fedora core 6 1.2949 + vserver 2.2.0
[linux-2.6.git] / fs / ext3 / resize.c
index d1fe21d..b73cba1 100644 (file)
@@ -8,7 +8,6 @@
  * This could probably be made into a module, because it is not often in use.
  */
 
-#include <linux/config.h>
 
 #define EXT3FS_DEBUG
 
@@ -28,23 +27,23 @@ static int verify_group_input(struct super_block *sb,
 {
        struct ext3_sb_info *sbi = EXT3_SB(sb);
        struct ext3_super_block *es = sbi->s_es;
-       unsigned start = le32_to_cpu(es->s_blocks_count);
-       unsigned end = start + input->blocks_count;
+       ext3_fsblk_t start = le32_to_cpu(es->s_blocks_count);
+       ext3_fsblk_t end = start + input->blocks_count;
        unsigned group = input->group;
-       unsigned itend = input->inode_table + EXT3_SB(sb)->s_itb_per_group;
+       ext3_fsblk_t itend = input->inode_table + sbi->s_itb_per_group;
        unsigned overhead = ext3_bg_has_super(sb, group) ?
                (1 + ext3_bg_num_gdb(sb, group) +
                 le16_to_cpu(es->s_reserved_gdt_blocks)) : 0;
-       unsigned metaend = start + overhead;
-       struct buffer_head *bh;
-       int free_blocks_count;
+       ext3_fsblk_t metaend = start + overhead;
+       struct buffer_head *bh = NULL;
+       ext3_grpblk_t free_blocks_count;
        int err = -EINVAL;
 
        input->free_blocks_count = free_blocks_count =
                input->blocks_count - 2 - overhead - sbi->s_itb_per_group;
 
        if (test_opt(sb, DEBUG))
-               printk("EXT3-fs: adding %s group %u: %u blocks "
+               printk(KERN_DEBUG "EXT3-fs: adding %s group %u: %u blocks "
                       "(%d free, %u reserved)\n",
                       ext3_bg_has_super(sb, input->group) ? "normal" :
                       "no-super", input->group, input->blocks_count,
@@ -64,7 +63,8 @@ static int verify_group_input(struct super_block *sb,
                ext3_warning(sb, __FUNCTION__, "Bad blocks count %u",
                             input->blocks_count);
        else if (!(bh = sb_bread(sb, end - 1)))
-               ext3_warning(sb, __FUNCTION__, "Cannot read last block (%u)",
+               ext3_warning(sb, __FUNCTION__,
+                            "Cannot read last block ("E3FSBLK")",
                             end - 1);
        else if (outside(input->block_bitmap, start, end))
                ext3_warning(sb, __FUNCTION__,
@@ -77,7 +77,7 @@ static int verify_group_input(struct super_block *sb,
        else if (outside(input->inode_table, start, end) ||
                 outside(itend - 1, start, end))
                ext3_warning(sb, __FUNCTION__,
-                            "Inode table not in group (blocks %u-%u)",
+                            "Inode table not in group (blocks %u-"E3FSBLK")",
                             input->inode_table, itend - 1);
        else if (input->inode_bitmap == input->block_bitmap)
                ext3_warning(sb, __FUNCTION__,
@@ -85,46 +85,53 @@ static int verify_group_input(struct super_block *sb,
                             input->block_bitmap);
        else if (inside(input->block_bitmap, input->inode_table, itend))
                ext3_warning(sb, __FUNCTION__,
-                            "Block bitmap (%u) in inode table (%u-%u)",
+                            "Block bitmap (%u) in inode table (%u-"E3FSBLK")",
                             input->block_bitmap, input->inode_table, itend-1);
        else if (inside(input->inode_bitmap, input->inode_table, itend))
                ext3_warning(sb, __FUNCTION__,
-                            "Inode bitmap (%u) in inode table (%u-%u)",
+                            "Inode bitmap (%u) in inode table (%u-"E3FSBLK")",
                             input->inode_bitmap, input->inode_table, itend-1);
        else if (inside(input->block_bitmap, start, metaend))
                ext3_warning(sb, __FUNCTION__,
-                            "Block bitmap (%u) in GDT table (%u-%u)",
+                            "Block bitmap (%u) in GDT table"
+                            " ("E3FSBLK"-"E3FSBLK")",
                             input->block_bitmap, start, metaend - 1);
        else if (inside(input->inode_bitmap, start, metaend))
                ext3_warning(sb, __FUNCTION__,
-                            "Inode bitmap (%u) in GDT table (%u-%u)",
+                            "Inode bitmap (%u) in GDT table"
+                            " ("E3FSBLK"-"E3FSBLK")",
                             input->inode_bitmap, start, metaend - 1);
        else if (inside(input->inode_table, start, metaend) ||
                 inside(itend - 1, start, metaend))
                ext3_warning(sb, __FUNCTION__,
-                            "Inode table (%u-%u) overlaps GDT table (%u-%u)",
+                            "Inode table (%u-"E3FSBLK") overlaps"
+                            "GDT table ("E3FSBLK"-"E3FSBLK")",
                             input->inode_table, itend - 1, start, metaend - 1);
-       else {
-               brelse(bh);
+       else
                err = 0;
-       }
+       brelse(bh);
 
