Merge branch 'for-2.6.35' of git://git.kernel.org/pub/scm/linux/kernel/git/broonie...
[safe/jmp/linux-2.6] / fs / ocfs2 / resize.c
index 848f729..3c3d673 100644 (file)
@@ -106,8 +106,8 @@ static int ocfs2_update_last_group_and_inode(handle_t *handle,
        mlog_entry("(new_clusters=%d, first_new_cluster = %u)\n",
                   new_clusters, first_new_cluster);
 
-       ret = ocfs2_journal_access(handle, bm_inode, group_bh,
-                                  OCFS2_JOURNAL_ACCESS_WRITE);
+       ret = ocfs2_journal_access_gd(handle, INODE_CACHE(bm_inode),
+                                     group_bh, OCFS2_JOURNAL_ACCESS_WRITE);
        if (ret < 0) {
                mlog_errno(ret);
                goto out;
@@ -141,8 +141,8 @@ static int ocfs2_update_last_group_and_inode(handle_t *handle,
        }
 
        /* update the inode accordingly. */
-       ret = ocfs2_journal_access(handle, bm_inode, bm_bh,
-                                  OCFS2_JOURNAL_ACCESS_WRITE);
+       ret = ocfs2_journal_access_di(handle, INODE_CACHE(bm_inode), bm_bh,
+                                     OCFS2_JOURNAL_ACCESS_WRITE);
        if (ret < 0) {
                mlog_errno(ret);
                goto out_rollback;
@@ -200,7 +200,7 @@ static int update_backups(struct inode * inode, u32 clusters, char *data)
                if (cluster > clusters)
                        break;
 
-               ret = ocfs2_read_block(osb, blkno, &backup, 0, NULL);
+               ret = ocfs2_read_blocks_sync(osb, blkno, 1, &backup);
                if (ret < 0) {
                        mlog_errno(ret);
                        break;
@@ -236,8 +236,8 @@ static void ocfs2_update_super_and_backups(struct inode *inode,
         * update the superblock last.
         * It doesn't matter if the write failed.
         */
-       ret = ocfs2_read_block(osb, OCFS2_SUPER_BLOCK_BLKNO,
-                              &super_bh, 0, NULL);
+       ret = ocfs2_read_blocks_sync(osb, OCFS2_SUPER_BLOCK_BLKNO, 1,
+                                    &super_bh);
        if (ret < 0) {
                mlog_errno(ret);
                goto out;
@@ -257,8 +257,7 @@ static void ocfs2_update_super_and_backups(struct inode *inode,
                ret = update_backups(inode, clusters, super_bh->b_data);
 
 out:
-       if (super_bh)
-               brelse(super_bh);
+       brelse(super_bh);
        if (ret)
                printk(KERN_WARNING "ocfs2: Failed to update super blocks on %s"
                        " during fs resize. This condition is not fatal,"
@@ -315,6 +314,10 @@ int ocfs2_group_extend(struct inode * inode, int new_clusters)
 
        fe = (struct ocfs2_dinode *)main_bm_bh->b_data;
 
+       /* main_bm_bh is validated by inode read inside ocfs2_inode_lock(),
+        * so any corruption is a code bug. */
+       BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
+
        if (le16_to_cpu(fe->id2.i_chain.cl_cpg) !=
                                 ocfs2_group_bitmap_size(osb->sb) * 8) {
                mlog(ML_ERROR, "The disk is too old and small. "
@@ -323,31 +326,18 @@ int ocfs2_group_extend(struct inode * inode, int new_clusters)
                goto out_unlock;
        }
 
-       if (!OCFS2_IS_VALID_DINODE(fe)) {
-               OCFS2_RO_ON_INVALID_DINODE(main_bm_inode->i_sb, fe);
-               ret = -EIO;
-               goto out_unlock;
-       }
-
        first_new_cluster = le32_to_cpu(fe->i_clusters);
        lgd_blkno = ocfs2_which_cluster_group(main_bm_inode,
                                              first_new_cluster - 1);
 
-       ret = ocfs2_read_block(osb, lgd_blkno, &group_bh, OCFS2_BH_CACHED,
-                              main_bm_inode);
+       ret = ocfs2_read_group_descriptor(main_bm_inode, fe, lgd_blkno,
+                                         &group_bh);
        if (ret < 0) {
                mlog_errno(ret);
                goto out_unlock;
        }
-
        group = (struct ocfs2_group_desc *)group_bh->b_data;
 
