]> bbs.cooldavid.org Git - net-next-2.6.git/blobdiff - fs/ext4/mballoc.c
ext4: Check return value of sb_getblk() and friends
[net-next-2.6.git] / fs / ext4 / mballoc.c
index 4b4ad4b7ce57ccb9ed8e2c773865323f1c0c646a..611c866ef3fe8cb3057732c4a18cf797ec1ec174 100644 (file)
 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.
@@ -2261,22 +2278,16 @@ 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;
        }
+       memset(meta_group_info[i], 0, kmem_cache_size(cachep));
        set_bit(EXT4_GROUP_INFO_NEED_INIT_BIT,
                &(meta_group_info[i]->bb_state));
 
@@ -2331,6 +2342,7 @@ 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 +2400,9 @@ 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 +2419,48 @@ 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 +2481,7 @@ 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 +2496,8 @@ 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 +2514,13 @@ 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 +2548,7 @@ 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 +2559,7 @@ 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) >>
@@ -2567,7 +2613,7 @@ static inline void ext4_issue_discard(struct super_block *sb,
        trace_ext4_discard_blocks(sb,
                        (unsigned long long) discard_block, count);
        ret = sb_issue_discard(sb, discard_block, count);
-       if (ret == EOPNOTSUPP) {
+       if (ret == -EOPNOTSUPP) {
                ext4_warning(sb, "discard not supported, disabling");
                clear_opt(EXT4_SB(sb)->s_mount_opt, DISCARD);
        }
@@ -2660,26 +2706,20 @@ static void ext4_remove_debugfs_entry(void)
 
 int __init init_ext4_mballoc(void)
 {
-       ext4_pspace_cachep =
-               kmem_cache_create("ext4_prealloc_space",
-                                    sizeof(struct ext4_prealloc_space),
-                                    0, SLAB_RECLAIM_ACCOUNT, NULL);
+       ext4_pspace_cachep = KMEM_CACHE(ext4_prealloc_space,
+                                       SLAB_RECLAIM_ACCOUNT);
        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);
+       ext4_ac_cachep = KMEM_CACHE(ext4_allocation_context,
+                                   SLAB_RECLAIM_ACCOUNT);
        if (ext4_ac_cachep == NULL) {
                kmem_cache_destroy(ext4_pspace_cachep);
                return -ENOMEM;
        }
 
-       ext4_free_ext_cachep =
-               kmem_cache_create("ext4_free_block_extents",
-                                    sizeof(struct ext4_free_data),
-                                    0, SLAB_RECLAIM_ACCOUNT, NULL);
+       ext4_free_ext_cachep = KMEM_CACHE(ext4_free_data,
+                                         SLAB_RECLAIM_ACCOUNT);
        if (ext4_free_ext_cachep == NULL) {
                kmem_cache_destroy(ext4_pspace_cachep);
                kmem_cache_destroy(ext4_ac_cachep);
@@ -2691,6 +2731,7 @@ 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 +2740,15 @@ 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();
 }
 
@@ -3535,8 +3585,7 @@ static int ext4_mb_new_preallocation(struct ext4_allocation_context *ac)
  */
 static noinline_for_stack int
 ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh,
-                       struct ext4_prealloc_space *pa,
-                       struct ext4_allocation_context *ac)
+                       struct ext4_prealloc_space *pa)
 {
        struct super_block *sb = e4b->bd_sb;
        struct ext4_sb_info *sbi = EXT4_SB(sb);
@@ -3554,11 +3603,6 @@ ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh,
        BUG_ON(group != e4b->bd_group && pa->pa_len != 0);
        end = bit + pa->pa_len;
 
-       if (ac) {
-               ac->ac_sb = sb;
-               ac->ac_inode = pa->pa_inode;
-       }
-
        while (bit < end) {
                bit = mb_find_next_zero_bit(bitmap_bh->b_data, end, bit);
                if (bit >= end)
@@ -3569,16 +3613,9 @@ ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh,
                         (unsigned) next - bit, (unsigned) group);
                free += next - bit;
 