        return err;
 }
 
 static struct buffer_head *bclean(handle_t *handle, struct super_block *sb,
-                                 unsigned long blk)
+                                 ext3_fsblk_t blk)
 {
        struct buffer_head *bh;
        int err;
 
        bh = sb_getblk(sb, blk);
-       set_buffer_uptodate(bh);
+       if (!bh)
+               return ERR_PTR(-EIO);
        if ((err = ext3_journal_get_write_access(handle, bh))) {
                brelse(bh);
                bh = ERR_PTR(err);
-       } else
+       } else {
+               lock_buffer(bh);
                memset(bh->b_data, 0, sb->s_blocksize);
+               set_buffer_uptodate(bh);
+               unlock_buffer(bh);
+       }
 
        return bh;
 }
@@ -154,27 +161,24 @@ static void mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
  * changing blocks outside the actual filesystem.  We still do journaling to
  * ensure the recovery is correct in case of a failure just after resize.
  * If any part of this fails, we simply abort the resize.
- *
- * We only pass inode because of the ext3 journal wrappers.
  */
-static int setup_new_group_blocks(struct super_block *sb, struct inode *inode,
+static int setup_new_group_blocks(struct super_block *sb,
                                  struct ext3_new_group_data *input)
 {
        struct ext3_sb_info *sbi = EXT3_SB(sb);
-       unsigned long start = input->group * sbi->s_blocks_per_group +
-               le32_to_cpu(sbi->s_es->s_first_data_block);
+       ext3_fsblk_t start = ext3_group_first_block_no(sb, input->group);
        int reserved_gdb = ext3_bg_has_super(sb, input->group) ?
                le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) : 0;
        unsigned long gdblocks = ext3_bg_num_gdb(sb, input->group);
        struct buffer_head *bh;
        handle_t *handle;
-       unsigned long block;
-       int bit;
+       ext3_fsblk_t block;
+       ext3_grpblk_t bit;
        int i;
        int err = 0, err2;
 
-       handle = ext3_journal_start(inode, reserved_gdb + gdblocks +
-                                   2 + sbi->s_itb_per_group);
+       handle = ext3_journal_start_sb(sb, reserved_gdb + gdblocks +
+                                      2 + sbi->s_itb_per_group);
        if (IS_ERR(handle))
                return PTR_ERR(handle);
 
@@ -202,12 +206,18 @@ static int setup_new_group_blocks(struct super_block *sb, struct inode *inode,
                ext3_debug("update backup group %#04lx (+%d)\n", block, bit);
 
                gdb = sb_getblk(sb, block);
-               set_buffer_uptodate(gdb);
+               if (!gdb) {
+                       err = -EIO;
+                       goto exit_bh;
+               }
                if ((err = ext3_journal_get_write_access(handle, gdb))) {
                        brelse(gdb);
                        goto exit_bh;
                }
-               memcpy(gdb->b_data, sbi->s_group_desc[i], bh->b_size);
+               lock_buffer(bh);
+               memcpy(gdb->b_data, sbi->s_group_desc[i]->b_data, bh->b_size);
+               set_buffer_uptodate(gdb);
+               unlock_buffer(bh);
                ext3_journal_dirty_metadata(handle, gdb);
                ext3_set_bit(bit, bh->b_data);
                brelse(gdb);
@@ -240,7 +250,7 @@ static int setup_new_group_blocks(struct super_block *sb, struct inode *inode,
             i < sbi->s_itb_per_group; i++, bit++, block++) {
                struct buffer_head *it;
 
