ext4: allocate struct ext4_allocation_context from a kmem cache

struct ext4_allocation_context is rather large, and this bloats
the stack of many functions which use it. Allocating it from
a named slab cache will alleviate this.

For example, with this change (on top of the noinline patch sent earlier):

-ext4_mb_new_blocks 200
+ext4_mb_new_blocks 40

-ext4_mb_free_blocks 344
+ext4_mb_free_blocks 168

-ext4_mb_release_inode_pa 216
+ext4_mb_release_inode_pa 40

-ext4_mb_release_group_pa 192
+ext4_mb_release_group_pa 24

Most of these stack-allocated structs are actually used only for
mballoc history; and in those cases often a smaller struct would do.
So changing that may be another way around it, at least for those
functions, if preferred. For now, in those cases where the ac
is only for history, an allocation failure simply skips the history
recording, and does not cause any other failures.


Signed-off-by: Eric Sandeen <[e-mail]>
Signed-off-by: Mingming Cao <[e-mail]>
Signed-off-by: "Theodore Ts'o" <[e-mail]>

fs/ext4/mballoc.c

···
 #define MB_DEFAULT_GROUP_PREALLOC	512
 
 static struct kmem_cache *ext4_pspace_cachep;
+static struct kmem_cache *ext4_ac_cachep;
 
 #ifdef EXT4_BB_MAX_BLOCKS
 #undef EXT4_BB_MAX_BLOCKS
···
 	if (ext4_pspace_cachep == NULL)
 		return -ENOMEM;
 
+	ext4_ac_cachep =
+		kmem_cache_create("ext4_alloc_context",
+				     sizeof(struct ext4_allocation_context),
+				     0, SLAB_RECLAIM_ACCOUNT, NULL);
+	if (ext4_ac_cachep == NULL) {
+		kmem_cache_destroy(ext4_pspace_cachep);
+		return -ENOMEM;
+	}
 #ifdef CONFIG_PROC_FS
 	proc_root_ext4 = proc_mkdir(EXT4_ROOT, proc_root_fs);
 	if (proc_root_ext4 == NULL)
 		printk(KERN_ERR "EXT4-fs: Unable to create %s\n", EXT4_ROOT);
 #endif
-
 	return 0;
 }
 
···
 {
 	/* XXX: synchronize_rcu(); */
 	kmem_cache_destroy(ext4_pspace_cachep);
+	kmem_cache_destroy(ext4_ac_cachep);
 #ifdef CONFIG_PROC_FS
 	remove_proc_entry(EXT4_ROOT, proc_root_fs);
 #endif
···
 				struct buffer_head *bitmap_bh,
 				struct ext4_prealloc_space *pa)
 {
-	struct ext4_allocation_context ac;
+	struct ext4_allocation_context *ac;
 	struct super_block *sb = e4b->bd_sb;
 	struct ext4_sb_info *sbi = EXT4_SB(sb);
 	unsigned long end;
···
 	BUG_ON(group != e4b->bd_group && pa->pa_len != 0);
 	end = bit + pa->pa_len;
 
-	ac.ac_sb = sb;
-	ac.ac_inode = pa->pa_inode;
-	ac.ac_op = EXT4_MB_HISTORY_DISCARD;
+	ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
+
+	if (ac) {
+		ac->ac_sb = sb;
+		ac->ac_inode = pa->pa_inode;
+		ac->ac_op = EXT4_MB_HISTORY_DISCARD;
+	}
 
 	while (bit < end) {
 		bit = ext4_find_next_zero_bit(bitmap_bh->b_data, end, bit);
···
 				(unsigned) group);
 		free += next - bit;
 
-		ac.ac_b_ex.fe_group = group;
-		ac.ac_b_ex.fe_start = bit;
-		ac.ac_b_ex.fe_len = next - bit;
-		ac.ac_b_ex.fe_logical = 0;
-		ext4_mb_store_history(&ac);
+		if (ac) {
+			ac->ac_b_ex.fe_group = group;
+			ac->ac_b_ex.fe_start = bit;
+			ac->ac_b_ex.fe_len = next - bit;
+			ac->ac_b_ex.fe_logical = 0;
+			ext4_mb_store_history(ac);
+		}
 
 		mb_free_blocks(pa->pa_inode, e4b, bit, next - bit);
 		bit = next + 1;
···
 	}
 	BUG_ON(free != pa->pa_free);
 	atomic_add(free, &sbi->s_mb_discarded);
+	if (ac)
+		kmem_cache_free(ext4_ac_cachep, ac);
 
 	return err;
 }
···
 static int ext4_mb_release_group_pa(struct ext4_buddy *e4b,
 				struct ext4_prealloc_space *pa)
 {
-	struct ext4_allocation_context ac;
+	struct ext4_allocation_context *ac;
 	struct super_block *sb = e4b->bd_sb;
 	ext4_group_t group;
 	ext4_grpblk_t bit;
 
-	ac.ac_op = EXT4_MB_HISTORY_DISCARD;
+	ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
+
+	if (ac)
+		ac->ac_op = EXT4_MB_HISTORY_DISCARD;
 
 	BUG_ON(pa->pa_deleted == 0);
 	ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, &bit);
···
 	mb_free_blocks(pa->pa_inode, e4b, bit, pa->pa_len);
 	atomic_add(pa->pa_len, &EXT4_SB(sb)->s_mb_discarded);
 
