Skip to contentSkip to conversation

erofs: get rid of erofs_get_meta_page()

PATCH In review Gao Xiang 3 participants 22 messages Jan 2, 2022 – Jan 4, 2022 EROFS file system

lore
Share & tools
lore

Gao Xiang

author
#0 lore raw

v2 cover erofs: get rid of erofs_get_meta_page()

Hi folks,

erofs_get_meta_page() is actually inflexible since it's too close to the page itself.

In order to prepare for folio and subpage features, introduce on-stack meta buffer descriptor instead and convert all erofs_get_meta_page() users to use it.

It can also be used for new potential backends such as fscache or mtd.

Patches are trivial.

Thanks,
Gao Xiang

changes since v1:
- fix unbalanced put_metabuf in xattrs.

Gao Xiang (5):
erofs: introduce meta buffer operations
erofs: use meta buffers for inode operations
erofs: use meta buffers for super operations
erofs: use meta buffers for xattr operations
erofs: use meta buffers for zmap operations

 fs/erofs/data.c     | 102 +++++++++++++++++++++++----------
 fs/erofs/inode.c    |  68 +++++++++++-----------
 fs/erofs/internal.h |  22 +++++++-
 fs/erofs/super.c    | 105 +++++++++-------------------------
 fs/erofs/xattr.c    | 134 +++++++++++++-------------------------------
 fs/erofs/zdata.c    |  23 +++-----
 fs/erofs/zmap.c     |  56 ++++++------------
 7 files changed, 214 insertions(+), 296 deletions(-)
Signature · 1 line
2.24.4
author
#1 lore raw

PATCH v2 1/5 erofs: introduce meta buffer operations

In order to support subpage and folio for all uncompressed files, introduce meta buffer descriptors, which can be effectively stored on stack, in place of meta page operations.

This converts the uncompressed data path to meta buffers.

---
 fs/erofs/data.c     | 97 +++++++++++++++++++++++++++++++++++----------
 fs/erofs/internal.h | 13 ++++++
 2 files changed, 89 insertions(+), 21 deletions(-)
fs/erofs/data.c +76 -21
index 4f98c76ec043..6495e16a50a9 100644
@@ -22,6 +22,56 @@ struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr)
 	return page;
 }
 
+void erofs_unmap_metabuf(struct erofs_buf *buf)
+{
+	if (buf->kmap_type == EROFS_KMAP)
+		kunmap(buf->page);
+	else if (buf->kmap_type == EROFS_KMAP_ATOMIC)
+		kunmap_atomic(buf->base);
+	buf->kmap_type = EROFS_NO_KMAP;
+}
+
+void erofs_put_metabuf(struct erofs_buf *buf)
+{
+	if (!buf->page)
+		return;
+	erofs_unmap_metabuf(buf);
+	put_page(buf->page);
+	buf->page = NULL;
+}
+
+void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
+			erofs_blk_t blkaddr, enum erofs_kmap_type type)
+{
+	struct address_space *const mapping = sb->s_bdev->bd_inode->i_mapping;
+	erofs_off_t offset = blknr_to_addr(blkaddr);
+	pgoff_t index = offset >> PAGE_SHIFT;
+	struct page *page = buf->page;
+
+	if (!page || page->index != index) {
+		erofs_put_metabuf(buf);
+		page = read_cache_page_gfp(mapping, index,
+				mapping_gfp_constraint(mapping, ~__GFP_FS));
+		if (IS_ERR(page))
+			return page;
+		/* should already be PageUptodate, no need to lock page */
+		buf->page = page;
+	}
+	if (buf->kmap_type == EROFS_NO_KMAP) {
+		if (type == EROFS_KMAP)
+			buf->base = kmap(page);
+		else if (type == EROFS_KMAP_ATOMIC)
+			buf->base = kmap_atomic(page);
+		buf->kmap_type = type;
+	} else if (buf->kmap_type != type) {
+		DBG_BUGON(1);
+		return ERR_PTR(-EFAULT);
+	}
+	if (type == EROFS_NO_KMAP)
+		return NULL;
+	return buf->base + (offset & ~PAGE_MASK);
+}
+
 static int erofs_map_blocks_flatmode(struct inode *inode,
 				     struct erofs_map_blocks *map,
 				     int flags)
@@ -31,7 +81,7 @@ static int erofs_map_blocks_flatmode(struct inode *inode,
 	struct erofs_inode *vi = EROFS_I(inode);
 	bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE);
 
-	nblocks = DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
+	nblocks = DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ);
 	lastblk = nblocks - tailendpacking;
 
 	/* there is no hole in flatmode */
@@ -72,10 +122,11 @@ static int erofs_map_blocks(struct inode *inode,
 	struct super_block *sb = inode->i_sb;
 	struct erofs_inode *vi = EROFS_I(inode);
 	struct erofs_inode_chunk_index *idx;
-	struct page *page;
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 	u64 chunknr;
 	unsigned int unit;
 	erofs_off_t pos;
+	void *kaddr;
 	int err = 0;
 
 	trace_erofs_map_blocks_enter(inode, map, flags);
@@ -101,9 +152,9 @@ static int erofs_map_blocks(struct inode *inode,
 	pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize +
 		    vi->xattr_isize, unit) + unit * chunknr;
 
-	page = erofs_get_meta_page(inode->i_sb, erofs_blknr(pos));
-	if (IS_ERR(page)) {
-		err = PTR_ERR(page);
+	kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos), EROFS_KMAP);
+	if (IS_ERR(kaddr)) {
+		err = PTR_ERR(kaddr);
 		goto out;
 	}
 	map->m_la = chunknr << vi->chunkbits;
@@ -112,7 +163,7 @@ static int erofs_map_blocks(struct inode *inode,
 
 	/* handle block map */
 	if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) {
-		__le32 *blkaddr = page_address(page) + erofs_blkoff(pos);
+		__le32 *blkaddr = kaddr + erofs_blkoff(pos);
 
 		if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) {
 			map->m_flags = 0;
@@ -123,7 +174,7 @@ static int erofs_map_blocks(struct inode *inode,
 		goto out_unlock;
 	}
 	/* parse chunk indexes */
-	idx = page_address(page) + erofs_blkoff(pos);
+	idx = kaddr + erofs_blkoff(pos);
 	switch (le32_to_cpu(idx->blkaddr)) {
 	case EROFS_NULL_ADDR:
 		map->m_flags = 0;
@@ -136,8 +187,7 @@ static int erofs_map_blocks(struct inode *inode,
 		break;
 	}
 out_unlock:
-	unlock_page(page);
-	put_page(page);
+	erofs_put_metabuf(&buf);
 out:
 	if (!err)
 		map->m_llen = map->m_plen;
@@ -226,16 +276,16 @@ static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 	}
 
 	if (map.m_flags & EROFS_MAP_META) {
-		struct page *ipage;
+		void *ptr;
+		struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 
 		iomap->type = IOMAP_INLINE;
-		ipage = erofs_get_meta_page(inode->i_sb,
-					    erofs_blknr(mdev.m_pa));
-		if (IS_ERR(ipage))
-			return PTR_ERR(ipage);
-		iomap->inline_data = page_address(ipage) +
-					erofs_blkoff(mdev.m_pa);
-		iomap->private = ipage;
+		ptr = erofs_read_metabuf(&buf, inode->i_sb,
+					 erofs_blknr(mdev.m_pa), EROFS_KMAP);
+		if (IS_ERR(ptr))
+			return PTR_ERR(ptr);
+		iomap->inline_data = ptr + erofs_blkoff(mdev.m_pa);
+		iomap->private = buf.base;
 	} else {
 		iomap->type = IOMAP_MAPPED;
 		iomap->addr = mdev.m_pa;
@@ -246,12 +296,17 @@ static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length,
 		ssize_t written, unsigned int flags, struct iomap *iomap)
 {
-	struct page *ipage = iomap->private;
+	void *ptr = iomap->private;
+
+	if (ptr) {
+		struct erofs_buf buf = {
+			.page = kmap_to_page(ptr),
+			.base = ptr,
+			.kmap_type = EROFS_KMAP,
+		};
 
-	if (ipage) {
 		DBG_BUGON(iomap->type != IOMAP_INLINE);
-		unlock_page(ipage);
-		put_page(ipage);
+		erofs_put_metabuf(&buf);
 	} else {
 		DBG_BUGON(iomap->type == IOMAP_INLINE);
 	}
fs/erofs/internal.h +13 -0
index fca3747d97be..7053f1c4171d 100644
@@ -251,6 +251,19 @@ static inline int erofs_wait_on_workgroup_freezed(struct erofs_workgroup *grp)
 #error erofs cannot be used in this platform
 #endif
 
+enum erofs_kmap_type {
+	EROFS_NO_KMAP,		/* don't map the buffer */
+	EROFS_KMAP,		/* use kmap() to map the buffer */
+	EROFS_KMAP_ATOMIC,	/* use kmap_atomic() to map the buffer */
+};
+
+struct erofs_buf {
+	struct page *page;
+	void *base;
+	enum erofs_kmap_type kmap_type;
+};
+#define __EROFS_BUF_INITIALIZER	((struct erofs_buf){ .page = NULL })
+
 #define ROOT_NID(sb)		((sb)->root_nid)
 
 #define erofs_blknr(addr)       ((addr) / EROFS_BLKSIZ)
Signature · 1 line
2.24.4
Reviewed-by on v2 1/5
#2 lore raw

Re: [PATCH v2 1/5] erofs: introduce meta buffer operations

Quoted 6 lines · Gao Xiang

In order to support subpage and folio for all uncompressed files, introduce meta buffer descriptors, which can be effectively stored on stack, in place of meta page operations.

This converts the uncompressed data path to meta buffers.

Quoted 203 lines
---
 fs/erofs/data.c     | 97 +++++++++++++++++++++++++++++++++++----------
 fs/erofs/internal.h | 13 ++++++
 2 files changed, 89 insertions(+), 21 deletions(-)
fs/erofs/data.c +76 -21
index 4f98c76ec043..6495e16a50a9 100644
@@ -22,6 +22,56 @@ struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr)
 	return page;
 }
 
+void erofs_unmap_metabuf(struct erofs_buf *buf)
+{
+	if (buf->kmap_type == EROFS_KMAP)
+		kunmap(buf->page);
+	else if (buf->kmap_type == EROFS_KMAP_ATOMIC)
+		kunmap_atomic(buf->base);
+	buf->kmap_type = EROFS_NO_KMAP;
+}
+
+void erofs_put_metabuf(struct erofs_buf *buf)
+{
+	if (!buf->page)
+		return;
+	erofs_unmap_metabuf(buf);
+	put_page(buf->page);
+	buf->page = NULL;
+}
+
+void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
+			erofs_blk_t blkaddr, enum erofs_kmap_type type)
+{
+	struct address_space *const mapping = sb->s_bdev->bd_inode->i_mapping;
+	erofs_off_t offset = blknr_to_addr(blkaddr);
+	pgoff_t index = offset >> PAGE_SHIFT;
+	struct page *page = buf->page;
+
+	if (!page || page->index != index) {
+		erofs_put_metabuf(buf);
+		page = read_cache_page_gfp(mapping, index,
+				mapping_gfp_constraint(mapping, ~__GFP_FS));
+		if (IS_ERR(page))
+			return page;
+		/* should already be PageUptodate, no need to lock page */
+		buf->page = page;
+	}
+	if (buf->kmap_type == EROFS_NO_KMAP) {
+		if (type == EROFS_KMAP)
+			buf->base = kmap(page);
+		else if (type == EROFS_KMAP_ATOMIC)
+			buf->base = kmap_atomic(page);
+		buf->kmap_type = type;
+	} else if (buf->kmap_type != type) {
+		DBG_BUGON(1);
+		return ERR_PTR(-EFAULT);
+	}
+	if (type == EROFS_NO_KMAP)
+		return NULL;
+	return buf->base + (offset & ~PAGE_MASK);
+}
+
 static int erofs_map_blocks_flatmode(struct inode *inode,
 				     struct erofs_map_blocks *map,
 				     int flags)
@@ -31,7 +81,7 @@ static int erofs_map_blocks_flatmode(struct inode *inode,
 	struct erofs_inode *vi = EROFS_I(inode);
 	bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE);
 
-	nblocks = DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
+	nblocks = DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ);
 	lastblk = nblocks - tailendpacking;
 
 	/* there is no hole in flatmode */
@@ -72,10 +122,11 @@ static int erofs_map_blocks(struct inode *inode,
 	struct super_block *sb = inode->i_sb;
 	struct erofs_inode *vi = EROFS_I(inode);
 	struct erofs_inode_chunk_index *idx;
-	struct page *page;
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 	u64 chunknr;
 	unsigned int unit;
 	erofs_off_t pos;
+	void *kaddr;
 	int err = 0;
 
 	trace_erofs_map_blocks_enter(inode, map, flags);
@@ -101,9 +152,9 @@ static int erofs_map_blocks(struct inode *inode,
 	pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize +
 		    vi->xattr_isize, unit) + unit * chunknr;
 
-	page = erofs_get_meta_page(inode->i_sb, erofs_blknr(pos));
-	if (IS_ERR(page)) {
-		err = PTR_ERR(page);
+	kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos), EROFS_KMAP);
+	if (IS_ERR(kaddr)) {
+		err = PTR_ERR(kaddr);
 		goto out;
 	}
 	map->m_la = chunknr << vi->chunkbits;
@@ -112,7 +163,7 @@ static int erofs_map_blocks(struct inode *inode,
 
 	/* handle block map */
 	if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) {
-		__le32 *blkaddr = page_address(page) + erofs_blkoff(pos);
+		__le32 *blkaddr = kaddr + erofs_blkoff(pos);
 
 		if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) {
 			map->m_flags = 0;
@@ -123,7 +174,7 @@ static int erofs_map_blocks(struct inode *inode,
 		goto out_unlock;
 	}
 	/* parse chunk indexes */
-	idx = page_address(page) + erofs_blkoff(pos);
+	idx = kaddr + erofs_blkoff(pos);
 	switch (le32_to_cpu(idx->blkaddr)) {
 	case EROFS_NULL_ADDR:
 		map->m_flags = 0;
@@ -136,8 +187,7 @@ static int erofs_map_blocks(struct inode *inode,
 		break;
 	}
 out_unlock:
-	unlock_page(page);
-	put_page(page);
+	erofs_put_metabuf(&buf);
 out:
 	if (!err)
 		map->m_llen = map->m_plen;
@@ -226,16 +276,16 @@ static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 	}
 
 	if (map.m_flags & EROFS_MAP_META) {
-		struct page *ipage;
+		void *ptr;
+		struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 
 		iomap->type = IOMAP_INLINE;
-		ipage = erofs_get_meta_page(inode->i_sb,
-					    erofs_blknr(mdev.m_pa));
-		if (IS_ERR(ipage))
-			return PTR_ERR(ipage);
-		iomap->inline_data = page_address(ipage) +
-					erofs_blkoff(mdev.m_pa);
-		iomap->private = ipage;
+		ptr = erofs_read_metabuf(&buf, inode->i_sb,
+					 erofs_blknr(mdev.m_pa), EROFS_KMAP);
+		if (IS_ERR(ptr))
+			return PTR_ERR(ptr);
+		iomap->inline_data = ptr + erofs_blkoff(mdev.m_pa);
+		iomap->private = buf.base;
 	} else {
 		iomap->type = IOMAP_MAPPED;
 		iomap->addr = mdev.m_pa;
@@ -246,12 +296,17 @@ static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length,
 		ssize_t written, unsigned int flags, struct iomap *iomap)
 {
-	struct page *ipage = iomap->private;
+	void *ptr = iomap->private;
+
+	if (ptr) {
+		struct erofs_buf buf = {
+			.page = kmap_to_page(ptr),
+			.base = ptr,
+			.kmap_type = EROFS_KMAP,
+		};
 
-	if (ipage) {
 		DBG_BUGON(iomap->type != IOMAP_INLINE);
-		unlock_page(ipage);
-		put_page(ipage);
+		erofs_put_metabuf(&buf);
 	} else {
 		DBG_BUGON(iomap->type == IOMAP_INLINE);
 	}
fs/erofs/internal.h +13 -0
index fca3747d97be..7053f1c4171d 100644
@@ -251,6 +251,19 @@ static inline int erofs_wait_on_workgroup_freezed(struct erofs_workgroup *grp)
 #error erofs cannot be used in this platform
 #endif
 
+enum erofs_kmap_type {
+	EROFS_NO_KMAP,		/* don't map the buffer */
+	EROFS_KMAP,		/* use kmap() to map the buffer */
+	EROFS_KMAP_ATOMIC,	/* use kmap_atomic() to map the buffer */
+};
+
+struct erofs_buf {
+	struct page *page;
+	void *base;
+	enum erofs_kmap_type kmap_type;
+};
+#define __EROFS_BUF_INITIALIZER	((struct erofs_buf){ .page = NULL })
+
 #define ROOT_NID(sb)		((sb)->root_nid)
 
 #define erofs_blknr(addr)       ((addr) / EROFS_BLKSIZ)
Signature · 1 line
2.24.4
maintainer Reviewed-by on v2 1/5
#3 lore raw

Re: [PATCH v2 1/5] erofs: introduce meta buffer operations

In reply to Gao Xiang
Quoted 17 lines · Gao Xiang

In order to support subpage and folio for all uncompressed files, introduce meta buffer descriptors, which can be effectively stored on stack, in place of meta page operations.

