Skip to contentSkip to conversation

erofs: iomap support for uncompressed cases

PATCH In review Gao Xiang 2 participants 8 messages Aug 5, 2021 – Aug 6, 2021

Subsystems (4)

v3 of 6 In review Latest is v6 →

lore
Share & tools
lore

Gao Xiang

author
#0 lore raw

v3 cover erofs: iomap support for uncompressed cases

Hi folks,

This patchset mainly adds EROFS iomap support for uncompressed cases I've planed for the next merge window.

The first 2 patches mainly deal with 2 new cases:
1) Direct I/O is useful in certain scenarios for uncompressed files. For example, double pagecache can be avoid by direct I/O when loop device is used for uncompressed files containing upper layer compressed filesystem.

2) DAX is quite useful for some container use cases in order to save guest memory extremely by using the minimal lightweight EROFS image. BTW, a bit more off this iomap topic, chunk-deduplicated regfile support is almost available (blob data support) for multi-layer container image use cases (aka. called RAFS v6, nydus [1] will support RAFS v6 (EROFS-compatible format) in the future and form a unified high-performance container image solution, which will be announced formally independently), which is also a small independent update.

The last patch relies on the previous iomap core update in order to convert tail-packing inline files into iomap, thus all uncompressed cases are handled with iomap properly.

Comments are welcome. Thanks for your time on reading this!

Thanks,
Gao Xiang

[1] https://github.com/dragonflyoss/image-service

changes since v2 (Chao):
- use filemap_read() instead;
- slightly adjust the order of a hunk in erofs_fill_inode();
- update brief documentation for dax option;
- use fs_put_dax() directly w/o NULL check;
- add a missing error handling for erofs_get_meta_page().

Gao Xiang (2):
erofs: dax support for non-tailpacking regular file
erofs: convert all uncompressed cases to iomap

Huang Jianan (1):
erofs: iomap support for non-tailpacking DIO

 Documentation/filesystems/erofs.rst |   2 +
 fs/erofs/Kconfig                    |   1 +
 fs/erofs/data.c                     | 338 ++++++++++++----------------
 fs/erofs/inode.c                    |   9 +-
 fs/erofs/internal.h                 |   4 +
 fs/erofs/super.c                    |  59 ++++-
 6 files changed, 213 insertions(+), 200 deletions(-)
Signature · 1 line
2.24.4
author
#1 lore raw

PATCH v3 1/3 erofs: iomap support for non-tailpacking DIO

From: Huang Jianan <[email protected]>

Add iomap support for non-tailpacking uncompressed data in order to support DIO and DAX.

Direct I/O is useful in certain scenarios for uncompressed files. For example, double pagecache can be avoid by direct I/O when loop device is used for uncompressed files containing upper layer compressed filesystem.

This adds iomap DIO support for non-tailpacking cases first and tail-packing inline files are handled in the follow-up patch.

---
 fs/erofs/Kconfig    |  1 +
 fs/erofs/data.c     | 94 +++++++++++++++++++++++++++++++++++++++++++++
 fs/erofs/inode.c    |  5 ++-
 fs/erofs/internal.h |  1 +
 4 files changed, 100 insertions(+), 1 deletion(-)
fs/erofs/Kconfig +1 -0
index 906af0c1998c..14b747026742 100644
@@ -3,6 +3,7 @@
 config EROFS_FS
 	tristate "EROFS filesystem support"
 	depends on BLOCK
+	select FS_IOMAP
 	select LIBCRC32C
 	help
 	  EROFS (Enhanced Read-Only File System) is a lightweight
fs/erofs/data.c +94 -0

