* [PATCH 2/4] erofs: use `struct erofs_device_info` for the primary device
2024-12-12 13:35 [PATCH 1/4] erofs: add erofs_sb_free() helper Gao Xiang
@ 2024-12-12 13:35 ` Gao Xiang
2024-12-12 14:37 ` Chao Yu
2024-12-12 13:35 ` [PATCH 3/4] erofs: reference `struct erofs_device_info` for erofs_map_dev Gao Xiang
` (2 subsequent siblings)
3 siblings, 1 reply; 10+ messages in thread
From: Gao Xiang @ 2024-12-12 13:35 UTC (permalink / raw)
To: linux-erofs; +Cc: LKML, Gao Xiang
Instead of just listing each one directly in `struct erofs_sb_info`
except that we still use `sb->s_bdev` for the primary block device.
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
---
fs/erofs/data.c | 10 +++-------
fs/erofs/internal.h | 7 ++-----
fs/erofs/super.c | 27 +++++++++++++--------------
3 files changed, 18 insertions(+), 26 deletions(-)
diff --git a/fs/erofs/data.c b/fs/erofs/data.c
index 1c49f8962021..0b0385fa3025 100644
--- a/fs/erofs/data.c
+++ b/fs/erofs/data.c
@@ -56,7 +56,7 @@ void erofs_init_metabuf(struct erofs_buf *buf, struct super_block *sb)
buf->file = NULL;
if (erofs_is_fileio_mode(sbi)) {
- buf->file = sbi->fdev; /* some fs like FUSE needs it */
+ buf->file = sbi->dif0.file; /* some fs like FUSE needs it */
buf->mapping = buf->file->f_mapping;
} else if (erofs_is_fscache_mode(sb))
buf->mapping = sbi->s_fscache->inode->i_mapping;
@@ -201,12 +201,8 @@ int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
erofs_off_t startoff, length;
int id;
- map->m_bdev = sb->s_bdev;
- map->m_daxdev = EROFS_SB(sb)->dax_dev;
- map->m_dax_part_off = EROFS_SB(sb)->dax_part_off;
- map->m_fscache = EROFS_SB(sb)->s_fscache;
- map->m_fp = EROFS_SB(sb)->fdev;
-
+ erofs_fill_from_devinfo(map, &EROFS_SB(sb)->dif0);
+ map->m_bdev = sb->s_bdev; /* use s_bdev for the primary device */
if (map->m_deviceid) {
down_read(&devs->rwsem);
dif = idr_find(&devs->tree, map->m_deviceid - 1);
diff --git a/fs/erofs/internal.h b/fs/erofs/internal.h
index 1c847c30a918..6e91fcdeba94 100644
--- a/fs/erofs/internal.h
+++ b/fs/erofs/internal.h
@@ -107,6 +107,7 @@ struct erofs_xattr_prefix_item {
};
struct erofs_sb_info {
+ struct erofs_device_info dif0;
struct erofs_mount_opts opt; /* options */
#ifdef CONFIG_EROFS_FS_ZIP
/* list for all registered superblocks, mainly for shrinker */
@@ -124,13 +125,9 @@ struct erofs_sb_info {
struct erofs_sb_lz4_info lz4;
#endif /* CONFIG_EROFS_FS_ZIP */
- struct file *fdev;
struct inode *packed_inode;
struct erofs_dev_context *devs;
- struct dax_device *dax_dev;
- u64 dax_part_off;
u64 total_blocks;
- u32 primarydevice_blocks;
u32 meta_blkaddr;
#ifdef CONFIG_EROFS_FS_XATTR
@@ -187,7 +184,7 @@ struct erofs_sb_info {
static inline bool erofs_is_fileio_mode(struct erofs_sb_info *sbi)
{
- return IS_ENABLED(CONFIG_EROFS_FS_BACKED_BY_FILE) && sbi->fdev;
+ return IS_ENABLED(CONFIG_EROFS_FS_BACKED_BY_FILE) && sbi->dif0.file;
}
static inline bool erofs_is_fscache_mode(struct super_block *sb)
diff --git a/fs/erofs/super.c b/fs/erofs/super.c
index 1ce5ca11d32b..934fb1513095 100644
--- a/fs/erofs/super.c
+++ b/fs/erofs/super.c
@@ -203,7 +203,7 @@ static int erofs_scan_devices(struct super_block *sb,
struct erofs_device_info *dif;
int id, err = 0;
- sbi->total_blocks = sbi->primarydevice_blocks;
+ sbi->total_blocks = sbi->dif0.blocks;
