btrfs: add ioctls and send/receive support for reading/writing compressed data

From: Omar Sandoval <[email protected]>

Encoded I/O in Btrfs needs to check a write with a given logical size without an iov_iter that matches that size (because the iov_iter we have is for the compressed data). So, factor out the parts of generic_write_check() that expect an iov_iter into a new __generic_write_checks() function and export that.

Signed-off-by Omar Sandoval <[email protected]>

---
 fs/read_write.c    | 40 ++++++++++++++++++++++------------------
 include/linux/fs.h |  1 +
 2 files changed, 23 insertions(+), 18 deletions(-)

fs/read_write.c +0 -0

index 0029ff2b0ca8..3bddd5ee7f64 100644
@@ -1633,6 +1633,26 @@ int generic_write_check_limits(struct file *file, loff_t pos, loff_t *count)
 	return 0;

Quoted 22 lines · Omar Sandoval

From: Omar Sandoval <[email protected]>

Encoded I/O in Btrfs needs to check a write with a given logical size without an iov_iter that matches that size (because the iov_iter we have is for the compressed data). So, factor out the parts of generic_write_check() that expect an iov_iter into a new __generic_write_checks() function and export that.

Signed-off-by Omar Sandoval <[email protected]>

---
 fs/read_write.c    | 40 ++++++++++++++++++++++------------------
 include/linux/fs.h |  1 +
 2 files changed, 23 insertions(+), 18 deletions(-)

fs/read_write.c +0 -0

index 0029ff2b0ca8..3bddd5ee7f64 100644
@@ -1633,6 +1633,26 @@ int generic_write_check_limits(struct file *file, loff_t pos, loff_t *count)
 	return 0;
 }
 

fs/read_write.c int generic_write_check_limits(struct file *file, loff_t pos, loff_t *count) · line 1643

Quoted 8 lines

fs/read_write.c +8 -0

+/* Like generic_write_checks(), but takes size of write instead of iter. */
+int __generic_write_checks(struct kiocb *iocb, loff_t *count)
+{
+	struct file *file = iocb->ki_filp;
+	struct inode *inode = file->f_mapping->host;
+
+	if (IS_SWAPFILE(inode))
+		return -ETXTBSY;

Missing 'if(!count) return 0' from original code ?

From: Omar Sandoval <[email protected]>

btrfs_csum_one_bio() loops over each filesystem block in the bio while keeping a cursor of its current logical position in the file in order to look up the ordered extent to add the checksums to. However, this doesn't make much sense for compressed extents, as a sector on disk does not correspond to a sector of decompressed file data. It happens to work because 1) the compressed bio always covers one ordered extent and 2) the size of the bio is always less than the size of the ordered extent. However, the second point will not always be true for encoded writes.

Let's add a boolean parameter to btrfs_csum_one_bio() to indicate that it can assume that the bio only covers one ordered extent. Since we're already changing the signature, let's get rid of the contig parameter and make it implied by the offset parameter, similar to the change we recently made to btrfs_lookup_bio_sums(). Additionally, let's rename nr_sectors to blockcount to make it clear that it's the number of filesystem blocks, not the number of 512-byte sectors.

Reviewed-by Josef Bacik <[email protected]>

Signed-off-by Omar Sandoval <[email protected]>

---
 fs/btrfs/compression.c |  5 +++--
 fs/btrfs/ctree.h       |  2 +-
 fs/btrfs/file-item.c   | 35 ++++++++++++++++-------------------
 fs/btrfs/inode.c       |  8 ++++----
 4 files changed, 24 insertions(+), 26 deletions(-)

fs/btrfs/compression.c +3 -2

index 7869ad12bc6e..e645b3c2f09a 100644
@@ -480,7 +480,8 @@ blk_status_t btrfs_submit_compressed_write(struct btrfs_inode *inode, u64 start,
 			BUG_ON(ret); /* -ENOMEM */
 
 			if (!skip_sum) {
-				ret = btrfs_csum_one_bio(inode, bio, start, 1);
+				ret = btrfs_csum_one_bio(inode, bio, start,
+							 true);
 				BUG_ON(ret); /* -ENOMEM */
 			}
 

@@ -516,7 +517,7 @@ blk_status_t btrfs_submit_compressed_write(struct btrfs_inode *inode, u64 start,
 	BUG_ON(ret); /* -ENOMEM */
 
 	if (!skip_sum) {
-		ret = btrfs_csum_one_bio(inode, bio, start, 1);
+		ret = btrfs_csum_one_bio(inode, bio, start, true);
 		BUG_ON(ret); /* -ENOMEM */
 	}
 

fs/btrfs/ctree.h +1 -1

index f07c82fafa04..be245b4b8efe 100644
@@ -3111,7 +3111,7 @@ int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
 			   struct btrfs_root *root,
 			   struct btrfs_ordered_sum *sums);
 blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
-				u64 file_start, int contig);
+				u64 offset, bool one_ordered);
 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
 			     struct list_head *list, int search_commit);
 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,

fs/btrfs/file-item.c +0 -0

index 2673c6ba7a4e..844fae923340 100644
@@ -614,28 +614,28 @@ int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
  * btrfs_csum_one_bio - Calculates checksums of the data contained inside a bio
  * @inode:	 Owner of the data inside the bio
  * @bio:	 Contains the data to be checksummed

fs/btrfs/file-item.c +3 -9

 	char *data;
 	struct bvec_iter iter;
 	struct bio_vec bvec;
 	int index;
-	int nr_sectors;
+	int blockcount;
 	unsigned long total_bytes = 0;
 	unsigned long this_sum_bytes = 0;
 	int i;
-	u64 offset;
 	unsigned nofs_flag;
 
 	nofs_flag = memalloc_nofs_save();

@@ -649,18 +649,13 @@ blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
 	sums->len = bio->bi_iter.bi_size;
 	INIT_LIST_HEAD(&sums->list);
 
-	if (contig)
-		offset = file_start;
-	else
-		offset = 0; /* shut up gcc */
-
 	sums->bytenr = bio->bi_iter.bi_sector << 9;
 	index = 0;
 
 	shash->tfm = fs_info->csum_shash;
 
 	bio_for_each_segment(bvec, bio, iter) {
-		if (!contig)
+		if (page_offsets)
 			offset = page_offset(bvec.bv_page) + bvec.bv_offset;
 
 		if (!ordered) {

@@ -668,13 +663,14 @@ blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
 			BUG_ON(!ordered); /* Logic error */
 		}
 
-		nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
+		blockcount = BTRFS_BYTES_TO_BLKS(fs_info,

fs/btrfs/file-item.c +0 -0

 			this_sum_bytes += fs_info->sectorsize;
 			total_bytes += fs_info->sectorsize;
 		}

Quoted 72 lines · Omar Sandoval

From: Omar Sandoval <[email protected]>

btrfs_csum_one_bio() loops over each filesystem block in the bio while keeping a cursor of its current logical position in the file in order to look up the ordered extent to add the checksums to. However, this doesn't make much sense for compressed extents, as a sector on disk does not correspond to a sector of decompressed file data. It happens to work because 1) the compressed bio always covers one ordered extent and 2) the size of the bio is always less than the size of the ordered extent. However, the second point will not always be true for encoded writes.

Let's add a boolean parameter to btrfs_csum_one_bio() to indicate that it can assume that the bio only covers one ordered extent. Since we're already changing the signature, let's get rid of the contig parameter and make it implied by the offset parameter, similar to the change we recently made to btrfs_lookup_bio_sums(). Additionally, let's rename nr_sectors to blockcount to make it clear that it's the number of filesystem blocks, not the number of 512-byte sectors.

Reviewed-by Josef Bacik <[email protected]>

Signed-off-by Omar Sandoval <[email protected]>

---
 fs/btrfs/compression.c |  5 +++--
 fs/btrfs/ctree.h       |  2 +-
 fs/btrfs/file-item.c   | 35 ++++++++++++++++-------------------
 fs/btrfs/inode.c       |  8 ++++----
 4 files changed, 24 insertions(+), 26 deletions(-)

fs/btrfs/compression.c +3 -2

index 7869ad12bc6e..e645b3c2f09a 100644
@@ -480,7 +480,8 @@ blk_status_t btrfs_submit_compressed_write(struct btrfs_inode *inode, u64 start,
 			BUG_ON(ret); /* -ENOMEM */
 
 			if (!skip_sum) {
-				ret = btrfs_csum_one_bio(inode, bio, start, 1);
+				ret = btrfs_csum_one_bio(inode, bio, start,
+							 true);
 				BUG_ON(ret); /* -ENOMEM */
 			}
 

@@ -516,7 +517,7 @@ blk_status_t btrfs_submit_compressed_write(struct btrfs_inode *inode, u64 start,
 	BUG_ON(ret); /* -ENOMEM */
 
 	if (!skip_sum) {
-		ret = btrfs_csum_one_bio(inode, bio, start, 1);
+		ret = btrfs_csum_one_bio(inode, bio, start, true);
 		BUG_ON(ret); /* -ENOMEM */
 	}
 

fs/btrfs/ctree.h +1 -1

index f07c82fafa04..be245b4b8efe 100644
@@ -3111,7 +3111,7 @@ int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
 			   struct btrfs_root *root,
 			   struct btrfs_ordered_sum *sums);
 blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
-				u64 file_start, int contig);
+				u64 offset, bool one_ordered);
 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
 			     struct list_head *list, int search_commit);
 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,

fs/btrfs/file-item.c +0 -0

index 2673c6ba7a4e..844fae923340 100644
@@ -614,28 +614,28 @@ int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
  * btrfs_csum_one_bio - Calculates checksums of the data contained inside a bio
  * @inode:	 Owner of the data inside the bio
  * @bio:	 Contains the data to be checksummed

fs/btrfs/file-item.c int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end, · line 621

Quoted 10 lines

fs/btrfs/file-item.c +5 -5

- * @file_start:  offset in file this bio begins to describe
- * @contig:	 Boolean. If true/1 means all bio vecs in this bio are
- *		 contiguous and they begin at @file_start in the file. False/0
- *		 means this bio can contain potentially discontiguous bio vecs
- *		 so the logical offset of each should be calculated separately.
+ * @offset:      If (u64)-1, @bio may contain discontiguous bio vecs, so the
+ *               file offsets are determined from the page offsets in the bio.
+ *               Otherwise, this is the starting file offset of the bio vecs in
+ *               @bio, which must be contiguous.
+ * @one_ordered: If true, @bio only refers to one ordered extent.

nit: I don't have strong preference but my gut feeling tells me "single_ordered" might be more explicit/informative. But unless someone else thinks the same one_ordered would also make do

fs/btrfs/file-item.c int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end, · line 630

Quoted 10 lines

fs/btrfs/file-item.c +2 -1

  */
 blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
-		       u64 file_start, int contig)
+				u64 offset, bool one_ordered)
 {
 	struct btrfs_fs_info *fs_info = inode->root->fs_info;
 	SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 	struct btrfs_ordered_sum *sums;
 	struct btrfs_ordered_extent *ordered = NULL;
+	const bool page_offsets = (offset == (u64)-1);

nit: Again, instead of page_offsets perhaps use_page_offsets, somewhat more explicit/informative.