This converts the uncompressed data path to meta buffers.

---
  fs/erofs/data.c     | 97 +++++++++++++++++++++++++++++++++++----------
  fs/erofs/internal.h | 13 ++++++
  2 files changed, 89 insertions(+), 21 deletions(-)
fs/erofs/data.c +0 -0
index 4f98c76ec043..6495e16a50a9 100644

fs/erofs/data.c struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr) · line 30

Quoted 9 lines
fs/erofs/data.c +6 -0
@@ -22,6 +22,56 @@ struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr)
  	return page;
  }
 
+void erofs_unmap_metabuf(struct erofs_buf *buf)
+{
+	if (buf->kmap_type == EROFS_KMAP)
+		kunmap(buf->page);
+	else if (buf->kmap_type == EROFS_KMAP_ATOMIC)
+		kunmap_atomic(buf->base);

Once user calls this function, .base should be invalidated.

buf->base = NULL;

Otherwise, it looks good to me.

Thanks,

Quoted 181 lines
fs/erofs/data.c +70 -21
+	buf->kmap_type = EROFS_NO_KMAP;
+}
+
+void erofs_put_metabuf(struct erofs_buf *buf)
+{
+	if (!buf->page)
+		return;
+	erofs_unmap_metabuf(buf);
+	put_page(buf->page);
+	buf->page = NULL;
+}
+
+void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
+			erofs_blk_t blkaddr, enum erofs_kmap_type type)
+{
+	struct address_space *const mapping = sb->s_bdev->bd_inode->i_mapping;
+	erofs_off_t offset = blknr_to_addr(blkaddr);
+	pgoff_t index = offset >> PAGE_SHIFT;
+	struct page *page = buf->page;
+
+	if (!page || page->index != index) {
+		erofs_put_metabuf(buf);
+		page = read_cache_page_gfp(mapping, index,
+				mapping_gfp_constraint(mapping, ~__GFP_FS));
+		if (IS_ERR(page))
+			return page;
+		/* should already be PageUptodate, no need to lock page */
+		buf->page = page;
+	}
+	if (buf->kmap_type == EROFS_NO_KMAP) {
+		if (type == EROFS_KMAP)
+			buf->base = kmap(page);
+		else if (type == EROFS_KMAP_ATOMIC)
+			buf->base = kmap_atomic(page);
+		buf->kmap_type = type;
+	} else if (buf->kmap_type != type) {
+		DBG_BUGON(1);
+		return ERR_PTR(-EFAULT);
+	}
+	if (type == EROFS_NO_KMAP)
+		return NULL;
+	return buf->base + (offset & ~PAGE_MASK);
+}
+
  static int erofs_map_blocks_flatmode(struct inode *inode,
  				     struct erofs_map_blocks *map,
  				     int flags)
@@ -31,7 +81,7 @@ static int erofs_map_blocks_flatmode(struct inode *inode,
  	struct erofs_inode *vi = EROFS_I(inode);
  	bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE);
 
-	nblocks = DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
+	nblocks = DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ);
  	lastblk = nblocks - tailendpacking;
 
  	/* there is no hole in flatmode */
@@ -72,10 +122,11 @@ static int erofs_map_blocks(struct inode *inode,
  	struct super_block *sb = inode->i_sb;
  	struct erofs_inode *vi = EROFS_I(inode);
  	struct erofs_inode_chunk_index *idx;
-	struct page *page;
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
  	u64 chunknr;
  	unsigned int unit;
  	erofs_off_t pos;
+	void *kaddr;
  	int err = 0;
 
  	trace_erofs_map_blocks_enter(inode, map, flags);
@@ -101,9 +152,9 @@ static int erofs_map_blocks(struct inode *inode,
  	pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize +
  		    vi->xattr_isize, unit) + unit * chunknr;
 
-	page = erofs_get_meta_page(inode->i_sb, erofs_blknr(pos));
-	if (IS_ERR(page)) {
-		err = PTR_ERR(page);
+	kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos), EROFS_KMAP);
+	if (IS_ERR(kaddr)) {
+		err = PTR_ERR(kaddr);
  		goto out;
  	}
  	map->m_la = chunknr << vi->chunkbits;
@@ -112,7 +163,7 @@ static int erofs_map_blocks(struct inode *inode,
 
  	/* handle block map */
  	if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) {
-		__le32 *blkaddr = page_address(page) + erofs_blkoff(pos);
+		__le32 *blkaddr = kaddr + erofs_blkoff(pos);
 
  		if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) {
  			map->m_flags = 0;
@@ -123,7 +174,7 @@ static int erofs_map_blocks(struct inode *inode,
  		goto out_unlock;
  	}
  	/* parse chunk indexes */
-	idx = page_address(page) + erofs_blkoff(pos);
+	idx = kaddr + erofs_blkoff(pos);
  	switch (le32_to_cpu(idx->blkaddr)) {
  	case EROFS_NULL_ADDR:
  		map->m_flags = 0;
@@ -136,8 +187,7 @@ static int erofs_map_blocks(struct inode *inode,
  		break;
  	}
  out_unlock:
-	unlock_page(page);
-	put_page(page);
+	erofs_put_metabuf(&buf);
  out:
  	if (!err)
  		map->m_llen = map->m_plen;
@@ -226,16 +276,16 @@ static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
  	}
 
  	if (map.m_flags & EROFS_MAP_META) {
-		struct page *ipage;
+		void *ptr;
+		struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 
  		iomap->type = IOMAP_INLINE;
-		ipage = erofs_get_meta_page(inode->i_sb,
-					    erofs_blknr(mdev.m_pa));
-		if (IS_ERR(ipage))
-			return PTR_ERR(ipage);
-		iomap->inline_data = page_address(ipage) +
-					erofs_blkoff(mdev.m_pa);
-		iomap->private = ipage;
+		ptr = erofs_read_metabuf(&buf, inode->i_sb,
+					 erofs_blknr(mdev.m_pa), EROFS_KMAP);
+		if (IS_ERR(ptr))
+			return PTR_ERR(ptr);
+		iomap->inline_data = ptr + erofs_blkoff(mdev.m_pa);
+		iomap->private = buf.base;
  	} else {
  		iomap->type = IOMAP_MAPPED;
  		iomap->addr = mdev.m_pa;
@@ -246,12 +296,17 @@ static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
  static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length,
  		ssize_t written, unsigned int flags, struct iomap *iomap)
  {
-	struct page *ipage = iomap->private;
+	void *ptr = iomap->private;
+
+	if (ptr) {
+		struct erofs_buf buf = {
+			.page = kmap_to_page(ptr),
+			.base = ptr,
+			.kmap_type = EROFS_KMAP,
+		};
 
-	if (ipage) {
  		DBG_BUGON(iomap->type != IOMAP_INLINE);
-		unlock_page(ipage);
-		put_page(ipage);
+		erofs_put_metabuf(&buf);
  	} else {
  		DBG_BUGON(iomap->type == IOMAP_INLINE);
  	}
fs/erofs/internal.h +13 -0
index fca3747d97be..7053f1c4171d 100644
@@ -251,6 +251,19 @@ static inline int erofs_wait_on_workgroup_freezed(struct erofs_workgroup *grp)
  #error erofs cannot be used in this platform
  #endif
 
+enum erofs_kmap_type {
+	EROFS_NO_KMAP,		/* don't map the buffer */
+	EROFS_KMAP,		/* use kmap() to map the buffer */
+	EROFS_KMAP_ATOMIC,	/* use kmap_atomic() to map the buffer */
+};
+
+struct erofs_buf {
+	struct page *page;
+	void *base;
+	enum erofs_kmap_type kmap_type;
+};
+#define __EROFS_BUF_INITIALIZER	((struct erofs_buf){ .page = NULL })
+
  #define ROOT_NID(sb)		((sb)->root_nid)
 
  #define erofs_blknr(addr)       ((addr) / EROFS_BLKSIZ)
author
#4 lore raw

Re: [PATCH v2 1/5] erofs: introduce meta buffer operations

Hi Chao,

Quoted 29 lines · Chao Yu
Quoted 17 lines · Gao Xiang

In order to support subpage and folio for all uncompressed files, introduce meta buffer descriptors, which can be effectively stored on stack, in place of meta page operations.

This converts the uncompressed data path to meta buffers.

---
  fs/erofs/data.c     | 97 +++++++++++++++++++++++++++++++++++----------
  fs/erofs/internal.h | 13 ++++++
  2 files changed, 89 insertions(+), 21 deletions(-)
fs/erofs/data.c +0 -0
index 4f98c76ec043..6495e16a50a9 100644

fs/erofs/data.c struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr) · line 29

Quoted 8 lines
fs/erofs/data.c +6 -0
@@ -22,6 +22,56 @@ struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr)
  	return page;
  }
+void erofs_unmap_metabuf(struct erofs_buf *buf)
+{
+	if (buf->kmap_type == EROFS_KMAP)
+		kunmap(buf->page);
+	else if (buf->kmap_type == EROFS_KMAP_ATOMIC)
+		kunmap_atomic(buf->base);

Once user calls this function, .base should be invalidated.

Quoted 1 line

buf->base = NULL;

Thanks for your suggestion!

Since buf->kmap_type will be switched to EROFS_NO_KMAP, so buf->base is completely ignored (IOWs, it's unnecessary to reset buf->base..), however I'm fine to set buf->base = NULL here either.

Thanks,
Gao Xiang

Quoted 7 lines

Otherwise, it looks good to me.

Thanks,

author
#5 lore raw

PATCH v2 2/5 erofs: use meta buffers for inode operations

In reply to Gao Xiang

Get rid of old erofs_get_meta_page() within inode operations by using on-stack meta buffers in order to prepare subpage and folio features.

---
 fs/erofs/inode.c    | 68 +++++++++++++++++++++------------------------
 fs/erofs/internal.h |  3 ++
 2 files changed, 35 insertions(+), 36 deletions(-)
fs/erofs/inode.c +32 -36
index 2345f1de438e..ff62f84f47d3 100644
@@ -13,8 +13,8 @@
  * the inode payload page if it's an extended inode) in order to fill
  * inline data if possible.
  */
-static struct page *erofs_read_inode(struct inode *inode,
-				     unsigned int *ofs)
+static void *erofs_read_inode(struct erofs_buf *buf,
+			      struct inode *inode, unsigned int *ofs)
 {
 	struct super_block *sb = inode->i_sb;
 	struct erofs_sb_info *sbi = EROFS_SB(sb);
@@ -22,7 +22,7 @@ static struct page *erofs_read_inode(struct inode *inode,
 	const erofs_off_t inode_loc = iloc(sbi, vi->nid);
 
 	erofs_blk_t blkaddr, nblks = 0;
-	struct page *page;
+	void *kaddr;
 	struct erofs_inode_compact *dic;
 	struct erofs_inode_extended *die, *copied = NULL;
 	unsigned int ifmt;
@@ -34,14 +34,14 @@ static struct page *erofs_read_inode(struct inode *inode,
 	erofs_dbg("%s, reading inode nid %llu at %u of blkaddr %u",
 		  __func__, vi->nid, *ofs, blkaddr);
 
-	page = erofs_get_meta_page(sb, blkaddr);
-	if (IS_ERR(page)) {
+	kaddr = erofs_read_metabuf(buf, sb, blkaddr, EROFS_KMAP);
+	if (IS_ERR(kaddr)) {
 		erofs_err(sb, "failed to get inode (nid: %llu) page, err %ld",
-			  vi->nid, PTR_ERR(page));
-		return page;
+			  vi->nid, PTR_ERR(kaddr));
+		return kaddr;
 	}
 
-	dic = page_address(page) + *ofs;
+	dic = kaddr + *ofs;
 	ifmt = le16_to_cpu(dic->i_format);
 
 	if (ifmt & ~EROFS_I_ALL) {
@@ -62,12 +62,12 @@ static struct page *erofs_read_inode(struct inode *inode,
 	switch (erofs_inode_version(ifmt)) {
 	case EROFS_INODE_LAYOUT_EXTENDED:
 		vi->inode_isize = sizeof(struct erofs_inode_extended);
-		/* check if the inode acrosses page boundary */
-		if (*ofs + vi->inode_isize <= PAGE_SIZE) {
+		/* check if the extended inode acrosses block boundary */
+		if (*ofs + vi->inode_isize <= EROFS_BLKSIZ) {
 			*ofs += vi->inode_isize;
 			die = (struct erofs_inode_extended *)dic;
 		} else {
-			const unsigned int gotten = PAGE_SIZE - *ofs;
+			const unsigned int gotten = EROFS_BLKSIZ - *ofs;
 
 			copied = kmalloc(vi->inode_isize, GFP_NOFS);
 			if (!copied) {
@@ -75,18 +75,16 @@ static struct page *erofs_read_inode(struct inode *inode,
 				goto err_out;
 			}
 			memcpy(copied, dic, gotten);
-			unlock_page(page);
-			put_page(page);
-
-			page = erofs_get_meta_page(sb, blkaddr + 1);
-			if (IS_ERR(page)) {
-				erofs_err(sb, "failed to get inode payload page (nid: %llu), err %ld",
-					  vi->nid, PTR_ERR(page));
+			kaddr = erofs_read_metabuf(buf, sb, blkaddr + 1,
+						   EROFS_KMAP);
+			if (IS_ERR(kaddr)) {
+				erofs_err(sb, "failed to get inode payload block (nid: %llu), err %ld",
+					  vi->nid, PTR_ERR(kaddr));
 				kfree(copied);
-				return page;
+				return kaddr;
 			}
 			*ofs = vi->inode_isize - gotten;
-			memcpy((u8 *)copied + gotten, page_address(page), *ofs);
+			memcpy((u8 *)copied + gotten, kaddr, *ofs);
 			die = copied;
 		}
 		vi->xattr_isize = erofs_xattr_ibody_size(die->i_xattr_icount);
@@ -200,7 +198,7 @@ static struct page *erofs_read_inode(struct inode *inode,
 		inode->i_blocks = roundup(inode->i_size, EROFS_BLKSIZ) >> 9;
 	else
 		inode->i_blocks = nblks << LOG_SECTORS_PER_BLOCK;
-	return page;
+	return kaddr;
 
 bogusimode:
 	erofs_err(inode->i_sb, "bogus i_mode (%o) @ nid %llu",
@@ -209,12 +207,11 @@ static struct page *erofs_read_inode(struct inode *inode,
 err_out:
 	DBG_BUGON(1);
 	kfree(copied);
-	unlock_page(page);
-	put_page(page);
+	erofs_put_metabuf(buf);
 	return ERR_PTR(err);
 }
 
-static int erofs_fill_symlink(struct inode *inode, void *data,
+static int erofs_fill_symlink(struct inode *inode, void *kaddr,
 			      unsigned int m_pofs)
 {
 	struct erofs_inode *vi = EROFS_I(inode);
@@ -222,7 +219,7 @@ static int erofs_fill_symlink(struct inode *inode, void *data,
 