if (!erofs_sb_has_device_table(sbi))
ondisk_extradevs = 0;
else
@@ -307,7 +307,7 @@ static int erofs_read_superblock(struct super_block *sb)
sbi->sb_size);
goto out;
}
- sbi->primarydevice_blocks = le32_to_cpu(dsb->blocks);
+ sbi->dif0.blocks = le32_to_cpu(dsb->blocks);
sbi->meta_blkaddr = le32_to_cpu(dsb->meta_blkaddr);
#ifdef CONFIG_EROFS_FS_XATTR
sbi->xattr_blkaddr = le32_to_cpu(dsb->xattr_blkaddr);
@@ -602,9 +602,8 @@ static int erofs_fc_fill_super(struct super_block *sb, struct fs_context *fc)
return -EINVAL;
}
- sbi->dax_dev = fs_dax_get_by_bdev(sb->s_bdev,
- &sbi->dax_part_off,
- NULL, NULL);
+ sbi->dif0.dax_dev = fs_dax_get_by_bdev(sb->s_bdev,
+ &sbi->dif0.dax_part_off, NULL, NULL);
}
err = erofs_read_superblock(sb);
@@ -627,7 +626,7 @@ static int erofs_fc_fill_super(struct super_block *sb, struct fs_context *fc)
}
if (test_opt(&sbi->opt, DAX_ALWAYS)) {
- if (!sbi->dax_dev) {
+ if (!sbi->dif0.dax_dev) {
errorfc(fc, "DAX unsupported by block device. Turning off DAX.");
clear_opt(&sbi->opt, DAX_ALWAYS);
} else if (sbi->blkszbits != PAGE_SHIFT) {
@@ -707,14 +706,13 @@ static int erofs_fc_get_tree(struct fs_context *fc)
if (!fc->source)
return invalf(fc, "No source specified");
-
file = filp_open(fc->source, O_RDONLY | O_LARGEFILE, 0);
if (IS_ERR(file))
return PTR_ERR(file);
- sbi->fdev = file;
+ sbi->dif0.file = file;
- if (S_ISREG(file_inode(sbi->fdev)->i_mode) &&
- sbi->fdev->f_mapping->a_ops->read_folio)
+ if (S_ISREG(file_inode(sbi->dif0.file)->i_mode) &&
+ sbi->dif0.file->f_mapping->a_ops->read_folio)
return get_tree_nodev(fc, erofs_fc_fill_super);
}
#endif
@@ -771,8 +769,8 @@ static void erofs_sb_free(struct erofs_sb_info *sbi)
erofs_free_dev_context(sbi->devs);
kfree(sbi->fsid);
kfree(sbi->domain_id);
- if (sbi->fdev)
- fput(sbi->fdev);
+ if (sbi->dif0.file)
+ fput(sbi->dif0.file);
kfree(sbi);
}
@@ -817,11 +815,12 @@ static void erofs_kill_sb(struct super_block *sb)
{
struct erofs_sb_info *sbi = EROFS_SB(sb);
- if ((IS_ENABLED(CONFIG_EROFS_FS_ONDEMAND) && sbi->fsid) || sbi->fdev)
+ if ((IS_ENABLED(CONFIG_EROFS_FS_ONDEMAND) && sbi->fsid) ||
+ sbi->dif0.file)
kill_anon_super(sb);
else
kill_block_super(sb);
- fs_put_dax(sbi->dax_dev, NULL);
+ fs_put_dax(sbi->dif0.dax_dev, NULL);
erofs_fscache_unregister_fs(sb);
erofs_sb_free(sbi);
sb->s_fs_info = NULL;
--
2.43.5
^ permalink raw reply [flat|nested] 10+ messages in thread* [PATCH 3/4] erofs: reference `struct erofs_device_info` for erofs_map_dev
2024-12-12 13:35 [PATCH 1/4] erofs: add erofs_sb_free() helper Gao Xiang
2024-12-12 13:35 ` [PATCH 2/4] erofs: use `struct erofs_device_info` for the primary device Gao Xiang
@ 2024-12-12 13:35 ` Gao Xiang
2024-12-12 14:42 ` Chao Yu
2024-12-12 23:54 ` [PATCH v2 " Gao Xiang
2024-12-12 13:35 ` [PATCH 4/4] erofs: use buffered I/O for file-backed mounts by default Gao Xiang
2024-12-12 14:29 ` [PATCH 1/4] erofs: add erofs_sb_free() helper Chao Yu
3 siblings, 2 replies; 10+ messages in thread
From: Gao Xiang @ 2024-12-12 13:35 UTC (permalink / raw)
To: linux-erofs; +Cc: LKML, Gao Xiang
Record `m_sb` and `m_dif` to replace `m_fscache`, `m_daxdev`, `m_fp`
and `m_dax_part_off` in order to simplify the codebase.
Note that `m_bdev` is still left since it can be assigned from
`sb->s_bdev` directly.