Quoted 39 lines

fs/btrfs/file-item.c +3 -9

 	char *data;
 	struct bvec_iter iter;
 	struct bio_vec bvec;
 	int index;
-	int nr_sectors;
+	int blockcount;
 	unsigned long total_bytes = 0;
 	unsigned long this_sum_bytes = 0;
 	int i;
-	u64 offset;
 	unsigned nofs_flag;
 
 	nofs_flag = memalloc_nofs_save();

@@ -649,18 +649,13 @@ blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
 	sums->len = bio->bi_iter.bi_size;
 	INIT_LIST_HEAD(&sums->list);
 
-	if (contig)
-		offset = file_start;
-	else
-		offset = 0; /* shut up gcc */
-
 	sums->bytenr = bio->bi_iter.bi_sector << 9;
 	index = 0;
 
 	shash->tfm = fs_info->csum_shash;
 
 	bio_for_each_segment(bvec, bio, iter) {
-		if (!contig)
+		if (page_offsets)
 			offset = page_offset(bvec.bv_page) + bvec.bv_offset;
 
 		if (!ordered) {

@@ -668,13 +663,14 @@ blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
 			BUG_ON(!ordered); /* Logic error */
 		}
 
-		nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
+		blockcount = BTRFS_BYTES_TO_BLKS(fs_info,

fs/btrfs/file-item.c blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio, · line 673

Quoted 10 lines

fs/btrfs/file-item.c +4 -3

 						 bvec.bv_len + fs_info->sectorsize
 						 - 1);
 
-		for (i = 0; i < nr_sectors; i++) {
-			if (offset >= ordered->file_offset + ordered->num_bytes ||
-			    offset < ordered->file_offset) {
+		for (i = 0; i < blockcount; i++) {
+			if (!one_ordered &&
+			    (offset >= ordered->file_offset + ordered->num_bytes ||
+			     offset < ordered->file_offset)) {

Since you are changing this hunk, how about using the in_range macro:

if (!one_ordered && !in_range(offset, ordered->file_offset, ordered->num_bytes) { foo

Though I think the "change the ordered extent now that we are working on a different range" code should be factored out in a separate function because currently it's somewhat breaking the flow of reading.

Quoted 2 lines

fs/btrfs/file-item.c +0 -0

 				unsigned long bytes_left;
 

fs/btrfs/file-item.c blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio, · line 705

Quoted 8 lines

fs/btrfs/file-item.c +2 -1

 				sums->len = this_sum_bytes;

@@ -705,7 +701,8 @@ blk_status_t btrfs_csum_one_bio(struct btrfs_inode *inode, struct bio *bio,
 					    sums->sums + index);
 			kunmap_atomic(data);
 			index += fs_info->csum_size;
-			offset += fs_info->sectorsize;
+			if (!one_ordered)
+				offset += fs_info->sectorsize;

Instead of adding one additional conditional op can't offset always be incremented but in the case of one_ordered then the !in_range check should always be false i.e we won't be using the offset to lookup a new OE?

Quoted 3 lines

fs/btrfs/file-item.c +0 -0

 			this_sum_bytes += fs_info->sectorsize;
 			total_bytes += fs_info->sectorsize;
 		}

<snip>

From: Omar Sandoval <[email protected]>

Currently, we only create ordered extents when ram_bytes == num_bytes and offset == 0. However, RWF_ENCODED writes may create extents which only refer to a subset of the full unencoded extent, so we need to plumb

Quoted 5 lines · Omar Sandoval

From: Omar Sandoval <[email protected]>

Currently, we only create ordered extents when ram_bytes == num_bytes and offset == 0. However, RWF_ENCODED writes may create extents which only refer to a subset of the full unencoded extent, so we need to plumb

Can you give an example of such a case?

From: Omar Sandoval <[email protected]>

Currently, an inline extent is always created after i_size is extended from btrfs_dirty_pages(). However, for encoded writes, we only want to update i_size after we successfully created the inline extent. Add an update_i_size parameter to cow_file_range_inline() and insert_inline_extent() and pass in the size of the extent rather than determining it from i_size. Since the start parameter is always passed as 0, get rid of it and simplify the logic in these two functions. While we're here, let's document the requirements for creating an inline extent.

Reviewed-by Josef Bacik <[email protected]>

Signed-off-by Omar Sandoval <[email protected]>

---
 fs/btrfs/inode.c | 100 +++++++++++++++++++++++------------------------
 1 file changed, 48 insertions(+), 52 deletions(-)

fs/btrfs/inode.c +21 -20

index 708d8ab098bc..0b5ff14aa7fd 100644
@@ -236,9 +236,10 @@ static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 static int insert_inline_extent(struct btrfs_trans_handle *trans,
 				struct btrfs_path *path, bool extent_inserted,
 				struct btrfs_root *root, struct inode *inode,
-				u64 start, size_t size, size_t compressed_size,
+				size_t size, size_t compressed_size,
 				int compress_type,
-				struct page **compressed_pages)
+				struct page **compressed_pages,
+				bool update_i_size)
 {
 	struct extent_buffer *leaf;
 	struct page *page = NULL;

@@ -247,7 +248,7 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
 	struct btrfs_file_extent_item *ei;
 	int ret;
 	size_t cur_size = size;
-	unsigned long offset;
+	u64 i_size;
 
 	ASSERT((compressed_size > 0 && compressed_pages) ||
 	       (compressed_size == 0 && !compressed_pages));

@@ -260,7 +261,7 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
 		size_t datasize;
 
 		key.objectid = btrfs_ino(BTRFS_I(inode));
-		key.offset = start;
+		key.offset = 0;
 		key.type = BTRFS_EXTENT_DATA_KEY;
 
 		datasize = btrfs_file_extent_calc_inline_size(cur_size);

@@ -297,12 +298,10 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
 		btrfs_set_file_extent_compression(leaf, ei,
 						  compress_type);
 	} else {
-		page = find_get_page(inode->i_mapping,
-				     start >> PAGE_SHIFT);
+		page = find_get_page(inode->i_mapping, 0);
 		btrfs_set_file_extent_compression(leaf, ei, 0);
 		kaddr = kmap_atomic(page);
-		offset = offset_in_page(start);
-		write_extent_buffer(leaf, kaddr + offset, ptr, size);
+		write_extent_buffer(leaf, kaddr, ptr, size);
 		kunmap_atomic(kaddr);
 		put_page(page);
 	}

@@ -313,8 +312,8 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
 	 * We align size to sectorsize for inline extents just for simplicity
 	 * sake.
 	 */
-	size = ALIGN(size, root->fs_info->sectorsize);
-	ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size);
+	ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
+					ALIGN(size, root->fs_info->sectorsize));
 	if (ret)
 		goto fail;
 

@@ -327,7 +326,13 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
 	 * before we unlock the pages.  Otherwise we
 	 * could end up racing with unlink.
 	 */
-	BTRFS_I(inode)->disk_i_size = inode->i_size;
+	i_size = i_size_read(inode);
+	if (update_i_size && size > i_size) {
+		i_size_write(inode, size);
+		i_size = size;
+	}
+	BTRFS_I(inode)->disk_i_size = i_size;
+
 fail:
 	return ret;
 }

@@ -338,35 +343,31 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
  * does the checks required to make sure the data is small enough
  * to fit as an inline extent.
  */
-static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start,
-					  u64 end, size_t compressed_size,
+static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
+					  size_t compressed_size,
 					  int compress_type,
-					  struct page **compressed_pages)
+					  struct page **compressed_pages,
+					  bool update_i_size)
 {
 	struct btrfs_drop_extents_args drop_args = { 0 };
 	struct btrfs_root *root = inode->root;
 	struct btrfs_fs_info *fs_info = root->fs_info;
 	struct btrfs_trans_handle *trans;
-	u64 isize = i_size_read(&inode->vfs_inode);
-	u64 actual_end = min(end + 1, isize);
-	u64 inline_len = actual_end - start;
-	u64 aligned_end = ALIGN(end, fs_info->sectorsize);
-	u64 data_len = inline_len;
+	u64 data_len = compressed_size ? compressed_size : size;
 	int ret;
 	struct btrfs_path *path;
 
-	if (compressed_size)

Quoted 5 lines · Omar Sandoval

From: Omar Sandoval <[email protected]>

Currently, an inline extent is always created after i_size is extended from btrfs_dirty_pages(). However, for encoded writes, we only want to update i_size after we successfully created the inline extent.

Quoted 126 lines

Add an
update_i_size parameter to cow_file_range_inline() and insert_inline_extent() and pass in the size of the extent rather than determining it from i_size. Since the start parameter is always passed as 0, get rid of it and simplify the logic in these two functions. While we're here, let's document the requirements for creating an inline extent.

Reviewed-by Josef Bacik <[email protected]>

Signed-off-by Omar Sandoval <[email protected]>

---
  fs/btrfs/inode.c | 100 +++++++++++++++++++++++------------------------
  1 file changed, 48 insertions(+), 52 deletions(-)

fs/btrfs/inode.c +26 -24

index 708d8ab098bc..0b5ff14aa7fd 100644
@@ -236,9 +236,10 @@ static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  static int insert_inline_extent(struct btrfs_trans_handle *trans,
  				struct btrfs_path *path, bool extent_inserted,
  				struct btrfs_root *root, struct inode *inode,
-				u64 start, size_t size, size_t compressed_size,
+				size_t size, size_t compressed_size,
  				int compress_type,
-				struct page **compressed_pages)
+				struct page **compressed_pages,
+				bool update_i_size)
  {
  	struct extent_buffer *leaf;
  	struct page *page = NULL;

@@ -247,7 +248,7 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
  	struct btrfs_file_extent_item *ei;
  	int ret;
  	size_t cur_size = size;
-	unsigned long offset;
+	u64 i_size;
 
  	ASSERT((compressed_size > 0 && compressed_pages) ||
  	       (compressed_size == 0 && !compressed_pages));

@@ -260,7 +261,7 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
  		size_t datasize;
 
  		key.objectid = btrfs_ino(BTRFS_I(inode));
-		key.offset = start;
+		key.offset = 0;
  		key.type = BTRFS_EXTENT_DATA_KEY;
 
  		datasize = btrfs_file_extent_calc_inline_size(cur_size);

@@ -297,12 +298,10 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
  		btrfs_set_file_extent_compression(leaf, ei,
  						  compress_type);
  	} else {
-		page = find_get_page(inode->i_mapping,
-				     start >> PAGE_SHIFT);
+		page = find_get_page(inode->i_mapping, 0);
  		btrfs_set_file_extent_compression(leaf, ei, 0);
  		kaddr = kmap_atomic(page);
-		offset = offset_in_page(start);
-		write_extent_buffer(leaf, kaddr + offset, ptr, size);
+		write_extent_buffer(leaf, kaddr, ptr, size);
  		kunmap_atomic(kaddr);
  		put_page(page);
  	}

