Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jbarnes...
[linux-2.6] / 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_gd(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_di(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_blocks_sync(osb, blkno, 1, &backup);
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_blocks_sync(osb, OCFS2_SUPER_BLOCK_BLKNO, 1,
240                                      &super_bh);
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         brelse(super_bh);
261         if (ret)
262                 printk(KERN_WARNING "ocfs2: Failed to update super blocks on %s"
263                         " during fs resize. This condition is not fatal,"
264                         " but fsck.ocfs2 should be run to fix it\n",
265                         osb->dev_str);
266         return;
267 }
268
269 /*
270  * Extend the filesystem to the new number of clusters specified.  This entry
271  * point is only used to extend the current filesystem to the end of the last
272  * existing group.
273  */
274 int ocfs2_group_extend(struct inode * inode, int new_clusters)
275 {
276         int ret;
277         handle_t *handle;
278         struct buffer_head *main_bm_bh = NULL;
279         struct buffer_head *group_bh = NULL;
280         struct inode *main_bm_inode = NULL;
281         struct ocfs2_dinode *fe = NULL;
282         struct ocfs2_group_desc *group = NULL;
283         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
284         u16 cl_bpc;
285         u32 first_new_cluster;
286         u64 lgd_blkno;
287
288         mlog_entry_void();
289
290         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
291                 return -EROFS;
292
293         if (new_clusters < 0)
294                 return -EINVAL;
295         else if (new_clusters == 0)
296                 return 0;
297
298         main_bm_inode = ocfs2_get_system_file_inode(osb,
299                                                     GLOBAL_BITMAP_SYSTEM_INODE,
300                                                     OCFS2_INVALID_SLOT);
301         if (!main_bm_inode) {
302                 ret = -EINVAL;
303                 mlog_errno(ret);
304                 goto out;
305         }
306
307         mutex_lock(&main_bm_inode->i_mutex);
308
309         ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
310         if (ret < 0) {
311                 mlog_errno(ret);
312                 goto out_mutex;
313         }
314
315         fe = (struct ocfs2_dinode *)main_bm_bh->b_data;
316
317         /* main_bm_bh is validated by inode read inside ocfs2_inode_lock(),
318          * so any corruption is a code bug. */
319         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
320
321         if (le16_to_cpu(fe->id2.i_chain.cl_cpg) !=
322                                  ocfs2_group_bitmap_size(osb->sb) * 8) {
323                 mlog(ML_ERROR, "The disk is too old and small. "
324                      "Force to do offline resize.");
325                 ret = -EINVAL;
326                 goto out_unlock;
327         }
328
329         first_new_cluster = le32_to_cpu(fe->i_clusters);
330         lgd_blkno = ocfs2_which_cluster_group(main_bm_inode,
331                                               first_new_cluster - 1);
332
333         ret = ocfs2_read_group_descriptor(main_bm_inode, fe, lgd_blkno,
334                                           &group_bh);
335         if (ret < 0) {
336                 mlog_errno(ret);
337                 goto out_unlock;
338         }
339         group = (struct ocfs2_group_desc *)group_bh->b_data;
340
341         cl_bpc = le16_to_cpu(fe->id2.i_chain.cl_bpc);
342         if (le16_to_cpu(group->bg_bits) / cl_bpc + new_clusters >
343                 le16_to_cpu(fe->id2.i_chain.cl_cpg)) {
344                 ret = -EINVAL;
345                 goto out_unlock;
346         }
347
348         mlog(0, "extend the last group at %llu, new clusters = %d\n",
349              (unsigned long long)le64_to_cpu(group->bg_blkno), new_clusters);
350
351         handle = ocfs2_start_trans(osb, OCFS2_GROUP_EXTEND_CREDITS);
352         if (IS_ERR(handle)) {
353                 mlog_errno(PTR_ERR(handle));
354                 ret = -EINVAL;
355                 goto out_unlock;
356         }
357
358         /* update the last group descriptor and inode. */
359         ret = ocfs2_update_last_group_and_inode(handle, main_bm_inode,
360                                                 main_bm_bh, group_bh,
361                                                 first_new_cluster,
362                                                 new_clusters);
363         if (ret) {
364                 mlog_errno(ret);
365                 goto out_commit;
366         }
367
368         ocfs2_update_super_and_backups(main_bm_inode, new_clusters);
369
370 out_commit:
371         ocfs2_commit_trans(osb, handle);
372 out_unlock:
373         brelse(group_bh);
374         brelse(main_bm_bh);
375
376         ocfs2_inode_unlock(main_bm_inode, 1);
377
378 out_mutex:
379         mutex_unlock(&main_bm_inode->i_mutex);
380         iput(main_bm_inode);
381
382 out:
383         mlog_exit_void();
384         return ret;
385 }
386
387 static int ocfs2_check_new_group(struct inode *inode,
388                                  struct ocfs2_dinode *di,
