xref: /openbmc/linux/fs/ocfs2/resize.c (revision bc000245)
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 #include <cluster/masklog.h>
31 
32 #include "ocfs2.h"
33 
34 #include "alloc.h"
35 #include "dlmglue.h"
36 #include "inode.h"
37 #include "journal.h"
38 #include "super.h"
39 #include "sysfile.h"
40 #include "uptodate.h"
41 #include "ocfs2_trace.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 	return backups;
86 }
87 
88 static int ocfs2_update_last_group_and_inode(handle_t *handle,
89 					     struct inode *bm_inode,
90 					     struct buffer_head *bm_bh,
91 					     struct buffer_head *group_bh,
92 					     u32 first_new_cluster,
93 					     int new_clusters)
94 {
95 	int ret = 0;
96 	struct ocfs2_super *osb = OCFS2_SB(bm_inode->i_sb);
97 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bm_bh->b_data;
98 	struct ocfs2_chain_list *cl = &fe->id2.i_chain;
99 	struct ocfs2_chain_rec *cr;
100 	struct ocfs2_group_desc *group;
101 	u16 chain, num_bits, backups = 0;
102 	u16 cl_bpc = le16_to_cpu(cl->cl_bpc);
103 	u16 cl_cpg = le16_to_cpu(cl->cl_cpg);
104 
105 	trace_ocfs2_update_last_group_and_inode(new_clusters,
106 						first_new_cluster);
107 
108 	ret = ocfs2_journal_access_gd(handle, INODE_CACHE(bm_inode),
109 				      group_bh, OCFS2_JOURNAL_ACCESS_WRITE);
110 	if (ret < 0) {
111 		mlog_errno(ret);
112 		goto out;
113 	}
114 
115 	group = (struct ocfs2_group_desc *)group_bh->b_data;
116 
117 	/* update the group first. */
118 	num_bits = new_clusters * cl_bpc;
119 	le16_add_cpu(&group->bg_bits, num_bits);
120 	le16_add_cpu(&group->bg_free_bits_count, num_bits);
121 
122 	/*
123 	 * check whether there are some new backup superblocks exist in
124 	 * this group and update the group bitmap accordingly.
125 	 */
126 	if (OCFS2_HAS_COMPAT_FEATURE(osb->sb,
127 				     OCFS2_FEATURE_COMPAT_BACKUP_SB)) {
128 		backups = ocfs2_calc_new_backup_super(bm_inode,
129 						     group,
130 						     new_clusters,
131 						     first_new_cluster,
132 						     cl_cpg, 1);
133 		le16_add_cpu(&group->bg_free_bits_count, -1 * backups);
134 	}
135 
136 	ocfs2_journal_dirty(handle, group_bh);
137 
138 	/* update the inode accordingly. */
139 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(bm_inode), bm_bh,
140 				      OCFS2_JOURNAL_ACCESS_WRITE);
141 	if (ret < 0) {
142 		mlog_errno(ret);
143 		goto out_rollback;
144 	}
145 
146 	chain = le16_to_cpu(group->bg_chain);
147 	cr = (&cl->cl_recs[chain]);
148 	le32_add_cpu(&cr->c_total, num_bits);
149 	le32_add_cpu(&cr->c_free, num_bits);
150 	le32_add_cpu(&fe->id1.bitmap1.i_total, num_bits);
151 	le32_add_cpu(&fe->i_clusters, new_clusters);
152 
153 	if (backups) {
154 		le32_add_cpu(&cr->c_free, -1 * backups);
155 		le32_add_cpu(&fe->id1.bitmap1.i_used, backups);
156 	}
157 
158 	spin_lock(&OCFS2_I(bm_inode)->ip_lock);
159 	OCFS2_I(bm_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
160 	le64_add_cpu(&fe->i_size, new_clusters << osb->s_clustersize_bits);
161 	spin_unlock(&OCFS2_I(bm_inode)->ip_lock);
162 	i_size_write(bm_inode, le64_to_cpu(fe->i_size));