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
---
fs/erofs/data.c | 26 ++++++++++----------------
fs/erofs/fileio.c | 2 +-
fs/erofs/internal.h | 6 ++----
3 files changed, 13 insertions(+), 21 deletions(-)
diff --git a/fs/erofs/data.c b/fs/erofs/data.c
index 0b0385fa3025..63840971976c 100644
--- a/fs/erofs/data.c
+++ b/fs/erofs/data.c
@@ -179,19 +179,13 @@ int erofs_map_blocks(struct inode *inode, struct erofs_map_blocks *map)
}
static void erofs_fill_from_devinfo(struct erofs_map_dev *map,
- struct erofs_device_info *dif)
+ struct super_block *sb, struct erofs_device_info *dif)
{
+ map->m_sb = sb;
+ map->m_dif = dif;
map->m_bdev = NULL;
- map->m_fp = NULL;
- if (dif->file) {
- if (S_ISBLK(file_inode(dif->file)->i_mode))
- map->m_bdev = file_bdev(dif->file);
- else
- map->m_fp = dif->file;
- }
- map->m_daxdev = dif->dax_dev;
- map->m_dax_part_off = dif->dax_part_off;
- map->m_fscache = dif->fscache;
+ if (dif->file && S_ISBLK(file_inode(dif->file)->i_mode))
+ map->m_bdev = file_bdev(dif->file);
}
int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
@@ -201,7 +195,7 @@ int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
erofs_off_t startoff, length;
int id;
- erofs_fill_from_devinfo(map, &EROFS_SB(sb)->dif0);
+ erofs_fill_from_devinfo(map, sb, &EROFS_SB(sb)->dif0);
map->m_bdev = sb->s_bdev; /* use s_bdev for the primary device */
if (map->m_deviceid) {
down_read(&devs->rwsem);
@@ -215,7 +209,7 @@ int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
up_read(&devs->rwsem);
return 0;
}
- erofs_fill_from_devinfo(map, dif);
+ erofs_fill_from_devinfo(map, sb, dif);
up_read(&devs->rwsem);
} else if (devs->extra_devices && !devs->flatdev) {
down_read(&devs->rwsem);
@@ -228,7 +222,7 @@ int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
if (map->m_pa >= startoff &&
map->m_pa < startoff + length) {
map->m_pa -= startoff;
- erofs_fill_from_devinfo(map, dif);
+ erofs_fill_from_devinfo(map, sb, dif);
break;
}
}
@@ -298,7 +292,7 @@ static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
iomap->offset = map.m_la;
if (flags & IOMAP_DAX)
- iomap->dax_dev = mdev.m_daxdev;
+ iomap->dax_dev = mdev.m_dif->dax_dev;
else
iomap->bdev = mdev.m_bdev;
iomap->length = map.m_llen;
@@ -327,7 +321,7 @@ static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
iomap->type = IOMAP_MAPPED;
iomap->addr = mdev.m_pa;
if (flags & IOMAP_DAX)
- iomap->addr += mdev.m_dax_part_off;
+ iomap->addr += mdev.m_dif->dax_part_off;
}
return 0;
}
diff --git a/fs/erofs/fileio.c b/fs/erofs/fileio.c
index 3af96b1e2c2a..a61b8faec651 100644
--- a/fs/erofs/fileio.c
+++ b/fs/erofs/fileio.c
@@ -67,7 +67,7 @@ static struct erofs_fileio_rq *erofs_fileio_rq_alloc(struct erofs_map_dev *mdev)
GFP_KERNEL | __GFP_NOFAIL);
bio_init(&rq->bio, NULL, rq->bvecs, BIO_MAX_VECS, REQ_OP_READ);
- rq->iocb.ki_filp = mdev->m_fp;
+ rq->iocb.ki_filp = mdev->m_dif->file;
return rq;
}
diff --git a/fs/erofs/internal.h b/fs/erofs/internal.h
index 6e91fcdeba94..59e7739b7c3b 100644
--- a/fs/erofs/internal.h
+++ b/fs/erofs/internal.h
@@ -354,11 +354,9 @@ enum {
};
struct erofs_map_dev {
- struct erofs_fscache *m_fscache;
+ struct super_block *m_sb;
+ struct erofs_device_info *m_dif;
struct block_device *m_bdev;
- struct dax_device *m_daxdev;
- struct file *m_fp;
- u64 m_dax_part_off;
erofs_off_t m_pa;
unsigned int m_deviceid;
--
2.43.5
^ permalink raw reply [flat|nested] 10+ messages in thread* Re: [PATCH 3/4] erofs: reference `struct erofs_device_info` for erofs_map_dev
2024-12-12 13:35 ` [PATCH 3/4] erofs: reference `struct erofs_device_info` for erofs_map_dev Gao Xiang
@ 2024-12-12 14:42 ` Chao Yu
2024-12-12 23:54 ` [PATCH v2 " Gao Xiang
1 sibling, 0 replies; 10+ messages in thread
From: Chao Yu @ 2024-12-12 14:42 UTC (permalink / raw)
To: Gao Xiang, linux-erofs; +Cc: Chao Yu, LKML
On 2024/12/12 21:35, Gao Xiang wrote:
> Record `m_sb` and `m_dif` to replace `m_fscache`, `m_daxdev`, `m_fp`
> and `m_dax_part_off` in order to simplify the codebase.
