* Re: [PATCH 1/2] f2fs: improve readahead for POSIX_FADV_WILLNEED
[not found] <20251121014202.1969909-1-jaegeuk@kernel.org>
@ 2025-11-21 6:49 ` Christoph Hellwig
2025-11-21 18:05 ` Jaegeuk Kim
0 siblings, 1 reply; 2+ messages in thread
From: Christoph Hellwig @ 2025-11-21 6:49 UTC (permalink / raw)
To: Jaegeuk Kim
Cc: linux-kernel, linux-f2fs-devel, Matthew Wilcox, Andrew Morton,
linux-fsdevel, linux-mm
On Fri, Nov 21, 2025 at 01:42:01AM +0000, Jaegeuk Kim wrote:
> This patch boosts readahead for POSIX_FADV_WILLNEED.
How? That's not a good changelog.
Also open coding the read-ahead logic is not a good idea. The only
f2fs-specific bits are the compression check, and the extent precaching,
but you surely should be able to share a read-ahead helper with common
code instead of duplicating the logic.
>
> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
> ---
> fs/f2fs/data.c | 61 ++++++++++++++++++++++++++++++++++++++++++++++++++
> fs/f2fs/f2fs.h | 1 +
> fs/f2fs/file.c | 9 +++++---
> 3 files changed, 68 insertions(+), 3 deletions(-)
>
> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> index a0433c8a4d84..d95974d79fb3 100644
> --- a/fs/f2fs/data.c
> +++ b/fs/f2fs/data.c
> @@ -2710,6 +2710,67 @@ static void f2fs_readahead(struct readahead_control *rac)
> f2fs_mpage_readpages(inode, rac, NULL);
> }
>
> +int f2fs_readahead_pages(struct file *file, loff_t offset, loff_t len)
> +{
> + struct inode *inode = file_inode(file);
> + struct address_space *mapping = file->f_mapping;
> + pgoff_t start_index = offset >> PAGE_SHIFT;
> + loff_t endbyte = offset + len;
> + pgoff_t end_index;
> + unsigned long nrpages;
> + unsigned long ra_pages = (16 * 1024 * 1024) / PAGE_SIZE;
> + DEFINE_READAHEAD(ractl, NULL, &file->f_ra, mapping, start_index);
> +
> + if (!S_ISREG(inode->i_mode))
> + return -EOPNOTSUPP;
> +
> + /* Should be read only. */
> + if (!(file->f_mode & FMODE_READ))
> + return -EBADF;
> +
> + /* Do not support compressed file for large folio. */
> + if (f2fs_compressed_file(inode))
> + return -EINVAL;
> +
> + if (!mapping || len < 0)
> + return -EINVAL;
> +
> + if (unlikely(!mapping->a_ops->read_folio && !mapping->a_ops->readahead))
> + return -EINVAL;
> +
> + /* Load extent cache at the first readahead. */
> + f2fs_precache_extents(inode);
> +
> + /*
> + * Careful about overflows. Len == 0 means "as much as possible". Use
> + * unsigned math because signed overflows are undefined and UBSan
> + * complains.
