DEVFS broken?

DEVFS broken?

[Lawrence Walton]

I am not sure how to debug this but it apears that 
in 2.5.3-pre1 and in 2.5.2-dj1 DEVFS is not working.
It started by terminals hanging and not being able to
shutdown.
I went to /dev/ and did a ls, it compleatly hangs that
terminal and I cannot kill ls.
I have the devfsd version from debian 1.3.21 .

-- 
*--* Mail: lawrence@otak.com
*--* Voice: 425.739.4247
*--* Fax: 425.827.9577
*--* HTTP://www.otak-k.com/~lawrence/
--------------------------------------
- - - - - - O t a k  i n c . - - - - - 

Re: DEVFS broken?

[Dan Chen]

[-- Attachment #1: Type: text/plain, Size: 6154 bytes --]

Also using Debian sid here (devfsd 1.3.21-1). Over the past two days
I've seen random nasties with devfs-v199.6 and v199.7 (I backed out
v199.7 in my local tree because my machine refuses to finish booting
otherwise). Machine: VIA VT82C693A/694x, PIII/1GHz SMP, 1GB HIGHMEM
enabled.

What follows is a series of crashes culled from /var/log/kern.log
(apologies regarding the format). Before each I'll explain what I did to
possibly invoke it. All are with devfs-v199.6 and devfsd-1.3.21 running
2.4.18-pre4 + ext3-2.4-0.9.17-2418p3 + ide.2.4.16.12102001:

# modprobe aic7xxx   //I tried to start an Eterm in X 4.1.0.1 afterward
-- snip --
Jan 16 22:54:50 opeth kernel: invalid operand: 0000
Jan 16 22:54:50 opeth kernel: CPU:    0
Jan 16 22:54:50 opeth kernel: EIP:    0010:[d_instantiate+17/68]
Tainted: P 
Jan 16 22:54:50 opeth kernel: EFLAGS: 00010202
Jan 16 22:54:50 opeth kernel: eax: 5a5a5a00   ebx: f6b5f560   ecx:
f7aa1b80   ed
x: f6b5f590
Jan 16 22:54:50 opeth kernel: esi: f7aa1b80   edi: f6b5f560   ebp:
f7a7f840   es
p: f7a41f18
Jan 16 22:54:50 opeth kernel: ds: 0018   es: 0018   ss: 0018
Jan 16 22:54:50 opeth kernel: Process devfsd (pid: 26,
stackpage=f7a41000)
Jan 16 22:54:50 opeth kernel: Stack: f77d9460 c0179b13 f6b5f560 f7aa1b80
f6b5f56
0 00000000 f7a41fa4 f7a83620 
Jan 16 22:54:50 opeth kernel:        c01409be f6b5f560 00000000 f7a41f74
c014118
1 f7a83620 f7a41f74 00000000 
Jan 16 22:54:50 opeth kernel:        f60d4000 00000000 f7a41fa4 00000009
0000000
9 f60d4005 00000000 f60d4004 
Jan 16 22:54:50 opeth kernel: Call Trace:
[devfs_d_revalidate_wait+231/276] [cac
hed_lookup+46/84] [link_path_walk+1409/2016] [path_walk+26/28]
[__user_walk+53/8
0] 
Jan 16 22:54:50 opeth kernel:    [sys_stat64+25/112] [sys_read+188/196]
[system_
call+51/56] 
Jan 16 22:54:50 opeth kernel: 
Jan 16 22:54:50 opeth kernel: Code: 0f 0b f0 fe 0d a0 66 2d c0 0f 88 43
09 00 00
 85 c9 74 12 8b 
 Jan 16 23:09:25 opeth kernel:  <7>VFS: Disk change detected on device
 ide1(22,0)
 Jan 16 23:55:23 opeth kernel: scsi0 : Adaptec AIC7XXX EISA/VLB/PCI SCSI
 HBA DRIV
 ER, Rev 6.2.4
 Jan 16 23:55:23 opeth kernel:         <Adaptec aic7860 Ultra SCSI
 adapter>
 Jan 16 23:55:23 opeth kernel:         aic7860: Ultra Single Channel A,
 SCSI Id=7
 , 3/253 SCBs
 Jan 16 23:55:23 opeth kernel: 
 Jan 16 23:55:39 opeth kernel:   Vendor: PLEXTOR   Model: CD-ROM PX-40TS
 Rev: 
 1.04
 Jan 16 23:55:39 opeth kernel:   Type:   CD-ROM
 ANSI 
 SCSI revision: 02
 Jan 16 23:55:39 opeth kernel: Attached scsi CD-ROM sr0 at scsi0,
 channel 0, id 3
 , lun 0
 Jan 16 23:55:39 opeth kernel: (scsi0:A:3): 20.000MB/s transfers
 (20.000MHz, offs
 et 15)
 Jan 16 23:55:39 opeth kernel: sr0: scsi-1 drive
 Jan 16 23:59:45 opeth kernel: invalid operand: 0000
 Jan 16 23:59:45 opeth kernel: CPU:    1
 Jan 16 23:59:45 opeth kernel: EIP:    0010:[d_instantiate+17/68]
 Tainted: P  
 Jan 16 23:59:45 opeth kernel: EFLAGS: 00010202
 Jan 16 23:59:45 opeth kernel: eax: 5a5a5a00   ebx: e7304660   ecx:
 e8d92e00   ed
 x: e7304690
 Jan 16 23:59:45 opeth kernel: esi: f742f520   edi: f7ea9140   ebp:
 e7304660   es
 p: ea08deb4
 Jan 16 23:59:45 opeth kernel: ds: 0018   es: 0018   ss: 0018
 Jan 16 23:59:45 opeth kernel: Process Eterm (pid: 4924,
 stackpage=ea08d000)
 Jan 16 23:59:45 opeth kernel: Stack: e8d92e00 c0179d4e e7304660
 e8d92e00 fffffff
 4 f7a7f840 ea08c000 e7304660 
 Jan 16 23:59:45 opeth kernel:        f7ea914c c0287c80 f3b1b600
 c0126938 f6fddb6
 0 f3b1b600 4038dc00 00000000 
 Jan 16 23:59:45 opeth kernel:        00000246 00000000 f7a7f840
 f7a7f8a8 f7a8362
 0 00000246 c0149613 c200f288 
 Jan 16 23:59:45 opeth kernel: Call Trace: [devfs_lookup+526/600]
 [handle_mm_faul
 t+92/184] [d_alloc+27/376] [real_lookup+122/264]
 [link_path_walk+1431/2016] 
 Jan 16 23:59:45 opeth kernel:    [path_walk+26/28] [__user_walk+53/80]
 [sys_stat
 64+25/112] [sys_ioctl+490/497] [system_call+51/56] 
 Jan 16 23:59:45 opeth kernel: 
 Jan 16 23:59:45 opeth kernel: Code: 0f 0b f0 fe 0d a0 66 2d c0 0f 88 43
 09 00 00
  85 c9 74 12 8b 
-- snip --

//this one occurred during boot
-- snip --
Jan 17 00:07:47 opeth kernel: invalid operand: 0000
Jan 17 00:07:47 opeth kernel: CPU:    0
Jan 17 00:07:47 opeth kernel: EIP:    0010:[d_instantiate+17/68]
Tainted: P 
Jan 17 00:07:47 opeth kernel: EFLAGS: 00010287
Jan 17 00:07:47 opeth kernel: eax: 5a5a5a00   ebx: f7228540   ecx:
f7215060   ed
x: f7228570
Jan 17 00:07:47 opeth kernel: esi: f7215060   edi: f7228540   ebp:
f7a810c0   es
p: f7a41f18
Jan 17 00:07:47 opeth kernel: ds: 0018   es: 0018   ss: 0018
Jan 17 00:07:47 opeth kernel: Process devfsd (pid: 26,
stackpage=f7a41000)
Jan 17 00:07:47 opeth kernel: Stack: f7252940 c0179b13 f7228540 f7215060
f722854
0 00000000 f7a41fa4 f7a82a20 
Jan 17 00:07:47 opeth kernel:        c01409be f7228540 00000000 f7a41f74
c014118
1 f7a82a20 f7a41f74 00000000 
Jan 17 00:07:47 opeth kernel:        f7854000 00000000 f7a41fa4 00000009
0000000
9 f7854005 00000000 f7854004 
Jan 17 00:07:47 opeth kernel: Call Trace:
[devfs_d_revalidate_wait+231/276] [cac
hed_lookup+46/84] [link_path_walk+1409/2016] [path_walk+26/28]
[__user_walk+53/8
0] 
Jan 17 00:07:47 opeth kernel:    [sys_stat64+25/112] [sys_read+188/196]
[system_
call+51/56] 
Jan 17 00:07:47 opeth kernel: 
Jan 17 00:07:47 opeth kernel: Code: 0f 0b f0 fe 0d a0 66 2d c0 0f 88 43
09 00 00 85 c9 74 12 8b
-- snip --