-	ac.ac_sb = sb;
-	ac.ac_inode = NULL;
-	ac.ac_b_ex.fe_group = group;
-	ac.ac_b_ex.fe_start = bit;
-	ac.ac_b_ex.fe_len = pa->pa_len;
-	ac.ac_b_ex.fe_logical = 0;
-	ext4_mb_store_history(&ac);
+	if (ac) {
+		ac->ac_sb = sb;
+		ac->ac_inode = NULL;
+		ac->ac_b_ex.fe_group = group;
+		ac->ac_b_ex.fe_start = bit;
+		ac->ac_b_ex.fe_len = pa->pa_len;
+		ac->ac_b_ex.fe_logical = 0;
+		ext4_mb_store_history(ac);
+		kmem_cache_free(ext4_ac_cachep, ac);
+	}
 
 	return 0;
 }
···
 ext4_fsblk_t ext4_mb_new_blocks(handle_t *handle,
 				 struct ext4_allocation_request *ar, int *errp)
 {
-	struct ext4_allocation_context ac;
+	struct ext4_allocation_context *ac = NULL;
 	struct ext4_sb_info *sbi;
 	struct super_block *sb;
 	ext4_fsblk_t block = 0;
···
 	}
 	inquota = ar->len;
 
+	ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
+	if (!ac) {
+		*errp = -ENOMEM;
+		return 0;
+	}
+
 	ext4_mb_poll_new_transaction(sb, handle);
 
-	*errp = ext4_mb_initialize_context(&ac, ar);
+	*errp = ext4_mb_initialize_context(ac, ar);
 	if (*errp) {
 		ar->len = 0;
 		goto out;
 	}
 
-	ac.ac_op = EXT4_MB_HISTORY_PREALLOC;
-	if (!ext4_mb_use_preallocated(&ac)) {
+	ac->ac_op = EXT4_MB_HISTORY_PREALLOC;
+	if (!ext4_mb_use_preallocated(ac)) {
 
-		ac.ac_op = EXT4_MB_HISTORY_ALLOC;
-		ext4_mb_normalize_request(&ac, ar);
+		ac->ac_op = EXT4_MB_HISTORY_ALLOC;
+		ext4_mb_normalize_request(ac, ar);
 
 repeat:
 		/* allocate space in core */
-		ext4_mb_regular_allocator(&ac);
+		ext4_mb_regular_allocator(ac);
 
 		/* as we've just preallocated more space than
 		 * user requested orinally, we store allocated
 		 * space in a special descriptor */
-		if (ac.ac_status == AC_STATUS_FOUND &&
-				ac.ac_o_ex.fe_len < ac.ac_b_ex.fe_len)
-			ext4_mb_new_preallocation(&ac);
+		if (ac->ac_status == AC_STATUS_FOUND &&
+				ac->ac_o_ex.fe_len < ac->ac_b_ex.fe_len)
+			ext4_mb_new_preallocation(ac);
 	}
 
-	if (likely(ac.ac_status == AC_STATUS_FOUND)) {
-		ext4_mb_mark_diskspace_used(&ac, handle);
+	if (likely(ac->ac_status == AC_STATUS_FOUND)) {
+		ext4_mb_mark_diskspace_used(ac, handle);
 		*errp = 0;
-		block = ext4_grp_offs_to_block(sb, &ac.ac_b_ex);
-		ar->len = ac.ac_b_ex.fe_len;
+		block = ext4_grp_offs_to_block(sb, &ac->ac_b_ex);
+		ar->len = ac->ac_b_ex.fe_len;
 	} else {
-		freed  = ext4_mb_discard_preallocations(sb, ac.ac_o_ex.fe_len);
+		freed  = ext4_mb_discard_preallocations(sb, ac->ac_o_ex.fe_len);
 		if (freed)
 			goto repeat;
 		*errp = -ENOSPC;
-		ac.ac_b_ex.fe_len = 0;
+		ac->ac_b_ex.fe_len = 0;
 		ar->len = 0;
-		ext4_mb_show_ac(&ac);
+		ext4_mb_show_ac(ac);
 	}
 
-	ext4_mb_release_context(&ac);
+	ext4_mb_release_context(ac);
 
 out:
 	if (ar->len < inquota)
 		DQUOT_FREE_BLOCK(ar->inode, inquota - ar->len);
 
+	kmem_cache_free(ext4_ac_cachep, ac);
 	return block;
 }
 static void ext4_mb_poll_new_transaction(struct super_block *sb,
···
 {
 	struct buffer_head *bitmap_bh = 0;
 	struct super_block *sb = inode->i_sb;
-	struct ext4_allocation_context ac;
+	struct ext4_allocation_context *ac = NULL;
 	struct ext4_group_desc *gdp;
 	struct ext4_super_block *es;
 	unsigned long overflow;
···
 
 	ext4_debug("freeing block %lu\n", block);
 
-	ac.ac_op = EXT4_MB_HISTORY_FREE;
-	ac.ac_inode = inode;
-	ac.ac_sb = sb;
+	ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
+	if (ac) {
+		ac->ac_op = EXT4_MB_HISTORY_FREE;
+		ac->ac_inode = inode;
+		ac->ac_sb = sb;
+	}
 
 do_more:
 	overflow = 0;
···
 	BUFFER_TRACE(bitmap_bh, "dirtied bitmap block");
 	err = ext4_journal_dirty_metadata(handle, bitmap_bh);
 
-	ac.ac_b_ex.fe_group = block_group;
-	ac.ac_b_ex.fe_start = bit;
-	ac.ac_b_ex.fe_len = count;
-	ext4_mb_store_history(&ac);
+	if (ac) {
+		ac->ac_b_ex.fe_group = block_group;
+		ac->ac_b_ex.fe_start = bit;
+		ac->ac_b_ex.fe_len = count;
+		ext4_mb_store_history(ac);
+	}
 
 	if (metadata) {
 		/* blocks being freed are metadata. these blocks shouldn't
···
 error_return:
 	brelse(bitmap_bh);
 	ext4_std_error(sb, err);
+	if (ac)
+		kmem_cache_free(ext4_ac_cachep, ac);
 	return;
 }