-               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;
-                       trace_ext4_mballoc_discard(ac);
-               }
-
-               trace_ext4_mb_release_inode_pa(sb, ac, pa, grp_blk_start + bit,
-                                              next - bit);
+               trace_ext4_mballoc_discard(sb, NULL, group, bit, next - bit);
+               trace_ext4_mb_release_inode_pa(sb, pa->pa_inode, pa,
+                                              grp_blk_start + bit, next - bit);
                mb_free_blocks(pa->pa_inode, e4b, bit, next - bit);
                bit = next + 1;
        }
@@ -3601,29 +3638,19 @@ ext4_mb_release_inode_pa(struct ext4_buddy *e4b, struct buffer_head *bitmap_bh,
 
 static noinline_for_stack int
 ext4_mb_release_group_pa(struct ext4_buddy *e4b,
-                               struct ext4_prealloc_space *pa,
-                               struct ext4_allocation_context *ac)
+                               struct ext4_prealloc_space *pa)
 {
        struct super_block *sb = e4b->bd_sb;
        ext4_group_t group;
        ext4_grpblk_t bit;
 
-       trace_ext4_mb_release_group_pa(sb, ac, pa);
+       trace_ext4_mb_release_group_pa(sb, pa);
        BUG_ON(pa->pa_deleted == 0);
        ext4_get_group_no_and_offset(sb, pa->pa_pstart, &group, &bit);
        BUG_ON(group != e4b->bd_group && pa->pa_len != 0);
        mb_free_blocks(pa->pa_inode, e4b, bit, pa->pa_len);
        atomic_add(pa->pa_len, &EXT4_SB(sb)->s_mb_discarded);
-
-       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;
-               trace_ext4_mballoc_discard(ac);
-       }
+       trace_ext4_mballoc_discard(sb, NULL, group, bit, pa->pa_len);
 
        return 0;
 }
@@ -3644,7 +3671,6 @@ ext4_mb_discard_group_preallocations(struct super_block *sb,
        struct ext4_group_info *grp = ext4_get_group_info(sb, group);
        struct buffer_head *bitmap_bh = NULL;
        struct ext4_prealloc_space *pa, *tmp;
-       struct ext4_allocation_context *ac;
        struct list_head list;
        struct ext4_buddy e4b;
        int err;
@@ -3673,9 +3699,6 @@ ext4_mb_discard_group_preallocations(struct super_block *sb,
                needed = EXT4_BLOCKS_PER_GROUP(sb) + 1;
 
        INIT_LIST_HEAD(&list);
-       ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
-       if (ac)
-               ac->ac_sb = sb;
 repeat:
        ext4_lock_group(sb, group);
        list_for_each_entry_safe(pa, tmp,
@@ -3730,9 +3753,9 @@ repeat:
                spin_unlock(pa->pa_obj_lock);
 
                if (pa->pa_type == MB_GROUP_PA)
-                       ext4_mb_release_group_pa(&e4b, pa, ac);
+                       ext4_mb_release_group_pa(&e4b, pa);
                else
-                       ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa, ac);
+                       ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa);
 
                list_del(&pa->u.pa_tmp_list);
                call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback);
@@ -3740,8 +3763,6 @@ repeat:
 
 out:
        ext4_unlock_group(sb, group);
-       if (ac)
-               kmem_cache_free(ext4_ac_cachep, ac);
        ext4_mb_unload_buddy(&e4b);
        put_bh(bitmap_bh);
        return free;
@@ -3762,7 +3783,6 @@ void ext4_discard_preallocations(struct inode *inode)
        struct super_block *sb = inode->i_sb;
        struct buffer_head *bitmap_bh = NULL;
        struct ext4_prealloc_space *pa, *tmp;
-       struct ext4_allocation_context *ac;
        ext4_group_t group = 0;
        struct list_head list;
        struct ext4_buddy e4b;
@@ -3778,11 +3798,6 @@ void ext4_discard_preallocations(struct inode *inode)
 
        INIT_LIST_HEAD(&list);
 
-       ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
-       if (ac) {
-               ac->ac_sb = sb;
-               ac->ac_inode = inode;
-       }
 repeat:
        /* first, collect all pa's in the inode */
        spin_lock(&ei->i_prealloc_lock);
@@ -3852,7 +3867,7 @@ repeat:
 
                ext4_lock_group(sb, group);
                list_del(&pa->pa_group_list);
-               ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa, ac);
+               ext4_mb_release_inode_pa(&e4b, bitmap_bh, pa);
                ext4_unlock_group(sb, group);
 
                ext4_mb_unload_buddy(&e4b);
@@ -3861,8 +3876,6 @@ repeat:
                list_del(&pa->u.pa_tmp_list);
                call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback);
        }
-       if (ac)
-               kmem_cache_free(ext4_ac_cachep, ac);
 }
 
