Skip to contentSkip to conversation

xfs: Call kiocb_modified() for buffered write

PATCH In review Xiao Yang 2 participants 5 messages Nov 16, 2022 – Jan 13, 2023 XFS filesystem

lore
Share & tools
lore

Xiao Yang

author
#0 lore raw

PATCH 1/1 xfs: Call kiocb_modified() for buffered write

kiocb_modified() should be used for sync/async buffered write because it will return -EAGAIN when IOCB_NOWAIT is set. Unfortunately, kiocb_modified() is used by the common xfs_file_write_checks() which is called by all types of write(i.e. buffered/direct/dax write). This issue makes generic/471 with xfs always get the following error:
--------------------------------------------------------
QA output created by 471
pwrite: Resource temporarily unavailable
wrote 8388608/8388608 bytes at offset 0
XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
pwrite: Resource temporarily unavailable
...
--------------------------------------------------------

Fixes 1aa91d9c9933 ("xfs: Add async buffered write support")
---
 fs/xfs/xfs_file.c | 3 +++
 1 file changed, 3 insertions(+)
fs/xfs/xfs_file.c +3 -0
index e462d39c840e..561fab3a49c7 100644
@@ -417,6 +417,9 @@ xfs_file_write_checks(
 		spin_unlock(&ip->i_flags_lock);
 
 out:
+	if (IS_DAX(inode) || (iocb->ki_flags & IOCB_DIRECT))
+		return file_modified(file);
+
 	return kiocb_modified(iocb);
 }
 
Signature · 1 line
2.21.0
#1 lore raw
Quoted 5 lines · Xiao Yang

kiocb_modified() should be used for sync/async buffered write because it will return -EAGAIN when IOCB_NOWAIT is set. Unfortunately, kiocb_modified() is used by the common xfs_file_write_checks() which is called by all types of write(i.e. buffered/direct/dax write). This issue makes generic/471 with xfs always get the following error:

Quoted 9 lines

--------------------------------------------------------
QA output created by 471
pwrite: Resource temporarily unavailable
wrote 8388608/8388608 bytes at offset 0
XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
pwrite: Resource temporarily unavailable
...
--------------------------------------------------------

There have been earlier discussions about this. Snippet from the earlier discussion:

"generic/471 complains because it expects any write done with RWF_NOWAIT to succeed as long as the blocks for the write are already instantiated. This isn't necessarily a correct assumption, as there are other conditions that can cause an RWF_NOWAIT write to fail with -EAGAIN even if the range is already there."

So the test itself probably needs fixing.

Quoted 20 lines
Fixes 1aa91d9c9933 ("xfs: Add async buffered write support")
---
 fs/xfs/xfs_file.c | 3 +++
 1 file changed, 3 insertions(+)
fs/xfs/xfs_file.c +3 -0
index e462d39c840e..561fab3a49c7 100644
@@ -417,6 +417,9 @@ xfs_file_write_checks(
 		spin_unlock(&ip->i_flags_lock);
 
 out:
+	if (IS_DAX(inode) || (iocb->ki_flags & IOCB_DIRECT))
+		return file_modified(file);
+
 	return kiocb_modified(iocb);
 }
 
#2 lore raw

On 2022/11/17 3:00, Stefan Roesch write:

Quoted 16 lines · Xiao Yang
Quoted 5 lines

kiocb_modified() should be used for sync/async buffered write because it will return -EAGAIN when IOCB_NOWAIT is set. Unfortunately, kiocb_modified() is used by the common xfs_file_write_checks() which is called by all types of write(i.e. buffered/direct/dax write). This issue makes generic/471 with xfs always get the following error:

Quoted 9 lines

--------------------------------------------------------
QA output created by 471
pwrite: Resource temporarily unavailable
wrote 8388608/8388608 bytes at offset 0
XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
pwrite: Resource temporarily unavailable
...
--------------------------------------------------------

Quoted 8 lines

There have been earlier discussions about this. Snippet from the earlier discussion:

"generic/471 complains because it expects any write done with RWF_NOWAIT to succeed as long as the blocks for the write are already instantiated. This isn't necessarily a correct assumption, as there are other conditions that can cause an RWF_NOWAIT write to fail with -EAGAIN even if the range is already there."

