2023-06-12 05:40:40

by Christoph Hellwig

[permalink] [raw]
Subject: [PATCH] ext4: set FMODE_CAN_ODIRECT instead of a dummy direct_IO method

Since commit a2ad63daa88b ("VFS: add FMODE_CAN_ODIRECT file flag") file
systems can just set the FMODE_CAN_ODIRECT flag at open time instead of
wiring up a dummy direct_IO method to indicate support for direct I/O.

Do that for ext4 so that noop_direct_IO can eventually be removed.

Signed-off-by: Christoph Hellwig <[email protected]>
---
fs/ext4/file.c | 2 +-
fs/ext4/inode.c | 4 ----
2 files changed, 1 insertion(+), 5 deletions(-)

diff --git a/fs/ext4/file.c b/fs/ext4/file.c
index d101b3b0c7dad8..a1d8ffbf571274 100644
--- a/fs/ext4/file.c
+++ b/fs/ext4/file.c
@@ -900,7 +900,7 @@ static int ext4_file_open(struct inode *inode, struct file *filp)
}

filp->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC |
- FMODE_DIO_PARALLEL_WRITE;
+ FMODE_DIO_PARALLEL_WRITE | FMODE_CAN_ODIRECT;
return dquot_file_open(inode, filp);
}

diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 02de439bf1f04e..b9c1cfa1864779 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -3539,7 +3539,6 @@ static const struct address_space_operations ext4_aops = {
.bmap = ext4_bmap,
.invalidate_folio = ext4_invalidate_folio,
.release_folio = ext4_release_folio,
- .direct_IO = noop_direct_IO,
.migrate_folio = buffer_migrate_folio,
.is_partially_uptodate = block_is_partially_uptodate,
.error_remove_page = generic_error_remove_page,
@@ -3556,7 +3555,6 @@ static const struct address_space_operations ext4_journalled_aops = {
.bmap = ext4_bmap,
.invalidate_folio = ext4_journalled_invalidate_folio,
.release_folio = ext4_release_folio,
- .direct_IO = noop_direct_IO,
.migrate_folio = buffer_migrate_folio_norefs,
.is_partially_uptodate = block_is_partially_uptodate,
.error_remove_page = generic_error_remove_page,
@@ -3573,7 +3571,6 @@ static const struct address_space_operations ext4_da_aops = {
.bmap = ext4_bmap,
.invalidate_folio = ext4_invalidate_folio,
.release_folio = ext4_release_folio,
- .direct_IO = noop_direct_IO,
.migrate_folio = buffer_migrate_folio,
.is_partially_uptodate = block_is_partially_uptodate,
.error_remove_page = generic_error_remove_page,
@@ -3582,7 +3579,6 @@ static const struct address_space_operations ext4_da_aops = {

static const struct address_space_operations ext4_dax_aops = {
.writepages = ext4_dax_writepages,
- .direct_IO = noop_direct_IO,
.dirty_folio = noop_dirty_folio,
.bmap = ext4_bmap,
.swap_activate = ext4_iomap_swap_activate,
--
2.39.2



2023-06-13 05:34:01

by Ritesh Harjani

[permalink] [raw]
Subject: Re: [PATCH] ext4: set FMODE_CAN_ODIRECT instead of a dummy direct_IO method

Christoph Hellwig <[email protected]> writes:

> Since commit a2ad63daa88b ("VFS: add FMODE_CAN_ODIRECT file flag") file
> systems can just set the FMODE_CAN_ODIRECT flag at open time instead of
> wiring up a dummy direct_IO method to indicate support for direct I/O.
>
> Do that for ext4 so that noop_direct_IO can eventually be removed.

Looks straight forward change to me.

However I got some query...

> diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
> index 02de439bf1f04e..b9c1cfa1864779 100644
> --- a/fs/ext4/inode.c
> +++ b/fs/ext4/inode.c
> @@ -3573,7 +3571,6 @@ static const struct address_space_operations ext4_da_aops = {
> .bmap = ext4_bmap,
> .invalidate_folio = ext4_invalidate_folio,
> .release_folio = ext4_release_folio,
> - .direct_IO = noop_direct_IO,
> .migrate_folio = buffer_migrate_folio,
> .is_partially_uptodate = block_is_partially_uptodate,
> .error_remove_page = generic_error_remove_page,
> @@ -3582,7 +3579,6 @@ static const struct address_space_operations ext4_da_aops = {
>
> static const struct address_space_operations ext4_dax_aops = {
> .writepages = ext4_dax_writepages,
> - .direct_IO = noop_direct_IO,
> .dirty_folio = noop_dirty_folio,
> .bmap = ext4_bmap,
> .swap_activate = ext4_iomap_swap_activate,

why do we require .direct_IO function op for any of the dax_aops?
IIUC, any inode if it supports DAX i.e. IS_DAX(inode), then it takes the
separate path in file read/write iter path.

so it should never do ->direct_IO on an inode which supports DAX right?

Maybe I am missing something and I will have to check when does
->direct_IO gets called for DAX...
... Or is it due to CONFIG_FS_DAX?

e.g.

ext4_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
{
struct inode *inode = file_inode(iocb->ki_filp);

if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
return -EIO;

#ifdef CONFIG_FS_DAX
if (IS_DAX(inode))
return ext4_dax_write_iter(iocb, from);
#endif
if (iocb->ki_flags & IOCB_DIRECT)
return ext4_dio_write_iter(iocb, from);
else
return ext4_buffered_write_iter(iocb, from);
}

-ritesh

2023-06-13 05:57:08

by Christoph Hellwig

[permalink] [raw]
Subject: Re: [PATCH] ext4: set FMODE_CAN_ODIRECT instead of a dummy direct_IO method

On Tue, Jun 13, 2023 at 11:00:19AM +0530, Ritesh Harjani wrote:
> why do we require .direct_IO function op for any of the dax_aops?
> IIUC, any inode if it supports DAX i.e. IS_DAX(inode), then it takes the
> separate path in file read/write iter path.
>
> so it should never do ->direct_IO on an inode which supports DAX right?

do_dentry_open rejects opens with O_DIRECT if FMODE_CAN_ODIRECT is
not set. So we either needs to set that manually or because there is a
->direct_IO if we want to keep supporting O_DIRECT opens for DAX
files, which we've traditionally supported.

2023-06-13 08:39:16

by Ritesh Harjani

[permalink] [raw]
Subject: Re: [PATCH] ext4: set FMODE_CAN_ODIRECT instead of a dummy direct_IO method

Christoph Hellwig <[email protected]> writes:

> On Tue, Jun 13, 2023 at 11:00:19AM +0530, Ritesh Harjani wrote:
>> why do we require .direct_IO function op for any of the dax_aops?
>> IIUC, any inode if it supports DAX i.e. IS_DAX(inode), then it takes the
>> separate path in file read/write iter path.
>>
>> so it should never do ->direct_IO on an inode which supports DAX right?
>
> do_dentry_open rejects opens with O_DIRECT if FMODE_CAN_ODIRECT is
> not set. So we either needs to set that manually or because there is a
> ->direct_IO if we want to keep supporting O_DIRECT opens for DAX
> files, which we've traditionally supported.

Ok, so for DAX it was mainly to support file opens with O_DIRECT semantics.

Thanks
-ritesh