You are not logged in.
I won't bother. Looks like others have reported what seems to be the same bug on the udisks project.
https://github.com/storaged-project/udi … 1702623620
libblockdev-3.0.3 was released yesterday with the workaround: https://github.com/storaged-project/lib … ag/3.0.3-1
While you are producing the issue under 3.0.3.
With either the patched or unpatched libnvme does `nvme sanitize-log` produce 'stack smashing detected'?
Offline
If https://github.com/storaged-project/lib … e33f594b11 is the discussed "workaround" then it seems irrelevant to the issue at hand and focus on a different libnvme feature.
bulletmark has mn = "PC611 NVMe SK hynix 512GB" and the linked bug mn: NE-1TB and mn: J.ZAO QL SERIES 1TB SSD so it's not limited to a vendor/model.
It might be required to build udisks w/o libnvme for the time being to [edit "avoid breaking"] a fundamental feature across the board.
Last edited by seth (2023-09-24 13:23:33)
Offline
While you are producing the issue under 3.0.3.
With either the patched or unpatched libnvme does `nvme sanitize-log` produce 'stack smashing detected'?
I posted a comment earlier on that same bug you quote loqs and after reading it I already tried to recreate the fault with sanitize-log but I can't recreate it no matter what I try.
I saw that it stated it was fixed in libblockdev 3.0.3 and if you look at my post there I pointed out my versions but that I am still seeing the fault.
Don't worry pursuing this anymore. I expected this was a serious bug that would affect many but clearly that is not the case. I will just pursue a fix independently via that udisks bug on github.
Offline
I will just pursue a fix independently via that udisks bug on github.
You are not going to open a new one given the fix does not work for you and the reproducer does not produce the issue for you?
Offline
@loqs, that opening bug description is pretty close to the same issue I have so while it remains open I will pursue it there.
Offline
Did you try whether
nvme sanitize-log -H /dev/nvme0n1
triggers the stack smashing in libnvme for you (you don't have to upgrade udisks or libblockdev for that)
https://github.com/storaged-project/udi … 1685977468 looks likes a brute force hack to copy the data from the kernel instead of just mapping it and is probably more what you require here, alternatively libblockdev needs to copy the data at some point, but in any event, that bug would be some miles away from the one you're currently latching onto.
Offline
@bulletmark have you tried with libnvme's git HEAD? https://github.com/storaged-project/udi … 1733643749
Edit:
Or adding the kernel parameter swiotlb=force so all accesses go through the bounce buffer which will deal with alignment.
Last edited by loqs (2023-09-25 13:43:26)
Offline
@loqs, I am just about to leave for a 6 day holiday with my family. Will try these if I get opportunity while away or certainly first thing when I get back (if the bug has not moved). Thanks for continuing to research this.
PS: Seems to be something that will get fixed in libblockdev.
Last edited by bulletmark (2023-09-25 22:21:00)
Offline