This adds a new field in ext4_group_info to cache the largest available
block range in a block group; and don't load the buddy pages until *after*
we've done a sanity check on the block group.
With large allocation requests (e.g., fallocate(), 8MiB) and relatively full
partitions, it's easy to have no block groups with a block extent large
enough to satisfy the input request length. This currently causes the loop
during cr == 0 in ext4_mb_regular_allocator() to load the buddy bitmap pages
for EVERY block group. That can be a lot of pages. The patch below allows
us to call ext4_mb_good_group() BEFORE we load the buddy pages.
Signed-off-by: Curt Wohlgemuth <[email protected]>
---
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index bf938cf..d266003 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -1678,6 +1678,7 @@ struct ext4_group_info {
ext4_grpblk_t bb_first_free; /* first free block */
ext4_grpblk_t bb_free; /* total free blocks */
ext4_grpblk_t bb_fragments; /* nr of freespace fragments */
+ ext4_grpblk_t bb_largest_free_order;/* order of largest frag in BG */
struct list_head bb_prealloc_list;
#ifdef DOUBLE_CHECK
void *bb_bitmap;
diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index b423a36..9b3621f 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -658,6 +658,26 @@ static void ext4_mb_mark_free_simple(struct
super_block *sb,
}
}
+
+/* Cache the order of the largest free extent we have available in this block
+ * group. */
+static void
+mb_set_largest_free_order(struct super_block *sb, struct ext4_group_info *grp)
+{
+ int i;
+ int bits;
+
+ grp->bb_largest_free_order = -1; /* uninit */
+
+ bits = sb->s_blocksize_bits + 1;
+ for (i = bits; i >= 0; i--) {
+ if (grp->bb_counters[i] > 0) {
+ grp->bb_largest_free_order = i;
+ break;
+ }
+ }
+}
+
static noinline_for_stack
void ext4_mb_generate_buddy(struct super_block *sb,
void *buddy, void *bitmap, ext4_group_t group)
@@ -700,6 +720,7 @@ void ext4_mb_generate_buddy(struct super_block *sb,
*/
grp->bb_free = free;
}
+ mb_set_largest_free_order(sb, grp);
clear_bit(EXT4_GROUP_INFO_NEED_INIT_BIT, &(grp->bb_state));
@@ -725,6 +746,9 @@ void ext4_mb_generate_buddy(struct super_block *sb,
* contain blocks_per_page (PAGE_CACHE_SIZE / blocksize) blocks.
* So it can have information regarding groups_per_page which
* is blocks_per_page/2
+ *
+ * Locking note: This routine takes the block group lock of all groups
+ * for this page; do not hold this lock when calling this routine!
*/
static int ext4_mb_init_cache(struct page *page, char *incore)
@@ -910,6 +934,11 @@ out:
return err;
}
+/*
+ * Locking note: This routine calls ext4_mb_init_cache(), which takes the
+ * block group lock of all groups for this page; do not hold the BG lock when
+ * calling this routine!
+ */
static noinline_for_stack
int ext4_mb_init_group(struct super_block *sb, ext4_group_t group)
{
@@ -1004,6 +1033,11 @@ err:
return ret;
}
+/*
+ * Locking note: This routine calls ext4_mb_init_cache(), which takes the
+ * block group lock of all groups for this page; do not hold the BG lock when
+ * calling this routine!