-               ext3_debug("clear inode block %#04x (+%ld)\n", block, bit);
+               ext3_debug("clear inode block %#04lx (+%d)\n", block, bit);
                if (IS_ERR(it = bclean(handle, sb, block))) {
                        err = PTR_ERR(it);
                        goto exit_bh;
@@ -283,8 +293,8 @@ exit_journal:
  * sequence of powers of 3, 5, and 7: 1, 3, 5, 7, 9, 25, 27, 49, 81, ...
  * For a non-sparse filesystem it will be every group: 1, 2, 3, 4, ...
  */
-unsigned ext3_list_backups(struct super_block *sb, unsigned *three,
-                          unsigned *five, unsigned *seven)
+static unsigned ext3_list_backups(struct super_block *sb, unsigned *three,
+                                 unsigned *five, unsigned *seven)
 {
        unsigned *min = three;
        int mult = 3;
@@ -320,19 +330,20 @@ unsigned ext3_list_backups(struct super_block *sb, unsigned *three,
 static int verify_reserved_gdb(struct super_block *sb,
                               struct buffer_head *primary)
 {
-       const unsigned long blk = primary->b_blocknr;
+       const ext3_fsblk_t blk = primary->b_blocknr;
        const unsigned long end = EXT3_SB(sb)->s_groups_count;
        unsigned three = 1;
        unsigned five = 5;
        unsigned seven = 7;
        unsigned grp;
-       __u32 *p = (__u32 *)primary->b_data;
+       __le32 *p = (__le32 *)primary->b_data;
        int gdbackups = 0;
 
        while ((grp = ext3_list_backups(sb, &three, &five, &seven)) < end) {
                if (le32_to_cpu(*p++) != grp * EXT3_BLOCKS_PER_GROUP(sb) + blk){
                        ext3_warning(sb, __FUNCTION__,
-                                    "reserved GDT %ld missing grp %d (%ld)\n",
+                                    "reserved GDT "E3FSBLK
+                                    " missing grp %d ("E3FSBLK")",
                                     blk, grp,
                                     grp * EXT3_BLOCKS_PER_GROUP(sb) + blk);
                        return -EINVAL;
@@ -364,17 +375,17 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
        struct super_block *sb = inode->i_sb;
        struct ext3_super_block *es = EXT3_SB(sb)->s_es;
        unsigned long gdb_num = input->group / EXT3_DESC_PER_BLOCK(sb);
-       unsigned long gdb_off = input->group % EXT3_DESC_PER_BLOCK(sb);
-       unsigned long gdblock = EXT3_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
+       ext3_fsblk_t gdblock = EXT3_SB(sb)->s_sbh->b_blocknr + 1 + gdb_num;
        struct buffer_head **o_group_desc, **n_group_desc;
        struct buffer_head *dind;
        int gdbackups;
        struct ext3_iloc iloc;
-       __u32 *data;
+       __le32 *data;
        int err;
 
        if (test_opt(sb, DEBUG))
-               printk("EXT3-fs: ext3_add_new_gdb: adding group block %lu\n",
+               printk(KERN_DEBUG
+                      "EXT3-fs: ext3_add_new_gdb: adding group block %lu\n",
                       gdb_num);
 
        /*
@@ -385,8 +396,8 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
        if (EXT3_SB(sb)->s_sbh->b_blocknr !=
            le32_to_cpu(EXT3_SB(sb)->s_es->s_first_data_block)) {
                ext3_warning(sb, __FUNCTION__,
-                            "won't resize using backup superblock at %lu\n",
-                            EXT3_SB(sb)->s_sbh->b_blocknr);
+                       "won't resize using backup superblock at %llu",
+                       (unsigned long long)EXT3_SB(sb)->s_sbh->b_blocknr);
                return -EPERM;
        }
 