 	/* if it cannot be handled with fast symlink scheme */
 	if (vi->datalayout != EROFS_INODE_FLAT_INLINE ||
-	    inode->i_size >= PAGE_SIZE) {
+	    inode->i_size >= EROFS_BLKSIZ) {
 		inode->i_op = &erofs_symlink_iops;
 		return 0;
 	}
@@ -232,8 +229,8 @@ static int erofs_fill_symlink(struct inode *inode, void *data,
 		return -ENOMEM;
 
 	m_pofs += vi->xattr_isize;
-	/* inline symlink data shouldn't cross page boundary as well */
-	if (m_pofs + inode->i_size > PAGE_SIZE) {
+	/* inline symlink data shouldn't cross block boundary */
+	if (m_pofs + inode->i_size > EROFS_BLKSIZ) {
 		kfree(lnk);
 		erofs_err(inode->i_sb,
 			  "inline data cross block boundary @ nid %llu",
@@ -241,8 +238,7 @@ static int erofs_fill_symlink(struct inode *inode, void *data,
 		DBG_BUGON(1);
 		return -EFSCORRUPTED;
 	}
-
-	memcpy(lnk, data + m_pofs, inode->i_size);
+	memcpy(lnk, kaddr + m_pofs, inode->i_size);
 	lnk[inode->i_size] = '\0';
 
 	inode->i_link = lnk;
@@ -253,16 +249,17 @@ static int erofs_fill_symlink(struct inode *inode, void *data,
 static int erofs_fill_inode(struct inode *inode, int isdir)
 {
 	struct erofs_inode *vi = EROFS_I(inode);
-	struct page *page;
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
+	void *kaddr;
 	unsigned int ofs;
 	int err = 0;
 
 	trace_erofs_fill_inode(inode, isdir);
 
 	/* read inode base data from disk */
-	page = erofs_read_inode(inode, &ofs);
-	if (IS_ERR(page))
-		return PTR_ERR(page);
+	kaddr = erofs_read_inode(&buf, inode, &ofs);
+	if (IS_ERR(kaddr))
+		return PTR_ERR(kaddr);
 
 	/* setup the new inode */
 	switch (inode->i_mode & S_IFMT) {
@@ -278,7 +275,7 @@ static int erofs_fill_inode(struct inode *inode, int isdir)
 		inode->i_fop = &erofs_dir_fops;
 		break;
 	case S_IFLNK:
-		err = erofs_fill_symlink(inode, page_address(page), ofs);
+		err = erofs_fill_symlink(inode, kaddr, ofs);
 		if (err)
 			goto out_unlock;
 		inode_nohighmem(inode);
@@ -302,8 +299,7 @@ static int erofs_fill_inode(struct inode *inode, int isdir)
 	inode->i_mapping->a_ops = &erofs_raw_access_aops;
 
 out_unlock:
-	unlock_page(page);
-	put_page(page);
+	erofs_put_metabuf(&buf);
 	return err;
 }
 
fs/erofs/internal.h +3 -0
index 7053f1c4171d..f1e4eb3025f6 100644
@@ -475,6 +475,9 @@ struct erofs_map_dev {
 /* data.c */
 extern const struct file_operations erofs_file_fops;
 struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr);
+void erofs_put_metabuf(struct erofs_buf *buf);
+void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
+			 erofs_blk_t blkaddr, enum erofs_kmap_type type);
 int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *dev);
 int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 		 u64 start, u64 len);
Signature · 1 line
2.24.4
Reviewed-by on v2 2/5
#6 lore raw

Re: [PATCH v2 2/5] erofs: use meta buffers for inode operations

Quoted 4 lines · Gao Xiang

Get rid of old erofs_get_meta_page() within inode operations by using on-stack meta buffers in order to prepare subpage and folio features.

Quoted 204 lines
---
 fs/erofs/inode.c    | 68 +++++++++++++++++++++------------------------
 fs/erofs/internal.h |  3 ++
 2 files changed, 35 insertions(+), 36 deletions(-)
fs/erofs/inode.c +32 -36
index 2345f1de438e..ff62f84f47d3 100644
@@ -13,8 +13,8 @@
  * the inode payload page if it's an extended inode) in order to fill
  * inline data if possible.
  */
-static struct page *erofs_read_inode(struct inode *inode,
-				     unsigned int *ofs)
+static void *erofs_read_inode(struct erofs_buf *buf,
+			      struct inode *inode, unsigned int *ofs)
 {
 	struct super_block *sb = inode->i_sb;
 	struct erofs_sb_info *sbi = EROFS_SB(sb);
@@ -22,7 +22,7 @@ static struct page *erofs_read_inode(struct inode *inode,
 	const erofs_off_t inode_loc = iloc(sbi, vi->nid);
 
 	erofs_blk_t blkaddr, nblks = 0;
-	struct page *page;
+	void *kaddr;
 	struct erofs_inode_compact *dic;
 	struct erofs_inode_extended *die, *copied = NULL;
 	unsigned int ifmt;
@@ -34,14 +34,14 @@ static struct page *erofs_read_inode(struct inode *inode,
 	erofs_dbg("%s, reading inode nid %llu at %u of blkaddr %u",
 		  __func__, vi->nid, *ofs, blkaddr);
 
-	page = erofs_get_meta_page(sb, blkaddr);
-	if (IS_ERR(page)) {
+	kaddr = erofs_read_metabuf(buf, sb, blkaddr, EROFS_KMAP);
+	if (IS_ERR(kaddr)) {
 		erofs_err(sb, "failed to get inode (nid: %llu) page, err %ld",
-			  vi->nid, PTR_ERR(page));
-		return page;
+			  vi->nid, PTR_ERR(kaddr));
+		return kaddr;
 	}
 
-	dic = page_address(page) + *ofs;
+	dic = kaddr + *ofs;
 	ifmt = le16_to_cpu(dic->i_format);
 
 	if (ifmt & ~EROFS_I_ALL) {
@@ -62,12 +62,12 @@ static struct page *erofs_read_inode(struct inode *inode,
 	switch (erofs_inode_version(ifmt)) {
 	case EROFS_INODE_LAYOUT_EXTENDED:
 		vi->inode_isize = sizeof(struct erofs_inode_extended);
-		/* check if the inode acrosses page boundary */
-		if (*ofs + vi->inode_isize <= PAGE_SIZE) {
+		/* check if the extended inode acrosses block boundary */
+		if (*ofs + vi->inode_isize <= EROFS_BLKSIZ) {
 			*ofs += vi->inode_isize;
 			die = (struct erofs_inode_extended *)dic;
 		} else {
-			const unsigned int gotten = PAGE_SIZE - *ofs;
+			const unsigned int gotten = EROFS_BLKSIZ - *ofs;
 
 			copied = kmalloc(vi->inode_isize, GFP_NOFS);
 			if (!copied) {
@@ -75,18 +75,16 @@ static struct page *erofs_read_inode(struct inode *inode,
 				goto err_out;
 			}
 			memcpy(copied, dic, gotten);
-			unlock_page(page);
-			put_page(page);
-
-			page = erofs_get_meta_page(sb, blkaddr + 1);
-			if (IS_ERR(page)) {
-				erofs_err(sb, "failed to get inode payload page (nid: %llu), err %ld",
-					  vi->nid, PTR_ERR(page));
+			kaddr = erofs_read_metabuf(buf, sb, blkaddr + 1,
+						   EROFS_KMAP);
+			if (IS_ERR(kaddr)) {
+				erofs_err(sb, "failed to get inode payload block (nid: %llu), err %ld",
+					  vi->nid, PTR_ERR(kaddr));
 				kfree(copied);
-				return page;
+				return kaddr;
 			}
 			*ofs = vi->inode_isize - gotten;
-			memcpy((u8 *)copied + gotten, page_address(page), *ofs);
+			memcpy((u8 *)copied + gotten, kaddr, *ofs);
 			die = copied;
 		}
 		vi->xattr_isize = erofs_xattr_ibody_size(die->i_xattr_icount);
@@ -200,7 +198,7 @@ static struct page *erofs_read_inode(struct inode *inode,
 		inode->i_blocks = roundup(inode->i_size, EROFS_BLKSIZ) >> 9;
 	else
 		inode->i_blocks = nblks << LOG_SECTORS_PER_BLOCK;
-	return page;
+	return kaddr;
 
 bogusimode:
 	erofs_err(inode->i_sb, "bogus i_mode (%o) @ nid %llu",
@@ -209,12 +207,11 @@ static struct page *erofs_read_inode(struct inode *inode,
 err_out:
 	DBG_BUGON(1);
 	kfree(copied);
-	unlock_page(page);
-	put_page(page);
+	erofs_put_metabuf(buf);
 	return ERR_PTR(err);
 }
 
-static int erofs_fill_symlink(struct inode *inode, void *data,
+static int erofs_fill_symlink(struct inode *inode, void *kaddr,
 			      unsigned int m_pofs)
 {
 	struct erofs_inode *vi = EROFS_I(inode);
@@ -222,7 +219,7 @@ static int erofs_fill_symlink(struct inode *inode, void *data,
 
 	/* if it cannot be handled with fast symlink scheme */
 	if (vi->datalayout != EROFS_INODE_FLAT_INLINE ||
-	    inode->i_size >= PAGE_SIZE) {
+	    inode->i_size >= EROFS_BLKSIZ) {
 		inode->i_op = &erofs_symlink_iops;
 		return 0;
 	}
@@ -232,8 +229,8 @@ static int erofs_fill_symlink(struct inode *inode, void *data,
 		return -ENOMEM;
 
 	m_pofs += vi->xattr_isize;
-	/* inline symlink data shouldn't cross page boundary as well */
-	if (m_pofs + inode->i_size > PAGE_SIZE) {
+	/* inline symlink data shouldn't cross block boundary */
+	if (m_pofs + inode->i_size > EROFS_BLKSIZ) {
 		kfree(lnk);
 		erofs_err(inode->i_sb,
 			  "inline data cross block boundary @ nid %llu",
@@ -241,8 +238,7 @@ static int erofs_fill_symlink(struct inode *inode, void *data,
 		DBG_BUGON(1);
 		return -EFSCORRUPTED;
 	}
-
-	memcpy(lnk, data + m_pofs, inode->i_size);
+	memcpy(lnk, kaddr + m_pofs, inode->i_size);
 	lnk[inode->i_size] = '\0';
 
 	inode->i_link = lnk;
@@ -253,16 +249,17 @@ static int erofs_fill_symlink(struct inode *inode, void *data,
 static int erofs_fill_inode(struct inode *inode, int isdir)
 {
 	struct erofs_inode *vi = EROFS_I(inode);
-	struct page *page;
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
+	void *kaddr;
 	unsigned int ofs;
 	int err = 0;
 
 	trace_erofs_fill_inode(inode, isdir);
 
 	/* read inode base data from disk */
-	page = erofs_read_inode(inode, &ofs);
-	if (IS_ERR(page))
-		return PTR_ERR(page);
+	kaddr = erofs_read_inode(&buf, inode, &ofs);
+	if (IS_ERR(kaddr))
+		return PTR_ERR(kaddr);
 
 	/* setup the new inode */
 	switch (inode->i_mode & S_IFMT) {
@@ -278,7 +275,7 @@ static int erofs_fill_inode(struct inode *inode, int isdir)
 		inode->i_fop = &erofs_dir_fops;
 		break;
 	case S_IFLNK:
-		err = erofs_fill_symlink(inode, page_address(page), ofs);
+		err = erofs_fill_symlink(inode, kaddr, ofs);
 		if (err)
 			goto out_unlock;
 		inode_nohighmem(inode);
@@ -302,8 +299,7 @@ static int erofs_fill_inode(struct inode *inode, int isdir)
 	inode->i_mapping->a_ops = &erofs_raw_access_aops;
 
 out_unlock:
-	unlock_page(page);
-	put_page(page);
+	erofs_put_metabuf(&buf);
 	return err;
 }
 
fs/erofs/internal.h +3 -0
index 7053f1c4171d..f1e4eb3025f6 100644
@@ -475,6 +475,9 @@ struct erofs_map_dev {
 /* data.c */
 extern const struct file_operations erofs_file_fops;
 struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr);
+void erofs_put_metabuf(struct erofs_buf *buf);
+void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
+			 erofs_blk_t blkaddr, enum erofs_kmap_type type);
 int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *dev);
 int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 		 u64 start, u64 len);
Signature · 1 line
2.24.4
author
#8 lore raw

PATCH v2 3/5 erofs: use meta buffers for super operations

In reply to Gao Xiang

Get rid of old erofs_get_meta_page() within super operations by using on-stack meta buffers in order to prepare subpage and folio features.

---
 fs/erofs/super.c | 105 ++++++++++++-----------------------------------
 1 file changed, 26 insertions(+), 79 deletions(-)
fs/erofs/super.c +26 -79
index 0724ad5fd6cf..38305fa2969b 100644
@@ -2,6 +2,7 @@
 /*
  * Copyright (C) 2017-2018 HUAWEI, Inc.
  *             https://www.huawei.com/
+ * Copyright (C) 2021, Alibaba Cloud
  */
 #include <linux/module.h>
 #include <linux/buffer_head.h>
@@ -124,80 +125,48 @@ static bool check_layout_compatibility(struct super_block *sb,
 
 #ifdef CONFIG_EROFS_FS_ZIP
 /* read variable-sized metadata, offset will be aligned by 4-byte */
-static void *erofs_read_metadata(struct super_block *sb, struct page **pagep,
+static void *erofs_read_metadata(struct super_block *sb, struct erofs_buf *buf,
 				 erofs_off_t *offset, int *lengthp)
 {
-	struct page *page = *pagep;
 	u8 *buffer, *ptr;
 	int len, i, cnt;
-	erofs_blk_t blk;
 
 	*offset = round_up(*offset, 4);
-	blk = erofs_blknr(*offset);
+	ptr = erofs_read_metabuf(buf, sb, erofs_blknr(*offset), EROFS_KMAP);
+	if (IS_ERR(ptr))
+		return ptr;
 
-	if (!page || page->index != blk) {
-		if (page) {
-			unlock_page(page);
-			put_page(page);
-		}
-		page = erofs_get_meta_page(sb, blk);
-		if (IS_ERR(page))
-			goto err_nullpage;
-	}
-
-	ptr = kmap(page);
 	len = le16_to_cpu(*(__le16 *)&ptr[erofs_blkoff(*offset)]);
 	if (!len)
 		len = U16_MAX + 1;
 	buffer = kmalloc(len, GFP_KERNEL);
-	if (!buffer) {
-		buffer = ERR_PTR(-ENOMEM);
-		goto out;
-	}
+	if (!buffer)
+		return ERR_PTR(-ENOMEM);
 	*offset += sizeof(__le16);
 	*lengthp = len;
 
 	for (i = 0; i < len; i += cnt) {
 		cnt = min(EROFS_BLKSIZ - (int)erofs_blkoff(*offset), len - i);
-		blk = erofs_blknr(*offset);
-
-		if (!page || page->index != blk) {
-			if (page) {
-				kunmap(page);
-				unlock_page(page);
-				put_page(page);
-			}
-			page = erofs_get_meta_page(sb, blk);
-			if (IS_ERR(page)) {
-				kfree(buffer);
-				goto err_nullpage;
-			}
-			ptr = kmap(page);
-		}
+		ptr = erofs_read_metabuf(buf, sb, erofs_blknr(*offset),
+					 EROFS_KMAP);
+		if (IS_ERR(ptr))
+			return ptr;
 		memcpy(buffer + i, ptr + erofs_blkoff(*offset), cnt);
 		*offset += cnt;
 	}
-out:
-	kunmap(page);
-	*pagep = page;
 	return buffer;
-err_nullpage:
-	*pagep = NULL;
-	return page;
 }
 
 static int erofs_load_compr_cfgs(struct super_block *sb,
 				 struct erofs_super_block *dsb)
 {
-	struct erofs_sb_info *sbi;
-	struct page *page;
+	struct erofs_sb_info *sbi = EROFS_SB(sb);
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 	unsigned int algs, alg;
 	erofs_off_t offset;
-	int size, ret;
+	int size, ret = 0;
 
-	sbi = EROFS_SB(sb);
 	sbi->available_compr_algs = le16_to_cpu(dsb->u1.available_compr_algs);
-
 	if (sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS) {
 		erofs_err(sb, "try to load compressed fs with unsupported algorithms %x",
 			  sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS);
@@ -205,21 +174,16 @@ static int erofs_load_compr_cfgs(struct super_block *sb,
 	}
 
 	offset = EROFS_SUPER_OFFSET + sbi->sb_size;
-	page = NULL;
 	alg = 0;
-	ret = 0;
-
 	for (algs = sbi->available_compr_algs; algs; algs >>= 1, ++alg) {
 		void *data;
 
 		if (!(algs & 1))
 			continue;
 
-		data = erofs_read_metadata(sb, &page, &offset, &size);
-		if (IS_ERR(data)) {
-			ret = PTR_ERR(data);
-			goto err;
-		}
+		data = erofs_read_metadata(sb, &buf, &offset, &size);
+		if (IS_ERR(data))
+			return PTR_ERR(data);
 
 		switch (alg) {
 		case Z_EROFS_COMPRESSION_LZ4:
@@ -234,13 +198,9 @@ static int erofs_load_compr_cfgs(struct super_block *sb,
 		}
 		kfree(data);
 		if (ret)
-			goto err;
-	}
-err:
-	if (page) {
-		unlock_page(page);
-		put_page(page);
+			break;
 	}
+	erofs_put_metabuf(&buf);
 	return ret;
 }
 #else
@@ -261,7 +221,7 @@ static int erofs_init_devices(struct super_block *sb,
 	struct erofs_sb_info *sbi = EROFS_SB(sb);
 	unsigned int ondisk_extradevs;
 	erofs_off_t pos;
-	struct page *page = NULL;
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 	struct erofs_device_info *dif;
 	struct erofs_deviceslot *dis;
 	void *ptr;
@@ -285,22 +245,13 @@ static int erofs_init_devices(struct super_block *sb,
 	pos = le16_to_cpu(dsb->devt_slotoff) * EROFS_DEVT_SLOT_SIZE;
 	down_read(&sbi->devs->rwsem);
 	idr_for_each_entry(&sbi->devs->tree, dif, id) {
-		erofs_blk_t blk = erofs_blknr(pos);
 		struct block_device *bdev;
 
-		if (!page || page->index != blk) {
-			if (page) {
-				kunmap(page);
-				unlock_page(page);
-				put_page(page);
-			}
-
-			page = erofs_get_meta_page(sb, blk);
-			if (IS_ERR(page)) {
-				up_read(&sbi->devs->rwsem);
-				return PTR_ERR(page);
-			}
-			ptr = kmap(page);
+		ptr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos),
+					 EROFS_KMAP);
+		if (IS_ERR(ptr)) {
+			up_read(&sbi->devs->rwsem);
+			return PTR_ERR(ptr);
 		}
 		dis = ptr + erofs_blkoff(pos);
 
@@ -320,11 +271,7 @@ static int erofs_init_devices(struct super_block *sb,
 	}
 err_out:
 	up_read(&sbi->devs->rwsem);
-	if (page) {
-		kunmap(page);
-		unlock_page(page);
-		put_page(page);
-	}
+	erofs_put_metabuf(&buf);
 	return err;
 }
 
Signature · 1 line
2.24.4
#9 lore raw

Re: [PATCH v2 3/5] erofs: use meta buffers for super operations

Quoted 75 lines · Gao Xiang

Get rid of old erofs_get_meta_page() within super operations by using on-stack meta buffers in order to prepare subpage and folio features.

---
 fs/erofs/super.c | 105 ++++++++++++-----------------------------------
 1 file changed, 26 insertions(+), 79 deletions(-)
fs/erofs/super.c +7 -27
index 0724ad5fd6cf..38305fa2969b 100644
@@ -2,6 +2,7 @@
 /*
  * Copyright (C) 2017-2018 HUAWEI, Inc.
  *             https://www.huawei.com/
+ * Copyright (C) 2021, Alibaba Cloud
  */
 #include <linux/module.h>
 #include <linux/buffer_head.h>
@@ -124,80 +125,48 @@ static bool check_layout_compatibility(struct super_block *sb,
 
 #ifdef CONFIG_EROFS_FS_ZIP
 /* read variable-sized metadata, offset will be aligned by 4-byte */
-static void *erofs_read_metadata(struct super_block *sb, struct page **pagep,
+static void *erofs_read_metadata(struct super_block *sb, struct erofs_buf *buf,
 				 erofs_off_t *offset, int *lengthp)
 {
-	struct page *page = *pagep;
 	u8 *buffer, *ptr;
 	int len, i, cnt;
-	erofs_blk_t blk;
 
 	*offset = round_up(*offset, 4);
-	blk = erofs_blknr(*offset);
+	ptr = erofs_read_metabuf(buf, sb, erofs_blknr(*offset), EROFS_KMAP);
+	if (IS_ERR(ptr))
+		return ptr;
 
-	if (!page || page->index != blk) {
-		if (page) {
-			unlock_page(page);
-			put_page(page);
-		}
-		page = erofs_get_meta_page(sb, blk);
-		if (IS_ERR(page))
-			goto err_nullpage;
-	}
-
-	ptr = kmap(page);
 	len = le16_to_cpu(*(__le16 *)&ptr[erofs_blkoff(*offset)]);
 	if (!len)
 		len = U16_MAX + 1;
 	buffer = kmalloc(len, GFP_KERNEL);
-	if (!buffer) {
-		buffer = ERR_PTR(-ENOMEM);
-		goto out;
-	}
+	if (!buffer)
+		return ERR_PTR(-ENOMEM);
 	*offset += sizeof(__le16);
 	*lengthp = len;
 
 	for (i = 0; i < len; i += cnt) {
 		cnt = min(EROFS_BLKSIZ - (int)erofs_blkoff(*offset), len - i);
-		blk = erofs_blknr(*offset);
-
-		if (!page || page->index != blk) {
-			if (page) {
-				kunmap(page);
-				unlock_page(page);
-				put_page(page);
-			}

fs/erofs/super.c static bool check_layout_compatibility(struct super_block *sb, · line 152

Quoted 10 lines
fs/erofs/super.c +3 -7
-			page = erofs_get_meta_page(sb, blk);
-			if (IS_ERR(page)) {
-				kfree(buffer);
-				goto err_nullpage;
-			}
-			ptr = kmap(page);
-		}
+		ptr = erofs_read_metabuf(buf, sb, erofs_blknr(*offset),
+					 EROFS_KMAP);
+		if (IS_ERR(ptr))

kfree(buffer) is missing.

Quoted 41 lines
fs/erofs/super.c +4 -14
+			return ptr;
 		memcpy(buffer + i, ptr + erofs_blkoff(*offset), cnt);
 		*offset += cnt;
 	}
-out:
-	kunmap(page);
-	*pagep = page;
 	return buffer;
-err_nullpage:
-	*pagep = NULL;
-	return page;
 }
 
 static int erofs_load_compr_cfgs(struct super_block *sb,
 				 struct erofs_super_block *dsb)
 {
-	struct erofs_sb_info *sbi;
-	struct page *page;
+	struct erofs_sb_info *sbi = EROFS_SB(sb);
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 	unsigned int algs, alg;
 	erofs_off_t offset;
-	int size, ret;
+	int size, ret = 0;
 
-	sbi = EROFS_SB(sb);
 	sbi->available_compr_algs = le16_to_cpu(dsb->u1.available_compr_algs);
-
 	if (sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS) {
 		erofs_err(sb, "try to load compressed fs with unsupported algorithms %x",
 			  sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS);
@@ -205,21 +174,16 @@ static int erofs_load_compr_cfgs(struct super_block *sb,
 	}
 
 	offset = EROFS_SUPER_OFFSET + sbi->sb_size;
-	page = NULL;
 	alg = 0;
-	ret = 0;
-
 	for (algs = sbi->available_compr_algs; algs; algs >>= 1, ++alg) {
 		void *data;

fs/erofs/super.c static int erofs_load_compr_cfgs(struct super_block *sb, · line 184

Quoted 10 lines
fs/erofs/super.c +1 -5
 
 		if (!(algs & 1))
 			continue;
 
-		data = erofs_read_metadata(sb, &page, &offset, &size);
-		if (IS_ERR(data)) {
-			ret = PTR_ERR(data);
-			goto err;
-		}
+		data = erofs_read_metadata(sb, &buf, &offset, &size);

if PTR_ERR(data) is -ENOMEM, buf needs to be put according to the current design.

Quoted 43 lines
fs/erofs/super.c +5 -14
+		if (IS_ERR(data))
+			return PTR_ERR(data);
 
 		switch (alg) {
 		case Z_EROFS_COMPRESSION_LZ4:
@@ -234,13 +198,9 @@ static int erofs_load_compr_cfgs(struct super_block *sb,
 		}
 		kfree(data);
 		if (ret)
-			goto err;
-	}
-err:
-	if (page) {
-		unlock_page(page);
-		put_page(page);
+			break;
 	}
+	erofs_put_metabuf(&buf);
 	return ret;
 }
 #else
@@ -261,7 +221,7 @@ static int erofs_init_devices(struct super_block *sb,
 	struct erofs_sb_info *sbi = EROFS_SB(sb);
 	unsigned int ondisk_extradevs;
 	erofs_off_t pos;
-	struct page *page = NULL;
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 	struct erofs_device_info *dif;
 	struct erofs_deviceslot *dis;
 	void *ptr;
@@ -285,22 +245,13 @@ static int erofs_init_devices(struct super_block *sb,
 	pos = le16_to_cpu(dsb->devt_slotoff) * EROFS_DEVT_SLOT_SIZE;
 	down_read(&sbi->devs->rwsem);
 	idr_for_each_entry(&sbi->devs->tree, dif, id) {
-		erofs_blk_t blk = erofs_blknr(pos);
 		struct block_device *bdev;
 
-		if (!page || page->index != blk) {
-			if (page) {
-				kunmap(page);
-				unlock_page(page);
-				put_page(page);
-			}

fs/erofs/super.c static int erofs_init_devices(struct super_block *sb, · line 252

Quoted 10 lines
fs/erofs/super.c +3 -7
-
-			page = erofs_get_meta_page(sb, blk);
-			if (IS_ERR(page)) {
-				up_read(&sbi->devs->rwsem);
-				return PTR_ERR(page);
-			}
-			ptr = kmap(page);
+		ptr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos),
+					 EROFS_KMAP);
+		if (IS_ERR(ptr)) {

Ditto.

Thanks,
liubo

Quoted 20 lines
fs/erofs/super.c +3 -5
+			up_read(&sbi->devs->rwsem);
+			return PTR_ERR(ptr);
 		}
 		dis = ptr + erofs_blkoff(pos);
 
@@ -320,11 +271,7 @@ static int erofs_init_devices(struct super_block *sb,
 	}
 err_out:
 	up_read(&sbi->devs->rwsem);
-	if (page) {
-		kunmap(page);
-		unlock_page(page);
-		put_page(page);
-	}
+	erofs_put_metabuf(&buf);
 	return err;
 }
 
Signature · 1 line
2.24.4
author
#10 lore raw

Re: [PATCH v2 3/5] erofs: use meta buffers for super operations

Hi Bo,

Quoted 1 line · Liu Bo

On Sun, Jan 02, 2022 at 12:00:15PM +0800, Gao Xiang wrote:

...

Quoted 110 lines

Review

Quoted 5 lines
(quoted diff) +3 -2
-			ptr = kmap(page);
-		}
+		ptr = erofs_read_metabuf(buf, sb, erofs_blknr(*offset),
+					 EROFS_KMAP);
+		if (IS_ERR(ptr))

kfree(buffer) is missing.

Quoted 41 lines
(quoted diff) +4 -14
+			return ptr;
 		memcpy(buffer + i, ptr + erofs_blkoff(*offset), cnt);
 		*offset += cnt;
 	}
-out:
-	kunmap(page);
-	*pagep = page;
 	return buffer;
-err_nullpage:
-	*pagep = NULL;
-	return page;
 }
 
 static int erofs_load_compr_cfgs(struct super_block *sb,
 				 struct erofs_super_block *dsb)
 {
-	struct erofs_sb_info *sbi;
-	struct page *page;
+	struct erofs_sb_info *sbi = EROFS_SB(sb);
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 	unsigned int algs, alg;
 	erofs_off_t offset;
-	int size, ret;
+	int size, ret = 0;
 
-	sbi = EROFS_SB(sb);
 	sbi->available_compr_algs = le16_to_cpu(dsb->u1.available_compr_algs);
-
 	if (sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS) {
 		erofs_err(sb, "try to load compressed fs with unsupported algorithms %x",
 			  sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS);
@@ -205,21 +174,16 @@ static int erofs_load_compr_cfgs(struct super_block *sb,
 	}
 
 	offset = EROFS_SUPER_OFFSET + sbi->sb_size;
-	page = NULL;
 	alg = 0;
-	ret = 0;
-
 	for (algs = sbi->available_compr_algs; algs; algs >>= 1, ++alg) {
 		void *data;

Review static int erofs_load_compr_cfgs(struct super_block *sb, · line 184

Quoted 10 lines
(quoted diff) +1 -5
 
 		if (!(algs & 1))
 			continue;
 
-		data = erofs_read_metadata(sb, &page, &offset, &size);
-		if (IS_ERR(data)) {
-			ret = PTR_ERR(data);
-			goto err;
-		}
+		data = erofs_read_metadata(sb, &buf, &offset, &size);

if PTR_ERR(data) is -ENOMEM, buf needs to be put according to the current design.

Quoted 47 lines
(quoted diff) +5 -18
+		if (IS_ERR(data))
+			return PTR_ERR(data);
 
 		switch (alg) {
 		case Z_EROFS_COMPRESSION_LZ4:
@@ -234,13 +198,9 @@ static int erofs_load_compr_cfgs(struct super_block *sb,
 		}
 		kfree(data);
 		if (ret)
-			goto err;
-	}
-err:
-	if (page) {
-		unlock_page(page);
-		put_page(page);
+			break;
 	}
+	erofs_put_metabuf(&buf);
 	return ret;
 }
 #else
@@ -261,7 +221,7 @@ static int erofs_init_devices(struct super_block *sb,
 	struct erofs_sb_info *sbi = EROFS_SB(sb);
 	unsigned int ondisk_extradevs;
 	erofs_off_t pos;
-	struct page *page = NULL;
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 	struct erofs_device_info *dif;
 	struct erofs_deviceslot *dis;
 	void *ptr;
@@ -285,22 +245,13 @@ static int erofs_init_devices(struct super_block *sb,
 	pos = le16_to_cpu(dsb->devt_slotoff) * EROFS_DEVT_SLOT_SIZE;
 	down_read(&sbi->devs->rwsem);
 	idr_for_each_entry(&sbi->devs->tree, dif, id) {
-		erofs_blk_t blk = erofs_blknr(pos);
 		struct block_device *bdev;
 
-		if (!page || page->index != blk) {
-			if (page) {
-				kunmap(page);
-				unlock_page(page);
-				put_page(page);
-			}
-
-			page = erofs_get_meta_page(sb, blk);
-			if (IS_ERR(page)) {
-				up_read(&sbi->devs->rwsem);
Quoted 9 lines

Review static int erofs_init_devices(struct super_block *sb, · line 252

Quoted 6 lines
(quoted diff) +3 -3
-				return PTR_ERR(page);
-			}
-			ptr = kmap(page);
+		ptr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos),
+					 EROFS_KMAP);
+		if (IS_ERR(ptr)) {

Ditto.

Thanks for catching these! I've fixed these as follows, please help check out again!

Thanks,
Gao Xiang

fs/erofs/super.c +10 -6
index 38305fa2969b..9e076303bb24 100644
@@ -149,8 +149,10 @@ static void *erofs_read_metadata(struct super_block *sb, struct erofs_buf *buf,
 		cnt = min(EROFS_BLKSIZ - (int)erofs_blkoff(*offset), len - i);
 		ptr = erofs_read_metabuf(buf, sb, erofs_blknr(*offset),
 					 EROFS_KMAP);
-		if (IS_ERR(ptr))
+		if (IS_ERR(ptr)) {
+			kfree(buffer);
 			return ptr;
+		}
 		memcpy(buffer + i, ptr + erofs_blkoff(*offset), cnt);
 		*offset += cnt;
 	}
@@ -182,8 +184,10 @@ static int erofs_load_compr_cfgs(struct super_block *sb,
 			continue;
 
 		data = erofs_read_metadata(sb, &buf, &offset, &size);
-		if (IS_ERR(data))
-			return PTR_ERR(data);
+		if (IS_ERR(data)) {
+			ret = PTR_ERR(data);
+			break;
+		}
 
 		switch (alg) {
 		case Z_EROFS_COMPRESSION_LZ4:
@@ -250,8 +254,8 @@ static int erofs_init_devices(struct super_block *sb,
 		ptr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos),
 					 EROFS_KMAP);
 		if (IS_ERR(ptr)) {
-			up_read(&sbi->devs->rwsem);
-			return PTR_ERR(ptr);
+			err = PTR_ERR(ptr);
+			break;
 		}
 		dis = ptr + erofs_blkoff(pos);
 
@@ -260,7 +264,7 @@ static int erofs_init_devices(struct super_block *sb,
 					  sb->s_type);
 		if (IS_ERR(bdev)) {
 			err = PTR_ERR(bdev);
-			goto err_out;
+			break;
 		}
 		dif->bdev = bdev;
 		dif->dax_dev = fs_dax_get_by_bdev(bdev);
author
#11 lore raw

PATCH v3 3/5 erofs: use meta buffers for super operations

In reply to Gao Xiang

Get rid of old erofs_get_meta_page() within super operations by using on-stack meta buffers in order to prepare subpage and folio features.

