| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <ffd3b5f4-4a43-4f5b-b53a-1849f4b2fb71@molgen.mpg.de>
Date: Sun, 15 Jun 2025 11:20:31 +0200
From: Paul Menzel <pmenzel@...gen.mpg.de>
To: Zheng Qixing <zhengqixing@...wei.com>,
Zheng Qixing <zhengqixing@...weicloud.com>
Cc: song@...nel.org, yukuai3@...wei.com, linux-raid@...r.kernel.org,
linux-kernel@...r.kernel.org, yi.zhang@...wei.com, yangerkun@...wei.com
Subject: Re: [PATCH] md/raid1,raid10: fix IO handle for REQ_NOWAIT
Dear Zheng,
Am 14.06.25 um 08:50 schrieb Zheng Qixing:
> Please disregard the previous reply email, as it contained garbled text.
Thank you for noticing this, and resending.
> 在 2025/6/13 16:02, Paul Menzel 写道:
>> Am 12.06.25 um 15:21 schrieb Zheng Qixing:
>>> From: Zheng Qixing <zhengqixing@...wei.com>
>>>
>>> IO with REQ_NOWAIT should not set R1BIO_Uptodate when it fails,
>>> and bad blocks should also be cleared when REQ_NOWAIT IO succeeds.
>>
>> It’d be great if you could add an explanation for the *should*. Why
>> should it not be done?
>>
>> Do you have a reproducer for this?
>
> If we set R1BIO_Uptodate when IO with REQ_NOWAIT fails, the request will
> return a success.
Understood. So no command to check for this automatically on a test system.
For the explanation, I guess my problem is, that I was not familiar with
REQ_NOWAIT, which means that it fails for blocked IO. (If I am correct.)
> But actually it should return BLK_STS_IOERR or BLK_STS_AGAIN, right?
Sorry, I do not know. Hopefully the maintainers can answer this.
>>> Fixes: 9f346f7d4ea7 ("md/raid1,raid10: don't handle IO error for REQ_RAHEAD and REQ_NOWAIT")
>>> Signed-off-by: Zheng Qixing <zhengqixing@...wei.com>
>>> ---
>>> drivers/md/raid1.c | 11 ++++++-----
>>> drivers/md/raid10.c | 9 +++++----
>>> 2 files changed, 11 insertions(+), 9 deletions(-)
>>>
>>> diff --git a/drivers/md/raid1.c b/drivers/md/raid1.c
>>> index 19c5a0ce5a40..a1cddd24b178 100644
>>> --- a/drivers/md/raid1.c
>>> +++ b/drivers/md/raid1.c
>>> @@ -455,13 +455,13 @@ static void raid1_end_write_request(struct bio *bio)
>>> struct md_rdev *rdev = conf->mirrors[mirror].rdev;
>>> sector_t lo = r1_bio->sector;
>>> sector_t hi = r1_bio->sector + r1_bio->sectors;
>>> - bool ignore_error = !raid1_should_handle_error(bio) ||
>>> - (bio->bi_status && bio_op(bio) == REQ_OP_DISCARD);
>>> + bool discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD;
>>
>> Excuse my ignorance. What is the difference between ignore and discard?
>
> REQ_OP_DISCARD is a operation type while REQ_NOWAIT is just a request flag.
>
> These two can be combined together. IO with REQ_NOWAIT can fail early, even
> though the storage medium is fine. So, we better handle this type of
> error specially.
>
> I hope this clarifies your doubts.
Sorry about being not clear enough. My question was more about changing
the naming of the variable.
>>> /*
>>> * 'one mirror IO has finished' event handler:
>>> */
>>> - if (bio->bi_status && !ignore_error) {
>>> + if (bio->bi_status && !discard_error &&
>>> + raid1_should_handle_error(bio)) {
>>> set_bit(WriteErrorSeen, &rdev->flags);
>>> if (!test_and_set_bit(WantReplacement, &rdev->flags))
>>> set_bit(MD_RECOVERY_NEEDED, &
>>> @@ -507,12 +507,13 @@ static void raid1_end_write_request(struct bio *bio)
>>> * check this here.
>>> */
>>> if (test_bit(In_sync, &rdev->flags) &&
>>> - !test_bit(Faulty, &rdev->flags))
>>> + !test_bit(Faulty, &rdev->flags) &&
>>> + (!bio->bi_status || discard_error))
>>> set_bit(R1BIO_Uptodate, &r1_bio->state);
>>> /* Maybe we can clear some bad blocks. */
>>> if (rdev_has_badblock(rdev, r1_bio->sector, r1_bio->sectors) &&
>>> - !ignore_error) {
>>> + !bio->bi_status) {
>>> r1_bio->bios[mirror] = IO_MADE_GOOD;
>>> set_bit(R1BIO_MadeGood, &r1_bio->state);
>>> }
>>> diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c
>>> index b74780af4c22..1848947b0a6d 100644
>>> --- a/drivers/md/raid10.c
>>> +++ b/drivers/md/raid10.c
>>> @@ -458,8 +458,8 @@ static void raid10_end_write_request(struct bio *bio)
>>> int slot, repl;
>>> struct md_rdev *rdev = NULL;
>>> struct bio *to_put = NULL;
>>> - bool ignore_error = !raid1_should_handle_error(bio) ||
>>> - (bio->bi_status && bio_op(bio) == REQ_OP_DISCARD);
>>> + bool discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD;
>>> + bool ignore_error = !raid1_should_handle_error(bio) || discard_error;
>>> dev = find_bio_disk(conf, r10_bio, bio, &slot, &repl);
>>> @@ -522,13 +522,14 @@ static void raid10_end_write_request(struct bio *bio)
>>> * check this here.
>>> */
>>> if (test_bit(In_sync, &rdev->flags) &&
>>> - !test_bit(Faulty, &rdev->flags))
>>> + !test_bit(Faulty, &rdev->flags) &&
>>> + (!bio->bi_status || discard_error))
>>> set_bit(R10BIO_Uptodate, &r10_bio->state);
>>> /* Maybe we can clear some bad blocks. */
>>> if (rdev_has_badblock(rdev, r10_bio->devs[slot].addr,
>>> r10_bio->sectors) &&
>>> - !ignore_error) {
>>> + !bio->bi_status) {
>>> bio_put(bio);
>>> if (repl)
>>> r10_bio->devs[slot].repl_bio = IO_MADE_GOOD;
Kind regards,
Paul
PS: As it’s two hunks only connected through REQ_NOWAIT, maybe make it
two commits: one for raid1 and one for raid10? Feel free to ignore.
Powered by blists - more mailing lists