[Bug 271766] dumpfs fails on geli devices: cylinder group checks failed
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=271766 Graham Perrin changed: What|Removed |Added Assignee|f...@freebsd.org |g...@freebsd.org Component|bin |kern Severity|Affects Only Me |Affects Some People CC||grahamper...@freebsd.org --- Comment #6 from Graham Perrin --- ^Triage, partly at the suggestion of RhodiumToad in IRC: * kern * severity * assignment to a list that's relevant to (GELI) encryption Key phrase: data corruption -- You are receiving this mail because: You are the assignee for the bug.
[Bug 271794] boot1.efi hangs on a system with gjournaled partitions
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=271794 Graham Perrin changed: What|Removed |Added Status|New |Open Assignee|b...@freebsd.org|g...@freebsd.org -- You are receiving this mail because: You are the assignee for the bug.
[Bug 271766] dumpfs fails on geli devices: cylinder group checks failed
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=271766 --- Comment #7 from Michael Laß --- (In reply to Michael Laß from comment #3) I was wondering why I could not reproduce this in 14 Current. The dumpfs issue is hidden by changes between 1297a704d99110644af1ae32840cee1b9523cb56 (still reproducible) and b21582ee03ec1394e08173e276df311979856e54 (not reproducible anymore). Most likely it comes down to: bf24d17fda75cc89a92248715d52a73f23adc89c: Have dumpfs(8) ignore superblock check-hash failures. However, this seems to only hide the issue in dumpfs and I assume that the actual data corruption issue in geli is still present in 14 Current. However, I'm not sure how to verify that. -- You are receiving this mail because: You are the assignee for the bug.
[Bug 271766] dumpfs fails on geli devices: cylinder group checks failed
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=271766 --- Comment #8 from Andrew "RhodiumToad" Gierth --- This just got much more serious: in trying to reproduce the bug more easily, I found that it is possible to panic the system with it; if the alignment of the user-provided buffer is not merely non-page-aligned, but also not word-aligned (for some wordsize, I don't yet know what), then: panic: general protection fault KDB: stack backtrace: [...] #3 0x807b9467 at trap_fatal+0x387 #4 0x8079378e at calltrap+0x8 #5 0x80696c53 at swcr_encdec+0x653 #6 0x8069635a at swcr_process+0x3a #7 0x806942bd at crypto_dispatch+0x15d #8 0x8124bab8 at g_eli_crypto_run+0x158 #9 0x81244c88 at g_eli_worker+0x308 -- You are receiving this mail because: You are the assignee for the bug.
[Bug 271766] dumpfs fails on geli devices: cylinder group checks failed
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=271766 Graham Perrin changed: What|Removed |Added Keywords||crash -- You are receiving this mail because: You are the assignee for the bug.
[Bug 271766] dumpfs fails on geli devices: cylinder group checks failed
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=271766 --- Comment #9 from Andrew "RhodiumToad" Gierth --- (In reply to Andrew "RhodiumToad" Gierth from comment #8) Correction: I can get crashes even for buffers with 16-byte alignment. Seems to be related to crypto_cursor_segment? I will post more info shortly. -- You are receiving this mail because: You are the assignee for the bug.
[Bug 271766] dumpfs fails on geli devices: cylinder group checks failed
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=271766 --- Comment #10 from Andrew "RhodiumToad" Gierth --- Created attachment 242577 --> https://bugs.freebsd.org/bugzilla/attachment.cgi?id=242577&action=edit Test program I've attached the test program I'm using to reproduce. Usage: md=$(mdconfig -s 1M -t swap) geli onetime -s 4096 $md # this will panic with software crypto, or if there are data discrepancies it will report them: ./a.out /dev/$md 1 The code where it panics has changed in 14-current so it's quite possible that the bug doesn't exist there. This program should make it easy to test. -- You are receiving this mail because: You are the assignee for the bug.
[Bug 271766] dumpfs fails on geli devices: cylinder group checks failed
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=271766
--- Comment #11 from Andrew "RhodiumToad" Gierth
---
(In reply to Andrew "RhodiumToad" Gierth from comment #10)
That should be
./a.out /dev/${md}.eli 1
of course.
--
You are receiving this mail because:
You are the assignee for the bug.
[Bug 271766] dumpfs fails on geli devices: cylinder group checks failed
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=271766 --- Comment #12 from Michael Laß --- I tested your code on 14-current and while there it does not seem to cause crashes, it still detects errors as soon as I specify strides smaller than 16. -- You are receiving this mail because: You are the assignee for the bug.
