On Tuesday, January 28, 2014 09:37:36 PM Craig Small wrote:
> tags 736628 help
> tags 736628 upstream
>
> On Mon, Jan 27, 2014 at 09:08:56AM -0800, Daniel Schepler wrote:
> > (gdb) p disks
> > $1 = (struct disk_stat **) 0x7fffffffeb08
> > (gdb) p *disks
> > $2 = (struct disk_stat *) 0x0
> > (gdb) p cDisk
>
> The problem seems to be that a partition is found before the disk
> appears. More specifically, the partition needs to be before *any*
> disk. (A different but related bug is that the library assumes
> all partitions appear immediately after the disk line).
>
> So: sda, sda1, sda2 = ok
> sda1,sda2,sda = crash
> sr0, sda1, sda2 = ok, but sda1 is "linked" to sr0
>
> The problem now is, under what conditions do these odd
> /proc/diskstats occur and what interpretation should
> vmstat make of the situation? Ignore the partitions?
> Raise an error? re-map the disk->partition links
> after reading the file?
I just did a bit more debugging using strace, and found that this happens just
before the crash:
access("/sys/block/sda", F_OK) = -1 ENOENT (No such file or directory)
--- SIGSEGV (Segmentation fault) @ 0 (0) ---
+++ killed by SIGSEGV +++
Segmentation fault
In fact, it looks like pbuilder doesn't mount /sys, and I can get vmstat to
work by hand mounting it:
root@frobozz:/# vmstat -p sda1
Segmentation fault
root@frobozz:/# mount -t sysfs sysfs /sys
root@frobozz:/# vmstat -p sda1
sda1 reads read sectors writes requested writes
727293 24757690 278900 32422536
root@frobozz:/# umount /sys
root@frobozz:/# vmstat -p sda1
Segmentation fault
--
Daniel Schepler
--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]