@@ -313,8 +312,8 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
  	 * We align size to sectorsize for inline extents just for simplicity
  	 * sake.
  	 */
-	size = ALIGN(size, root->fs_info->sectorsize);
-	ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size);
+	ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
+					ALIGN(size, root->fs_info->sectorsize));
  	if (ret)
  		goto fail;
 

@@ -327,7 +326,13 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
  	 * before we unlock the pages.  Otherwise we
  	 * could end up racing with unlink.
  	 */
-	BTRFS_I(inode)->disk_i_size = inode->i_size;
+	i_size = i_size_read(inode);
+	if (update_i_size && size > i_size) {
+		i_size_write(inode, size);
+		i_size = size;
+	}
+	BTRFS_I(inode)->disk_i_size = i_size;
+
  fail:
  	return ret;
  }

@@ -338,35 +343,31 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
   * does the checks required to make sure the data is small enough
   * to fit as an inline extent.
   */
-static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start,
-					  u64 end, size_t compressed_size,
+static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
+					  size_t compressed_size,
  					  int compress_type,
-					  struct page **compressed_pages)
+					  struct page **compressed_pages,
+					  bool update_i_size)
  {
  	struct btrfs_drop_extents_args drop_args = { 0 };
  	struct btrfs_root *root = inode->root;
  	struct btrfs_fs_info *fs_info = root->fs_info;
  	struct btrfs_trans_handle *trans;
-	u64 isize = i_size_read(&inode->vfs_inode);
-	u64 actual_end = min(end + 1, isize);
-	u64 inline_len = actual_end - start;
-	u64 aligned_end = ALIGN(end, fs_info->sectorsize);
-	u64 data_len = inline_len;
+	u64 data_len = compressed_size ? compressed_size : size;
  	int ret;
  	struct btrfs_path *path;
 
-	if (compressed_size)
-		data_len = compressed_size;
-
-	if (start > 0 ||
-	    actual_end > fs_info->sectorsize ||
+	/*
+	 * We can create an inline extent if it ends at or beyond the current
+	 * i_size, is no larger than a sector (decompressed), and the (possibly
+	 * compressed) data fits in a leaf and the configured maximum inline
+	 * size.

fs/btrfs/inode.c static int insert_inline_extent(struct btrfs_trans_handle *trans, · line 365

Quoted 1 line

fs/btrfs/inode.c +1 -0

+	 */

Urgh, just some days ago Qu was talking about how awkward it is to have mixed extents in a file. And now, AFAIU, you are making them more likely since now they can be created not just at the beginning of the file but also after i_size write. While this won't be a problem in and of itself it goes just the opposite way of us trying to shrink the possible cases when we can have mixed extents.

Qu what is your take on that?

<snip>

Quoted 8 lines · Nikolay Borisov

Quoted 5 lines · Omar Sandoval

From: Omar Sandoval <[email protected]>

Currently, an inline extent is always created after i_size is extended from btrfs_dirty_pages(). However, for encoded writes, we only want to update i_size after we successfully created the inline extent.

To me, the idea of write first then update isize is just going to cause tons of inline extent related prblems.

The current example is falloc, which only update the isize after the falloc finishes.

This behavior has already bothered me quite a lot, as it can easily create mixed inline and regular extents.

Can't we remember the old isize (with proper locking), enlarge isize (with holes filled), do the write.

If something wrong happened, we truncate the isize back to its old isize.

Quoted 126 lines

Quoted 126 lines

Add an
update_i_size parameter to cow_file_range_inline() and insert_inline_extent() and pass in the size of the extent rather than determining it from i_size. Since the start parameter is always passed as 0, get rid of it and simplify the logic in these two functions. While we're here, let's document the requirements for creating an inline extent.

Reviewed-by Josef Bacik <[email protected]>

Signed-off-by Omar Sandoval <[email protected]>

---
  fs/btrfs/inode.c | 100 +++++++++++++++++++++++------------------------
  1 file changed, 48 insertions(+), 52 deletions(-)

fs/btrfs/inode.c +26 -24

index 708d8ab098bc..0b5ff14aa7fd 100644
@@ -236,9 +236,10 @@ static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  static int insert_inline_extent(struct btrfs_trans_handle *trans,
  				struct btrfs_path *path, bool extent_inserted,
  				struct btrfs_root *root, struct inode *inode,
-				u64 start, size_t size, size_t compressed_size,
+				size_t size, size_t compressed_size,
  				int compress_type,
-				struct page **compressed_pages)
+				struct page **compressed_pages,
+				bool update_i_size)
  {
  	struct extent_buffer *leaf;
  	struct page *page = NULL;

@@ -247,7 +248,7 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
  	struct btrfs_file_extent_item *ei;
  	int ret;
  	size_t cur_size = size;
-	unsigned long offset;
+	u64 i_size;
 
  	ASSERT((compressed_size > 0 && compressed_pages) ||
  	       (compressed_size == 0 && !compressed_pages));

@@ -260,7 +261,7 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
  		size_t datasize;
 
  		key.objectid = btrfs_ino(BTRFS_I(inode));
-		key.offset = start;
+		key.offset = 0;
  		key.type = BTRFS_EXTENT_DATA_KEY;
 
  		datasize = btrfs_file_extent_calc_inline_size(cur_size);

@@ -297,12 +298,10 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
  		btrfs_set_file_extent_compression(leaf, ei,
  						  compress_type);
  	} else {
-		page = find_get_page(inode->i_mapping,
-				     start >> PAGE_SHIFT);
+		page = find_get_page(inode->i_mapping, 0);
  		btrfs_set_file_extent_compression(leaf, ei, 0);
  		kaddr = kmap_atomic(page);
-		offset = offset_in_page(start);
-		write_extent_buffer(leaf, kaddr + offset, ptr, size);
+		write_extent_buffer(leaf, kaddr, ptr, size);
  		kunmap_atomic(kaddr);
  		put_page(page);
  	}

@@ -313,8 +312,8 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
  	 * We align size to sectorsize for inline extents just for simplicity
  	 * sake.
  	 */
-	size = ALIGN(size, root->fs_info->sectorsize);
-	ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size);
+	ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
+					ALIGN(size, root->fs_info->sectorsize));
  	if (ret)
  		goto fail;
 

@@ -327,7 +326,13 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
  	 * before we unlock the pages.  Otherwise we
  	 * could end up racing with unlink.
  	 */
-	BTRFS_I(inode)->disk_i_size = inode->i_size;
+	i_size = i_size_read(inode);
+	if (update_i_size && size > i_size) {
+		i_size_write(inode, size);
+		i_size = size;
+	}
+	BTRFS_I(inode)->disk_i_size = i_size;
+
  fail:
  	return ret;
  }

@@ -338,35 +343,31 @@ static int insert_inline_extent(struct btrfs_trans_handle *trans,
   * does the checks required to make sure the data is small enough
   * to fit as an inline extent.
   */
-static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start,
-					  u64 end, size_t compressed_size,
+static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
+					  size_t compressed_size,
  					  int compress_type,
-					  struct page **compressed_pages)
+					  struct page **compressed_pages,
+					  bool update_i_size)
  {
  	struct btrfs_drop_extents_args drop_args = { 0 };
  	struct btrfs_root *root = inode->root;
  	struct btrfs_fs_info *fs_info = root->fs_info;
  	struct btrfs_trans_handle *trans;
-	u64 isize = i_size_read(&inode->vfs_inode);
-	u64 actual_end = min(end + 1, isize);
-	u64 inline_len = actual_end - start;
-	u64 aligned_end = ALIGN(end, fs_info->sectorsize);
-	u64 data_len = inline_len;
+	u64 data_len = compressed_size ? compressed_size : size;
  	int ret;
  	struct btrfs_path *path;
 
-	if (compressed_size)
-		data_len = compressed_size;
-
-	if (start > 0 ||
-	    actual_end > fs_info->sectorsize ||
+	/*
+	 * We can create an inline extent if it ends at or beyond the current
+	 * i_size, is no larger than a sector (decompressed), and the (possibly
+	 * compressed) data fits in a leaf and the configured maximum inline
+	 * size.

Quoted 8 lines

fs/btrfs/inode.c static int insert_inline_extent(struct btrfs_trans_handle *trans, · line 365

Quoted 1 line

fs/btrfs/inode.c +1 -0

+	 */

Urgh, just some days ago Qu was talking about how awkward it is to have mixed extents in a file. And now, AFAIU, you are making them more likely since now they can be created not just at the beginning of the file but also after i_size write. While this won't be a problem in and of itself it goes just the opposite way of us trying to shrink the possible cases when we can have mixed extents.

Tree-checker should reject such inline extent at non-zero offset.

Quoted 1 line

Qu what is your take on that?

My question is, why encoded write needs to bother the inline extents at all?

My intuition of such encoded write is, it should not create inline extents at all.

Or is there any special use-case involved for encoded write?

Quoted 18 lines · Qu Wenruo

Quoted 8 lines · Nikolay Borisov

Quoted 5 lines · Omar Sandoval

From: Omar Sandoval <[email protected]>

Currently, an inline extent is always created after i_size is extended from btrfs_dirty_pages(). However, for encoded writes, we only want to update i_size after we successfully created the inline extent.

To me, the idea of write first then update isize is just going to cause tons of inline extent related prblems.

The current example is falloc, which only update the isize after the falloc finishes.

Quoted 2 lines

This behavior has already bothered me quite a lot, as it can easily create mixed inline and regular extents.

Do you have an example of how this would happen? I have the inode and extent bits locked during an encoded write, and I see that fallocate does the same.

Quoted 6 lines

Can't we remember the old isize (with proper locking), enlarge isize (with holes filled), do the write.

If something wrong happened, we truncate the isize back to its old isize.

Quoted 9 lines

Quoted 7 lines

Urgh, just some days ago Qu was talking about how awkward it is to have mixed extents in a file. And now, AFAIU, you are making them more likely since now they can be created not just at the beginning of the file but also after i_size write. While this won't be a problem in and of itself it goes just the opposite way of us trying to shrink the possible cases when we can have mixed extents.

Tree-checker should reject such inline extent at non-zero offset.

This change does not allow creating inline extents at a non-zero offset.

Quoted 4 lines

Quoted 1 line

Qu what is your take on that?

My question is, why encoded write needs to bother the inline extents at all?

Quoted 4 lines

My intuition of such encoded write is, it should not create inline extents at all.

Or is there any special use-case involved for encoded write?

We create compressed inline extents with normal writes. We should be able to send and receive them without converting them into regular extents.

Quoted 19 lines · Omar Sandoval

Quoted 18 lines · Qu Wenruo

Quoted 8 lines · Nikolay Borisov

Quoted 5 lines · Omar Sandoval

From: Omar Sandoval <[email protected]>

Currently, an inline extent is always created after i_size is extended from btrfs_dirty_pages(). However, for encoded writes, we only want to update i_size after we successfully created the inline extent.