+ */
static noinline_for_stack int
ext4_mb_load_buddy(struct super_block *sb, ext4_group_t group,
struct ext4_buddy *e4b)
@@ -1299,6 +1333,7 @@ static void mb_free_blocks(struct inode *inode,
struct ext4_buddy *e4b,
buddy = buddy2;
} while (1);
}
+ mb_set_largest_free_order(sb, e4b->bd_info);
mb_check_buddy(e4b);
}
@@ -1427,6 +1462,7 @@ static int mb_mark_used(struct ext4_buddy *e4b,
struct ext4_free_extent *ex)
e4b->bd_info->bb_counters[ord]++;
e4b->bd_info->bb_counters[ord]++;
}
+ mb_set_largest_free_order(e4b->bd_sb, e4b->bd_info);
mb_set_bits(EXT4_MB_BITMAP(e4b), ex->fe_start, len0);
mb_check_buddy(e4b);
@@ -1821,16 +1857,22 @@ void ext4_mb_scan_aligned(struct
ext4_allocation_context *ac,
}
}
+/* This is now called BEFORE we load the buddy bitmap. */
static int ext4_mb_good_group(struct ext4_allocation_context *ac,
ext4_group_t group, int cr)
{
unsigned free, fragments;
- unsigned i, bits;
int flex_size = ext4_flex_bg_size(EXT4_SB(ac->ac_sb));
struct ext4_group_info *grp = ext4_get_group_info(ac->ac_sb, group);
BUG_ON(cr < 0 || cr >= 4);
- BUG_ON(EXT4_MB_GRP_NEED_INIT(grp));
+
+ /* We only do this if the grp has never been initialized */
+ if (unlikely(EXT4_MB_GRP_NEED_INIT(grp))) {
+ int ret = ext4_mb_init_group(ac->ac_sb, group);
+ if (ret)
+ return 0;
+ }
free = grp->bb_free;
fragments = grp->bb_fragments;
@@ -1843,17 +1885,17 @@ static int ext4_mb_good_group(struct
ext4_allocation_context *ac,
case 0:
BUG_ON(ac->ac_2order == 0);
+ if (grp->bb_largest_free_order < ac->ac_2order) {
+ return 0;
+ }
+
/* Avoid using the first bg of a flexgroup for data files */
if ((ac->ac_flags & EXT4_MB_HINT_DATA) &&
(flex_size >= EXT4_FLEX_SIZE_DIR_ALLOC_SCHEME) &&
((group % flex_size) == 0))
return 0;
- bits = ac->ac_sb->s_blocksize_bits + 1;
- for (i = ac->ac_2order; i <= bits; i++)
- if (grp->bb_counters[i] > 0)
- return 1;
- break;
+ return 1;
case 1:
if ((free / fragments) >= ac->ac_g_ex.fe_len)
return 1;
@@ -2024,15 +2066,13 @@ repeat:
group = ac->ac_g_ex.fe_group;
for (i = 0; i < ngroups; group++, i++) {
- struct ext4_group_info *grp;
struct ext4_group_desc *desc;
if (group == ngroups)
group = 0;
- /* quick check to skip empty groups */
- grp = ext4_get_group_info(sb, group);
- if (grp->bb_free == 0)
+ /* This now checks without needing the buddy page */
+ if (!ext4_mb_good_group(ac, group, cr))
continue;
err = ext4_mb_load_buddy(sb, group, &e4b);
@@ -2040,12 +2080,6 @@ repeat:
goto out;
ext4_lock_group(sb, group);
- if (!ext4_mb_good_group(ac, group, cr)) {
- /* someone did allocation from this group */
- ext4_unlock_group(sb, group);
- ext4_mb_release_desc(&e4b);
- continue;
- }
ac->ac_groups_scanned++;
desc = ext4_get_group_desc(sb, group, NULL);
@@ -2255,6 +2289,7 @@ int ext4_mb_add_groupinfo(struct super_block
*sb, ext4_group_t group,
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_ROOT;
+ meta_group_info[i]->bb_largest_free_order = -1; /* uninit */
#ifdef DOUBLE_CHECK
{
On Fri, May 07, 2010 at 09:00:16AM -0700, Curt Wohlgemuth wrote:
> This adds a new field in ext4_group_info to cache the largest available
> block range in a block group; and don't load the buddy pages until *after*
> we've done a sanity check on the block group.
>
> With large allocation requests (e.g., fallocate(), 8MiB) and relatively full
> partitions, it's easy to have no block groups with a block extent large
> enough to satisfy the input request length. This currently causes the loop
> during cr == 0 in ext4_mb_regular_allocator() to load the buddy bitmap pages
> for EVERY block group. That can be a lot of pages. The patch below allows
> us to call ext4_mb_good_group() BEFORE we load the buddy pages.
>
>Signed-off-by: Curt Wohlgemuth <[email protected]>
Added to the ext4 patch queue.
- Ted