109 lines not on this page · raw message

fs/erofs/inode.c +4 -1
index aa8a0d770ba3..f296e1a2661e 100644
@@ -247,7 +247,10 @@ static int erofs_fill_inode(struct inode *inode, int isdir)
 	switch (inode->i_mode & S_IFMT) {
 	case S_IFREG:
 		inode->i_op = &erofs_generic_iops;
-		inode->i_fop = &generic_ro_fops;
+		if (erofs_inode_is_data_compressed(vi->datalayout))
+			inode->i_fop = &generic_ro_fops;
+		else
+			inode->i_fop = &erofs_file_fops;
 		break;
 	case S_IFDIR:
 		inode->i_op = &erofs_dir_iops;
fs/erofs/internal.h +1 -0
index 543c2ff97d30..2669c785d548 100644
@@ -371,6 +371,7 @@ static inline int z_erofs_map_blocks_iter(struct inode *inode,
 #endif	/* !CONFIG_EROFS_FS_ZIP */
 
 /* data.c */
+extern const struct file_operations erofs_file_fops;
 struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr);
 
 /* inode.c */
Signature · 1 line
2.24.4
author
#2 lore raw

PATCH v3 2/3 erofs: dax support for non-tailpacking regular file

In reply to Gao Xiang

DAX is quite useful for some VM use cases in order to save guest memory extremely with minimal lightweight EROFS.

In order to prepare for such use cases, add preliminary dax support for non-tailpacking regular files for now.

Tested with the DRAM-emulated PMEM and the EROFS image generated by "mkfs.erofs -Enoinline_data enwik9.fsdax.img enwik9"

---
 Documentation/filesystems/erofs.rst |  2 +
 fs/erofs/data.c                     | 42 +++++++++++++++++++-
 fs/erofs/inode.c                    |  4 ++
 fs/erofs/internal.h                 |  3 ++
 fs/erofs/super.c                    | 59 ++++++++++++++++++++++++++++-
 5 files changed, 106 insertions(+), 4 deletions(-)
Documentation/filesystems/erofs.rst +2 -0
index 832839fcf4c3..ffd2ae7be511 100644
@@ -84,6 +84,8 @@ cache_strategy=%s      Select a strategy for cached decompression from now on:
                                    It still does in-place I/O decompression
                                    for the rest compressed physical clusters.
 		       ==========  =============================================
+dax                    Use direct access (no page cache).  See
+                       Documentation/filesystems/dax.rst.
 ===================    =========================================================
 
 On-disk details
fs/erofs/data.c +40 -2

68 lines not on this page · raw message

fs/erofs/inode.c +4 -0
index f296e1a2661e..92728da1d206 100644
@@ -174,6 +174,10 @@ static struct page *erofs_read_inode(struct inode *inode,
 	inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec;
 	inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec;
 
+	inode->i_flags &= ~S_DAX;
+	if (test_opt(&sbi->ctx, DAX_ALWAYS) && S_ISREG(inode->i_mode) &&
+	    vi->datalayout == EROFS_INODE_FLAT_PLAIN)
+		inode->i_flags |= S_DAX;
 	if (!nblks)
 		/* measure inode.i_blocks as generic filesystems */
 		inode->i_blocks = roundup(inode->i_size, EROFS_BLKSIZ) >> 9;
fs/erofs/internal.h +3 -0
index 2669c785d548..7c9abfc93109 100644
@@ -83,6 +83,7 @@ struct erofs_sb_info {
 
 	struct erofs_sb_lz4_info lz4;
 #endif	/* CONFIG_EROFS_FS_ZIP */
+	struct dax_device *dax_dev;
 	u32 blocks;
 	u32 meta_blkaddr;
 #ifdef CONFIG_EROFS_FS_XATTR
@@ -115,6 +116,8 @@ struct erofs_sb_info {
 /* Mount flags set via mount options or defaults */
 #define EROFS_MOUNT_XATTR_USER		0x00000010
 #define EROFS_MOUNT_POSIX_ACL		0x00000020
+#define EROFS_MOUNT_DAX_ALWAYS		0x00000040
+#define EROFS_MOUNT_DAX_NEVER		0x00000080
 
 #define clear_opt(ctx, option)	((ctx)->mount_opt &= ~EROFS_MOUNT_##option)
 #define set_opt(ctx, option)	((ctx)->mount_opt |= EROFS_MOUNT_##option)
fs/erofs/super.c +57 -2

119 lines not on this page · raw message

Signature · 1 line
2.24.4
maintainer Reviewed-by on v3 2/3
#3 lore raw

Re: [PATCH v3 2/3] erofs: dax support for non-tailpacking regular file

Quoted 25 lines · Gao Xiang

DAX is quite useful for some VM use cases in order to save guest memory extremely with minimal lightweight EROFS.

