Skip to contentSkip to conversation

xfsdump | xfsrestore resulting in files->orphanage

DISCUSSION L A Walsh 2 participants 2 messages Mar 25, 2021

lore
Share & tools
lore

L A Walsh

author
#0 lore raw

oops, forgot to cc list.

-------- Original Message --------

Quoted 5 lines · Eric Sandeen

This is a bug in root inode detection that Gao has fixed, and I really need to merge.

In the short term, you might try an older xfsdump version, 3.1.6 or earlier.

---
In the short term -- I was dumping from a dumpdir
for a partition (just to make a copy of it on the new disk), but there was no real requirement to do so, so I just dumped from the "source" dir, which for whatever reason didn't have the problem.

My final try would have been to use rsync or such.

Quoted 2 lines

(Assuming you don't actually have a bind mount)

---
Not on that partition...
3.1.6? Hasn't 318 been out for quite a while?

I looked through my bins only have 312 and 314 (and 318)... tried 314, but it started out with the same inode confusion -- didn't wait until it started spitting out any other errors.

Quoted 3 lines

Sorry about that.

-Eric

---
No prob. Hey, one thing else you might wanna fix in
xfsdump/restore that was fixed in xfs_fsr, is to put a posix_fadvise64(file->fd, offset, len, POSIX_FADV_DONTNEED) before the read in xfsdump and on the write in xfs_restore.

That way they'll let go of memory and won't end up
with pegged memory through-out the 'copy' - unless it is already there, and then I have some other problem :-( . But used all 8M of my swap (normally doesn't swap at all), shows a cpu load of 3.4, and over 100% in wait-state.

Just a hopeful suggestion. I _think_ Dave put the
call in xfs_fsr (it would clear out memory everytime it ran, like xfsdump/restore does now ;^)).

Thanks again for the possible cause...

#1 lore raw

Hi,

Quoted 18 lines · L A Walsh

oops, forgot to cc list.

-------- Original Message --------

Quoted 5 lines · Eric Sandeen

This is a bug in root inode detection that Gao has fixed, and I really need to merge.

In the short term, you might try an older xfsdump version, 3.1.6 or earlier.

---
In the short term -- I was dumping from a dumpdir
for a partition (just to make a copy of it on the new disk), but there was no real requirement to do so, so I just dumped from
the "source" dir, which for whatever reason didn't have the problem.

My final try would have been to use rsync or such.

Quoted 1 line
Quoted 8 lines
Quoted 1 line

(Assuming you don't actually have a bind mount)

---
Not on that partition...
3.1.6? Hasn't 318 been out for quite a while?

I looked through my bins only have 312 and 314 (and 318)... tried 314, but it started out with the same inode confusion -- didn't wait until it started spitting out any other errors.

Not sure get all your point.

That is because commit ("xfsdump: handle bind mount targets") was included in 3.1.7. So it would be better to confirm if 3.1.6 works as expected.

In principle, 3.1.6 should work if applying to non-bindmount root dir. (I mean if no new non-identify potential issue here)

Thanks,
Gao Xiang