]> git.karo-electronics.de Git - mv-sheeva.git/blob - fs/ocfs2/resize.c
[PATCH 1/2] ocfs2: Add group extend for online resize
[mv-sheeva.git] / fs / ocfs2 / resize.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * resize.c
5  *
6  * volume resize.
7  * Inspired by ext3/resize.c.
8  *
9  * Copyright (C) 2007 Oracle.  All rights reserved.
10  *
11  * This program is free software; you can redistribute it and/or
12  * modify it under the terms of the GNU General Public
13  * License as published by the Free Software Foundation; either
14  * version 2 of the License, or (at your option) any later version.
15  *
16  * This program is distributed in the hope that it will be useful,
17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
19  * General Public License for more details.
20  *
21  * You should have received a copy of the GNU General Public
22  * License along with this program; if not, write to the
23  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
24  * Boston, MA 021110-1307, USA.
25  */
26
27 #include <linux/fs.h>
28 #include <linux/types.h>
29
30 #define MLOG_MASK_PREFIX ML_DISK_ALLOC
31 #include <cluster/masklog.h>
32
33 #include "ocfs2.h"
34
35 #include "alloc.h"
36 #include "dlmglue.h"
37 #include "inode.h"
38 #include "journal.h"
39 #include "super.h"
40 #include "sysfile.h"
41 #include "uptodate.h"
42
43 #include "buffer_head_io.h"
44 #include "suballoc.h"
45 #include "resize.h"
46
47 /*
48  * Check whether there are new backup superblocks exist
49  * in the last group. If there are some, mark them or clear
50  * them in the bitmap.
51  *
52  * Return how many backups we find in the last group.
53  */
54 static u16 ocfs2_calc_new_backup_super(struct inode *inode,
55                                        struct ocfs2_group_desc *gd,
56                                        int new_clusters,
57                                        u32 first_new_cluster,
58                                        u16 cl_cpg,
59                                        int set)
60 {
61         int i;
62         u16 backups = 0;
63         u32 cluster;
64         u64 blkno, gd_blkno, lgd_blkno = le64_to_cpu(gd->bg_blkno);
65
66         for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
67                 blkno = ocfs2_backup_super_blkno(inode->i_sb, i);
68                 cluster = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
69
70                 gd_blkno = ocfs2_which_cluster_group(inode, cluster);
71                 if (gd_blkno < lgd_blkno)
72                         continue;
73                 else if (gd_blkno > lgd_blkno)
74                         break;
75
76                 if (set)
77                         ocfs2_set_bit(cluster % cl_cpg,
78                                       (unsigned long *)gd->bg_bitmap);
79                 else
80                         ocfs2_clear_bit(cluster % cl_cpg,
81                                         (unsigned long *)gd->bg_bitmap);
82                 backups++;
83         }
84
85         mlog_exit_void();
86         return backups;
87 }
88
89 static int ocfs2_update_last_group_and_inode(handle_t *handle,
90                                              struct inode *bm_inode,
91                                              struct buffer_head *bm_bh,
92                                              struct buffer_head *group_bh,
93                                              u32 first_new_cluster,
94                                              int new_clusters)
95 {
96         int ret = 0;
97         struct ocfs2_super *osb = OCFS2_SB(bm_inode->i_sb);
98         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bm_bh->b_data;
99         struct ocfs2_chain_list *cl = &fe->id2.i_chain;
100         struct ocfs2_chain_rec *cr;
101         struct ocfs2_group_desc *group;
102         u16 chain, num_bits, backups = 0;
103         u16 cl_bpc = le16_to_cpu(cl->cl_bpc);
104         u16 cl_cpg = le16_to_cpu(cl->cl_cpg);
105
106         mlog_entry("(new_clusters=%d, first_new_cluster = %u)\n",
107                    new_clusters, first_new_cluster);
108
109         ret = ocfs2_journal_access(handle, bm_inode, group_bh,
110                                    OCFS2_JOURNAL_ACCESS_WRITE);
111         if (ret < 0) {
112                 mlog_errno(ret);
113                 goto out;
114         }
115
116         group = (struct ocfs2_group_desc *)group_bh->b_data;
117
118         /* update the group first. */
119         num_bits = new_clusters * cl_bpc;
120         le16_add_cpu(&group->bg_bits, num_bits);
121         le16_add_cpu(&group->bg_free_bits_count, num_bits);
122
123         /*
124          * check whether there are some new backup superblocks exist in
125          * this group and update the group bitmap accordingly.
