Public bug reported: [Impact] Google: "We discovered an issue with upstream patch 52cb7f8f177878b4f22397b9c4d2c8f743766be3 that caused RW nfs mounts to be mounted as RO when there was an existing nfs RO mount on the same node."
The break commit was found on jammy:gke/gkeop and noble:gke/gkeop. [Fix] Apply the fix proposed on lkml at https://lore.kernel.org/lkml/20250304130533.549840-1-lilingfe...@huawei.com/T/ The commit titled "nfs: clear SB_RDONLY before getting superblock" should address the issue. Because it has not been applied to any trees, it will be initially applied as an UBUNTU: SAUCE: patch. [Test] A reproducer was included with the initial bug report at https://www.spinics.net/lists/linux-nfs/msg109544.html # Reproducer mkdir -p /export/{stuff,things}/dir{1,2,3,4} echo '/export/stuff *(rw)' >> /etc/exports echo '/export/things *(rw)' >> /etc/exports systemctl restart nfs-server mount -t nfs -o ro,vers=3 localhost:/export/stuff /mnt/stuff mount -t nfs -o rw,vers=3 localhost:/export/things /mnt/things grep -w nfs /proc/mounts # note that both mountpoints are ro, despite the explicit ro/rw options # reversing the order of mounts gives a different result After the fix, the read only mount should be properly tagged as read only, and the read/write mount should be properly tagged as read/write. [Where problems could occur] This is a fairly small patch which directly addresses a reproducible issue, so its scope of influence should be pretty limited. We do take some risk by applying this patch before it has been approved by upstream maintainers, so this won't be going into generic kernels until then. ** Affects: linux-gke (Ubuntu) Importance: Undecided Status: New ** Affects: linux-gkeop (Ubuntu) Importance: Undecided Status: New ** Affects: linux-gke (Ubuntu Jammy) Importance: Undecided Status: New ** Affects: linux-gkeop (Ubuntu Jammy) Importance: Undecided Status: New ** Affects: linux-gke (Ubuntu Noble) Importance: Undecided Status: New ** Affects: linux-gkeop (Ubuntu Noble) Importance: Undecided Status: New ** Also affects: linux-gkeop (Ubuntu) Importance: Undecided Status: New ** Also affects: linux-gke (Ubuntu Jammy) Importance: Undecided Status: New ** Also affects: linux-gkeop (Ubuntu Jammy) Importance: Undecided Status: New ** Also affects: linux-gke (Ubuntu Noble) Importance: Undecided Status: New ** Also affects: linux-gkeop (Ubuntu Noble) Importance: Undecided Status: New -- You received this bug notification because you are a member of Kernel Packages, which is subscribed to linux-gke in Ubuntu. https://bugs.launchpad.net/bugs/2108862 Title: Regression in NFS ro mount behavior Status in linux-gke package in Ubuntu: New Status in linux-gkeop package in Ubuntu: New Status in linux-gke source package in Jammy: New Status in linux-gkeop source package in Jammy: New Status in linux-gke source package in Noble: New Status in linux-gkeop source package in Noble: New Bug description: [Impact] Google: "We discovered an issue with upstream patch 52cb7f8f177878b4f22397b9c4d2c8f743766be3 that caused RW nfs mounts to be mounted as RO when there was an existing nfs RO mount on the same node." The break commit was found on jammy:gke/gkeop and noble:gke/gkeop. [Fix] Apply the fix proposed on lkml at https://lore.kernel.org/lkml/20250304130533.549840-1-lilingfe...@huawei.com/T/ The commit titled "nfs: clear SB_RDONLY before getting superblock" should address the issue. Because it has not been applied to any trees, it will be initially applied as an UBUNTU: SAUCE: patch. [Test] A reproducer was included with the initial bug report at https://www.spinics.net/lists/linux-nfs/msg109544.html # Reproducer mkdir -p /export/{stuff,things}/dir{1,2,3,4} echo '/export/stuff *(rw)' >> /etc/exports echo '/export/things *(rw)' >> /etc/exports systemctl restart nfs-server mount -t nfs -o ro,vers=3 localhost:/export/stuff /mnt/stuff mount -t nfs -o rw,vers=3 localhost:/export/things /mnt/things grep -w nfs /proc/mounts # note that both mountpoints are ro, despite the explicit ro/rw options # reversing the order of mounts gives a different result After the fix, the read only mount should be properly tagged as read only, and the read/write mount should be properly tagged as read/write. [Where problems could occur] This is a fairly small patch which directly addresses a reproducible issue, so its scope of influence should be pretty limited. We do take some risk by applying this patch before it has been approved by upstream maintainers, so this won't be going into generic kernels until then. To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/linux-gke/+bug/2108862/+subscriptions -- Mailing list: https://launchpad.net/~kernel-packages Post to : kernel-packages@lists.launchpad.net Unsubscribe : https://launchpad.net/~kernel-packages More help : https://help.launchpad.net/ListHelp