163 
164 	ocfs2_journal_dirty(handle, bm_bh);
165 
166 out_rollback:
167 	if (ret < 0) {
168 		ocfs2_calc_new_backup_super(bm_inode,
169 					    group,
170 					    new_clusters,
171 					    first_new_cluster,
172 					    cl_cpg, 0);
173 		le16_add_cpu(&group->bg_free_bits_count, backups);
174 		le16_add_cpu(&group->bg_bits, -1 * num_bits);
175 		le16_add_cpu(&group->bg_free_bits_count, -1 * num_bits);
176 	}
177 out:
178 	if (ret)
179 		mlog_errno(ret);
180 	return ret;
181 }
182 
183 static int update_backups(struct inode * inode, u32 clusters, char *data)
184 {
185 	int i, ret = 0;
186 	u32 cluster;
187 	u64 blkno;
188 	struct buffer_head *backup = NULL;
189 	struct ocfs2_dinode *backup_di = NULL;
190 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
191 
192 	/* calculate the real backups we need to update. */
193 	for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
194 		blkno = ocfs2_backup_super_blkno(inode->i_sb, i);
195 		cluster = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
196 		if (cluster > clusters)
197 			break;
198 
199 		ret = ocfs2_read_blocks_sync(osb, blkno, 1, &backup);
200 		if (ret < 0) {
201 			mlog_errno(ret);
202 			break;
203 		}
204 
205 		memcpy(backup->b_data, data, inode->i_sb->s_blocksize);
206 
207 		backup_di = (struct ocfs2_dinode *)backup->b_data;
208 		backup_di->i_blkno = cpu_to_le64(blkno);
209 
210 		ret = ocfs2_write_super_or_backup(osb, backup);
211 		brelse(backup);
212 		backup = NULL;
213 		if (ret < 0) {
214 			mlog_errno(ret);
215 			break;
216 		}
217 	}
218 
219 	return ret;
220 }
221 
222 static void ocfs2_update_super_and_backups(struct inode *inode,
223 					   int new_clusters)
224 {
225 	int ret;
226 	u32 clusters = 0;
227 	struct buffer_head *super_bh = NULL;
228 	struct ocfs2_dinode *super_di = NULL;
229 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
230 
231 	/*
232 	 * update the superblock last.
233 	 * It doesn't matter if the write failed.
234 	 */
235 	ret = ocfs2_read_blocks_sync(osb, OCFS2_SUPER_BLOCK_BLKNO, 1,
236 				     &super_bh);
237 	if (ret < 0) {
238 		mlog_errno(ret);
239 		goto out;
240 	}
241 
242 	super_di = (struct ocfs2_dinode *)super_bh->b_data;
243 	le32_add_cpu(&super_di->i_clusters, new_clusters);
244 	clusters = le32_to_cpu(super_di->i_clusters);
245 
246 	ret = ocfs2_write_super_or_backup(osb, super_bh);
247 	if (ret < 0) {
248 		mlog_errno(ret);
249 		goto out;
250 	}
251 
252 	if (OCFS2_HAS_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_COMPAT_BACKUP_SB))
253 		ret = update_backups(inode, clusters, super_bh->b_data);
254 
255 out:
256 	brelse(super_bh);
257 	if (ret)
258 		printk(KERN_WARNING "ocfs2: Failed to update super blocks on %s"
259 			" during fs resize. This condition is not fatal,"
260 			" but fsck.ocfs2 should be run to fix it\n",
261 			osb->dev_str);
262 	return;
263 }
264 
265 /*
266  * Extend the filesystem to the new number of clusters specified.  This entry
267  * point is only used to extend the current filesystem to the end of the last
268  * existing group.