>
> Note that `m_bdev` is still left since it can be assigned from
> `sb->s_bdev` directly.
>
> Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Reviewed-by: Chao Yu <chao@kernel.org>
Thanks,
^ permalink raw reply [flat|nested] 10+ messages in thread
* [PATCH v2 3/4] erofs: reference `struct erofs_device_info` for erofs_map_dev
2024-12-12 13:35 ` [PATCH 3/4] erofs: reference `struct erofs_device_info` for erofs_map_dev Gao Xiang
2024-12-12 14:42 ` Chao Yu
@ 2024-12-12 23:54 ` Gao Xiang
1 sibling, 0 replies; 10+ messages in thread
From: Gao Xiang @ 2024-12-12 23:54 UTC (permalink / raw)
To: linux-erofs; +Cc: LKML, Gao Xiang, Chao Yu
Record `m_sb` and `m_dif` to replace `m_fscache`, `m_daxdev`, `m_fp`
and `m_dax_part_off` in order to simplify the codebase.
Note that `m_bdev` is still left since it can be assigned from
`sb->s_bdev` directly.
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
---
v1: https://lore.kernel.org/r/20241212133504.2047178-3-hsiangkao@linux.alibaba.com
changes since v1:
- fix fscache build.
fs/erofs/data.c | 26 ++++++++++----------------
fs/erofs/fileio.c | 2 +-
fs/erofs/fscache.c | 4 ++--
fs/erofs/internal.h | 6 ++----
4 files changed, 15 insertions(+), 23 deletions(-)
diff --git a/fs/erofs/data.c b/fs/erofs/data.c
index 0b0385fa3025..63840971976c 100644
--- a/fs/erofs/data.c
+++ b/fs/erofs/data.c
@@ -179,19 +179,13 @@ int erofs_map_blocks(struct inode *inode, struct erofs_map_blocks *map)
}
static void erofs_fill_from_devinfo(struct erofs_map_dev *map,
- struct erofs_device_info *dif)
+ struct super_block *sb, struct erofs_device_info *dif)
{
+ map->m_sb = sb;
+ map->m_dif = dif;
map->m_bdev = NULL;
- map->m_fp = NULL;
- if (dif->file) {
- if (S_ISBLK(file_inode(dif->file)->i_mode))
- map->m_bdev = file_bdev(dif->file);
- else
- map->m_fp = dif->file;
- }
- map->m_daxdev = dif->dax_dev;
- map->m_dax_part_off = dif->dax_part_off;
- map->m_fscache = dif->fscache;
+ if (dif->file && S_ISBLK(file_inode(dif->file)->i_mode))
+ map->m_bdev = file_bdev(dif->file);
}
int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
@@ -201,7 +195,7 @@ int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
erofs_off_t startoff, length;
int id;
- erofs_fill_from_devinfo(map, &EROFS_SB(sb)->dif0);
+ erofs_fill_from_devinfo(map, sb, &EROFS_SB(sb)->dif0);
map->m_bdev = sb->s_bdev; /* use s_bdev for the primary device */
if (map->m_deviceid) {
down_read(&devs->rwsem);
@@ -215,7 +209,7 @@ int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
up_read(&devs->rwsem);
return 0;
}
- erofs_fill_from_devinfo(map, dif);
+ erofs_fill_from_devinfo(map, sb, dif);
up_read(&devs->rwsem);
} else if (devs->extra_devices && !devs->flatdev) {
down_read(&devs->rwsem);
@@ -228,7 +222,7 @@ int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
if (map->m_pa >= startoff &&
map->m_pa < startoff + length) {
map->m_pa -= startoff;
- erofs_fill_from_devinfo(map, dif);
+ erofs_fill_from_devinfo(map, sb, dif);
break;
}
}
@@ -298,7 +292,7 @@ static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
iomap->offset = map.m_la;
if (flags & IOMAP_DAX)
- iomap->dax_dev = mdev.m_daxdev;
