xref: /kernel/linux/linux-5.10/fs/ocfs2/suballoc.c (revision 8c2ecf20)
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
4 *
5 * suballoc.c
6 *
7 * metadata alloc and free
8 * Inspired by ext3 block groups.
9 *
10 * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
11 */
12
13#include <linux/fs.h>
14#include <linux/types.h>
15#include <linux/slab.h>
16#include <linux/highmem.h>
17
18#include <cluster/masklog.h>
19
20#include "ocfs2.h"
21
22#include "alloc.h"
23#include "blockcheck.h"
24#include "dlmglue.h"
25#include "inode.h"
26#include "journal.h"
27#include "localalloc.h"
28#include "suballoc.h"
29#include "super.h"
30#include "sysfile.h"
31#include "uptodate.h"
32#include "ocfs2_trace.h"
33
34#include "buffer_head_io.h"
35
36#define NOT_ALLOC_NEW_GROUP		0
37#define ALLOC_NEW_GROUP			0x1
38#define ALLOC_GROUPS_FROM_GLOBAL	0x2
39
40#define OCFS2_MAX_TO_STEAL		1024
41
42struct ocfs2_suballoc_result {
43	u64		sr_bg_blkno;	/* The bg we allocated from.  Set
44					   to 0 when a block group is
45					   contiguous. */
46	u64		sr_bg_stable_blkno; /*
47					     * Doesn't change, always
48					     * set to target block
49					     * group descriptor
50					     * block.
51					     */
52	u64		sr_blkno;	/* The first allocated block */
53	unsigned int	sr_bit_offset;	/* The bit in the bg */
54	unsigned int	sr_bits;	/* How many bits we claimed */
55};
56
57static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
58{
59	if (res->sr_blkno == 0)
60		return 0;
61
62	if (res->sr_bg_blkno)
63		return res->sr_bg_blkno;
64
65	return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset);
66}
67
68static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
69static int ocfs2_block_group_fill(handle_t *handle,
70				  struct inode *alloc_inode,
71				  struct buffer_head *bg_bh,
72				  u64 group_blkno,
73				  unsigned int group_clusters,
74				  u16 my_chain,
75				  struct ocfs2_chain_list *cl);
76static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
77				   struct inode *alloc_inode,
78				   struct buffer_head *bh,
79				   u64 max_block,
80				   u64 *last_alloc_group,
81				   int flags);
82
83static int ocfs2_cluster_group_search(struct inode *inode,
84				      struct buffer_head *group_bh,
85				      u32 bits_wanted, u32 min_bits,
86				      u64 max_block,
87				      struct ocfs2_suballoc_result *res);
88static int ocfs2_block_group_search(struct inode *inode,
89				    struct buffer_head *group_bh,
90				    u32 bits_wanted, u32 min_bits,
91				    u64 max_block,
92				    struct ocfs2_suballoc_result *res);
93static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
94				     handle_t *handle,
95				     u32 bits_wanted,
96				     u32 min_bits,
97				     struct ocfs2_suballoc_result *res);
98static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
99					 int nr);
100static int ocfs2_relink_block_group(handle_t *handle,
101				    struct inode *alloc_inode,
102				    struct buffer_head *fe_bh,
103				    struct buffer_head *bg_bh,
104				    struct buffer_head *prev_bg_bh,
105				    u16 chain);
106static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
107						     u32 wanted);
108static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
109						   u64 bg_blkno,
110						   u16 bg_bit_off);
111static inline void ocfs2_block_to_cluster_group(struct inode *inode,
112						u64 data_blkno,
113						u64 *bg_blkno,
114						u16 *bg_bit_off);
115static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
116					     u32 bits_wanted, u64 max_block,
117					     int flags,
118					     struct ocfs2_alloc_context **ac);
119
120void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
121{
122	struct inode *inode = ac->ac_inode;
123
124	if (inode) {
125		if (ac->ac_which != OCFS2_AC_USE_LOCAL)
126			ocfs2_inode_unlock(inode, 1);
127
128		inode_unlock(inode);
129
130		iput(inode);
131		ac->ac_inode = NULL;
132	}
133	brelse(ac->ac_bh);
134	ac->ac_bh = NULL;
135	ac->ac_resv = NULL;
136	kfree(ac->ac_find_loc_priv);
137	ac->ac_find_loc_priv = NULL;
138}
139
140void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
141{
142	ocfs2_free_ac_resource(ac);
143	kfree(ac);
144}
145
146static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
147{
148	return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
149}
150
151#define do_error(fmt, ...)						\
152do {									\
153	if (resize)							\
154		mlog(ML_ERROR, fmt, ##__VA_ARGS__);			\
155	else								\
156		return ocfs2_error(sb, fmt, ##__VA_ARGS__);		\
157} while (0)
158
159static int ocfs2_validate_gd_self(struct super_block *sb,
160				  struct buffer_head *bh,
161				  int resize)
162{
163	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
164
165	if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
166		do_error("Group descriptor #%llu has bad signature %.*s\n",
167			 (unsigned long long)bh->b_blocknr, 7,
168			 gd->bg_signature);
169	}
170
171	if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
172		do_error("Group descriptor #%llu has an invalid bg_blkno of %llu\n",
173			 (unsigned long long)bh->b_blocknr,
174			 (unsigned long long)le64_to_cpu(gd->bg_blkno));
175	}
176
177	if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
178		do_error("Group descriptor #%llu has an invalid fs_generation of #%u\n",
179			 (unsigned long long)bh->b_blocknr,
180			 le32_to_cpu(gd->bg_generation));
181	}
182
183	if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
184		do_error("Group descriptor #%llu has bit count %u but claims that %u are free\n",
185			 (unsigned long long)bh->b_blocknr,
186			 le16_to_cpu(gd->bg_bits),
187			 le16_to_cpu(gd->bg_free_bits_count));
188	}
189
190	if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
191		do_error("Group descriptor #%llu has bit count %u but max bitmap bits of %u\n",
192			 (unsigned long long)bh->b_blocknr,
193			 le16_to_cpu(gd->bg_bits),
194			 8 * le16_to_cpu(gd->bg_size));
195	}
196
197	return 0;
198}
199
200static int ocfs2_validate_gd_parent(struct super_block *sb,
201				    struct ocfs2_dinode *di,
202				    struct buffer_head *bh,
203				    int resize)
204{
205	unsigned int max_bits;
206	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
207
208	if (di->i_blkno != gd->bg_parent_dinode) {
209		do_error("Group descriptor #%llu has bad parent pointer (%llu, expected %llu)\n",
210			 (unsigned long long)bh->b_blocknr,
211			 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
212			 (unsigned long long)le64_to_cpu(di->i_blkno));
213	}
214
215	max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
216	if (le16_to_cpu(gd->bg_bits) > max_bits) {
217		do_error("Group descriptor #%llu has bit count of %u\n",
218			 (unsigned long long)bh->b_blocknr,
219			 le16_to_cpu(gd->bg_bits));
220	}
221
222	/* In resize, we may meet the case bg_chain == cl_next_free_rec. */
223	if ((le16_to_cpu(gd->bg_chain) >
224	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
225	    ((le16_to_cpu(gd->bg_chain) ==
226	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
227		do_error("Group descriptor #%llu has bad chain %u\n",
228			 (unsigned long long)bh->b_blocknr,
229			 le16_to_cpu(gd->bg_chain));
230	}
231
232	return 0;
233}
234
235#undef do_error
236
237/*
238 * This version only prints errors.  It does not fail the filesystem, and
239 * exists only for resize.
