Hi Ted, Quoting Theodore Ts'o (2025-06-24 21:04:56) > On Tue, Jun 24, 2025 at 04:53:46PM +0200, Anton Khirnov wrote: > > Package: e2fsprogs > > Version: 1.47.2-1+b1 > > Severity: normal > > > > Dear Maintainer, > > as a part of my backup system, I create read-only LVM snapshots of ext4 > > partitions. On my newest machine, mounting these snapshots sometimes > > fails with > > fsconfig() failed: Structure needs cleaning. > > accompanied by > > ext4_mark_recovery_complete:6264: comm mount: Orphan file not empty on > > read-only fs. > > in dmesg. Upon investigating, I found out that the orphan_file fs > > feature, which recently became enabled by default, is the culprit. > > > > This behaviour seems suboptimal to me. Would it be possible to make such > > filesystems mountable? If not, perhaps this feature should not be on by > > default, or at least there should be a warning about this. > > Hi Anton, > > When you say read-only snapshots, do you mean that you created with > some command like this: > > # lvcreate --size 100M --name snap --snapshot --permission r /dev/vg/source
yes, this is what I meant > I'd need to see your kernel messages to be certain, but this > requirement of needing to be able to write to the snapshot device is > there regardless of whether the orphan file feature is enabled or not. > So I'm not sure how you came to the conclusion that the orphan_file > feature is "at fault", but I'm pretty sure that's not the case. I assumed orphan_file is the cause because 1) The kernel error message mentioned it. 2) It is enabled only on the failing machine. The relevant kernel messages are just these: [1798344.446943] EXT4-fs (dm-9): write access unavailable, skipping orphan cleanup [1798344.446949] EXT4-fs (dm-9): recovery complete [1798344.446952] EXT4-fs error (device dm-9): ext4_mark_recovery_complete:6264: comm mount: Orphan file not empty on read-only fs. [1798344.446970] EXT4-fs (dm-9): mount failed I have been running this procedure daily on 30+ ext4 filesystems for 5+ years, and NEVER had the mount fail in this manner. Also, I investigated some more after filing the bugreport, and lvcreate is apparently supposed to freeze the filesystem while creating the snapshot. Looking at ext4_freeze() in fs/ext4/super.c it seems like it's marking the snapshot as not needing recovery unconditionally, but only clearing orphan_present if the orphan file is empty. If I understand this correctly, this might be the reason why it never fails without orphan_file, but does fail with it. > Finally, why are you creating read-only LVM snapshots? The whole > point of snapshots is that you can allow read-write snapshots, and > they won't affect the original source volume. Just create the > snapshots without the "--permissions r" option, and then you can mount > the file system with -o ro, and things will work fine. Yes, we may > need to replay the journal before continuing but that doesn't take a > lot of disk space. Using --size 100M should be *more* than enough for > pretty much any file system. For backups, I want the fs to be static. Back when I wrote this code, it seemed reasonable to just make the snapshot read-only, so nothing running on the system could modify it accidentally. As it has always worked fine until now, I had no reason to dwell on it any further. I suppose I drop the '--permission r' option and mount the snapshot -oro instead, but the fact that any change is needed at all still strikes me as not quite right. Cheers, -- Anton Khirnov
