lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:	Wed, 8 Sep 2010 14:51:23 -0400
From:	Ted Ts'o <tytso@....edu>
To:	Curt Wohlgemuth <curtw@...gle.com>
Cc:	ext4 development <linux-ext4@...r.kernel.org>
Subject: Re: [PATCH] ext4: use dedicated slab caches for group_info
 structures

Thanks, Curt.

This is what I've checked into the ext4 tree.  There are a few minor
checkpatch cleanups, but it's basically what you've sent me.

	   	     	      		     - Ted

>From 6e8a8e99d450636f5dd65a129253927507c9a006 Mon Sep 17 00:00:00 2001
From: Curt Wohlgemuth <curtw@...gle.com>
Date: Wed, 8 Sep 2010 14:44:50 -0400
Subject: [PATCH] ext4: use dedicated slab caches for group_info structures

ext4_group_info structures are currently allocated with kmalloc().
With a typical 4K block size, these are 136 bytes each -- meaning
they'll each consume a 256-byte slab object.  On a system with many
ext4 large partitions, that's a lot of wasted kernel slab space.
(E.g., a single 1TB partition will have about 8000 block groups, using
about 2MB of slab, of which nearly 1MB is wasted.)

This patch creates an array of slab pointers created as needed --
depending on the superblock block size -- and uses these slabs to
allocate the group info objects.

Google-Bug-Id: 2980809

Signed-off-by: Curt Wohlgemuth <curtw@...gle.com>
Signed-off-by: "Theodore Ts'o" <tytso@....edu>
---
 fs/ext4/ext4.h    |    1 +
 fs/ext4/mballoc.c |   97 +++++++++++++++++++++++++++++++++++++++++-----------
 2 files changed, 77 insertions(+), 21 deletions(-)

diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 889ec9d..b364b9d 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -205,6 +205,7 @@ typedef struct ext4_io_end {
 #define EXT4_MIN_BLOCK_SIZE		1024
 #define	EXT4_MAX_BLOCK_SIZE		65536
 #define EXT4_MIN_BLOCK_LOG_SIZE		10
+#define EXT4_MAX_BLOCK_LOG_SIZE		16
 #ifdef __KERNEL__
 # define EXT4_BLOCK_SIZE(s)		((s)->s_blocksize)
 #else
diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index 4b4ad4b..3d99432 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -338,6 +338,14 @@
 static struct kmem_cache *ext4_pspace_cachep;
 static struct kmem_cache *ext4_ac_cachep;
 static struct kmem_cache *ext4_free_ext_cachep;
+
+/* We create slab caches for groupinfo data structures based on the
+ * superblock block size.  There will be one per mounted filesystem for
+ * each unique s_blocksize_bits */
+#define NR_GRPINFO_CACHES 	\
+	(EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE + 1)
+static struct kmem_cache *ext4_groupinfo_caches[NR_GRPINFO_CACHES];
+
 static void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
 					ext4_group_t group);
 static void ext4_mb_generate_from_freelist(struct super_block *sb, void *bitmap,
@@ -2233,15 +2241,24 @@ static const struct file_operations ext4_mb_seq_groups_fops = {
 	.release	= seq_release,
 };
 
+static struct kmem_cache *get_groupinfo_cache(int blocksize_bits)
+{
+	int cache_index = blocksize_bits - EXT4_MIN_BLOCK_LOG_SIZE;
+	struct kmem_cache *cachep = ext4_groupinfo_caches[cache_index];
+
+	BUG_ON(!cachep);
+	return cachep;
+}
 
 /* Create and initialize ext4_group_info data for the given group. */
 int ext4_mb_add_groupinfo(struct super_block *sb, ext4_group_t group,
 			  struct ext4_group_desc *desc)
 {
-	int i, len;
+	int i;
 	int metalen = 0;
 	struct ext4_sb_info *sbi = EXT4_SB(sb);
 	struct ext4_group_info **meta_group_info;
+	struct kmem_cache *cachep = get_groupinfo_cache(sb->s_blocksize_bits);
 
 	/*
 	 * First check if this group is the first of a reserved block.
@@ -2261,18 +2278,11 @@ int ext4_mb_add_groupinfo(struct super_block *sb, ext4_group_t group,
 			meta_group_info;
 	}
 
-	/*
-	 * calculate needed size. if change bb_counters size,
-	 * don't forget about ext4_mb_generate_buddy()
-	 */
-	len = offsetof(typeof(**meta_group_info),
-		       bb_counters[sb->s_blocksize_bits + 2]);
-
 	meta_group_info =
 		sbi->s_group_info[group >> EXT4_DESC_PER_BLOCK_BITS(sb)];
 	i = group & (EXT4_DESC_PER_BLOCK(sb) - 1);
 
-	meta_group_info[i] = kzalloc(len, GFP_KERNEL);
+	meta_group_info[i] = kmem_cache_alloc(cachep, GFP_KERNEL);
 	if (meta_group_info[i] == NULL) {
 		printk(KERN_ERR "EXT4-fs: can't allocate buddy mem\n");
 		goto exit_group_info;
@@ -2331,6 +2341,7 @@ static int ext4_mb_init_backend(struct super_block *sb)
 	int num_meta_group_infos_max;
 	int array_size;
 	struct ext4_group_desc *desc;
+	struct kmem_cache *cachep;
 
