]> pilppa.org Git - linux-2.6-omap-h63xx.git/blobdiff - fs/ext4/ialloc.c
Ext4: Uninitialized Block Groups
[linux-2.6-omap-h63xx.git] / fs / ext4 / ialloc.c
index 427f83066a0da425b75757c68430768bba7428ea..c61f37fd3f05e4d72061791d79600024c858df97 100644 (file)
@@ -28,6 +28,7 @@
 
 #include "xattr.h"
 #include "acl.h"
+#include "group.h"
 
 /*
  * ialloc.c contains the inodes allocation and deallocation routines
  * the free blocks count in the block.
  */
 
+/*
+ * To avoid calling the atomic setbit hundreds or thousands of times, we only
+ * need to use it within a single byte (to ensure we get endianness right).
+ * We can use memset for the rest of the bitmap as there are no other users.
+ */
+void mark_bitmap_end(int start_bit, int end_bit, char *bitmap)
+{
+       int i;
+
+       if (start_bit >= end_bit)
+               return;
+
+       ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit);
+       for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++)
+               ext4_set_bit(i, bitmap);
+       if (i < end_bit)
+               memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3);
+}
+
+/* Initializes an uninitialized inode bitmap */
+unsigned ext4_init_inode_bitmap(struct super_block *sb,
+                               struct buffer_head *bh, int block_group,
+                               struct ext4_group_desc *gdp)
+{
+       struct ext4_sb_info *sbi = EXT4_SB(sb);
+
+       J_ASSERT_BH(bh, buffer_locked(bh));
+
+       /* If checksum is bad mark all blocks and inodes use to prevent
+        * allocation, essentially implementing a per-group read-only flag. */
+       if (!ext4_group_desc_csum_verify(sbi, block_group, gdp)) {
+               ext4_error(sb, __FUNCTION__, "Checksum bad for group %u\n",
+                          block_group);
+               gdp->bg_free_blocks_count = 0;
+               gdp->bg_free_inodes_count = 0;
+               gdp->bg_itable_unused = 0;
+               memset(bh->b_data, 0xff, sb->s_blocksize);
+               return 0;
+       }
+
+       memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8);
+       mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), EXT4_BLOCKS_PER_GROUP(sb),
+                       bh->b_data);
+
+       return EXT4_INODES_PER_GROUP(sb);
+}
 
 /*
  * Read the inode allocation bitmap for a given block_group, reading
@@ -59,8 +106,20 @@ read_inode_bitmap(struct super_block * sb, unsigned long block_group)
        desc = ext4_get_group_desc(sb, block_group, NULL);
        if (!desc)
                goto error_out;
-
-       bh = sb_bread(sb, ext4_inode_bitmap(sb, desc));
+       if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
+               bh = sb_getblk(sb, ext4_inode_bitmap(sb, desc));
+               if (!buffer_uptodate(bh)) {
+                       lock_buffer(bh);
+                       if (!buffer_uptodate(bh)) {
+                               ext4_init_inode_bitmap(sb, bh, block_group,
+                                                      desc);
+                               set_buffer_uptodate(bh);
+                       }
+                       unlock_buffer(bh);
+               }
+       } else {
+               bh = sb_bread(sb, ext4_inode_bitmap(sb, desc));
+       }
        if (!bh)
                ext4_error(sb, "read_inode_bitmap",
                            "Cannot read inode bitmap - "
@@ -169,6 +228,8 @@ void ext4_free_inode (handle_t *handle, struct inode * inode)
                        if (is_directory)
                                gdp->bg_used_dirs_count = cpu_to_le16(
                                  le16_to_cpu(gdp->bg_used_dirs_count) - 1);
+                       gdp->bg_checksum = ext4_group_desc_csum(sbi,
+                                                       block_group, gdp);
                        spin_unlock(sb_bgl_lock(sbi, block_group));
                        percpu_counter_inc(&sbi->s_freeinodes_counter);
                        if (is_directory)
@@ -204,14 +265,13 @@ static int find_group_dir(struct super_block *sb, struct inode *parent)
        int ngroups = EXT4_SB(sb)->s_groups_count;
        unsigned int freei, avefreei;
        struct ext4_group_desc *desc, *best_desc = NULL;
-       struct buffer_head *bh;
        int group, best_group = -1;
 
        freei = percpu_counter_read_positive(&EXT4_SB(sb)->s_freeinodes_counter);
        avefreei = freei / ngroups;
 