389                                  struct ocfs2_new_group_input *input,
390                                  struct buffer_head *group_bh)
391 {
392         int ret;
393         struct ocfs2_group_desc *gd =
394                 (struct ocfs2_group_desc *)group_bh->b_data;
395         u16 cl_bpc = le16_to_cpu(di->id2.i_chain.cl_bpc);
396
397         ret = ocfs2_check_group_descriptor(inode->i_sb, di, group_bh);
398         if (ret)
399                 goto out;
400
401         ret = -EINVAL;
402         if (le16_to_cpu(gd->bg_chain) != input->chain)
403                 mlog(ML_ERROR, "Group descriptor # %llu has bad chain %u "
404                      "while input has %u set.\n",
405                      (unsigned long long)le64_to_cpu(gd->bg_blkno),
406                      le16_to_cpu(gd->bg_chain), input->chain);
407         else if (le16_to_cpu(gd->bg_bits) != input->clusters * cl_bpc)
408                 mlog(ML_ERROR, "Group descriptor # %llu has bit count %u but "
409                      "input has %u clusters set\n",
410                      (unsigned long long)le64_to_cpu(gd->bg_blkno),
411                      le16_to_cpu(gd->bg_bits), input->clusters);
412         else if (le16_to_cpu(gd->bg_free_bits_count) != input->frees * cl_bpc)
413                 mlog(ML_ERROR, "Group descriptor # %llu has free bit count %u "
414                      "but it should have %u set\n",
415                      (unsigned long long)le64_to_cpu(gd->bg_blkno),
416                      le16_to_cpu(gd->bg_bits),
417                      input->frees * cl_bpc);
418         else
419                 ret = 0;
420
421 out:
422         return ret;
423 }
424
425 static int ocfs2_verify_group_and_input(struct inode *inode,
426                                         struct ocfs2_dinode *di,
427                                         struct ocfs2_new_group_input *input,
428                                         struct buffer_head *group_bh)
429 {
430         u16 cl_count = le16_to_cpu(di->id2.i_chain.cl_count);
431         u16 cl_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
432         u16 next_free = le16_to_cpu(di->id2.i_chain.cl_next_free_rec);
433         u32 cluster = ocfs2_blocks_to_clusters(inode->i_sb, input->group);
434         u32 total_clusters = le32_to_cpu(di->i_clusters);
435         int ret = -EINVAL;
436
437         if (cluster < total_clusters)
438                 mlog(ML_ERROR, "add a group which is in the current volume.\n");
439         else if (input->chain >= cl_count)
440                 mlog(ML_ERROR, "input chain exceeds the limit.\n");
441         else if (next_free != cl_count && next_free != input->chain)
442                 mlog(ML_ERROR,
443                      "the add group should be in chain %u\n", next_free);
444         else if (total_clusters + input->clusters < total_clusters)
445                 mlog(ML_ERROR, "add group's clusters overflow.\n");
446         else if (input->clusters > cl_cpg)
447                 mlog(ML_ERROR, "the cluster exceeds the maximum of a group\n");
448         else if (input->frees > input->clusters)
449                 mlog(ML_ERROR, "the free cluster exceeds the total clusters\n");
450         else if (total_clusters % cl_cpg != 0)
451                 mlog(ML_ERROR,
452                      "the last group isn't full. Use group extend first.\n");
453         else if (input->group != ocfs2_which_cluster_group(inode, cluster))
454                 mlog(ML_ERROR, "group blkno is invalid\n");
455         else if ((ret = ocfs2_check_new_group(inode, di, input, group_bh)))
456                 mlog(ML_ERROR, "group descriptor check failed.\n");
457         else
458                 ret = 0;
459
460         return ret;
461 }
462
463 /* Add a new group descriptor to global_bitmap. */
464 int ocfs2_group_add(struct inode *inode, struct ocfs2_new_group_input *input)
465 {
466         int ret;
467         handle_t *handle;
468         struct buffer_head *main_bm_bh = NULL;
469         struct inode *main_bm_inode = NULL;
470         struct ocfs2_dinode *fe = NULL;
471         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
472         struct buffer_head *group_bh = NULL;
473         struct ocfs2_group_desc *group = NULL;
474         struct ocfs2_chain_list *cl;
475         struct ocfs2_chain_rec *cr;
476         u16 cl_bpc;
477
478         mlog_entry_void();
479
480         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
481                 return -EROFS;
482
483         main_bm_inode = ocfs2_get_system_file_inode(osb,
484                                                     GLOBAL_BITMAP_SYSTEM_INODE,
485                                                     OCFS2_INVALID_SLOT);
486         if (!main_bm_inode) {
487                 ret = -EINVAL;
488                 mlog_errno(ret);
489                 goto out;
490         }
491
492         mutex_lock(&main_bm_inode->i_mutex);
493
494         ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
495         if (ret < 0) {
496                 mlog_errno(ret);
497                 goto out_mutex;
498         }
499
500         fe = (struct ocfs2_dinode *)main_bm_bh->b_data;
501
502         if (le16_to_cpu(fe->id2.i_chain.cl_cpg) !=
503                                  ocfs2_group_bitmap_size(osb->sb) * 8) {
504                 mlog(ML_ERROR, "The disk is too old and small."
