On 26/05/2023 18:40, Mickaël Salaün wrote:
On 21/05/2023 23:13, Richard Weinberger wrote:
----- Ursprüngliche Mail -----
Von: "Mickaël Salaün" mic@digikod.net hostfs creates a new inode for each opened or created file, which created useless inode allocations and forbade identifying a host file with a kernel inode.
Fix this uncommon filesystem behavior by tying kernel inodes to host file's inode and device IDs. Even if the host filesystem inodes may be recycled, this cannot happen while a file referencing it is open, which is the case with hostfs. It should be noted that hostfs inode IDs may not be unique for the same hostfs superblock because multiple host's (backed) superblocks may be used.
Delete inodes when dropping them to force backed host's file descriptors closing.
This enables to entirely remove ARCH_EPHEMERAL_INODES, and then makes Landlock fully supported by UML. This is very useful for testing (ongoing and backported) changes.
Removing ARCH_EPHEMERAL_INODES should be a patch on its own, IMHO.
OK, I'll do that in the next series.
Well, I added ARCH_EPHEMERAL_INODES for Landlock specifically because of this hostfs inconsistency, and it is not used by anything else in the kernel: https://git.kernel.org/torvalds/c/cb2c7d1a1776 I then think it makes sense to remove this Kconfig option with the hostfs change. Moreover, this protects against erroneously backporting the ARCH_EPHEMERAL_INODES change, which would silently introduce a bug for Landlock.
These changes also factor out and simplify some helpers thanks to the new hostfs_inode_update() and the hostfs_iget() revamp: read_name(), hostfs_create(), hostfs_lookup(), hostfs_mknod(), and hostfs_fill_sb_common().
A following commit with new Landlock tests check this new hostfs inode consistency.
Cc: Anton Ivanov anton.ivanov@cambridgegreys.com Cc: Johannes Berg johannes@sipsolutions.net Cc: Richard Weinberger richard@nod.at Cc: stable@vger.kernel.org # 5.15.x: ce72750f04d6: hostfs: Fix writeback of dirty pages Cc: stable@vger.kernel.org # 5.15+
I'm not sure whether this patch qualifies as stable material. While I fully agree that the current behavoir is odd, nothing user visible is really broken so far.
I added the ARCH_EPHEMERAL_INODES knob to avoid unexpected behavior. Thanks to that there is no regression for Landlock, but it's unfortunate that we could not use UML to test old kernel versions. According to this odd behavior, I guess some user space may not work with hostfs because of this issue, hence this Cc. I can remove it if you think it is not the case.
Signed-off-by: Mickaël Salaün mic@digikod.net Link: https://lore.kernel.org/r/20230309165455.175131-2-mic@digikod.net
Other than that, patch looks good to me.
Good, I'll send a new series with your suggestions.
Can I add your Signed-off-by to this patch (without touching ARCH_EPHEMERAL_INODES changes, but removing the Cc stable)?
Are you OK for me to push this patch (with the whole series) in the Landlock and next tree?
I'll send a new series splitting the Landlock tests to make a patch dedicated to Landlock with hostfs tests (not backported), and with another patch containing backportable and independent new Landlock FS tests.