Skip to contentSkip to conversation

f2fs: improve readahead for POSIX_FADV_WILLNEED

PATCH In review Jaegeuk Kim 2 participants 6 messages Nov 21, 2025 – Jan 22

Subsystems (4)
lore
Share & tools
lore

Jaegeuk Kim

author maintainer
#0 lore raw

PATCH 1/2 f2fs: improve readahead for POSIX_FADV_WILLNEED

This patch boosts readahead for POSIX_FADV_WILLNEED.

---
 fs/f2fs/data.c | 61 ++++++++++++++++++++++++++++++++++++++++++++++++++
 fs/f2fs/f2fs.h |  1 +
 fs/f2fs/file.c |  9 +++++---
 3 files changed, 68 insertions(+), 3 deletions(-)
fs/f2fs/data.c +61 -0
index a0433c8a4d84..d95974d79fb3 100644
@@ -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);
fs/f2fs/f2fs.h +1 -0
index 3340db04a7c2..934287cc5624 100644
@@ -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,
fs/f2fs/file.c +6 -3
index d7047ca6b98d..b6f71efd6d2a 100644
@@ -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);
Signature · 1 line
2.52.0.487.g5c8c507ade-goog
author maintainer
#1 lore raw

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

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

---
 fs/f2fs/data.c              |  3 +++
 include/trace/events/f2fs.h | 43 +++++++++++++++++++++++++++++++++++++
 2 files changed, 46 insertions(+)
fs/f2fs/data.c +3 -0
index d95974d79fb3..c80d7960b652 100644
@@ -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;
 }
 
include/trace/events/f2fs.h +43 -0

49 lines not on this page · raw message

Signature · 1 line
2.52.0.487.g5c8c507ade-goog
#2 lore raw
In reply to Jaegeuk Kim
Quoted 1 line · Jaegeuk Kim

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.

Quoted 116 lines
---
 fs/f2fs/data.c | 61 ++++++++++++++++++++++++++++++++++++++++++++++++++
 fs/f2fs/f2fs.h |  1 +
 fs/f2fs/file.c |  9 +++++---
 3 files changed, 68 insertions(+), 3 deletions(-)
fs/f2fs/data.c +61 -0
index a0433c8a4d84..d95974d79fb3 100644
@@ -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);
fs/f2fs/f2fs.h +1 -0
index 3340db04a7c2..934287cc5624 100644
@@ -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,
fs/f2fs/file.c +6 -3
index d7047ca6b98d..b6f71efd6d2a 100644
@@ -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);
Signature · 1 line
2.52.0.487.g5c8c507ade-goog

---end quoted text---

author maintainer
#3 lore raw
Quoted 9 lines · Christoph Hellwig
Quoted 1 line · Jaegeuk Kim

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.

Quoted 118 lines
Quoted 117 lines
---
 fs/f2fs/data.c | 61 ++++++++++++++++++++++++++++++++++++++++++++++++++
 fs/f2fs/f2fs.h |  1 +
 fs/f2fs/file.c |  9 +++++---
 3 files changed, 68 insertions(+), 3 deletions(-)
fs/f2fs/data.c +61 -0
index a0433c8a4d84..d95974d79fb3 100644
@@ -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);
fs/f2fs/f2fs.h +1 -0
index 3340db04a7c2..934287cc5624 100644
@@ -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,
fs/f2fs/file.c +6 -3
index d7047ca6b98d..b6f71efd6d2a 100644
@@ -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);
Signature · 1 line
2.52.0.487.g5c8c507ade-goog

---end quoted text---

#4 lore raw
Quoted 4 lines · Jaegeuk Kim
Quoted 3 lines · Christoph Hellwig
Quoted 1 line · Jaegeuk Kim

This patch boosts readahead for POSIX_FADV_WILLNEED.

Quoted 8 lines
Quoted 6 lines

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?

author maintainer
#5 lore raw
Quoted 7 lines · Christoph Hellwig
Quoted 6 lines · Jaegeuk Kim
Quoted 5 lines · Christoph Hellwig
Quoted 1 line · Jaegeuk Kim

This patch boosts readahead for POSIX_FADV_WILLNEED.

How? That's not a good changelog.

Quoted 8 lines
Quoted 6 lines
Quoted 4 lines

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/[email protected]/