@@ -406,10 +417,10 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
                goto exit_bh;
        }
 
-       data = (__u32 *)dind->b_data;
+       data = (__le32 *)dind->b_data;
        if (le32_to_cpu(data[gdb_num % EXT3_ADDR_PER_BLOCK(sb)]) != gdblock) {
                ext3_warning(sb, __FUNCTION__,
-                            "new group %u GDT block %lu not reserved\n",
+                            "new group %u GDT block "E3FSBLK" not reserved",
                             input->group, gdblock);
                err = -EINVAL;
                goto exit_dind;
@@ -428,8 +439,8 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
        if ((err = ext3_reserve_inode_write(handle, inode, &iloc)))
                goto exit_dindj;
 
-       n_group_desc = (struct buffer_head **)kmalloc((gdb_num + 1) *
-                               sizeof(struct buffer_head *), GFP_KERNEL);
+       n_group_desc = kmalloc((gdb_num + 1) * sizeof(struct buffer_head *),
+                       GFP_KERNEL);
        if (!n_group_desc) {
                err = -ENOMEM;
                ext3_warning (sb, __FUNCTION__,
@@ -446,10 +457,6 @@ static int add_new_gdb(handle_t *handle, struct inode *inode,
         * these blocks, because they are marked as in-use from being in the
         * reserved inode, and will become GDT blocks (primary and backup).
         */
-       /*
-       printk("removing block %d = %ld from dindir %ld[%ld]\n",
-              ((__u32 *)(dind->b_data))[gdb_off], gdblock, dind->b_blocknr,
-              gdb_num); */
        data[gdb_num % EXT3_ADDR_PER_BLOCK(sb)] = 0;
        ext3_journal_dirty_metadata(handle, dind);
        brelse(dind);
@@ -511,8 +518,8 @@ static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
        struct buffer_head **primary;
        struct buffer_head *dind;
        struct ext3_iloc iloc;
-       unsigned long blk;
-       __u32 *data, *end;
+       ext3_fsblk_t blk;
+       __le32 *data, *end;
        int gdbackups = 0;
        int res, i;
        int err;
@@ -529,15 +536,17 @@ static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
        }
 
        blk = EXT3_SB(sb)->s_sbh->b_blocknr + 1 + EXT3_SB(sb)->s_gdb_count;
-       data = (__u32 *)dind->b_data + EXT3_SB(sb)->s_gdb_count;
-       end = (__u32 *)dind->b_data + EXT3_ADDR_PER_BLOCK(sb);
+       data = (__le32 *)dind->b_data + EXT3_SB(sb)->s_gdb_count;
+       end = (__le32 *)dind->b_data + EXT3_ADDR_PER_BLOCK(sb);
 
        /* Get each reserved primary GDT block and verify it holds backups */
        for (res = 0; res < reserved_gdb; res++, blk++) {
                if (le32_to_cpu(*data) != blk) {
                        ext3_warning(sb, __FUNCTION__,
-                                    "reserved block %lu not at offset %ld\n",
-                                    blk, (long)(data - (__u32 *)dind->b_data));
+                                    "reserved block "E3FSBLK
+                                    " not at offset %ld",
+                                    blk,
+                                    (long)(data - (__le32 *)dind->b_data));
                        err = -EINVAL;
                        goto exit_bh;
                }
@@ -552,7 +561,7 @@ static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
                        goto exit_bh;
                }
                if (++data >= end)
-                       data = (__u32 *)dind->b_data;
+                       data = (__le32 *)dind->b_data;
        }
 
        for (i = 0; i < reserved_gdb; i++) {
@@ -576,7 +585,7 @@ static int reserve_backup_gdb(handle_t *handle, struct inode *inode,
        blk = input->group * EXT3_BLOCKS_PER_GROUP(sb);
        for (i = 0; i < reserved_gdb; i++) {
                int err2;
-               data = (__u32 *)primary[i]->b_data;
+               data = (__le32 *)primary[i]->b_data;
                /* printk("reserving backup %lu[%u] = %lu\n",
                       primary[i]->b_blocknr, gdbackups,
                       blk + primary[i]->b_blocknr); */
@@ -613,10 +622,8 @@ exit_free:
  * 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 only pass inode because of the ext3 journal wrappers.
  */
-static void update_backups(struct super_block *sb, struct inode *inode,
+static void update_backups(struct super_block *sb,
                           int blk_off, char *data, int size)
 {
        struct ext3_sb_info *sbi = EXT3_SB(sb);
@@ -630,7 +637,7 @@ static void update_backups(struct super_block *sb, struct inode *inode,
        handle_t *handle;
        int err = 0, err2;
 
