[cc io_uring]
On Thu, May 12, 2022 at 06:52:37PM +0200, Christian Brauner wrote:
> From: "Christian Brauner (Microsoft)" <brauner@xxxxxxxxxx>
>
> Hey everyone,
>
> Please note that this patch series contains patches that will be
> rejected by the fstests mailing list because of the amount of changes
> they contain. So tools like b4 will not be able to find the whole patch
> series on a mailing list. In case it's helpful I've added the
> "fstests.vfstest.for-next" tag which can be pulled. Otherwise it's
> possible to simply use the patch series as it appears in your inbox.
>
> All vfstests pass:
[...]
> #### xfs ####
> ubuntu@imp1-vm:~/src/git/xfstests$ sudo ./check -g idmapped
> FSTYP -- xfs (debug)
> PLATFORM -- Linux/x86_64 imp1-vm 5.18.0-rc4-fs-mnt-hold-writers-8a2e2350494f #107 SMP PREEMPT_DYNAMIC Mon May 9 12:12:34 UTC 2022
> MKFS_OPTIONS -- -f /dev/sda4
> MOUNT_OPTIONS -- /dev/sda4 /mnt/scratch
>
> generic/633 58s ... 58s
> generic/644 62s ... 60s
> generic/645 161s ... 161s
> generic/656 62s ... 63s
> xfs/152 133s ... 133s
> xfs/153 94s ... 92s
> Ran: generic/633 generic/644 generic/645 generic/656 xfs/152 xfs/153
> Passed all 6 tests
I'm not sure if it's this series that has introduced a test bug or
triggered a latent issue in the kernel, but I've started seeing
generic/633 throw audit subsystem warnings on a single test machine
as of late Friday:
[ 7285.015888] WARNING: CPU: 3 PID: 2147118 at kernel/auditsc.c:2035 __audit_syscall_entry+0x113/0x140
[ 7285.019973] Modules linked in:
[ 7285.021281] CPU: 3 PID: 2147118 Comm: vfstest Not tainted 5.18.0-rc7-dgc+ #1250
[ 7285.024341] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.15.0-1 04/01/2014
[ 7285.027782] RIP: 0010:__audit_syscall_entry+0x113/0x140
[ 7285.029923] Code: 24 e8 c1 6b ff ff 48 8b 34 24 85 c0 48 8b 54 24 08 48 8b 4c 24 10 4c 8b 44 24 18 0f 84 72 ff ff ff 48 83 c4 20 5b 5d 41 5c c3 <0f> 0b 85 c0 75 14 48 83 c4 20 48 c7 c7 70 45 7f 82 5b 5d 41 5c e9
[ 7285.037563] RSP: 0018:ffffc900012f7ed0 EFLAGS: 00010202
[ 7285.039748] RAX: 0000000000000000 RBX: ffff8880aaf18800 RCX: 000000000000003c
[ 7285.043126] RDX: 00000000000000e7 RSI: 0000000000000000 RDI: ffff888102104f00
[ 7285.046120] RBP: 00000000000000e7 R08: fffffffffffffe2c R09: 0000000000000002
[ 7285.049108] R10: 0000000000000001 R11: 0000000000000000 R12: 0000000000000000
[ 7285.052058] R13: ffffc900012f7f58 R14: 00000000000000e7 R15: 0000000000000000
[ 7285.055030] FS: 00007f7906d6c740(0000) GS:ffff88813bd80000(0000) knlGS:0000000000000000
[ 7285.058396] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 7285.060788] CR2: 00007f3ffa7e9bb8 CR3: 000000010bb00000 CR4: 00000000000006e0
[ 7285.063735] Call Trace:
[ 7285.064796] <TASK>
[ 7285.065759] syscall_trace_enter.constprop.0+0x122/0x1a0
[ 7285.067978] do_syscall_64+0x16/0x80
[ 7285.069497] entry_SYSCALL_64_after_hwframe+0x44/0xae
[ 7285.071590] RIP: 0033:0x7f7906e35f49
[ 7285.073118] Code: 00 4c 8b 05 29 6f 10 00 be e7 00 00 00 ba 3c 00 00 00 eb 12 0f 1f 44 00 00 89 d0 0f 05 48 3d 00 f0 ff ff 77 1c f4 89 f0 0f 05 <48> 3d 00 f0 ff ff 76 e7 f7 d8 64 41 89 00 eb df 0f 1f 80 00 00 00
[ 7285.078021] RSP: 002b:00007ffeee52db88 EFLAGS: 00000246 ORIG_RAX: 00000000000000e7
[ 7285.079995] RAX: ffffffffffffffda RBX: 00007f7906f39920 RCX: 00007f7906e35f49