+ iomap->dax_dev = mdev.m_dif->dax_dev;
else
iomap->bdev = mdev.m_bdev;
iomap->length = map.m_llen;
@@ -327,7 +321,7 @@ static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
iomap->type = IOMAP_MAPPED;
iomap->addr = mdev.m_pa;
if (flags & IOMAP_DAX)
- iomap->addr += mdev.m_dax_part_off;
+ iomap->addr += mdev.m_dif->dax_part_off;
}
return 0;
}
diff --git a/fs/erofs/fileio.c b/fs/erofs/fileio.c
index 3af96b1e2c2a..a61b8faec651 100644
--- a/fs/erofs/fileio.c
+++ b/fs/erofs/fileio.c
@@ -67,7 +67,7 @@ static struct erofs_fileio_rq *erofs_fileio_rq_alloc(struct erofs_map_dev *mdev)
GFP_KERNEL | __GFP_NOFAIL);
bio_init(&rq->bio, NULL, rq->bvecs, BIO_MAX_VECS, REQ_OP_READ);
- rq->iocb.ki_filp = mdev->m_fp;
+ rq->iocb.ki_filp = mdev->m_dif->file;
return rq;
}
diff --git a/fs/erofs/fscache.c b/fs/erofs/fscache.c
index fda16eedafb5..1268baa2ecc4 100644
--- a/fs/erofs/fscache.c
+++ b/fs/erofs/fscache.c
@@ -198,7 +198,7 @@ struct bio *erofs_fscache_bio_alloc(struct erofs_map_dev *mdev)
io = kmalloc(sizeof(*io), GFP_KERNEL | __GFP_NOFAIL);
bio_init(&io->bio, NULL, io->bvecs, BIO_MAX_VECS, REQ_OP_READ);
- io->io.private = mdev->m_fscache->cookie;
+ io->io.private = mdev->m_dif->fscache->cookie;
io->io.end_io = erofs_fscache_bio_endio;
refcount_set(&io->io.ref, 1);
return &io->bio;
@@ -316,7 +316,7 @@ static int erofs_fscache_data_read_slice(struct erofs_fscache_rq *req)
if (!io)
return -ENOMEM;
iov_iter_xarray(&io->iter, ITER_DEST, &mapping->i_pages, pos, count);
- ret = erofs_fscache_read_io_async(mdev.m_fscache->cookie,
+ ret = erofs_fscache_read_io_async(mdev.m_dif->fscache->cookie,
mdev.m_pa + (pos - map.m_la), io);
erofs_fscache_req_io_put(io);
diff --git a/fs/erofs/internal.h b/fs/erofs/internal.h
index 6e91fcdeba94..59e7739b7c3b 100644
--- a/fs/erofs/internal.h
+++ b/fs/erofs/internal.h
@@ -354,11 +354,9 @@ enum {
};
struct erofs_map_dev {
- struct erofs_fscache *m_fscache;
+ struct super_block *m_sb;
+ struct erofs_device_info *m_dif;
struct block_device *m_bdev;
- struct dax_device *m_daxdev;
- struct file *m_fp;
- u64 m_dax_part_off;
erofs_off_t m_pa;
unsigned int m_deviceid;
--
2.43.5
^ permalink raw reply [flat|nested] 10+ messages in thread
* [PATCH 4/4] erofs: use buffered I/O for file-backed mounts by default
2024-12-12 13:35 [PATCH 1/4] erofs: add erofs_sb_free() helper Gao Xiang
2024-12-12 13:35 ` [PATCH 2/4] erofs: use `struct erofs_device_info` for the primary device Gao Xiang
2024-12-12 13:35 ` [PATCH 3/4] erofs: reference `struct erofs_device_info` for erofs_map_dev Gao Xiang
@ 2024-12-12 13:35 ` Gao Xiang
2024-12-12 13:43 ` [PATCH v2 " Gao Xiang
2024-12-12 14:29 ` [PATCH 1/4] erofs: add erofs_sb_free() helper Chao Yu
3 siblings, 1 reply; 10+ messages in thread
From: Gao Xiang @ 2024-12-12 13:35 UTC (permalink / raw)
To: linux-erofs; +Cc: LKML, Gao Xiang, Derek McGowan
For many use cases (e.g. container images are just fetched from remote),
performance will be impacted if underlay page cache is up-to-date but
direct i/o will flush dirty pages first.
Instead, let's use buffered I/O by default and add a (re)mount option to
explicitly enable direct I/O if underlay files are supported too.