126          */
127         if (OCFS2_HAS_COMPAT_FEATURE(osb->sb,
128                                      OCFS2_FEATURE_COMPAT_BACKUP_SB)) {
129                 backups = ocfs2_calc_new_backup_super(bm_inode,
130                                                      group,
131                                                      new_clusters,
132                                                      first_new_cluster,
133                                                      cl_cpg, 1);
134                 le16_add_cpu(&group->bg_free_bits_count, -1 * backups);
135         }
136
137         ret = ocfs2_journal_dirty(handle, group_bh);
138         if (ret < 0) {
139                 mlog_errno(ret);
140                 goto out_rollback;
141         }
142
143         /* update the inode accordingly. */
144         ret = ocfs2_journal_access(handle, bm_inode, bm_bh,
145                                    OCFS2_JOURNAL_ACCESS_WRITE);
146         if (ret < 0) {
147                 mlog_errno(ret);
148                 goto out_rollback;
149         }
150
151         chain = le16_to_cpu(group->bg_chain);
152         cr = (&cl->cl_recs[chain]);
153         le32_add_cpu(&cr->c_total, num_bits);
154         le32_add_cpu(&cr->c_free, num_bits);
155         le32_add_cpu(&fe->id1.bitmap1.i_total, num_bits);
156         le32_add_cpu(&fe->i_clusters, new_clusters);
157
158         if (backups) {
159                 le32_add_cpu(&cr->c_free, -1 * backups);
160                 le32_add_cpu(&fe->id1.bitmap1.i_used, backups);
161         }
162
163         spin_lock(&OCFS2_I(bm_inode)->ip_lock);
164         OCFS2_I(bm_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
165         le64_add_cpu(&fe->i_size, new_clusters << osb->s_clustersize_bits);
166         spin_unlock(&OCFS2_I(bm_inode)->ip_lock);
167         i_size_write(bm_inode, le64_to_cpu(fe->i_size));
168
169         ocfs2_journal_dirty(handle, bm_bh);
170
171 out_rollback:
172         if (ret < 0) {
173                 ocfs2_calc_new_backup_super(bm_inode,
174                                             group,
175                                             new_clusters,
176                                             first_new_cluster,
177                                             cl_cpg, 0);
178                 le16_add_cpu(&group->bg_free_bits_count, backups);
179                 le16_add_cpu(&group->bg_bits, -1 * num_bits);
180                 le16_add_cpu(&group->bg_free_bits_count, -1 * num_bits);
181         }
182 out:
183         mlog_exit(ret);
184         return ret;
185 }
186
187 static int update_backups(struct inode * inode, u32 clusters, char *data)
188 {
189         int i, ret = 0;
190         u32 cluster;
191         u64 blkno;
192         struct buffer_head *backup = NULL;
193         struct ocfs2_dinode *backup_di = NULL;
194         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
195
196         /* calculate the real backups we need to update. */
197         for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
198                 blkno = ocfs2_backup_super_blkno(inode->i_sb, i);
199                 cluster = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
200                 if (cluster > clusters)
201                         break;
202
203                 ret = ocfs2_read_block(osb, blkno, &backup, 0, NULL);
204                 if (ret < 0) {
205                         mlog_errno(ret);
206                         break;
207                 }
208
209                 memcpy(backup->b_data, data, inode->i_sb->s_blocksize);
210
211                 backup_di = (struct ocfs2_dinode *)backup->b_data;
212                 backup_di->i_blkno = cpu_to_le64(blkno);
213
214                 ret = ocfs2_write_super_or_backup(osb, backup);
215                 brelse(backup);
216                 backup = NULL;
217                 if (ret < 0) {
218                         mlog_errno(ret);
219                         break;
220                 }
221         }
222
223         return ret;
224 }
225
226 static void ocfs2_update_super_and_backups(struct inode *inode,
227                                            int new_clusters)
228 {
229         int ret;
230         u32 clusters = 0;
231         struct buffer_head *super_bh = NULL;
232         struct ocfs2_dinode *super_di = NULL;
233         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
234
235         /*
236          * update the superblock last.
237          * It doesn't matter if the write failed.
238          */
239         ret = ocfs2_read_block(osb, OCFS2_SUPER_BLOCK_BLKNO,
240                                &super_bh, 0, NULL);
241         if (ret < 0) {
242                 mlog_errno(ret);
243                 goto out;
244         }
245
246         super_di = (struct ocfs2_dinode *)super_bh->b_data;
247         le32_add_cpu(&super_di->i_clusters, new_clusters);
248         clusters = le32_to_cpu(super_di->i_clusters);
249
250         ret = ocfs2_write_super_or_backup(osb, super_bh);
251         if (ret < 0) {
252                 mlog_errno(ret);
253                 goto out;
254         }
255
256         if (OCFS2_HAS_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_COMPAT_BACKUP_SB))
257                 ret = update_backups(inode, clusters, super_bh->b_data);
258
259 out:
260         if (super_bh)
261                 brelse(super_bh);
262         if (ret)
263                 printk(KERN_WARNING "ocfs2: Failed to update super blocks on %s"
264                         " during fs resize. This condition is not fatal,"
265                         " but fsck.ocfs2 should be run to fix it\n",
266                         osb->dev_str);
267         return;
268 }
269
270 /*
271  * Extend the filesystem to the new number of clusters specified.  This entry
272  * point is only used to extend the current filesystem to the end of the last
273  * existing group.