In order to prepare for such use cases, add preliminary dax support for non-tailpacking regular files for now.

Tested with the DRAM-emulated PMEM and the EROFS image generated by "mkfs.erofs -Enoinline_data enwik9.fsdax.img enwik9"

---
  Documentation/filesystems/erofs.rst |  2 +
  fs/erofs/data.c                     | 42 +++++++++++++++++++-
  fs/erofs/inode.c                    |  4 ++
  fs/erofs/internal.h                 |  3 ++
  fs/erofs/super.c                    | 59 ++++++++++++++++++++++++++++-
  5 files changed, 106 insertions(+), 4 deletions(-)
Documentation/filesystems/erofs.rst +0 -0
index 832839fcf4c3..ffd2ae7be511 100644

Documentation/filesystems/erofs.rst cache_strategy=%s Select a strategy for cached decompression from now on: · line 87

Quoted 4 lines
Documentation/filesystems/erofs.rst +1 -0
@@ -84,6 +84,8 @@ cache_strategy=%s      Select a strategy for cached decompression from now on:
                                     It still does in-place I/O decompression
                                     for the rest compressed physical clusters.
  		       ==========  =============================================
+dax                    Use direct access (no page cache).  See

dax or dax=%s

Otherwise, it looks good to me.

Thanks,

author
#4 lore raw

Re: [PATCH v3 2/3] erofs: dax support for non-tailpacking regular file

Hi Chao,

Quoted 33 lines · Chao Yu
Quoted 25 lines · Gao Xiang

DAX is quite useful for some VM use cases in order to save guest memory extremely with minimal lightweight EROFS.

In order to prepare for such use cases, add preliminary dax support for non-tailpacking regular files for now.

Tested with the DRAM-emulated PMEM and the EROFS image generated by "mkfs.erofs -Enoinline_data enwik9.fsdax.img enwik9"

---
  Documentation/filesystems/erofs.rst |  2 +
  fs/erofs/data.c                     | 42 +++++++++++++++++++-
  fs/erofs/inode.c                    |  4 ++
  fs/erofs/internal.h                 |  3 ++
  fs/erofs/super.c                    | 59 ++++++++++++++++++++++++++++-
  5 files changed, 106 insertions(+), 4 deletions(-)
Documentation/filesystems/erofs.rst +0 -0
index 832839fcf4c3..ffd2ae7be511 100644

Documentation/filesystems/erofs.rst cache_strategy=%s Select a strategy for cached decompression from now on: · line 87

Quoted 4 lines
Documentation/filesystems/erofs.rst +1 -0
@@ -84,6 +84,8 @@ cache_strategy=%s      Select a strategy for cached decompression from now on:
                                     It still does in-place I/O decompression
                                     for the rest compressed physical clusters.
  		       ==========  =============================================
+dax                    Use direct access (no page cache).  See

dax or dax=%s

Quoted 4 lines

Otherwise, it looks good to me.

Thanks, it originally inherited from filesystems/ext2.rst, I will update this into
dax, dax={always,never} .....

Since for such image vm-shared memory scenario, no need to add per-file DAX (at least for now.)

Thanks,
Gao Xiang

Quoted 1 line

Thanks,

maintainer
#5 lore raw

Re: [PATCH v3 2/3] erofs: dax support for non-tailpacking regular file

Xiang,

Quoted 3 lines · Gao Xiang

Thanks, it originally inherited from filesystems/ext2.rst, I will update this into
dax, dax={always,never} .....

Above change looks fine to me, thanks.