The startup time for containerd is improved as below:
unpack 1st run non-1st runs
EROFS snapshotter buffered I/O file 4.586404265s 0.308s 0.198s
EROFS snapshotter direct I/O file 4.581742849s 2.238s 0.222s
EROFS snapshotter loop 4.596023152s 0.346s 0.201s
Overlayfs snapshotter 5.382851037s 0.206s 0.214s
Fixes: fb176750266a ("erofs: add file-backed mount support")
Cc: Derek McGowan <derek@mcg.dev>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
---
fs/erofs/fileio.c | 7 +++++--
fs/erofs/internal.h | 1 +
fs/erofs/super.c | 21 +++++++++++++--------
3 files changed, 19 insertions(+), 10 deletions(-)
diff --git a/fs/erofs/fileio.c b/fs/erofs/fileio.c
index a61b8faec651..33f8539dda4a 100644
--- a/fs/erofs/fileio.c
+++ b/fs/erofs/fileio.c
@@ -9,6 +9,7 @@ struct erofs_fileio_rq {
struct bio_vec bvecs[BIO_MAX_VECS];
struct bio bio;
struct kiocb iocb;
+ struct super_block *sb;
};
struct erofs_fileio {
@@ -52,8 +53,9 @@ static void erofs_fileio_rq_submit(struct erofs_fileio_rq *rq)
rq->iocb.ki_pos = rq->bio.bi_iter.bi_sector << SECTOR_SHIFT;
rq->iocb.ki_ioprio = get_current_ioprio();
rq->iocb.ki_complete = erofs_fileio_ki_complete;
- rq->iocb.ki_flags = (rq->iocb.ki_filp->f_mode & FMODE_CAN_ODIRECT) ?
- IOCB_DIRECT : 0;
+ if (test_opt(&EROFS_SB(rq->sb)->opt, DIRECT_IO) &&
+ rq->iocb.ki_filp->f_mode & FMODE_CAN_ODIRECT)
+ rq->iocb.ki_flags = IOCB_DIRECT;
iov_iter_bvec(&iter, ITER_DEST, rq->bvecs, rq->bio.bi_vcnt,
rq->bio.bi_iter.bi_size);
ret = vfs_iocb_iter_read(rq->iocb.ki_filp, &rq->iocb, &iter);
@@ -68,6 +70,7 @@ static struct erofs_fileio_rq *erofs_fileio_rq_alloc(struct erofs_map_dev *mdev)
bio_init(&rq->bio, NULL, rq->bvecs, BIO_MAX_VECS, REQ_OP_READ);
rq->iocb.ki_filp = mdev->m_dif->file;
+ rq->sb = mdev->m_sb;
return rq;
}
diff --git a/fs/erofs/internal.h b/fs/erofs/internal.h
index 59e7739b7c3b..bc46000ccc6b 100644
--- a/fs/erofs/internal.h
+++ b/fs/erofs/internal.h
@@ -177,6 +177,7 @@ struct erofs_sb_info {
#define EROFS_MOUNT_POSIX_ACL 0x00000020
#define EROFS_MOUNT_DAX_ALWAYS 0x00000040
#define EROFS_MOUNT_DAX_NEVER 0x00000080
+#define EROFS_MOUNT_DIRECT_IO 0x00000100
#define clear_opt(opt, option) ((opt)->mount_opt &= ~EROFS_MOUNT_##option)
#define set_opt(opt, option) ((opt)->mount_opt |= EROFS_MOUNT_##option)
diff --git a/fs/erofs/super.c b/fs/erofs/super.c
index 934fb1513095..cbb9566cc6dc 100644
--- a/fs/erofs/super.c
+++ b/fs/erofs/super.c
@@ -364,14 +364,8 @@ static void erofs_default_options(struct erofs_sb_info *sbi)
}
enum {
- Opt_user_xattr,
- Opt_acl,
- Opt_cache_strategy,
- Opt_dax,
- Opt_dax_enum,
- Opt_device,
- Opt_fsid,
- Opt_domain_id,
+ Opt_user_xattr, Opt_acl, Opt_cache_strategy, Opt_dax, Opt_dax_enum,
+ Opt_device, Opt_fsid, Opt_domain_id, Opt_directio,
Opt_err
};
@@ -398,6 +392,7 @@ static const struct fs_parameter_spec erofs_fs_parameters[] = {
fsparam_string("device", Opt_device),
fsparam_string("fsid", Opt_fsid),
fsparam_string("domain_id", Opt_domain_id),
+ fsparam_flag_no("directio", Opt_directio),
{}
};
@@ -511,6 +506,16 @@ static int erofs_fc_parse_param(struct fs_context *fc,
errorfc(fc, "%s option not supported", erofs_fs_parameters[opt].name);
break;
#endif
+ case Opt_directio:
+#ifdef CONFIG_EROFS_FS_BACKED_BY_FILE
+ if (result.boolean)
+ set_opt(&sbi->opt, DIRECT_IO);
+ else
+ clear_opt(&sbi->opt, DIRECT_IO);
+#else
+ errorfc(fc, "%s option not supported", erofs_fs_parameters[opt].name);
+#endif
+ break;
default:
return -ENOPARAM;
}
--
2.43.5
^ permalink raw reply [flat|nested] 10+ messages in thread* [PATCH v2 4/4] erofs: use buffered I/O for file-backed mounts by default
2024-12-12 13:35 ` [PATCH 4/4] erofs: use buffered I/O for file-backed mounts by default Gao Xiang
@ 2024-12-12 13:43 ` Gao Xiang
2024-12-12 14:46 ` Chao Yu
0 siblings, 1 reply; 10+ messages in thread
From: Gao Xiang @ 2024-12-12 13:43 UTC (permalink / raw)
To: linux-erofs; +Cc: LKML, Gao Xiang, Derek McGowan
For many use cases (e.g. container images are just fetched from remote),
performance will be impacted if underlay page cache is up-to-date but
direct i/o flushes dirty pages first.