274  */
275 int ocfs2_group_extend(struct inode * inode, int new_clusters)
276 {
277         int ret;
278         handle_t *handle;
279         struct buffer_head *main_bm_bh = NULL;
280         struct buffer_head *group_bh = NULL;
281         struct inode *main_bm_inode = NULL;
282         struct ocfs2_dinode *fe = NULL;
283         struct ocfs2_group_desc *group = NULL;
284         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
285         u16 cl_bpc;
286         u32 first_new_cluster;
287         u64 lgd_blkno;
288
289         mlog_entry_void();
290
291         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
292                 return -EROFS;
293
294         if (new_clusters < 0)
295                 return -EINVAL;
296         else if (new_clusters == 0)
297                 return 0;
298
299         main_bm_inode = ocfs2_get_system_file_inode(osb,
300                                                     GLOBAL_BITMAP_SYSTEM_INODE,
301                                                     OCFS2_INVALID_SLOT);
302         if (!main_bm_inode) {
303                 ret = -EINVAL;
304                 mlog_errno(ret);
305                 goto out;
306         }
307
308         mutex_lock(&main_bm_inode->i_mutex);
309
310         ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
311         if (ret < 0) {
312                 mlog_errno(ret);
313                 goto out_mutex;
314         }
315
316         fe = (struct ocfs2_dinode *)main_bm_bh->b_data;
317
318         if (le16_to_cpu(fe->id2.i_chain.cl_cpg) !=
319                                  ocfs2_group_bitmap_size(osb->sb) * 8) {
320                 mlog(ML_ERROR, "The disk is too old and small. "
321                      "Force to do offline resize.");
322                 ret = -EINVAL;
323                 goto out_unlock;
324         }
325
326         if (!OCFS2_IS_VALID_DINODE(fe)) {
327                 OCFS2_RO_ON_INVALID_DINODE(main_bm_inode->i_sb, fe);
328                 ret = -EIO;
329                 goto out_unlock;
330         }
331
332         first_new_cluster = le32_to_cpu(fe->i_clusters);
333         lgd_blkno = ocfs2_which_cluster_group(main_bm_inode,
334                                               first_new_cluster - 1);
335
336         ret = ocfs2_read_block(osb, lgd_blkno, &group_bh, OCFS2_BH_CACHED,
337                                main_bm_inode);
338         if (ret < 0) {
339                 mlog_errno(ret);
340                 goto out_unlock;
341         }
342
343         group = (struct ocfs2_group_desc *)group_bh->b_data;
344
345         ret = ocfs2_check_group_descriptor(inode->i_sb, fe, group);
346         if (ret) {
347                 mlog_errno(ret);
348                 goto out_unlock;
349         }
350
351         cl_bpc = le16_to_cpu(fe->id2.i_chain.cl_bpc);
352         if (le16_to_cpu(group->bg_bits) / cl_bpc + new_clusters >
353                 le16_to_cpu(fe->id2.i_chain.cl_cpg)) {
354                 ret = -EINVAL;
355                 goto out_unlock;
356         }
357
358         mlog(0, "extend the last group at %llu, new clusters = %d\n",
359              le64_to_cpu(group->bg_blkno), new_clusters);
360
361         handle = ocfs2_start_trans(osb, OCFS2_GROUP_EXTEND_CREDITS);
362         if (IS_ERR(handle)) {
363                 mlog_errno(PTR_ERR(handle));
364                 ret = -EINVAL;
365                 goto out_unlock;
366         }
367
368         /* update the last group descriptor and inode. */
369         ret = ocfs2_update_last_group_and_inode(handle, main_bm_inode,
370                                                 main_bm_bh, group_bh,
371                                                 first_new_cluster,
372                                                 new_clusters);
373         if (ret) {
374                 mlog_errno(ret);
375                 goto out_commit;
376         }
377
378         ocfs2_update_super_and_backups(main_bm_inode, new_clusters);
379
380 out_commit:
381         ocfs2_commit_trans(osb, handle);
382 out_unlock:
383         if (group_bh)
384                 brelse(group_bh);
385
386         if (main_bm_bh)
387                 brelse(main_bm_bh);
388
389         ocfs2_inode_unlock(main_bm_inode, 1);
390
391 out_mutex:
392         mutex_unlock(&main_bm_inode->i_mutex);
393         iput(main_bm_inode);
394
395 out:
396         mlog_exit_void();
397         return ret;
398 }