[patch 23/39] ext4: Use EXT4_GROUP_INFO_NEED_INIT_BIT during resize

[Greg KH <gregkh@suse.de>]

[-- Attachment #1: ext4-use-ext4_group_info_need_init_bit-during-resize.patch --]
[-- Type: text/plain, Size: 16726 bytes --]


2.6.28-stable review patch.  If anyone has any objections, please let us know.

------------------

From: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>

(cherry picked from commit 920313a726e04fef0f2c0bcb04ad8229c0e700d8)

The new groups added during resize are flagged as
need_init group. Make sure we properly initialize these
groups. When we have block size < page size and we are adding
new groups the page may still be marked uptodate even though
we haven't initialized the group. While forcing the init
of buddy cache we need to make sure other groups part of the
same page of buddy cache is not using the cache.
group_info->alloc_sem is added to ensure the same.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

---
 fs/ext4/balloc.c  |   21 ++--
 fs/ext4/ext4.h    |    7 -
 fs/ext4/mballoc.c |  259 +++++++++++++++++++++++++++++++++++++++++-------------
 fs/ext4/mballoc.h |    3 
 fs/ext4/resize.c  |   49 +---------
 5 files changed, 229 insertions(+), 110 deletions(-)

--- a/fs/ext4/balloc.c
+++ b/fs/ext4/balloc.c
@@ -381,6 +381,7 @@ void ext4_add_groupblocks(handle_t *hand
 	ext4_debug("Adding block(s) %llu-%llu\n", block, block + count - 1);
 
 	ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
+	grp = ext4_get_group_info(sb, block_group);
 	/*
 	 * Check to see if we are freeing blocks across a group
 	 * boundary.
@@ -425,7 +426,11 @@ void ext4_add_groupblocks(handle_t *hand
 	err = ext4_journal_get_write_access(handle, gd_bh);
 	if (err)
 		goto error_return;
-
+	/*
+	 * make sure we don't allow a parallel init on other groups in the
+	 * same buddy cache
+	 */
+	down_write(&grp->alloc_sem);
 	for (i = 0, blocks_freed = 0; i < count; i++) {
 		BUFFER_TRACE(bitmap_bh, "clear bit");
 		if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
@@ -450,6 +455,13 @@ void ext4_add_groupblocks(handle_t *hand
 		sbi->s_flex_groups[flex_group].free_blocks += blocks_freed;
 		spin_unlock(sb_bgl_lock(sbi, flex_group));
 	}
+	/*
+	 * request to reload the buddy with the
+	 * new bitmap information
+	 */
+	set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));
+	ext4_mb_update_group_info(grp, blocks_freed);
+	up_write(&grp->alloc_sem);
 
 	/* We dirtied the bitmap block */
 	BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
@@ -461,13 +473,6 @@ void ext4_add_groupblocks(handle_t *hand
 	if (!err)
 		err = ret;
 	sb->s_dirt = 1;
-	/*
-	 * request to reload the buddy with the
-	 * new bitmap information
-	 */
-	grp = ext4_get_group_info(sb, block_group);
-	set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));
-	ext4_mb_update_group_info(grp, blocks_freed);
 
 error_return:
 	brelse(bitmap_bh);
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -1056,12 +1056,13 @@ extern int __init init_ext4_mballoc(void
 extern void exit_ext4_mballoc(void);
 extern void ext4_mb_free_blocks(handle_t *, struct inode *,
 		unsigned long, unsigned long, int, unsigned long *);
-extern int ext4_mb_add_more_groupinfo(struct super_block *sb,
+extern int ext4_mb_add_groupinfo(struct super_block *sb,
 		ext4_group_t i, struct ext4_group_desc *desc);
 extern void ext4_mb_update_group_info(struct ext4_group_info *grp,
 		ext4_grpblk_t add);
-
-
+extern int ext4_mb_get_buddy_cache_lock(struct super_block *, ext4_group_t);
+extern void ext4_mb_put_buddy_cache_lock(struct super_block *,
+						ext4_group_t, int);
 /* inode.c */
 int ext4_forget(handle_t *handle, int is_metadata, struct inode *inode,
 		struct buffer_head *bh, ext4_fsblk_t blocknr);
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -886,18 +886,20 @@ static noinline_for_stack int
 ext4_mb_load_buddy(struct super_block *sb, ext4_group_t group,
 					struct ext4_buddy *e4b)
 {
-	struct ext4_sb_info *sbi = EXT4_SB(sb);
-	struct inode *inode = sbi->s_buddy_cache;
 	int blocks_per_page;
 	int block;
 	int pnum;
 	int poff;
 	struct page *page;
 	int ret;
+	struct ext4_group_info *grp;
+	struct ext4_sb_info *sbi = EXT4_SB(sb);
+	struct inode *inode = sbi->s_buddy_cache;
 