505                      " Force to do offline resize.");
506                 ret = -EINVAL;
507                 goto out_unlock;
508         }
509
510         ret = ocfs2_read_blocks_sync(osb, input->group, 1, &group_bh);
511         if (ret < 0) {
512                 mlog(ML_ERROR, "Can't read the group descriptor # %llu "
513                      "from the device.", (unsigned long long)input->group);
514                 goto out_unlock;
515         }
516
517         ocfs2_set_new_buffer_uptodate(inode, group_bh);
518
519         ret = ocfs2_verify_group_and_input(main_bm_inode, fe, input, group_bh);
520         if (ret) {
521                 mlog_errno(ret);
522                 goto out_unlock;
523         }
524
525         mlog(0, "Add a new group  %llu in chain = %u, length = %u\n",
526              (unsigned long long)input->group, input->chain, input->clusters);
527
528         handle = ocfs2_start_trans(osb, OCFS2_GROUP_ADD_CREDITS);
529         if (IS_ERR(handle)) {
530                 mlog_errno(PTR_ERR(handle));
531                 ret = -EINVAL;
532                 goto out_unlock;
533         }
534
535         cl_bpc = le16_to_cpu(fe->id2.i_chain.cl_bpc);
536         cl = &fe->id2.i_chain;
537         cr = &cl->cl_recs[input->chain];
538
539         ret = ocfs2_journal_access_gd(handle, main_bm_inode, group_bh,
540                                       OCFS2_JOURNAL_ACCESS_WRITE);
541         if (ret < 0) {
542                 mlog_errno(ret);
543                 goto out_commit;
544         }
545
546         group = (struct ocfs2_group_desc *)group_bh->b_data;
547         group->bg_next_group = cr->c_blkno;
548
549         ret = ocfs2_journal_dirty(handle, group_bh);
550         if (ret < 0) {
551                 mlog_errno(ret);
552                 goto out_commit;
553         }
554
555         ret = ocfs2_journal_access_di(handle, main_bm_inode, main_bm_bh,
556                                       OCFS2_JOURNAL_ACCESS_WRITE);
557         if (ret < 0) {
558                 mlog_errno(ret);
559                 goto out_commit;
560         }
561
562         if (input->chain == le16_to_cpu(cl->cl_next_free_rec)) {
563                 le16_add_cpu(&cl->cl_next_free_rec, 1);
564                 memset(cr, 0, sizeof(struct ocfs2_chain_rec));
565         }
566
567         cr->c_blkno = cpu_to_le64(input->group);
568         le32_add_cpu(&cr->c_total, input->clusters * cl_bpc);
569         le32_add_cpu(&cr->c_free, input->frees * cl_bpc);
570
571         le32_add_cpu(&fe->id1.bitmap1.i_total, input->clusters *cl_bpc);
572         le32_add_cpu(&fe->id1.bitmap1.i_used,
573                      (input->clusters - input->frees) * cl_bpc);
574         le32_add_cpu(&fe->i_clusters, input->clusters);
575
576         ocfs2_journal_dirty(handle, main_bm_bh);
577
578         spin_lock(&OCFS2_I(main_bm_inode)->ip_lock);
579         OCFS2_I(main_bm_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
580         le64_add_cpu(&fe->i_size, input->clusters << osb->s_clustersize_bits);
581         spin_unlock(&OCFS2_I(main_bm_inode)->ip_lock);
582         i_size_write(main_bm_inode, le64_to_cpu(fe->i_size));
583
584         ocfs2_update_super_and_backups(main_bm_inode, input->clusters);
585
586 out_commit:
587         ocfs2_commit_trans(osb, handle);
588 out_unlock:
589         brelse(group_bh);
590         brelse(main_bm_bh);
591
592         ocfs2_inode_unlock(main_bm_inode, 1);
593
594 out_mutex:
595         mutex_unlock(&main_bm_inode->i_mutex);
596         iput(main_bm_inode);
597
598 out:
599         mlog_exit_void();
600         return ret;
601 }