ext4: make grpinfo slab cache names static
authorEric Sandeen <sandeen@redhat.com>
Sat, 12 Feb 2011 13:12:18 +0000 (08:12 -0500)
committerTheodore Ts'o <tytso@mit.edu>
Sat, 12 Feb 2011 13:12:18 +0000 (08:12 -0500)
In 2.6.37 I was running into oopses with repeated module
loads & unloads.  I tracked this down to:

fb1813f4 ext4: use dedicated slab caches for group_info structures

(this was in addition to the features advert unload problem)

The kstrdup & subsequent kfree of the cache name was causing
a double free.  In slub, at least, if I read it right it allocates
& frees the name itself, slab seems to do something different...
so in slub I think we were leaking -our- cachep->name, and double
freeing the one allocated by slub.

After getting lost in slab/slub/slob a bit, I just looked at other
sized-caches that get allocated.  jbd2, biovec, sgpool all do it
more or less the way jbd2 does.  Below patch follows the jbd2
method of dynamically allocating a cache at mount time from
a list of static names.

(This might also possibly fix a race creating the caches with
parallel mounts running).

[Folded in a fix from Dan Carpenter which fixed an off-by-one error in
the original patch]

Cc: stable@kernel.org
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
fs/ext4/mballoc.c

index 851f49b2f9d24e6715dae75459b344019c7351fe..d1fe09aea73dc92419b0bf574694b6add33d7d8e 100644 (file)
@@ -342,10 +342,15 @@ 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)
+#define NR_GRPINFO_CACHES 8
 static struct kmem_cache *ext4_groupinfo_caches[NR_GRPINFO_CACHES];
 
+static const char *ext4_groupinfo_slab_names[NR_GRPINFO_CACHES] = {
+       "ext4_groupinfo_1k", "ext4_groupinfo_2k", "ext4_groupinfo_4k",
+       "ext4_groupinfo_8k", "ext4_groupinfo_16k", "ext4_groupinfo_32k",
+       "ext4_groupinfo_64k", "ext4_groupinfo_128k"
+};
+
 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,
@@ -2414,6 +2419,55 @@ err_freesgi:
        return -ENOMEM;
 }
 
+static void ext4_groupinfo_destroy_slabs(void)
+{
+       int i;
+
+       for (i = 0; i < NR_GRPINFO_CACHES; i++) {
+               if (ext4_groupinfo_caches[i])
+                       kmem_cache_destroy(ext4_groupinfo_caches[i]);
+               ext4_groupinfo_caches[i] = NULL;
+       }
+}
+
+static int ext4_groupinfo_create_slab(size_t size)
+{
+       static DEFINE_MUTEX(ext4_grpinfo_slab_create_mutex);
+       int slab_size;
+       int blocksize_bits = order_base_2(size);
+       int cache_index = blocksize_bits - EXT4_MIN_BLOCK_LOG_SIZE;
+       struct kmem_cache *cachep;
+
+       if (cache_index >= NR_GRPINFO_CACHES)
+               return -EINVAL;
+
+       if (unlikely(cache_index < 0))
+               cache_index = 0;
+
+       mutex_lock(&ext4_grpinfo_slab_create_mutex);
+       if (ext4_groupinfo_caches[cache_index]) {
+               mutex_unlock(&ext4_grpinfo_slab_create_mutex);
+               return 0;       /* Already created */
+       }
+
+       slab_size = offsetof(struct ext4_group_info,
+                               bb_counters[blocksize_bits + 2]);
+
+       cachep = kmem_cache_create(ext4_groupinfo_slab_names[cache_index],
+                                       slab_size, 0, SLAB_RECLAIM_ACCOUNT,
+                                       NULL);
+
+       mutex_unlock(&ext4_grpinfo_slab_create_mutex);
+       if (!cachep) {
+               printk(KERN_EMERG "EXT4: no memory for groupinfo slab cache\n");
+               return -ENOMEM;
+       }
+
+       ext4_groupinfo_caches[cache_index] = cachep;
+
+       return 0;
+}
+
 int ext4_mb_init(struct super_block *sb, int needs_recovery)
 {
        struct ext4_sb_info *sbi = EXT4_SB(sb);
@@ -2421,9 +2475,6 @@ 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);
 
@@ -2440,30 +2491,9 @@ int ext4_mb_init(struct super_block *sb, int needs_recovery)
                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;
-       }
+       ret = ext4_groupinfo_create_slab(sb->s_blocksize);
+       if (ret < 0)
+               goto out;
 
        /* order 0 is regular bitmap */
        sbi->s_mb_maxs[0] = sb->s_blocksize << 3;
@@ -2520,7 +2550,6 @@ out:
        if (ret) {
                kfree(sbi->s_mb_offsets);
                kfree(sbi->s_mb_maxs);
-               kfree(namep);
        }
        return ret;
 }
@@ -2734,7 +2763,6 @@ int __init ext4_init_mballoc(void)
 
 void ext4_exit_mballoc(void)
 {
-       int i;
        /*
         * Wait for completion of call_rcu()'s on ext4_pspace_cachep
         * before destroying the slab cache.
@@ -2743,15 +2771,7 @@ void ext4_exit_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_groupinfo_destroy_slabs();
        ext4_remove_debugfs_entry();
 }