240 */
241int ocfs2_check_group_descriptor(struct super_block *sb,
242				 struct ocfs2_dinode *di,
243				 struct buffer_head *bh)
244{
245	int rc;
246	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
247
248	BUG_ON(!buffer_uptodate(bh));
249
250	/*
251	 * If the ecc fails, we return the error but otherwise
252	 * leave the filesystem running.  We know any error is
253	 * local to this block.
254	 */
255	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
256	if (rc) {
257		mlog(ML_ERROR,
258		     "Checksum failed for group descriptor %llu\n",
259		     (unsigned long long)bh->b_blocknr);
260	} else
261		rc = ocfs2_validate_gd_self(sb, bh, 1);
262	if (!rc)
263		rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
264
265	return rc;
266}
267
268static int ocfs2_validate_group_descriptor(struct super_block *sb,
269					   struct buffer_head *bh)
270{
271	int rc;
272	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
273
274	trace_ocfs2_validate_group_descriptor(
275					(unsigned long long)bh->b_blocknr);
276
277	BUG_ON(!buffer_uptodate(bh));
278
279	/*
280	 * If the ecc fails, we return the error but otherwise
281	 * leave the filesystem running.  We know any error is
282	 * local to this block.
283	 */
284	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
285	if (rc)
286		return rc;
287
288	/*
289	 * Errors after here are fatal.
290	 */
291
292	return ocfs2_validate_gd_self(sb, bh, 0);
293}
294
295int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
296				u64 gd_blkno, struct buffer_head **bh)
297{
298	int rc;
299	struct buffer_head *tmp = *bh;
300
301	rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
302			      ocfs2_validate_group_descriptor);
303	if (rc)
304		goto out;
305
306	rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
307	if (rc) {
308		brelse(tmp);
309		goto out;
310	}
311
312	/* If ocfs2_read_block() got us a new bh, pass it up. */
313	if (!*bh)
314		*bh = tmp;
315
316out:
317	return rc;
318}
319
320static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
321					  struct ocfs2_group_desc *bg,
322					  struct ocfs2_chain_list *cl,
323					  u64 p_blkno, unsigned int clusters)
324{
325	struct ocfs2_extent_list *el = &bg->bg_list;
326	struct ocfs2_extent_rec *rec;
327
328	BUG_ON(!ocfs2_supports_discontig_bg(osb));
329	if (!el->l_next_free_rec)
330		el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
331	rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
332	rec->e_blkno = cpu_to_le64(p_blkno);
333	rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
334				  le16_to_cpu(cl->cl_bpc));
335	rec->e_leaf_clusters = cpu_to_le16(clusters);
336	le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
337	le16_add_cpu(&bg->bg_free_bits_count,
338		     clusters * le16_to_cpu(cl->cl_bpc));
339	le16_add_cpu(&el->l_next_free_rec, 1);
340}
341
342static int ocfs2_block_group_fill(handle_t *handle,
343				  struct inode *alloc_inode,
344				  struct buffer_head *bg_bh,
345				  u64 group_blkno,
346				  unsigned int group_clusters,
347				  u16 my_chain,
348				  struct ocfs2_chain_list *cl)
349{
350	int status = 0;
351	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
352	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
353	struct super_block * sb = alloc_inode->i_sb;
354
355	if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
356		status = ocfs2_error(alloc_inode->i_sb,
357				     "group block (%llu) != b_blocknr (%llu)\n",
358				     (unsigned long long)group_blkno,
359				     (unsigned long long) bg_bh->b_blocknr);
360		goto bail;
361	}
362
363	status = ocfs2_journal_access_gd(handle,
364					 INODE_CACHE(alloc_inode),
365					 bg_bh,
366					 OCFS2_JOURNAL_ACCESS_CREATE);
367	if (status < 0) {
368		mlog_errno(status);
369		goto bail;
370	}
371
372	memset(bg, 0, sb->s_blocksize);
373	strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
374	bg->bg_generation = cpu_to_le32(osb->fs_generation);
375	bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
376						osb->s_feature_incompat));
377	bg->bg_chain = cpu_to_le16(my_chain);
378	bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
379	bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
380	bg->bg_blkno = cpu_to_le64(group_blkno);
381	if (group_clusters == le16_to_cpu(cl->cl_cpg))
382		bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
383	else
384		ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
385					      group_clusters);
386
387	/* set the 1st bit in the bitmap to account for the descriptor block */
388	ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
389	bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
390
391	ocfs2_journal_dirty(handle, bg_bh);
392
393	/* There is no need to zero out or otherwise initialize the
394	 * other blocks in a group - All valid FS metadata in a block
395	 * group stores the superblock fs_generation value at
396	 * allocation time. */
397
398bail:
399	if (status)
400		mlog_errno(status);
401	return status;
402}
403
404static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
405{
406	u16 curr, best;
407
408	best = curr = 0;
409	while (curr < le16_to_cpu(cl->cl_count)) {
410		if (le32_to_cpu(cl->cl_recs[best].c_total) >
411		    le32_to_cpu(cl->cl_recs[curr].c_total))
412			best = curr;
413		curr++;
414	}
415	return best;
416}
417
418static struct buffer_head *
419ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
420			       struct inode *alloc_inode,
421			       struct ocfs2_alloc_context *ac,
422			       struct ocfs2_chain_list *cl)
423{
424	int status;
425	u32 bit_off, num_bits;
426	u64 bg_blkno;
427	struct buffer_head *bg_bh;
428	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
429
430	status = ocfs2_claim_clusters(handle, ac,
431				      le16_to_cpu(cl->cl_cpg), &bit_off,
432				      &num_bits);
433	if (status < 0) {
434		if (status != -ENOSPC)
435			mlog_errno(status);
436		goto bail;
437	}
438
439	/* setup the group */
440	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
441	trace_ocfs2_block_group_alloc_contig(
442	     (unsigned long long)bg_blkno, alloc_rec);
443
444	bg_bh = sb_getblk(osb->sb, bg_blkno);
445	if (!bg_bh) {
446		status = -ENOMEM;
447		mlog_errno(status);
448		goto bail;
449	}
450	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
451
452	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
453					bg_blkno, num_bits, alloc_rec, cl);
454	if (status < 0) {
455		brelse(bg_bh);
456		mlog_errno(status);
457	}
458
459bail:
460	return status ? ERR_PTR(status) : bg_bh;
461}
462
463static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
464					handle_t *handle,
465					struct ocfs2_alloc_context *ac,
466					unsigned int min_bits,
467					u32 *bit_off, u32 *num_bits)
468{
469	int status = 0;
470
471	while (min_bits) {
472		status = ocfs2_claim_clusters(handle, ac, min_bits,
473					      bit_off, num_bits);
474		if (status != -ENOSPC)
475			break;
476
477		min_bits >>= 1;
478	}
479
480	return status;
481}
482
483static int ocfs2_block_group_grow_discontig(handle_t *handle,
484					    struct inode *alloc_inode,
485					    struct buffer_head *bg_bh,
486					    struct ocfs2_alloc_context *ac,
487					    struct ocfs2_chain_list *cl,
488					    unsigned int min_bits)
489{
490	int status;
491	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
492	struct ocfs2_group_desc *bg =
493		(struct ocfs2_group_desc *)bg_bh->b_data;
494	unsigned int needed = le16_to_cpu(cl->cl_cpg) -
495			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
496	u32 p_cpos, clusters;
497	u64 p_blkno;
498	struct ocfs2_extent_list *el = &bg->bg_list;
499
500	status = ocfs2_journal_access_gd(handle,
501					 INODE_CACHE(alloc_inode),
502					 bg_bh,
503					 OCFS2_JOURNAL_ACCESS_CREATE);
504	if (status < 0) {
505		mlog_errno(status);
506		goto bail;
507	}
508
509	while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
510				le16_to_cpu(el->l_count))) {
511		if (min_bits > needed)
512			min_bits = needed;
513		status = ocfs2_block_group_claim_bits(osb, handle, ac,
514						      min_bits, &p_cpos,
515						      &clusters);
516		if (status < 0) {
517			if (status != -ENOSPC)
518				mlog_errno(status);
519			goto bail;
520		}
521		p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
522		ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
523					      clusters);
524
525		min_bits = clusters;
526		needed = le16_to_cpu(cl->cl_cpg) -
527			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
528	}
529
530	if (needed > 0) {
531		/*
532		 * We have used up all the extent rec but can't fill up
533		 * the cpg. So bail out.