269  */
270 int ocfs2_group_extend(struct inode * inode, int new_clusters)
271 {
272 	int ret;
273 	handle_t *handle;
274 	struct buffer_head *main_bm_bh = NULL;
275 	struct buffer_head *group_bh = NULL;
276 	struct inode *main_bm_inode = NULL;
277 	struct ocfs2_dinode *fe = NULL;
278 	struct ocfs2_group_desc *group = NULL;
279 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
280 	u16 cl_bpc;
281 	u32 first_new_cluster;
282 	u64 lgd_blkno;
283 
284 	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
285 		return -EROFS;
286 
287 	if (new_clusters < 0)
288 		return -EINVAL;
289 	else if (new_clusters == 0)
290 		return 0;
291 
292 	main_bm_inode = ocfs2_get_system_file_inode(osb,
293 						    GLOBAL_BITMAP_SYSTEM_INODE,
294 						    OCFS2_INVALID_SLOT);
295 	if (!main_bm_inode) {
296 		ret = -EINVAL;
297 		mlog_errno(ret);
298 		goto out;
299 	}
300 
301 	mutex_lock(&main_bm_inode->i_mutex);
302 
303 	ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
304 	if (ret < 0) {
305 		mlog_errno(ret);
306 		goto out_mutex;
307 	}
308 
309 	fe = (struct ocfs2_dinode *)main_bm_bh->b_data;
310 
311 	/* main_bm_bh is validated by inode read inside ocfs2_inode_lock(),
312 	 * so any corruption is a code bug. */
313 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
314 
315 	if (le16_to_cpu(fe->id2.i_chain.cl_cpg) !=
316 		ocfs2_group_bitmap_size(osb->sb, 0,
317 					osb->s_feature_incompat) * 8) {
318 		mlog(ML_ERROR, "The disk is too old and small. "
319 		     "Force to do offline resize.");
320 		ret = -EINVAL;
321 		goto out_unlock;
322 	}
323 
324 	first_new_cluster = le32_to_cpu(fe->i_clusters);
325 	lgd_blkno = ocfs2_which_cluster_group(main_bm_inode,
326 					      first_new_cluster - 1);
327 
328 	ret = ocfs2_read_group_descriptor(main_bm_inode, fe, lgd_blkno,
329 					  &group_bh);
330 	if (ret < 0) {
331 		mlog_errno(ret);
332 		goto out_unlock;
333 	}
334 	group = (struct ocfs2_group_desc *)group_bh->b_data;
335 
336 	cl_bpc = le16_to_cpu(fe->id2.i_chain.cl_bpc);
337 	if (le16_to_cpu(group->bg_bits) / cl_bpc + new_clusters >
338 		le16_to_cpu(fe->id2.i_chain.cl_cpg)) {
339 		ret = -EINVAL;
340 		goto out_unlock;
341 	}
342 
343 
344 	trace_ocfs2_group_extend(
345 	     (unsigned long long)le64_to_cpu(group->bg_blkno), new_clusters);
346 
347 	handle = ocfs2_start_trans(osb, OCFS2_GROUP_EXTEND_CREDITS);
348 	if (IS_ERR(handle)) {
349 		mlog_errno(PTR_ERR(handle));
350 		ret = -EINVAL;
351 		goto out_unlock;
352 	}
353 
354 	/* update the last group descriptor and inode. */
355 	ret = ocfs2_update_last_group_and_inode(handle, main_bm_inode,
356 						main_bm_bh, group_bh,
357 						first_new_cluster,
358 						new_clusters);
359 	if (ret) {
360 		mlog_errno(ret);
361 		goto out_commit;
362 	}
363 
364 	ocfs2_update_super_and_backups(main_bm_inode, new_clusters);
365 
366 out_commit:
367 	ocfs2_commit_trans(osb, handle);
368 out_unlock:
369 	brelse(group_bh);
370 	brelse(main_bm_bh);
371 
372 	ocfs2_inode_unlock(main_bm_inode, 1);
373 
374 out_mutex:
375 	mutex_unlock(&main_bm_inode->i_mutex);
376 	iput(main_bm_inode);
377 
378 out:
379 	return ret;
380 }
381 
382 static int ocfs2_check_new_group(struct inode *inode,
383 				 struct ocfs2_dinode *di,
384 				 struct ocfs2_new_group_input *input,
385 				 struct buffer_head *group_bh)
386 {
387 	int ret;
388 	struct ocfs2_group_desc *gd =
389 		(struct ocfs2_group_desc *)group_bh->b_data;
390 	u16 cl_bpc = le16_to_cpu(di->id2.i_chain.cl_bpc);
391 
392 	ret = ocfs2_check_group_descriptor(inode->i_sb, di, group_bh);
393 	if (ret)
394 		goto out;
395 
396 	ret = -EINVAL;
397 	if (le16_to_cpu(gd->bg_chain) != input->chain)
398 		mlog(ML_ERROR, "Group descriptor # %llu has bad chain %u "
399 		     "while input has %u set.\n",