 	/* This is the number of blocks used by GDT */
 	num_meta_group_infos = (ngroups + EXT4_DESC_PER_BLOCK(sb) -
@@ -2388,8 +2399,9 @@ static int ext4_mb_init_backend(struct super_block *sb)
 	return 0;
 
 err_freebuddy:
+	cachep = get_groupinfo_cache(sb->s_blocksize_bits);
 	while (i-- > 0)
-		kfree(ext4_get_group_info(sb, i));
+		kmem_cache_free(cachep, ext4_get_group_info(sb, i));
 	i = num_meta_group_infos;
 	while (i-- > 0)
 		kfree(sbi->s_group_info[i]);
@@ -2406,19 +2418,48 @@ int ext4_mb_init(struct super_block *sb, int needs_recovery)
 	unsigned offset;
 	unsigned max;
 	int ret;
+	int cache_index;
+	struct kmem_cache *cachep;
+	char *namep = NULL;
 
 	i = (sb->s_blocksize_bits + 2) * sizeof(*sbi->s_mb_offsets);
 
 	sbi->s_mb_offsets = kmalloc(i, GFP_KERNEL);
 	if (sbi->s_mb_offsets == NULL) {
-		return -ENOMEM;
+		ret = -ENOMEM;
+		goto out;
 	}
 
 	i = (sb->s_blocksize_bits + 2) * sizeof(*sbi->s_mb_maxs);
 	sbi->s_mb_maxs = kmalloc(i, GFP_KERNEL);
 	if (sbi->s_mb_maxs == NULL) {
-		kfree(sbi->s_mb_offsets);
-		return -ENOMEM;
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	cache_index = sb->s_blocksize_bits - EXT4_MIN_BLOCK_LOG_SIZE;
+	cachep = ext4_groupinfo_caches[cache_index];
+	if (!cachep) {
+		char name[32];
+		int len = offsetof(struct ext4_group_info,
+					bb_counters[sb->s_blocksize_bits + 2]);
+
+		sprintf(name, "ext4_groupinfo_%d", sb->s_blocksize_bits);
+		namep = kstrdup(name, GFP_KERNEL);
+		if (!namep) {
+			ret = -ENOMEM;
+			goto out;
+		}
+
+		/* Need to free the kmem_cache_name() when we
+		 * destroy the slab */
+		cachep = kmem_cache_create(namep, len, 0,
+					     SLAB_RECLAIM_ACCOUNT, NULL);
+		if (!cachep) {
+			ret = -ENOMEM;
+			goto out;
+		}
+		ext4_groupinfo_caches[cache_index] = cachep;
 	}
 
 	/* order 0 is regular bitmap */
@@ -2439,9 +2480,7 @@ int ext4_mb_init(struct super_block *sb, int needs_recovery)
 	/* init file for buddy data */
 	ret = ext4_mb_init_backend(sb);
 	if (ret != 0) {
-		kfree(sbi->s_mb_offsets);
-		kfree(sbi->s_mb_maxs);
-		return ret;
+		goto out;
 	}
 
 	spin_lock_init(&sbi->s_md_lock);
@@ -2456,9 +2495,8 @@ int ext4_mb_init(struct super_block *sb, int needs_recovery)
 
 	sbi->s_locality_groups = alloc_percpu(struct ext4_locality_group);
 	if (sbi->s_locality_groups == NULL) {
-		kfree(sbi->s_mb_offsets);
-		kfree(sbi->s_mb_maxs);
-		return -ENOMEM;
+		ret = -ENOMEM;
+		goto out;
 	}
 	for_each_possible_cpu(i) {
 		struct ext4_locality_group *lg;
@@ -2475,7 +2513,13 @@ int ext4_mb_init(struct super_block *sb, int needs_recovery)
 
 	if (sbi->s_journal)
 		sbi->s_journal->j_commit_callback = release_blocks_on_commit;
-	return 0;
+out:
+	if (ret) {
+		kfree(sbi->s_mb_offsets);
+		kfree(sbi->s_mb_maxs);
+		kfree(namep);
+	}
+	return ret;
 }
 
 /* need to called with the ext4 group lock held */
@@ -2503,6 +2547,7 @@ int ext4_mb_release(struct super_block *sb)
 	int num_meta_group_infos;
 	struct ext4_group_info *grinfo;
 	struct ext4_sb_info *sbi = EXT4_SB(sb);
+	struct kmem_cache *cachep = get_groupinfo_cache(sb->s_blocksize_bits);
 
 	if (sbi->s_group_info) {
 		for (i = 0; i < ngroups; i++) {
@@ -2513,7 +2558,7 @@ int ext4_mb_release(struct super_block *sb)
 			ext4_lock_group(sb, i);
 			ext4_mb_cleanup_pa(grinfo);
 			ext4_unlock_group(sb, i);
-			kfree(grinfo);
+			kmem_cache_free(cachep, grinfo);
 		}
 		num_meta_group_infos = (ngroups +
 				EXT4_DESC_PER_BLOCK(sb) - 1) >>
@@ -2691,6 +2736,7 @@ int __init init_ext4_mballoc(void)
 
 void exit_ext4_mballoc(void)
 {
+	int i;
 	/*
 	 * Wait for completion of call_rcu()'s on ext4_pspace_cachep
 	 * before destroying the slab cache.
@@ -2699,6 +2745,15 @@ void exit_ext4_mballoc(void)
 	kmem_cache_destroy(ext4_pspace_cachep);
 	kmem_cache_destroy(ext4_ac_cachep);
 	kmem_cache_destroy(ext4_free_ext_cachep);
+
+	for (i = 0; i < NR_GRPINFO_CACHES; i++) {
+		struct kmem_cache *cachep = ext4_groupinfo_caches[i];
+		if (cachep) {
+			char *name = (char *)kmem_cache_name(cachep);
+			kmem_cache_destroy(cachep);
+			kfree(name);
+		}
+	}
 	ext4_remove_debugfs_entry();
 }
 
-- 
1.7.1

--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Powered by blists - more mailing lists