zdb gives Input/Output error

[FransUrbo]

(Possibly) related to #1848, #1862 and #1933.

Any attempt to run (most) zdb commands gives Input/Output error. Trying on an exported pool, with the cache file removed, gives a segmentation fault:

celia(SSD1):~# zdb -c -e -p /dev/disk/by-id share 8328
Segmentation fault

Running it through gdb shows:

[...]
Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0xf6f68b10 (LWP 12467)]
0xf76906ab in spa_history_log_internal (spa=0xaa52c1a0, operation=0xf76cc1d4 "import", tx=0x0,
    fmt=0xf76d7db4 "pool version %llu; software version %llu/%d; uts %s %s %s %s") at ../../module/zfs/spa_history.c:488
488     ../../module/zfs/spa_history.c: No such file or directory.
        in ../../module/zfs/spa_history.c
(gdb) bt
#0  0xf76906ab in spa_history_log_internal (spa=0xaa52c1a0, operation=0xf76cc1d4 "import", tx=0x0,
    fmt=0xf76d7db4 "pool version %llu; software version %llu/%d; uts %s %s %s %s") at ../../module/zfs/spa_history.c:488
#1  0xf7690789 in spa_history_log_version (spa=0xaa52c1a0, operation=0xf76cc1d4 "import") at ../../module/zfs/spa_history.c:549
#2  0xf768b82f in spa_import (pool=0x9d661c8 "share", config=0x9d661d0, props=0x0, flags=1) at ../../module/zfs/spa.c:4045
#3  0x08054807 in main ()

[behlendorf]

@FransUrbo The history code changed substantially with the libzfs_core changes which went in just after 0.6.2. It would be helpful to instrument spa_history_log_internal to see exactly where things go wrong, it's not obvious to me from the gdb stack. You could also try disabling spa_history_log_version for now, it's entirely non-critical and could just be commented out to see if there are other issues.

[FransUrbo]

It is the dmu_tx_create_dd() in spa_history_log_internal() that triggers the seg fault...

Doing a blame on spa.c, I see a lot of rougue's hand here. I'll have a talk to @lundman about this...

--EDIT-- Actually, that seems to be wrong. It looked a little weird, so I did a diff between my version and the vanilla ZoL code and no spa_history_log* stuff is touched.

[FransUrbo]

Disabling spa_history_log_version() stops the seg fault, but then I'm back to I/O error:

On an exported pool:

celia(SSD1):~# zdb -c -e -p /dev/disk/by-id share 8328
zdb: can't open 'share': Input/output error
celia(SSD1):~# zdb -c -e -p /dev/disk/by-id share 0 8328
zdb: can't open 'share': Input/output error

Which is the same as on an imported pool:

celia(SSD1):~# zdb -c share 8328
zdb: can't open 'share': Input/output error
celia(SSD1):~# zdb -c share 0 8328
zdb: can't open 'share': Input/output error

[FransUrbo]

I'm closing this, because it's now quite obvious that it's the zfs-crypto patch that fucked everything up.