Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ryusuke...
[safe/jmp/linux-2.6] / fs / ext4 / resize.c
index b6ec184..3cfc343 100644 (file)
@@ -15,7 +15,6 @@
 #include <linux/slab.h>
 
 #include "ext4_jbd2.h"
-#include "group.h"
 
 #define outside(b, first, last)        ((b) < (first) || (b) >= (last))
 #define inside(b, first, last) ((b) >= (first) && (b) < (last))
@@ -50,7 +49,7 @@ static int verify_group_input(struct super_block *sb,
        ext4_get_group_no_and_offset(sb, start, NULL, &offset);
        if (group != sbi->s_groups_count)
                ext4_warning(sb, __func__,
-                            "Cannot add at group %u (only %lu groups)",
+                            "Cannot add at group %u (only %u groups)",
                             input->group, sbi->s_groups_count);
        else if (offset != 0)
                        ext4_warning(sb, __func__, "Last group not full");
@@ -149,7 +148,7 @@ static int extend_or_restart_transaction(handle_t *handle, int thresh,
 {
        int err;
 
-       if (handle->h_buffer_credits >= thresh)
+       if (ext4_handle_has_enough_credits(handle, thresh))
                return 0;
 
        err = ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA);
@@ -193,7 +192,7 @@ static int setup_new_group_blocks(struct super_block *sb,
        if (IS_ERR(handle))
                return PTR_ERR(handle);
 
-       lock_super(sb);
+       mutex_lock(&sbi->s_resize_lock);
        if (input->group != sbi->s_groups_count) {
                err = -EBUSY;
                goto exit_journal;
@@ -232,7 +231,7 @@ static int setup_new_group_blocks(struct super_block *sb,
                memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, gdb->b_size);
                set_buffer_uptodate(gdb);
                unlock_buffer(gdb);
-               ext4_journal_dirty_metadata(handle, gdb);
+               ext4_handle_dirty_metadata(handle, NULL, gdb);
                ext4_set_bit(bit, bh->b_data);
                brelse(gdb);
        }
@@ -251,7 +250,7 @@ static int setup_new_group_blocks(struct super_block *sb,
                        err = PTR_ERR(bh);
                        goto exit_bh;
                }
-               ext4_journal_dirty_metadata(handle, gdb);
+               ext4_handle_dirty_metadata(handle, NULL, gdb);
                ext4_set_bit(bit, bh->b_data);
                brelse(gdb);
        }
@@ -276,7 +275,7 @@ static int setup_new_group_blocks(struct super_block *sb,
                        err = PTR_ERR(it);
                        goto exit_bh;
                }
-               ext4_journal_dirty_metadata(handle, it);
+               ext4_handle_dirty_metadata(handle, NULL, it);
                brelse(it);
                ext4_set_bit(bit, bh->b_data);
        }
@@ -284,11 +283,9 @@ static int setup_new_group_blocks(struct super_block *sb,
        if ((err = extend_or_restart_transaction(handle, 2, bh)))
                goto exit_bh;
 
-       mark_bitmap_end(input->blocks_count, EXT4_BLOCKS_PER_GROUP(sb),
-                       bh->b_data);
-       ext4_journal_dirty_metadata(handle, bh);
+       mark_bitmap_end(input->blocks_count, sb->s_blocksize * 8, bh->b_data);
+       ext4_handle_dirty_metadata(handle, NULL, bh);
        brelse(bh);
-
        /* Mark unused entries in inode bitmap used */
        ext4_debug("clear inode bitmap %#04llx (+%llu)\n",
                   input->inode_bitmap, input->inode_bitmap - start);
@@ -297,14 +294,14 @@ static int setup_new_group_blocks(struct super_block *sb,
                goto exit_journal;
        }
 
-       mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), EXT4_BLOCKS_PER_GROUP(sb),
+       mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), sb->s_blocksize * 8,
                        bh->b_data);
-       ext4_journal_dirty_metadata(handle, bh);
+       ext4_handle_dirty_metadata(handle, NULL, bh);
 exit_bh:
        brelse(bh);
 
 exit_journal:
-       unlock_super(sb);
+       mutex_unlock(&sbi->s_resize_lock);
        if ((err2 = ext4_journal_stop(handle)) && !err)
                err = err2;
 
