[Stgt-devel] [PATCH] Proof of Concept -> Separate thread init/tear-down from backing store open/close.

[Richard Sharpe]

On Wed, Jul 30, 2008 at 5:50 AM, FUJITA Tomonori
<fujita.tomonori at lab.ntt.co.jp> wrote:
> On Wed, 30 Jul 2008 17:44:25 +1000
> Mark Harvey <markh794 at gmail.com> wrote:
>
>> Patch also included as an attachment - just in case.
>>
>> >From 715e434fff633ff8f346d181aeae3f27c9564553 Mon Sep 17 00:00:00 2001
>> From: Mark Harvey <markh794 at gmail.com>
>> Date: Wed, 30 Jul 2008 17:27:28 +1000
>> Subject: Separate thread init/tear-down from backing store open/close.
>>
>> Fix segfault when lu created without a backing store.
>>  - Devices defined as 'removable' are able to be configured without a backing
>>    store defined.
>>  - thread init via bs_init() which is called for all logical units
>>  - thread tear-down via bs_exit() which is called for all logical units.
>>  - bs_open() limited to opening backing store path (called when required).
>>  - bs_close() limited to closing backing store fd (called when required).
>>
>> Note: bs_aio and bs_mmap compile-tested only.
>>
>> Signed-off-by: Mark Harvey <markh794 at gmail.com>
>> ---
>>  usr/bs_aio.c  |   25 +++++++++++++++++--------
>>  usr/bs_mmap.c |   22 ++++++++++++----------
>>  usr/bs_rdwr.c |   25 ++++++++++++++-----------
>>  usr/bs_ssc.c  |   25 +++++++++++++------------
>>  usr/target.c  |   15 ++++++++++++---
>>  usr/tgtd.h    |    2 ++
>>  6 files changed, 70 insertions(+), 44 deletions(-)
>
> Thanks, looks nice.
>
> We need to check the return value of bs_init but the rest is fine by
> me (I've attached a patch).
>
> Have you tested this?
>
> If it works fine for ssc and mmc, I'll apply the patches.

I have tested the patch with SSC with a two drive, 8 tape library,
loaded two tapes, ran two concurrent tar backups, then unloaded one
tape and then did another tar backup to the other tape drive and all
seems to work OK.

I will try the same with MMC.