---
changes since v2:
- add a missing kfree(buffer);
- forget to put_metabuf in erofs_load_compr_cfgs;
- fotget to put_metabuf in erofs_init_devices.

 fs/erofs/super.c | 104 ++++++++++++-----------------------------------
 1 file changed, 27 insertions(+), 77 deletions(-)
fs/erofs/super.c +27 -77
index 0724ad5fd6cf..5c137647fa8a 100644
@@ -2,6 +2,7 @@
 /*
  * Copyright (C) 2017-2018 HUAWEI, Inc.
  *             https://www.huawei.com/
+ * Copyright (C) 2021, Alibaba Cloud
  */
 #include <linux/module.h>
 #include <linux/buffer_head.h>
@@ -124,80 +125,50 @@ static bool check_layout_compatibility(struct super_block *sb,
 
 #ifdef CONFIG_EROFS_FS_ZIP
 /* read variable-sized metadata, offset will be aligned by 4-byte */
-static void *erofs_read_metadata(struct super_block *sb, struct page **pagep,
+static void *erofs_read_metadata(struct super_block *sb, struct erofs_buf *buf,
 				 erofs_off_t *offset, int *lengthp)
 {
-	struct page *page = *pagep;
 	u8 *buffer, *ptr;
 	int len, i, cnt;
-	erofs_blk_t blk;
 
 	*offset = round_up(*offset, 4);
-	blk = erofs_blknr(*offset);
+	ptr = erofs_read_metabuf(buf, sb, erofs_blknr(*offset), EROFS_KMAP);
+	if (IS_ERR(ptr))
+		return ptr;
 
-	if (!page || page->index != blk) {
-		if (page) {
-			unlock_page(page);
-			put_page(page);
-		}
-		page = erofs_get_meta_page(sb, blk);
-		if (IS_ERR(page))
-			goto err_nullpage;
-	}
-
-	ptr = kmap(page);
 	len = le16_to_cpu(*(__le16 *)&ptr[erofs_blkoff(*offset)]);
 	if (!len)
 		len = U16_MAX + 1;
 	buffer = kmalloc(len, GFP_KERNEL);
-	if (!buffer) {
-		buffer = ERR_PTR(-ENOMEM);
-		goto out;
-	}
+	if (!buffer)
+		return ERR_PTR(-ENOMEM);
 	*offset += sizeof(__le16);
 	*lengthp = len;
 
 	for (i = 0; i < len; i += cnt) {
 		cnt = min(EROFS_BLKSIZ - (int)erofs_blkoff(*offset), len - i);
-		blk = erofs_blknr(*offset);
-
-		if (!page || page->index != blk) {
-			if (page) {
-				kunmap(page);
-				unlock_page(page);
-				put_page(page);
-			}
-			page = erofs_get_meta_page(sb, blk);
-			if (IS_ERR(page)) {
-				kfree(buffer);
-				goto err_nullpage;
-			}
-			ptr = kmap(page);
+		ptr = erofs_read_metabuf(buf, sb, erofs_blknr(*offset),
+					 EROFS_KMAP);
+		if (IS_ERR(ptr)) {
+			kfree(buffer);
+			return ptr;
 		}
 		memcpy(buffer + i, ptr + erofs_blkoff(*offset), cnt);
 		*offset += cnt;
 	}
-out:
-	kunmap(page);
-	*pagep = page;
 	return buffer;
-err_nullpage:
-	*pagep = NULL;
-	return page;
 }
 
 static int erofs_load_compr_cfgs(struct super_block *sb,
 				 struct erofs_super_block *dsb)
 {
-	struct erofs_sb_info *sbi;
-	struct page *page;
+	struct erofs_sb_info *sbi = EROFS_SB(sb);
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 	unsigned int algs, alg;
 	erofs_off_t offset;
-	int size, ret;
+	int size, ret = 0;
 
-	sbi = EROFS_SB(sb);
 	sbi->available_compr_algs = le16_to_cpu(dsb->u1.available_compr_algs);
-
 	if (sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS) {
 		erofs_err(sb, "try to load compressed fs with unsupported algorithms %x",
 			  sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS);
@@ -205,20 +176,17 @@ static int erofs_load_compr_cfgs(struct super_block *sb,
 	}
 
 	offset = EROFS_SUPER_OFFSET + sbi->sb_size;
-	page = NULL;
 	alg = 0;
-	ret = 0;
-
 	for (algs = sbi->available_compr_algs; algs; algs >>= 1, ++alg) {
 		void *data;
 
 		if (!(algs & 1))
 			continue;
 
-		data = erofs_read_metadata(sb, &page, &offset, &size);
+		data = erofs_read_metadata(sb, &buf, &offset, &size);
 		if (IS_ERR(data)) {
 			ret = PTR_ERR(data);
-			goto err;
+			break;
 		}
 
 		switch (alg) {
@@ -234,13 +202,9 @@ static int erofs_load_compr_cfgs(struct super_block *sb,
 		}
 		kfree(data);
 		if (ret)
-			goto err;
-	}
-err:
-	if (page) {
-		unlock_page(page);
-		put_page(page);
+			break;
 	}
+	erofs_put_metabuf(&buf);
 	return ret;
 }
 #else
@@ -261,7 +225,7 @@ static int erofs_init_devices(struct super_block *sb,
 	struct erofs_sb_info *sbi = EROFS_SB(sb);
 	unsigned int ondisk_extradevs;
 	erofs_off_t pos;
-	struct page *page = NULL;
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 	struct erofs_device_info *dif;
 	struct erofs_deviceslot *dis;
 	void *ptr;
@@ -285,22 +249,13 @@ static int erofs_init_devices(struct super_block *sb,
 	pos = le16_to_cpu(dsb->devt_slotoff) * EROFS_DEVT_SLOT_SIZE;
 	down_read(&sbi->devs->rwsem);
 	idr_for_each_entry(&sbi->devs->tree, dif, id) {
-		erofs_blk_t blk = erofs_blknr(pos);
 		struct block_device *bdev;
 
-		if (!page || page->index != blk) {
-			if (page) {
-				kunmap(page);
-				unlock_page(page);
-				put_page(page);
-			}
-
-			page = erofs_get_meta_page(sb, blk);
-			if (IS_ERR(page)) {
-				up_read(&sbi->devs->rwsem);
-				return PTR_ERR(page);
-			}
-			ptr = kmap(page);
+		ptr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos),
+					 EROFS_KMAP);
+		if (IS_ERR(ptr)) {
+			err = PTR_ERR(ptr);
+			break;
 		}
 		dis = ptr + erofs_blkoff(pos);
 
@@ -309,7 +264,7 @@ static int erofs_init_devices(struct super_block *sb,
 					  sb->s_type);
 		if (IS_ERR(bdev)) {
 			err = PTR_ERR(bdev);
-			goto err_out;
+			break;
 		}
 		dif->bdev = bdev;
 		dif->dax_dev = fs_dax_get_by_bdev(bdev);
@@ -318,13 +273,8 @@ static int erofs_init_devices(struct super_block *sb,
 		sbi->total_blocks += dif->blocks;
 		pos += EROFS_DEVT_SLOT_SIZE;
 	}
-err_out:
 	up_read(&sbi->devs->rwsem);
-	if (page) {
-		kunmap(page);
-		unlock_page(page);
-		put_page(page);
-	}
+	erofs_put_metabuf(&buf);
 	return err;
 }
 
Signature · 1 line
2.24.4
Reviewed-by on v3 3/5
#12 lore raw

Re: [PATCH v3 3/5] erofs: use meta buffers for super operations

Quoted 4 lines · Gao Xiang

Get rid of old erofs_get_meta_page() within super operations by using on-stack meta buffers in order to prepare subpage and folio features.

Quoted 220 lines

---
changes since v2:
- add a missing kfree(buffer);
- forget to put_metabuf in erofs_load_compr_cfgs;
- fotget to put_metabuf in erofs_init_devices.

 fs/erofs/super.c | 104 ++++++++++++-----------------------------------
 1 file changed, 27 insertions(+), 77 deletions(-)
fs/erofs/super.c +27 -77
index 0724ad5fd6cf..5c137647fa8a 100644
@@ -2,6 +2,7 @@
 /*
  * Copyright (C) 2017-2018 HUAWEI, Inc.
  *             https://www.huawei.com/
+ * Copyright (C) 2021, Alibaba Cloud
  */
 #include <linux/module.h>
 #include <linux/buffer_head.h>
@@ -124,80 +125,50 @@ static bool check_layout_compatibility(struct super_block *sb,
 
 #ifdef CONFIG_EROFS_FS_ZIP
 /* read variable-sized metadata, offset will be aligned by 4-byte */
-static void *erofs_read_metadata(struct super_block *sb, struct page **pagep,
+static void *erofs_read_metadata(struct super_block *sb, struct erofs_buf *buf,
 				 erofs_off_t *offset, int *lengthp)
 {
-	struct page *page = *pagep;
 	u8 *buffer, *ptr;
 	int len, i, cnt;
-	erofs_blk_t blk;
 
 	*offset = round_up(*offset, 4);
-	blk = erofs_blknr(*offset);
+	ptr = erofs_read_metabuf(buf, sb, erofs_blknr(*offset), EROFS_KMAP);
+	if (IS_ERR(ptr))
+		return ptr;
 
-	if (!page || page->index != blk) {
-		if (page) {
-			unlock_page(page);
-			put_page(page);
-		}
-		page = erofs_get_meta_page(sb, blk);
-		if (IS_ERR(page))
-			goto err_nullpage;
-	}
-
-	ptr = kmap(page);
 	len = le16_to_cpu(*(__le16 *)&ptr[erofs_blkoff(*offset)]);
 	if (!len)
 		len = U16_MAX + 1;
 	buffer = kmalloc(len, GFP_KERNEL);
-	if (!buffer) {
-		buffer = ERR_PTR(-ENOMEM);
-		goto out;
-	}
+	if (!buffer)
+		return ERR_PTR(-ENOMEM);
 	*offset += sizeof(__le16);
 	*lengthp = len;
 
 	for (i = 0; i < len; i += cnt) {
 		cnt = min(EROFS_BLKSIZ - (int)erofs_blkoff(*offset), len - i);
-		blk = erofs_blknr(*offset);
-
-		if (!page || page->index != blk) {
-			if (page) {
-				kunmap(page);
-				unlock_page(page);
-				put_page(page);
-			}
-			page = erofs_get_meta_page(sb, blk);
-			if (IS_ERR(page)) {
-				kfree(buffer);
-				goto err_nullpage;
-			}
-			ptr = kmap(page);
+		ptr = erofs_read_metabuf(buf, sb, erofs_blknr(*offset),
+					 EROFS_KMAP);
+		if (IS_ERR(ptr)) {
+			kfree(buffer);
+			return ptr;
 		}
 		memcpy(buffer + i, ptr + erofs_blkoff(*offset), cnt);
 		*offset += cnt;
 	}
-out:
-	kunmap(page);
-	*pagep = page;
 	return buffer;
-err_nullpage:
-	*pagep = NULL;
-	return page;
 }
 
 static int erofs_load_compr_cfgs(struct super_block *sb,
 				 struct erofs_super_block *dsb)
 {
-	struct erofs_sb_info *sbi;
-	struct page *page;
+	struct erofs_sb_info *sbi = EROFS_SB(sb);
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 	unsigned int algs, alg;
 	erofs_off_t offset;
-	int size, ret;
+	int size, ret = 0;
 
-	sbi = EROFS_SB(sb);
 	sbi->available_compr_algs = le16_to_cpu(dsb->u1.available_compr_algs);
-
 	if (sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS) {
 		erofs_err(sb, "try to load compressed fs with unsupported algorithms %x",
 			  sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS);
@@ -205,20 +176,17 @@ static int erofs_load_compr_cfgs(struct super_block *sb,
 	}
 
 	offset = EROFS_SUPER_OFFSET + sbi->sb_size;
-	page = NULL;
 	alg = 0;
-	ret = 0;
-
 	for (algs = sbi->available_compr_algs; algs; algs >>= 1, ++alg) {
 		void *data;
 
 		if (!(algs & 1))
 			continue;
 
-		data = erofs_read_metadata(sb, &page, &offset, &size);
+		data = erofs_read_metadata(sb, &buf, &offset, &size);
 		if (IS_ERR(data)) {
 			ret = PTR_ERR(data);
-			goto err;
+			break;
 		}
 
 		switch (alg) {
@@ -234,13 +202,9 @@ static int erofs_load_compr_cfgs(struct super_block *sb,
 		}
 		kfree(data);
 		if (ret)
-			goto err;
-	}
-err:
-	if (page) {
-		unlock_page(page);
-		put_page(page);
+			break;
 	}
+	erofs_put_metabuf(&buf);
 	return ret;
 }
 #else
@@ -261,7 +225,7 @@ static int erofs_init_devices(struct super_block *sb,
 	struct erofs_sb_info *sbi = EROFS_SB(sb);
 	unsigned int ondisk_extradevs;
 	erofs_off_t pos;
-	struct page *page = NULL;
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 	struct erofs_device_info *dif;
 	struct erofs_deviceslot *dis;
 	void *ptr;
@@ -285,22 +249,13 @@ static int erofs_init_devices(struct super_block *sb,
 	pos = le16_to_cpu(dsb->devt_slotoff) * EROFS_DEVT_SLOT_SIZE;
 	down_read(&sbi->devs->rwsem);
 	idr_for_each_entry(&sbi->devs->tree, dif, id) {
-		erofs_blk_t blk = erofs_blknr(pos);
 		struct block_device *bdev;
 
-		if (!page || page->index != blk) {
-			if (page) {
-				kunmap(page);
-				unlock_page(page);
-				put_page(page);
-			}
-
-			page = erofs_get_meta_page(sb, blk);
-			if (IS_ERR(page)) {
-				up_read(&sbi->devs->rwsem);
-				return PTR_ERR(page);
-			}
-			ptr = kmap(page);
+		ptr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos),
+					 EROFS_KMAP);
+		if (IS_ERR(ptr)) {
+			err = PTR_ERR(ptr);
+			break;
 		}
 		dis = ptr + erofs_blkoff(pos);
 
@@ -309,7 +264,7 @@ static int erofs_init_devices(struct super_block *sb,
 					  sb->s_type);
 		if (IS_ERR(bdev)) {
 			err = PTR_ERR(bdev);
-			goto err_out;
+			break;
 		}
 		dif->bdev = bdev;
 		dif->dax_dev = fs_dax_get_by_bdev(bdev);
@@ -318,13 +273,8 @@ static int erofs_init_devices(struct super_block *sb,
 		sbi->total_blocks += dif->blocks;
 		pos += EROFS_DEVT_SLOT_SIZE;
 	}
-err_out:
 	up_read(&sbi->devs->rwsem);
-	if (page) {
-		kunmap(page);
-		unlock_page(page);
-		put_page(page);
-	}
+	erofs_put_metabuf(&buf);
 	return err;
 }
 
Signature · 1 line
2.24.4
author
#14 lore raw

PATCH v2 4/5 erofs: use meta buffers for xattr operations

In reply to Gao Xiang

Get rid of old erofs_get_meta_page() within xattr operations by using on-stack meta buffers in order to prepare subpage and folio features.