@@ -486,12 +483,12 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
         * reserved inode, and will become GDT blocks (primary and backup).
         */
        data[gdb_num % EXT4_ADDR_PER_BLOCK(sb)] = 0;
-       ext4_journal_dirty_metadata(handle, dind);
+       ext4_handle_dirty_metadata(handle, NULL, dind);
        brelse(dind);
        inode->i_blocks -= (gdbackups + 1) * sb->s_blocksize >> 9;
        ext4_mark_iloc_dirty(handle, inode, &iloc);
        memset((*primary)->b_data, 0, sb->s_blocksize);
-       ext4_journal_dirty_metadata(handle, *primary);
+       ext4_handle_dirty_metadata(handle, NULL, *primary);
 
        o_group_desc = EXT4_SB(sb)->s_group_desc;
        memcpy(n_group_desc, o_group_desc,
@@ -502,7 +499,7 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
        kfree(o_group_desc);
 
        le16_add_cpu(&es->s_reserved_gdt_blocks, -1);
-       ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
+       ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
 
        return 0;
 
@@ -618,7 +615,7 @@ static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
                       primary[i]->b_blocknr, gdbackups,
                       blk + primary[i]->b_blocknr); */
                data[gdbackups] = cpu_to_le32(blk + primary[i]->b_blocknr);
-               err2 = ext4_journal_dirty_metadata(handle, primary[i]);
+               err2 = ext4_handle_dirty_metadata(handle, NULL, primary[i]);
                if (!err)
                        err = err2;
        }
@@ -645,11 +642,12 @@ exit_free:
  * important part is that the new block and inode counts are in the backup
  * superblocks, and the location of the new group metadata in the GDT backups.
  *
- * We do not need lock_super() for this, because these blocks are not
- * otherwise touched by the filesystem code when it is mounted.  We don't
- * need to worry about last changing from sbi->s_groups_count, because the
- * worst that can happen is that we do not copy the full number of backups
- * at this time.  The resize which changed s_groups_count will backup again.
+ * We do not need take the s_resize_lock for this, because these
+ * blocks are not otherwise touched by the filesystem code when it is
+ * mounted.  We don't need to worry about last changing from
+ * sbi->s_groups_count, because the worst that can happen is that we
+ * do not copy the full number of backups at this time.  The resize
+ * which changed s_groups_count will backup again.
  */
 static void update_backups(struct super_block *sb,
                           int blk_off, char *data, int size)
@@ -676,7 +674,8 @@ static void update_backups(struct super_block *sb,
                struct buffer_head *bh;
 
                /* Out of journal space, and can't get more - abort - so sad */
-               if (handle->h_buffer_credits == 0 &&
+               if (ext4_handle_valid(handle) &&
+                   handle->h_buffer_credits == 0 &&
                    ext4_journal_extend(handle, EXT4_MAX_TRANS_DATA) &&
                    (err = ext4_journal_restart(handle, EXT4_MAX_TRANS_DATA)))
                        break;
@@ -696,7 +695,7 @@ static void update_backups(struct super_block *sb,
                        memset(bh->b_data + size, 0, rest);
                set_buffer_uptodate(bh);
                unlock_buffer(bh);
-               ext4_journal_dirty_metadata(handle, bh);
+               ext4_handle_dirty_metadata(handle, NULL, bh);
                brelse(bh);
        }
        if ((err2 = ext4_journal_stop(handle)) && !err)
@@ -715,7 +714,7 @@ static void update_backups(struct super_block *sb,
 exit_err:
        if (err) {
                ext4_warning(sb, __func__,
-                            "can't update backup for group %lu (err %d), "
+                            "can't update backup for group %u (err %d), "
                             "forcing fsck on next reboot", group, err);
                sbi->s_mount_state &= ~EXT4_VALID_FS;
                sbi->s_es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
@@ -761,13 +760,13 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
 
        if (ext4_blocks_count(es) + input->blocks_count <
            ext4_blocks_count(es)) {
-               ext4_warning(sb, __func__, "blocks_count overflow\n");
+               ext4_warning(sb, __func__, "blocks_count overflow");
                return -EINVAL;
        }
 
        if (le32_to_cpu(es->s_inodes_count) + EXT4_INODES_PER_GROUP(sb) <
            le32_to_cpu(es->s_inodes_count)) {
-               ext4_warning(sb, __func__, "inodes_count overflow\n");
+               ext4_warning(sb, __func__, "inodes_count overflow");
                return -EINVAL;
        }
 
