Two crashes on raid0 error path (during a member device removal)

Bug #1836806 reported by Guilherme G. Piccoli on 2019-07-16
This bug affects 1 person
Affects Status Importance Assigned to Milestone
linux (Ubuntu)
Status tracked in Eoan
Guilherme G. Piccoli
Guilherme G. Piccoli
Guilherme G. Piccoli
Guilherme G. Piccoli

Bug Description


* During raid0 error path testing, by removing one member of the array, we've noticed after kernel 4.18 we can trigger a crash depending if there's I/O in-flight during the array removal. When debugging the issue, a second problem was found, that could cause a different crash.

* For the first and more relevant problem, commit cd4a4ae4683d
("block: don't use blocking queue entered for recursive bio submits") introduced the flag BIO_QUEUE_ENTERED in order BIOs that were split do bypass the blocking queue entering routine and use the live non-blocking version. What happens with md/raid0 though is that their BIOs have their underlying device changed to the physical disk (array member). If we remove this physical disk (or if it fails), we could have one BIO that had the flag changed to BIO_QUEUE_ENTERED and had the device changed to the removed array member (before its removal); this bio then skips a lot of checks in generic_make_request_checks(), triggering the following crash:

BUG: unable to handle kernel NULL pointer dereference at 0000000000000155
PGD 0 P4D 0
Oops: 0000 [#1] SMP PTI
RIP: 0010:blk_throtl_bio+0x45/0x970
Call Trace:
 raid0_make_request+0x184/0x620 [raid0]
 ? raid0_make_request+0x184/0x620 [raid0]

* When debugging the above issue, by rebuilding the kernel with CONFIG_BLK_CGROUP=n we've noticed a different crash. Commit 37f9579f4c31 ("blk-mq: Avoid that submitting a bio concurrently with device removal triggers a crash") introduced a NULL pointer dereference in generic_make_request(), that manifests as:

BUG: unable to handle kernel NULL pointer dereference at 0000000000000078
PGD 0 P4D 0
Oops: 0000 [#1] SMP PTI
RIP: 0010:generic_make_request+0x32b/0x400
Call Trace:

* For both the issues, we have simple patches that are present in linux-stable but not in Linus tree.
## For issue 1 (md removal crash):
869eec894663 ("md/raid0: Do not bypass blocking queue entered for raid0 bios")

## For issue 2 (generic_make_request() NULL dereference):
c9d8d3e9d7a0 ("block: Fix a NULL pointer dereference in generic_make_request()")

The reasoning for both patches not being present in Linus tree is explained in the commit messages, but in summary Ming Lei submitted a major clean-up series at the same time I've submitted both patches, it wouldn't make sense to accept my patches to soon after remove the code paths with his clean-up. But Ming's series rely on legacy I/O path removal, and so it's very hard to backport. Hence maintainers suggested me to submit my small fixes to stable tree only.

[Test case]

For both cases, the test is the same, the only change being a kernel config option. To reproduce issue 1 (md removal crash), a regular Ubuntu kernel config is enough. For the issue 2, a kernel rebuild with CONFIG_BLK_CGROUP=n is necessary.

Steps to reproduce:

a) Create a raid0 md array with 2 NVMe devices as members, and mount it with an ext4 filesystem.

b) Run the following oneliner (supposing the raid0 is mounted in /mnt):
(dd of=/mnt/tmp if=/dev/zero bs=1M count=999 &); sleep 0.3;\
echo 1 > /sys/block/nvme1n1/device/device/remove
(whereas nvme1n1 is the 2nd array member)

[Regression potential]

The fixes are self-contained and small, both validated by a great number of subsystem maintainers (including block, raid and stable). Commit c9d8d3e9d7a0 was also validated by the author of the offender patch it fixes, and has no functional change. Commit 869eec894663 has only raid0 driver as scope, and fall-backs raid0 to a previous behavior before the introduction of BIO_QUEUE_ENTERED flag (which indeed increases the amount of checks performed in BIOs), so the regression potential is low and restricted to raid0.

CVE References

Changed in linux (Ubuntu Bionic):
status: New → Fix Released
Changed in linux (Ubuntu Eoan):
status: Confirmed → Fix Released
Changed in linux (Ubuntu Disco):
status: New → In Progress
importance: Undecided → Medium
Changed in linux (Ubuntu Bionic):
importance: Undecided → Medium
Changed in linux (Ubuntu Disco):
assignee: nobody → Guilherme G. Piccoli (gpiccoli)
Changed in linux (Ubuntu Bionic):
assignee: nobody → Guilherme G. Piccoli (gpiccoli)
Changed in linux (Ubuntu Cosmic):
status: New → Won't Fix
importance: Undecided → Medium
assignee: nobody → Guilherme G. Piccoli (gpiccoli)
description: updated
Guilherme G. Piccoli (gpiccoli) wrote :

This issue affects only kernels after 4.17 and before 5.2, hence it's fixed on Bionic and Eoan, and won't be fixed in Cosmic (4.18) since it is EOL.

Patches submitted to kernel-team ML:



Changed in linux (Ubuntu Disco):
status: In Progress → Fix Committed
Brad Figg (brad-figg) on 2019-07-24
tags: added: cscc

This bug is awaiting verification that the kernel in -proposed solves the problem. Please test the kernel and update this bug with the results. If the problem is solved, change the tag 'verification-needed-disco' to 'verification-done-disco'. If the problem still exists, change the tag 'verification-needed-disco' to 'verification-failed-disco'.

If verification is not done by 5 working days from today, this fix will be dropped from the source code, and this bug will be closed.

See for documentation how to enable and use -proposed. Thank you!

tags: added: verification-needed-disco
Guilherme G. Piccoli (gpiccoli) wrote :

Verified in disco kernel 5.0.22-generic (that is available in -proposed pocket), using the test cases described in the patches.


tags: added: verification-done-disco
removed: verification-needed-disco
Launchpad Janitor (janitor) wrote :

This bug was fixed in the package linux - 5.0.0-25.26

linux (5.0.0-25.26) disco; urgency=medium

  * CVE-2019-1125
    - x86/cpufeatures: Carve out CQM features retrieval
    - x86/cpufeatures: Combine word 11 and 12 into a new scattered features word
    - x86/speculation: Prepare entry code for Spectre v1 swapgs mitigations
    - x86/speculation: Enable Spectre v1 swapgs mitigations
    - x86/entry/64: Use JMP instead of JMPQ
    - x86/speculation/swapgs: Exclude ATOMs from speculation through SWAPGS

 -- Kleber Sacilotto de Souza <email address hidden> Thu, 01 Aug 2019 12:04:35 +0200

Changed in linux (Ubuntu Disco):
status: Fix Committed → Fix Released
To post a comment you must log in.
This report contains Public information  Edit
Everyone can see this information.

Other bug subscribers