-       handle = ext3_journal_start(inode, EXT3_MAX_TRANS_DATA);
+       handle = ext3_journal_start_sb(sb, EXT3_MAX_TRANS_DATA);
        if (IS_ERR(handle)) {
                group = 1;
                err = PTR_ERR(handle);
@@ -647,14 +654,20 @@ static void update_backups(struct super_block *sb, struct inode *inode,
                        break;
 
                bh = sb_getblk(sb, group * bpg + blk_off);
-               set_buffer_uptodate(bh);
-               ext3_debug(sb, __FUNCTION__, "update metadata backup %#04lx\n",
-                          bh->b_blocknr);
+               if (!bh) {
+                       err = -EIO;
+                       break;
+               }
+               ext3_debug("update metadata backup %#04lx\n",
+                         (unsigned long)bh->b_blocknr);
                if ((err = ext3_journal_get_write_access(handle, bh)))
                        break;
+               lock_buffer(bh);
                memcpy(bh->b_data, data, size);
                if (rest)
                        memset(bh->b_data + size, 0, rest);
+               set_buffer_uptodate(bh);
+               unlock_buffer(bh);
                ext3_journal_dirty_metadata(handle, bh);
                brelse(bh);
        }
@@ -675,9 +688,9 @@ exit_err:
        if (err) {
                ext3_warning(sb, __FUNCTION__,
                             "can't update backup for group %d (err %d), "
-                            "forcing fsck on next reboot\n", group, err);
+                            "forcing fsck on next reboot", group, err);
                sbi->s_mount_state &= ~EXT3_VALID_FS;
-               sbi->s_es->s_state &= ~cpu_to_le16(EXT3_VALID_FS);
+               sbi->s_es->s_state &= cpu_to_le16(~EXT3_VALID_FS);
                mark_buffer_dirty(sbi->s_sbh);
        }
 }
@@ -704,7 +717,6 @@ int ext3_group_add(struct super_block *sb, struct ext3_new_group_data *input)
        struct buffer_head *primary = NULL;
        struct ext3_group_desc *gdp;
        struct inode *inode = NULL;
-       struct inode bogus;
        handle_t *handle;
        int gdb_off, gdb_num;
        int err, err2;
@@ -715,34 +727,42 @@ int ext3_group_add(struct super_block *sb, struct ext3_new_group_data *input)
        if (gdb_off == 0 && !EXT3_HAS_RO_COMPAT_FEATURE(sb,
                                        EXT3_FEATURE_RO_COMPAT_SPARSE_SUPER)) {
                ext3_warning(sb, __FUNCTION__,
-                            "Can't resize non-sparse filesystem further\n");
+                            "Can't resize non-sparse filesystem further");
                return -EPERM;
        }
 
+       if (le32_to_cpu(es->s_blocks_count) + input->blocks_count <
+           le32_to_cpu(es->s_blocks_count)) {
+               ext3_warning(sb, __FUNCTION__, "blocks_count overflow\n");
+               return -EINVAL;
+       }
+
+       if (le32_to_cpu(es->s_inodes_count) + EXT3_INODES_PER_GROUP(sb) <
+           le32_to_cpu(es->s_inodes_count)) {
+               ext3_warning(sb, __FUNCTION__, "inodes_count overflow\n");
+               return -EINVAL;
+       }
+
        if (reserved_gdb || gdb_off == 0) {
                if (!EXT3_HAS_COMPAT_FEATURE(sb,
                                             EXT3_FEATURE_COMPAT_RESIZE_INODE)){
                        ext3_warning(sb, __FUNCTION__,
-                                    "No reserved GDT blocks, can't resize\n");
+                                    "No reserved GDT blocks, can't resize");
                        return -EPERM;
                }
                inode = iget(sb, EXT3_RESIZE_INO);
                if (!inode || is_bad_inode(inode)) {
                        ext3_warning(sb, __FUNCTION__,
-                                    "Error opening resize inode\n");
+                                    "Error opening resize inode");
                        iput(inode);
                        return -ENOENT;
                }
-       } else {
-               /* Used only for ext3 journal wrapper functions to get sb */
-               inode = &bogus;
-               bogus.i_sb = sb;
        }
 
        if ((err = verify_group_input(sb, input)))
                goto exit_put;
 