 	mb_debug("load group %lu\n", group);
 
 	blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
+	grp = ext4_get_group_info(sb, group);
 
 	e4b->bd_blkbits = sb->s_blocksize_bits;
 	e4b->bd_info = ext4_get_group_info(sb, group);
@@ -905,6 +907,15 @@ ext4_mb_load_buddy(struct super_block *s
 	e4b->bd_group = group;
 	e4b->bd_buddy_page = NULL;
 	e4b->bd_bitmap_page = NULL;
+	e4b->alloc_semp = &grp->alloc_sem;
+
+	/* Take the read lock on the group alloc
+	 * sem. This would make sure a parallel
+	 * ext4_mb_init_group happening on other
+	 * groups mapped by the page is blocked
+	 * till we are done with allocation
+	 */
+	down_read(e4b->alloc_semp);
 
 	/*
 	 * the buddy cache inode stores the block bitmap
@@ -920,6 +931,14 @@ ext4_mb_load_buddy(struct super_block *s
 	page = find_get_page(inode->i_mapping, pnum);
 	if (page == NULL || !PageUptodate(page)) {
 		if (page)
+			/*
+			 * drop the page reference and try
+			 * to get the page with lock. If we
+			 * are not uptodate that implies
+			 * somebody just created the page but
+			 * is yet to initialize the same. So
+			 * wait for it to initialize.
+			 */
 			page_cache_release(page);
 		page = find_or_create_page(inode->i_mapping, pnum, GFP_NOFS);
 		if (page) {
@@ -985,6 +1004,9 @@ err:
 		page_cache_release(e4b->bd_buddy_page);
 	e4b->bd_buddy = NULL;
 	e4b->bd_bitmap = NULL;
+
+	/* Done with the buddy cache */
+	up_read(e4b->alloc_semp);
 	return ret;
 }
 
@@ -994,6 +1016,8 @@ static void ext4_mb_release_desc(struct 
 		page_cache_release(e4b->bd_bitmap_page);
 	if (e4b->bd_buddy_page)
 		page_cache_release(e4b->bd_buddy_page);
+	/* Done with the buddy cache */
+	up_read(e4b->alloc_semp);
 }
 