Instead, let's use buffered I/O by default to keep in sync with loop
devices and add a (re)mount option to explicitly enable direct I/O
if supported by the underlying files.
The container startup time is improved as below:
unpack 1st run non-1st runs
EROFS snapshotter buffered I/O file 4.586404265s 0.308s 0.198s
EROFS snapshotter direct I/O file 4.581742849s 2.238s 0.222s
EROFS snapshotter loop 4.596023152s 0.346s 0.201s
Overlayfs snapshotter 5.382851037s 0.206s 0.214s
Fixes: fb176750266a ("erofs: add file-backed mount support")
Cc: Derek McGowan <derek@mcg.dev>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
---
v1: https://lore.kernel.org/r/20241212133504.2047178-4-hsiangkao@linux.alibaba.com
changes since v1 (obsoleted version by mistake):
- refine commit message;
- add `directio` in erofs_show_options() too.
fs/erofs/fileio.c | 7 +++++--
fs/erofs/internal.h | 1 +
fs/erofs/super.c | 23 +++++++++++++++--------
3 files changed, 21 insertions(+), 10 deletions(-)
diff --git a/fs/erofs/fileio.c b/fs/erofs/fileio.c
index a61b8faec651..33f8539dda4a 100644
--- a/fs/erofs/fileio.c
+++ b/fs/erofs/fileio.c
@@ -9,6 +9,7 @@ struct erofs_fileio_rq {
struct bio_vec bvecs[BIO_MAX_VECS];
struct bio bio;
struct kiocb iocb;
+ struct super_block *sb;
};
struct erofs_fileio {
@@ -52,8 +53,9 @@ static void erofs_fileio_rq_submit(struct erofs_fileio_rq *rq)
rq->iocb.ki_pos = rq->bio.bi_iter.bi_sector << SECTOR_SHIFT;
rq->iocb.ki_ioprio = get_current_ioprio();
rq->iocb.ki_complete = erofs_fileio_ki_complete;
- rq->iocb.ki_flags = (rq->iocb.ki_filp->f_mode & FMODE_CAN_ODIRECT) ?
- IOCB_DIRECT : 0;
+ if (test_opt(&EROFS_SB(rq->sb)->opt, DIRECT_IO) &&
+ rq->iocb.ki_filp->f_mode & FMODE_CAN_ODIRECT)
+ rq->iocb.ki_flags = IOCB_DIRECT;
iov_iter_bvec(&iter, ITER_DEST, rq->bvecs, rq->bio.bi_vcnt,
rq->bio.bi_iter.bi_size);
ret = vfs_iocb_iter_read(rq->iocb.ki_filp, &rq->iocb, &iter);
@@ -68,6 +70,7 @@ static struct erofs_fileio_rq *erofs_fileio_rq_alloc(struct erofs_map_dev *mdev)
bio_init(&rq->bio, NULL, rq->bvecs, BIO_MAX_VECS, REQ_OP_READ);
rq->iocb.ki_filp = mdev->m_dif->file;
+ rq->sb = mdev->m_sb;
return rq;
}
diff --git a/fs/erofs/internal.h b/fs/erofs/internal.h
index 59e7739b7c3b..bc46000ccc6b 100644
--- a/fs/erofs/internal.h
+++ b/fs/erofs/internal.h
@@ -177,6 +177,7 @@ struct erofs_sb_info {
#define EROFS_MOUNT_POSIX_ACL 0x00000020
#define EROFS_MOUNT_DAX_ALWAYS 0x00000040
#define EROFS_MOUNT_DAX_NEVER 0x00000080
+#define EROFS_MOUNT_DIRECT_IO 0x00000100
#define clear_opt(opt, option) ((opt)->mount_opt &= ~EROFS_MOUNT_##option)
#define set_opt(opt, option) ((opt)->mount_opt |= EROFS_MOUNT_##option)
diff --git a/fs/erofs/super.c b/fs/erofs/super.c
index 9044907354e1..f5956474bfde 100644
--- a/fs/erofs/super.c
+++ b/fs/erofs/super.c
@@ -364,14 +364,8 @@ static void erofs_default_options(struct erofs_sb_info *sbi)
}
enum {
- Opt_user_xattr,
- Opt_acl,
- Opt_cache_strategy,
- Opt_dax,
- Opt_dax_enum,
- Opt_device,
- Opt_fsid,
- Opt_domain_id,
+ Opt_user_xattr, Opt_acl, Opt_cache_strategy, Opt_dax, Opt_dax_enum,
+ Opt_device, Opt_fsid, Opt_domain_id, Opt_directio,
Opt_err
};
@@ -398,6 +392,7 @@ static const struct fs_parameter_spec erofs_fs_parameters[] = {