-       if ((err = setup_new_group_blocks(sb, inode, input)))
+       if ((err = setup_new_group_blocks(sb, input)))
                goto exit_put;
 
        /*
@@ -752,17 +772,19 @@ int ext3_group_add(struct super_block *sb, struct ext3_new_group_data *input)
         * are adding a group with superblock/GDT backups  we will also
         * modify each of the reserved GDT dindirect blocks.
         */
-       handle = ext3_journal_start(inode, ext3_bg_has_super(sb, input->group) ?
-                                   3 + reserved_gdb : 4);
+       handle = ext3_journal_start_sb(sb,
+                                      ext3_bg_has_super(sb, input->group) ?
+                                      3 + reserved_gdb : 4);
        if (IS_ERR(handle)) {
                err = PTR_ERR(handle);
                goto exit_put;
        }
 
        lock_super(sb);
-       if (input->group != EXT3_SB(sb)->s_groups_count) {
+       if (input->group != sbi->s_groups_count) {
                ext3_warning(sb, __FUNCTION__,
-                            "multiple resizers run on filesystem!\n");
+                            "multiple resizers run on filesystem!");
+               err = -EBUSY;
                goto exit_journal;
        }
 
@@ -786,7 +808,26 @@ int ext3_group_add(struct super_block *sb, struct ext3_new_group_data *input)
        } else if ((err = add_new_gdb(handle, inode, input, &primary)))
                goto exit_journal;
 
-       /* Finally update group descriptor block for new group */
+       /*
+        * OK, now we've set up the new group.  Time to make it active.
+        *
+        * Current kernels don't lock all allocations via lock_super(),
+        * 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.
+        *
+        * The key field here is sbi->s_groups_count: as long as
+        * that retains its old value, nobody is going to access the new
+        * group.
+        *
+        * So first we update all the descriptor metadata for the new
+        * group; then we update the total disk blocks count; then we
+        * update the groups count to enable the group; then finally we
+        * update the free space counts so that the system can start
+        * using the new disk blocks.
+        */
+
+       /* Update group descriptor block for new group */
        gdp = (struct ext3_group_desc *)primary->b_data + gdb_off;
 
        gdp->bg_block_bitmap = cpu_to_le32(input->block_bitmap);
@@ -795,39 +836,76 @@ int ext3_group_add(struct super_block *sb, struct ext3_new_group_data *input)
        gdp->bg_free_blocks_count = cpu_to_le16(input->free_blocks_count);
        gdp->bg_free_inodes_count = cpu_to_le16(EXT3_INODES_PER_GROUP(sb));
 
