ltp directio test causes oops on 2.4.20pre5aa2

ltp directio test causes oops on 2.4.20pre5aa2

[rwhron]

The linux test project diotest1 running on reiserfs gave oops:

No modules in ksyms, skipping objects
Warning (read_lsmod): no symbols in lsmod, is /proc/modules a valid lsmod file?
Unable to handle kernel NULL pointer dereference at virtual address 00000012
8015d6d3
*pde = 00000000
Oops: 0000 2.4.20-pre5aa2 #1 Mon Sep 9 17:31:33 EDT 2002
CPU:    0
EIP:    0010:[<8015d6d3>]    Not tainted
Using defaults from ksymoops -t elf32-i386 -a i386
EFLAGS: 00010256
eax: 00000002   ebx: 8d019e50   ecx: 00000000   edx: 851c8840
esi: 00000000   edi: 00000000   ebp: 8d019e08   esp: 8d019d00
ds: 0018   es: 0018   ss: 0018
Process diotest1 (pid: 2965, stackpage=8d019000)
Stack: 8d019e50 00000000 00000000 8358a240 001c8454 8016bc2c 00000000 001c8454
       b3740740 8015ea94 8d019d68 8d019f04 bff9d800 b3740740 8d019d68 8358a240
       8358a240 00000000 00000000 8d019d6c 0000003f 0000b474 00000000 00000000
Call Trace:    [<8016bc2c>] [<8015ea94>] [<8015b513>] [<8015d55a>] [<801388f9>]
  [<80121000>] [<8012463d>] [<80124796>] [<801604a5>] [<8015d540>] [<80128965>]
  [<8012a7b0>] [<80126022>] [<8012a8b1>] [<80135796>] [<8010880b>]
Code: 8a 48 10 0f a5 f7 d3 e6 f6 c1 20 74 04 89 f7 31 f6 83 c6 01


>>EIP; 8015d6d3 <reiserfs_get_block+43/e50>   <=====

Trace; 8016bc2c <pathrelse+1c/30>
Trace; 8015ea94 <reiserfs_update_sd+164/180>
Trace; 8015b513 <reiserfs_add_entry+403/430>
Trace; 8015d55a <reiserfs_get_block_direct_io+1a/40>
Trace; 801388f9 <generic_direct_IO+c9/140>
Trace; 80121000 <sys_setreuid+1a0/1b0>
Trace; 8012463d <get_user_pages+fd/190>
Trace; 80124796 <map_user_kiobuf+c6/110>
Trace; 801604a5 <reiserfs_direct_io+25/30>
Trace; 8015d540 <reiserfs_get_block_direct_io+0/40>
Trace; 80128965 <generic_file_direct_IO+1a5/220>
Trace; 8012a7b0 <do_generic_file_write+630/6c0>
Trace; 80126022 <__vma_link+62/b0>
Trace; 8012a8b1 <generic_file_write+71/b0>
Trace; 80135796 <sys_write+96/f0>
Trace; 8010880b <system_call+33/38>

The subsequent diotest[2-6] also printed oops.

diotest1 is unkillable:
USER       PID %CPU %MEM   VSZ  RSS TTY      STAT START   TIME COMMAND
root      2977  0.0  0.0  1336  408 tty1     D    16:39   0:00 diotest1 -b 65536

-- 
Randy Hron
http://home.earthlink.net/~rwhron/kernel/bigbox.html

Re: ltp directio test causes oops on 2.4.20pre5aa2

[Andrea Arcangeli]

while merging the new nfs stuff with nfs-o_direct this bit was left out:

> diff -u --recursive --new-file linux-2.4.18-pathconf/mm/filemap.c linux-2.4.18-odirect/mm/filemap.c
> --- linux-2.4.18-pathconf/mm/filemap.c	Wed Feb 20 17:14:28 2002
> +++ linux-2.4.18-odirect/mm/filemap.c	Wed Feb 20 17:17:19 2002
> @@ -1515,7 +1515,7 @@
594a592,606
> --- 2.4.19pre3aa1/fs/reiserfs/inode.c.~1~	Tue Mar 12 00:07:18 2002
> +++ 2.4.19pre3aa1/fs/reiserfs/inode.c	Tue Mar 12 01:24:21 2002
> @@ -2161,10 +2161,11 @@
>  	}
>  }
>  
> -static int reiserfs_direct_io(int rw, struct inode *inode, 
> +static int reiserfs_direct_io(int rw, struct file * filp,
>                                struct kiobuf *iobuf, unsigned long blocknr,
>  			      int blocksize) 
>  {
> +    struct inode * inode = filp->f_dentry->d_inode->i_mapping->host;
>      return generic_direct_IO(rw, inode, iobuf, blocknr, blocksize,
>                               reiserfs_get_block_direct_io) ;
>  }

Trond could you include the reiserfs part in your nfs-o_direct too?
(ext3 has o_direct in -aa only so you don't need it) I'll keep them in
separate files so they won't be forgotten again.  In the meantime you
can test with this incremental fix:

--- 2.4.19pre3aa1/fs/reiserfs/inode.c.~1~	Tue Mar 12 00:07:18 2002
+++ 2.4.19pre3aa1/fs/reiserfs/inode.c	Tue Mar 12 01:24:21 2002
@@ -2161,10 +2161,11 @@
 	}
 }
 
-static int reiserfs_direct_io(int rw, struct inode *inode, 
+static int reiserfs_direct_io(int rw, struct file * filp,
                               struct kiobuf *iobuf, unsigned long blocknr,
 			      int blocksize) 
 {
+    struct inode * inode = filp->f_dentry->d_inode->i_mapping->host;
     return generic_direct_IO(rw, inode, iobuf, blocknr, blocksize,
                              reiserfs_get_block_direct_io) ;
 }
--- 2.4.20pre5aa2/fs/ext3/inode.c.~1~	Mon Sep  9 02:38:08 2002
+++ 2.4.20pre5aa2/fs/ext3/inode.c	Tue Sep 10 05:22:18 2002
@@ -1385,9 +1385,10 @@ static int ext3_releasepage(struct page 
 }
 
 static int
-ext3_direct_IO(int rw, struct inode *inode, struct kiobuf *iobuf,
+ext3_direct_IO(int rw, struct file * filp, struct kiobuf *iobuf,
 		unsigned long blocknr, int blocksize)
 {
+	struct inode * inode = filp->f_dentry->d_inode->i_mapping->host;
 	struct ext3_inode_info *ei = EXT3_I(inode);
 	handle_t *handle = NULL;
 	int ret;

thanks for the as usual accurate report,

Andrea

PS. Later I'll make sure the new sched can't introduce starvation, then
    I'll upload a new -aa with these few bits.

Re: ltp directio test causes oops on 2.4.20pre5aa2

[rwhron]

> In the meantime you
> can test with this incremental fix:

I've ran ltp on reiserfs and ext3 and 2.4.19pre5aa2 
passes the ltp directio tests with no oops.  The new 
nanosleep tests in ltp-20020910 pass too.

Thanks!
-- 
Randy Hron
http://home.earthlink.net/~rwhron/kernel/bigbox.html