Quoted 3 lines

Since for such image vm-shared memory scenario, no need to add per-file DAX (at least for now.)

Agreed.

Thanks,

author
#6 lore raw

PATCH v3 3/3 erofs: convert all uncompressed cases to iomap

In reply to Gao Xiang

Since tail-packing inline has been supported by iomap now, let's convert all EROFS uncompressed data I/O to iomap, which is pretty straight-forward.

---
 fs/erofs/data.c | 290 +++++++++---------------------------------------
 1 file changed, 51 insertions(+), 239 deletions(-)
fs/erofs/data.c +51 -239
index 6bbfcf16bf28..a8a483b5caac 100644
@@ -9,29 +9,6 @@
 #include <linux/dax.h>
 #include <trace/events/erofs.h>
 
-static void erofs_readendio(struct bio *bio)
-{
-	struct bio_vec *bvec;
-	blk_status_t err = bio->bi_status;
-	struct bvec_iter_all iter_all;
-
-	bio_for_each_segment_all(bvec, bio, iter_all) {
-		struct page *page = bvec->bv_page;
-
-		/* page is already locked */
-		DBG_BUGON(PageUptodate(page));
-
-		if (err)
-			SetPageError(page);
-		else
-			SetPageUptodate(page);
-
-		unlock_page(page);
-		/* page could be reclaimed now */
-	}
-	bio_put(bio);
-}
-
 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;
@@ -109,206 +86,6 @@ static int erofs_map_blocks_flatmode(struct inode *inode,
 	return err;
 }
 
