pkay even when I started with pfix=ram and built a new pup_save I got the inode errors
trapping them they appear to start with 2 inode which state
deleted inode 6283 has zero dtime fix - yes
deleted inode 6284 has zero dtime fix - yes
which fixes that and breaks
then in Pass5 I get
Inode bitmap diff 6283 - 6284 fix - yes
which fixes that and breaks
Free count wrong group #3 fix - yes
I tried googling the initial error (zero dtime) and found this interesting exchange which seems to imply that a sync has to be forced prior to unmounting in order to flush memory mapped files.
Because of the way I work I would prefer to not perform an fsck on startup - but I can see that others might need this.
Is there a way to choose to NOT run it maybe with a boot setting barry, I would prefer that to just hacking the routine out of the initrd ?
http://savannah.gnu.org/bugs/?17123
21:12 < youpi> Mmm, ext2fs doesn't seem to have a S_startup_dosync() handler, is that intended ? I'm not sure it performs sync on shutdown
21:13 < youpi> I mean, if I call sync before halt for shutting my hurd box down, I have not error on reboot fsck
21:13 < youpi> else I have quite some
21:14 < tschwinge> youpi: I can confirm that behaviour.
21:14 < antrik> that would explain a lot...
21:14 < youpi> and between sync and halt, some process could write on the disk, so calling sync by hand isn't a solution
21:22 < antrik> well, someone claimed the hurd ext2fs is written in such a manner that the data on disk should be consistent at any time... if that's the case, maybe sync wasn't considered necessary, or something
21:23 < youpi> !
21:23 < youpi> yeah, data is consistent
21:23 < youpi> I've not noticed any data loss
21:24 < youpi> but deleted but memory-living inodes are not deleted on the disk
21:24 < youpi> "deleted but memory-living", I meant
21:26 < youpi> (hence the "deleted inode xxx has zero dtime" fsck correction)
21:28 < bing> youpi: ogi would be a great guy to ask this questoin of
21:28 < youpi> and what about memory-mapped files ?
21:28 < youpi> they need be flushed
21:34 < youpi> hum, sorry yet, ext2fs has one, coming from libdiskfs, but
21:35 < youpi> how is it that we still have errors ?
21:36 < antrik> ?
21:38 < youpi> libdiskfs has a S_startup_dosync() which calls diskfs_sync_everything() in ext2, which should do the work
21:42 < antrik> ah, OK
Monday 07/17/2006 at 09:09 UTC, original submission:
14:57 <youpi> schlesix: the non-umounting file system on system halt is not easy
14:57 <youpi> because for now tasks are not all killed
14:58 <youpi> (that's why you get "deleted inode foo has zero dtime: try touch blip, mmap("blip");pause() &, rm blip, blip will be a deleted inode with zero dtime (because the program running mmap() is still alive when ext2fs gets killed)
George