400 		     (unsigned long long)le64_to_cpu(gd->bg_blkno),
401 		     le16_to_cpu(gd->bg_chain), input->chain);
402 	else if (le16_to_cpu(gd->bg_bits) != input->clusters * cl_bpc)
403 		mlog(ML_ERROR, "Group descriptor # %llu has bit count %u but "
404 		     "input has %u clusters set\n",
405 		     (unsigned long long)le64_to_cpu(gd->bg_blkno),
406 		     le16_to_cpu(gd->bg_bits), input->clusters);
407 	else if (le16_to_cpu(gd->bg_free_bits_count) != input->frees * cl_bpc)
408 		mlog(ML_ERROR, "Group descriptor # %llu has free bit count %u "
409 		     "but it should have %u set\n",
410 		     (unsigned long long)le64_to_cpu(gd->bg_blkno),
411 		     le16_to_cpu(gd->bg_bits),
412 		     input->frees * cl_bpc);
413 	else
414 		ret = 0;
415 
416 out:
417 	return ret;
418 }
419 
420 static int ocfs2_verify_group_and_input(struct inode *inode,
421 					struct ocfs2_dinode *di,
422 					struct ocfs2_new_group_input *input,
423 					struct buffer_head *group_bh)
424 {
425 	u16 cl_count = le16_to_cpu(di->id2.i_chain.cl_count);
426 	u16 cl_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
427 	u16 next_free = le16_to_cpu(di->id2.i_chain.cl_next_free_rec);
428 	u32 cluster = ocfs2_blocks_to_clusters(inode->i_sb, input->group);
429 	u32 total_clusters = le32_to_cpu(di->i_clusters);
430 	int ret = -EINVAL;
431 
432 	if (cluster < total_clusters)
433 		mlog(ML_ERROR, "add a group which is in the current volume.\n");
434 	else if (input->chain >= cl_count)
435 		mlog(ML_ERROR, "input chain exceeds the limit.\n");
436 	else if (next_free != cl_count && next_free != input->chain)
437 		mlog(ML_ERROR,
438 		     "the add group should be in chain %u\n", next_free);
439 	else if (total_clusters + input->clusters < total_clusters)
440 		mlog(ML_ERROR, "add group's clusters overflow.\n");
441 	else if (input->clusters > cl_cpg)
442 		mlog(ML_ERROR, "the cluster exceeds the maximum of a group\n");
443 	else if (input->frees > input->clusters)
444 		mlog(ML_ERROR, "the free cluster exceeds the total clusters\n");
445 	else if (total_clusters % cl_cpg != 0)
446 		mlog(ML_ERROR,
447 		     "the last group isn't full. Use group extend first.\n");
448 	else if (input->group != ocfs2_which_cluster_group(inode, cluster))
449 		mlog(ML_ERROR, "group blkno is invalid\n");
450 	else if ((ret = ocfs2_check_new_group(inode, di, input, group_bh)))
451 		mlog(ML_ERROR, "group descriptor check failed.\n");
452 	else
453 		ret = 0;
454 
455 	return ret;
456 }
457 
458 /* Add a new group descriptor to global_bitmap. */
459 int ocfs2_group_add(struct inode *inode, struct ocfs2_new_group_input *input)
460 {
461 	int ret;
462 	handle_t *handle;
463 	struct buffer_head *main_bm_bh = NULL;
464 	struct inode *main_bm_inode = NULL;
465 	struct ocfs2_dinode *fe = NULL;
466 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
467 	struct buffer_head *group_bh = NULL;
468 	struct ocfs2_group_desc *group = NULL;
469 	struct ocfs2_chain_list *cl;
470 	struct ocfs2_chain_rec *cr;
471 	u16 cl_bpc;
472 	u64 bg_ptr;
473 
474 	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
475 		return -EROFS;
476 
477 	main_bm_inode = ocfs2_get_system_file_inode(osb,
478 						    GLOBAL_BITMAP_SYSTEM_INODE,
479 						    OCFS2_INVALID_SLOT);
480 	if (!main_bm_inode) {
481 		ret = -EINVAL;
482 		mlog_errno(ret);
483 		goto out;
484 	}
485 
486 	mutex_lock(&main_bm_inode->i_mutex);
487 
488 	ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
489 	if (ret < 0) {
490 		mlog_errno(ret);
491 		goto out_mutex;
492 	}
493 
494 	fe = (struct ocfs2_dinode *)main_bm_bh->b_data;
495 
496 	if (le16_to_cpu(fe->id2.i_chain.cl_cpg) !=
497 		ocfs2_group_bitmap_size(osb->sb, 0,
498 					osb->s_feature_incompat) * 8) {
499 		mlog(ML_ERROR, "The disk is too old and small."
