cd + geom -> panic on media change

[Andriy Gapon]

On 15/02/2017 22:33, Andriy Gapon wrote:
> 
> I had an Audio CD in a DVD drive.
> I booted with that CD in and this is how it was reported:
> cd0: <Optiarc DVD RW AD-7191S 1.02> Removable CD-ROM SCSI device
> cd0: 150.000MB/s transfers (SATA 1.x, UDMA5, ATAPI 12bytes, PIO 8192bytes)
> cd0: 392MB (200919 2048 byte sectors)
> 
> I see that there is some special code in scsi_cd.c that sets sector size to 2352
> for Audio CDs, but that was not reflected in the report quoted above.
> 
> Later I popped out the CD (using the physical eject button, if that matters) and
> popped in a UDF formatted DVD disk.  That resulted in the following panic:
> 
> panic: wrong offset 472559440 for sectorsize 2048


By the way, I see that all requests going down pass through g_io_check() which
would reject such a request with EINVAL.
Why do we have the KASSERT in g_io_request() then?

> KDB: stack backtrace:
> db_trace_self_wrapper() at 0xffffffff8043517b = db_trace_self_wrapper+0x2b/frame
> 0xfffffe0504bc8870
> kdb_backtrace() at 0xffffffff80685289 = kdb_backtrace+0x39/frame 0xfffffe0504bc8920
> vpanic() at 0xffffffff8064ce8c = vpanic+0x14c/frame 0xfffffe0504bc8960
> panic() at 0xffffffff8064cbd3 = panic+0x43/frame 0xfffffe0504bc89c0
> g_io_request() at 0xffffffff805c3b81 = g_io_request+0x3e1/frame 0xfffffe0504bc8a00
> g_read_data() at 0xffffffff805c48e7 = g_read_data+0x77/frame 0xfffffe0504bc8a40
> g_part_gpt_probe() at 0xffffffff805df0d1 = g_part_gpt_probe+0x111/frame
> 0xfffffe0504bc8a80
> G_PART_PROBE() at 0xffffffff805daa0b = G_PART_PROBE+0x4b/frame 0xfffffe0504bc8aa0
> g_part_probe() at 0xffffffff805da366 = g_part_probe+0xc6/frame 0xfffffe0504bc8af0
> g_part_taste() at 0xffffffff805d8de7 = g_part_taste+0x147/frame 0xfffffe0504bc8b30
> g_new_provider_event() at 0xffffffff805c7a0b = g_new_provider_event+0x10b/frame
> 0xfffffe0504bc8b50
> one_event() at 0xffffffff805c2a4f = one_event+0xff/frame 0xfffffe0504bc8b70
> g_run_events() at 0xffffffff805c2875 = g_run_events+0x65/frame 0xfffffe0504bc8b90
> g_event_procbody() at 0xffffffff805c4f38 = g_event_procbody+0x58/frame
> 0xfffffe0504bc8ba0
> fork_exit() at 0xffffffff80614800 = fork_exit+0xd0/frame 0xfffffe0504bc8bf0
> fork_trampoline() at 0xffffffff80837a7e = fork_trampoline+0xe/frame
> 0xfffffe0504bc8bf0
> 
> The actual DVD has size 680368 x 2048.
> I did some basic math and 472559440 == 200919 * 2352 - 2048.
> Given that the code was probing GPT, the offset is consistent with mediasize
> still being the old size of the Audio CD while the sector size being 2048.
> Seems like the disk properties were not properly updated before posting the new
> provider event.
> 
> Patches and suggestions are welcome :)
> Thank you!
> 


-- 
Andriy Gapon