Re: 2.6.25-rc6: kernel BUG at fs/sysfs/file.c:89

[Greg KH <greg@kroah.com>]

On Fri, Mar 21, 2008 at 10:26:06PM -0700, Andrew Morton wrote:
> On Sat, 22 Mar 2008 02:26:34 +0100 "Rafael J. Wysocki" <rjw@sisk.pl> wrote:
> 
> > On Friday, 21 of March 2008, Christian Kujau wrote:
> > > Hi,
> > 
> > Hi,
> > 
> > > while attempting to test the patch which should fix the dm-crypt hang[0], 
> > > I applied the diff to -rc6. After rebooting I increased disk i/o and 
> > > shortly after the box rebooted - I noticed this earlier with -rc5 too, but 
> > > had no debug messages at hand. Well, the box is running again and tar(1) 
> > > is reading with 32MB/s for some time now - but is not rebooting, which is 
> > > good :)
> > > (I changed kernel.panic and kernel.panic_on_oops to zero).
> > > However, it's printing stuff like this:
> > 
> > This looks worrisome (some experts added to the CC list).
> > 
> > > [ 2632.423419] ------------[ cut here ]------------
> > > [ 2632.423762] kernel BUG at fs/sysfs/file.c:89!
> > > [ 2632.424089] invalid opcode: 0000 [#1] 
> > > [ 2632.424398] Modules linked in: sha256_generic act_police sch_ingress cls_u32 sch_sfq sch_cbq ipt_ULOG x_tables nfsd lockd nfs_acl auth_rpcgss exportfs tun sunrpc fuse twofish_i586 twofish_common eeprom w83l785ts asb100 hwmon_vid usb_storage zd1211rw firmware_class mac80211 i2c_nforce2 snd_intel8x0 snd_ac97_codec i2c_core cfg80211 ac97_bus snd_pcm snd_timer snd soundcore snd_page_alloc
> > > [ 2632.426407] 
> > > [ 2632.426651] Pid: 10297, comm: tar Not tainted (2.6.25-rc6 #3)
> > > [ 2632.427019] EIP: 0060:[<c01a8add>] EFLAGS: 00010212 CPU: 0
> > > [ 2632.427386] EIP is at sysfs_read_file+0xdd/0xf0
> > > [ 2632.427717] EAX: 00000001 EBX: f5691f00 ECX: 00000003 EDX: f7ca1e70
> > > [ 2632.428100] ESI: 00001000 EDI: f7ca1ed0 EBP: f5691f14 ESP: f6f3ff54
> > > [ 2632.428482]  DS: 007b ES: 007b FS: 0000 GS: 0033 SS: 0068
> > > [ 2632.428758] Process tar (pid: 10297, ti=f6f3f000 task=f6290d60 task.ti=f6f3f000)
> > > [ 2632.428921] Stack: c0164ff7 f5ca6000 00001000 0866ce00 c0550d50 f7d8fab8 f776b180 0866ce00 
> > > [ 2632.429174]        c01a8a00 00001000 c016868d f6f3ffa0 00000003 f776b180 fffffff7 0866ce00 
> > > [ 2632.429430]        f6f3f000 c0168a91 f6f3ffa0 00000000 00000000 00000000 00000003 00001000 
> > > [ 2632.429688] Call Trace:
> > > [ 2632.429884]  [<c0164ff7>] kmem_cache_free+0xa7/0xf0
> > > [ 2632.430023]  [<c01a8a00>] sysfs_read_file+0x0/0xf0
> > > [ 2632.430159]  [<c016868d>] vfs_read+0x9d/0x140
> > > [ 2632.430333]  [<c0168a91>] sys_read+0x41/0x70
> > > [ 2632.430657]  [<c0102dee>] sysenter_past_esp+0x5f/0xa5
> > > [ 2632.431006]  =======================
> > > [ 2632.431308] Code: 18 89 f0 5b 5e 5f 5d c3 be ed ff ff ff eb e8 b8 d0 00 00 00 be f4 ff ff ff e8 e0 31 fa ff 85 c0 89 43 0c 0f 85 6d ff ff ff eb cc <0f> 0b 90 eb fd 8d b4 26 00 00 00 00 8d bc 27 00 00 00 00 83 ec 
> > > [ 2632.432467] EIP: [<c01a8add>] sysfs_read_file+0xdd/0xf0 SS:ESP 0068:f6f3ff54
> > > [ 2632.432630] ---[ end trace 9959fa4f11f46a8e ]---
> > > 
> > > Please see the full dmesg and .config: http://nerdbynature.de/bits/2.6.25-rc6/
> 
> Is MD (ie: raid) in use?
> 
> If so, the below (already merged) patch should fix this crash.
> 
> If this patch does not fix it then please apply this debug patch:
> 
> http://userweb.kernel.org/~akpm/mmotm/broken-out/gregkh-driver-driver-core-debug-for-bad-dev_attr_show-return-value.patch
> 
> then rerun the test.
> 
> Greg/Kay:
> 
> - why is this BUG_ON now triggering for people?

Someone is overflowing the sysfs buffer now, it's only a PAGE_SIZE big,
which should be fine for a simple one-value-per-file system.

> - we're killing people's machines.  Can we switch it to a
>   WARN_ON()+fix-up-the-mess?

The patch above does just that, it's queued up for .26, but I have no
objection to add it now if you want.

thanks,

greg k-h