Bug 2218844
| Summary: | [RHEL 9] rename() duplicates paths for a file on NFSv4 volume | ||
|---|---|---|---|
| Product: | Red Hat Enterprise Linux 9 | Reporter: | Zhi Li <yieli> |
| Component: | kernel | Assignee: | Jeff Layton <jlayton> |
| kernel sub component: | NFS | QA Contact: | Zhi Li <yieli> |
| Status: | CLOSED ERRATA | Docs Contact: | |
| Severity: | unspecified | ||
| Priority: | unspecified | CC: | bcodding, jiyin, jlayton, jwboyer, nfs-team, shdunne, xzhou, yoyang |
| Version: | 9.3 | Keywords: | Regression, Triaged |
| Target Milestone: | rc | ||
| Target Release: | --- | ||
| Hardware: | Unspecified | ||
| OS: | Unspecified | ||
| Whiteboard: | |||
| Fixed In Version: | kernel-5.14.0-362.4.1.el9_3 | Doc Type: | If docs needed, set a value |
| Doc Text: | Story Points: | --- | |
| Clone Of: | Environment: | ||
| Last Closed: | 2023-11-07 08:49:05 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: | 2209174 | ||
Hi team, Has anyone looked into this regression? Finally bisected back to: 638e3e7d9493 nfsd: use the getattr operation to fetch i_version Could be the assumption about change attr for directories is incorrect? I am digging in, but maybe you see the problem right away Jeff? This doesn't reproduce on:
615e95831ec3 (HEAD -> bisect) Merge tag 'v6.6-vfs.ctime' of git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs
Jeff suspects this is the xfs change attr issue he's already fixed upstream, and has a couple of directions to choose from for a fix here, so handing this over.
Here's my single-system reproducer (commented lines are for a non-restart bisect attempt):
#!/bin/bash
set -o xtrace
set -o errexit
umount -a -t nfs4
sudo rmmod nfsv4 nfsv3 nfs || true
#sudo systemctl stop nfs-server
#sudo umount /proc/fs/nfsd || true
#sleep 1
#sudo rmmod nfsd rpcsec_gss_krb5 auth_rpcgss lockd nfs_acl || true
#
#LOCALVERSION="" make -C /devel/linux-nfs -j10 M=fs/nfs
#LOCALVERSION="" make -C /devel/linux-nfs -j10 M=fs/nfsd
#
#sudo insmod /devel/linux-nfs/fs/nfs/nfs.ko
#sudo insmod /devel/linux-nfs/fs/nfs/nfsv4.ko
#sudo modprobe lockd
#sudo modprobe nfs_acl
#sudo modprobe auth_rpcgss
#sudo modprobe rpcsec_gss_krb5
#sudo insmod /devel/linux-nfs/fs/nfsd/nfsd.ko
sudo systemctl start nfs-server
mkdir /mnt/localhost1 || true
mkdir /mnt/localhost2 || true
mount -t nfs -ov4,sec=sys,noac,nosharecache localhost:/exports /mnt/localhost1
mount -t nfs -ov4,sec=sys,noac,nosharecache localhost:/exports /mnt/localhost2
rm -rf /exports/{A,B}
mkdir /mnt/localhost1/{A,B}
touch /mnt/localhost1/A/f
mv /mnt/localhost1/A/f /mnt/localhost1/B/
cat /mnt/localhost2/B/f
mv /mnt/localhost2/B/f /mnt/localhost2/A/
mv /mnt/localhost1/A/f /mnt/localhost1/B/
umount -a -t nfs4
This turns out to be an ancient nfsd bug, that only manifested with the changes to how we fetch and present the change attr. Patch posted here for upstream:
https://lore.kernel.org/linux-nfs/ZPyMyv1nNFV2whKP@tissot.1015granger.net/T/#t
Moving to VERIFIED according to the comment#26. Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory (Important: kernel security, bug fix, and enhancement update), and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://access.redhat.com/errata/RHSA-2023:6583 |
Description of problem: -rename() duplicates paths for a file on NFSv4 volume. Version-Release number of selected component (if applicable): kernel-5.14.0-331.el9.x86_64 How reproducible: always Steps to Reproduce: 1. Mount a NFSv4 volume with noac on node1 and node2. [root@node1 ~]# mount -o noac server:/srv/export /mnt/nfs [root@node2 ~]# mount -o noac server:/srv/export /mnt/nfs 2. Create directories and a file. [root@node1 ~]# mkdir /mnt/nfs/{A,B}; touch /mnt/nfs/A/f 3. Move the file from A to B on the node1. [root@node1 ~]# mv /mnt/nfs/A/f /mnt/nfs/B/ 4. cat the file on the node2, and move it from B to A. [root@node2 ~]# cat /mnt/nfs/B/f; mv /mnt/nfs/B/f /mnt/nfs/A/ 5. Move it again from A to B on the node1. [root@node1 ~]# mv /mnt/nfs/A/f /mnt/nfs/B/ 6. cat the file again on the node2, and move it from B to A again. [root@node2 ~]# cat /mnt/nfs/B/f; mv /mnt/nfs/B/f /mnt/nfs/A/ mv: `/mnt/nfs/B/f' and `/mnt/nfs/A/f' are the same file Actual results: - File paths are duplicated, and mv is failed with "... are the same file". Expected results - No duplicate, and mv is succeeded. Additional info: This issue exists on rhel9.3 but does not occur on rhel8.9, please confirm if this is a valid bug for rhel9. Beaker job: Test passed on rhel8.9(kernel-4.18.0-498.el8): https://beaker.engineering.redhat.com/jobs/7988933 Test failed on rhel9.3 (kernel-5.14.0-331.el9) https://beaker.engineering.redhat.com/jobs/8017247