534		 */
535		status = -ENOSPC;
536		goto bail;
537	}
538
539	ocfs2_journal_dirty(handle, bg_bh);
540
541bail:
542	return status;
543}
544
545static void ocfs2_bg_alloc_cleanup(handle_t *handle,
546				   struct ocfs2_alloc_context *cluster_ac,
547				   struct inode *alloc_inode,
548				   struct buffer_head *bg_bh)
549{
550	int i, ret;
551	struct ocfs2_group_desc *bg;
552	struct ocfs2_extent_list *el;
553	struct ocfs2_extent_rec *rec;
554
555	if (!bg_bh)
556		return;
557
558	bg = (struct ocfs2_group_desc *)bg_bh->b_data;
559	el = &bg->bg_list;
560	for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
561		rec = &el->l_recs[i];
562		ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
563					  cluster_ac->ac_bh,
564					  le64_to_cpu(rec->e_blkno),
565					  le16_to_cpu(rec->e_leaf_clusters));
566		if (ret)
567			mlog_errno(ret);
568		/* Try all the clusters to free */
569	}
570
571	ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
572	brelse(bg_bh);
573}
574
575static struct buffer_head *
576ocfs2_block_group_alloc_discontig(handle_t *handle,
577				  struct inode *alloc_inode,
578				  struct ocfs2_alloc_context *ac,
579				  struct ocfs2_chain_list *cl)
580{
581	int status;
582	u32 bit_off, num_bits;
583	u64 bg_blkno;
584	unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
585	struct buffer_head *bg_bh = NULL;
586	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
587	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
588
589	if (!ocfs2_supports_discontig_bg(osb)) {
590		status = -ENOSPC;
591		goto bail;
592	}
593
594	status = ocfs2_extend_trans(handle,
595				    ocfs2_calc_bg_discontig_credits(osb->sb));
596	if (status) {
597		mlog_errno(status);
598		goto bail;
599	}
600
601	/*
602	 * We're going to be grabbing from multiple cluster groups.
603	 * We don't have enough credits to relink them all, and the
604	 * cluster groups will be staying in cache for the duration of
605	 * this operation.
606	 */
607	ac->ac_disable_chain_relink = 1;
608
609	/* Claim the first region */
610	status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
611					      &bit_off, &num_bits);
612	if (status < 0) {
613		if (status != -ENOSPC)
614			mlog_errno(status);
615		goto bail;
616	}
617	min_bits = num_bits;
618
619	/* setup the group */
620	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
621	trace_ocfs2_block_group_alloc_discontig(
622				(unsigned long long)bg_blkno, alloc_rec);
623
624	bg_bh = sb_getblk(osb->sb, bg_blkno);
625	if (!bg_bh) {
626		status = -ENOMEM;
627		mlog_errno(status);
628		goto bail;
629	}
630	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
631
632	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
633					bg_blkno, num_bits, alloc_rec, cl);
634	if (status < 0) {
635		mlog_errno(status);
636		goto bail;
637	}
638
639	status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
640						  bg_bh, ac, cl, min_bits);
641	if (status)
642		mlog_errno(status);
643
644bail:
645	if (status)
646		ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
647	return status ? ERR_PTR(status) : bg_bh;
648}
649
650/*
651 * We expect the block group allocator to already be locked.
652 */
653static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
654				   struct inode *alloc_inode,
655				   struct buffer_head *bh,
656				   u64 max_block,
657				   u64 *last_alloc_group,
658				   int flags)
659{
660	int status, credits;
661	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
662	struct ocfs2_chain_list *cl;
663	struct ocfs2_alloc_context *ac = NULL;
664	handle_t *handle = NULL;
665	u16 alloc_rec;
666	struct buffer_head *bg_bh = NULL;
667	struct ocfs2_group_desc *bg;
668
669	BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
670
671	cl = &fe->id2.i_chain;
672	status = ocfs2_reserve_clusters_with_limit(osb,
673						   le16_to_cpu(cl->cl_cpg),
674						   max_block, flags, &ac);
675	if (status < 0) {
676		if (status != -ENOSPC)
677			mlog_errno(status);
678		goto bail;
679	}
680
681	credits = ocfs2_calc_group_alloc_credits(osb->sb,
682						 le16_to_cpu(cl->cl_cpg));
683	handle = ocfs2_start_trans(osb, credits);
684	if (IS_ERR(handle)) {
685		status = PTR_ERR(handle);
686		handle = NULL;
687		mlog_errno(status);
688		goto bail;
689	}
690
691	if (last_alloc_group && *last_alloc_group != 0) {
692		trace_ocfs2_block_group_alloc(
693				(unsigned long long)*last_alloc_group);
694		ac->ac_last_group = *last_alloc_group;
695	}
696
697	bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
698					       ac, cl);
699	if (PTR_ERR(bg_bh) == -ENOSPC)
700		bg_bh = ocfs2_block_group_alloc_discontig(handle,
701							  alloc_inode,
702							  ac, cl);
703	if (IS_ERR(bg_bh)) {
704		status = PTR_ERR(bg_bh);
705		bg_bh = NULL;
706		if (status != -ENOSPC)
707			mlog_errno(status);
708		goto bail;
709	}
710	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
711
712	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
713					 bh, OCFS2_JOURNAL_ACCESS_WRITE);
714	if (status < 0) {
715		mlog_errno(status);
716		goto bail;
717	}
718
719	alloc_rec = le16_to_cpu(bg->bg_chain);
720	le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
721		     le16_to_cpu(bg->bg_free_bits_count));
722	le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
723		     le16_to_cpu(bg->bg_bits));
724	cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
725	if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
726		le16_add_cpu(&cl->cl_next_free_rec, 1);
727
728	le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
729					le16_to_cpu(bg->bg_free_bits_count));
730	le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
731	le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
732
733	ocfs2_journal_dirty(handle, bh);
734
735	spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
736	OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
737	fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
738					     le32_to_cpu(fe->i_clusters)));
739	spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
740	i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
741	alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
742	ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
743
744	status = 0;
745
746	/* save the new last alloc group so that the caller can cache it. */
747	if (last_alloc_group)
748		*last_alloc_group = ac->ac_last_group;
749
750bail:
751	if (handle)
752		ocfs2_commit_trans(osb, handle);
753
754	if (ac)
755		ocfs2_free_alloc_context(ac);
756
757	brelse(bg_bh);
758
759	if (status)
760		mlog_errno(status);
761	return status;
762}
763
764static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
765				       struct ocfs2_alloc_context *ac,
766				       int type,
767				       u32 slot,
768				       u64 *last_alloc_group,
769				       int flags)
770{
771	int status;
772	u32 bits_wanted = ac->ac_bits_wanted;
773	struct inode *alloc_inode;
774	struct buffer_head *bh = NULL;
775	struct ocfs2_dinode *fe;
776	u32 free_bits;
777
778	alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
779	if (!alloc_inode) {
780		mlog_errno(-EINVAL);
781		return -EINVAL;
782	}
783
784	inode_lock(alloc_inode);
785
786	status = ocfs2_inode_lock(alloc_inode, &bh, 1);
787	if (status < 0) {
788		inode_unlock(alloc_inode);
789		iput(alloc_inode);
790
791		mlog_errno(status);
792		return status;
793	}
794
795	ac->ac_inode = alloc_inode;
796	ac->ac_alloc_slot = slot;
797
798	fe = (struct ocfs2_dinode *) bh->b_data;
799
800	/* The bh was validated by the inode read inside
801	 * ocfs2_inode_lock().  Any corruption is a code bug. */
802	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
803
804	if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
805		status = ocfs2_error(alloc_inode->i_sb,
806				     "Invalid chain allocator %llu\n",
807				     (unsigned long long)le64_to_cpu(fe->i_blkno));
808		goto bail;
809	}
810
811	free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
812		le32_to_cpu(fe->id1.bitmap1.i_used);
813
814	if (bits_wanted > free_bits) {
815		/* cluster bitmap never grows */
816		if (ocfs2_is_cluster_bitmap(alloc_inode)) {
817			trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
818								free_bits);
819			status = -ENOSPC;
820			goto bail;
821		}
822
823		if (!(flags & ALLOC_NEW_GROUP)) {
824			trace_ocfs2_reserve_suballoc_bits_no_new_group(
825						slot, bits_wanted, free_bits);
826			status = -ENOSPC;
827			goto bail;
828		}
829
830		status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
831						 ac->ac_max_block,
832						 last_alloc_group, flags);
833		if (status < 0) {
834			if (status != -ENOSPC)
835				mlog_errno(status);
836			goto bail;
837		}
838		atomic_inc(&osb->alloc_stats.bg_extends);
839
840		/* You should never ask for this much metadata */
841		BUG_ON(bits_wanted >
842		       (le32_to_cpu(fe->id1.bitmap1.i_total)
843			- le32_to_cpu(fe->id1.bitmap1.i_used)));
844	}
845
846	get_bh(bh);
847	ac->ac_bh = bh;
848bail:
849	brelse(bh);
850
851	if (status)
852		mlog_errno(status);
853	return status;
854}
855
856static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
857{
858	spin_lock(&osb->osb_lock);
859	osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
860	spin_unlock(&osb->osb_lock);
861	atomic_set(&osb->s_num_inodes_stolen, 0);
862}
863
864static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
865{
866	spin_lock(&osb->osb_lock);
867	osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
868	spin_unlock(&osb->osb_lock);
869	atomic_set(&osb->s_num_meta_stolen, 0);
870}
871
872void ocfs2_init_steal_slots(struct ocfs2_super *osb)
873{
874	ocfs2_init_inode_steal_slot(osb);
875	ocfs2_init_meta_steal_slot(osb);
876}
877
878static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
879{
880	spin_lock(&osb->osb_lock);
881	if (type == INODE_ALLOC_SYSTEM_INODE)
882		osb->s_inode_steal_slot = (u16)slot;
883	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
884		osb->s_meta_steal_slot = (u16)slot;
885	spin_unlock(&osb->osb_lock);
886}
887
888static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
889{
890	int slot = OCFS2_INVALID_SLOT;
891
892	spin_lock(&osb->osb_lock);
893	if (type == INODE_ALLOC_SYSTEM_INODE)
894		slot = osb->s_inode_steal_slot;
895	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
896		slot = osb->s_meta_steal_slot;
897	spin_unlock(&osb->osb_lock);
898
899	return slot;
900}
901
902static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
903{
904	return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
905}
906
907static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
908{
909	return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
910}
911
912static int ocfs2_steal_resource(struct ocfs2_super *osb,
913				struct ocfs2_alloc_context *ac,
914				int type)
915{
916	int i, status = -ENOSPC;
917	int slot = __ocfs2_get_steal_slot(osb, type);
918
919	/* Start to steal resource from the first slot after ours. */
920	if (slot == OCFS2_INVALID_SLOT)
921		slot = osb->slot_num + 1;
922
923	for (i = 0; i < osb->max_slots; i++, slot++) {
924		if (slot == osb->max_slots)
925			slot = 0;
926
927		if (slot == osb->slot_num)
928			continue;
929
930		status = ocfs2_reserve_suballoc_bits(osb, ac,
931						     type,
932						     (u32)slot, NULL,
933						     NOT_ALLOC_NEW_GROUP);
934		if (status >= 0) {
935			__ocfs2_set_steal_slot(osb, slot, type);
936			break;
937		}
938
939		ocfs2_free_ac_resource(ac);
940	}
941
942	return status;
943}
944
945static int ocfs2_steal_inode(struct ocfs2_super *osb,
946			     struct ocfs2_alloc_context *ac)
947{
948	return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
949}
950
951static int ocfs2_steal_meta(struct ocfs2_super *osb,
952			    struct ocfs2_alloc_context *ac)
953{
954	return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
955}
956
957int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
958				      int blocks,
959				      struct ocfs2_alloc_context **ac)
960{
961	int status;
962	int slot = ocfs2_get_meta_steal_slot(osb);
963
964	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
965	if (!(*ac)) {
966		status = -ENOMEM;
967		mlog_errno(status);
968		goto bail;
969	}
970
971	(*ac)->ac_bits_wanted = blocks;
972	(*ac)->ac_which = OCFS2_AC_USE_META;
973	(*ac)->ac_group_search = ocfs2_block_group_search;
974
975	if (slot != OCFS2_INVALID_SLOT &&
976		atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
977		goto extent_steal;
978
979	atomic_set(&osb->s_num_meta_stolen, 0);
980	status = ocfs2_reserve_suballoc_bits(osb, (*ac),
981					     EXTENT_ALLOC_SYSTEM_INODE,
982					     (u32)osb->slot_num, NULL,
983					     ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
984
985
986	if (status >= 0) {
987		status = 0;
988		if (slot != OCFS2_INVALID_SLOT)
989			ocfs2_init_meta_steal_slot(osb);
990		goto bail;
991	} else if (status < 0 && status != -ENOSPC) {
992		mlog_errno(status);
993		goto bail;
994	}
995
996	ocfs2_free_ac_resource(*ac);
997
998extent_steal:
999	status = ocfs2_steal_meta(osb, *ac);
1000	atomic_inc(&osb->s_num_meta_stolen);
1001	if (status < 0) {
1002		if (status != -ENOSPC)
1003			mlog_errno(status);
1004		goto bail;
1005	}
1006
1007	status = 0;
1008bail:
1009	if ((status < 0) && *ac) {
1010		ocfs2_free_alloc_context(*ac);
1011		*ac = NULL;
1012	}
1013
1014	if (status)
1015		mlog_errno(status);
1016	return status;
1017}
1018
1019int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1020			       struct ocfs2_extent_list *root_el,
1021			       struct ocfs2_alloc_context **ac)
1022{
1023	return ocfs2_reserve_new_metadata_blocks(osb,
1024					ocfs2_extend_meta_needed(root_el),
1025					ac);
1026}
1027
1028int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
1029			    struct ocfs2_alloc_context **ac)
1030{
1031	int status;
1032	int slot = ocfs2_get_inode_steal_slot(osb);
1033	u64 alloc_group;
1034
1035	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1036	if (!(*ac)) {
1037		status = -ENOMEM;
1038		mlog_errno(status);
1039		goto bail;
1040	}
1041
1042	(*ac)->ac_bits_wanted = 1;
1043	(*ac)->ac_which = OCFS2_AC_USE_INODE;
1044
1045	(*ac)->ac_group_search = ocfs2_block_group_search;
1046
1047	/*
1048	 * stat(2) can't handle i_ino > 32bits, so we tell the
1049	 * lower levels not to allocate us a block group past that
1050	 * limit.  The 'inode64' mount option avoids this behavior.
1051	 */
1052	if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1053		(*ac)->ac_max_block = (u32)~0U;
1054
1055	/*
1056	 * slot is set when we successfully steal inode from other nodes.
1057	 * It is reset in 3 places:
1058	 * 1. when we flush the truncate log
1059	 * 2. when we complete local alloc recovery.
1060	 * 3. when we successfully allocate from our own slot.
1061	 * After it is set, we will go on stealing inodes until we find the
1062	 * need to check our slots to see whether there is some space for us.
1063	 */
1064	if (slot != OCFS2_INVALID_SLOT &&
1065	    atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
1066		goto inode_steal;
1067
1068	atomic_set(&osb->s_num_inodes_stolen, 0);
1069	alloc_group = osb->osb_inode_alloc_group;
1070	status = ocfs2_reserve_suballoc_bits(osb, *ac,
1071					     INODE_ALLOC_SYSTEM_INODE,
1072					     (u32)osb->slot_num,
1073					     &alloc_group,
1074					     ALLOC_NEW_GROUP |
1075					     ALLOC_GROUPS_FROM_GLOBAL);
1076	if (status >= 0) {
1077		status = 0;
1078
1079		spin_lock(&osb->osb_lock);
1080		osb->osb_inode_alloc_group = alloc_group;
1081		spin_unlock(&osb->osb_lock);
1082		trace_ocfs2_reserve_new_inode_new_group(
1083			(unsigned long long)alloc_group);
1084
1085		/*
1086		 * Some inodes must be freed by us, so try to allocate
1087		 * from our own next time.
1088		 */
1089		if (slot != OCFS2_INVALID_SLOT)
1090			ocfs2_init_inode_steal_slot(osb);
1091		goto bail;
1092	} else if (status < 0 && status != -ENOSPC) {
1093		mlog_errno(status);
1094		goto bail;
1095	}
1096
1097	ocfs2_free_ac_resource(*ac);
1098
1099inode_steal:
1100	status = ocfs2_steal_inode(osb, *ac);
1101	atomic_inc(&osb->s_num_inodes_stolen);
1102	if (status < 0) {
1103		if (status != -ENOSPC)
1104			mlog_errno(status);
1105		goto bail;
1106	}
1107
1108	status = 0;
1109bail:
1110	if ((status < 0) && *ac) {
1111		ocfs2_free_alloc_context(*ac);
1112		*ac = NULL;
1113	}
1114
1115	if (status)
1116		mlog_errno(status);
1117	return status;
1118}
1119
1120/* local alloc code has to do the same thing, so rather than do this
1121 * twice.. */
1122int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1123				      struct ocfs2_alloc_context *ac)
1124{
1125	int status;
1126
1127	ac->ac_which = OCFS2_AC_USE_MAIN;
1128	ac->ac_group_search = ocfs2_cluster_group_search;
1129
1130	status = ocfs2_reserve_suballoc_bits(osb, ac,
1131					     GLOBAL_BITMAP_SYSTEM_INODE,
1132					     OCFS2_INVALID_SLOT, NULL,
1133					     ALLOC_NEW_GROUP);
1134	if (status < 0 && status != -ENOSPC)
1135		mlog_errno(status);
1136
1137	return status;
1138}
1139
1140/* Callers don't need to care which bitmap (local alloc or main) to
1141 * use so we figure it out for them, but unfortunately this clutters
1142 * things a bit. */
1143static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1144					     u32 bits_wanted, u64 max_block,
1145					     int flags,
1146					     struct ocfs2_alloc_context **ac)
1147{
1148	int status, ret = 0;
1149	int retried = 0;
1150
1151	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1152	if (!(*ac)) {
1153		status = -ENOMEM;
1154		mlog_errno(status);
1155		goto bail;
1156	}
1157
1158	(*ac)->ac_bits_wanted = bits_wanted;
1159	(*ac)->ac_max_block = max_block;
1160
1161	status = -ENOSPC;
1162	if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1163	    ocfs2_alloc_should_use_local(osb, bits_wanted)) {
1164		status = ocfs2_reserve_local_alloc_bits(osb,
1165							bits_wanted,
1166							*ac);
1167		if ((status < 0) && (status != -ENOSPC)) {
1168			mlog_errno(status);
1169			goto bail;
1170		}
1171	}
1172
1173	if (status == -ENOSPC) {
1174retry:
1175		status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1176		/* Retry if there is sufficient space cached in truncate log */
1177		if (status == -ENOSPC && !retried) {
1178			retried = 1;
1179			ocfs2_inode_unlock((*ac)->ac_inode, 1);
1180			inode_unlock((*ac)->ac_inode);
1181
1182			ret = ocfs2_try_to_free_truncate_log(osb, bits_wanted);
1183			if (ret == 1) {
1184				iput((*ac)->ac_inode);
1185				(*ac)->ac_inode = NULL;
1186				goto retry;
1187			}
1188
1189			if (ret < 0)
1190				mlog_errno(ret);
1191
1192			inode_lock((*ac)->ac_inode);
1193			ret = ocfs2_inode_lock((*ac)->ac_inode, NULL, 1);
1194			if (ret < 0) {
1195				mlog_errno(ret);
1196				inode_unlock((*ac)->ac_inode);
1197				iput((*ac)->ac_inode);
1198				(*ac)->ac_inode = NULL;
1199				goto bail;
1200			}
1201		}
1202		if (status < 0) {
1203			if (status != -ENOSPC)
1204				mlog_errno(status);
1205			goto bail;
1206		}
1207	}
1208
1209	status = 0;
1210bail:
1211	if ((status < 0) && *ac) {
1212		ocfs2_free_alloc_context(*ac);
1213		*ac = NULL;
1214	}
1215
1216	if (status)
1217		mlog_errno(status);
1218	return status;
1219}
1220
1221int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1222			   u32 bits_wanted,
1223			   struct ocfs2_alloc_context **ac)
1224{
1225	return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1226						 ALLOC_NEW_GROUP, ac);
1227}
1228
1229/*
1230 * More or less lifted from ext3. I'll leave their description below:
1231 *
1232 * "For ext3 allocations, we must not reuse any blocks which are
1233 * allocated in the bitmap buffer's "last committed data" copy.  This
1234 * prevents deletes from freeing up the page for reuse until we have
1235 * committed the delete transaction.
1236 *
1237 * If we didn't do this, then deleting something and reallocating it as
1238 * data would allow the old block to be overwritten before the
1239 * transaction committed (because we force data to disk before commit).
1240 * This would lead to corruption if we crashed between overwriting the
1241 * data and committing the delete.
1242 *
1243 * @@@ We may want to make this allocation behaviour conditional on
1244 * data-writes at some point, and disable it for metadata allocations or
1245 * sync-data inodes."
1246 *
1247 * Note: OCFS2 already does this differently for metadata vs data
1248 * allocations, as those bitmaps are separate and undo access is never
1249 * called on a metadata group descriptor.
1250 */
1251static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1252					 int nr)
1253{
1254	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1255	struct journal_head *jh;
1256	int ret;
1257
1258	if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1259		return 0;
1260
1261	jh = jbd2_journal_grab_journal_head(bg_bh);
1262	if (!jh)
1263		return 1;
1264
1265	spin_lock(&jh->b_state_lock);
1266	bg = (struct ocfs2_group_desc *) jh->b_committed_data;
1267	if (bg)
1268		ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1269	else
1270		ret = 1;
1271	spin_unlock(&jh->b_state_lock);
1272	jbd2_journal_put_journal_head(jh);
1273
1274	return ret;
1275}
1276
1277static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1278					     struct buffer_head *bg_bh,
1279					     unsigned int bits_wanted,
1280					     unsigned int total_bits,
1281					     struct ocfs2_suballoc_result *res)
1282{
1283	void *bitmap;
1284	u16 best_offset, best_size;
1285	int offset, start, found, status = 0;
1286	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1287
1288	/* Callers got this descriptor from
1289	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1290	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1291
1292	found = start = best_offset = best_size = 0;
1293	bitmap = bg->bg_bitmap;
1294
1295	while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1296		if (offset == total_bits)
1297			break;
1298
1299		if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1300			/* We found a zero, but we can't use it as it
1301			 * hasn't been put to disk yet! */
1302			found = 0;
1303			start = offset + 1;
1304		} else if (offset == start) {
1305			/* we found a zero */
1306			found++;
1307			/* move start to the next bit to test */
1308			start++;
1309		} else {
1310			/* got a zero after some ones */
1311			found = 1;
1312			start = offset + 1;
1313		}
1314		if (found > best_size) {
1315			best_size = found;
1316			best_offset = start - found;
1317		}
1318		/* we got everything we needed */
1319		if (found == bits_wanted) {
1320			/* mlog(0, "Found it all!\n"); */
1321			break;
1322		}
1323	}
1324
1325	if (best_size) {
1326		res->sr_bit_offset = best_offset;
1327		res->sr_bits = best_size;
1328	} else {
1329		status = -ENOSPC;
1330		/* No error log here -- see the comment above
1331		 * ocfs2_test_bg_bit_allocatable */
1332	}
1333
1334	return status;
1335}
1336
1337int ocfs2_block_group_set_bits(handle_t *handle,
1338					     struct inode *alloc_inode,
1339					     struct ocfs2_group_desc *bg,
1340					     struct buffer_head *group_bh,
1341					     unsigned int bit_off,
1342					     unsigned int num_bits)
1343{
1344	int status;
1345	void *bitmap = bg->bg_bitmap;
1346	int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1347
1348	/* All callers get the descriptor via
1349	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1350	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1351	BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1352
1353	trace_ocfs2_block_group_set_bits(bit_off, num_bits);
1354
1355	if (ocfs2_is_cluster_bitmap(alloc_inode))
1356		journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1357
1358	status = ocfs2_journal_access_gd(handle,
1359					 INODE_CACHE(alloc_inode),
1360					 group_bh,
1361					 journal_type);
1362	if (status < 0) {
1363		mlog_errno(status);
1364		goto bail;
1365	}
1366
1367	le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1368	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1369		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
1370				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
1371				   le16_to_cpu(bg->bg_bits),
1372				   le16_to_cpu(bg->bg_free_bits_count),
1373				   num_bits);
1374	}
1375	while(num_bits--)
1376		ocfs2_set_bit(bit_off++, bitmap);
1377
1378	ocfs2_journal_dirty(handle, group_bh);
1379
1380bail:
1381	return status;
1382}
1383
1384/* find the one with the most empty bits */
1385static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1386{
1387	u16 curr, best;
1388
1389	BUG_ON(!cl->cl_next_free_rec);
1390
1391	best = curr = 0;
1392	while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1393		if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1394		    le32_to_cpu(cl->cl_recs[best].c_free))
1395			best = curr;
1396		curr++;
1397	}
1398
1399	BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1400	return best;
1401}
1402
1403static int ocfs2_relink_block_group(handle_t *handle,
1404				    struct inode *alloc_inode,
1405				    struct buffer_head *fe_bh,
1406				    struct buffer_head *bg_bh,
1407				    struct buffer_head *prev_bg_bh,
1408				    u16 chain)
1409{
1410	int status;
1411	/* there is a really tiny chance the journal calls could fail,
1412	 * but we wouldn't want inconsistent blocks in *any* case. */
1413	u64 bg_ptr, prev_bg_ptr;
1414	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1415	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1416	struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1417
1418	/* The caller got these descriptors from
1419	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1420	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1421	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1422
1423	trace_ocfs2_relink_block_group(
1424		(unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1425		(unsigned long long)le64_to_cpu(bg->bg_blkno),
1426		(unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1427
1428	bg_ptr = le64_to_cpu(bg->bg_next_group);
1429	prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1430
1431	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1432					 prev_bg_bh,
1433					 OCFS2_JOURNAL_ACCESS_WRITE);
1434	if (status < 0)
1435		goto out;
1436
1437	prev_bg->bg_next_group = bg->bg_next_group;
1438	ocfs2_journal_dirty(handle, prev_bg_bh);
1439
1440	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1441					 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1442	if (status < 0)
1443		goto out_rollback_prev_bg;
1444
1445	bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1446	ocfs2_journal_dirty(handle, bg_bh);
1447
1448	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1449					 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1450	if (status < 0)
1451		goto out_rollback_bg;
1452
1453	fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1454	ocfs2_journal_dirty(handle, fe_bh);
1455
1456out:
1457	if (status < 0)
1458		mlog_errno(status);
1459	return status;
1460
1461out_rollback_bg:
1462	bg->bg_next_group = cpu_to_le64(bg_ptr);
1463out_rollback_prev_bg:
1464	prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1465	goto out;
1466}
1467
1468static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1469						     u32 wanted)
1470{
1471	return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1472}
1473
1474/* return 0 on success, -ENOSPC to keep searching and any other < 0
1475 * value on error. */
1476static int ocfs2_cluster_group_search(struct inode *inode,
1477				      struct buffer_head *group_bh,
1478				      u32 bits_wanted, u32 min_bits,
1479				      u64 max_block,
1480				      struct ocfs2_suballoc_result *res)
1481{
1482	int search = -ENOSPC;
1483	int ret;
1484	u64 blkoff;
1485	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1486	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1487	unsigned int max_bits, gd_cluster_off;
1488
1489	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1490
1491	if (gd->bg_free_bits_count) {
1492		max_bits = le16_to_cpu(gd->bg_bits);
1493
1494		/* Tail groups in cluster bitmaps which aren't cpg
1495		 * aligned are prone to partial extension by a failed
1496		 * fs resize. If the file system resize never got to
1497		 * update the dinode cluster count, then we don't want
1498		 * to trust any clusters past it, regardless of what
1499		 * the group descriptor says. */
1500		gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1501							  le64_to_cpu(gd->bg_blkno));
1502		if ((gd_cluster_off + max_bits) >
1503		    OCFS2_I(inode)->ip_clusters) {
1504			max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1505			trace_ocfs2_cluster_group_search_wrong_max_bits(
1506				(unsigned long long)le64_to_cpu(gd->bg_blkno),
1507				le16_to_cpu(gd->bg_bits),
1508				OCFS2_I(inode)->ip_clusters, max_bits);
1509		}
1510
1511		ret = ocfs2_block_group_find_clear_bits(osb,
1512							group_bh, bits_wanted,
1513							max_bits, res);
1514		if (ret)
1515			return ret;
1516
1517		if (max_block) {
1518			blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1519							  gd_cluster_off +
1520							  res->sr_bit_offset +
1521							  res->sr_bits);
1522			trace_ocfs2_cluster_group_search_max_block(
1523				(unsigned long long)blkoff,
1524				(unsigned long long)max_block);
1525			if (blkoff > max_block)
1526				return -ENOSPC;
1527		}
1528
1529		/* ocfs2_block_group_find_clear_bits() might
1530		 * return success, but we still want to return
1531		 * -ENOSPC unless it found the minimum number
1532		 * of bits. */
1533		if (min_bits <= res->sr_bits)
1534			search = 0; /* success */
1535		else if (res->sr_bits) {
1536			/*
1537			 * Don't show bits which we'll be returning
1538			 * for allocation to the local alloc bitmap.
1539			 */
1540			ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
1541		}
1542	}
1543
1544	return search;
1545}
1546
1547static int ocfs2_block_group_search(struct inode *inode,
1548				    struct buffer_head *group_bh,
1549				    u32 bits_wanted, u32 min_bits,
1550				    u64 max_block,
1551				    struct ocfs2_suballoc_result *res)
1552{
1553	int ret = -ENOSPC;
1554	u64 blkoff;
1555	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1556
1557	BUG_ON(min_bits != 1);
1558	BUG_ON(ocfs2_is_cluster_bitmap(inode));
1559
1560	if (bg->bg_free_bits_count) {
1561		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1562							group_bh, bits_wanted,
1563							le16_to_cpu(bg->bg_bits),
1564							res);
1565		if (!ret && max_block) {
1566			blkoff = le64_to_cpu(bg->bg_blkno) +
1567				res->sr_bit_offset + res->sr_bits;
1568			trace_ocfs2_block_group_search_max_block(
1569				(unsigned long long)blkoff,
1570				(unsigned long long)max_block);
1571			if (blkoff > max_block)
1572				ret = -ENOSPC;
1573		}
1574	}
1575
1576	return ret;
1577}
1578
1579int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1580				       handle_t *handle,
1581				       struct buffer_head *di_bh,
1582				       u32 num_bits,
1583				       u16 chain)
1584{
1585	int ret;
1586	u32 tmp_used;
1587	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1588	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1589
1590	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1591				      OCFS2_JOURNAL_ACCESS_WRITE);
1592	if (ret < 0) {
1593		mlog_errno(ret);
1594		goto out;
1595	}
1596
1597	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1598	di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1599	le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1600	ocfs2_journal_dirty(handle, di_bh);
1601
1602out:
1603	return ret;
1604}
1605
1606void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
1607				       struct buffer_head *di_bh,
1608				       u32 num_bits,
1609				       u16 chain)
1610{
1611	u32 tmp_used;
1612	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1613	struct ocfs2_chain_list *cl;
1614
1615	cl = (struct ocfs2_chain_list *)&di->id2.i_chain;
1616	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1617	di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits);
1618	le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits);
1619}
1620
1621static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1622					 struct ocfs2_extent_rec *rec,
1623					 struct ocfs2_chain_list *cl)
1624{
1625	unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1626	unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1627	unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1628
1629	if (res->sr_bit_offset < bitoff)
1630		return 0;
1631	if (res->sr_bit_offset >= (bitoff + bitcount))
1632		return 0;
1633	res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1634		(res->sr_bit_offset - bitoff);
1635	if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1636		res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1637	return 1;
1638}
1639
1640static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1641					  struct ocfs2_group_desc *bg,
1642					  struct ocfs2_suballoc_result *res)
1643{
1644	int i;
1645	u64 bg_blkno = res->sr_bg_blkno;  /* Save off */
1646	struct ocfs2_extent_rec *rec;
1647	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1648	struct ocfs2_chain_list *cl = &di->id2.i_chain;
1649
1650	if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1651		res->sr_blkno = 0;
1652		return;
1653	}
1654
1655	res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1656	res->sr_bg_blkno = 0;  /* Clear it for contig block groups */
1657	if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1658	    !bg->bg_list.l_next_free_rec)
1659		return;
1660
1661	for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1662		rec = &bg->bg_list.l_recs[i];
1663		if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1664			res->sr_bg_blkno = bg_blkno;  /* Restore */
1665			break;
1666		}
1667	}
1668}
1669
1670static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1671				  handle_t *handle,
1672				  u32 bits_wanted,
1673				  u32 min_bits,
1674				  struct ocfs2_suballoc_result *res,
1675				  u16 *bits_left)
1676{
1677	int ret;
1678	struct buffer_head *group_bh = NULL;
1679	struct ocfs2_group_desc *gd;
1680	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1681	struct inode *alloc_inode = ac->ac_inode;
1682
1683	ret = ocfs2_read_group_descriptor(alloc_inode, di,
1684					  res->sr_bg_blkno, &group_bh);
1685	if (ret < 0) {
1686		mlog_errno(ret);
1687		return ret;
1688	}
1689
1690	gd = (struct ocfs2_group_desc *) group_bh->b_data;
1691	ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1692				  ac->ac_max_block, res);
1693	if (ret < 0) {
1694		if (ret != -ENOSPC)
1695			mlog_errno(ret);
1696		goto out;
1697	}
1698
1699	if (!ret)
1700		ocfs2_bg_discontig_fix_result(ac, gd, res);
1701
1702	/*
1703	 * sr_bg_blkno might have been changed by
1704	 * ocfs2_bg_discontig_fix_result
1705	 */
1706	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1707
1708	if (ac->ac_find_loc_only)
1709		goto out_loc_only;
1710
1711	ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1712					       res->sr_bits,
1713					       le16_to_cpu(gd->bg_chain));
1714	if (ret < 0) {
1715		mlog_errno(ret);
1716		goto out;
1717	}
1718
1719	ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1720					 res->sr_bit_offset, res->sr_bits);
1721	if (ret < 0) {
1722		ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh,
1723					       res->sr_bits,
1724					       le16_to_cpu(gd->bg_chain));
1725		mlog_errno(ret);
1726	}
1727
1728out_loc_only:
1729	*bits_left = le16_to_cpu(gd->bg_free_bits_count);
1730
1731out:
1732	brelse(group_bh);
1733
1734	return ret;
1735}
1736
1737static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1738			      handle_t *handle,
1739			      u32 bits_wanted,
1740			      u32 min_bits,
1741			      struct ocfs2_suballoc_result *res,
1742			      u16 *bits_left)
1743{
1744	int status;
1745	u16 chain;
1746	u64 next_group;
1747	struct inode *alloc_inode = ac->ac_inode;
1748	struct buffer_head *group_bh = NULL;
1749	struct buffer_head *prev_group_bh = NULL;
1750	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1751	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1752	struct ocfs2_group_desc *bg;
1753
1754	chain = ac->ac_chain;
1755	trace_ocfs2_search_chain_begin(
1756		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1757		bits_wanted, chain);
1758
1759	status = ocfs2_read_group_descriptor(alloc_inode, fe,