 
@@ -1694,6 +1718,173 @@ static int ext4_mb_good_group(struct ext
 	return 0;
 }
 
+/*
+ * lock the group_info alloc_sem of all the groups
+ * belonging to the same buddy cache page. This
+ * make sure other parallel operation on the buddy
+ * cache doesn't happen  whild holding the buddy cache
+ * lock
+ */
+int ext4_mb_get_buddy_cache_lock(struct super_block *sb, ext4_group_t group)
+{
+	int i;
+	int block, pnum;
+	int blocks_per_page;
+	int groups_per_page;
+	ext4_group_t first_group;
+	struct ext4_group_info *grp;
+
+	blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
+	/*
+	 * the buddy cache inode stores the block bitmap
+	 * and buddy information in consecutive blocks.
+	 * So for each group we need two blocks.
+	 */
+	block = group * 2;
+	pnum = block / blocks_per_page;
+	first_group = pnum * blocks_per_page / 2;
+
+	groups_per_page = blocks_per_page >> 1;
+	if (groups_per_page == 0)
+		groups_per_page = 1;
+	/* read all groups the page covers into the cache */
+	for (i = 0; i < groups_per_page; i++) {
+
+		if ((first_group + i) >= EXT4_SB(sb)->s_groups_count)
+			break;
+		grp = ext4_get_group_info(sb, first_group + i);
+		/* take all groups write allocation
+		 * semaphore. This make sure there is
+		 * no block allocation going on in any
+		 * of that groups
+		 */
+		down_write(&grp->alloc_sem);
+	}
+	return i;
+}
+
+void ext4_mb_put_buddy_cache_lock(struct super_block *sb,
+					ext4_group_t group, int locked_group)
+{
+	int i;
+	int block, pnum;
+	int blocks_per_page;
+	ext4_group_t first_group;
+	struct ext4_group_info *grp;
+
+	blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
+	/*
+	 * the buddy cache inode stores the block bitmap
+	 * and buddy information in consecutive blocks.
+	 * So for each group we need two blocks.
+	 */
+	block = group * 2;
+	pnum = block / blocks_per_page;
+	first_group = pnum * blocks_per_page / 2;
+	/* release locks on all the groups */
+	for (i = 0; i < locked_group; i++) {
+
+		grp = ext4_get_group_info(sb, first_group + i);
+		/* take all groups write allocation
+		 * semaphore. This make sure there is
+		 * no block allocation going on in any
+		 * of that groups
+		 */
+		up_write(&grp->alloc_sem);
+	}
+
+}
+
+static int ext4_mb_init_group(struct super_block *sb, ext4_group_t group)
+{
+
+	int ret;
+	void *bitmap;
+	int blocks_per_page;
+	int block, pnum, poff;
+	int num_grp_locked = 0;
+	struct ext4_group_info *this_grp;
+	struct ext4_sb_info *sbi = EXT4_SB(sb);
+	struct inode *inode = sbi->s_buddy_cache;
+	struct page *page = NULL, *bitmap_page = NULL;
+
+	mb_debug("init group %lu\n", group);
+	blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
+	this_grp = ext4_get_group_info(sb, group);
+	/*
+	 * This ensures we don't add group
+	 * to this buddy cache via resize
+	 */
+	num_grp_locked =  ext4_mb_get_buddy_cache_lock(sb, group);
+	if (!EXT4_MB_GRP_NEED_INIT(this_grp)) {
+		/*
+		 * somebody initialized the group
+		 * return without doing anything
+		 */
+		ret = 0;
+		goto err;
+	}
+	/*
+	 * the buddy cache inode stores the block bitmap
+	 * and buddy information in consecutive blocks.
+	 * So for each group we need two blocks.
+	 */
+	block = group * 2;
+	pnum = block / blocks_per_page;
+	poff = block % blocks_per_page;
+	page = find_or_create_page(inode->i_mapping, pnum, GFP_NOFS);
+	if (page) {
+		BUG_ON(page->mapping != inode->i_mapping);
+		ret = ext4_mb_init_cache(page, NULL);
+		if (ret) {
+			unlock_page(page);
+			goto err;
+		}
+		unlock_page(page);
+	}
+	if (page == NULL || !PageUptodate(page)) {
+		ret = -EIO;
+		goto err;
+	}
+	mark_page_accessed(page);
+	bitmap_page = page;
+	bitmap = page_address(page) + (poff * sb->s_blocksize);
+
+	/* init buddy cache */
+	block++;
+	pnum = block / blocks_per_page;
+	poff = block % blocks_per_page;
+	page = find_or_create_page(inode->i_mapping, pnum, GFP_NOFS);
+	if (page == bitmap_page) {
+		/*
+		 * If both the bitmap and buddy are in
+		 * the same page we don't need to force
+		 * init the buddy
+		 */
+		unlock_page(page);
+	} else if (page) {
+		BUG_ON(page->mapping != inode->i_mapping);
+		ret = ext4_mb_init_cache(page, bitmap);
+		if (ret) {
+			unlock_page(page);
+			goto err;
+		}
+		unlock_page(page);
+	}
+	if (page == NULL || !PageUptodate(page)) {
+		ret = -EIO;
+		goto err;
+	}
+	mark_page_accessed(page);
+err:
+	ext4_mb_put_buddy_cache_lock(sb, group, num_grp_locked);
+	if (bitmap_page)
+		page_cache_release(bitmap_page);
+	if (page)
+		page_cache_release(page);
+	return ret;
+}
+
 static noinline_for_stack int
 ext4_mb_regular_allocator(struct ext4_allocation_context *ac)
 {
@@ -1777,7 +1968,7 @@ repeat:
 				group = 0;
 