@@ -787,6 +786,7 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
                }
        }
 
+
        if ((err = verify_group_input(sb, input)))
                goto exit_put;
 
@@ -808,7 +808,7 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
                goto exit_put;
        }
 
-       lock_super(sb);
+       mutex_lock(&sbi->s_resize_lock);
        if (input->group != sbi->s_groups_count) {
                ext4_warning(sb, __func__,
                             "multiple resizers run on filesystem!");
@@ -839,7 +839,7 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
         /*
          * OK, now we've set up the new group.  Time to make it active.
          *
-         * Current kernels don't lock all allocations via lock_super(),
+         * We do not lock all allocations via s_resize_lock
          * so we have to be safe wrt. concurrent accesses the group
          * data.  So we need to be careful to set all of the relevant
          * group descriptor data etc. *before* we enable the group.
@@ -859,18 +859,20 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
        gdp = (struct ext4_group_desc *)((char *)primary->b_data +
                                         gdb_off * EXT4_DESC_SIZE(sb));
 
+       memset(gdp, 0, EXT4_DESC_SIZE(sb));
        ext4_block_bitmap_set(sb, gdp, input->block_bitmap); /* LV FIXME */
        ext4_inode_bitmap_set(sb, gdp, input->inode_bitmap); /* LV FIXME */
        ext4_inode_table_set(sb, gdp, input->inode_table); /* LV FIXME */
-       gdp->bg_free_blocks_count = cpu_to_le16(input->free_blocks_count);
-       gdp->bg_free_inodes_count = cpu_to_le16(EXT4_INODES_PER_GROUP(sb));
+       ext4_free_blks_set(sb, gdp, input->free_blocks_count);
+       ext4_free_inodes_set(sb, gdp, EXT4_INODES_PER_GROUP(sb));
+       gdp->bg_flags = cpu_to_le16(EXT4_BG_INODE_ZEROED);
        gdp->bg_checksum = ext4_group_desc_csum(sbi, input->group, gdp);
 
        /*
         * We can allocate memory for mb_alloc based on the new group
         * descriptor
         */
-       err = ext4_mb_add_more_groupinfo(sb, input->group, gdp);
+       err = ext4_mb_add_groupinfo(sb, input->group, gdp);
        if (err)
                goto exit_journal;
 
@@ -894,12 +896,12 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
         *
         * The precise rules we use are:
         *
-        * * Writers of s_groups_count *must* hold lock_super
+        * * Writers of s_groups_count *must* hold s_resize_lock
         * AND
         * * Writers must perform a smp_wmb() after updating all dependent
         *   data and before modifying the groups count
         *
-        * * Readers must hold lock_super() over the access
+        * * Readers must hold s_resize_lock over the access
         * OR
         * * Readers must perform an smp_rmb() after reading the groups count
         *   and before reading any dependent data.
@@ -915,7 +917,7 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
        /* Update the global fs size fields */
        sbi->s_groups_count++;
 
-       ext4_journal_dirty_metadata(handle, primary);
+       ext4_handle_dirty_metadata(handle, NULL, primary);
 
        /* Update the reserved block counts only once the new group is
         * active. */
@@ -931,17 +933,17 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
        if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG)) {
                ext4_group_t flex_group;
                flex_group = ext4_flex_group(sbi, input->group);
-               sbi->s_flex_groups[flex_group].free_blocks +=
-                       input->free_blocks_count;
-               sbi->s_flex_groups[flex_group].free_inodes +=
-                       EXT4_INODES_PER_GROUP(sb);
+               atomic_add(input->free_blocks_count,
+                          &sbi->s_flex_groups[flex_group].free_blocks);
+               atomic_add(EXT4_INODES_PER_GROUP(sb),
+                          &sbi->s_flex_groups[flex_group].free_inodes);
        }
 
-       ext4_journal_dirty_metadata(handle, sbi->s_sbh);
+       ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
        sb->s_dirt = 1;
 
 exit_journal:
-       unlock_super(sb);
+       mutex_unlock(&sbi->s_resize_lock);
        if ((err2 = ext4_journal_stop(handle)) && !err)
                err = err2;
        if (!err) {
@@ -975,13 +977,11 @@ int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
        struct buffer_head *bh;
        handle_t *handle;
        int err;
-       unsigned long freed_blocks;
        ext4_group_t group;
-       struct ext4_group_info *grp;
 