-       EXT3_SB(sb)->s_groups_count++;
-       ext3_journal_dirty_metadata(handle, primary);
-
-       /* Update superblock with new block counts */
+       /*
+        * Make the new blocks and inodes valid next.  We do this before
+        * increasing the group count so that once the group is enabled,
+        * all of its blocks and inodes are already valid.
+        *
+        * We always allocate group-by-group, then block-by-block or
+        * inode-by-inode within a group, so enabling these
+        * blocks/inodes before the group is live won't actually let us
+        * allocate the new space yet.
+        */
        es->s_blocks_count = cpu_to_le32(le32_to_cpu(es->s_blocks_count) +
                input->blocks_count);
-       es->s_free_blocks_count =
-               cpu_to_le32(le32_to_cpu(es->s_free_blocks_count) +
-                           input->free_blocks_count);
-       es->s_r_blocks_count = cpu_to_le32(le32_to_cpu(es->s_r_blocks_count) +
-               input->reserved_blocks);
        es->s_inodes_count = cpu_to_le32(le32_to_cpu(es->s_inodes_count) +
                EXT3_INODES_PER_GROUP(sb));
-       es->s_free_inodes_count =
-               cpu_to_le32(le32_to_cpu(es->s_free_inodes_count) +
-                           EXT3_INODES_PER_GROUP(sb));
-       ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
+
+       /*
+        * We need to protect s_groups_count against other CPUs seeing
+        * inconsistent state in the superblock.
+        *
+        * The precise rules we use are:
+        *
+        * * Writers of s_groups_count *must* hold lock_super
+        * 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
+        * OR
+        * * Readers must perform an smp_rmb() after reading the groups count
+        *   and before reading any dependent data.
+        *
+        * NB. These rules can be relaxed when checking the group count
+        * while freeing data, as we can only allocate from a block
+        * group after serialising against the group count, and we can
+        * only then free after serialising in turn against that
+        * allocation.
+        */
+       smp_wmb();
+
+       /* Update the global fs size fields */
+       sbi->s_groups_count++;
+
+       ext3_journal_dirty_metadata(handle, primary);
+
+       /* Update the reserved block counts only once the new group is
+        * active. */
+       es->s_r_blocks_count = cpu_to_le32(le32_to_cpu(es->s_r_blocks_count) +
+               input->reserved_blocks);
+
+       /* Update the free space counts */
+       percpu_counter_mod(&sbi->s_freeblocks_counter,
+                          input->free_blocks_count);
+       percpu_counter_mod(&sbi->s_freeinodes_counter,
+                          EXT3_INODES_PER_GROUP(sb));
+
+       ext3_journal_dirty_metadata(handle, sbi->s_sbh);
        sb->s_dirt = 1;
 
 exit_journal:
        unlock_super(sb);
-       handle->h_sync = 1;
        if ((err2 = ext3_journal_stop(handle)) && !err)
                err = err2;
        if (!err) {
-               update_backups(sb, inode, sbi->s_sbh->b_blocknr, (char *)es,
+               update_backups(sb, sbi->s_sbh->b_blocknr, (char *)es,
                               sizeof(struct ext3_super_block));
-               update_backups(sb, inode, primary->b_blocknr, primary->b_data,
+               update_backups(sb, primary->b_blocknr, primary->b_data,
                               primary->b_size);
        }
 exit_put:
-       if (inode != &bogus)
-               iput(inode);
+       iput(inode);
        return err;
 } /* ext3_group_add */
 
@@ -841,27 +919,40 @@ exit_put:
  * GDT blocks are reserved to grow to the desired size.
  */
 int ext3_group_extend(struct super_block *sb, struct ext3_super_block *es,
-                     unsigned long n_blocks_count)
+                     ext3_fsblk_t n_blocks_count)
 {
-       unsigned long o_blocks_count;
+       ext3_fsblk_t o_blocks_count;
        unsigned long o_groups_count;
-       unsigned long last;
-       int add;
-       struct inode *inode;
+       ext3_grpblk_t last;
+       ext3_grpblk_t add;
        struct buffer_head * bh;
        handle_t *handle;
        int err;
+       unsigned long freed_blocks;
 
+       /* 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. */
        o_blocks_count = le32_to_cpu(es->s_blocks_count);
        o_groups_count = EXT3_SB(sb)->s_groups_count;
 
        if (test_opt(sb, DEBUG))
-               printk("EXT3-fs: extending last group from %lu to %lu blocks\n",
+               printk(KERN_DEBUG "EXT3-fs: extending last group from "E3FSBLK" uto "E3FSBLK" blocks\n",
                       o_blocks_count, n_blocks_count);
 
        if (n_blocks_count == 0 || n_blocks_count == o_blocks_count)
                return 0;
 