Hi Stefan,

Thanks for your reply.
Could you give me the URL about the earlier discussions?

kiocb_modified() makes all types of write always get -EAGAIN when RWF_NOWAIT is set. I don't think this patch[1] is correct because it changed the original logic. The original logic only makes buffered write get -EOPNOTSUPP when RWF_NOWAIT is set.
---------------------------------------------
static int file_modified_flags(struct file *file, int flags)
{
...
if (flags & IOCB_NOWAIT)
return -EAGAIN;
...
}
int kiocb_modified(struct kiocb *iocb)

{
        return file_modified_flags(iocb->ki_filp, iocb->ki_flags);
}

---------------------------------------------
PS: kiocb_modified() is used by the common xfs_file_write_checks() which is called by all types of write(i.e. buffered/direct/dax write).

Quoted 2 lines

So the test itself probably needs fixing.

In my opinion, both kernel and the test probably need to be fixed.

[1] 1aa91d9c9933 ("xfs: Add async buffered write support")

Best Regards,
Xiao Yang

#3 lore raw

Hi

Kindly ping. ^_^

Best Regards,
Xiao Yang

-----Original Message-----
From: Yang, Xiao/杨 晓 <[email protected]>
Sent: 2022年11月17日 10:28
To: Stefan Roesch <[email protected]>; [email protected]; [email protected]
Cc: [email protected]; Ruan, Shiyang/阮 世阳 <[email protected]>
Subject: Re: [PATCH] xfs: Call kiocb_modified() for buffered write

On 2022/11/17 3:00, Stefan Roesch write:

Quoted 15 lines · Xiao Yang
Quoted 4 lines

kiocb_modified() should be used for sync/async buffered write because it will return -EAGAIN when IOCB_NOWAIT is set. Unfortunately, kiocb_modified() is used by the common xfs_file_write_checks() which is called by all types of write(i.e. buffered/direct/dax write).

Quoted 9 lines

This issue makes generic/471 with xfs always get the following error:
--------------------------------------------------------
QA output created by 471
pwrite: Resource temporarily unavailable wrote 8388608/8388608 bytes at offset 0 XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
pwrite: Resource temporarily unavailable ...
--------------------------------------------------------

Quoted 8 lines

There have been earlier discussions about this. Snippet from the earlier discussion:

"generic/471 complains because it expects any write done with RWF_NOWAIT to succeed as long as the blocks for the write are already instantiated. This isn't necessarily a correct assumption, as there are other conditions that can cause an RWF_NOWAIT write to fail with -EAGAIN even if the range is already there."

Hi Stefan,

Thanks for your reply.
Could you give me the URL about the earlier discussions?

kiocb_modified() makes all types of write always get -EAGAIN when RWF_NOWAIT is set. I don't think this patch[1] is correct because it changed the original logic. The original logic only makes buffered write get -EOPNOTSUPP when RWF_NOWAIT is set.
---------------------------------------------
static int file_modified_flags(struct file *file, int flags) { ...
if (flags & IOCB_NOWAIT)
return -EAGAIN;
...
}
int kiocb_modified(struct kiocb *iocb)

{
         return file_modified_flags(iocb->ki_filp, iocb->ki_flags); }

---------------------------------------------
PS: kiocb_modified() is used by the common xfs_file_write_checks() which is called by all types of write(i.e. buffered/direct/dax write).

Quoted 2 lines

So the test itself probably needs fixing.

In my opinion, both kernel and the test probably need to be fixed.