-static inline struct bio *erofs_read_raw_page(struct bio *bio,
-					      struct address_space *mapping,
-					      struct page *page,
-					      erofs_off_t *last_block,
-					      unsigned int nblocks,
-					      unsigned int *eblks,
-					      bool ra)
-{
-	struct inode *const inode = mapping->host;
-	struct super_block *const sb = inode->i_sb;
-	erofs_off_t current_block = (erofs_off_t)page->index;
-	int err;
-
-	DBG_BUGON(!nblocks);
-
-	if (PageUptodate(page)) {
-		err = 0;
-		goto has_updated;
-	}
-
-	/* note that for readpage case, bio also equals to NULL */
-	if (bio &&
-	    (*last_block + 1 != current_block || !*eblks)) {
-submit_bio_retry:
-		submit_bio(bio);
-		bio = NULL;
-	}
-
-	if (!bio) {
-		struct erofs_map_blocks map = {
-			.m_la = blknr_to_addr(current_block),
-		};
-		erofs_blk_t blknr;
-		unsigned int blkoff;
-
-		err = erofs_map_blocks_flatmode(inode, &map, EROFS_GET_BLOCKS_RAW);
-		if (err)
-			goto err_out;
-
-		/* zero out the holed page */
-		if (!(map.m_flags & EROFS_MAP_MAPPED)) {
-			zero_user_segment(page, 0, PAGE_SIZE);
-			SetPageUptodate(page);
-
-			/* imply err = 0, see erofs_map_blocks */
-			goto has_updated;
-		}
-
-		/* for RAW access mode, m_plen must be equal to m_llen */
-		DBG_BUGON(map.m_plen != map.m_llen);
-
-		blknr = erofs_blknr(map.m_pa);
-		blkoff = erofs_blkoff(map.m_pa);
-
-		/* deal with inline page */
-		if (map.m_flags & EROFS_MAP_META) {
-			void *vsrc, *vto;
-			struct page *ipage;
-
-			DBG_BUGON(map.m_plen > PAGE_SIZE);
-
-			ipage = erofs_get_meta_page(inode->i_sb, blknr);
-
-			if (IS_ERR(ipage)) {
-				err = PTR_ERR(ipage);
-				goto err_out;
-			}
-
-			vsrc = kmap_atomic(ipage);
-			vto = kmap_atomic(page);
-			memcpy(vto, vsrc + blkoff, map.m_plen);
-			memset(vto + map.m_plen, 0, PAGE_SIZE - map.m_plen);
-			kunmap_atomic(vto);
-			kunmap_atomic(vsrc);
-			flush_dcache_page(page);
-
-			SetPageUptodate(page);
-			/* TODO: could we unlock the page earlier? */
-			unlock_page(ipage);
-			put_page(ipage);
-
-			/* imply err = 0, see erofs_map_blocks */
-			goto has_updated;
-		}
-
-		/* pa must be block-aligned for raw reading */
-		DBG_BUGON(erofs_blkoff(map.m_pa));
-
-		/* max # of continuous pages */
-		if (nblocks > DIV_ROUND_UP(map.m_plen, PAGE_SIZE))
-			nblocks = DIV_ROUND_UP(map.m_plen, PAGE_SIZE);
-
-		*eblks = bio_max_segs(nblocks);
-		bio = bio_alloc(GFP_NOIO, *eblks);
-
-		bio->bi_end_io = erofs_readendio;
-		bio_set_dev(bio, sb->s_bdev);
-		bio->bi_iter.bi_sector = (sector_t)blknr <<
-			LOG_SECTORS_PER_BLOCK;
-		bio->bi_opf = REQ_OP_READ | (ra ? REQ_RAHEAD : 0);
-	}
-
-	err = bio_add_page(bio, page, PAGE_SIZE, 0);
-	/* out of the extent or bio is full */
-	if (err < PAGE_SIZE)
-		goto submit_bio_retry;
-	--*eblks;
-	*last_block = current_block;
-	return bio;
-
-err_out:
-	/* for sync reading, set page error immediately */
-	if (!ra) {
-		SetPageError(page);
-		ClearPageUptodate(page);
-	}
-has_updated:
-	unlock_page(page);
-
-	/* if updated manually, continuous pages has a gap */
-	if (bio)
-		submit_bio(bio);
-	return err ? ERR_PTR(err) : NULL;
-}
-
-/*
- * since we dont have write or truncate flows, so no inode
- * locking needs to be held at the moment.
- */
-static int erofs_raw_access_readpage(struct file *file, struct page *page)
-{
-	erofs_off_t last_block;
-	unsigned int eblks;
-	struct bio *bio;
-
-	trace_erofs_readpage(page, true);
-
-	bio = erofs_read_raw_page(NULL, page->mapping,
-				  page, &last_block, 1, &eblks, false);
-
-	if (IS_ERR(bio))
-		return PTR_ERR(bio);
-
-	if (bio)
-		submit_bio(bio);
-	return 0;
-}
-
-static void erofs_raw_access_readahead(struct readahead_control *rac)
-{
-	erofs_off_t last_block;
-	unsigned int eblks;
-	struct bio *bio = NULL;
-	struct page *page;
-
-	trace_erofs_readpages(rac->mapping->host, readahead_index(rac),
-			readahead_count(rac), true);
-
-	while ((page = readahead_page(rac))) {
-		prefetchw(&page->flags);
-
-		bio = erofs_read_raw_page(bio, rac->mapping, page, &last_block,
-				readahead_count(rac), &eblks, true);
-
-		/* all the page errors are ignored when readahead */
-		if (IS_ERR(bio)) {
-			pr_err("%s, readahead error at page %lu of nid %llu\n",
-			       __func__, page->index,
-			       EROFS_I(rac->mapping->host)->nid);
-
-			bio = NULL;
-		}
-
-		put_page(page);
-	}
-
-	if (bio)
-		submit_bio(bio);
-}
-
-static sector_t erofs_bmap(struct address_space *mapping, sector_t block)
-{
-	struct inode *inode = mapping->host;
-	struct erofs_map_blocks map = {
-		.m_la = blknr_to_addr(block),
-	};
-
-	if (EROFS_I(inode)->datalayout == EROFS_INODE_FLAT_INLINE) {
-		erofs_blk_t blks = i_size_read(inode) >> LOG_BLOCK_SIZE;
-
-		if (block >> LOG_SECTORS_PER_BLOCK >= blks)
-			return 0;
-	}
-
-	if (!erofs_map_blocks_flatmode(inode, &map, EROFS_GET_BLOCKS_RAW))
-		return erofs_blknr(map.m_pa);
-
-	return 0;
-}
-
 static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 		unsigned int flags, struct iomap *iomap, struct iomap *srcmap)
 {
@@ -327,6 +104,7 @@ static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 	iomap->offset = map.m_la;
 	iomap->length = map.m_llen;
 	iomap->flags = 0;
+	iomap->private = NULL;
 