> + */
> + if (!len || endbyte > i_size_read(inode) || endbyte < len)
> + endbyte = i_size_read(inode) - 1;
> + else
> + endbyte--; /* inclusive */
> +
> + /* First and last PARTIAL page! */
> + end_index = endbyte >> PAGE_SHIFT;
> +
> + if (start_index > end_index)
> + return 0;
> +
> + nrpages = end_index - start_index + 1;
> +
> + while (nrpages) {
> + unsigned long this_chunk = min(nrpages, ra_pages);
> +
> + ractl.ra->ra_pages = this_chunk;
> +
> + page_cache_sync_ra(&ractl, this_chunk << 1);
> +
> + nrpages -= this_chunk;
> + }
> + return 0;
> +}
> +
> int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
> {
> struct inode *inode = fio_inode(fio);
> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> index 3340db04a7c2..934287cc5624 100644
> --- a/fs/f2fs/f2fs.h
> +++ b/fs/f2fs/f2fs.h
> @@ -4047,6 +4047,7 @@ int f2fs_init_bio_entry_cache(void);
> void f2fs_destroy_bio_entry_cache(void);
> void f2fs_submit_read_bio(struct f2fs_sb_info *sbi, struct bio *bio,
> enum page_type type);
> +int f2fs_readahead_pages(struct file *file, loff_t offset, loff_t len);
> int f2fs_init_write_merge_io(struct f2fs_sb_info *sbi);
> void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type);
> void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
> diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
> index d7047ca6b98d..b6f71efd6d2a 100644
> --- a/fs/f2fs/file.c
> +++ b/fs/f2fs/file.c
> @@ -5305,9 +5305,12 @@ static int f2fs_file_fadvise(struct file *filp, loff_t offset, loff_t len,
> filp->f_mode &= ~FMODE_RANDOM;
> spin_unlock(&filp->f_lock);
> return 0;
> - } else if (advice == POSIX_FADV_WILLNEED && offset == 0) {
> - /* Load extent cache at the first readahead. */
> - f2fs_precache_extents(inode);
> + } else if (advice == POSIX_FADV_WILLNEED) {
> + if (offset == 0 && len == -1) {
> + f2fs_precache_extents(inode);
> + return 0;
> + }
> + return f2fs_readahead_pages(filp, offset, len);
> }
>
> err = generic_fadvise(filp, offset, len, advice);
> --
> 2.52.0.487.g5c8c507ade-goog
>
>
---end quoted text---
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH 1/2] f2fs: improve readahead for POSIX_FADV_WILLNEED
2025-11-21 6:49 ` [PATCH 1/2] f2fs: improve readahead for POSIX_FADV_WILLNEED Christoph Hellwig
@ 2025-11-21 18:05 ` Jaegeuk Kim
0 siblings, 0 replies; 2+ messages in thread
From: Jaegeuk Kim @ 2025-11-21 18:05 UTC (permalink / raw)
To: Christoph Hellwig
Cc: linux-kernel, linux-f2fs-devel, Matthew Wilcox, Andrew Morton,
linux-fsdevel, linux-mm
On 11/20, Christoph Hellwig wrote:
> On Fri, Nov 21, 2025 at 01:42:01AM +0000, Jaegeuk Kim wrote:
> > This patch boosts readahead for POSIX_FADV_WILLNEED.
>
> How? That's not a good changelog.
>
> Also open coding the read-ahead logic is not a good idea. The only
> f2fs-specific bits are the compression check, and the extent precaching,
> but you surely should be able to share a read-ahead helper with common
> code instead of duplicating the logic.
Ok, let me try to write up and post a generic version of the changes.
>
> >
> > Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
> > ---
> > fs/f2fs/data.c | 61 ++++++++++++++++++++++++++++++++++++++++++++++++++
> > fs/f2fs/f2fs.h | 1 +
> > fs/f2fs/file.c | 9 +++++---
> > 3 files changed, 68 insertions(+), 3 deletions(-)
> >
> > diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
> > index a0433c8a4d84..d95974d79fb3 100644
> > --- a/fs/f2fs/data.c
> > +++ b/fs/f2fs/data.c
> > @@ -2710,6 +2710,67 @@ static void f2fs_readahead(struct readahead_control *rac)
> > f2fs_mpage_readpages(inode, rac, NULL);
> > }
> >
> > +int f2fs_readahead_pages(struct file *file, loff_t offset, loff_t len)
> > +{
> > + struct inode *inode = file_inode(file);
> > + struct address_space *mapping = file->f_mapping;
> > + pgoff_t start_index = offset >> PAGE_SHIFT;
> > + loff_t endbyte = offset + len;
> > + pgoff_t end_index;
> > + unsigned long nrpages;
> > + unsigned long ra_pages = (16 * 1024 * 1024) / PAGE_SIZE;
> > + DEFINE_READAHEAD(ractl, NULL, &file->f_ra, mapping, start_index);
> > +
> > + if (!S_ISREG(inode->i_mode))
> > + return -EOPNOTSUPP;
> > +
> > + /* Should be read only. */
> > + if (!(file->f_mode & FMODE_READ))
> > + return -EBADF;
> > +
> > + /* Do not support compressed file for large folio. */
> > + if (f2fs_compressed_file(inode))
> > + return -EINVAL;
> > +
> > + if (!mapping || len < 0)
> > + return -EINVAL;
> > +
> > + if (unlikely(!mapping->a_ops->read_folio && !mapping->a_ops->readahead))
> > + return -EINVAL;
> > +
> > + /* Load extent cache at the first readahead. */
> > + f2fs_precache_extents(inode);
> > +
> > + /*
> > + * Careful about overflows. Len == 0 means "as much as possible". Use
> > + * unsigned math because signed overflows are undefined and UBSan
> > + * complains.
