Hang observed while reading a block device using a simple block driver using the file as a back-end data store

Hang observed while reading a block device using a simple block driver using the file as a back-end data store

[Rohit B]

[-- Attachment #1: Type: text/plain, Size: 3534 bytes --]

Hi,
I am trying to implement a block driver and facing an issue. Please find 
attached a simple code snippet which reproduces this hang. I have used and 
modified the Simple Block driver(sbull) example from Linux Device Drivers (3rd 
Ed) 
to simulate the problem. The example has been modified to use a file as the 
back-end instead of memory. 

The problem that I see with this example is that, when an attempt is made to 
read 
the block device using dd, the vfs_read function call in thesbull_make_request 
hangs. 
This sometimes causes the insmod of sbull to hang. And almost always causes the 
dd command to hang. This hang is only observed in the linux kernel version 
2.6.27.45-0.1 (SLES 11). I have also tried the same code example on 
Liunx 2.6.16.46-0.12 (SLES 10 and its service packs), but the hang is not 
observed there. While I understand that this is not a normal use-case, a block 
driver will not use a file as a back-end, I wanted to understand how to fix my 
example code to solve this hang problem for the latest kernel. Any help in this 
regard is highly appreciated.

Distribution Used SLES 11
Kernels checked: 2.6.27.45-0.1
Description of the Source Attached: This attached source uses the simple 
no-queue version and defines a make-request. The make-request reads the file for 

the data and services the bio. 

Steps to reproduce the problem:
1.      untar the sbull.tar.gz and cd into sbull and do a make. Make also 
creates a 512MB file in the current
directory which is used as the back-end.
2.      insmod sbull.ko, a block device /dev/sbulla is created.
3.      dd if=/dev/sbulla of=./file (hangs)

Stack Trace of the Hang:
crash> bt 4289
PID: 4289   TASK: ffff880026120200  CPU: 1   COMMAND: "dd"
#0 [ffff8800260a15c8] schedule at ffffffff8049c0a7
#1 [ffff8800260a16d0] io_schedule at ffffffff8049c1c6
#2 [ffff8800260a16f0] sync_page at ffffffff80282d5d
#3 [ffff8800260a1700] __lock_page_killable at ffffffff8028305f
#4 [ffff8800260a1760] do_generic_file_read at ffffffff80283e84
#5 [ffff8800260a17f0] generic_file_aio_read at ffffffff80284a94
#6 [ffff8800260a1880] do_sync_read at ffffffff802b1880
#7 [ffff8800260a19b0] vfs_read at ffffffff802b227a
#8 [ffff8800260a19e0] sbull_make_request at ffffffffa02bd3f3
#9 [ffff8800260a1a40] generic_make_request at ffffffff8033a7f8
#10 [ffff8800260a1ae0] submit_bio at ffffffff8033a901
#11 [ffff8800260a1b20] submit_bh at ffffffff802d2ae2
#12 [ffff8800260a1b40] block_read_full_page at ffffffff802d5b08
#13 [ffff8800260a1c00] __do_page_cache_readahead at ffffffff8028aeed
#14 [ffff8800260a1c80] ondemand_readahead at ffffffff8028b22b
#15 [ffff8800260a1cc0] do_generic_file_read at ffffffff80283d52
#16 [ffff8800260a1d50] generic_file_aio_read at ffffffff80284a94
#17 [ffff8800260a1de0] do_sync_read at ffffffff802b1880
#18 [ffff8800260a1f10] vfs_read at ffffffff802b227a
#19 [ffff8800260a1f40] sys_read at ffffffff802b23df
#20 [ffff8800260a1f80] system_call_fastpath at ffffffff8020bffb
    RIP: 00007ffabd782ee0  RSP: 00007fff3b4131f8  RFLAGS: 00000217
    RAX: 0000000000000000  RBX: ffffffff8020bffb  RCX: 00007ffabd782ee0
    RDX: 0000000000000200  RSI: 0000000000611000  RDI: 0000000000000000
    RBP: 0000000000000000   R8: 0000000000000000   R9: 0000000000000000
    R10: 0000000000000000  R11: 0000000000000246  R12: 0000000000611000
    R13: 0000000000000200  R14: 0000000000000800  R15: 0000000000000000
    ORIG_RAX: 0000000000000000  CS: 0033  SS: 002b
crash>


Rgds,
Rohit


[-- Attachment #2: sbull.tar.gz --]
[-- Type: application/x-gzip, Size: 3569 bytes --]