+       if (n_blocks_count > (sector_t)(~0ULL) >> (sb->s_blocksize_bits - 9)) {
+               printk(KERN_ERR "EXT3-fs: filesystem on %s:"
+                       " too large to resize to %lu blocks safely\n",
+                       sb->s_id, n_blocks_count);
+               if (sizeof(sector_t) < 8)
+                       ext3_warning(sb, __FUNCTION__,
+                       "CONFIG_LBD not enabled\n");
+               return -EINVAL;
+       }
+
        if (n_blocks_count < o_blocks_count) {
                ext3_warning(sb, __FUNCTION__,
                             "can't shrink FS - resize aborted");
@@ -874,18 +965,24 @@ int ext3_group_extend(struct super_block *sb, struct ext3_super_block *es,
 
        if (last == 0) {
                ext3_warning(sb, __FUNCTION__,
-                            "need to use ext2online to resize further\n");
+                            "need to use ext2online to resize further");
                return -EPERM;
        }
 
        add = EXT3_BLOCKS_PER_GROUP(sb) - last;
 
+       if (o_blocks_count + add < o_blocks_count) {
+               ext3_warning(sb, __FUNCTION__, "blocks_count overflow");
+               return -EINVAL;
+       }
+
        if (o_blocks_count + add > n_blocks_count)
                add = n_blocks_count - o_blocks_count;
 
        if (o_blocks_count + add < n_blocks_count)
                ext3_warning(sb, __FUNCTION__,
-                            "will only finish group (%lu blocks, %u new)",
+                            "will only finish group ("E3FSBLK
+                            " blocks, %u new)",
                             o_blocks_count + add, add);
 
        /* See if the device is actually as big as what was requested */
@@ -897,20 +994,10 @@ int ext3_group_extend(struct super_block *sb, struct ext3_super_block *es,
        }
        brelse(bh);
 
-       /* Get a bogus inode to "free" the new blocks in this group. */
-       if (!(inode = new_inode(sb))) {
-               ext3_warning(sb, __FUNCTION__,
-                            "error getting dummy resize inode");
-               return -ENOMEM;
-       }
-       inode->i_ino = 0;
-
-       EXT3_I(inode)->i_state = EXT3_STATE_RESIZE;
-
        /* We will update the superblock, one block bitmap, and
         * one group descriptor via ext3_free_blocks().
         */
-       handle = ext3_journal_start(inode, 3);
+       handle = ext3_journal_start_sb(sb, 3);
        if (IS_ERR(handle)) {
                err = PTR_ERR(handle);
                ext3_warning(sb, __FUNCTION__, "error %d on journal start",err);
@@ -920,7 +1007,8 @@ int ext3_group_extend(struct super_block *sb, struct ext3_super_block *es,
        lock_super(sb);
        if (o_blocks_count != le32_to_cpu(es->s_blocks_count)) {
                ext3_warning(sb, __FUNCTION__,
-                            "multiple resizers run on filesystem!\n");
+                            "multiple resizers run on filesystem!");
+               unlock_super(sb);
                err = -EBUSY;
                goto exit_put;
        }
@@ -937,20 +1025,18 @@ int ext3_group_extend(struct super_block *sb, struct ext3_super_block *es,
        ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
        sb->s_dirt = 1;
        unlock_super(sb);
-       ext3_debug("freeing blocks %ld through %ld\n", o_blocks_count,
+       ext3_debug("freeing blocks %lu through "E3FSBLK"\n", o_blocks_count,
                   o_blocks_count + add);
-       ext3_free_blocks(handle, inode, o_blocks_count, add);
-       ext3_debug("freed blocks %ld through %ld\n", o_blocks_count,
+       ext3_free_blocks_sb(handle, sb, o_blocks_count, add, &freed_blocks);
+       ext3_debug("freed blocks "E3FSBLK" through "E3FSBLK"\n", o_blocks_count,
                   o_blocks_count + add);
        if ((err = ext3_journal_stop(handle)))
                goto exit_put;
        if (test_opt(sb, DEBUG))
-               printk("EXT3-fs: extended group to %u blocks\n",
+               printk(KERN_DEBUG "EXT3-fs: extended group to %u blocks\n",
                       le32_to_cpu(es->s_blocks_count));
-       update_backups(sb, inode, EXT3_SB(sb)->s_sbh->b_blocknr, (char *)es,
+       update_backups(sb, EXT3_SB(sb)->s_sbh->b_blocknr, (char *)es,
                       sizeof(struct ext3_super_block));
 exit_put:
-       iput(inode);
-
        return err;
 } /* ext3_group_extend */