        /* We don't need to worry about locking wrt other resizers just
         * yet: we're going to revalidate es->s_blocks_count after
-        * taking lock_super() below. */
+        * taking the s_resize_lock below. */
        o_blocks_count = ext4_blocks_count(es);
        o_groups_count = EXT4_SB(sb)->s_groups_count;
 
@@ -997,8 +997,7 @@ int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
                        " too large to resize to %llu blocks safely\n",
                        sb->s_id, n_blocks_count);
                if (sizeof(sector_t) < 8)
-                       ext4_warning(sb, __func__,
-                       "CONFIG_LBD not enabled\n");
+                       ext4_warning(sb, __func__, "CONFIG_LBDAF not enabled");
                return -EINVAL;
        }
 
@@ -1052,11 +1051,11 @@ int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
                goto exit_put;
        }
 
-       lock_super(sb);
+       mutex_lock(&EXT4_SB(sb)->s_resize_lock);
        if (o_blocks_count != ext4_blocks_count(es)) {
                ext4_warning(sb, __func__,
                             "multiple resizers run on filesystem!");
-               unlock_super(sb);
+               mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
                ext4_journal_stop(handle);
                err = -EBUSY;
                goto exit_put;
@@ -1066,67 +1065,23 @@ int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es,
                                                 EXT4_SB(sb)->s_sbh))) {
                ext4_warning(sb, __func__,
                             "error %d on journal write access", err);
-               unlock_super(sb);
+               mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
                ext4_journal_stop(handle);
                goto exit_put;
        }
        ext4_blocks_count_set(es, o_blocks_count + add);
-       ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
+       ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
        sb->s_dirt = 1;
-       unlock_super(sb);
+       mutex_unlock(&EXT4_SB(sb)->s_resize_lock);
        ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count,
                   o_blocks_count + add);
-       ext4_free_blocks_sb(handle, sb, o_blocks_count, add, &freed_blocks);
+       /* We add the blocks to the bitmap and set the group need init bit */
+       ext4_add_groupblocks(handle, sb, o_blocks_count, add);
        ext4_debug("freed blocks %llu through %llu\n", o_blocks_count,
                   o_blocks_count + add);
        if ((err = ext4_journal_stop(handle)))
                goto exit_put;
 
-       /*
-        * Mark mballoc pages as not up to date so that they will be updated
-        * next time they are loaded by ext4_mb_load_buddy.
-        *
-        * XXX Bad, Bad, BAD!!!  We should not be overloading the
-        * Uptodate flag, particularly on thte bitmap bh, as way of
-        * hinting to ext4_mb_load_buddy() that it needs to be
-        * overloaded.  A user could take a LVM snapshot, then do an
-        * on-line fsck, and clear the uptodate flag, and this would
-        * not be a bug in userspace, but a bug in the kernel.  FIXME!!!
-        */
-       {
-               struct ext4_sb_info *sbi = EXT4_SB(sb);
-               struct inode *inode = sbi->s_buddy_cache;
-               int blocks_per_page;
-               int block;
-               int pnum;
-               struct page *page;
-
-               /* Set buddy page as not up to date */
-               blocks_per_page = PAGE_CACHE_SIZE / sb->s_blocksize;
-               block = group * 2;
-               pnum = block / blocks_per_page;
-               page = find_get_page(inode->i_mapping, pnum);
-               if (page != NULL) {
-                       ClearPageUptodate(page);
-                       page_cache_release(page);
-               }
-
-               /* Set bitmap page as not up to date */
-               block++;
-               pnum = block / blocks_per_page;
-               page = find_get_page(inode->i_mapping, pnum);
-               if (page != NULL) {
-                       ClearPageUptodate(page);
-                       page_cache_release(page);
-               }
-
-               /* Get the info on the last group */
-               grp = ext4_get_group_info(sb, group);
-
-               /* Update free blocks in group info */
-               ext4_mb_update_group_info(grp, add);
-       }
-
        if (test_opt(sb, DEBUG))
                printk(KERN_DEBUG "EXT4-fs: extended group to %llu blocks\n",
                       ext4_blocks_count(es));