To me, the idea of write first then update isize is just going to cause tons of inline extent related prblems.

The current example is falloc, which only update the isize after the falloc finishes.

Quoted 6 lines

Quoted 2 lines

This behavior has already bothered me quite a lot, as it can easily create mixed inline and regular extents.

Do you have an example of how this would happen? I have the inode and extent bits locked during an encoded write, and I see that fallocate does the same.

xfs_io -f -c "pwrite 0 1K" -c "sync" -c "falloc 0 4k" -c "pwrite 4k 4k"

The [0, 1K) will be written as inline without doubt.

Then we go to falloc, it will try to zero the range [1K, 4K), but it doesn't increase the isize.
Thus the page [0, 4k) will still be written back as inline, since isize is still 1K.

Later [4K, 8K) will be written back as regular, causing mixed extents.

Quoted 5 lines

Quoted 5 lines

Can't we remember the old isize (with proper locking), enlarge isize (with holes filled), do the write.

If something wrong happened, we truncate the isize back to its old isize.

[...]

Quoted 16 lines

Quoted 9 lines

Quoted 7 lines

Urgh, just some days ago Qu was talking about how awkward it is to have mixed extents in a file. And now, AFAIU, you are making them more likely since now they can be created not just at the beginning of the file but also after i_size write. While this won't be a problem in and of itself it goes just the opposite way of us trying to shrink the possible cases when we can have mixed extents.

Tree-checker should reject such inline extent at non-zero offset.

This change does not allow creating inline extents at a non-zero offset.

Quoted 4 lines

Quoted 1 line

Qu what is your take on that?

My question is, why encoded write needs to bother the inline extents at all?

Quoted 9 lines

Quoted 4 lines

My intuition of such encoded write is, it should not create inline extents at all.

Or is there any special use-case involved for encoded write?

We create compressed inline extents with normal writes. We should be able to send and receive them without converting them into regular extents.

But my first impression for any encoded write is that, they should work like DIO, thus everything should be sectorsize aligned.

Then why could they create inline extent? As inline extent can only be possible when the isize is smaller than sectorsize.

Quoted 38 lines · Qu Wenruo

Quoted 19 lines · Omar Sandoval

Quoted 18 lines · Qu Wenruo

Quoted 8 lines · Nikolay Borisov

Quoted 5 lines · Omar Sandoval

From: Omar Sandoval <[email protected]>

Currently, an inline extent is always created after i_size is extended from btrfs_dirty_pages(). However, for encoded writes, we only want to update i_size after we successfully created the inline extent.

To me, the idea of write first then update isize is just going to cause tons of inline extent related prblems.

The current example is falloc, which only update the isize after the falloc finishes.

Quoted 6 lines

Quoted 2 lines

This behavior has already bothered me quite a lot, as it can easily create mixed inline and regular extents.

Do you have an example of how this would happen? I have the inode and extent bits locked during an encoded write, and I see that fallocate does the same.

xfs_io -f -c "pwrite 0 1K" -c "sync" -c "falloc 0 4k" -c "pwrite 4k 4k"

The [0, 1K) will be written as inline without doubt.

Then we go to falloc, it will try to zero the range [1K, 4K), but it doesn't increase the isize.
Thus the page [0, 4k) will still be written back as inline, since isize is still 1K.

Quoted 1 line

Later [4K, 8K) will be written back as regular, causing mixed extents.

I'll have to read fallocate more closely to follow what's going on here and figure out if it applies to encoded writes. Please help me out if you see how this would be an issue with encoded writes.

Quoted 32 lines

Quoted 6 lines

Quoted 6 lines

Can't we remember the old isize (with proper locking), enlarge isize (with holes filled), do the write.

If something wrong happened, we truncate the isize back to its old isize.

[...]

Quoted 16 lines

Quoted 9 lines

Quoted 7 lines

Urgh, just some days ago Qu was talking about how awkward it is to have mixed extents in a file. And now, AFAIU, you are making them more likely since now they can be created not just at the beginning of the file but also after i_size write. While this won't be a problem in and of itself it goes just the opposite way of us trying to shrink the possible cases when we can have mixed extents.

Tree-checker should reject such inline extent at non-zero offset.

This change does not allow creating inline extents at a non-zero offset.

Quoted 4 lines

Quoted 1 line

Qu what is your take on that?

My question is, why encoded write needs to bother the inline extents at all?

Quoted 9 lines

Quoted 4 lines

My intuition of such encoded write is, it should not create inline extents at all.

Or is there any special use-case involved for encoded write?

We create compressed inline extents with normal writes. We should be able to send and receive them without converting them into regular extents.

Quoted 5 lines

But my first impression for any encoded write is that, they should work like DIO, thus everything should be sectorsize aligned.

Then why could they create inline extent? As inline extent can only be possible when the isize is smaller than sectorsize.

ENCODED_WRITE is not defined as "O_DIRECT, but encoded". It happens to have some resemblance to O_DIRECT because we have alignment requirements for new extents and because we bypass the page cache, but there's no reason to copy arbitrary restrictions from O_DIRECT. If someone is using ENCODED_WRITE to write compressed data, then they care about space efficiency, so we should make efficient use of inline extents.

Quoted 39 lines · Omar Sandoval

Quoted 38 lines · Qu Wenruo

Quoted 19 lines · Omar Sandoval

Quoted 18 lines · Qu Wenruo

Quoted 8 lines · Nikolay Borisov

Quoted 5 lines · Omar Sandoval

From: Omar Sandoval <[email protected]>

Currently, an inline extent is always created after i_size is extended from btrfs_dirty_pages(). However, for encoded writes, we only want to update i_size after we successfully created the inline extent.

To me, the idea of write first then update isize is just going to cause tons of inline extent related prblems.

The current example is falloc, which only update the isize after the falloc finishes.

Quoted 6 lines

Quoted 2 lines

This behavior has already bothered me quite a lot, as it can easily create mixed inline and regular extents.

Do you have an example of how this would happen? I have the inode and extent bits locked during an encoded write, and I see that fallocate does the same.

xfs_io -f -c "pwrite 0 1K" -c "sync" -c "falloc 0 4k" -c "pwrite 4k 4k"

The [0, 1K) will be written as inline without doubt.

Then we go to falloc, it will try to zero the range [1K, 4K), but it doesn't increase the isize.
Thus the page [0, 4k) will still be written back as inline, since isize is still 1K.

Quoted 5 lines

Quoted 1 line

Later [4K, 8K) will be written back as regular, causing mixed extents.

I'll have to read fallocate more closely to follow what's going on here and figure out if it applies to encoded writes. Please help me out if you see how this would be an issue with encoded writes.

This won't cause anything wrong, if the encoded writes follows the existing inline extents requirement (always at offset 0).

Otherwise, the read path could be affected to handle inlined extent at non-zero offset.

Quoted 38 lines

Quoted 32 lines

Quoted 6 lines

Quoted 6 lines

Can't we remember the old isize (with proper locking), enlarge isize (with holes filled), do the write.

If something wrong happened, we truncate the isize back to its old isize.

[...]

Quoted 16 lines

Quoted 9 lines

Quoted 7 lines

Urgh, just some days ago Qu was talking about how awkward it is to have mixed extents in a file. And now, AFAIU, you are making them more likely since now they can be created not just at the beginning of the file but also after i_size write. While this won't be a problem in and of itself it goes just the opposite way of us trying to shrink the possible cases when we can have mixed extents.

Tree-checker should reject such inline extent at non-zero offset.

This change does not allow creating inline extents at a non-zero offset.

Quoted 4 lines

Quoted 1 line

Qu what is your take on that?

My question is, why encoded write needs to bother the inline extents at all?

Quoted 9 lines

Quoted 4 lines

My intuition of such encoded write is, it should not create inline extents at all.

Or is there any special use-case involved for encoded write?

We create compressed inline extents with normal writes. We should be able to send and receive them without converting them into regular extents.

Quoted 5 lines

But my first impression for any encoded write is that, they should work like DIO, thus everything should be sectorsize aligned.

Then why could they create inline extent? As inline extent can only be possible when the isize is smaller than sectorsize.

Quoted 7 lines

ENCODED_WRITE is not defined as "O_DIRECT, but encoded". It happens to have some resemblance to O_DIRECT because we have alignment requirements for new extents and because we bypass the page cache, but there's no reason to copy arbitrary restrictions from O_DIRECT. If someone is using ENCODED_WRITE to write compressed data, then they care about space efficiency, so we should make efficient use of inline extents.

Then as long as the inline extent requirement for 0 offset is still followed, I'll be fine with that.

But for non-zero offset inline extent? It looks like a much larger change, and may affect read path.

So I'd prefer we keep the 0 offset requirement for inline extent, and find a better way to work around.

From: Omar Sandoval <[email protected]>

In order to allow sending and receiving compressed data without decompressing it, we need an interface to write pre-compressed data directly to the filesystem and the matching interface to read compressed data without decompressing it. This adds the definitions for ioctls to do that and detailed explanations of how to use them.

Signed-off-by Omar Sandoval <[email protected]>

---
 include/uapi/linux/btrfs.h | 132 +++++++++++++++++++++++++++++++++++++
 1 file changed, 132 insertions(+)

include/uapi/linux/btrfs.h +117 -0

index d7d3cfead056..95da52955894 100644
@@ -861,6 +861,134 @@ struct btrfs_ioctl_get_subvol_rootref_args {
 		__u8 align[7];
 };
 
