Bug 1465619

Summary:	kernel/tcmu-runner: fix uio_poll crash during uio device removal
Product:	Red Hat Enterprise Linux 7	Reporter:	Mike Christie <mchristi>
Component:	tcmu-runner	Assignee:	Mike Christie <mchristi>
Status:	CLOSED NOTABUG	QA Contact:	Storage QE <storage-qe>
Severity:	unspecified	Docs Contact:
Priority:	unspecified
Version:	7.4	CC:	mchristi
Target Milestone:	rc
Target Release:	---
Hardware:	Unspecified
OS:	Unspecified
Whiteboard:
Fixed In Version:		Doc Type:	If docs needed, set a value
Doc Text:		Story Points:	---
Clone Of:
Clones:	1474188 (view as bug list)		Environment:
Last Closed:	2017-09-06 21:56:37 UTC	Type:	Bug
Regression:	---	Mount Type:	---
Documentation:	---	CRM:
Verified Versions:		Category:	---
oVirt Team:	---	RHEL 7.3 requirements from Atomic Host:
Cloudforms Team:	---	Target Upstream Version:
Embargoed:
Bug Depends On:
Bug Blocks:	1474188

Description Mike Christie 2017-06-27 19:06:03 UTC

Description of problem:

During uio device removal we hit this trace:

[110889.918952] Call Trace:
[110889.920407]  [<ffffffff812172c7>] do_sys_poll+0x327/0x580
[110889.921844]  [<ffffffff810cdcbc>] ? dequeue_entity+0x11c/0x5d0
[110889.923264]  [<ffffffff810cd7d4>] ? update_curr+0x104/0x190
[110889.924657]  [<ffffffff810ca2de>] ? account_entity_dequeue+0xae/0xd0
[110889.926022]  [<ffffffff810cdcbc>] ? dequeue_entity+0x11c/0x5d0
[110889.927361]  [<ffffffff810ce58e>] ? dequeue_task_fair+0x41e/0x660
[110889.928681]  [<ffffffff810c7735>] ? sched_clock_cpu+0x85/0xc0
[110889.929969]  [<ffffffff81215e00>] ? poll_select_copy_remaining+0x150/0x150
[110889.931248]  [<ffffffff8109b62b>] ? recalc_sigpending+0x1b/0x50
[110889.932493]  [<ffffffff8109b756>] ? dequeue_signal+0x86/0x170
[110889.933721]  [<ffffffff8109e28f>] ? get_signal_to_deliver+0xff/0x5e0
[110889.934935]  [<ffffffff810f4f00>] ? futex_wake+0x90/0x180
[110889.936109]  [<ffffffff8102a467>] ? do_signal+0x57/0x6c0
[110889.937271]  [<ffffffff8133da6d>] ? list_del+0xd/0x30
[110889.938404]  [<ffffffff810b16b1>] ? remove_wait_queue+0x31/0x40
[110889.939531]  [<ffffffff81217569>] do_restart_poll+0x49/0x90
[110889.940641]  [<ffffffff816b423d>] ? int_signal+0x12/0x17
[110889.941710]  [<ffffffff8109e8ef>] sys_restart_syscall+0x1f/0x30
[110889.942753]  [<ffffffff816b3f89>] system_call_fastpath+0x16/0x1b
[110889.943767] Code: ff ff c3 0f 1f 84 00 00 00 00 00 66 66 66 66 90
55 b8 fb ff ff ff 48 89 e5 41 54 53 4c 8b a7 a8 00 00 00 49 8b 1c 24
48 8b 4b 40 <48> 83 b9 d0 01 00 00 00 75 06 5b 41 5c 5d c3 90 48 85 f6
74 19
[110889.945863] RIP  [<ffffffffc0304080>] uio_poll+0x20/0x70 [uio]
[110889.946851]  RSP <ffff88065de8fb08>



It looks like the problem is if the app is shutting down the user of a device and gets a signal while waiting in a poll call (before do_poll has called do_pollfd and that has done a fdget), the app shutting down the device will proceed to do the close on the uio device and the kernel will eventually do a uio_unregister_device which frees the uio device resources. However, the kernel could restart the poll like we see above. The restarted poll call could then race with the app shutdown and call into the uio_poll callout and reference memory freed by uio_unregister_device.

To fix this:

- It looks like the userspace app could use ppoll instead of poll since it does not support restarting poll from the kernel.

- Maybe the poll code could do a fdget/fget on the uio device's fd/file when it decides to the the restart (note uio does not have refcounting so we would need to add that to support this



Version-Release number of selected component (if applicable):


How reproducible:


Steps to Reproduce:
1.
2.
3.

Actual results:


Expected results:


Additional info: