[OpenAFS-devel] [CSL #301175] openafs-1.4.1 kernel panics

[Rainer Toebbicke]

thomas@cs.wisc.edu wrote:

> 
> afs_osi_Alloc: Can't vmalloc 4524 bytes.
> crget: No more memory for creds!
> Unable to handle kernel NULL pointer dereference at virtual
> address 00000000
>  printing eip:
> f8ddf800
> *pde = 2192b001
> *pte = 69107067
> Oops: 0002
> sg ide-scsi openafs iptable_filter ip_tables e1000 microcode keybdev
> mousedev hid input usb-uhci usbcore ext3 jbd 3w-9xxx sd_mod scsi_mod
> CPU: 3
> EIP: 0060:[<f8ddf800>] Tainted: P
> EFLAGS: 00010282
> 
> EIP is at osi_Panic [openafs] 0x20 (2.4.21-40.ELsmp/i686)
> eax: 00000021 ebx: f3d82000 ecx: 00000000 edx: c0387e98
> esi: eefe9019 edi: 00000000 ebp: f8e19a40 esp: f3d83e9c
> ds: 0068 es: 0068 ss: 0068
> Process imapd (pid: 24128, stackpage=f3d83000)
> Stack: f8e04c78 416ab962 00000296 f1a20c80 000066fc 00000001
> 00000001 f8dea664
> 
>        f8e04c78 416ab962 00000296 f1a20c80 f3d82000 f3d82000
>        f3d82000 f8dea71f
>        00000001 00000001 eefe9019 f8df0bd7 c0456780 f0c3f280
>        f18db6ec f3d82000
> [<f8dea664>] crget [openafs] 0xe4 (0xf3d83eb8)
> [<f8e04c78>] .rodata.str1.4 [openafs] 0x437c (0xf3d83ebc)
> [<f8dea71f>] crref [openafs] 0xf (0xf3d83ed8)
> [<f8df0bd7>] afs_linux_permission [openafs] 0x17 (0xf3d83ee8)
> [<c017391f>] permission [kernel] 0x4f (0xf3d83f08)
> [<f8def7a0>] afs_linux_lookup [openafs] 0x0 (0xf3d83f14)
> [<c0173de6>] link_path_walk [kernel] 0x76 (0xf3d83f18)
> [<c0174779>] path_lookup [kernel] 0x39 (0xf3d83f58)
> [<c0174ac9>] __user_walk [kernel] 0x49 (0xf3d83f68)
> [<c016f6ee>] sys_stat64 [kernel] 0x2e (0xf3d83f84)
> [<c02af06f>] no_timing [kernel] 0x7 (0xf3d83fc0)
> 
> (See openafs-1.4.1/src/afs/LINUX/osi_cred.c).
> 
> 


We are actually seeing similar crashes on our multi-access machines. 
Now, we *are* already running with Hartmut Reuter's patch that reduces 
the saved tokens to what is actually needed. So there is something else.

With kdump (I believe the more recent ones compile correctly at least 
on i386) you can find out how many PAGs are in use, 'kdump -users' 
shows something like over 2000 on some of our systems which I find 
surprising.

I wonder if there is a problem with PAG garbage collection. But even 
without Hartmut's patch the amount of space actually wasted is 
worrying but should not be fatal.


-- 
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Rainer Toebbicke
European Laboratory for Particle Physics(CERN) - Geneva, Switzerland
Phone: +41 22 767 8985       Fax: +41 22 767 7155