 			/* quick check to skip empty groups */
-			grp = ext4_get_group_info(ac->ac_sb, group);
+			grp = ext4_get_group_info(sb, group);
 			if (grp->bb_free == 0)
 				continue;
 
@@ -1790,10 +1981,9 @@ repeat:
 				 * we need full data about the group
 				 * to make a good selection
 				 */
-				err = ext4_mb_load_buddy(sb, group, &e4b);
+				err = ext4_mb_init_group(sb, group);
 				if (err)
 					goto out;
-				ext4_mb_release_desc(&e4b);
 			}
 
 			/*
@@ -2302,6 +2492,7 @@ int ext4_mb_add_groupinfo(struct super_b
 	}
 
 	INIT_LIST_HEAD(&meta_group_info[i]->bb_prealloc_list);
+	init_rwsem(&meta_group_info[i]->alloc_sem);
 	meta_group_info[i]->bb_free_root.rb_node = NULL;;
 
 #ifdef DOUBLE_CHECK
@@ -2329,54 +2520,6 @@ exit_meta_group_info:
 } /* ext4_mb_add_groupinfo */
 
 /*
- * Add a group to the existing groups.
- * This function is used for online resize
- */
-int ext4_mb_add_more_groupinfo(struct super_block *sb, ext4_group_t group,
-			       struct ext4_group_desc *desc)
-{
-	struct ext4_sb_info *sbi = EXT4_SB(sb);
-	struct inode *inode = sbi->s_buddy_cache;
-	int blocks_per_page;
-	int block;
-	int pnum;
-	struct page *page;
-	int err;
-
-	/* Add group based on group descriptor*/
-	err = ext4_mb_add_groupinfo(sb, group, desc);
-	if (err)
-		return err;
-
-	/*
-	 * Cache pages containing dynamic mb_alloc datas (buddy and bitmap
-	 * datas) are set not up to date so that they will be re-initilaized
-	 * during the next call to ext4_mb_load_buddy
-	 */
-
-	/* Set buddy page as not up to date */
-	blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
-	block = group * 2;
-	pnum = block / blocks_per_page;
-	page = find_get_page(inode->i_mapping, pnum);
-	if (page != NULL) {
-		ClearPageUptodate(page);
-		page_cache_release(page);
-	}
-
-	/* Set bitmap page as not up to date */
-	block++;
-	pnum = block / blocks_per_page;
-	page = find_get_page(inode->i_mapping, pnum);
-	if (page != NULL) {
-		ClearPageUptodate(page);
-		page_cache_release(page);
-	}
-
-	return 0;
-}
-
-/*
  * Update an existing group.
  * This function is used for online resize
  */
@@ -4585,11 +4728,6 @@ do_more:
 	err = ext4_journal_get_write_access(handle, gd_bh);
 	if (err)
 		goto error_return;
-
-	err = ext4_mb_load_buddy(sb, block_group, &e4b);
-	if (err)
-		goto error_return;
-
 #ifdef AGGRESSIVE_CHECK
 	{
 		int i;
@@ -4603,6 +4741,8 @@ do_more:
 	/* We dirtied the bitmap block */
 	BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
 	err = ext4_journal_dirty_metadata(handle, bitmap_bh);
+	if (err)
+		goto error_return;
 
 	if (ac) {
 		ac->ac_b_ex.fe_group = block_group;
@@ -4611,6 +4751,9 @@ do_more:
 		ext4_mb_store_history(ac);
 	}
 
+	err = ext4_mb_load_buddy(sb, block_group, &e4b);
+	if (err)
+		goto error_return;
 	if (metadata) {
 		/* blocks being freed are metadata. these blocks shouldn't
 		 * be used until this transaction is committed */
--- a/fs/ext4/mballoc.h
+++ b/fs/ext4/mballoc.h
@@ -20,6 +20,7 @@
 #include <linux/version.h>
 #include <linux/blkdev.h>
 #include <linux/marker.h>
+#include <linux/mutex.h>
 #include "ext4_jbd2.h"
 #include "ext4.h"
 #include "group.h"
@@ -130,6 +131,7 @@ struct ext4_group_info {
 #ifdef DOUBLE_CHECK
 	void		*bb_bitmap;
 #endif
+	struct rw_semaphore alloc_sem;
 	unsigned short	bb_counters[];
 };
 