 /*
@@ -4060,14 +4073,10 @@ ext4_mb_discard_lg_preallocations(struct super_block *sb,
        struct ext4_buddy e4b;
        struct list_head discard_list;
        struct ext4_prealloc_space *pa, *tmp;
-       struct ext4_allocation_context *ac;
 
        mb_debug(1, "discard locality group preallocation\n");
 
        INIT_LIST_HEAD(&discard_list);
-       ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
-       if (ac)
-               ac->ac_sb = sb;
 
        spin_lock(&lg->lg_prealloc_lock);
        list_for_each_entry_rcu(pa, &lg->lg_prealloc_list[order],
@@ -4119,15 +4128,13 @@ ext4_mb_discard_lg_preallocations(struct super_block *sb,
                }
                ext4_lock_group(sb, group);
                list_del(&pa->pa_group_list);
-               ext4_mb_release_group_pa(&e4b, pa, ac);
+               ext4_mb_release_group_pa(&e4b, pa);
                ext4_unlock_group(sb, group);
 
                ext4_mb_unload_buddy(&e4b);
                list_del(&pa->u.pa_tmp_list);
                call_rcu(&(pa)->u.pa_rcu, ext4_mb_pa_callback);
        }
-       if (ac)
-               kmem_cache_free(ext4_ac_cachep, ac);
 }
 
 /*
@@ -4491,7 +4498,6 @@ void ext4_free_blocks(handle_t *handle, struct inode *inode,
 {
        struct buffer_head *bitmap_bh = NULL;
        struct super_block *sb = inode->i_sb;
-       struct ext4_allocation_context *ac = NULL;
        struct ext4_group_desc *gdp;
        unsigned long freed = 0;
        unsigned int overflow;
@@ -4531,6 +4537,8 @@ void ext4_free_blocks(handle_t *handle, struct inode *inode,
                        if (!bh)
                                tbh = sb_find_get_block(inode->i_sb,
                                                        block + i);
+                       if (unlikely(!tbh))
+                               continue;
                        ext4_forget(handle, flags & EXT4_FREE_BLOCKS_METADATA,
                                    inode, tbh, block + i);
                }
@@ -4546,12 +4554,6 @@ void ext4_free_blocks(handle_t *handle, struct inode *inode,
        if (!ext4_should_writeback_data(inode))
                flags |= EXT4_FREE_BLOCKS_METADATA;
 
-       ac = kmem_cache_alloc(ext4_ac_cachep, GFP_NOFS);
-       if (ac) {
-               ac->ac_inode = inode;
-               ac->ac_sb = sb;
-       }
-
 do_more:
        overflow = 0;
        ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
@@ -4609,12 +4611,7 @@ do_more:
                        BUG_ON(!mb_test_bit(bit + i, bitmap_bh->b_data));
        }
 #endif
-       if (ac) {
-               ac->ac_b_ex.fe_group = block_group;
-               ac->ac_b_ex.fe_start = bit;
-               ac->ac_b_ex.fe_len = count;
-               trace_ext4_mballoc_free(ac);
-       }
+       trace_ext4_mballoc_free(sb, inode, block_group, bit, count);
 
        err = ext4_mb_load_buddy(sb, block_group, &e4b);
        if (err)
@@ -4640,12 +4637,12 @@ do_more:
                 * with group lock held. generate_buddy look at
                 * them with group lock_held
                 */
+               if (test_opt(sb, DISCARD))
+                       ext4_issue_discard(sb, block_group, bit, count);
                ext4_lock_group(sb, block_group);
                mb_clear_bits(bitmap_bh->b_data, bit, count);
                mb_free_blocks(inode, &e4b, bit, count);
                ext4_mb_return_to_preallocation(inode, &e4b, block, count);
-               if (test_opt(sb, DISCARD))
-                       ext4_issue_discard(sb, block_group, bit, count);
        }
 
        ret = ext4_free_blks_count(sb, gdp) + count;
@@ -4685,7 +4682,5 @@ error_return:
                dquot_free_block(inode, freed);
        brelse(bitmap_bh);
        ext4_std_error(sb, err);
-       if (ac)
-               kmem_cache_free(ext4_ac_cachep, ac);
        return;
 }