[ 7285.081869] RDX: 000000000000003c RSI: 00000000000000e7 RDI: 0000000000000000
[ 7285.083729] RBP: 0000000000000000 R08: ffffffffffffff88 R09: 0000000000000001
[ 7285.085594] R10: fffffffffffffe2c R11: 0000000000000246 R12: 00007f7906f39920
[ 7285.087457] R13: 0000000000000001 R14: 00007f7906f3ee28 R15: 0000000000000000
[ 7285.089320] </TASK>
[ 7285.089949] ---[ end trace 0000000000000000 ]---
[ 7285.091182] audit_panic: 22 callbacks suppressed
[ 7285.091185] audit: unrecoverable error in audit_syscall_entry()
Adn faddr_to_line tells me it is this:
void __audit_syscall_entry(int major, unsigned long a1, unsigned long a2,
unsigned long a3, unsigned long a4)
{
struct audit_context *context = audit_context();
enum audit_state state;
if (!audit_enabled || !context)
return;
>>>>>> WARN_ON(context->context != AUDIT_CTX_UNUSED); <<<<<<
WARN_ON(context->name_count);
if (context->context != AUDIT_CTX_UNUSED || context->name_count) {
audit_panic("unrecoverable error in audit_syscall_entry()");
return;
}
.....
I have no clue how the audit subsystem works, I don't even know how
to enable it, and I've never seen audit messages on the console of
this test machine. I have other test machines that have audit
enabled, and they do not dump warnings like this - the only thing I
see in the logs for those machines is this:
run xfstest generic/633
process 'vfstest' launched '/dev/fd/4/file1' with NULL argv: empty string added
XFS (pmem0): Unmounting Filesystem
XFS (pmem0): Mounting V5 Filesystem
XFS (pmem0): Ending clean mount
run xfstest generic/634
That's waht I was seeing from this test machine earlier in the week,
too - I've been running 5.18-rc7 as the base kernel all week - so
I'm not sure .....
Oooooohhhh.
/* The per-task audit context. */
struct audit_context {
int dummy; /* must be the first element */
enum {
AUDIT_CTX_UNUSED, /* audit_context is currently unused */
AUDIT_CTX_SYSCALL, /* in use by syscall */
AUDIT_CTX_URING, /* in use by io_uring */
} context;
....
And that reminded me of something. I commented on #xfs on Friday
afternoon:
[20/5/22 15:04] <dchinner> so of the 3.5 hours run time on the machine that jsut completed, it looks like about a dozen tests are responsible for an hour of that runtime...
[20/5/22 15:05] <dchinner> but it was a clean run with no failures in 1055 tests run.
[20/5/22 15:06] <dchinner> But there's some WTFs like this in it:
[20/5/22 15:06] <dchinner> generic/678 [not run] kernel does not support IO_URING
[20/5/22 15:08] <dchinner> yet the same kernel on a different machine:
[20/5/22 15:08] <dchinner> generic/678 11s ... 3s
[20/5/22 15:08] <dchinner> and they have the same userspace, too....
Yeah, the machine that complained about "kernel does not support
IO_URING" is the one that is throwing these warnings now. It wasn't
that the kernel didn't support io-uring, it was that the machine was
missing the liburing-dev library. I installed it and rebuilt
fstests. These audit failures co-incide with the first test runs
with io-uring enabled. And vfstest uses io_uring if fstests enables
it.
Hence this now smells like a pre-existing issue - either a test bug
or an io_uring task audit context leak. Anyone got any ideas?
Cheers,
Dave.
--
Dave Chinner
david@xxxxxxxxxxxxx