        for (group = 0; group < ngroups; group++) {
-               desc = ext4_get_group_desc (sb, group, &bh);
+               desc = ext4_get_group_desc (sb, group, NULL);
                if (!desc || !desc->bg_free_inodes_count)
                        continue;
                if (le16_to_cpu(desc->bg_free_inodes_count) < avefreei)
@@ -269,7 +329,6 @@ static int find_group_orlov(struct super_block *sb, struct inode *parent)
        ext4_grpblk_t min_blocks;
        int group = -1, i;
        struct ext4_group_desc *desc;
-       struct buffer_head *bh;
 
        freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter);
        avefreei = freei / ngroups;
@@ -287,7 +346,7 @@ static int find_group_orlov(struct super_block *sb, struct inode *parent)
                parent_group = (unsigned)group % ngroups;
                for (i = 0; i < ngroups; i++) {
                        group = (parent_group + i) % ngroups;
-                       desc = ext4_get_group_desc (sb, group, &bh);
+                       desc = ext4_get_group_desc (sb, group, NULL);
                        if (!desc || !desc->bg_free_inodes_count)
                                continue;
                        if (le16_to_cpu(desc->bg_used_dirs_count) >= best_ndir)
@@ -322,7 +381,7 @@ static int find_group_orlov(struct super_block *sb, struct inode *parent)
 
        for (i = 0; i < ngroups; i++) {
                group = (parent_group + i) % ngroups;
-               desc = ext4_get_group_desc (sb, group, &bh);
+               desc = ext4_get_group_desc (sb, group, NULL);
                if (!desc || !desc->bg_free_inodes_count)
                        continue;
                if (le16_to_cpu(desc->bg_used_dirs_count) >= max_dirs)
@@ -337,7 +396,7 @@ static int find_group_orlov(struct super_block *sb, struct inode *parent)
 fallback:
        for (i = 0; i < ngroups; i++) {
                group = (parent_group + i) % ngroups;
-               desc = ext4_get_group_desc (sb, group, &bh);
+               desc = ext4_get_group_desc (sb, group, NULL);
                if (!desc || !desc->bg_free_inodes_count)
                        continue;
                if (le16_to_cpu(desc->bg_free_inodes_count) >= avefreei)
@@ -361,14 +420,13 @@ static int find_group_other(struct super_block *sb, struct inode *parent)
        int parent_group = EXT4_I(parent)->i_block_group;
        int ngroups = EXT4_SB(sb)->s_groups_count;
        struct ext4_group_desc *desc;
-       struct buffer_head *bh;
        int group, i;
 
        /*
         * Try to place the inode in its parent directory
         */
        group = parent_group;
-       desc = ext4_get_group_desc (sb, group, &bh);
+       desc = ext4_get_group_desc (sb, group, NULL);
        if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
                        le16_to_cpu(desc->bg_free_blocks_count))
                return group;
@@ -392,7 +450,7 @@ static int find_group_other(struct super_block *sb, struct inode *parent)
                group += i;
                if (group >= ngroups)
                        group -= ngroups;
-               desc = ext4_get_group_desc (sb, group, &bh);
+               desc = ext4_get_group_desc (sb, group, NULL);
                if (desc && le16_to_cpu(desc->bg_free_inodes_count) &&
                                le16_to_cpu(desc->bg_free_blocks_count))
                        return group;
@@ -406,7 +464,7 @@ static int find_group_other(struct super_block *sb, struct inode *parent)
        for (i = 0; i < ngroups; i++) {
                if (++group >= ngroups)
                        group = 0;
-               desc = ext4_get_group_desc (sb, group, &bh);
+               desc = ext4_get_group_desc (sb, group, NULL);
                if (desc && le16_to_cpu(desc->bg_free_inodes_count))
                        return group;
        }
@@ -438,7 +496,7 @@ struct inode *ext4_new_inode(handle_t *handle, struct inode * dir, int mode)
        struct ext4_sb_info *sbi;
        int err = 0;
        struct inode *ret;
-       int i;
+       int i, free = 0;
 