---
 fs/erofs/xattr.c | 134 ++++++++++++++---------------------------------
 1 file changed, 40 insertions(+), 94 deletions(-)
fs/erofs/xattr.c +40 -94
index 01c581e93c5f..437ff6e7293d 100644
@@ -8,33 +8,13 @@
 
 struct xattr_iter {
 	struct super_block *sb;
-	struct page *page;
+	struct erofs_buf buf;
 	void *kaddr;
 
 	erofs_blk_t blkaddr;
 	unsigned int ofs;
 };
 
-static inline void xattr_iter_end(struct xattr_iter *it, bool atomic)
-{
-	/* the only user of kunmap() is 'init_inode_xattrs' */
-	if (!atomic)
-		kunmap(it->page);
-	else
-		kunmap_atomic(it->kaddr);
-
-	unlock_page(it->page);
-	put_page(it->page);
-}
-
-static inline void xattr_iter_end_final(struct xattr_iter *it)
-{
-	if (!it->page)
-		return;
-
-	xattr_iter_end(it, true);
-}
-
 static int init_inode_xattrs(struct inode *inode)
 {
 	struct erofs_inode *const vi = EROFS_I(inode);
@@ -43,7 +23,6 @@ static int init_inode_xattrs(struct inode *inode)
 	struct erofs_xattr_ibody_header *ih;
 	struct super_block *sb;
 	struct erofs_sb_info *sbi;
-	bool atomic_map;
 	int ret = 0;
 
 	/* the most case is that xattrs of this inode are initialized. */
@@ -91,26 +70,23 @@ static int init_inode_xattrs(struct inode *inode)
 
 	sb = inode->i_sb;
 	sbi = EROFS_SB(sb);
+	it.buf = __EROFS_BUF_INITIALIZER;
 	it.blkaddr = erofs_blknr(iloc(sbi, vi->nid) + vi->inode_isize);
 	it.ofs = erofs_blkoff(iloc(sbi, vi->nid) + vi->inode_isize);
 
-	it.page = erofs_get_meta_page(sb, it.blkaddr);
-	if (IS_ERR(it.page)) {
-		ret = PTR_ERR(it.page);
+	/* read in shared xattr array (non-atomic, see kmalloc below) */
+	it.kaddr = erofs_read_metabuf(&it.buf, sb, it.blkaddr, EROFS_KMAP);
+	if (IS_ERR(it.kaddr)) {
+		ret = PTR_ERR(it.kaddr);
 		goto out_unlock;
 	}
 
-	/* read in shared xattr array (non-atomic, see kmalloc below) */
-	it.kaddr = kmap(it.page);
-	atomic_map = false;
-
 	ih = (struct erofs_xattr_ibody_header *)(it.kaddr + it.ofs);
-
 	vi->xattr_shared_count = ih->h_shared_count;
 	vi->xattr_shared_xattrs = kmalloc_array(vi->xattr_shared_count,
 						sizeof(uint), GFP_KERNEL);
 	if (!vi->xattr_shared_xattrs) {
-		xattr_iter_end(&it, atomic_map);
+		erofs_put_metabuf(&it.buf);
 		ret = -ENOMEM;
 		goto out_unlock;
 	}
@@ -122,25 +98,22 @@ static int init_inode_xattrs(struct inode *inode)
 		if (it.ofs >= EROFS_BLKSIZ) {
 			/* cannot be unaligned */
 			DBG_BUGON(it.ofs != EROFS_BLKSIZ);
-			xattr_iter_end(&it, atomic_map);
 
-			it.page = erofs_get_meta_page(sb, ++it.blkaddr);
-			if (IS_ERR(it.page)) {
+			it.kaddr = erofs_read_metabuf(&it.buf, sb, ++it.blkaddr,
+						      EROFS_KMAP);
+			if (IS_ERR(it.kaddr)) {
 				kfree(vi->xattr_shared_xattrs);
 				vi->xattr_shared_xattrs = NULL;
-				ret = PTR_ERR(it.page);
+				ret = PTR_ERR(it.kaddr);
 				goto out_unlock;
 			}
-
-			it.kaddr = kmap_atomic(it.page);
-			atomic_map = true;
 			it.ofs = 0;
 		}
 		vi->xattr_shared_xattrs[i] =
 			le32_to_cpu(*(__le32 *)(it.kaddr + it.ofs));
 		it.ofs += sizeof(__le32);
 	}
-	xattr_iter_end(&it, atomic_map);
+	erofs_put_metabuf(&it.buf);
 
 	/* paired with smp_mb() at the beginning of the function. */
 	smp_mb();
@@ -172,19 +145,11 @@ static inline int xattr_iter_fixup(struct xattr_iter *it)
 	if (it->ofs < EROFS_BLKSIZ)
 		return 0;
 
-	xattr_iter_end(it, true);
-
 	it->blkaddr += erofs_blknr(it->ofs);
-
-	it->page = erofs_get_meta_page(it->sb, it->blkaddr);
-	if (IS_ERR(it->page)) {
-		int err = PTR_ERR(it->page);
-
-		it->page = NULL;
-		return err;
-	}
-
-	it->kaddr = kmap_atomic(it->page);
+	it->kaddr = erofs_read_metabuf(&it->buf, it->sb, it->blkaddr,
+				       EROFS_KMAP_ATOMIC);
+	if (IS_ERR(it->kaddr))
+		return PTR_ERR(it->kaddr);
 	it->ofs = erofs_blkoff(it->ofs);
 	return 0;
 }
@@ -207,11 +172,10 @@ static int inline_xattr_iter_begin(struct xattr_iter *it,
 	it->blkaddr = erofs_blknr(iloc(sbi, vi->nid) + inline_xattr_ofs);
 	it->ofs = erofs_blkoff(iloc(sbi, vi->nid) + inline_xattr_ofs);
 
-	it->page = erofs_get_meta_page(inode->i_sb, it->blkaddr);
-	if (IS_ERR(it->page))
-		return PTR_ERR(it->page);
-
-	it->kaddr = kmap_atomic(it->page);
+	it->kaddr = erofs_read_metabuf(&it->buf, inode->i_sb, it->blkaddr,
+				       EROFS_KMAP_ATOMIC);
+	if (IS_ERR(it->kaddr))
+		return PTR_ERR(it->kaddr);
 	return vi->xattr_isize - xattr_header_sz;
 }
 
@@ -272,7 +236,7 @@ static int xattr_foreach(struct xattr_iter *it,
 			it->ofs = 0;
 		}
 
-		slice = min_t(unsigned int, PAGE_SIZE - it->ofs,
+		slice = min_t(unsigned int, EROFS_BLKSIZ - it->ofs,
 			      entry.e_name_len - processed);
 
 		/* handle name */
@@ -307,7 +271,7 @@ static int xattr_foreach(struct xattr_iter *it,
 			it->ofs = 0;
 		}
 
-		slice = min_t(unsigned int, PAGE_SIZE - it->ofs,
+		slice = min_t(unsigned int, EROFS_BLKSIZ - it->ofs,
 			      value_sz - processed);
 		op->value(it, processed, it->kaddr + it->ofs, slice);
 		it->ofs += slice;
@@ -386,8 +350,6 @@ static int inline_getxattr(struct inode *inode, struct getxattr_iter *it)
 		if (ret != -ENOATTR)
 			break;
 	}
-	xattr_iter_end_final(&it->it);
-
 	return ret ? ret : it->buffer_size;
 }
 
@@ -404,26 +366,16 @@ static int shared_getxattr(struct inode *inode, struct getxattr_iter *it)
 			xattrblock_addr(sbi, vi->xattr_shared_xattrs[i]);
 
 		it->it.ofs = xattrblock_offset(sbi, vi->xattr_shared_xattrs[i]);
-
-		if (!i || blkaddr != it->it.blkaddr) {
-			if (i)
-				xattr_iter_end(&it->it, true);
-
-			it->it.page = erofs_get_meta_page(sb, blkaddr);
-			if (IS_ERR(it->it.page))
-				return PTR_ERR(it->it.page);
-
-			it->it.kaddr = kmap_atomic(it->it.page);
-			it->it.blkaddr = blkaddr;
-		}
+		it->it.kaddr = erofs_read_metabuf(&it->it.buf, sb, blkaddr,
+						  EROFS_KMAP_ATOMIC);
+		if (IS_ERR(it->it.kaddr))
+			return PTR_ERR(it->it.kaddr);
+		it->it.blkaddr = blkaddr;
 
 		ret = xattr_foreach(&it->it, &find_xattr_handlers, NULL);
 		if (ret != -ENOATTR)
 			break;
 	}
-	if (vi->xattr_shared_count)
-		xattr_iter_end_final(&it->it);
-
 	return ret ? ret : it->buffer_size;
 }
 
@@ -452,10 +404,11 @@ int erofs_getxattr(struct inode *inode, int index,
 		return ret;
 
 	it.index = index;
-
 	it.name.len = strlen(name);
 	if (it.name.len > EROFS_NAME_LEN)
 		return -ERANGE;
+
+	it.it.buf = __EROFS_BUF_INITIALIZER;
 	it.name.name = name;
 
 	it.buffer = buffer;
@@ -465,6 +418,7 @@ int erofs_getxattr(struct inode *inode, int index,
 	ret = inline_getxattr(inode, &it);
 	if (ret == -ENOATTR)
 		ret = shared_getxattr(inode, &it);
+	erofs_put_metabuf(&it.it.buf);
 	return ret;
 }
 
@@ -607,7 +561,6 @@ static int inline_listxattr(struct listxattr_iter *it)
 		if (ret)
 			break;
 	}
-	xattr_iter_end_final(&it->it);
 	return ret ? ret : it->buffer_ofs;
 }
 
@@ -625,25 +578,16 @@ static int shared_listxattr(struct listxattr_iter *it)
 			xattrblock_addr(sbi, vi->xattr_shared_xattrs[i]);
 
 		it->it.ofs = xattrblock_offset(sbi, vi->xattr_shared_xattrs[i]);
-		if (!i || blkaddr != it->it.blkaddr) {
-			if (i)
-				xattr_iter_end(&it->it, true);
-
-			it->it.page = erofs_get_meta_page(sb, blkaddr);
-			if (IS_ERR(it->it.page))
-				return PTR_ERR(it->it.page);
-
-			it->it.kaddr = kmap_atomic(it->it.page);
-			it->it.blkaddr = blkaddr;
-		}
+		it->it.kaddr = erofs_read_metabuf(&it->it.buf, sb, blkaddr,
+						  EROFS_KMAP_ATOMIC);
+		if (IS_ERR(it->it.kaddr))
+			return PTR_ERR(it->it.kaddr);
+		it->it.blkaddr = blkaddr;
 
 		ret = xattr_foreach(&it->it, &list_xattr_handlers, NULL);
 		if (ret)
 			break;
 	}
-	if (vi->xattr_shared_count)
-		xattr_iter_end_final(&it->it);
-
 	return ret ? ret : it->buffer_ofs;
 }
 
@@ -659,6 +603,7 @@ ssize_t erofs_listxattr(struct dentry *dentry,
 	if (ret)
 		return ret;
 
+	it.it.buf = __EROFS_BUF_INITIALIZER;
 	it.dentry = dentry;
 	it.buffer = buffer;
 	it.buffer_size = buffer_size;
@@ -667,9 +612,10 @@ ssize_t erofs_listxattr(struct dentry *dentry,
 	it.it.sb = dentry->d_sb;
 
 	ret = inline_listxattr(&it);
-	if (ret < 0 && ret != -ENOATTR)
-		return ret;
-	return shared_listxattr(&it);
+	if (ret >= 0 || ret == -ENOATTR)
+		ret = shared_listxattr(&it);
+	erofs_put_metabuf(&it.it.buf);
+	return ret;
 }
 
 #ifdef CONFIG_EROFS_FS_POSIX_ACL
Signature · 1 line
2.24.4
Reviewed-by on v2 4/5
#15 lore raw

Re: [PATCH v2 4/5] erofs: use meta buffers for xattr operations

Quoted 4 lines · Gao Xiang

Get rid of old erofs_get_meta_page() within xattr operations by using on-stack meta buffers in order to prepare subpage and folio features.

Quoted 299 lines
---
 fs/erofs/xattr.c | 134 ++++++++++++++---------------------------------
 1 file changed, 40 insertions(+), 94 deletions(-)
fs/erofs/xattr.c +40 -94
index 01c581e93c5f..437ff6e7293d 100644
@@ -8,33 +8,13 @@
 
 struct xattr_iter {
 	struct super_block *sb;
-	struct page *page;
+	struct erofs_buf buf;
 	void *kaddr;
 
 	erofs_blk_t blkaddr;
 	unsigned int ofs;
 };
 
-static inline void xattr_iter_end(struct xattr_iter *it, bool atomic)
-{
-	/* the only user of kunmap() is 'init_inode_xattrs' */
-	if (!atomic)
-		kunmap(it->page);
-	else
-		kunmap_atomic(it->kaddr);
-
-	unlock_page(it->page);
-	put_page(it->page);
-}
-
-static inline void xattr_iter_end_final(struct xattr_iter *it)
-{
-	if (!it->page)
-		return;
-
-	xattr_iter_end(it, true);
-}
-
 static int init_inode_xattrs(struct inode *inode)
 {
 	struct erofs_inode *const vi = EROFS_I(inode);
@@ -43,7 +23,6 @@ static int init_inode_xattrs(struct inode *inode)
 	struct erofs_xattr_ibody_header *ih;
 	struct super_block *sb;
 	struct erofs_sb_info *sbi;
-	bool atomic_map;
 	int ret = 0;
 
 	/* the most case is that xattrs of this inode are initialized. */
@@ -91,26 +70,23 @@ static int init_inode_xattrs(struct inode *inode)
 
 	sb = inode->i_sb;
 	sbi = EROFS_SB(sb);
+	it.buf = __EROFS_BUF_INITIALIZER;
 	it.blkaddr = erofs_blknr(iloc(sbi, vi->nid) + vi->inode_isize);
 	it.ofs = erofs_blkoff(iloc(sbi, vi->nid) + vi->inode_isize);
 
-	it.page = erofs_get_meta_page(sb, it.blkaddr);
-	if (IS_ERR(it.page)) {
-		ret = PTR_ERR(it.page);
+	/* read in shared xattr array (non-atomic, see kmalloc below) */
+	it.kaddr = erofs_read_metabuf(&it.buf, sb, it.blkaddr, EROFS_KMAP);
+	if (IS_ERR(it.kaddr)) {
+		ret = PTR_ERR(it.kaddr);
 		goto out_unlock;
 	}
 
-	/* read in shared xattr array (non-atomic, see kmalloc below) */
-	it.kaddr = kmap(it.page);
-	atomic_map = false;
-
 	ih = (struct erofs_xattr_ibody_header *)(it.kaddr + it.ofs);
-
 	vi->xattr_shared_count = ih->h_shared_count;
 	vi->xattr_shared_xattrs = kmalloc_array(vi->xattr_shared_count,
 						sizeof(uint), GFP_KERNEL);
 	if (!vi->xattr_shared_xattrs) {
-		xattr_iter_end(&it, atomic_map);
+		erofs_put_metabuf(&it.buf);
 		ret = -ENOMEM;
 		goto out_unlock;
 	}
@@ -122,25 +98,22 @@ static int init_inode_xattrs(struct inode *inode)
 		if (it.ofs >= EROFS_BLKSIZ) {
 			/* cannot be unaligned */
 			DBG_BUGON(it.ofs != EROFS_BLKSIZ);
-			xattr_iter_end(&it, atomic_map);
 
-			it.page = erofs_get_meta_page(sb, ++it.blkaddr);
-			if (IS_ERR(it.page)) {
+			it.kaddr = erofs_read_metabuf(&it.buf, sb, ++it.blkaddr,
+						      EROFS_KMAP);
+			if (IS_ERR(it.kaddr)) {
 				kfree(vi->xattr_shared_xattrs);
 				vi->xattr_shared_xattrs = NULL;
-				ret = PTR_ERR(it.page);
+				ret = PTR_ERR(it.kaddr);
 				goto out_unlock;
 			}
-
-			it.kaddr = kmap_atomic(it.page);
-			atomic_map = true;
 			it.ofs = 0;
 		}
 		vi->xattr_shared_xattrs[i] =
 			le32_to_cpu(*(__le32 *)(it.kaddr + it.ofs));
 		it.ofs += sizeof(__le32);
 	}
-	xattr_iter_end(&it, atomic_map);
+	erofs_put_metabuf(&it.buf);
 