[1] 1aa91d9c9933 ("xfs: Add async buffered write support")

Best Regards,
Xiao Yang

#4 lore raw

This has been discussed in https://lore.kernel.org/linux-xfs/[email protected]/

Here is Jens comment:

From: Jens Axboe <[email protected]>
To: "Darrick J. Wong" <[email protected]>, fstests <[email protected]>
Cc: [email protected], [email protected], [email protected],
[email protected], [email protected], [email protected], [email protected], [email protected], [email protected], Stefan Roesch <[email protected]>
Subject: Re: generic/471 regression with async buffered writes?
Date: Thu, 18 Aug 2022 11:00:38 -0600 [thread overview]
Message-ID: <[email protected]> (raw)
In-Reply-To: <Yv5quvRMZXlDXED/@magnolia>

Quoted 24 lines · Darrick J. Wong

Hi everyone,

I noticed the following fstest failure on XFS on 6.0-rc1 that wasn't there in 5.19:

471.out.bad renamed from 471.out +3 -5
@@ -2,12 +2,10 @@
 pwrite: Resource temporarily unavailable
 wrote 8388608/8388608 bytes at offset 0
 XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
-RWF_NOWAIT time is within limits.
+pwrite: Resource temporarily unavailable
+(standard_in) 1: syntax error
+RWF_NOWAIT took  seconds
 00000000:  aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa  ................
 *
-00200000:  bb bb bb bb bb bb bb bb bb bb bb bb bb bb bb bb  ................
-*
-00300000:  aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa aa  ................
-*
 read 8388608/8388608 bytes at offset 0
 XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
Quoted 2 lines

Is this related to the async buffered write changes, or should I keep looking? AFAICT nobody else has mentioned problems with 471...

The test is just broken. It made some odd assumptions on what RWF_NOWAIT means with buffered writes. There's been a discussion on it previously, I'll see if I can find the links. IIRC, the tldr is that the test doesn't really tie RWF_NOWAIT to whether we'll block or not.

Signature · 1 line
Jens Axboe
Quoted 71 lines · On 1/12/23 8:55 PM, [email protected] wrote:

!-------------------------------------------------------------------|
This Message Is From an External Sender

|-------------------------------------------------------------------!

Hi

Kindly ping. ^_^

Best Regards,
Xiao Yang

-----Original Message-----
From: Yang, Xiao/杨 晓 <[email protected]>
Sent: 2022年11月17日 10:28
To: Stefan Roesch <[email protected]>; [email protected]; [email protected]
Cc: [email protected]; Ruan, Shiyang/阮 世阳 <[email protected]>
Subject: Re: [PATCH] xfs: Call kiocb_modified() for buffered write

On 2022/11/17 3:00, Stefan Roesch write:

Quoted 15 lines · Xiao Yang
Quoted 4 lines

kiocb_modified() should be used for sync/async buffered write because it will return -EAGAIN when IOCB_NOWAIT is set. Unfortunately, kiocb_modified() is used by the common xfs_file_write_checks() which is called by all types of write(i.e. buffered/direct/dax write).

Quoted 9 lines

This issue makes generic/471 with xfs always get the following error:
--------------------------------------------------------
QA output created by 471
pwrite: Resource temporarily unavailable wrote 8388608/8388608 bytes at offset 0 XXX Bytes, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
pwrite: Resource temporarily unavailable ...
--------------------------------------------------------

Quoted 8 lines

There have been earlier discussions about this. Snippet from the earlier discussion:

"generic/471 complains because it expects any write done with RWF_NOWAIT to succeed as long as the blocks for the write are already instantiated. This isn't necessarily a correct assumption, as there are other conditions that can cause an RWF_NOWAIT write to fail with -EAGAIN even if the range is already there."

Hi Stefan,

Thanks for your reply.
Could you give me the URL about the earlier discussions?

kiocb_modified() makes all types of write always get -EAGAIN when RWF_NOWAIT is set. I don't think this patch[1] is correct because it changed the original logic. The original logic only makes buffered write get -EOPNOTSUPP when RWF_NOWAIT is set.
---------------------------------------------
static int file_modified_flags(struct file *file, int flags) { ...
if (flags & IOCB_NOWAIT)
return -EAGAIN;
...
}
int kiocb_modified(struct kiocb *iocb)

{
         return file_modified_flags(iocb->ki_filp, iocb->ki_flags); }

---------------------------------------------
PS: kiocb_modified() is used by the common xfs_file_write_checks() which is called by all types of write(i.e. buffered/direct/dax write).

Quoted 2 lines

So the test itself probably needs fixing.

In my opinion, both kernel and the test probably need to be fixed.

[1] 1aa91d9c9933 ("xfs: Add async buffered write support")

Best Regards,
Xiao Yang