+/*
+ * Data and metadata for an encoded read or write.
+ *
+ * Encoded I/O bypasses any encoding automatically done by the filesystem (e.g.,
+ * compression). This can be used to read the compressed contents of a file or
+ * write pre-compressed data directly to a file.
+ *
+ * BTRFS_IOC_ENCODED_READ and BTRFS_IOC_ENCODED_WRITE are essentially
+ * preadv/pwritev with additional metadata about how the data is encoded and the
+ * size of the unencoded data.
+ *
+ * BTRFS_IOC_ENCODED_READ fills the given iovecs with the encoded data, fills
+ * the metadata fields, and returns the size of the encoded data. It reads one
+ * extent per call. It can also read data which is not encoded.
+ *
+ * BTRFS_IOC_ENCODED_WRITE uses the metadata fields, writes the encoded data
+ * from the iovecs, and returns the size of the encoded data. Note that the
+ * encoded data is not validated when it is written; if it is not valid (e.g.,
+ * it cannot be decompressed), then a subsequent read may return an error.
+ *
+ * Since the filesystem page cache contains decoded data, encoded I/O bypasses
+ * the page cache. Encoded I/O requires CAP_SYS_ADMIN.
+ */
+struct btrfs_ioctl_encoded_io_args {
+	/* Input parameters for both reads and writes. */
+
+	/*
+	 * iovecs containing encoded data.
+	 *
+	 * For reads, if the size of the encoded data is larger than the sum of
+	 * iov[n].iov_len for 0 <= n < iovcnt, then the ioctl fails with
+	 * ENOBUFS.
+	 *
+	 * For writes, the size of the encoded data is the sum of iov[n].iov_len
+	 * for 0 <= n < iovcnt. This must be less than 128 KiB (this limit may
+	 * increase in the future). This must also be less than or equal to
+	 * unencoded_len.
+	 */
+	const struct iovec __user *iov;
+	/* Number of iovecs. */
+	unsigned long iovcnt;
+	/*
+	 * Offset in file.
+	 *
+	 * For writes, must be aligned to the sector size of the filesystem.
+	 */
+	__s64 offset;
+	/* Currently must be zero. */
+	__u64 flags;
+
+	/*
+	 * For reads, the following members are filled in with the metadata for
+	 * the encoded data.
+	 * For writes, the following members must be set to the metadata for the
+	 * encoded data.
+	 */
+
+	/*
+	 * Length of the data in the file.
+	 *
+	 * Must be less than or equal to unencoded_len - unencoded_offset. For
+	 * writes, must be aligned to the sector size of the filesystem unless
+	 * the data ends at or beyond the current end of the file.
+	 */
+	__u64 len;
+	/*
+	 * Length of the unencoded (i.e., decrypted and decompressed) data.
+	 *
+	 * For writes, must be no more than 128 KiB (this limit may increase in
+	 * the future). If the unencoded data is actually longer than
+	 * unencoded_len, then it is truncated; if it is shorter, then it is
+	 * extended with zeroes.
+	 */
+	__u64 unencoded_len;
+	/*
+	 * Offset from the first byte of the unencoded data to the first byte of
+	 * logical data in the file.
+	 *
+	 * Must be less than unencoded_len.
+	 */
+	__u64 unencoded_offset;
+	/*
+	 * BTRFS_ENCODED_IO_COMPRESSION_* type.
+	 *
+	 * For writes, must not be BTRFS_ENCODED_IO_COMPRESSION_NONE.
+	 */
+	__u32 compression;
+	/* Currently always BTRFS_ENCODED_IO_ENCRYPTION_NONE. */
+	__u32 encryption;
+	/*
+	 * Reserved for future expansion.
+	 *
+	 * For reads, always returned as zero. Users should check for non-zero
+	 * bytes. If there are any, then the kernel has a newer version of this
+	 * structure with additional information that the user definition is
+	 * missing.
+	 *
+	 * For writes, must be zeroed.
+	 */
+	__u8 reserved[32];
+};
+
+/* Data is not compressed. */
+#define BTRFS_ENCODED_IO_COMPRESSION_NONE 0
+/* Data is compressed as a single zlib stream. */
+#define BTRFS_ENCODED_IO_COMPRESSION_ZLIB 1
+/*
+ * Data is compressed as a single zstd frame with the windowLog compression
+ * parameter set to no more than 17.
+ */
+#define BTRFS_ENCODED_IO_COMPRESSION_ZSTD 2
+/*
+ * Data is compressed page by page (using the page size indicated by the name of
+ * the constant) with LZO1X and wrapped in the format documented in
+ * fs/btrfs/lzo.c. For writes, the compression page size must match the
+ * filesystem page size.
+ */

include/uapi/linux/btrfs.h +5 -0

+
 /* Error codes as returned by the kernel */
 enum btrfs_err_code {
 	BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET = 1,

@@ -989,5 +1117,9 @@ enum btrfs_err_code {
 				struct btrfs_ioctl_ino_lookup_user_args)
 #define BTRFS_IOC_SNAP_DESTROY_V2 _IOW(BTRFS_IOCTL_MAGIC, 63, \
 				struct btrfs_ioctl_vol_args_v2)
+#define BTRFS_IOC_ENCODED_READ _IOR(BTRFS_IOCTL_MAGIC, 64, \
+				    struct btrfs_ioctl_encoded_io_args)
+#define BTRFS_IOC_ENCODED_WRITE _IOW(BTRFS_IOCTL_MAGIC, 64, \
+				     struct btrfs_ioctl_encoded_io_args)
 
 #endif /* _UAPI_LINUX_BTRFS_H */

Quoted 142 lines · Omar Sandoval

From: Omar Sandoval <[email protected]>

In order to allow sending and receiving compressed data without decompressing it, we need an interface to write pre-compressed data directly to the filesystem and the matching interface to read compressed data without decompressing it. This adds the definitions for ioctls to do that and detailed explanations of how to use them.

Signed-off-by Omar Sandoval <[email protected]>

---
 include/uapi/linux/btrfs.h | 132 +++++++++++++++++++++++++++++++++++++
 1 file changed, 132 insertions(+)

include/uapi/linux/btrfs.h +121 -0

index d7d3cfead056..95da52955894 100644
@@ -861,6 +861,134 @@ struct btrfs_ioctl_get_subvol_rootref_args {
 		__u8 align[7];
 };
 
+/*
+ * Data and metadata for an encoded read or write.
+ *
+ * Encoded I/O bypasses any encoding automatically done by the filesystem (e.g.,
+ * compression). This can be used to read the compressed contents of a file or
+ * write pre-compressed data directly to a file.
+ *
+ * BTRFS_IOC_ENCODED_READ and BTRFS_IOC_ENCODED_WRITE are essentially
+ * preadv/pwritev with additional metadata about how the data is encoded and the
+ * size of the unencoded data.
+ *
+ * BTRFS_IOC_ENCODED_READ fills the given iovecs with the encoded data, fills
+ * the metadata fields, and returns the size of the encoded data. It reads one
+ * extent per call. It can also read data which is not encoded.
+ *
+ * BTRFS_IOC_ENCODED_WRITE uses the metadata fields, writes the encoded data
+ * from the iovecs, and returns the size of the encoded data. Note that the
+ * encoded data is not validated when it is written; if it is not valid (e.g.,
+ * it cannot be decompressed), then a subsequent read may return an error.
+ *
+ * Since the filesystem page cache contains decoded data, encoded I/O bypasses
+ * the page cache. Encoded I/O requires CAP_SYS_ADMIN.
+ */
+struct btrfs_ioctl_encoded_io_args {
+	/* Input parameters for both reads and writes. */
+
+	/*
+	 * iovecs containing encoded data.
+	 *
+	 * For reads, if the size of the encoded data is larger than the sum of
+	 * iov[n].iov_len for 0 <= n < iovcnt, then the ioctl fails with
+	 * ENOBUFS.
+	 *
+	 * For writes, the size of the encoded data is the sum of iov[n].iov_len
+	 * for 0 <= n < iovcnt. This must be less than 128 KiB (this limit may
+	 * increase in the future). This must also be less than or equal to
+	 * unencoded_len.
+	 */
+	const struct iovec __user *iov;
+	/* Number of iovecs. */
+	unsigned long iovcnt;
+	/*
+	 * Offset in file.
+	 *
+	 * For writes, must be aligned to the sector size of the filesystem.
+	 */
+	__s64 offset;
+	/* Currently must be zero. */
+	__u64 flags;
+
+	/*
+	 * For reads, the following members are filled in with the metadata for
+	 * the encoded data.
+	 * For writes, the following members must be set to the metadata for the
+	 * encoded data.
+	 */
+
+	/*
+	 * Length of the data in the file.
+	 *
+	 * Must be less than or equal to unencoded_len - unencoded_offset. For
+	 * writes, must be aligned to the sector size of the filesystem unless
+	 * the data ends at or beyond the current end of the file.
+	 */
+	__u64 len;
+	/*
+	 * Length of the unencoded (i.e., decrypted and decompressed) data.
+	 *
+	 * For writes, must be no more than 128 KiB (this limit may increase in
+	 * the future). If the unencoded data is actually longer than
+	 * unencoded_len, then it is truncated; if it is shorter, then it is
+	 * extended with zeroes.
+	 */
+	__u64 unencoded_len;
+	/*
+	 * Offset from the first byte of the unencoded data to the first byte of
+	 * logical data in the file.
+	 *
+	 * Must be less than unencoded_len.
+	 */
+	__u64 unencoded_offset;
+	/*
+	 * BTRFS_ENCODED_IO_COMPRESSION_* type.
+	 *
+	 * For writes, must not be BTRFS_ENCODED_IO_COMPRESSION_NONE.
+	 */
+	__u32 compression;
+	/* Currently always BTRFS_ENCODED_IO_ENCRYPTION_NONE. */
+	__u32 encryption;
+	/*
+	 * Reserved for future expansion.
+	 *
+	 * For reads, always returned as zero. Users should check for non-zero
+	 * bytes. If there are any, then the kernel has a newer version of this
+	 * structure with additional information that the user definition is
+	 * missing.
+	 *
+	 * For writes, must be zeroed.
+	 */
+	__u8 reserved[32];
+};
+
+/* Data is not compressed. */
+#define BTRFS_ENCODED_IO_COMPRESSION_NONE 0
+/* Data is compressed as a single zlib stream. */
+#define BTRFS_ENCODED_IO_COMPRESSION_ZLIB 1
+/*
+ * Data is compressed as a single zstd frame with the windowLog compression
+ * parameter set to no more than 17.
+ */
+#define BTRFS_ENCODED_IO_COMPRESSION_ZSTD 2
+/*
+ * Data is compressed page by page (using the page size indicated by the name of
+ * the constant) with LZO1X and wrapped in the format documented in
+ * fs/btrfs/lzo.c. For writes, the compression page size must match the
+ * filesystem page size.
+ */
+#define BTRFS_ENCODED_IO_COMPRESSION_LZO_4K 3
+#define BTRFS_ENCODED_IO_COMPRESSION_LZO_8K 4
+#define BTRFS_ENCODED_IO_COMPRESSION_LZO_16K 5
+#define BTRFS_ENCODED_IO_COMPRESSION_LZO_32K 6

include/uapi/linux/btrfs.h struct btrfs_ioctl_get_subvol_rootref_args { · line 990

Quoted 6 lines

include/uapi/linux/btrfs.h +6 -0

+#define BTRFS_ENCODED_IO_COMPRESSION_LZO_64K 7
+#define BTRFS_ENCODED_IO_COMPRESSION_TYPES 8
+
+/* Data is not encrypted. */
+#define BTRFS_ENCODED_IO_ENCRYPTION_NONE 0
+#define BTRFS_ENCODED_IO_ENCRYPTION_TYPES 1

How about an enums for encryption/compression.

From: Omar Sandoval <[email protected]>

There are 4 main cases:

1. Inline extents: we copy the data straight out of the extent buffer.
2. Hole/preallocated extents: we fill in zeroes.
3. Regular, uncompressed extents: we read the sectors we need directly from disk.
4. Regular, compressed extents: we read the entire compressed extent from disk and indicate what subset of the decompressed extent is in the file.

This initial implementation simplifies a few things that can be improved in the future:

- We hold the inode lock during the operation.
- Cases 1, 3, and 4 allocate temporary memory to read into before copying out to userspace.
- We don't do read repair, because it turns out that read repair is currently broken for compressed data.