500 		     " Force to do offline resize.");
501 		ret = -EINVAL;
502 		goto out_unlock;
503 	}
504 
505 	ret = ocfs2_read_blocks_sync(osb, input->group, 1, &group_bh);
506 	if (ret < 0) {
507 		mlog(ML_ERROR, "Can't read the group descriptor # %llu "
508 		     "from the device.", (unsigned long long)input->group);
509 		goto out_unlock;
510 	}
511 
512 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), group_bh);
513 
514 	ret = ocfs2_verify_group_and_input(main_bm_inode, fe, input, group_bh);
515 	if (ret) {
516 		mlog_errno(ret);
517 		goto out_free_group_bh;
518 	}
519 
520 	trace_ocfs2_group_add((unsigned long long)input->group,
521 			       input->chain, input->clusters, input->frees);
522 
523 	handle = ocfs2_start_trans(osb, OCFS2_GROUP_ADD_CREDITS);
524 	if (IS_ERR(handle)) {
525 		mlog_errno(PTR_ERR(handle));
526 		ret = -EINVAL;
527 		goto out_free_group_bh;
528 	}
529 
530 	cl_bpc = le16_to_cpu(fe->id2.i_chain.cl_bpc);
531 	cl = &fe->id2.i_chain;
532 	cr = &cl->cl_recs[input->chain];
533 
534 	ret = ocfs2_journal_access_gd(handle, INODE_CACHE(main_bm_inode),
535 				      group_bh, OCFS2_JOURNAL_ACCESS_WRITE);
536 	if (ret < 0) {
537 		mlog_errno(ret);
538 		goto out_commit;
539 	}
540 
541 	group = (struct ocfs2_group_desc *)group_bh->b_data;
542 	bg_ptr = le64_to_cpu(group->bg_next_group);
543 	group->bg_next_group = cr->c_blkno;
544 	ocfs2_journal_dirty(handle, group_bh);
545 
546 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(main_bm_inode),
547 				      main_bm_bh, OCFS2_JOURNAL_ACCESS_WRITE);
548 	if (ret < 0) {
549 		group->bg_next_group = cpu_to_le64(bg_ptr);
550 		mlog_errno(ret);
551 		goto out_commit;
552 	}
553 
554 	if (input->chain == le16_to_cpu(cl->cl_next_free_rec)) {
555 		le16_add_cpu(&cl->cl_next_free_rec, 1);
556 		memset(cr, 0, sizeof(struct ocfs2_chain_rec));
557 	}
558 
559 	cr->c_blkno = cpu_to_le64(input->group);
560 	le32_add_cpu(&cr->c_total, input->clusters * cl_bpc);
561 	le32_add_cpu(&cr->c_free, input->frees * cl_bpc);
562 
563 	le32_add_cpu(&fe->id1.bitmap1.i_total, input->clusters *cl_bpc);
564 	le32_add_cpu(&fe->id1.bitmap1.i_used,
565 		     (input->clusters - input->frees) * cl_bpc);
566 	le32_add_cpu(&fe->i_clusters, input->clusters);
567 
568 	ocfs2_journal_dirty(handle, main_bm_bh);
569 
570 	spin_lock(&OCFS2_I(main_bm_inode)->ip_lock);
571 	OCFS2_I(main_bm_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
572 	le64_add_cpu(&fe->i_size, input->clusters << osb->s_clustersize_bits);
573 	spin_unlock(&OCFS2_I(main_bm_inode)->ip_lock);
574 	i_size_write(main_bm_inode, le64_to_cpu(fe->i_size));
575 
576 	ocfs2_update_super_and_backups(main_bm_inode, input->clusters);
577 
578 out_commit:
579 	ocfs2_commit_trans(osb, handle);
580 
581 out_free_group_bh:
582 	brelse(group_bh);
583 
584 out_unlock:
585 	brelse(main_bm_bh);
586 
587 	ocfs2_inode_unlock(main_bm_inode, 1);
588 
589 out_mutex:
590 	mutex_unlock(&main_bm_inode->i_mutex);
591 	iput(main_bm_inode);
592 
593 out:
594 	return ret;
595 }
596