From: Damien Le Moal <Damien.LeMoal@wdc.com>
To: Ming Lei <ming.lei@redhat.com>
Cc: Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>,
Bob Liu <bob.liu@oracle.com>, "axboe@kernel.dk" <axboe@kernel.dk>,
"linux-block@vger.kernel.org" <linux-block@vger.kernel.org>
Subject: Re: [PATCH] block: Bail out iteration functions upon SIGKILL.
Date: Wed, 13 Nov 2019 07:11:36 +0000 [thread overview]
Message-ID: <BYAPR04MB581697D2911048D329DFF5B2E7760@BYAPR04MB5816.namprd04.prod.outlook.com> (raw)
In-Reply-To: 20191113065523.GA1985@ming.t460p
On 2019/11/13 15:55, Ming Lei wrote:
> On Wed, Nov 13, 2019 at 01:54:14AM +0000, Damien Le Moal wrote:
>> On 2019/11/12 23:48, Tetsuo Handa wrote:
>> [...]
>>>>> +static int blk_should_abort(struct bio *bio)
>>>>> +{
>>>>> + int ret;
>>>>> +
>>>>> + cond_resched();
>>>>> + if (!fatal_signal_pending(current))
>>>>> + return 0;
>>>>> + ret = submit_bio_wait(bio);
>>>>
>>>> This will change the behavior of __blkdev_issue_discard() to a sync IO
>>>> execution instead of the current async execution since submit_bio_wait()
>>>> call is the responsibility of the caller (e.g. blkdev_issue_discard()).
>>>> Have you checked if users of __blkdev_issue_discard() are OK with that ?
>>>> f2fs, ext4, xfs, dm and nvme use this function.
>>>
>>> I'm not sure...
>>>
>>>>
>>>> Looking at f2fs, this does not look like it is going to work as expected
>>>> since the bio setup, including end_io callback, is done after this
>>>> function is called and a regular submit_bio() execution is being used.
>>>
>>> Then, just breaking the iteration like below?
>>> nvmet_bdev_execute_write_zeroes() ignores -EINTR if "*biop = bio;" is done. Is that no problem?
>>>
>>> --- a/block/blk-lib.c
>>> +++ b/block/blk-lib.c
>>> @@ -7,6 +7,7 @@
>>> #include <linux/bio.h>
>>> #include <linux/blkdev.h>
>>> #include <linux/scatterlist.h>
>>> +#include <linux/sched/signal.h>
>>>
>>> #include "blk.h"
>>>
>>> @@ -30,6 +31,7 @@ int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
>>> struct bio *bio = *biop;
>>> unsigned int op;
>>> sector_t bs_mask;
>>> + int ret = 0;
>>>
>>> if (!q)
>>> return -ENXIO;
>>> @@ -76,10 +78,14 @@ int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
>>> * is disabled.
>>> */
>>> cond_resched();
>>> + if (fatal_signal_pending(current)) {
>>> + ret = -EINTR;
>>> + break;
>>> + }
>>> }
>>>
>>> *biop = bio;
>>> - return 0;
>>> + return ret;
>>
>> This will leak a bio as blkdev_issue_discard() executes the bio only in
>> the case "if (!ret && bio)". So that does not work as is, unless all
>> callers of __blkdev_issue_discard() are also changed. Same problem for
>> the other __blkdev_issue_xxx() functions.
>>
>> Looking more into this, if an error is returned here, no bio should be
>> returned and we need to make sure that all started bios are also
>> completed. So your helper blk_should_abort() did the right thing calling
>> submit_bio_wait(). However, I Think it would be better to fail
>> immediately the current loop bio instead of executing it and then
>> reporting the -EINTR error, unconditionally, regardless of what the
>> started bios completion status is.
>>
>> This could be done with the help of a function like this, very similar
>> to submit_bio_wait().
>>
>> void bio_chain_end_wait(struct bio *bio)
>> {
>> DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map);
>>
>> bio->bi_private = &done;
>> bio->bi_end_io = submit_bio_wait_endio;
>> bio->bi_opf |= REQ_SYNC;
>> bio_endio(bio);
>> wait_for_completion_io(&done);
>> }
>>
>> And then your helper function becomes something like this:
>>
>> static int blk_should_abort(struct bio *bio)
>> {
>> int ret;
>>
>> cond_resched();
>> if (!fatal_signal_pending(current))
>> return 0;
>>
>> if (bio_flagged(bio, BIO_CHAIN))
>> bio_chain_end_wait(bio);
>> bio_put(bio);
>>
>> return -EINTR;
>> }
>>
>> Thoughts ?
>
> DISCARD request can be quite big, and any sync bio submission may cause
> serious performance regression.
Yes indeed. But if the bio issuing loop is interrupted with discard BIOs
already issued, I do not think there is any other choice but to wait for
their completion before returning.
> Not mention blkdev_issue_discard() may be called in non-block context.
This loop is calling cond_resched(), which checks might_sleep(). So
certainly this function can block, no ?
--
Damien Le Moal
Western Digital Research
next prev parent reply other threads:[~2019-11-13 7:11 UTC|newest]
Thread overview: 14+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-10-23 7:56 INFO: task syz-executor can't die for more than 143 seconds. (2) syzbot
2019-10-24 10:08 ` Tetsuo Handa
2019-10-28 8:51 ` Bob Liu
2019-11-08 11:41 ` [PATCH] block: Bail out iteration functions upon SIGKILL Tetsuo Handa
2019-11-08 18:13 ` Chaitanya Kulkarni
2019-11-08 22:18 ` Chaitanya Kulkarni
2019-11-12 4:05 ` Damien Le Moal
2019-11-12 14:47 ` Tetsuo Handa
2019-11-13 1:54 ` Damien Le Moal
2019-11-13 6:55 ` Ming Lei
2019-11-13 7:11 ` Damien Le Moal [this message]
2019-11-13 7:49 ` Ming Lei
2019-11-15 10:05 ` Tetsuo Handa
2019-11-18 0:02 ` Damien Le Moal
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=BYAPR04MB581697D2911048D329DFF5B2E7760@BYAPR04MB5816.namprd04.prod.outlook.com \
--to=damien.lemoal@wdc.com \
--cc=axboe@kernel.dk \
--cc=bob.liu@oracle.com \
--cc=linux-block@vger.kernel.org \
--cc=ming.lei@redhat.com \
--cc=penguin-kernel@i-love.sakura.ne.jp \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).