-       ret = ocfs2_check_group_descriptor(inode->i_sb, fe, group);
-       if (ret) {
-               mlog_errno(ret);
-               goto out_unlock;
-       }
-
        cl_bpc = le16_to_cpu(fe->id2.i_chain.cl_bpc);
        if (le16_to_cpu(group->bg_bits) / cl_bpc + new_clusters >
                le16_to_cpu(fe->id2.i_chain.cl_cpg)) {
@@ -356,7 +346,7 @@ int ocfs2_group_extend(struct inode * inode, int new_clusters)
        }
 
        mlog(0, "extend the last group at %llu, new clusters = %d\n",
-            le64_to_cpu(group->bg_blkno), new_clusters);
+            (unsigned long long)le64_to_cpu(group->bg_blkno), new_clusters);
 
        handle = ocfs2_start_trans(osb, OCFS2_GROUP_EXTEND_CREDITS);
        if (IS_ERR(handle)) {
@@ -380,11 +370,224 @@ int ocfs2_group_extend(struct inode * inode, int new_clusters)
 out_commit:
        ocfs2_commit_trans(osb, handle);
 out_unlock:
-       if (group_bh)
-               brelse(group_bh);
+       brelse(group_bh);
+       brelse(main_bm_bh);
+
+       ocfs2_inode_unlock(main_bm_inode, 1);
+
+out_mutex:
+       mutex_unlock(&main_bm_inode->i_mutex);
+       iput(main_bm_inode);
+
+out:
+       mlog_exit_void();
+       return ret;
+}
+
+static int ocfs2_check_new_group(struct inode *inode,
+                                struct ocfs2_dinode *di,
+                                struct ocfs2_new_group_input *input,
+                                struct buffer_head *group_bh)
+{
+       int ret;
+       struct ocfs2_group_desc *gd =
+               (struct ocfs2_group_desc *)group_bh->b_data;
+       u16 cl_bpc = le16_to_cpu(di->id2.i_chain.cl_bpc);
+
+       ret = ocfs2_check_group_descriptor(inode->i_sb, di, group_bh);
+       if (ret)
+               goto out;
+
+       ret = -EINVAL;
+       if (le16_to_cpu(gd->bg_chain) != input->chain)
+               mlog(ML_ERROR, "Group descriptor # %llu has bad chain %u "
+                    "while input has %u set.\n",
+                    (unsigned long long)le64_to_cpu(gd->bg_blkno),
+                    le16_to_cpu(gd->bg_chain), input->chain);
+       else if (le16_to_cpu(gd->bg_bits) != input->clusters * cl_bpc)
+               mlog(ML_ERROR, "Group descriptor # %llu has bit count %u but "
+                    "input has %u clusters set\n",
+                    (unsigned long long)le64_to_cpu(gd->bg_blkno),
+                    le16_to_cpu(gd->bg_bits), input->clusters);
+       else if (le16_to_cpu(gd->bg_free_bits_count) != input->frees * cl_bpc)
+               mlog(ML_ERROR, "Group descriptor # %llu has free bit count %u "
+                    "but it should have %u set\n",
+                    (unsigned long long)le64_to_cpu(gd->bg_blkno),
+                    le16_to_cpu(gd->bg_bits),
+                    input->frees * cl_bpc);
+       else
+               ret = 0;
+
+out:
+       return ret;
+}
+
+static int ocfs2_verify_group_and_input(struct inode *inode,
+                                       struct ocfs2_dinode *di,
+                                       struct ocfs2_new_group_input *input,
+                                       struct buffer_head *group_bh)
+{
+       u16 cl_count = le16_to_cpu(di->id2.i_chain.cl_count);
+       u16 cl_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
+       u16 next_free = le16_to_cpu(di->id2.i_chain.cl_next_free_rec);
+       u32 cluster = ocfs2_blocks_to_clusters(inode->i_sb, input->group);
+       u32 total_clusters = le32_to_cpu(di->i_clusters);
+       int ret = -EINVAL;
+
+       if (cluster < total_clusters)
+               mlog(ML_ERROR, "add a group which is in the current volume.\n");
+       else if (input->chain >= cl_count)
+               mlog(ML_ERROR, "input chain exceeds the limit.\n");
+       else if (next_free != cl_count && next_free != input->chain)
+               mlog(ML_ERROR,
+                    "the add group should be in chain %u\n", next_free);
+       else if (total_clusters + input->clusters < total_clusters)
+               mlog(ML_ERROR, "add group's clusters overflow.\n");
+       else if (input->clusters > cl_cpg)
+               mlog(ML_ERROR, "the cluster exceeds the maximum of a group\n");
+       else if (input->frees > input->clusters)
+               mlog(ML_ERROR, "the free cluster exceeds the total clusters\n");
+       else if (total_clusters % cl_cpg != 0)
+               mlog(ML_ERROR,
+                    "the last group isn't full. Use group extend first.\n");
+       else if (input->group != ocfs2_which_cluster_group(inode, cluster))
+               mlog(ML_ERROR, "group blkno is invalid\n");
+       else if ((ret = ocfs2_check_new_group(inode, di, input, group_bh)))
+               mlog(ML_ERROR, "group descriptor check failed.\n");
+       else
+               ret = 0;
+
+       return ret;
+}
+
+/* Add a new group descriptor to global_bitmap. */
+int ocfs2_group_add(struct inode *inode, struct ocfs2_new_group_input *input)
+{
+       int ret;
+       handle_t *handle;
+       struct buffer_head *main_bm_bh = NULL;
+       struct inode *main_bm_inode = NULL;
+       struct ocfs2_dinode *fe = NULL;
+       struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+       struct buffer_head *group_bh = NULL;
+       struct ocfs2_group_desc *group = NULL;
+       struct ocfs2_chain_list *cl;
+       struct ocfs2_chain_rec *cr;
+       u16 cl_bpc;
+
+       mlog_entry_void();
+
+       if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
+               return -EROFS;
+
+       main_bm_inode = ocfs2_get_system_file_inode(osb,
+                                                   GLOBAL_BITMAP_SYSTEM_INODE,
+                                                   OCFS2_INVALID_SLOT);
+       if (!main_bm_inode) {
+               ret = -EINVAL;
+               mlog_errno(ret);
+               goto out;
+       }
 