 	/* paired with smp_mb() at the beginning of the function. */
 	smp_mb();
@@ -172,19 +145,11 @@ static inline int xattr_iter_fixup(struct xattr_iter *it)
 	if (it->ofs < EROFS_BLKSIZ)
 		return 0;
 
-	xattr_iter_end(it, true);
-
 	it->blkaddr += erofs_blknr(it->ofs);
-
-	it->page = erofs_get_meta_page(it->sb, it->blkaddr);
-	if (IS_ERR(it->page)) {
-		int err = PTR_ERR(it->page);
-
-		it->page = NULL;
-		return err;
-	}
-
-	it->kaddr = kmap_atomic(it->page);
+	it->kaddr = erofs_read_metabuf(&it->buf, it->sb, it->blkaddr,
+				       EROFS_KMAP_ATOMIC);
+	if (IS_ERR(it->kaddr))
+		return PTR_ERR(it->kaddr);
 	it->ofs = erofs_blkoff(it->ofs);
 	return 0;
 }
@@ -207,11 +172,10 @@ static int inline_xattr_iter_begin(struct xattr_iter *it,
 	it->blkaddr = erofs_blknr(iloc(sbi, vi->nid) + inline_xattr_ofs);
 	it->ofs = erofs_blkoff(iloc(sbi, vi->nid) + inline_xattr_ofs);
 
-	it->page = erofs_get_meta_page(inode->i_sb, it->blkaddr);
-	if (IS_ERR(it->page))
-		return PTR_ERR(it->page);
-
-	it->kaddr = kmap_atomic(it->page);
+	it->kaddr = erofs_read_metabuf(&it->buf, inode->i_sb, it->blkaddr,
+				       EROFS_KMAP_ATOMIC);
+	if (IS_ERR(it->kaddr))
+		return PTR_ERR(it->kaddr);
 	return vi->xattr_isize - xattr_header_sz;
 }
 
@@ -272,7 +236,7 @@ static int xattr_foreach(struct xattr_iter *it,
 			it->ofs = 0;
 		}
 
-		slice = min_t(unsigned int, PAGE_SIZE - it->ofs,
+		slice = min_t(unsigned int, EROFS_BLKSIZ - it->ofs,
 			      entry.e_name_len - processed);
 
 		/* handle name */
@@ -307,7 +271,7 @@ static int xattr_foreach(struct xattr_iter *it,
 			it->ofs = 0;
 		}
 
-		slice = min_t(unsigned int, PAGE_SIZE - it->ofs,
+		slice = min_t(unsigned int, EROFS_BLKSIZ - it->ofs,
 			      value_sz - processed);
 		op->value(it, processed, it->kaddr + it->ofs, slice);
 		it->ofs += slice;
@@ -386,8 +350,6 @@ static int inline_getxattr(struct inode *inode, struct getxattr_iter *it)
 		if (ret != -ENOATTR)
 			break;
 	}
-	xattr_iter_end_final(&it->it);
-
 	return ret ? ret : it->buffer_size;
 }
 
@@ -404,26 +366,16 @@ static int shared_getxattr(struct inode *inode, struct getxattr_iter *it)
 			xattrblock_addr(sbi, vi->xattr_shared_xattrs[i]);
 
 		it->it.ofs = xattrblock_offset(sbi, vi->xattr_shared_xattrs[i]);
-
-		if (!i || blkaddr != it->it.blkaddr) {
-			if (i)
-				xattr_iter_end(&it->it, true);
-
-			it->it.page = erofs_get_meta_page(sb, blkaddr);
-			if (IS_ERR(it->it.page))
-				return PTR_ERR(it->it.page);
-
-			it->it.kaddr = kmap_atomic(it->it.page);
-			it->it.blkaddr = blkaddr;
-		}
+		it->it.kaddr = erofs_read_metabuf(&it->it.buf, sb, blkaddr,
+						  EROFS_KMAP_ATOMIC);
+		if (IS_ERR(it->it.kaddr))
+			return PTR_ERR(it->it.kaddr);
+		it->it.blkaddr = blkaddr;
 
 		ret = xattr_foreach(&it->it, &find_xattr_handlers, NULL);
 		if (ret != -ENOATTR)
 			break;
 	}
-	if (vi->xattr_shared_count)
-		xattr_iter_end_final(&it->it);
-
 	return ret ? ret : it->buffer_size;
 }
 
@@ -452,10 +404,11 @@ int erofs_getxattr(struct inode *inode, int index,
 		return ret;
 
 	it.index = index;
-
 	it.name.len = strlen(name);
 	if (it.name.len > EROFS_NAME_LEN)
 		return -ERANGE;
+
+	it.it.buf = __EROFS_BUF_INITIALIZER;
 	it.name.name = name;
 
 	it.buffer = buffer;
@@ -465,6 +418,7 @@ int erofs_getxattr(struct inode *inode, int index,
 	ret = inline_getxattr(inode, &it);
 	if (ret == -ENOATTR)
 		ret = shared_getxattr(inode, &it);
+	erofs_put_metabuf(&it.it.buf);
 	return ret;
 }
 
@@ -607,7 +561,6 @@ static int inline_listxattr(struct listxattr_iter *it)
 		if (ret)
 			break;
 	}
-	xattr_iter_end_final(&it->it);
 	return ret ? ret : it->buffer_ofs;
 }
 
@@ -625,25 +578,16 @@ static int shared_listxattr(struct listxattr_iter *it)
 			xattrblock_addr(sbi, vi->xattr_shared_xattrs[i]);
 
 		it->it.ofs = xattrblock_offset(sbi, vi->xattr_shared_xattrs[i]);
-		if (!i || blkaddr != it->it.blkaddr) {
-			if (i)
-				xattr_iter_end(&it->it, true);
-
-			it->it.page = erofs_get_meta_page(sb, blkaddr);
-			if (IS_ERR(it->it.page))
-				return PTR_ERR(it->it.page);
-
-			it->it.kaddr = kmap_atomic(it->it.page);
-			it->it.blkaddr = blkaddr;
-		}
+		it->it.kaddr = erofs_read_metabuf(&it->it.buf, sb, blkaddr,
+						  EROFS_KMAP_ATOMIC);
+		if (IS_ERR(it->it.kaddr))
+			return PTR_ERR(it->it.kaddr);
+		it->it.blkaddr = blkaddr;
 
 		ret = xattr_foreach(&it->it, &list_xattr_handlers, NULL);
 		if (ret)
 			break;
 	}
-	if (vi->xattr_shared_count)
-		xattr_iter_end_final(&it->it);
-
 	return ret ? ret : it->buffer_ofs;
 }
 
@@ -659,6 +603,7 @@ ssize_t erofs_listxattr(struct dentry *dentry,
 	if (ret)
 		return ret;
 
+	it.it.buf = __EROFS_BUF_INITIALIZER;
 	it.dentry = dentry;
 	it.buffer = buffer;
 	it.buffer_size = buffer_size;
@@ -667,9 +612,10 @@ ssize_t erofs_listxattr(struct dentry *dentry,
 	it.it.sb = dentry->d_sb;
 
 	ret = inline_listxattr(&it);
-	if (ret < 0 && ret != -ENOATTR)
-		return ret;
-	return shared_listxattr(&it);
+	if (ret >= 0 || ret == -ENOATTR)
+		ret = shared_listxattr(&it);
+	erofs_put_metabuf(&it.it.buf);
+	return ret;
 }
 
 #ifdef CONFIG_EROFS_FS_POSIX_ACL
Signature · 1 line
2.24.4
author
#17 lore raw

PATCH v2 5/5 erofs: use meta buffers for zmap operations

In reply to Gao Xiang

Get rid of old erofs_get_meta_page() within zmap operations by using on-stack meta buffers in order to prepare subpage and folio features.

Finally, erofs_get_meta_page() is useless. Get rid of it!

---
 fs/erofs/data.c     | 13 -----------
 fs/erofs/internal.h |  6 ++---
 fs/erofs/zdata.c    | 23 ++++++++-----------
 fs/erofs/zmap.c     | 56 +++++++++++++--------------------------------
 4 files changed, 28 insertions(+), 70 deletions(-)
fs/erofs/data.c +0 -13
index 6495e16a50a9..187f19f8a9a1 100644
@@ -9,19 +9,6 @@
 #include <linux/dax.h>
 #include <trace/events/erofs.h>
 
-struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr)
-{
-	struct address_space *const mapping = sb->s_bdev->bd_inode->i_mapping;
-	struct page *page;
-
-	page = read_cache_page_gfp(mapping, blkaddr,
-				   mapping_gfp_constraint(mapping, ~__GFP_FS));
-	/* should already be PageUptodate */
-	if (!IS_ERR(page))
-		lock_page(page);
-	return page;
-}
-
 void erofs_unmap_metabuf(struct erofs_buf *buf)
 {
 	if (buf->kmap_type == EROFS_KMAP)
fs/erofs/internal.h +3 -3
index f1e4eb3025f6..3db494a398b2 100644
@@ -419,14 +419,14 @@ enum {
 #define EROFS_MAP_FULL_MAPPED	(1 << BH_FullMapped)
 
 struct erofs_map_blocks {
+	struct erofs_buf buf;
+
 	erofs_off_t m_pa, m_la;
 	u64 m_plen, m_llen;
 
 	unsigned short m_deviceid;
 	char m_algorithmformat;
 	unsigned int m_flags;
-
-	struct page *mpage;
 };
 
 /* Flags used by erofs_map_blocks_flatmode() */
@@ -474,7 +474,7 @@ struct erofs_map_dev {
 
 /* data.c */
 extern const struct file_operations erofs_file_fops;
-struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr);
+void erofs_unmap_metabuf(struct erofs_buf *buf);
 void erofs_put_metabuf(struct erofs_buf *buf);
 void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
 			 erofs_blk_t blkaddr, enum erofs_kmap_type type);
fs/erofs/zdata.c +9 -14
index 49da3931b2e3..498b7666efe8 100644
@@ -698,20 +698,18 @@ static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe,
 		goto err_out;
 
 	if (z_erofs_is_inline_pcluster(clt->pcl)) {
-		struct page *mpage;
+		void *mp;
 
-		mpage = erofs_get_meta_page(inode->i_sb,
-					    erofs_blknr(map->m_pa));
-		if (IS_ERR(mpage)) {
-			err = PTR_ERR(mpage);
+		mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb,
+					erofs_blknr(map->m_pa), EROFS_NO_KMAP);
+		if (IS_ERR(mp)) {
+			err = PTR_ERR(mp);
 			erofs_err(inode->i_sb,
 				  "failed to get inline page, err %d", err);
 			goto err_out;
 		}
-		/* TODO: new subpage feature will get rid of it */
-		unlock_page(mpage);
-
-		WRITE_ONCE(clt->pcl->compressed_pages[0], mpage);
+		get_page(fe->map.buf.page);
+		WRITE_ONCE(clt->pcl->compressed_pages[0], fe->map.buf.page);
 		clt->mode = COLLECT_PRIMARY_FOLLOWED_NOINPLACE;
 	} else {
 		/* preload all compressed pages (can change mode if needed) */
@@ -1529,9 +1527,7 @@ static int z_erofs_readpage(struct file *file, struct page *page)
 	if (err)
 		erofs_err(inode->i_sb, "failed to read, err [%d]", err);
 
-	if (f.map.mpage)
-		put_page(f.map.mpage);
-
+	erofs_put_metabuf(&f.map.buf);
 	erofs_release_pages(&pagepool);
 	return err;
 }
@@ -1576,8 +1572,7 @@ static void z_erofs_readahead(struct readahead_control *rac)
 
 	z_erofs_runqueue(inode->i_sb, &f, &pagepool,
 			 z_erofs_get_sync_decompress_policy(sbi, nr_pages));
-	if (f.map.mpage)
-		put_page(f.map.mpage);
+	erofs_put_metabuf(&f.map.buf);
 	erofs_release_pages(&pagepool);
 }
 
fs/erofs/zmap.c +16 -40

104 lines not on this page · raw message

Signature · 1 line
2.24.4
Reviewed-by on v2 5/5
#18 lore raw

Re: [PATCH v2 5/5] erofs: use meta buffers for zmap operations

Quoted 5 lines · Gao Xiang

Get rid of old erofs_get_meta_page() within zmap operations by using on-stack meta buffers in order to prepare subpage and folio features.

Finally, erofs_get_meta_page() is useless. Get rid of it!

Quoted 233 lines
---
 fs/erofs/data.c     | 13 -----------
 fs/erofs/internal.h |  6 ++---
 fs/erofs/zdata.c    | 23 ++++++++-----------
 fs/erofs/zmap.c     | 56 +++++++++++++--------------------------------
 4 files changed, 28 insertions(+), 70 deletions(-)
fs/erofs/data.c +0 -13
index 6495e16a50a9..187f19f8a9a1 100644
@@ -9,19 +9,6 @@
 #include <linux/dax.h>
 #include <trace/events/erofs.h>
 
-struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr)
-{
-	struct address_space *const mapping = sb->s_bdev->bd_inode->i_mapping;
-	struct page *page;
-
-	page = read_cache_page_gfp(mapping, blkaddr,
-				   mapping_gfp_constraint(mapping, ~__GFP_FS));
-	/* should already be PageUptodate */
-	if (!IS_ERR(page))
-		lock_page(page);
-	return page;
-}
-
 void erofs_unmap_metabuf(struct erofs_buf *buf)
 {
 	if (buf->kmap_type == EROFS_KMAP)
fs/erofs/internal.h +3 -3
index f1e4eb3025f6..3db494a398b2 100644
@@ -419,14 +419,14 @@ enum {
 #define EROFS_MAP_FULL_MAPPED	(1 << BH_FullMapped)
 
 struct erofs_map_blocks {
+	struct erofs_buf buf;
+
 	erofs_off_t m_pa, m_la;
 	u64 m_plen, m_llen;
 
 	unsigned short m_deviceid;
 	char m_algorithmformat;
 	unsigned int m_flags;
-
-	struct page *mpage;
 };
 
 /* Flags used by erofs_map_blocks_flatmode() */
@@ -474,7 +474,7 @@ struct erofs_map_dev {
 
 /* data.c */
 extern const struct file_operations erofs_file_fops;
-struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr);
+void erofs_unmap_metabuf(struct erofs_buf *buf);
 void erofs_put_metabuf(struct erofs_buf *buf);
 void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
 			 erofs_blk_t blkaddr, enum erofs_kmap_type type);
fs/erofs/zdata.c +9 -14
index 49da3931b2e3..498b7666efe8 100644
@@ -698,20 +698,18 @@ static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe,
 		goto err_out;
 
 	if (z_erofs_is_inline_pcluster(clt->pcl)) {
-		struct page *mpage;
+		void *mp;
 
-		mpage = erofs_get_meta_page(inode->i_sb,
-					    erofs_blknr(map->m_pa));
-		if (IS_ERR(mpage)) {
-			err = PTR_ERR(mpage);
+		mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb,
+					erofs_blknr(map->m_pa), EROFS_NO_KMAP);
+		if (IS_ERR(mp)) {
+			err = PTR_ERR(mp);
 			erofs_err(inode->i_sb,
 				  "failed to get inline page, err %d", err);
 			goto err_out;
 		}
-		/* TODO: new subpage feature will get rid of it */
-		unlock_page(mpage);
-
-		WRITE_ONCE(clt->pcl->compressed_pages[0], mpage);
+		get_page(fe->map.buf.page);
+		WRITE_ONCE(clt->pcl->compressed_pages[0], fe->map.buf.page);
 		clt->mode = COLLECT_PRIMARY_FOLLOWED_NOINPLACE;
 	} else {
 		/* preload all compressed pages (can change mode if needed) */
@@ -1529,9 +1527,7 @@ static int z_erofs_readpage(struct file *file, struct page *page)
 	if (err)
 		erofs_err(inode->i_sb, "failed to read, err [%d]", err);
 