> > + */
> > + if (!len || endbyte > i_size_read(inode) || endbyte < len)
> > + endbyte = i_size_read(inode) - 1;
> > + else
> > + endbyte--; /* inclusive */
> > +
> > + /* First and last PARTIAL page! */
> > + end_index = endbyte >> PAGE_SHIFT;
> > +
> > + if (start_index > end_index)
> > + return 0;
> > +
> > + nrpages = end_index - start_index + 1;
> > +
> > + while (nrpages) {
> > + unsigned long this_chunk = min(nrpages, ra_pages);
> > +
> > + ractl.ra->ra_pages = this_chunk;
> > +
> > + page_cache_sync_ra(&ractl, this_chunk << 1);
> > +
> > + nrpages -= this_chunk;
> > + }
> > + return 0;
> > +}
> > +
> > int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
> > {
> > struct inode *inode = fio_inode(fio);
> > diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> > index 3340db04a7c2..934287cc5624 100644
> > --- a/fs/f2fs/f2fs.h
> > +++ b/fs/f2fs/f2fs.h
> > @@ -4047,6 +4047,7 @@ int f2fs_init_bio_entry_cache(void);
> > void f2fs_destroy_bio_entry_cache(void);
> > void f2fs_submit_read_bio(struct f2fs_sb_info *sbi, struct bio *bio,
> > enum page_type type);
> > +int f2fs_readahead_pages(struct file *file, loff_t offset, loff_t len);
> > int f2fs_init_write_merge_io(struct f2fs_sb_info *sbi);
> > void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type);
> > void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
> > diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
> > index d7047ca6b98d..b6f71efd6d2a 100644
> > --- a/fs/f2fs/file.c
> > +++ b/fs/f2fs/file.c
> > @@ -5305,9 +5305,12 @@ static int f2fs_file_fadvise(struct file *filp, loff_t offset, loff_t len,
> > filp->f_mode &= ~FMODE_RANDOM;
> > spin_unlock(&filp->f_lock);
> > return 0;
> > - } else if (advice == POSIX_FADV_WILLNEED && offset == 0) {
> > - /* Load extent cache at the first readahead. */
> > - f2fs_precache_extents(inode);
> > + } else if (advice == POSIX_FADV_WILLNEED) {
> > + if (offset == 0 && len == -1) {
> > + f2fs_precache_extents(inode);
> > + return 0;
> > + }
> > + return f2fs_readahead_pages(filp, offset, len);
> > }
> >
> > err = generic_fadvise(filp, offset, len, advice);
> > --
> > 2.52.0.487.g5c8c507ade-goog
> >
> >
> ---end quoted text---
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2025-11-21 18:05 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
[not found] <20251121014202.1969909-1-jaegeuk@kernel.org>
2025-11-21 6:49 ` [PATCH 1/2] f2fs: improve readahead for POSIX_FADV_WILLNEED Christoph Hellwig
2025-11-21 18:05 ` Jaegeuk Kim
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox