resize.c

  1// SPDX-License-Identifier: GPL-2.0-or-later
  2/*
  3 * resize.c
  4 *
  5 * volume resize.
  6 * Inspired by ext3/resize.c.
  7 *
  8 * Copyright (C) 2007 Oracle.  All rights reserved.
  9 */
 10
 11#include <linux/fs.h>
 12#include <linux/types.h>
 13
 14#include <cluster/masklog.h>
 15
 16#include "ocfs2.h"
 17
 18#include "alloc.h"
 19#include "dlmglue.h"
 20#include "inode.h"
 21#include "journal.h"
 22#include "super.h"
 23#include "sysfile.h"
 24#include "uptodate.h"
 25#include "ocfs2_trace.h"
 26
 27#include "buffer_head_io.h"
 28#include "suballoc.h"
 29#include "resize.h"
 30
 31/*
 32 * Check whether there are new backup superblocks exist
 33 * in the last group. If there are some, mark them or clear
 34 * them in the bitmap.
 35 *
 36 * Return how many backups we find in the last group.
 37 */
 38static u16 ocfs2_calc_new_backup_super(struct inode *inode,
 39				       struct ocfs2_group_desc *gd,
 40				       u16 cl_cpg,
 41				       u16 old_bg_clusters,
 42				       int set)
 43{
 44	int i;
 45	u16 backups = 0;
 46	u32 cluster, lgd_cluster;
 47	u64 blkno, gd_blkno, lgd_blkno = le64_to_cpu(gd->bg_blkno);
 48
 49	for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
 50		blkno = ocfs2_backup_super_blkno(inode->i_sb, i);
 51		cluster = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
 52
 53		gd_blkno = ocfs2_which_cluster_group(inode, cluster);
 54		if (gd_blkno < lgd_blkno)
 55			continue;
 56		else if (gd_blkno > lgd_blkno)
 57			break;
 58
 59		/* check if already done backup super */
 60		lgd_cluster = ocfs2_blocks_to_clusters(inode->i_sb, lgd_blkno);
 61		lgd_cluster += old_bg_clusters;
 62		if (lgd_cluster >= cluster)
 63			continue;
 64
 65		if (set)
 66			ocfs2_set_bit(cluster % cl_cpg,
 67				      (unsigned long *)gd->bg_bitmap);
 68		else
 69			ocfs2_clear_bit(cluster % cl_cpg,
 70					(unsigned long *)gd->bg_bitmap);
 71		backups++;
 72	}
 73
 74	return backups;
 75}
 76
 77static int ocfs2_update_last_group_and_inode(handle_t *handle,
 78					     struct inode *bm_inode,
 79					     struct buffer_head *bm_bh,
 80					     struct buffer_head *group_bh,
 81					     u32 first_new_cluster,
 82					     int new_clusters)
 83{
 84	int ret = 0;
 85	struct ocfs2_super *osb = OCFS2_SB(bm_inode->i_sb);
 86	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bm_bh->b_data;
 87	struct ocfs2_chain_list *cl = &fe->id2.i_chain;
 88	struct ocfs2_chain_rec *cr;
 89	struct ocfs2_group_desc *group;
 90	u16 chain, num_bits, backups = 0;
 91	u16 cl_bpc = le16_to_cpu(cl->cl_bpc);
 92	u16 cl_cpg = le16_to_cpu(cl->cl_cpg);
 93	u16 old_bg_clusters;
 94	u16 contig_bits;
 95	__le16 old_bg_contig_free_bits;
 96
 97	trace_ocfs2_update_last_group_and_inode(new_clusters,
 98						first_new_cluster);
 99
100	ret = ocfs2_journal_access_gd(handle, INODE_CACHE(bm_inode),
101				      group_bh, OCFS2_JOURNAL_ACCESS_WRITE);
102	if (ret < 0) {
103		mlog_errno(ret);
104		goto out;
105	}
106
107	group = (struct ocfs2_group_desc *)group_bh->b_data;
108
109	old_bg_clusters = le16_to_cpu(group->bg_bits) / cl_bpc;
110	/* update the group first. */
111	num_bits = new_clusters * cl_bpc;
112	le16_add_cpu(&group->bg_bits, num_bits);
113	le16_add_cpu(&group->bg_free_bits_count, num_bits);
114
115	/*
116	 * check whether there are some new backup superblocks exist in
117	 * this group and update the group bitmap accordingly.
118	 */
119	if (OCFS2_HAS_COMPAT_FEATURE(osb->sb,
120				     OCFS2_FEATURE_COMPAT_BACKUP_SB)) {
121		backups = ocfs2_calc_new_backup_super(bm_inode,
122						     group,
123						     cl_cpg, old_bg_clusters, 1);
124		le16_add_cpu(&group->bg_free_bits_count, -1 * backups);
125	}
126
127	contig_bits = ocfs2_find_max_contig_free_bits(group->bg_bitmap,
128					le16_to_cpu(group->bg_bits), 0);
129	old_bg_contig_free_bits = group->bg_contig_free_bits;
130	group->bg_contig_free_bits = cpu_to_le16(contig_bits);
131
132	ocfs2_journal_dirty(handle, group_bh);
133
134	/* update the inode accordingly. */
135	ret = ocfs2_journal_access_di(handle, INODE_CACHE(bm_inode), bm_bh,
136				      OCFS2_JOURNAL_ACCESS_WRITE);
137	if (ret < 0) {
138		mlog_errno(ret);
139		goto out_rollback;
140	}
141
142	chain = le16_to_cpu(group->bg_chain);
143	cr = (&cl->cl_recs[chain]);
144	le32_add_cpu(&cr->c_total, num_bits);
145	le32_add_cpu(&cr->c_free, num_bits);
146	le32_add_cpu(&fe->id1.bitmap1.i_total, num_bits);
147	le32_add_cpu(&fe->i_clusters, new_clusters);
148
149	if (backups) {
150		le32_add_cpu(&cr->c_free, -1 * backups);
151		le32_add_cpu(&fe->id1.bitmap1.i_used, backups);
152	}
153
154	spin_lock(&OCFS2_I(bm_inode)->ip_lock);
155	OCFS2_I(bm_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
156	le64_add_cpu(&fe->i_size, (u64)new_clusters << osb->s_clustersize_bits);
157	spin_unlock(&OCFS2_I(bm_inode)->ip_lock);
158	i_size_write(bm_inode, le64_to_cpu(fe->i_size));
159
160	ocfs2_journal_dirty(handle, bm_bh);
161
162out_rollback:
163	if (ret < 0) {
164		ocfs2_calc_new_backup_super(bm_inode,
165					    group,
166					    cl_cpg, old_bg_clusters, 0);
167		le16_add_cpu(&group->bg_free_bits_count, backups);
168		le16_add_cpu(&group->bg_bits, -1 * num_bits);
169		le16_add_cpu(&group->bg_free_bits_count, -1 * num_bits);
170		group->bg_contig_free_bits = old_bg_contig_free_bits;
171	}
172out:
173	if (ret)
174		mlog_errno(ret);
175	return ret;
176}
177
178static int update_backups(struct inode * inode, u32 clusters, char *data)
179{
180	int i, ret = 0;
181	u32 cluster;
182	u64 blkno;
183	struct buffer_head *backup = NULL;
184	struct ocfs2_dinode *backup_di = NULL;
185	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
186
187	/* calculate the real backups we need to update. */
188	for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
189		blkno = ocfs2_backup_super_blkno(inode->i_sb, i);
190		cluster = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
191		if (cluster >= clusters)
192			break;
193
194		ret = ocfs2_read_blocks_sync(osb, blkno, 1, &backup);
195		if (ret < 0) {
196			mlog_errno(ret);
197			break;
198		}
199
200		memcpy(backup->b_data, data, inode->i_sb->s_blocksize);
201
202		backup_di = (struct ocfs2_dinode *)backup->b_data;
203		backup_di->i_blkno = cpu_to_le64(blkno);
204
205		ret = ocfs2_write_super_or_backup(osb, backup);
206		brelse(backup);
207		backup = NULL;
208		if (ret < 0) {
209			mlog_errno(ret);
210			break;
211		}
212	}
213
214	return ret;
215}
216
217static void ocfs2_update_super_and_backups(struct inode *inode,
218					   int new_clusters)
219{
220	int ret;
221	u32 clusters = 0;
222	struct buffer_head *super_bh = NULL;
223	struct ocfs2_dinode *super_di = NULL;
224	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
225
226	/*
227	 * update the superblock last.
228	 * It doesn't matter if the write failed.
229	 */
230	ret = ocfs2_read_blocks_sync(osb, OCFS2_SUPER_BLOCK_BLKNO, 1,
231				     &super_bh);
232	if (ret < 0) {
233		mlog_errno(ret);
234		goto out;
235	}
236
237	super_di = (struct ocfs2_dinode *)super_bh->b_data;
238	le32_add_cpu(&super_di->i_clusters, new_clusters);
239	clusters = le32_to_cpu(super_di->i_clusters);
240
241	ret = ocfs2_write_super_or_backup(osb, super_bh);
242	if (ret < 0) {
243		mlog_errno(ret);
244		goto out;
245	}
246
247	if (OCFS2_HAS_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_COMPAT_BACKUP_SB))
248		ret = update_backups(inode, clusters, super_bh->b_data);
249
250out:
251	brelse(super_bh);
252	if (ret)
253		printk(KERN_WARNING "ocfs2: Failed to update super blocks on %s"
254			" during fs resize. This condition is not fatal,"
255			" but fsck.ocfs2 should be run to fix it\n",
256			osb->dev_str);
257	return;
258}
259
260/*
261 * Extend the filesystem to the new number of clusters specified.  This entry
262 * point is only used to extend the current filesystem to the end of the last
263 * existing group.
264 */
265int ocfs2_group_extend(struct inode * inode, int new_clusters)
266{
267	int ret;
268	handle_t *handle;
269	struct buffer_head *main_bm_bh = NULL;
270	struct buffer_head *group_bh = NULL;
271	struct inode *main_bm_inode = NULL;
272	struct ocfs2_dinode *fe = NULL;
273	struct ocfs2_group_desc *group = NULL;
274	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
275	u16 cl_bpc;
276	u32 first_new_cluster;
277	u64 lgd_blkno;
278
279	if (unlikely(ocfs2_emergency_state(osb)))
280		return -EROFS;
281
282	if (new_clusters < 0)
283		return -EINVAL;
284	else if (new_clusters == 0)
285		return 0;
286
287	main_bm_inode = ocfs2_get_system_file_inode(osb,
288						    GLOBAL_BITMAP_SYSTEM_INODE,
289						    OCFS2_INVALID_SLOT);
290	if (!main_bm_inode) {
291		ret = -EINVAL;
292		mlog_errno(ret);
293		goto out;
294	}
295
296	inode_lock(main_bm_inode);
297
298	ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
299	if (ret < 0) {
300		mlog_errno(ret);
301		goto out_mutex;
302	}
303
304	fe = (struct ocfs2_dinode *)main_bm_bh->b_data;
305
306	/* JBD-managed buffers can bypass validation, so treat this as corruption. */
307	if (!OCFS2_IS_VALID_DINODE(fe)) {
308		ret = ocfs2_error(main_bm_inode->i_sb,
309				  "Invalid dinode #%llu\n",
310				  (unsigned long long)OCFS2_I(main_bm_inode)->ip_blkno);
311		goto out_unlock;
312	}
313
314	if (le16_to_cpu(fe->id2.i_chain.cl_cpg) !=
315		ocfs2_group_bitmap_size(osb->sb, 0,
316					osb->s_feature_incompat) * 8) {
317		mlog(ML_ERROR, "The disk is too old and small. "
318		     "Force to do offline resize.");
319		ret = -EINVAL;
320		goto out_unlock;
321	}
322
323	first_new_cluster = le32_to_cpu(fe->i_clusters);
324	lgd_blkno = ocfs2_which_cluster_group(main_bm_inode,
325					      first_new_cluster - 1);
326
327	ret = ocfs2_read_group_descriptor(main_bm_inode, fe, lgd_blkno,
328					  &group_bh);
329	if (ret < 0) {
330		mlog_errno(ret);
331		goto out_unlock;
332	}
333	group = (struct ocfs2_group_desc *)group_bh->b_data;
334
335	cl_bpc = le16_to_cpu(fe->id2.i_chain.cl_bpc);
336	if (le16_to_cpu(group->bg_bits) / cl_bpc + new_clusters >
337		le16_to_cpu(fe->id2.i_chain.cl_cpg)) {
338		ret = -EINVAL;
339		goto out_unlock;
340	}
341
342
343	trace_ocfs2_group_extend(
344	     (unsigned long long)le64_to_cpu(group->bg_blkno), new_clusters);
345
346	handle = ocfs2_start_trans(osb, OCFS2_GROUP_EXTEND_CREDITS);
347	if (IS_ERR(handle)) {
348		mlog_errno(PTR_ERR(handle));
349		ret = -EINVAL;
350		goto out_unlock;
351	}
352
353	/* update the last group descriptor and inode. */
354	ret = ocfs2_update_last_group_and_inode(handle, main_bm_inode,
355						main_bm_bh, group_bh,
356						first_new_cluster,
357						new_clusters);
358	if (ret) {
359		mlog_errno(ret);
360		goto out_commit;
361	}
362
363	ocfs2_update_super_and_backups(main_bm_inode, new_clusters);
364
365out_commit:
366	ocfs2_commit_trans(osb, handle);
367out_unlock:
368	brelse(group_bh);
369	brelse(main_bm_bh);
370
371	ocfs2_inode_unlock(main_bm_inode, 1);
372
373out_mutex:
374	inode_unlock(main_bm_inode);
375	iput(main_bm_inode);
376
377out:
378	return ret;
379}
380
381static int ocfs2_check_new_group(struct inode *inode,
382				 struct ocfs2_dinode *di,
383				 struct ocfs2_new_group_input *input,
384				 struct buffer_head *group_bh)
385{
386	int ret;
387	struct ocfs2_group_desc *gd =
388		(struct ocfs2_group_desc *)group_bh->b_data;
389	u16 cl_bpc = le16_to_cpu(di->id2.i_chain.cl_bpc);
390
391	ret = ocfs2_check_group_descriptor(inode->i_sb, di, group_bh);
392	if (ret)
393		goto out;
394
395	ret = -EINVAL;
396	if (le16_to_cpu(gd->bg_chain) != input->chain)
397		mlog(ML_ERROR, "Group descriptor # %llu has bad chain %u "
398		     "while input has %u set.\n",
399		     (unsigned long long)le64_to_cpu(gd->bg_blkno),
400		     le16_to_cpu(gd->bg_chain), input->chain);
401	else if (le16_to_cpu(gd->bg_bits) != input->clusters * cl_bpc)
402		mlog(ML_ERROR, "Group descriptor # %llu has bit count %u but "
403		     "input has %u clusters set\n",
404		     (unsigned long long)le64_to_cpu(gd->bg_blkno),
405		     le16_to_cpu(gd->bg_bits), input->clusters);
406	else if (le16_to_cpu(gd->bg_free_bits_count) != input->frees * cl_bpc)
407		mlog(ML_ERROR, "Group descriptor # %llu has free bit count %u "
408		     "but it should have %u set\n",
409		     (unsigned long long)le64_to_cpu(gd->bg_blkno),
410		     le16_to_cpu(gd->bg_bits),
411		     input->frees * cl_bpc);
412	else
413		ret = 0;
414
415out:
416	return ret;
417}
418
419static int ocfs2_verify_group_and_input(struct inode *inode,
420					struct ocfs2_dinode *di,
421					struct ocfs2_new_group_input *input,
422					struct buffer_head *group_bh)
423{
424	u16 cl_count = le16_to_cpu(di->id2.i_chain.cl_count);
425	u16 cl_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
426	u16 next_free = le16_to_cpu(di->id2.i_chain.cl_next_free_rec);
427	u32 cluster = ocfs2_blocks_to_clusters(inode->i_sb, input->group);
428	u32 total_clusters = le32_to_cpu(di->i_clusters);
429	int ret = -EINVAL;
430
431	if (cluster < total_clusters)
432		mlog(ML_ERROR, "add a group which is in the current volume.\n");
433	else if (input->chain >= cl_count)
434		mlog(ML_ERROR, "input chain exceeds the limit.\n");
435	else if (next_free != cl_count && next_free != input->chain)
436		mlog(ML_ERROR,
437		     "the add group should be in chain %u\n", next_free);
438	else if (total_clusters + input->clusters < total_clusters)
439		mlog(ML_ERROR, "add group's clusters overflow.\n");
440	else if (input->clusters > cl_cpg)
441		mlog(ML_ERROR, "the cluster exceeds the maximum of a group\n");
442	else if (input->frees > input->clusters)
443		mlog(ML_ERROR, "the free cluster exceeds the total clusters\n");
444	else if (total_clusters % cl_cpg != 0)
445		mlog(ML_ERROR,
446		     "the last group isn't full. Use group extend first.\n");
447	else if (input->group != ocfs2_which_cluster_group(inode, cluster))
448		mlog(ML_ERROR, "group blkno is invalid\n");
449	else if ((ret = ocfs2_check_new_group(inode, di, input, group_bh)))
450		mlog(ML_ERROR, "group descriptor check failed.\n");
451	else
452		ret = 0;
453
454	return ret;
455}
456
457/* Add a new group descriptor to global_bitmap. */
458int ocfs2_group_add(struct inode *inode, struct ocfs2_new_group_input *input)
459{
460	int ret;
461	handle_t *handle;
462	struct buffer_head *main_bm_bh = NULL;
463	struct inode *main_bm_inode = NULL;
464	struct ocfs2_dinode *fe = NULL;
465	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
466	struct buffer_head *group_bh = NULL;
467	struct ocfs2_group_desc *group = NULL;
468	struct ocfs2_chain_list *cl;
469	struct ocfs2_chain_rec *cr;
470	u16 cl_bpc;
471	u64 bg_ptr;
472
473	if (unlikely(ocfs2_emergency_state(osb)))
474		return -EROFS;
475
476	main_bm_inode = ocfs2_get_system_file_inode(osb,
477						    GLOBAL_BITMAP_SYSTEM_INODE,
478						    OCFS2_INVALID_SLOT);
479	if (!main_bm_inode) {
480		ret = -EINVAL;
481		mlog_errno(ret);
482		goto out;
483	}
484
485	inode_lock(main_bm_inode);
486
487	ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
488	if (ret < 0) {
489		mlog_errno(ret);
490		goto out_mutex;
491	}
492
493	fe = (struct ocfs2_dinode *)main_bm_bh->b_data;
494
495	if (le16_to_cpu(fe->id2.i_chain.cl_cpg) !=
496		ocfs2_group_bitmap_size(osb->sb, 0,
497					osb->s_feature_incompat) * 8) {
498		mlog(ML_ERROR, "The disk is too old and small."
499		     " Force to do offline resize.");
500		ret = -EINVAL;
501		goto out_unlock;
502	}
503
504	ret = ocfs2_read_blocks_sync(osb, input->group, 1, &group_bh);
505	if (ret < 0) {
506		mlog(ML_ERROR, "Can't read the group descriptor # %llu "
507		     "from the device.", (unsigned long long)input->group);
508		goto out_unlock;
509	}
510
511	ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), group_bh);
512
513	ret = ocfs2_verify_group_and_input(main_bm_inode, fe, input, group_bh);
514	if (ret) {
515		mlog_errno(ret);
516		goto out_free_group_bh;
517	}
518
519	trace_ocfs2_group_add((unsigned long long)input->group,
520			       input->chain, input->clusters, input->frees);
521
522	handle = ocfs2_start_trans(osb, OCFS2_GROUP_ADD_CREDITS);
523	if (IS_ERR(handle)) {
524		mlog_errno(PTR_ERR(handle));
525		ret = -EINVAL;
526		goto out_free_group_bh;
527	}
528
529	cl_bpc = le16_to_cpu(fe->id2.i_chain.cl_bpc);
530	cl = &fe->id2.i_chain;
531	cr = &cl->cl_recs[input->chain];
532
533	ret = ocfs2_journal_access_gd(handle, INODE_CACHE(main_bm_inode),
534				      group_bh, OCFS2_JOURNAL_ACCESS_WRITE);
535	if (ret < 0) {
536		mlog_errno(ret);
537		goto out_commit;
538	}
539
540	group = (struct ocfs2_group_desc *)group_bh->b_data;
541	bg_ptr = le64_to_cpu(group->bg_next_group);
542	group->bg_next_group = cr->c_blkno;
543	ocfs2_journal_dirty(handle, group_bh);
544
545	ret = ocfs2_journal_access_di(handle, INODE_CACHE(main_bm_inode),
546				      main_bm_bh, OCFS2_JOURNAL_ACCESS_WRITE);
547	if (ret < 0) {
548		group->bg_next_group = cpu_to_le64(bg_ptr);
549		mlog_errno(ret);
550		goto out_commit;
551	}
552
553	if (input->chain == le16_to_cpu(cl->cl_next_free_rec)) {
554		le16_add_cpu(&cl->cl_next_free_rec, 1);
555		memset(cr, 0, sizeof(struct ocfs2_chain_rec));
556	}
557
558	cr->c_blkno = cpu_to_le64(input->group);
559	le32_add_cpu(&cr->c_total, input->clusters * cl_bpc);
560	le32_add_cpu(&cr->c_free, input->frees * cl_bpc);
561
562	le32_add_cpu(&fe->id1.bitmap1.i_total, input->clusters *cl_bpc);
563	le32_add_cpu(&fe->id1.bitmap1.i_used,
564		     (input->clusters - input->frees) * cl_bpc);
565	le32_add_cpu(&fe->i_clusters, input->clusters);
566
567	ocfs2_journal_dirty(handle, main_bm_bh);
568
569	spin_lock(&OCFS2_I(main_bm_inode)->ip_lock);
570	OCFS2_I(main_bm_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
571	le64_add_cpu(&fe->i_size, (u64)input->clusters << osb->s_clustersize_bits);
572	spin_unlock(&OCFS2_I(main_bm_inode)->ip_lock);
573	i_size_write(main_bm_inode, le64_to_cpu(fe->i_size));
574
575	ocfs2_update_super_and_backups(main_bm_inode, input->clusters);
576
577out_commit:
578	ocfs2_commit_trans(osb, handle);
579
580out_free_group_bh:
581	if (ret < 0)
582		ocfs2_remove_from_cache(INODE_CACHE(inode), group_bh);
583	brelse(group_bh);
584
585out_unlock:
586	brelse(main_bm_bh);
587
588	ocfs2_inode_unlock(main_bm_inode, 1);
589
590out_mutex:
591	inode_unlock(main_bm_inode);
592	iput(main_bm_inode);
593
594out:
595	return ret;
596}