 	if (!(map.m_flags & EROFS_MAP_MAPPED)) {
 		iomap->type = IOMAP_HOLE;
@@ -336,20 +114,63 @@ static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 		return 0;
 	}
 
-	/* that shouldn't happen for now */
 	if (map.m_flags & EROFS_MAP_META) {
-		DBG_BUGON(1);
-		return -ENOTBLK;
+		struct page *ipage;
+
+		iomap->type = IOMAP_INLINE;
+		ipage = erofs_get_meta_page(inode->i_sb,
+					    erofs_blknr(map.m_pa));
+		if (IS_ERR(ipage))
+			return PTR_ERR(ipage);
+		iomap->inline_data = page_address(ipage) +
+					erofs_blkoff(map.m_pa);
+		iomap->private = ipage;
+	} else {
+		iomap->type = IOMAP_MAPPED;
+		iomap->addr = map.m_pa;
 	}
-	iomap->type = IOMAP_MAPPED;
-	iomap->addr = map.m_pa;
 	return 0;
 }
 
+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;
+
+	if (ipage) {
+		DBG_BUGON(iomap->type != IOMAP_INLINE);
+		unlock_page(ipage);
+		put_page(ipage);
+	} else {
+		DBG_BUGON(iomap->type == IOMAP_INLINE);
+	}
+	return written;
+}
+
 const struct iomap_ops erofs_iomap_ops = {
 	.iomap_begin = erofs_iomap_begin,
+	.iomap_end = erofs_iomap_end,
 };
 
+/*
+ * since we dont have write or truncate flows, so no inode
+ * locking needs to be held at the moment.
+ */
+static int erofs_readpage(struct file *file, struct page *page)
+{
+	return iomap_readpage(page, &erofs_iomap_ops);
+}
+
+static void erofs_readahead(struct readahead_control *rac)
+{
+	return iomap_readahead(rac, &erofs_iomap_ops);
+}
+
+static sector_t erofs_bmap(struct address_space *mapping, sector_t block)
+{
+	return iomap_bmap(mapping, block, &erofs_iomap_ops);
+}
+
 static int erofs_prepare_dio(struct kiocb *iocb, struct iov_iter *to)
 {
 	struct inode *inode = file_inode(iocb->ki_filp);
@@ -365,15 +186,6 @@ static int erofs_prepare_dio(struct kiocb *iocb, struct iov_iter *to)
 
 	if (align & blksize_mask)
 		return -EINVAL;
-
-	/*
-	 * Temporarily fall back tail-packing inline to buffered I/O instead
-	 * since tail-packing inline support relies on an iomap core update.
-	 */
-	if (EROFS_I(inode)->datalayout == EROFS_INODE_FLAT_INLINE &&
-	    iocb->ki_pos + iov_iter_count(to) >
-			rounddown(inode->i_size, EROFS_BLKSIZ))
-		return 1;
 	return 0;
 }
 
@@ -401,8 +213,8 @@ static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 
 /* for uncompressed (aligned) files and raw access for other files */
 const struct address_space_operations erofs_raw_access_aops = {
-	.readpage = erofs_raw_access_readpage,
-	.readahead = erofs_raw_access_readahead,
+	.readpage = erofs_readpage,
+	.readahead = erofs_readahead,
 	.bmap = erofs_bmap,
 	.direct_IO = noop_direct_IO,
 };
Signature · 1 line
2.24.4