fsparam_string("device", Opt_device),
fsparam_string("fsid", Opt_fsid),
fsparam_string("domain_id", Opt_domain_id),
+ fsparam_flag_no("directio", Opt_directio),
{}
};
@@ -511,6 +506,16 @@ static int erofs_fc_parse_param(struct fs_context *fc,
errorfc(fc, "%s option not supported", erofs_fs_parameters[opt].name);
break;
#endif
+ case Opt_directio:
+#ifdef CONFIG_EROFS_FS_BACKED_BY_FILE
+ if (result.boolean)
+ set_opt(&sbi->opt, DIRECT_IO);
+ else
+ clear_opt(&sbi->opt, DIRECT_IO);
+#else
+ errorfc(fc, "%s option not supported", erofs_fs_parameters[opt].name);
+#endif
+ break;
default:
return -ENOPARAM;
}
@@ -948,6 +953,8 @@ static int erofs_show_options(struct seq_file *seq, struct dentry *root)
seq_puts(seq, ",dax=always");
if (test_opt(opt, DAX_NEVER))
seq_puts(seq, ",dax=never");
+ if (erofs_is_fileio_mode(sbi) && test_opt(opt, DIRECT_IO))
+ seq_puts(seq, ",directio");
#ifdef CONFIG_EROFS_FS_ONDEMAND
if (sbi->fsid)
seq_printf(seq, ",fsid=%s", sbi->fsid);
--
2.43.5
^ permalink raw reply [flat|nested] 10+ messages in thread* Re: [PATCH v2 4/4] erofs: use buffered I/O for file-backed mounts by default
2024-12-12 13:43 ` [PATCH v2 " Gao Xiang
@ 2024-12-12 14:46 ` Chao Yu
0 siblings, 0 replies; 10+ messages in thread
From: Chao Yu @ 2024-12-12 14:46 UTC (permalink / raw)
To: Gao Xiang, linux-erofs; +Cc: Chao Yu, Derek McGowan, LKML
On 2024/12/12 21:43, Gao Xiang wrote:
> For many use cases (e.g. container images are just fetched from remote),
> performance will be impacted if underlay page cache is up-to-date but
> direct i/o flushes dirty pages first.
>
> Instead, let's use buffered I/O by default to keep in sync with loop
> devices and add a (re)mount option to explicitly enable direct I/O
> if supported by the underlying files.
>
> The container startup time is improved as below:
> unpack 1st run non-1st runs
> EROFS snapshotter buffered I/O file 4.586404265s 0.308s 0.198s
> EROFS snapshotter direct I/O file 4.581742849s 2.238s 0.222s
> EROFS snapshotter loop 4.596023152s 0.346s 0.201s
> Overlayfs snapshotter 5.382851037s 0.206s 0.214s
>
> Fixes: fb176750266a ("erofs: add file-backed mount support")
> Cc: Derek McGowan <derek@mcg.dev>
> Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Reviewed-by: Chao Yu <chao@kernel.org>
Thanks,
^ permalink raw reply [flat|nested] 10+ messages in thread
* Re: [PATCH 1/4] erofs: add erofs_sb_free() helper
2024-12-12 13:35 [PATCH 1/4] erofs: add erofs_sb_free() helper Gao Xiang
` (2 preceding siblings ...)
2024-12-12 13:35 ` [PATCH 4/4] erofs: use buffered I/O for file-backed mounts by default Gao Xiang
@ 2024-12-12 14:29 ` Chao Yu
3 siblings, 0 replies; 10+ messages in thread
From: Chao Yu @ 2024-12-12 14:29 UTC (permalink / raw)
To: Gao Xiang, linux-erofs; +Cc: Chao Yu, LKML
On 2024/12/12 21:35, Gao Xiang wrote:
> Unify the common parts of erofs_fc_free() and erofs_kill_sb() as
> erofs_sb_free().
>
> Thus, fput() in erofs_fc_get_tree() is no longer needed, too.
>
> Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
Reviewed-by: Chao Yu <chao@kernel.org>
Thanks,
^ permalink raw reply [flat|nested] 10+ messages in thread