-       if (main_bm_bh)
-               brelse(main_bm_bh);
+       mutex_lock(&main_bm_inode->i_mutex);
+
+       ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
+       if (ret < 0) {
+               mlog_errno(ret);
+               goto out_mutex;
+       }
+
+       fe = (struct ocfs2_dinode *)main_bm_bh->b_data;
+
+       if (le16_to_cpu(fe->id2.i_chain.cl_cpg) !=
+                                ocfs2_group_bitmap_size(osb->sb) * 8) {
+               mlog(ML_ERROR, "The disk is too old and small."
+                    " Force to do offline resize.");
+               ret = -EINVAL;
+               goto out_unlock;
+       }
+
+       ret = ocfs2_read_blocks_sync(osb, input->group, 1, &group_bh);
+       if (ret < 0) {
+               mlog(ML_ERROR, "Can't read the group descriptor # %llu "
+                    "from the device.", (unsigned long long)input->group);
+               goto out_unlock;
+       }
+
+       ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), group_bh);
+
+       ret = ocfs2_verify_group_and_input(main_bm_inode, fe, input, group_bh);
+       if (ret) {
+               mlog_errno(ret);
+               goto out_unlock;
+       }
+
+       mlog(0, "Add a new group  %llu in chain = %u, length = %u\n",
+            (unsigned long long)input->group, input->chain, input->clusters);
+
+       handle = ocfs2_start_trans(osb, OCFS2_GROUP_ADD_CREDITS);
+       if (IS_ERR(handle)) {
+               mlog_errno(PTR_ERR(handle));
+               ret = -EINVAL;
+               goto out_unlock;
+       }
+
+       cl_bpc = le16_to_cpu(fe->id2.i_chain.cl_bpc);
+       cl = &fe->id2.i_chain;
+       cr = &cl->cl_recs[input->chain];
+
+       ret = ocfs2_journal_access_gd(handle, INODE_CACHE(main_bm_inode),
+                                     group_bh, OCFS2_JOURNAL_ACCESS_WRITE);
+       if (ret < 0) {
+               mlog_errno(ret);
+               goto out_commit;
+       }
+
+       group = (struct ocfs2_group_desc *)group_bh->b_data;
+       group->bg_next_group = cr->c_blkno;
+
+       ret = ocfs2_journal_dirty(handle, group_bh);
+       if (ret < 0) {
+               mlog_errno(ret);
+               goto out_commit;
+       }
+
+       ret = ocfs2_journal_access_di(handle, INODE_CACHE(main_bm_inode),
+                                     main_bm_bh, OCFS2_JOURNAL_ACCESS_WRITE);
+       if (ret < 0) {
+               mlog_errno(ret);
+               goto out_commit;
+       }
+
+       if (input->chain == le16_to_cpu(cl->cl_next_free_rec)) {
+               le16_add_cpu(&cl->cl_next_free_rec, 1);
+               memset(cr, 0, sizeof(struct ocfs2_chain_rec));
+       }
+
+       cr->c_blkno = cpu_to_le64(input->group);
+       le32_add_cpu(&cr->c_total, input->clusters * cl_bpc);
+       le32_add_cpu(&cr->c_free, input->frees * cl_bpc);
+
+       le32_add_cpu(&fe->id1.bitmap1.i_total, input->clusters *cl_bpc);
+       le32_add_cpu(&fe->id1.bitmap1.i_used,
+                    (input->clusters - input->frees) * cl_bpc);
+       le32_add_cpu(&fe->i_clusters, input->clusters);
+
+       ocfs2_journal_dirty(handle, main_bm_bh);
+
+       spin_lock(&OCFS2_I(main_bm_inode)->ip_lock);
+       OCFS2_I(main_bm_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
+       le64_add_cpu(&fe->i_size, input->clusters << osb->s_clustersize_bits);
+       spin_unlock(&OCFS2_I(main_bm_inode)->ip_lock);
+       i_size_write(main_bm_inode, le64_to_cpu(fe->i_size));
+
+       ocfs2_update_super_and_backups(main_bm_inode, input->clusters);
+
+out_commit:
+       ocfs2_commit_trans(osb, handle);
+out_unlock:
+       brelse(group_bh);
+       brelse(main_bm_bh);
 
        ocfs2_inode_unlock(main_bm_inode, 1);