.787227] blk_update_request+0x224/0x380
[ 6973.787994] blk_mq_end_request+0x1a/0x130
[ 6973.788739] blk_complete_reqs+0x35/0x50
[ 6973.789456] __do_softirq+0xd7/0x2c8
[ 6973.790114] ? sort_range+0x20/0x20
[ 6973.790763] run_ksoftirqd+0x2a/0x40
[ 6973.791400] smpboot_thread_fn+0xb5/0x150
[ 6973.792114] kthread+0x10b/0x130
[ 6973.792724] ? set_kthread_struct+0x50/0x50
[ 6973.793491] ret_from_fork+0x1f/0x40
The reason is that raid0 doesn't do anything in raid0_quiesce. During
the level change in level_store, it needs
to make sure all inflight bios to finish. We can add a count to do
this in raid0. Is it a good way?
There is a count mddev->active_io in mddev. But it's decreased in
md_handle_request rather than in the callback
endio function. Is it right to decrease active_io in callbcak endio function?
I think it is a race between ios and takeover action, since
mddev_suspend called by
level_store should ensure no io is submitted to array at that time by below.
wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0)
However it can't guarantee new io comes after mddev_suspend because the
empty
raid0 quiesce. Maybe we can do something like this.
diff --git a/drivers/md/md.c b/drivers/md/md.c
index afaf36b2f6ab..0fdc07008bcf 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -3891,6 +3891,7 @@ level_store(struct mddev *mddev, const char *buf,
size_t len)
if (rv)
return rv;
+ set_bit(MD_TAKENOVER, &mddev->flags);
if (mddev->pers == NULL) {
strncpy(mddev->clevel, buf, slen);
if (mddev->clevel[slen-1] == '\n')
@@ -4074,6 +4075,7 @@ level_store(struct mddev *mddev, const char *buf,
size_t len)
rv = len;
out_unlock:
mddev_unlock(mddev);
+ clear_bit(MD_TAKENOVER, &mddev->flags);
return rv;
}
diff --git a/drivers/md/md.h b/drivers/md/md.h
index b4e2d8b87b61..e77da3bc6a72 100644
--- a/drivers/md/md.h
+++ b/drivers/md/md.h
@@ -255,6 +255,7 @@ struct md_cluster_info;
* array is ready yet.
* @MD_BROKEN: This is used to stop writes and mark array as failed.
* @MD_DELETED: This device is being deleted
+ * @MD_TAKENOVER: This device is being taken over to another personality.
*
* change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added
*/
@@ -272,6 +273,7 @@ enum mddev_flags {
MD_NOT_READY,
MD_BROKEN,
MD_DELETED,
+ MD_TAKENOVER,
};
enum mddev_sb_flags {
diff --git a/drivers/md/raid0.c b/drivers/md/raid0.c
index 78addfe4a0c9..b00ae87c42d3 100644
--- a/drivers/md/raid0.c
+++ b/drivers/md/raid0.c
@@ -523,6 +523,9 @@ static bool raid0_make_request(struct mddev *mddev,
struct bio *bio)
unsigned chunk_sects;
unsigned sectors;
+ if (test_bit(MD_TAKENOVER, &mddev->flags))
+ return true;
+
if (unlikely(bio->bi_opf & REQ_PREFLUSH)
&& md_flush_request(mddev, bio))
return true;
Another way is set mddev->pers to NULL before mddev_suspend in
level_store, then move
"mddev->pers = pers" after mddev_resume, this way makes md_submit_bio
returns earlier.
I guess this way is more simple and straightforward.
diff --git a/drivers/md/md.c b/drivers/md/md.c
index afaf36b2f6ab..5a7134e3c182 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -3979,6 +3979,9 @@ level_store(struct mddev *mddev, const char *buf,
size_t len)
goto out_unlock;
}
+ /* Only accept IO after takeover is done */
+ mddev->pers = NULL;
+
/* Looks like we have a winner */
mddev_suspend(mddev);
mddev_detach(mddev);
@@ -3986,7 +3989,6 @@ level_store(struct mddev *mddev, const char *buf,
size_t len)
spin_lock(&mddev->lock);
oldpers = mddev->pers;
oldpriv = mddev->private;
- mddev->pers = pers;
mddev->private = priv;
strscpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
mddev->level = mddev->new_level;
@@ -4067,6 +4069,7 @@ level_store(struct mddev *mddev, const char *buf,
size_t len)
pers->run(mddev);
set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
mddev_resume(mddev);
+ mddev->pers = pers;
if (!mddev->thread)
md_update_sb(mddev, 1);
sysfs_notify_dirent_safe(mddev->sysfs_level);
Thanks,
Guoqing