Signed-off-by Omar Sandoval <[email protected]>

---
 fs/btrfs/ctree.h |   4 +
 fs/btrfs/inode.c | 489 +++++++++++++++++++++++++++++++++++++++++++++++
 fs/btrfs/ioctl.c | 111 +++++++++++
 3 files changed, 604 insertions(+)

fs/btrfs/ctree.h +4 -0

index 973616d80080..b68d8ea42f6e 100644
@@ -3219,6 +3219,10 @@ int btrfs_writepage_cow_fixup(struct page *page);
 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
 					  struct page *page, u64 start,
 					  u64 end, bool uptodate);
+struct btrfs_ioctl_encoded_io_args;
+ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
+			   struct btrfs_ioctl_encoded_io_args *encoded);
+
 extern const struct dentry_operations btrfs_dentry_operations;
 extern const struct iomap_ops btrfs_dio_iomap_ops;
 extern const struct iomap_dio_ops btrfs_dio_ops;

fs/btrfs/inode.c +357 -0

index 0b5ff14aa7fd..6d7fae859fb5 100644
@@ -10496,6 +10496,495 @@ void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
 	}
 }
 
+static int btrfs_encoded_io_compression_from_extent(int compress_type)
+{
+	switch (compress_type) {
+	case BTRFS_COMPRESS_NONE:
+		return BTRFS_ENCODED_IO_COMPRESSION_NONE;
+	case BTRFS_COMPRESS_ZLIB:
+		return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
+	case BTRFS_COMPRESS_LZO:
+		/*
+		 * The LZO format depends on the page size. 64k is the maximum
+		 * sectorsize (and thus page size) that we support.
+		 */
+		if (PAGE_SIZE < SZ_4K || PAGE_SIZE > SZ_64K)
+			return -EINVAL;
+		return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + (PAGE_SHIFT - 12);
+	case BTRFS_COMPRESS_ZSTD:
+		return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
+	default:
+		return -EUCLEAN;
+	}
+}
+
+static ssize_t btrfs_encoded_read_inline(
+				struct kiocb *iocb,
+				struct iov_iter *iter, u64 start,
+				u64 lockend,
+				struct extent_state **cached_state,
+				u64 extent_start, size_t count,
+				struct btrfs_ioctl_encoded_io_args *encoded,
+				bool *unlocked)
+{
+	struct inode *inode = file_inode(iocb->ki_filp);
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	struct btrfs_file_extent_item *item;
+	u64 ram_bytes;
+	unsigned long ptr;
+	void *tmp;
+	ssize_t ret;
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		ret = -ENOMEM;
+		goto out;
+	}
+	ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root, path,
+				       btrfs_ino(BTRFS_I(inode)), extent_start,
+				       0);
+	if (ret) {
+		if (ret > 0) {
+			/* The extent item disappeared? */
+			ret = -EIO;
+		}
+		goto out;
+	}
+	leaf = path->nodes[0];
+	item = btrfs_item_ptr(leaf, path->slots[0],
+			      struct btrfs_file_extent_item);
+
+	ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
+	ptr = btrfs_file_extent_inline_start(item);
+
+	encoded->len = (min_t(u64, extent_start + ram_bytes, inode->i_size) -
+			iocb->ki_pos);
+	ret = btrfs_encoded_io_compression_from_extent(
+				 btrfs_file_extent_compression(leaf, item));
+	if (ret < 0)
+		goto out;
+	encoded->compression = ret;
+	if (encoded->compression) {
+		size_t inline_size;
+
+		inline_size = btrfs_file_extent_inline_item_len(leaf,
+						btrfs_item_nr(path->slots[0]));
+		if (inline_size > count) {
+			ret = -ENOBUFS;
+			goto out;
+		}
+		count = inline_size;
+		encoded->unencoded_len = ram_bytes;
+		encoded->unencoded_offset = iocb->ki_pos - extent_start;
+	} else {
+		encoded->len = encoded->unencoded_len = count =
+			min_t(u64, count, encoded->len);
+		ptr += iocb->ki_pos - extent_start;
+	}
+
+	tmp = kmalloc(count, GFP_NOFS);
+	if (!tmp) {
+		ret = -ENOMEM;
+		goto out;
+	}
+	read_extent_buffer(leaf, tmp, ptr, count);
+	btrfs_release_path(path);
+	unlock_extent_cached(io_tree, start, lockend, cached_state);
+	inode_unlock_shared(inode);
+	*unlocked = true;
+
+	ret = copy_to_iter(tmp, count, iter);
+	if (ret != count)
+		ret = -EFAULT;
+	kfree(tmp);
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+struct btrfs_encoded_read_private {
+	struct inode *inode;
+	wait_queue_head_t wait;
+	atomic_t pending;
+	blk_status_t status;
+	bool skip_csum;
+};
+
+static blk_status_t submit_encoded_read_bio(struct inode *inode,
+					    struct bio *bio, int mirror_num,
+					    unsigned long bio_flags)
+{
+	struct btrfs_encoded_read_private *priv = bio->bi_private;
+	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
+	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
+	blk_status_t ret;
+
+	if (!priv->skip_csum) {
+		ret = btrfs_lookup_bio_sums(inode, bio, NULL);
+		if (ret)
+			return ret;
+	}
+
+	ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
+	if (ret) {
+		btrfs_io_bio_free_csum(io_bio);
+		return ret;
+	}
+
+	atomic_inc(&priv->pending);
+	ret = btrfs_map_bio(fs_info, bio, mirror_num);
+	if (ret) {
+		atomic_dec(&priv->pending);
+		btrfs_io_bio_free_csum(io_bio);
+	}
+	return ret;
+}
+
+static blk_status_t btrfs_encoded_read_check_bio(struct btrfs_io_bio *io_bio)
+{
+	const bool uptodate = io_bio->bio.bi_status == BLK_STS_OK;
+	struct btrfs_encoded_read_private *priv = io_bio->bio.bi_private;
+	struct inode *inode = priv->inode;
+	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
+	u32 sectorsize = fs_info->sectorsize;
+	struct bio_vec *bvec;
+	struct bvec_iter_all iter_all;
+	u64 start = io_bio->logical;
+	u32 bio_offset = 0;
+
+	if (priv->skip_csum || !uptodate)
+		return io_bio->bio.bi_status;
+
+	bio_for_each_segment_all(bvec, &io_bio->bio, iter_all) {
+		unsigned int i, nr_sectors, pgoff;
+
+		nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
+		pgoff = bvec->bv_offset;
+		for (i = 0; i < nr_sectors; i++) {
+			ASSERT(pgoff < PAGE_SIZE);
+			if (check_data_csum(inode, io_bio, bio_offset,
+					    bvec->bv_page, pgoff, start))
+				return BLK_STS_IOERR;
+			start += sectorsize;
+			bio_offset += sectorsize;
+			pgoff += sectorsize;
+		}
+	}
+	return BLK_STS_OK;
+}
+
+static void btrfs_encoded_read_endio(struct bio *bio)
+{
+	struct btrfs_encoded_read_private *priv = bio->bi_private;
+	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
+	blk_status_t status;
+
+	status = btrfs_encoded_read_check_bio(io_bio);
+	if (status) {
+		/*
+		 * The memory barrier implied by the atomic_dec_return() here
+		 * pairs with the memory barrier implied by the
+		 * atomic_dec_return() or io_wait_event() in
+		 * btrfs_encoded_read_regular_fill_pages() to ensure that this
+		 * write is observed before the load of status in
+		 * btrfs_encoded_read_regular_fill_pages().
+		 */
+		WRITE_ONCE(priv->status, status);
+	}
+	if (!atomic_dec_return(&priv->pending))
+		wake_up(&priv->wait);
+	btrfs_io_bio_free_csum(io_bio);
+	bio_put(bio);
+}
+
+static int btrfs_encoded_read_regular_fill_pages(struct inode *inode, u64 offset,
+						 u64 disk_io_size, struct page **pages)
+{
+	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
+	struct btrfs_encoded_read_private priv = {
+		.inode = inode,
+		.pending = ATOMIC_INIT(1),
+		.skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM,
+	};
+	unsigned long i = 0;
+	u64 cur = 0;
+	int ret;
+
+	init_waitqueue_head(&priv.wait);
+	/*
+	 * Submit bios for the extent, splitting due to bio or stripe limits as
+	 * necessary.
+	 */
+	while (cur < disk_io_size) {
+		struct extent_map *em;
+		struct btrfs_io_geometry geom;
+		struct bio *bio = NULL;
+		u64 remaining;
+
+		em = btrfs_get_chunk_map(fs_info, offset + cur,
+					 disk_io_size - cur);
+		if (IS_ERR(em)) {
+			ret = PTR_ERR(em);
+		} else {
+			ret = btrfs_get_io_geometry(fs_info, em, BTRFS_MAP_READ,
+						    offset + cur, &geom);
+			free_extent_map(em);
+		}
+		if (ret) {
+			WRITE_ONCE(priv.status, errno_to_blk_status(ret));
+			break;
+		}
+		remaining = min(geom.len, disk_io_size - cur);
+		while (bio || remaining) {
+			size_t bytes = min_t(u64, remaining, PAGE_SIZE);
+
+			if (!bio) {
+				bio = btrfs_bio_alloc(offset + cur);
+				bio->bi_end_io = btrfs_encoded_read_endio;
+				bio->bi_private = &priv;
+				bio->bi_opf = REQ_OP_READ;
+			}
+
+			if (!bytes ||
+			    bio_add_page(bio, pages[i], bytes, 0) < bytes) {
+				blk_status_t status;
+
+				status = submit_encoded_read_bio(inode, bio, 0,
+								 0);
+				if (status) {
+					WRITE_ONCE(priv.status, status);
+					bio_put(bio);
+					goto out;
+				}
+				bio = NULL;
+				continue;
+			}
+
+			i++;
+			cur += bytes;
+			remaining -= bytes;
+		}
+	}
+
+out:
+	if (atomic_dec_return(&priv.pending))
+		io_wait_event(priv.wait, !atomic_read(&priv.pending));
+	/* See btrfs_encoded_read_endio() for ordering. */
+	return blk_status_to_errno(READ_ONCE(priv.status));
+}
+
+static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
+					  struct iov_iter *iter,
+					  u64 start, u64 lockend,
+					  struct extent_state **cached_state,
+					  u64 offset, u64 disk_io_size,
+					  size_t count, bool compressed,
+					  bool *unlocked)
+{
+	struct inode *inode = file_inode(iocb->ki_filp);
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	struct page **pages;
+	unsigned long nr_pages, i;
+	u64 cur;
+	size_t page_offset;
+	ssize_t ret;
+
+	nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
+	pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
+	if (!pages)
+		return -ENOMEM;
+	for (i = 0; i < nr_pages; i++) {
+		pages[i] = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
+		if (!pages[i]) {
+			ret = -ENOMEM;
+			goto out;
+		}
+	}
+
+	ret = btrfs_encoded_read_regular_fill_pages(inode, offset, disk_io_size,
+						    pages);
+	if (ret)
+		goto out;
+
+	unlock_extent_cached(io_tree, start, lockend, cached_state);
+	inode_unlock_shared(inode);
+	*unlocked = true;
+
+	if (compressed) {
+		i = 0;
+		page_offset = 0;
+	} else {
+		i = (iocb->ki_pos - start) >> PAGE_SHIFT;
+		page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
+	}
+	cur = 0;
+	while (cur < count) {
+		size_t bytes = min_t(size_t, count - cur,
+				     PAGE_SIZE - page_offset);
+
+		if (copy_page_to_iter(pages[i], page_offset, bytes,
+				      iter) != bytes) {
+			ret = -EFAULT;
+			goto out;
+		}
+		i++;
+		cur += bytes;
+		page_offset = 0;
+	}
+	ret = count;
+out:
+	for (i = 0; i < nr_pages; i++) {
+		if (pages[i])
+			__free_page(pages[i]);
+	}
+	kfree(pages);
+	return ret;
+}
+
+ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
+			   struct btrfs_ioctl_encoded_io_args *encoded)
+{
+	struct inode *inode = file_inode(iocb->ki_filp);
+	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	ssize_t ret;
+	size_t count = iov_iter_count(iter);
+	u64 start, lockend, offset, disk_io_size;
+	struct extent_state *cached_state = NULL;

