[PATCH 1/2] f2fs: improve readahead for POSIX_FADV_WILLNEED

[PATCH 1/2] f2fs: improve readahead for POSIX_FADV_WILLNEED

[Jaegeuk Kim]

This patch boosts readahead for POSIX_FADV_WILLNEED.

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

[PATCH 2/2] f2fs: add a tracepoint for readahead pages

[Jaegeuk Kim]

This patch adds a tracepoint to track the f2fs_readahead_pages().

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
---
 fs/f2fs/data.c              |  3 +++
 include/trace/events/f2fs.h | 43 +++++++++++++++++++++++++++++++++++++
 2 files changed, 46 insertions(+)

diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c
index d95974d79fb3..c80d7960b652 100644
--- a/fs/f2fs/data.c
+++ b/fs/f2fs/data.c
@@ -2757,6 +2757,8 @@ int f2fs_readahead_pages(struct file *file, loff_t offset, loff_t len)
 	if (start_index > end_index)
 		return 0;
 
+	trace_f2fs_readahead_pages_start(inode, start_index, end_index);
+
 	nrpages = end_index - start_index + 1;
 
 	while (nrpages) {
@@ -2768,6 +2770,7 @@ int f2fs_readahead_pages(struct file *file, loff_t offset, loff_t len)
 
 		nrpages -= this_chunk;
 	}
+	trace_f2fs_readahead_pages_end(inode, start_index, end_index);
 	return 0;
 }
 
diff --git a/include/trace/events/f2fs.h b/include/trace/events/f2fs.h
index d406b047c50b..bb98a9655ec9 100644
--- a/include/trace/events/f2fs.h
+++ b/include/trace/events/f2fs.h
@@ -618,6 +618,49 @@ TRACE_EVENT(f2fs_fadvise,
 		__entry->advice)
 );
 
+DECLARE_EVENT_CLASS(f2fs__readahead_pages,
+
+	TP_PROTO(struct inode *inode, pgoff_t start, pgoff_t end),
+
+	TP_ARGS(inode, start, end),
+
+	TP_STRUCT__entry(
+		__field(dev_t,	dev)
+		__field(ino_t,	ino)
+		__field(loff_t, size)
+		__field(pgoff_t, start)
+		__field(pgoff_t, end)
+	),
+
+	TP_fast_assign(
+		__entry->dev	= inode->i_sb->s_dev;
+		__entry->ino	= inode->i_ino;
+		__entry->size	= inode->i_size;
+		__entry->start = start;
+		__entry->end	= end;
+	),
+
+	TP_printk("dev = (%d,%d), ino = %lu, i_size = %lld start: %lu, end: %lu",
+		show_dev_ino(__entry),
+		(unsigned long long)__entry->size,
+		__entry->start,
+		__entry->end)
+);
+
+DEFINE_EVENT(f2fs__readahead_pages, f2fs_readahead_pages_start,
+
+	TP_PROTO(struct inode *inode, pgoff_t start, pgoff_t end),
+
+	TP_ARGS(inode, start, end)
+);
+
+DEFINE_EVENT(f2fs__readahead_pages, f2fs_readahead_pages_end,
+
+	TP_PROTO(struct inode *inode, pgoff_t start, pgoff_t end),
+
+	TP_ARGS(inode, start, end)
+);
+
 TRACE_EVENT(f2fs_map_blocks,
 	TP_PROTO(struct inode *inode, struct f2fs_map_blocks *map, int flag,
 		 int ret),
-- 
2.52.0.487.g5c8c507ade-goog

Re: [PATCH 1/2] f2fs: improve readahead for POSIX_FADV_WILLNEED

[Christoph Hellwig]

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---

Re: [PATCH 1/2] f2fs: improve readahead for POSIX_FADV_WILLNEED

[Jaegeuk Kim]

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---

Re: [PATCH 1/2] f2fs: improve readahead for POSIX_FADV_WILLNEED

[Christoph Hellwig]

On Fri, Nov 21, 2025 at 06:05:01PM +0000, Jaegeuk Kim wrote:
> 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.

Did this go anywhere?

Re: [PATCH 1/2] f2fs: improve readahead for POSIX_FADV_WILLNEED

[Jaegeuk Kim]

On 01/22, Christoph Hellwig wrote:
> On Fri, Nov 21, 2025 at 06:05:01PM +0000, Jaegeuk Kim wrote:
> > 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.
> 
> Did this go anywhere?

Here.

https://lore.kernel.org/lkml/20251202013212.964298-1-jaegeuk@kernel.org/