tux3: crash upon writing data with bs=1M

[OGAWA Hirofumi]

Daniel Phillips <phillips at phunq.net> writes:

> On Wednesday, January 02, 2013 03:27:38 PM Dongsu Park wrote:
>>   # insmod ./user/kernel/tux3.ko
>>   # ./user/tux3 mkfs -b 4096 /dev/vdb
>>   # mount -t tux3 /dev/vdb /mnt/tux3
>>   # dd if=/dev/zero of=/mnt/tux3/test1 bs=1M count=256
>>
>> ---------------------------------------------------------------------
>> [  259.269921] tux3_fill_super: s_blocksize 4096
>> [  339.038377] balloc: couldn't balloc: blocks 46390
>> [  339.038434] ------------[ cut here ]------------
>> [  339.038736] kernel BUG at /home/dpark/Dev/tux3/user/kernel/buffer_writeback.c:67!
>>
>> [  339.040036]  [<ffffffffa03f495a>] ? tux3_write_begin+0x3a/0x570 [tux3]
>> [  339.040036]  [<ffffffffa03f2190>] ? blockget+0xd0/0xd0 [tux3]
>> [  339.040036]  [<ffffffffa03fca83>] ? tux3_flush_inode+0x323/0x3e0 [tux3]
>> [  339.040036]  [<ffffffffa03fcbd4>] ? tux3_flush_inodes+0x94/0xb0 [tux3]
>> [  339.040036]  [<ffffffffa03ee435>] ? flush_delta_work+0x135/0x5a0 [tux3]
>> [  339.040036]  [<ffffffffa03ee300>] ? setup_sb+0x30/0x30 [tux3]
>> [  339.040036]  [<ffffffffa03ee300>] ? setup_sb+0x30/0x30 [tux3]
>> [  339.040036]  [<ffffffff810609b6>] ? kthread+0xc6/0xd0
>> [  339.040036]  [<ffffffff814fcec4>] ? kernel_thread_helper+0x4/0x10
>> ...
>
> Thanks for the report, and thanks for proving that you can build Tux3 as a 
> module. Tux3 hit an assert because bufvec->contig is not empty as it should 
> be, definitely a bug. Let the debugging begin.

Hi,

Thanks for the report. This is known issue what I'm trying to fix now.
Current balloc() supports only limited length of blocks (maximum is
blocksize * 8 blocks for now).

But, the inode of this delta is trying to allocate 46390 blocks
(limitation is 32768 on blocksize==4096). So, it exceeds current
limitation.

Workaround for this is, write data separately, and sync for each write. 

Well, I'll fix this. But I'm going to rewrite balloc() more or less, not
just fix, so it may take a time than usual.

Thanks.
-- 
OGAWA Hirofumi <hirofumi at mail.parknet.co.jp>