[Devel] [PATCH RHEL7 COMMIT] ms/ext4: more efficient SEEK_DATA implementation

Konstantin Khorenko khorenko at virtuozzo.com
Fri Apr 28 05:16:16 PDT 2017


The commit is pushed to "branch-rh7-3.10.0-514.16.1.vz7.32.x-ovz" and will appear at https://src.openvz.org/scm/ovz/vzkernel.git
after rh7-3.10.0-514.16.1.vz7.30.8
------>
commit d67d98bac6fd3f34052ba38dcd8fb79dfcd99d5c
Author: Jan Kara <jack at suse.cz>
Date:   Fri Apr 28 16:16:16 2017 +0400

    ms/ext4: more efficient SEEK_DATA implementation
    
    commit 2d90c160e5f1d784e180f1e1458d56eee4d7f4f4 upstream.
    
    Using SEEK_DATA in a huge sparse file can easily lead to sotflockups as
    ext4_seek_data() iterates hole block-by-block. Fix the problem by using
    returned hole size from ext4_map_blocks() and thus skip the hole in one
    go.
    
    Update also SEEK_HOLE implementation to follow the same pattern as
    SEEK_DATA to make future maintenance easier.
    
    Furthermore we add cond_resched() to both ext4_seek_data() and
    ext4_seek_hole() to avoid softlockups in case evil user creates huge
    fragmented file and we have to go through lots of extents.
    
    Signed-off-by: Jan Kara <jack at suse.cz>
    Signed-off-by: Theodore Ts'o <tytso at mit.edu>
    
    https://jira.sw.ru/browse/PSBM-65551
    Signed-off-by: Andrey Ryabinin <aryabinin at virtuozzo.com>
---
 fs/ext4/ext4.h  |  3 ++
 fs/ext4/file.c  | 97 +++++++++++++++++++++------------------------------------
 fs/ext4/inode.c | 67 +++++++++++++++++++++++++++++++++++++++
 3 files changed, 106 insertions(+), 61 deletions(-)

diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 8fcb65e..1cd9648 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -2180,6 +2180,9 @@ extern void ext4_da_update_reserve_space(struct inode *inode,
 					int used, int quota_claim);
 extern int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk,
 			      ext4_fsblk_t pblk, ext4_lblk_t len);
+extern int ext4_get_next_extent(struct inode *inode, ext4_lblk_t lblk,
+				unsigned int map_len,
+				struct extent_status *result);
 
 /* indirect.c */
 extern int ext4_ind_map_blocks(handle_t *handle, struct inode *inode,
diff --git a/fs/ext4/file.c b/fs/ext4/file.c
index c63d937..cea271f 100644
--- a/fs/ext4/file.c
+++ b/fs/ext4/file.c
@@ -425,7 +425,7 @@ static int ext4_file_open(struct inode * inode, struct file * filp)
  */
 static int ext4_find_unwritten_pgoff(struct inode *inode,
 				     int whence,
-				     struct ext4_map_blocks *map,
+				     ext4_lblk_t end_blk,
 				     loff_t *offset)
 {
 	struct pagevec pvec;
@@ -440,7 +440,7 @@ static int ext4_find_unwritten_pgoff(struct inode *inode,
 	blkbits = inode->i_sb->s_blocksize_bits;
 	startoff = *offset;
 	lastoff = startoff;
-	endoff = (loff_t)(map->m_lblk + map->m_len) << blkbits;
+	endoff = (loff_t)end_blk << blkbits;
 
 	index = startoff >> PAGE_CACHE_SHIFT;
 	end = endoff >> PAGE_CACHE_SHIFT;
@@ -558,12 +558,11 @@ out:
 static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 {
 	struct inode *inode = file->f_mapping->host;
-	struct ext4_map_blocks map;
 	struct extent_status es;
 	ext4_lblk_t start, last, end;
 	loff_t dataoff, isize;
 	int blkbits;
-	int ret = 0;
+	int ret;
 
 	mutex_lock(&inode->i_mutex);
 
@@ -580,41 +579,32 @@ static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 	dataoff = offset;
 
 	do {
-		map.m_lblk = last;
-		map.m_len = end - last + 1;
-		ret = ext4_map_blocks(NULL, inode, &map, 0);
-		if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
-			if (last != start)
-				dataoff = (loff_t)last << blkbits;
-			break;
+		ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
+		if (ret <= 0) {
+			/* No extent found -> no data */
+			if (ret == 0)
+				ret = -ENXIO;
+			mutex_unlock(&inode->i_mutex);
+			return ret;
 		}
 
-		/*
-		 * If there is a delay extent at this offset,
-		 * it will be as a data.
-		 */
-		ext4_es_find_delayed_extent_range(inode, last, last, &es);
-		if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
-			if (last != start)
-				dataoff = (loff_t)last << blkbits;
+		last = es.es_lblk;
+		if (last != start)
+			dataoff = (loff_t)last << blkbits;
+		if (!ext4_es_is_unwritten(&es))
 			break;
-		}
 