-	if (f.map.mpage)
-		put_page(f.map.mpage);
-
+	erofs_put_metabuf(&f.map.buf);
 	erofs_release_pages(&pagepool);
 	return err;
 }
@@ -1576,8 +1572,7 @@ static void z_erofs_readahead(struct readahead_control *rac)
 
 	z_erofs_runqueue(inode->i_sb, &f, &pagepool,
 			 z_erofs_get_sync_decompress_policy(sbi, nr_pages));
-	if (f.map.mpage)
-		put_page(f.map.mpage);
+	erofs_put_metabuf(&f.map.buf);
 	erofs_release_pages(&pagepool);
 }
 
fs/erofs/zmap.c +16 -40
index 1037ac17b7a6..18d7fd1a5064 100644
@@ -35,7 +35,7 @@ static int z_erofs_fill_inode_lazy(struct inode *inode)
 	struct super_block *const sb = inode->i_sb;
 	int err, headnr;
 	erofs_off_t pos;
-	struct page *page;
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
 	void *kaddr;
 	struct z_erofs_map_header *h;
 
@@ -61,14 +61,13 @@ static int z_erofs_fill_inode_lazy(struct inode *inode)
 
 	pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize +
 		    vi->xattr_isize, 8);
-	page = erofs_get_meta_page(sb, erofs_blknr(pos));
-	if (IS_ERR(page)) {
-		err = PTR_ERR(page);
+	kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos),
+				   EROFS_KMAP_ATOMIC);
+	if (IS_ERR(kaddr)) {
+		err = PTR_ERR(kaddr);
 		goto out_unlock;
 	}
 
-	kaddr = kmap_atomic(page);
-
 	h = kaddr + erofs_blkoff(pos);
 	vi->z_advise = le16_to_cpu(h->h_advise);
 	vi->z_algorithmtype[0] = h->h_algorithmtype & 15;
@@ -101,20 +100,19 @@ static int z_erofs_fill_inode_lazy(struct inode *inode)
 		goto unmap_done;
 	}
 unmap_done:
-	kunmap_atomic(kaddr);
-	unlock_page(page);
-	put_page(page);
+	erofs_put_metabuf(&buf);
 	if (err)
 		goto out_unlock;
 
 	if (vi->z_advise & Z_EROFS_ADVISE_INLINE_PCLUSTER) {
-		struct erofs_map_blocks map = { .mpage = NULL };
+		struct erofs_map_blocks map = {
+			.buf = __EROFS_BUF_INITIALIZER
+		};
 
 		vi->z_idata_size = le16_to_cpu(h->h_idata_size);
 		err = z_erofs_do_map_blocks(inode, &map,
 					    EROFS_GET_BLOCKS_FINDTAIL);
-		if (map.mpage)
-			put_page(map.mpage);
+		erofs_put_metabuf(&map.buf);
 
 		if (!map.m_plen ||
 		    erofs_blkoff(map.m_pa) + map.m_plen > EROFS_BLKSIZ) {
@@ -151,31 +149,11 @@ static int z_erofs_reload_indexes(struct z_erofs_maprecorder *m,
 				  erofs_blk_t eblk)
 {
 	struct super_block *const sb = m->inode->i_sb;
-	struct erofs_map_blocks *const map = m->map;
-	struct page *mpage = map->mpage;
-
-	if (mpage) {
-		if (mpage->index == eblk) {
-			if (!m->kaddr)
-				m->kaddr = kmap_atomic(mpage);
-			return 0;
-		}
 
-		if (m->kaddr) {
-			kunmap_atomic(m->kaddr);
-			m->kaddr = NULL;
-		}
-		put_page(mpage);
-	}
-
-	mpage = erofs_get_meta_page(sb, eblk);
-	if (IS_ERR(mpage)) {
-		map->mpage = NULL;
-		return PTR_ERR(mpage);
-	}
-	m->kaddr = kmap_atomic(mpage);
-	unlock_page(mpage);
-	map->mpage = mpage;
+	m->kaddr = erofs_read_metabuf(&m->map->buf, sb, eblk,
+				      EROFS_KMAP_ATOMIC);
+	if (IS_ERR(m->kaddr))
+		return PTR_ERR(m->kaddr);
 	return 0;
 }
 
@@ -711,8 +689,7 @@ static int z_erofs_do_map_blocks(struct inode *inode,
 			map->m_flags |= EROFS_MAP_FULL_MAPPED;
 	}
 unmap_out:
-	if (m.kaddr)
-		kunmap_atomic(m.kaddr);
+	erofs_unmap_metabuf(&m.map->buf);
 
 out:
 	erofs_dbg("%s, m_la %llu m_pa %llu m_llen %llu m_plen %llu m_flags 0%o",
@@ -759,8 +736,7 @@ static int z_erofs_iomap_begin_report(struct inode *inode, loff_t offset,
 	struct erofs_map_blocks map = { .m_la = offset };
 
 	ret = z_erofs_map_blocks_iter(inode, &map, EROFS_GET_BLOCKS_FIEMAP);
-	if (map.mpage)
-		put_page(map.mpage);
+	erofs_put_metabuf(&map.buf);
 	if (ret < 0)
 		return ret;
 
Signature · 1 line
2.24.4
maintainer
#19 lore raw

Re: [PATCH v2 5/5] erofs: use meta buffers for zmap operations

In reply to Gao Xiang
Quoted 87 lines · Gao Xiang

Get rid of old erofs_get_meta_page() within zmap operations by using on-stack meta buffers in order to prepare subpage and folio features.

Finally, erofs_get_meta_page() is useless. Get rid of it!

---
  fs/erofs/data.c     | 13 -----------
  fs/erofs/internal.h |  6 ++---
  fs/erofs/zdata.c    | 23 ++++++++-----------
  fs/erofs/zmap.c     | 56 +++++++++++++--------------------------------
  4 files changed, 28 insertions(+), 70 deletions(-)
fs/erofs/data.c +0 -13
index 6495e16a50a9..187f19f8a9a1 100644
@@ -9,19 +9,6 @@
  #include <linux/dax.h>
  #include <trace/events/erofs.h>
 
-struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr)
-{
-	struct address_space *const mapping = sb->s_bdev->bd_inode->i_mapping;
-	struct page *page;
-
-	page = read_cache_page_gfp(mapping, blkaddr,
-				   mapping_gfp_constraint(mapping, ~__GFP_FS));
-	/* should already be PageUptodate */
-	if (!IS_ERR(page))
-		lock_page(page);
-	return page;
-}
-
  void erofs_unmap_metabuf(struct erofs_buf *buf)
  {
  	if (buf->kmap_type == EROFS_KMAP)
fs/erofs/internal.h +3 -3
index f1e4eb3025f6..3db494a398b2 100644
@@ -419,14 +419,14 @@ enum {
  #define EROFS_MAP_FULL_MAPPED	(1 << BH_FullMapped)
 
  struct erofs_map_blocks {
+	struct erofs_buf buf;
+
  	erofs_off_t m_pa, m_la;
  	u64 m_plen, m_llen;
 
  	unsigned short m_deviceid;
  	char m_algorithmformat;
  	unsigned int m_flags;
-
-	struct page *mpage;
  };
 
  /* Flags used by erofs_map_blocks_flatmode() */
@@ -474,7 +474,7 @@ struct erofs_map_dev {
 
  /* data.c */
  extern const struct file_operations erofs_file_fops;
-struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr);
+void erofs_unmap_metabuf(struct erofs_buf *buf);
  void erofs_put_metabuf(struct erofs_buf *buf);
  void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
  			 erofs_blk_t blkaddr, enum erofs_kmap_type type);
fs/erofs/zdata.c +4 -5
index 49da3931b2e3..498b7666efe8 100644
@@ -698,20 +698,18 @@ static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe,
  		goto err_out;
 
  	if (z_erofs_is_inline_pcluster(clt->pcl)) {
-		struct page *mpage;
+		void *mp;
 
-		mpage = erofs_get_meta_page(inode->i_sb,
-					    erofs_blknr(map->m_pa));
-		if (IS_ERR(mpage)) {
-			err = PTR_ERR(mpage);
+		mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb,
+					erofs_blknr(map->m_pa), EROFS_NO_KMAP);
+		if (IS_ERR(mp)) {

fs/erofs/zdata.c static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, · line 711

Quoted 10 lines
fs/erofs/zdata.c +2 -4
+			err = PTR_ERR(mp);
  			erofs_err(inode->i_sb,
  				  "failed to get inline page, err %d", err);
  			goto err_out;
  		}
-		/* TODO: new subpage feature will get rid of it */
-		unlock_page(mpage);
-
-		WRITE_ONCE(clt->pcl->compressed_pages[0], mpage);
+		get_page(fe->map.buf.page);

Comparing to previous implementation, it adds an extra reference on the page, why?

Thanks,

Quoted 139 lines
fs/erofs/zdata.c +3 -5
+		WRITE_ONCE(clt->pcl->compressed_pages[0], fe->map.buf.page);
  		clt->mode = COLLECT_PRIMARY_FOLLOWED_NOINPLACE;
  	} else {
  		/* preload all compressed pages (can change mode if needed) */
@@ -1529,9 +1527,7 @@ static int z_erofs_readpage(struct file *file, struct page *page)
  	if (err)
  		erofs_err(inode->i_sb, "failed to read, err [%d]", err);
 
-	if (f.map.mpage)
-		put_page(f.map.mpage);
-
+	erofs_put_metabuf(&f.map.buf);
  	erofs_release_pages(&pagepool);
  	return err;
  }
@@ -1576,8 +1572,7 @@ static void z_erofs_readahead(struct readahead_control *rac)
 
  	z_erofs_runqueue(inode->i_sb, &f, &pagepool,
  			 z_erofs_get_sync_decompress_policy(sbi, nr_pages));
-	if (f.map.mpage)
-		put_page(f.map.mpage);
+	erofs_put_metabuf(&f.map.buf);
  	erofs_release_pages(&pagepool);
  }
 
fs/erofs/zmap.c +16 -40
index 1037ac17b7a6..18d7fd1a5064 100644
@@ -35,7 +35,7 @@ static int z_erofs_fill_inode_lazy(struct inode *inode)
  	struct super_block *const sb = inode->i_sb;
  	int err, headnr;
  	erofs_off_t pos;
-	struct page *page;
+	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
  	void *kaddr;
  	struct z_erofs_map_header *h;
 
@@ -61,14 +61,13 @@ static int z_erofs_fill_inode_lazy(struct inode *inode)
 
  	pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize +
  		    vi->xattr_isize, 8);
-	page = erofs_get_meta_page(sb, erofs_blknr(pos));
-	if (IS_ERR(page)) {
-		err = PTR_ERR(page);
+	kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos),
+				   EROFS_KMAP_ATOMIC);
+	if (IS_ERR(kaddr)) {
+		err = PTR_ERR(kaddr);
  		goto out_unlock;
  	}
 
-	kaddr = kmap_atomic(page);
-
  	h = kaddr + erofs_blkoff(pos);
  	vi->z_advise = le16_to_cpu(h->h_advise);
  	vi->z_algorithmtype[0] = h->h_algorithmtype & 15;
@@ -101,20 +100,19 @@ static int z_erofs_fill_inode_lazy(struct inode *inode)
  		goto unmap_done;
  	}
  unmap_done:
-	kunmap_atomic(kaddr);
-	unlock_page(page);
-	put_page(page);
+	erofs_put_metabuf(&buf);
  	if (err)
  		goto out_unlock;
 
  	if (vi->z_advise & Z_EROFS_ADVISE_INLINE_PCLUSTER) {
-		struct erofs_map_blocks map = { .mpage = NULL };
+		struct erofs_map_blocks map = {
+			.buf = __EROFS_BUF_INITIALIZER
+		};
 
  		vi->z_idata_size = le16_to_cpu(h->h_idata_size);
  		err = z_erofs_do_map_blocks(inode, &map,
  					    EROFS_GET_BLOCKS_FINDTAIL);
-		if (map.mpage)
-			put_page(map.mpage);
+		erofs_put_metabuf(&map.buf);
 
  		if (!map.m_plen ||
  		    erofs_blkoff(map.m_pa) + map.m_plen > EROFS_BLKSIZ) {
@@ -151,31 +149,11 @@ static int z_erofs_reload_indexes(struct z_erofs_maprecorder *m,
  				  erofs_blk_t eblk)
  {
  	struct super_block *const sb = m->inode->i_sb;
-	struct erofs_map_blocks *const map = m->map;
-	struct page *mpage = map->mpage;
-
-	if (mpage) {
-		if (mpage->index == eblk) {
-			if (!m->kaddr)
-				m->kaddr = kmap_atomic(mpage);
-			return 0;
-		}
 
-		if (m->kaddr) {
-			kunmap_atomic(m->kaddr);
-			m->kaddr = NULL;
-		}
-		put_page(mpage);
-	}
-
-	mpage = erofs_get_meta_page(sb, eblk);
-	if (IS_ERR(mpage)) {
-		map->mpage = NULL;
-		return PTR_ERR(mpage);
-	}
-	m->kaddr = kmap_atomic(mpage);
-	unlock_page(mpage);
-	map->mpage = mpage;
+	m->kaddr = erofs_read_metabuf(&m->map->buf, sb, eblk,
+				      EROFS_KMAP_ATOMIC);
+	if (IS_ERR(m->kaddr))
+		return PTR_ERR(m->kaddr);
  	return 0;
  }
 
@@ -711,8 +689,7 @@ static int z_erofs_do_map_blocks(struct inode *inode,
  			map->m_flags |= EROFS_MAP_FULL_MAPPED;
  	}
  unmap_out:
-	if (m.kaddr)
-		kunmap_atomic(m.kaddr);
+	erofs_unmap_metabuf(&m.map->buf);
 
  out:
  	erofs_dbg("%s, m_la %llu m_pa %llu m_llen %llu m_plen %llu m_flags 0%o",
@@ -759,8 +736,7 @@ static int z_erofs_iomap_begin_report(struct inode *inode, loff_t offset,
  	struct erofs_map_blocks map = { .m_la = offset };
 
  	ret = z_erofs_map_blocks_iter(inode, &map, EROFS_GET_BLOCKS_FIEMAP);
-	if (map.mpage)
-		put_page(map.mpage);
+	erofs_put_metabuf(&map.buf);
  	if (ret < 0)
  		return ret;
 
author
#20 lore raw

Re: [PATCH v2 5/5] erofs: use meta buffers for zmap operations

Hi Chao,

Quoted 1 line · Chao Yu

On 2022/1/2 12:00, Gao Xiang wrote:

...

Quoted 18 lines
Quoted 18 lines
fs/erofs/zdata.c +5 -5
@@ -698,20 +698,18 @@ static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe,
  		goto err_out;
  	if (z_erofs_is_inline_pcluster(clt->pcl)) {
-		struct page *mpage;
+		void *mp;
-		mpage = erofs_get_meta_page(inode->i_sb,
-					    erofs_blknr(map->m_pa));
-		if (IS_ERR(mpage)) {
-			err = PTR_ERR(mpage);
+		mp = erofs_read_metabuf(&fe->map.buf, inode->i_sb,
+					erofs_blknr(map->m_pa), EROFS_NO_KMAP);
+		if (IS_ERR(mp)) {
+			err = PTR_ERR(mp);
  			erofs_err(inode->i_sb,
  				  "failed to get inline page, err %d", err);
  			goto err_out;
Quoted 8 lines

fs/erofs/zdata.c static int z_erofs_do_read_page(struct z_erofs_decompress_frontend *fe, · line 711

Quoted 6 lines
fs/erofs/zdata.c +1 -4
  		}
-		/* TODO: new subpage feature will get rid of it */
-		unlock_page(mpage);
-
-		WRITE_ONCE(clt->pcl->compressed_pages[0], mpage);
+		get_page(fe->map.buf.page);

Comparing to previous implementation, it adds an extra reference on the page, why?

Thanks for the question. Previously, erofs_get_meta_page was called independently without reusing zmap mpage, so the page refcount had no relationship with zmap mpage.

However, now we reuse zmap metabuf instead(fe->map.buf), so an extra page refcount is needed since zmap metabuf will be released at the end of readpage or readahead...

Thanks,
Gao Xiang

maintainer Reviewed-by
#21 lore raw

Re: [PATCH v2 5/5] erofs: use meta buffers for zmap operations

Quoted 7 lines · Gao Xiang

Thanks for the question. Previously, erofs_get_meta_page was called independently without reusing zmap mpage, so the page refcount had no relationship with zmap mpage.

However, now we reuse zmap metabuf instead(fe->map.buf), so an extra page refcount is needed since zmap metabuf will be released at the end of readpage or readahead...

Thanks for the explanation.

Thanks,