Re: [PATCH] reiserfs: handle cnode allocation failure gracefully

[Jeff Mahoney <jeffm@suse.com>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hans Reiser wrote:
> Jeff Mahoney wrote:
> 
>> If an external device is used for a journal, by default it will use the
>> entire device. The reiserfs journal code allocates structures per journal
>> block when it mounts the file system. If the journal device is too large, and
>> memory cannot be allocated for the structures, it will continue and ultimately
>> panic when it can't pull one off the free list.
>>
>> This patch handles the allocation failure gracefully and prints an error
>> message at mount time.
>>
>> Changes: Updated error message to be more descriptive to the user.
>>
>> Signed-off-by: Jeff Mahoney <jeffm@suse.com>
>>
>> diff -ruNpX dontdiff linux-2.6.13.orig/fs/reiserfs/journal.c linux-2.6.13.fixed/fs/reiserfs/journal.c
>> --- linux-2.6.13.orig/fs/reiserfs/journal.c	2005-09-16 11:42:58.000000000 -0400
>> +++ linux-2.6.13.fixed/fs/reiserfs/journal.c	2005-11-23 19:14:17.000000000 -0500
>> @@ -2757,6 +2757,13 @@ int journal_init(struct super_block *p_s
>> 	journal->j_cnode_used = 0;
>> 	journal->j_must_wait = 0;
>>
>> +	if (journal->j_cnode_free == 0) {
>> +        	reiserfs_warning(p_s_sb, "journal-2004: Journal cnode memory "
>> +		                 "allocation failed. Journal is too large "
>> +				 "for available memory.");
>> +        	goto free_and_return;
>> +	}
>> +
>> 	init_journal_hash(p_s_sb);
>> 	jl = journal->j_current_jl;
>> 	jl->j_list_bitmap = get_list_bitmap(p_s_sb, jl);
>>
>>  
>>
> This is better than handling it ungracefully but..... can there be a
> memory shortage for some other reason and then this prints the wrong
> diagnostic? Forgive me, it is hard to know the answer looking at the
> patch by itself.....

Well, the error message is always true. ;) The failure will always be
because it couldn't allocate enough memory to support the internal
housekeeping for the journal. However, it's possible for other causes
than JUST the journal being oversized to cause the allocation failure.
On x86, the vmalloc area is limited to ~128MB. If other sources
(including other reiserfs file systems) are taking up most of the
vmalloc area, it could fail with a standard journal size. When the
vmalloc limit isn't a factor, memory could simply be unavailable to fill
the request, even a small one.

In most cases, though, it's a result of allocate_cnodes() asking for too
much memory. Using the standard 8192 block journal size, a 32-bit system
will allocate 288K for cnodes. If an external 8 GB partition is used,
2*1024*1024 blocks will be needed, causing 72 MB of RAM to be allocated
for cnodes alone.

I wouldn't be opposed to changing the error message to:
journal-2004: Journal cnode memory allocation failed (%lu bytes).
Journal may be too large or your system does not have enough available
memory.

- -Jeff

- --
Jeff Mahoney
SUSE Labs
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFDi3YwLPWxlyuTD7IRAugMAKCQjNIalS04Ul8447C8l3OvDbioAACeOjea
R6Z256+nUh5o7Wd6bhawRIw=
=rZIi
-----END PGP SIGNATURE-----