 		/*
 		 * If there is a unwritten extent at this offset,
 		 * it will be as a data or a hole according to page
 		 * cache that has data or not.
 		 */
-		if (map.m_flags & EXT4_MAP_UNWRITTEN) {
-			int unwritten;
-			unwritten = ext4_find_unwritten_pgoff(inode, SEEK_DATA,
-							      &map, &dataoff);
-			if (unwritten)
-				break;
-		}
-
-		last++;
+		if (ext4_find_unwritten_pgoff(inode, SEEK_DATA,
+					      es.es_lblk + es.es_len, &dataoff))
+			break;
+		last += es.es_len;
 		dataoff = (loff_t)last << blkbits;
+		cond_resched();
 	} while (last <= end);
 
 	mutex_unlock(&inode->i_mutex);
@@ -631,12 +621,11 @@ static loff_t ext4_seek_data(struct file *file, loff_t offset, loff_t maxsize)
 static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 {
 	struct inode *inode = file->f_mapping->host;
-	struct ext4_map_blocks map;
 	struct extent_status es;
 	ext4_lblk_t start, last, end;
 	loff_t holeoff, isize;
 	int blkbits;
-	int ret = 0;
+	int ret;
 
 	mutex_lock(&inode->i_mutex);
 
@@ -653,44 +642,30 @@ static loff_t ext4_seek_hole(struct file *file, loff_t offset, loff_t maxsize)
 	holeoff = offset;
 
 	do {
-		map.m_lblk = last;
-		map.m_len = end - last + 1;
-		ret = ext4_map_blocks(NULL, inode, &map, 0);
-		if (ret > 0 && !(map.m_flags & EXT4_MAP_UNWRITTEN)) {
-			last += ret;
-			holeoff = (loff_t)last << blkbits;
-			continue;
+		ret = ext4_get_next_extent(inode, last, end - last + 1, &es);
+		if (ret < 0) {
+			mutex_unlock(&inode->i_mutex);
+			return ret;
 		}
-
-		/*
-		 * If there is a delay extent at this offset,
-		 * we will skip this extent.
-		 */
-		ext4_es_find_delayed_extent_range(inode, last, last, &es);
-		if (es.es_len != 0 && in_range(last, es.es_lblk, es.es_len)) {
-			last = es.es_lblk + es.es_len;
-			holeoff = (loff_t)last << blkbits;
-			continue;
+		/* Found a hole? */
+		if (ret == 0 || es.es_lblk > last) {
+			if (last != start)
+				holeoff = (loff_t)last << blkbits;
+			break;
 		}
-
 		/*
 		 * If there is a unwritten extent at this offset,
 		 * it will be as a data or a hole according to page
 		 * cache that has data or not.
 		 */
-		if (map.m_flags & EXT4_MAP_UNWRITTEN) {
-			int unwritten;
-			unwritten = ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
-							      &map, &holeoff);
-			if (!unwritten) {
-				last += ret;
-				holeoff = (loff_t)last << blkbits;
-				continue;
-			}
-		}
+		if (ext4_es_is_unwritten(&es) &&
+		    ext4_find_unwritten_pgoff(inode, SEEK_HOLE,
+					      last + es.es_len, &holeoff))
+			break;
 
-		/* find a hole */
-		break;
+		last += es.es_len;
+		holeoff = (loff_t)last << blkbits;
+		cond_resched();
 	} while (last <= end);
 
 	mutex_unlock(&inode->i_mutex);
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 83b58ae..b51b85c 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -5433,3 +5433,70 @@ int ext4_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 
 	return err;
 }
+
+/*
+ * Find the first extent at or after @lblk in an inode that is not a hole.
+ * Search for @map_len blocks at most. The extent is returned in @result.
+ *
+ * The function returns 1 if we found an extent. The function returns 0 in
+ * case there is no extent at or after @lblk and in that case also sets
+ * @result->es_len to 0. In case of error, the error code is returned.
+ */
+int ext4_get_next_extent(struct inode *inode, ext4_lblk_t lblk,
+			 unsigned int map_len, struct extent_status *result)
+{
+	struct ext4_map_blocks map;
+	struct extent_status es = {};
+	int ret;
+
+	map.m_lblk = lblk;
+	map.m_len = map_len;
+
+	/*
+	 * For non-extent based files this loop may iterate several times since
+	 * we do not determine full hole size.
+	 */
+	while (map.m_len > 0) {
+		ret = ext4_map_blocks(NULL, inode, &map, 0);
+		if (ret < 0)
+			return ret;
+		/* There's extent covering m_lblk? Just return it. */
+		if (ret > 0) {
+			int status;
+
+			ext4_es_store_pblock(result, map.m_pblk);
+			result->es_lblk = map.m_lblk;
+			result->es_len = map.m_len;
+			if (map.m_flags & EXT4_MAP_UNWRITTEN)
+				status = EXTENT_STATUS_UNWRITTEN;
+			else
+				status = EXTENT_STATUS_WRITTEN;
+			ext4_es_store_status(result, status);
+			return 1;
+		}
+		ext4_es_find_delayed_extent_range(inode, map.m_lblk,
+						  map.m_lblk + map.m_len - 1,
+						  &es);
+		/* Is delalloc data before next block in extent tree? */
+		if (es.es_len && es.es_lblk < map.m_lblk + map.m_len) {
+			ext4_lblk_t offset = 0;
+
+			if (es.es_lblk < lblk)
+				offset = lblk - es.es_lblk;
+			result->es_lblk = es.es_lblk + offset;
+			ext4_es_store_pblock(result,
+					     ext4_es_pblock(&es) + offset);
+			result->es_len = es.es_len - offset;
+			ext4_es_store_status(result, ext4_es_status(&es));
+
+			return 1;
+		}
+		/* There's a hole at m_lblk, advance us after it */
+		map.m_lblk += map.m_len;
+		map_len -= map.m_len;
+		map.m_len = map_len;
+		cond_resched();
+	}
+	result->es_len = 0;
+	return 0;
+}


More information about the Devel mailing list