@@ -250,6 +252,7 @@ struct ext4_buddy {
 	struct super_block *bd_sb;
 	__u16 bd_blkbits;
 	ext4_group_t bd_group;
+	struct rw_semaphore *alloc_semp;
 };
 #define EXT4_MB_BITMAP(e4b)	((e4b)->bd_bitmap)
 #define EXT4_MB_BUDDY(e4b)	((e4b)->bd_buddy)
--- a/fs/ext4/resize.c
+++ b/fs/ext4/resize.c
@@ -747,6 +747,7 @@ int ext4_group_add(struct super_block *s
 	struct inode *inode = NULL;
 	handle_t *handle;
 	int gdb_off, gdb_num;
+	int num_grp_locked = 0;
 	int err, err2;
 
 	gdb_num = input->group / EXT4_DESC_PER_BLOCK(sb);
@@ -787,6 +788,7 @@ int ext4_group_add(struct super_block *s
 		}
 	}
 
+
 	if ((err = verify_group_input(sb, input)))
 		goto exit_put;
 
@@ -855,6 +857,7 @@ int ext4_group_add(struct super_block *s
          * using the new disk blocks.
          */
 
+	num_grp_locked = ext4_mb_get_buddy_cache_lock(sb, input->group);
 	/* Update group descriptor block for new group */
 	gdp = (struct ext4_group_desc *)((char *)primary->b_data +
 					 gdb_off * EXT4_DESC_SIZE(sb));
@@ -870,9 +873,11 @@ int ext4_group_add(struct super_block *s
 	 * We can allocate memory for mb_alloc based on the new group
 	 * descriptor
 	 */
-	err = ext4_mb_add_more_groupinfo(sb, input->group, gdp);
-	if (err)
+	err = ext4_mb_add_groupinfo(sb, input->group, gdp);
+	if (err) {
+		ext4_mb_put_buddy_cache_lock(sb, input->group, num_grp_locked);
 		goto exit_journal;
+	}
 
 	/*
 	 * Make the new blocks and inodes valid next.  We do this before
@@ -914,6 +919,7 @@ int ext4_group_add(struct super_block *s
 
 	/* Update the global fs size fields */
 	sbi->s_groups_count++;
+	ext4_mb_put_buddy_cache_lock(sb, input->group, num_grp_locked);
 
 	ext4_journal_dirty_metadata(handle, primary);
 
@@ -1081,45 +1087,6 @@ int ext4_group_extend(struct super_block
 	if ((err = ext4_journal_stop(handle)))
 		goto exit_put;
 
-	/*
-	 * Mark mballoc pages as not up to date so that they will be updated
-	 * next time they are loaded by ext4_mb_load_buddy.
-	 *
-	 * XXX Bad, Bad, BAD!!!  We should not be overloading the
-	 * Uptodate flag, particularly on thte bitmap bh, as way of
-	 * hinting to ext4_mb_load_buddy() that it needs to be
-	 * overloaded.  A user could take a LVM snapshot, then do an
-	 * on-line fsck, and clear the uptodate flag, and this would
-	 * not be a bug in userspace, but a bug in the kernel.  FIXME!!!
-	 */
-	{
-		struct ext4_sb_info *sbi = EXT4_SB(sb);
-		struct inode *inode = sbi->s_buddy_cache;
-		int blocks_per_page;
-		int block;
-		int pnum;
-		struct page *page;
-
-		/* Set buddy page as not up to date */
-		blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
-		block = group * 2;
-		pnum = block / blocks_per_page;
-		page = find_get_page(inode->i_mapping, pnum);
-		if (page != NULL) {
-			ClearPageUptodate(page);
-			page_cache_release(page);
-		}
-
-		/* Set bitmap page as not up to date */
-		block++;
-		pnum = block / blocks_per_page;
-		page = find_get_page(inode->i_mapping, pnum);
-		if (page != NULL) {
-			ClearPageUptodate(page);
-			page_cache_release(page);
-		}
-	}