        /* Cannot create files in a deleted directory */
        if (!dir || !dir->i_nlink)
@@ -520,11 +578,13 @@ repeat_in_this_group:
        goto out;
 
 got:
-       ino += group * EXT4_INODES_PER_GROUP(sb) + 1;
-       if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) {
-               ext4_error (sb, "ext4_new_inode",
-                           "reserved inode or inode > inodes count - "
-                           "block_group = %d, inode=%lu", group, ino);
+       ino++;
+       if ((group == 0 && ino < EXT4_FIRST_INO(sb)) ||
+           ino > EXT4_INODES_PER_GROUP(sb)) {
+               ext4_error(sb, __FUNCTION__,
+                          "reserved inode or inode > inodes count - "
+                          "block_group = %d, inode=%lu", group,
+                          ino + group * EXT4_INODES_PER_GROUP(sb));
                err = -EIO;
                goto fail;
        }
@@ -532,13 +592,78 @@ got:
        BUFFER_TRACE(bh2, "get_write_access");
        err = ext4_journal_get_write_access(handle, bh2);
        if (err) goto fail;
+
+       /* We may have to initialize the block bitmap if it isn't already */
+       if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM) &&
+           gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
+               struct buffer_head *block_bh = read_block_bitmap(sb, group);
+
+               BUFFER_TRACE(block_bh, "get block bitmap access");
+               err = ext4_journal_get_write_access(handle, block_bh);
+               if (err) {
+                       brelse(block_bh);
+                       goto fail;
+               }
+
+               free = 0;
+               spin_lock(sb_bgl_lock(sbi, group));
+               /* recheck and clear flag under lock if we still need to */
+               if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) {
+                       gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT);
+                       free = ext4_free_blocks_after_init(sb, group, gdp);
+                       gdp->bg_free_blocks_count = cpu_to_le16(free);
+               }
+               spin_unlock(sb_bgl_lock(sbi, group));
+
+               /* Don't need to dirty bitmap block if we didn't change it */
+               if (free) {
+                       BUFFER_TRACE(block_bh, "dirty block bitmap");
+                       err = ext4_journal_dirty_metadata(handle, block_bh);
+               }
+
+               brelse(block_bh);
+               if (err)
+                       goto fail;
+       }
+
        spin_lock(sb_bgl_lock(sbi, group));
+       /* If we didn't allocate from within the initialized part of the inode
+        * table then we need to initialize up to this inode. */
+       if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
+               if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
+                       gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
+
+                       /* When marking the block group with
+                        * ~EXT4_BG_INODE_UNINIT we don't want to depend
+                        * on the value of bg_itable_unsed even though
+                        * mke2fs could have initialized the same for us.
+                        * Instead we calculated the value below
+                        */
+
+                       free = 0;
+               } else {
+                       free = EXT4_INODES_PER_GROUP(sb) -
+                               le16_to_cpu(gdp->bg_itable_unused);
+               }
+
+               /*
+                * Check the relative inode number against the last used
+                * relative inode number in this group. if it is greater
+                * we need to  update the bg_itable_unused count
+                *
+                */
+               if (ino > free)
+                       gdp->bg_itable_unused =
+                               cpu_to_le16(EXT4_INODES_PER_GROUP(sb) - ino);
+       }
+
        gdp->bg_free_inodes_count =
                cpu_to_le16(le16_to_cpu(gdp->bg_free_inodes_count) - 1);
        if (S_ISDIR(mode)) {
                gdp->bg_used_dirs_count =
                        cpu_to_le16(le16_to_cpu(gdp->bg_used_dirs_count) + 1);
        }
+       gdp->bg_checksum = ext4_group_desc_csum(sbi, group, gdp);
        spin_unlock(sb_bgl_lock(sbi, group));
        BUFFER_TRACE(bh2, "call ext4_journal_dirty_metadata");
        err = ext4_journal_dirty_metadata(handle, bh2);
@@ -560,7 +685,7 @@ got:
                inode->i_gid = current->fsgid;
        inode->i_mode = mode;
 
-       inode->i_ino = ino;
+       inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb);
        /* This is the optimal IO size (for stat), not the fs block size */
        inode->i_blocks = 0;
        inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime =
@@ -576,11 +701,6 @@ got:
        /* dirsync only applies to directories */
        if (!S_ISDIR(mode))
                ei->i_flags &= ~EXT4_DIRSYNC_FL;
-#ifdef EXT4_FRAGMENTS
-       ei->i_faddr = 0;
-       ei->i_frag_no = 0;
-       ei->i_frag_size = 0;
-#endif
        ei->i_file_acl = 0;
        ei->i_dir_acl = 0;
        ei->i_dtime = 0;