Interestingly enough, I applied rml's preempt-kernel-rml-2.4.18-pre4-1,
recompiled, rebooted, and have yet to see an oops, though I believe it's
just dumb luck thus far. Any ideas?

On Thu, Jan 17, 2002 at 09:12:29AM -0800, Lawrence Walton wrote:
> I am not sure how to debug this but it apears that 
> in 2.5.3-pre1 and in 2.5.2-dj1 DEVFS is not working.
> It started by terminals hanging and not being able to
> shutdown.
> I went to /dev/ and did a ls, it compleatly hangs that
> terminal and I cannot kill ls.
> I have the devfsd version from debian 1.3.21 .

-- 
Dan Chen                 crimsun@email.unc.edu
GPG key:   www.unc.edu/~crimsun/pubkey.gpg.asc

[-- Attachment #2: Type: application/pgp-signature, Size: 232 bytes --]

Re: DEVFS broken?

[Helge Hafting]

Lawrence Walton wrote:
> 
> I am not sure how to debug this but it apears that
> in 2.5.3-pre1 and in 2.5.2-dj1 DEVFS is not working.
> It started by terminals hanging and not being able to
> shutdown.
> I went to /dev/ and did a ls, it compleatly hangs that
> terminal and I cannot kill ls.
> I have the devfsd version from debian 1.3.21 .

I run into that inability to shutdown occationally.
There is an easy fix though:

kill -SIGUSR1 1
This is the documented way of dealing with a
remounted /dev.  shutdown, init, telinit
work normally after that.  Unless there
are other errors of course.

Helge Hafting

Re: DEVFS broken?

[Richard Gooch]

Dan Chen writes:
> Also using Debian sid here (devfsd 1.3.21-1). Over the past two days
> I've seen random nasties with devfs-v199.6 and v199.7 (I backed out
> v199.7 in my local tree because my machine refuses to finish booting
> otherwise). Machine: VIA VT82C693A/694x, PIII/1GHz SMP, 1GB HIGHMEM
> enabled.
> 
> What follows is a series of crashes culled from /var/log/kern.log
> (apologies regarding the format). Before each I'll explain what I did to
> possibly invoke it. All are with devfs-v199.6 and devfsd-1.3.21 running
> 2.4.18-pre4 + ext3-2.4-0.9.17-2418p3 + ide.2.4.16.12102001:
> 
> # modprobe aic7xxx   //I tried to start an Eterm in X 4.1.0.1 afterward
> -- snip --
> Jan 16 22:54:50 opeth kernel: invalid operand: 0000

In future, please just send dmesg output, rather than
/var/log/kern.log output. The former doesn't have all the
date+hostname+" kernel: " crap that syslog puts in.

devfs-patch-v199.6 has a race that causes the Oops. devfs-patch-v199.7
fixes this race, but unfortunately had a silly oversight which could
cause deadlocks under some circumstances (and of course several days
of testing on my box didn't show it:-().

Please apply this patch on top of devfs-patch-v199.7 or on top of
plain 2.5.2, and let me know the result.

				Regards,

					Richard....
Permanent: rgooch@atnf.csiro.au
Current:   rgooch@ras.ucalgary.ca

diff -urN linux-2.5.3-pre2/fs/devfs/base.c linux/fs/devfs/base.c
--- linux-2.5.3-pre2/fs/devfs/base.c	Mon Jan 14 10:40:29 2002
+++ linux/fs/devfs/base.c	Sun Jan 20 12:09:55 2002
@@ -1,6 +1,6 @@
 /*  devfs (Device FileSystem) driver.
 
-    Copyright (C) 1998-2001  Richard Gooch
+    Copyright (C) 1998-2002  Richard Gooch
 
     This library is free software; you can redistribute it and/or
     modify it under the terms of the GNU Library General Public
@@ -604,6 +604,9 @@
     20020113   Richard Gooch <rgooch@atnf.csiro.au>
 	       Fixed (rare, old) race in <devfs_lookup>.
   v1.9
+    20020120   Richard Gooch <rgooch@atnf.csiro.au>
+	       Fixed deadlock bug in <devfs_d_revalidate_wait>.
+  v1.10
 */
 #include <linux/types.h>
 #include <linux/errno.h>
@@ -636,7 +639,7 @@
 #include <asm/bitops.h>
 #include <asm/atomic.h>
 
-#define DEVFS_VERSION            "1.9 (20020113)"
+#define DEVFS_VERSION            "1.10 (20020120)"
 
 #define DEVFS_NAME "devfs"
 
@@ -2878,13 +2881,16 @@
     struct devfs_lookup_struct *lookup_info = dentry->d_fsdata;
     DECLARE_WAITQUEUE (wait, current);
 
-    if ( !dentry->d_inode && is_devfsd_or_child (fs_info) )
+    if ( is_devfsd_or_child (fs_info) )
     {
 	devfs_handle_t de = lookup_info->de;
 	struct inode *inode;
 
-	DPRINTK (DEBUG_I_LOOKUP, "(%s): dentry: %p de: %p by: \"%s\"\n",
-		 dentry->d_name.name, dentry, de, current->comm);
+	DPRINTK (DEBUG_I_LOOKUP,
+		 "(%s): dentry: %p inode: %p de: %p by: \"%s\"\n",
+		 dentry->d_name.name, dentry, dentry->d_inode, de,
+		 current->comm);
+	if (dentry->d_inode) return 1;
 	if (de == NULL)
 	{
 	    read_lock (&parent->u.dir.lock);

Re: DEVFS broken?

[Dan Chen]

[-- Attachment #1: Type: text/plain, Size: 469 bytes --]

Grrr, I realized that I sent undecoded output instead of the ksymoops I
had here. I'll apply this patch and get back to you.

On Sun, Jan 20, 2002 at 01:04:36PM -0700, Richard Gooch wrote:
> In future, please just send dmesg output, rather than
> /var/log/kern.log output. The former doesn't have all the
> date+hostname+" kernel: " crap that syslog puts in.

-- 
Dan Chen                 crimsun@email.unc.edu
GPG key:   www.unc.edu/~crimsun/pubkey.gpg.asc

[-- Attachment #2: Type: application/pgp-signature, Size: 232 bytes --]

Re: DEVFS broken?

[Dan Chen]

[-- Attachment #1: Type: text/plain, Size: 338 bytes --]

Working well so far: booted and no oopses yet. Thanks!

On Sun, Jan 20, 2002 at 01:04:36PM -0700, Richard Gooch wrote:
> Please apply this patch on top of devfs-patch-v199.7 or on top of
> plain 2.5.2, and let me know the result.

-- 
Dan Chen                 crimsun@email.unc.edu
GPG key:   www.unc.edu/~crimsun/pubkey.gpg.asc

[-- Attachment #2: Type: application/pgp-signature, Size: 232 bytes --]

Re: DEVFS broken?

[Dan Chen]

[-- Attachment #1: Type: text/plain, Size: 211 bytes --]

Correct.

On Sun, Jan 20, 2002 at 05:50:26PM -0700, Richard Gooch wrote:
> I assume no deadlock either?

-- 
Dan Chen                 crimsun@email.unc.edu
GPG key:   www.unc.edu/~crimsun/pubkey.gpg.asc

[-- Attachment #2: Type: application/pgp-signature, Size: 232 bytes --]