-
Notifications
You must be signed in to change notification settings - Fork 72
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
invalid bkey u64s 8 type accounting: fsck cannot repair [6f03ad7] #735
Comments
Same thing here, also on 6.11-rc4, didn't get those on 6.11-rc1, so presumably something broke between those? |
okay, that's something ...
but ... as I said ... I don't know why. |
Running
in dmesg and exits. |
yes, I assume it's a caching error. running bcachefs job without any fsck gives this error here as well, which is why I recommend all fscks even though not useful |
fsck does not seem to help here :( |
@koverstreet : any ideas? EDIT: to show exactly what I did which fixed all bcachefs volumes for me:
(I'm waiting a bit here, but I didn't have the patience for i to become 40, so I pressed ctrl-c during "sleep 10" after a while)
|
OK, I think I figured out what's happening with the fix at least: every offline fsck tries to reapply the journal, runs into the broken journal entries, fixes one of them, then exits. My first offline fsck (well not really first, but first in this batch) started replaying at 243662, the following one at 243664, and seemingly so on. This is definitely an fsck bug, at the very least. Now, what is actually causing those entries to be wrong, I don't know. |
Actually, never mind, looks like the fsck is what's creating the journal entries, because there's always only one of them according to |
Also, I just tried updating to 6.11-rc5 and that doesn't seem to change anything. |
I had this problem with bcachefs-testing prior to rc4's release, and downgraded to rc3 each time I had the problem. Upgrading to rc4 didn't provoke it again. By the way, I changed my upgrade methodology when I moved to rc4: Before, I was running: Now, I run: I still run it from a rescue environment, because I don't trust the fsck not to OOM yet. I may blindly try booting into the OS next time, whenever that is, since it seems to either finish or not in only 20 minutes now, instead of the previous two hours. Another difference is I'm also running bcachefs-tools 1.11.0 now. |
Are you guys mounting with -o fsck,fix_errors? |
I wasn't initially but I am now. |
Result? And are you on IRC? |
Same errors in dmesg every time. It's currently 4AM for me (don't ask), but I can hop on IRC in maybe 3-4 hours. |
Ok; it's sounding like something must be wrong with the repair dump, so if I can get a metadata dump that should make it quick for me to debug |
Some poking on IRC later, it looks like what actually happened here was:
|
following up on issue #732 :
mounting / running bcachefs fsck complains about bkey accounting errors:
these "invalid bkey" errors happen at every mount / umount on every fs I have (currently I got 3 of them).
A similar issue is happening at bcachefs fsck.
I tried using GDB to see what happened, and I saw bch2_bkey_val_validate returns BCH_ERR_fsck_delete_bkey at btree_io.c:1223, so I assumed it would be deleted. This seems not true.
Any suggestions?
The text was updated successfully, but these errors were encountered: