Some UFS storage gives slower performance on FUA than write+cache_flush.
Let's give a way to manage it.
Signed-off-by: Jaegeuk Kim <[email protected]>
---
Documentation/ABI/testing/sysfs-fs-f2fs | 7 +++++++
fs/f2fs/data.c | 2 ++
fs/f2fs/f2fs.h | 1 +
fs/f2fs/sysfs.c | 2 ++
4 files changed, 12 insertions(+)
diff --git a/Documentation/ABI/testing/sysfs-fs-f2fs b/Documentation/ABI/testing/sysfs-fs-f2fs
index 9b583dd0298b..cd96b09d7182 100644
--- a/Documentation/ABI/testing/sysfs-fs-f2fs
+++ b/Documentation/ABI/testing/sysfs-fs-f2fs
@@ -434,6 +434,7 @@ Date: April 2020
Contact: "Daeho Jeong" <[email protected]>
Description: Give a way to change iostat_period time. 3secs by default.
The new iostat trace gives stats gap given the period.
+
What: /sys/fs/f2fs/<disk>/max_io_bytes
Date: December 2020
Contact: "Jaegeuk Kim" <[email protected]>
@@ -442,6 +443,12 @@ Description: This gives a control to limit the bio size in f2fs.
whereas, if it has a certain bytes value, f2fs won't submit a
bio larger than that size.
+What: /sys/fs/f2fs/<disk>/no_fua_dio
+Date: May 2022
+Contact: "Jaegeuk Kim" <[email protected]>
+Description: This gives a signal to iomap, which should not use FUA for
+ direct IOs. Default: 0.
+
What: /sys/fs/f2fs/<disk>/stat/sb_status
Date: December 2020
Contact: "Chao Yu" <[email protected]>
diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index f5f2b7233982..23486486eab2 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -4153,6 +4153,8 @@ static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
if ((inode->i_state & I_DIRTY_DATASYNC) ||
offset + length > i_size_read(inode))
iomap->flags |= IOMAP_F_DIRTY;
+ if (F2FS_I_SB(inode)->no_fua_dio)
+ iomap->flags |= IOMAP_F_DIRTY;
return 0;
}
diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index e10838879538..c2400ea0080b 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -1671,6 +1671,7 @@ struct f2fs_sb_info {
int dir_level; /* directory level */
int readdir_ra; /* readahead inode in readdir */
u64 max_io_bytes; /* max io bytes to merge IOs */
+ int no_fua_dio; /* avoid FUA in DIO */
block_t user_block_count; /* # of user blocks */
block_t total_valid_block_count; /* # of valid blocks */
diff --git a/fs/f2fs/sysfs.c b/fs/f2fs/sysfs.c
index 4c50aedd5144..24d628ca92cc 100644
--- a/fs/f2fs/sysfs.c
+++ b/fs/f2fs/sysfs.c
@@ -771,6 +771,7 @@ F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, iostat_period_ms, iostat_period_ms);
#endif
F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, readdir_ra, readdir_ra);
F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, max_io_bytes, max_io_bytes);
+F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, no_fua_dio, no_fua_dio);
F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, gc_pin_file_thresh, gc_pin_file_threshold);
F2FS_RW_ATTR(F2FS_SBI, f2fs_super_block, extension_list, extension_list);
#ifdef CONFIG_F2FS_FAULT_INJECTION
@@ -890,6 +891,7 @@ static struct attribute *f2fs_attrs[] = {
#endif
ATTR_LIST(readdir_ra),
ATTR_LIST(max_io_bytes),
+ ATTR_LIST(no_fua_dio),
ATTR_LIST(gc_pin_file_thresh),
ATTR_LIST(extension_list),
#ifdef CONFIG_F2FS_FAULT_INJECTION
--
2.36.1.124.g0e6072fb45-goog
On Fri, May 27, 2022 at 09:33:55PM +0000, Eric Biggers wrote:
> [+Cc linux-block for FUA, and linux-xfs for iomap]
linux-fsdevel should really be used for iomap stuff...
>
> On Fri, May 27, 2022 at 01:59:55PM -0700, Jaegeuk Kim wrote:
> > Some UFS storage gives slower performance on FUA than write+cache_flush.
> > Let's give a way to manage it.
> >
> > Signed-off-by: Jaegeuk Kim <[email protected]>
>
> Should the driver even be saying that it has FUA support in this case? If the
> driver didn't claim FUA support, that would also solve this problem.
Agreed, this is a hardware problem that need to addressed with a
driver quirk to stop it advertising FUA support. The high level
fs/iomap code should always issue FUA writes where possible and
the lower layers tell the block layer whether to issue the FUA as
a FUA or write+cache flush pair.
And, quite frankly, exposing this sort of "hardware needs help" knob
as a sysfs variable is exactly the sort of thing we should never do.
Users have no idea how to tune stuff like this correctly (even if
they knew it existed!), yet we know exactly what hardware has this
problem and the kernel already has mechanisms that would allow it to
just Do The Right Thing. IOWs, we can fix this without the user even
having to know that they have garbage hardware that needs special
help....
Cheers,
Dave.
--
Dave Chinner
[email protected]
On 05/28, Dave Chinner wrote:
> On Fri, May 27, 2022 at 09:33:55PM +0000, Eric Biggers wrote:
> > [+Cc linux-block for FUA, and linux-xfs for iomap]
>
> linux-fsdevel should really be used for iomap stuff...
>
> >
> > On Fri, May 27, 2022 at 01:59:55PM -0700, Jaegeuk Kim wrote:
> > > Some UFS storage gives slower performance on FUA than write+cache_flush.
> > > Let's give a way to manage it.
> > >
> > > Signed-off-by: Jaegeuk Kim <[email protected]>
> >
> > Should the driver even be saying that it has FUA support in this case? If the
> > driver didn't claim FUA support, that would also solve this problem.
>
> Agreed, this is a hardware problem that need to addressed with a
> driver quirk to stop it advertising FUA support. The high level
> fs/iomap code should always issue FUA writes where possible and
> the lower layers tell the block layer whether to issue the FUA as
> a FUA or write+cache flush pair.
I was thinking to turn off FUA in driver side quickly tho, one concern
was the bandwidth vs. latency. What if the device can support FUA having
short latency while giving low bandwidth? In that case, we still have
a room to utilize FUA for small-sized writes such as filesystem metadata
writes, but avoid DIO w/ FUA for sequential write stream. Is this just
HW problem? Or, does SW need to use FUA more efficiently?
>
> And, quite frankly, exposing this sort of "hardware needs help" knob
> as a sysfs variable is exactly the sort of thing we should never do.
>
> Users have no idea how to tune stuff like this correctly (even if
> they knew it existed!), yet we know exactly what hardware has this
> problem and the kernel already has mechanisms that would allow it to
> just Do The Right Thing. IOWs, we can fix this without the user even
> having to know that they have garbage hardware that needs special
> help....
>
> Cheers,
>
> Dave.
> --
> Dave Chinner
> [email protected]
On Fri, May 27, 2022 at 05:26:32PM -0700, Jaegeuk Kim wrote:
> On 05/28, Dave Chinner wrote:
> > On Fri, May 27, 2022 at 09:33:55PM +0000, Eric Biggers wrote:
> > > [+Cc linux-block for FUA, and linux-xfs for iomap]
> >
> > linux-fsdevel should really be used for iomap stuff...
> >
> > >
> > > On Fri, May 27, 2022 at 01:59:55PM -0700, Jaegeuk Kim wrote:
> > > > Some UFS storage gives slower performance on FUA than write+cache_flush.
> > > > Let's give a way to manage it.
> > > >
> > > > Signed-off-by: Jaegeuk Kim <[email protected]>
> > >
> > > Should the driver even be saying that it has FUA support in this case? If the
> > > driver didn't claim FUA support, that would also solve this problem.
> >
> > Agreed, this is a hardware problem that need to addressed with a
> > driver quirk to stop it advertising FUA support. The high level
> > fs/iomap code should always issue FUA writes where possible and
> > the lower layers tell the block layer whether to issue the FUA as
> > a FUA or write+cache flush pair.
>
> I was thinking to turn off FUA in driver side quickly tho, one concern
> was the bandwidth vs. latency. What if the device can support FUA having
> short latency while giving low bandwidth?
Seriously, how is a user supposed to know this sort of thing about
the hardware they are using? They don't, and to expect them to not
only know about the existing of a weird sysfs knob, let alone how it
applies to their hardware and their workload is totally
unreasonable.
If the hardware has non-deterministic FUA write performance, or
requires very careful switch over between cache flushes and FUA to
get the most out of the hardware, then that's not something we can
tune or optimise for - that's just broken hardware and the drive
should quirk the brokeness away so nobody has to care about it. Tell
the hardware manufacturer to fix their hardware, don't try to hack
around it in software and then expect the user to know how to tune
for that broken hardware.
> In that case, we still have
> a room to utilize FUA for small-sized writes such as filesystem metadata
> writes, but avoid DIO w/ FUA for sequential write stream.
Strawman.
We don't use FUA for normal DIO writes - they only get used for
O_DSYNC writes, in which case we either use FUA if the device
supports it, or we do a normal write followed by a cache flush.
If there are metadata updates that the O_DSYNC needs to also flush,
we don't use FUA by let the fileystem issue a cache flush in the
most optimal possible after the write completes.
Either way, using O_DSYNC DIO writes for streaming, sequential data
is a really poor choice for an application to make. Normal DIO
writes followed by fdatasync() to flush the metadata and caches once
will be much faster and far more efficient than a metadata and cache
flush after every single data write, FUA or not.
-Dave.
--
Dave Chinner
[email protected]
[+Cc linux-block for FUA, and linux-xfs for iomap]
On Fri, May 27, 2022 at 01:59:55PM -0700, Jaegeuk Kim wrote:
> Some UFS storage gives slower performance on FUA than write+cache_flush.
> Let's give a way to manage it.
>
> Signed-off-by: Jaegeuk Kim <[email protected]>
Should the driver even be saying that it has FUA support in this case? If the
driver didn't claim FUA support, that would also solve this problem.
> ---
> Documentation/ABI/testing/sysfs-fs-f2fs | 7 +++++++
> fs/f2fs/data.c | 2 ++
> fs/f2fs/f2fs.h | 1 +
> fs/f2fs/sysfs.c | 2 ++
> 4 files changed, 12 insertions(+)
>
> diff --git a/Documentation/ABI/testing/sysfs-fs-f2fs b/Documentation/ABI/testing/sysfs-fs-f2fs
> index 9b583dd0298b..cd96b09d7182 100644
> --- a/Documentation/ABI/testing/sysfs-fs-f2fs
> +++ b/Documentation/ABI/testing/sysfs-fs-f2fs
> @@ -434,6 +434,7 @@ Date: April 2020
> Contact: "Daeho Jeong" <[email protected]>
> Description: Give a way to change iostat_period time. 3secs by default.
> The new iostat trace gives stats gap given the period.
> +
> What: /sys/fs/f2fs/<disk>/max_io_bytes
> Date: December 2020
> Contact: "Jaegeuk Kim" <[email protected]>
> @@ -442,6 +443,12 @@ Description: This gives a control to limit the bio size in f2fs.
> whereas, if it has a certain bytes value, f2fs won't submit a
> bio larger than that size.
>
> +What: /sys/fs/f2fs/<disk>/no_fua_dio
> +Date: May 2022
> +Contact: "Jaegeuk Kim" <[email protected]>
> +Description: This gives a signal to iomap, which should not use FUA for
> + direct IOs. Default: 0.
iomap is an implementation detail, so it shouldn't be mentioned in UAPI
documentation. UAPI documentation should describe user-visible behavior only.
> +
> What: /sys/fs/f2fs/<disk>/stat/sb_status
> Date: December 2020
> Contact: "Chao Yu" <[email protected]>
> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> index f5f2b7233982..23486486eab2 100644
> --- a/fs/f2fs/data.c
> +++ b/fs/f2fs/data.c
> @@ -4153,6 +4153,8 @@ static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
> if ((inode->i_state & I_DIRTY_DATASYNC) ||
> offset + length > i_size_read(inode))
> iomap->flags |= IOMAP_F_DIRTY;
> + if (F2FS_I_SB(inode)->no_fua_dio)
> + iomap->flags |= IOMAP_F_DIRTY;
This is overloading the IOMAP_F_DIRTY flag to mean something other than dirty.
Perhaps this flag needs to be renamed, or a new flag should be added?
> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> index e10838879538..c2400ea0080b 100644
> --- a/fs/f2fs/f2fs.h
> +++ b/fs/f2fs/f2fs.h
> @@ -1671,6 +1671,7 @@ struct f2fs_sb_info {
> int dir_level; /* directory level */
> int readdir_ra; /* readahead inode in readdir */
> u64 max_io_bytes; /* max io bytes to merge IOs */
> + int no_fua_dio; /* avoid FUA in DIO */
Make this a bool?
> diff --git a/fs/f2fs/sysfs.c b/fs/f2fs/sysfs.c
> index 4c50aedd5144..24d628ca92cc 100644
> --- a/fs/f2fs/sysfs.c
> +++ b/fs/f2fs/sysfs.c
> @@ -771,6 +771,7 @@ F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, iostat_period_ms, iostat_period_ms);
> #endif
> F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, readdir_ra, readdir_ra);
> F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, max_io_bytes, max_io_bytes);
> +F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, no_fua_dio, no_fua_dio);
> F2FS_RW_ATTR(F2FS_SBI, f2fs_sb_info, gc_pin_file_thresh, gc_pin_file_threshold);
> F2FS_RW_ATTR(F2FS_SBI, f2fs_super_block, extension_list, extension_list);
> #ifdef CONFIG_F2FS_FAULT_INJECTION
> @@ -890,6 +891,7 @@ static struct attribute *f2fs_attrs[] = {
> #endif
> ATTR_LIST(readdir_ra),
> ATTR_LIST(max_io_bytes),
> + ATTR_LIST(no_fua_dio),
Where is it validated that only valid values (0 or 1) can be written to this
file?
- Eric