fs/btrfs/inode.c +14 -0

+	}
+	start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
+	/*
+	 * We don't know how long the extent containing iocb->ki_pos is, but if
+	 * it's compressed we know that it won't be longer than this.
+	 */
+	lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
+
+	for (;;) {
+		struct btrfs_ordered_extent *ordered;
+
+		ret = btrfs_wait_ordered_range(inode, start,
+					       lockend - start + 1);
+		if (ret)

fs/btrfs/inode.c +98 -0

+
+	em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start,
+			      lockend - start + 1);
+	if (IS_ERR(em)) {
+		ret = PTR_ERR(em);
+		goto out_unlock_extent;
+	}
+
+	if (em->block_start == EXTENT_MAP_INLINE) {
+		u64 extent_start = em->start;
+
+		/*
+		 * For inline extents we get everything we need out of the
+		 * extent item.
+		 */
+		free_extent_map(em);
+		em = NULL;
+		ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
+						&cached_state, extent_start,
+						count, encoded, &unlocked);
+		goto out;
+	}
+
+	/*
+	 * We only want to return up to EOF even if the extent extends beyond
+	 * that.
+	 */
+	encoded->len = (min_t(u64, extent_map_end(em), inode->i_size) -
+			iocb->ki_pos);
+	if (em->block_start == EXTENT_MAP_HOLE ||
+	    test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
+		offset = EXTENT_MAP_HOLE;
+		encoded->len = encoded->unencoded_len = count =
+			min_t(u64, count, encoded->len);
+	} else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
+		offset = em->block_start;
+		/*
+		 * Bail if the buffer isn't large enough to return the whole
+		 * compressed extent.
+		 */
+		if (em->block_len > count) {
+			ret = -ENOBUFS;
+			goto out_em;
+		}
+		disk_io_size = count = em->block_len;
+		encoded->unencoded_len = em->ram_bytes;
+		encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
+		ret = btrfs_encoded_io_compression_from_extent(
+							     em->compress_type);
+		if (ret < 0)
+			goto out_em;
+		encoded->compression = ret;
+	} else {
+		offset = em->block_start + (start - em->start);
+		if (encoded->len > count)
+			encoded->len = count;
+		/*
+		 * Don't read beyond what we locked. This also limits the page
+		 * allocations that we'll do.
+		 */
+		disk_io_size = min(lockend + 1,
+				   iocb->ki_pos + encoded->len) - start;
+		encoded->len = encoded->unencoded_len = count =
+			start + disk_io_size - iocb->ki_pos;
+		disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
+	}
+	free_extent_map(em);
+	em = NULL;
+
+	if (offset == EXTENT_MAP_HOLE) {
+		unlock_extent_cached(io_tree, start, lockend, &cached_state);
+		inode_unlock_shared(inode);
+		unlocked = true;
+		ret = iov_iter_zero(count, iter);
+		if (ret != count)
+			ret = -EFAULT;
+	} else {
+		ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
+						 &cached_state, offset,
+						 disk_io_size, count,
+						 encoded->compression,
+						 &unlocked);
+	}
+
+out:
+	if (ret >= 0)
+		iocb->ki_pos += encoded->len;
+out_em:
+	free_extent_map(em);
+out_unlock_extent:
+	if (!unlocked)
+		unlock_extent_cached(io_tree, start, lockend, &cached_state);
+out_unlock_inode:
+	if (!unlocked)
+		inode_unlock_shared(inode);
+	return ret;
+}
+
 #ifdef CONFIG_SWAP
 /*
  * Add an entry indicating a block group or device which is pinned by a

fs/btrfs/inode.c +84 -0

+	size_t copy_end_kernel = offsetofend(struct btrfs_ioctl_encoded_io_args,
+					     flags);
+	size_t copy_end;
+	struct iovec iovstack[UIO_FASTIOV];
+	struct iovec *iov = iovstack;
+	struct iov_iter iter;
+	loff_t pos;
+	struct kiocb kiocb;
+	ssize_t ret;
+
+	if (!capable(CAP_SYS_ADMIN)) {
+		ret = -EPERM;
+		goto out_acct;
+	}
+
+	if (compat) {
+#if defined(CONFIG_64BIT) && defined(CONFIG_COMPAT)
+		struct btrfs_ioctl_encoded_io_args_32 args32;
+
+		copy_end = offsetofend(struct btrfs_ioctl_encoded_io_args_32,
+				       flags);
+		if (copy_from_user(&args32, argp, copy_end)) {
+			ret = -EFAULT;
+			goto out_acct;
+		}
+		args.iov = compat_ptr(args32.iov);
+		args.iovcnt = args32.iovcnt;
+		args.offset = args32.offset;
+		args.flags = args32.flags;
+#else
+		return -ENOTTY;
+#endif
+	} else {
+		copy_end = copy_end_kernel;
+		if (copy_from_user(&args, argp, copy_end)) {
+			ret = -EFAULT;
+			goto out_acct;
+		}
+	}
+	if (args.flags != 0) {
+		ret = -EINVAL;
+		goto out_acct;
+	}
+	memset((char *)&args + copy_end_kernel, 0,
+	       sizeof(args) - copy_end_kernel);
+
+	ret = import_iovec(READ, args.iov, args.iovcnt, ARRAY_SIZE(iovstack),
+			   &iov, &iter);
+	if (ret < 0)
+		goto out_acct;
+
+	if (iov_iter_count(&iter) == 0) {
+		ret = 0;
+		goto out_iov;
+	}
+	pos = args.offset;
+	ret = rw_verify_area(READ, file, &pos, args.len);
+	if (ret < 0)
+		goto out_iov;
+
+	init_sync_kiocb(&kiocb, file);
+	ret = kiocb_set_rw_flags(&kiocb, 0);
+	if (ret)
+		goto out_iov;
+	kiocb.ki_pos = pos;
+
+	ret = btrfs_encoded_read(&kiocb, &iter, &args);
+	if (ret >= 0) {
+		fsnotify_access(file);
+		if (copy_to_user(argp + copy_end,
+				 (char *)&args + copy_end_kernel,
+				 sizeof(args) - copy_end_kernel))
+			ret = -EFAULT;
+	}
+
+out_iov:
+	kfree(iov);
+out_acct:
+	if (ret > 0)
+		add_rchar(current, ret);
+	inc_syscr(current);
+	return ret;
+}
+
 long btrfs_ioctl(struct file *file, unsigned int
 		cmd, unsigned long arg)
 {

Quoted 414 lines · Omar Sandoval

From: Omar Sandoval <[email protected]>

There are 4 main cases:

1. Inline extents: we copy the data straight out of the extent buffer.
2. Hole/preallocated extents: we fill in zeroes.
3. Regular, uncompressed extents: we read the sectors we need directly from disk.
4. Regular, compressed extents: we read the entire compressed extent from disk and indicate what subset of the decompressed extent is in the file.

This initial implementation simplifies a few things that can be improved in the future:

- We hold the inode lock during the operation.
- Cases 1, 3, and 4 allocate temporary memory to read into before copying out to userspace.
- We don't do read repair, because it turns out that read repair is currently broken for compressed data.

Signed-off-by Omar Sandoval <[email protected]>

---
 fs/btrfs/ctree.h |   4 +
 fs/btrfs/inode.c | 489 +++++++++++++++++++++++++++++++++++++++++++++++
 fs/btrfs/ioctl.c | 111 +++++++++++
 3 files changed, 604 insertions(+)

fs/btrfs/ctree.h +4 -0

index 973616d80080..b68d8ea42f6e 100644
@@ -3219,6 +3219,10 @@ int btrfs_writepage_cow_fixup(struct page *page);
 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
 					  struct page *page, u64 start,
 					  u64 end, bool uptodate);
+struct btrfs_ioctl_encoded_io_args;
+ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
+			   struct btrfs_ioctl_encoded_io_args *encoded);
+
 extern const struct dentry_operations btrfs_dentry_operations;
 extern const struct iomap_ops btrfs_dio_iomap_ops;
 extern const struct iomap_dio_ops btrfs_dio_ops;

fs/btrfs/inode.c +363 -0

index 0b5ff14aa7fd..6d7fae859fb5 100644
@@ -10496,6 +10496,495 @@ void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
 	}
 }
 
+static int btrfs_encoded_io_compression_from_extent(int compress_type)
+{
+	switch (compress_type) {
+	case BTRFS_COMPRESS_NONE:
+		return BTRFS_ENCODED_IO_COMPRESSION_NONE;
+	case BTRFS_COMPRESS_ZLIB:
+		return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
+	case BTRFS_COMPRESS_LZO:
+		/*
+		 * The LZO format depends on the page size. 64k is the maximum
+		 * sectorsize (and thus page size) that we support.
+		 */
+		if (PAGE_SIZE < SZ_4K || PAGE_SIZE > SZ_64K)
+			return -EINVAL;
+		return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + (PAGE_SHIFT - 12);
+	case BTRFS_COMPRESS_ZSTD:
+		return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
+	default:
+		return -EUCLEAN;
+	}
+}
+
+static ssize_t btrfs_encoded_read_inline(
+				struct kiocb *iocb,
+				struct iov_iter *iter, u64 start,
+				u64 lockend,
+				struct extent_state **cached_state,
+				u64 extent_start, size_t count,
+				struct btrfs_ioctl_encoded_io_args *encoded,
+				bool *unlocked)
+{
+	struct inode *inode = file_inode(iocb->ki_filp);
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	struct btrfs_path *path;
+	struct extent_buffer *leaf;
+	struct btrfs_file_extent_item *item;
+	u64 ram_bytes;
+	unsigned long ptr;
+	void *tmp;
+	ssize_t ret;
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		ret = -ENOMEM;
+		goto out;
+	}
+	ret = btrfs_lookup_file_extent(NULL, BTRFS_I(inode)->root, path,
+				       btrfs_ino(BTRFS_I(inode)), extent_start,
+				       0);
+	if (ret) {
+		if (ret > 0) {
+			/* The extent item disappeared? */
+			ret = -EIO;
+		}
+		goto out;
+	}
+	leaf = path->nodes[0];
+	item = btrfs_item_ptr(leaf, path->slots[0],
+			      struct btrfs_file_extent_item);
+
+	ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
+	ptr = btrfs_file_extent_inline_start(item);
+
+	encoded->len = (min_t(u64, extent_start + ram_bytes, inode->i_size) -
+			iocb->ki_pos);
+	ret = btrfs_encoded_io_compression_from_extent(
+				 btrfs_file_extent_compression(leaf, item));
+	if (ret < 0)
+		goto out;
+	encoded->compression = ret;
+	if (encoded->compression) {
+		size_t inline_size;
+
+		inline_size = btrfs_file_extent_inline_item_len(leaf,
+						btrfs_item_nr(path->slots[0]));
+		if (inline_size > count) {
+			ret = -ENOBUFS;
+			goto out;
+		}
+		count = inline_size;
+		encoded->unencoded_len = ram_bytes;
+		encoded->unencoded_offset = iocb->ki_pos - extent_start;
+	} else {
+		encoded->len = encoded->unencoded_len = count =
+			min_t(u64, count, encoded->len);
+		ptr += iocb->ki_pos - extent_start;
+	}
+
+	tmp = kmalloc(count, GFP_NOFS);
+	if (!tmp) {
+		ret = -ENOMEM;
+		goto out;
+	}
+	read_extent_buffer(leaf, tmp, ptr, count);
+	btrfs_release_path(path);
+	unlock_extent_cached(io_tree, start, lockend, cached_state);
+	inode_unlock_shared(inode);
+	*unlocked = true;
+
+	ret = copy_to_iter(tmp, count, iter);
+	if (ret != count)
+		ret = -EFAULT;
+	kfree(tmp);
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+struct btrfs_encoded_read_private {
+	struct inode *inode;
+	wait_queue_head_t wait;
+	atomic_t pending;
+	blk_status_t status;
+	bool skip_csum;
+};
+
+static blk_status_t submit_encoded_read_bio(struct inode *inode,
+					    struct bio *bio, int mirror_num,
+					    unsigned long bio_flags)
+{
+	struct btrfs_encoded_read_private *priv = bio->bi_private;
+	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
+	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
+	blk_status_t ret;
+
+	if (!priv->skip_csum) {
+		ret = btrfs_lookup_bio_sums(inode, bio, NULL);
+		if (ret)
+			return ret;
+	}
+
+	ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
+	if (ret) {
+		btrfs_io_bio_free_csum(io_bio);
+		return ret;
+	}
+
+	atomic_inc(&priv->pending);
+	ret = btrfs_map_bio(fs_info, bio, mirror_num);
+	if (ret) {
+		atomic_dec(&priv->pending);
+		btrfs_io_bio_free_csum(io_bio);
+	}
+	return ret;
+}
+
+static blk_status_t btrfs_encoded_read_check_bio(struct btrfs_io_bio *io_bio)
+{
+	const bool uptodate = io_bio->bio.bi_status == BLK_STS_OK;
+	struct btrfs_encoded_read_private *priv = io_bio->bio.bi_private;
+	struct inode *inode = priv->inode;
+	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
+	u32 sectorsize = fs_info->sectorsize;
+	struct bio_vec *bvec;
+	struct bvec_iter_all iter_all;
+	u64 start = io_bio->logical;
+	u32 bio_offset = 0;
+
+	if (priv->skip_csum || !uptodate)
+		return io_bio->bio.bi_status;
+
+	bio_for_each_segment_all(bvec, &io_bio->bio, iter_all) {
+		unsigned int i, nr_sectors, pgoff;
+
+		nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
+		pgoff = bvec->bv_offset;
+		for (i = 0; i < nr_sectors; i++) {
+			ASSERT(pgoff < PAGE_SIZE);
+			if (check_data_csum(inode, io_bio, bio_offset,
+					    bvec->bv_page, pgoff, start))
+				return BLK_STS_IOERR;
+			start += sectorsize;
+			bio_offset += sectorsize;
+			pgoff += sectorsize;
+		}
+	}
+	return BLK_STS_OK;
+}
+
+static void btrfs_encoded_read_endio(struct bio *bio)
+{
+	struct btrfs_encoded_read_private *priv = bio->bi_private;
+	struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
+	blk_status_t status;
+
+	status = btrfs_encoded_read_check_bio(io_bio);
+	if (status) {
+		/*
+		 * The memory barrier implied by the atomic_dec_return() here
+		 * pairs with the memory barrier implied by the
+		 * atomic_dec_return() or io_wait_event() in
+		 * btrfs_encoded_read_regular_fill_pages() to ensure that this
+		 * write is observed before the load of status in
+		 * btrfs_encoded_read_regular_fill_pages().
+		 */
+		WRITE_ONCE(priv->status, status);
+	}
+	if (!atomic_dec_return(&priv->pending))
+		wake_up(&priv->wait);
+	btrfs_io_bio_free_csum(io_bio);
+	bio_put(bio);
+}
+
+static int btrfs_encoded_read_regular_fill_pages(struct inode *inode, u64 offset,
+						 u64 disk_io_size, struct page **pages)
+{
+	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
+	struct btrfs_encoded_read_private priv = {
+		.inode = inode,
+		.pending = ATOMIC_INIT(1),
+		.skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM,
+	};
+	unsigned long i = 0;
+	u64 cur = 0;
+	int ret;
+
+	init_waitqueue_head(&priv.wait);
+	/*
+	 * Submit bios for the extent, splitting due to bio or stripe limits as
+	 * necessary.
+	 */
+	while (cur < disk_io_size) {
+		struct extent_map *em;
+		struct btrfs_io_geometry geom;
+		struct bio *bio = NULL;
+		u64 remaining;
+
+		em = btrfs_get_chunk_map(fs_info, offset + cur,
+					 disk_io_size - cur);
+		if (IS_ERR(em)) {
+			ret = PTR_ERR(em);
+		} else {
+			ret = btrfs_get_io_geometry(fs_info, em, BTRFS_MAP_READ,
+						    offset + cur, &geom);
+			free_extent_map(em);
+		}
+		if (ret) {
+			WRITE_ONCE(priv.status, errno_to_blk_status(ret));
+			break;
+		}
+		remaining = min(geom.len, disk_io_size - cur);
+		while (bio || remaining) {
+			size_t bytes = min_t(u64, remaining, PAGE_SIZE);
+
+			if (!bio) {
+				bio = btrfs_bio_alloc(offset + cur);
+				bio->bi_end_io = btrfs_encoded_read_endio;
+				bio->bi_private = &priv;
+				bio->bi_opf = REQ_OP_READ;
+			}
+
+			if (!bytes ||
+			    bio_add_page(bio, pages[i], bytes, 0) < bytes) {
+				blk_status_t status;
+
+				status = submit_encoded_read_bio(inode, bio, 0,
+								 0);
+				if (status) {
+					WRITE_ONCE(priv.status, status);
+					bio_put(bio);
+					goto out;
+				}
+				bio = NULL;
+				continue;
+			}
+
+			i++;
+			cur += bytes;
+			remaining -= bytes;
+		}
+	}
+
+out:
+	if (atomic_dec_return(&priv.pending))
+		io_wait_event(priv.wait, !atomic_read(&priv.pending));
+	/* See btrfs_encoded_read_endio() for ordering. */
+	return blk_status_to_errno(READ_ONCE(priv.status));
+}
+
+static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
+					  struct iov_iter *iter,
+					  u64 start, u64 lockend,
+					  struct extent_state **cached_state,
+					  u64 offset, u64 disk_io_size,
+					  size_t count, bool compressed,
+					  bool *unlocked)
+{
+	struct inode *inode = file_inode(iocb->ki_filp);
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	struct page **pages;
+	unsigned long nr_pages, i;
+	u64 cur;
+	size_t page_offset;
+	ssize_t ret;
+
+	nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
+	pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
+	if (!pages)
+		return -ENOMEM;
+	for (i = 0; i < nr_pages; i++) {
+		pages[i] = alloc_page(GFP_NOFS | __GFP_HIGHMEM);
+		if (!pages[i]) {
+			ret = -ENOMEM;
+			goto out;
+		}
+	}
+
+	ret = btrfs_encoded_read_regular_fill_pages(inode, offset, disk_io_size,
+						    pages);
+	if (ret)
+		goto out;
+
+	unlock_extent_cached(io_tree, start, lockend, cached_state);
+	inode_unlock_shared(inode);
+	*unlocked = true;
+
+	if (compressed) {
+		i = 0;
+		page_offset = 0;
+	} else {
+		i = (iocb->ki_pos - start) >> PAGE_SHIFT;
+		page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
+	}
+	cur = 0;
+	while (cur < count) {
+		size_t bytes = min_t(size_t, count - cur,
+				     PAGE_SIZE - page_offset);
+
+		if (copy_page_to_iter(pages[i], page_offset, bytes,
+				      iter) != bytes) {
+			ret = -EFAULT;
+			goto out;
+		}
+		i++;
+		cur += bytes;
+		page_offset = 0;
+	}
+	ret = count;
+out:
+	for (i = 0; i < nr_pages; i++) {
+		if (pages[i])
+			__free_page(pages[i]);
+	}
+	kfree(pages);
+	return ret;
+}
+
+ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
+			   struct btrfs_ioctl_encoded_io_args *encoded)
+{
+	struct inode *inode = file_inode(iocb->ki_filp);
+	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
+	struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
+	ssize_t ret;
+	size_t count = iov_iter_count(iter);
+	u64 start, lockend, offset, disk_io_size;
+	struct extent_state *cached_state = NULL;
+	struct extent_map *em;
+	bool unlocked = false;
+
+	file_accessed(iocb->ki_filp);
+
+	inode_lock_shared(inode);

fs/btrfs/inode.c void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end) · line 10865

Quoted 4 lines

fs/btrfs/inode.c +4 -0

+
+	if (iocb->ki_pos >= inode->i_size) {
+		inode_unlock_shared(inode);
+		return 0;

Don't we need to signal beyond EOF reads somehow. As it stands returning 0 means returning zeroed portion of btrfs_ioctl_encoded_io_args to user space?

Quoted 20 lines

fs/btrfs/inode.c +20 -0

+	}
+	start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
+	/*
+	 * We don't know how long the extent containing iocb->ki_pos is, but if
+	 * it's compressed we know that it won't be longer than this.
+	 */
+	lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
+
+	for (;;) {
+		struct btrfs_ordered_extent *ordered;
+
+		ret = btrfs_wait_ordered_range(inode, start,
+					       lockend - start + 1);
+		if (ret)
+			goto out_unlock_inode;
+		lock_extent_bits(io_tree, start, lockend, &cached_state);
+		ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
+						     lockend - start + 1);
+		if (!ordered)
+			break;