1760					     le64_to_cpu(cl->cl_recs[chain].c_blkno),
1761					     &group_bh);
1762	if (status < 0) {
1763		mlog_errno(status);
1764		goto bail;
1765	}
1766	bg = (struct ocfs2_group_desc *) group_bh->b_data;
1767
1768	status = -ENOSPC;
1769	/* for now, the chain search is a bit simplistic. We just use
1770	 * the 1st group with any empty bits. */
1771	while ((status = ac->ac_group_search(alloc_inode, group_bh,
1772					     bits_wanted, min_bits,
1773					     ac->ac_max_block,
1774					     res)) == -ENOSPC) {
1775		if (!bg->bg_next_group)
1776			break;
1777
1778		brelse(prev_group_bh);
1779		prev_group_bh = NULL;
1780
1781		next_group = le64_to_cpu(bg->bg_next_group);
1782		prev_group_bh = group_bh;
1783		group_bh = NULL;
1784		status = ocfs2_read_group_descriptor(alloc_inode, fe,
1785						     next_group, &group_bh);
1786		if (status < 0) {
1787			mlog_errno(status);
1788			goto bail;
1789		}
1790		bg = (struct ocfs2_group_desc *) group_bh->b_data;
1791	}
1792	if (status < 0) {
1793		if (status != -ENOSPC)
1794			mlog_errno(status);
1795		goto bail;
1796	}
1797
1798	trace_ocfs2_search_chain_succ(
1799		(unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1800
1801	res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1802
1803	BUG_ON(res->sr_bits == 0);
1804	if (!status)
1805		ocfs2_bg_discontig_fix_result(ac, bg, res);
1806
1807	/*
1808	 * sr_bg_blkno might have been changed by
1809	 * ocfs2_bg_discontig_fix_result
1810	 */
1811	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1812
1813	/*
1814	 * Keep track of previous block descriptor read. When
1815	 * we find a target, if we have read more than X
1816	 * number of descriptors, and the target is reasonably
1817	 * empty, relink him to top of his chain.
1818	 *
1819	 * We've read 0 extra blocks and only send one more to
1820	 * the transaction, yet the next guy to search has a
1821	 * much easier time.
1822	 *
1823	 * Do this *after* figuring out how many bits we're taking out
1824	 * of our target group.
1825	 */
1826	if (!ac->ac_disable_chain_relink &&
1827	    (prev_group_bh) &&
1828	    (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1829		status = ocfs2_relink_block_group(handle, alloc_inode,
1830						  ac->ac_bh, group_bh,
1831						  prev_group_bh, chain);
1832		if (status < 0) {
1833			mlog_errno(status);
1834			goto bail;
1835		}
1836	}
1837
1838	if (ac->ac_find_loc_only)
1839		goto out_loc_only;
1840
1841	status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1842						  ac->ac_bh, res->sr_bits,
1843						  chain);
1844	if (status) {
1845		mlog_errno(status);
1846		goto bail;
1847	}
1848
1849	status = ocfs2_block_group_set_bits(handle,
1850					    alloc_inode,
1851					    bg,
1852					    group_bh,
1853					    res->sr_bit_offset,
1854					    res->sr_bits);
1855	if (status < 0) {
1856		ocfs2_rollback_alloc_dinode_counts(alloc_inode,
1857					ac->ac_bh, res->sr_bits, chain);
1858		mlog_errno(status);
1859		goto bail;
1860	}
1861
1862	trace_ocfs2_search_chain_end(
1863			(unsigned long long)le64_to_cpu(fe->i_blkno),
1864			res->sr_bits);
1865
1866out_loc_only:
1867	*bits_left = le16_to_cpu(bg->bg_free_bits_count);
1868bail:
1869	brelse(group_bh);
1870	brelse(prev_group_bh);
1871
1872	if (status)
1873		mlog_errno(status);
1874	return status;
1875}
1876
1877/* will give out up to bits_wanted contiguous bits. */
1878static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1879				     handle_t *handle,
1880				     u32 bits_wanted,
1881				     u32 min_bits,
1882				     struct ocfs2_suballoc_result *res)
1883{
1884	int status;
1885	u16 victim, i;
1886	u16 bits_left = 0;
1887	u64 hint = ac->ac_last_group;
1888	struct ocfs2_chain_list *cl;
1889	struct ocfs2_dinode *fe;
1890
1891	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1892	BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1893	BUG_ON(!ac->ac_bh);
1894
1895	fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1896
1897	/* The bh was validated by the inode read during
1898	 * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1899	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1900
1901	if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1902	    le32_to_cpu(fe->id1.bitmap1.i_total)) {
1903		status = ocfs2_error(ac->ac_inode->i_sb,
1904				     "Chain allocator dinode %llu has %u used bits but only %u total\n",
1905				     (unsigned long long)le64_to_cpu(fe->i_blkno),
1906				     le32_to_cpu(fe->id1.bitmap1.i_used),
1907				     le32_to_cpu(fe->id1.bitmap1.i_total));
1908		goto bail;
1909	}
1910
1911	res->sr_bg_blkno = hint;
1912	if (res->sr_bg_blkno) {
1913		/* Attempt to short-circuit the usual search mechanism
1914		 * by jumping straight to the most recently used
1915		 * allocation group. This helps us maintain some
1916		 * contiguousness across allocations. */
1917		status = ocfs2_search_one_group(ac, handle, bits_wanted,
1918						min_bits, res, &bits_left);
1919		if (!status)
1920			goto set_hint;
1921		if (status < 0 && status != -ENOSPC) {
1922			mlog_errno(status);
1923			goto bail;
1924		}
1925	}
1926
1927	cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1928
1929	victim = ocfs2_find_victim_chain(cl);
1930	ac->ac_chain = victim;
1931
1932	status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1933				    res, &bits_left);
1934	if (!status) {
1935		if (ocfs2_is_cluster_bitmap(ac->ac_inode))
1936			hint = res->sr_bg_blkno;
1937		else
1938			hint = ocfs2_group_from_res(res);
1939		goto set_hint;
1940	}
1941	if (status < 0 && status != -ENOSPC) {
1942		mlog_errno(status);
1943		goto bail;
1944	}
1945
1946	trace_ocfs2_claim_suballoc_bits(victim);
1947
1948	/* If we didn't pick a good victim, then just default to
1949	 * searching each chain in order. Don't allow chain relinking
1950	 * because we only calculate enough journal credits for one
1951	 * relink per alloc. */
1952	ac->ac_disable_chain_relink = 1;
1953	for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1954		if (i == victim)
1955			continue;
1956		if (!cl->cl_recs[i].c_free)
1957			continue;
1958
1959		ac->ac_chain = i;
1960		status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1961					    res, &bits_left);
1962		if (!status) {
1963			hint = ocfs2_group_from_res(res);
1964			break;
1965		}
1966		if (status < 0 && status != -ENOSPC) {
1967			mlog_errno(status);
1968			goto bail;
1969		}
1970	}
1971
1972set_hint:
1973	if (status != -ENOSPC) {
1974		/* If the next search of this group is not likely to
1975		 * yield a suitable extent, then we reset the last
1976		 * group hint so as to not waste a disk read */
1977		if (bits_left < min_bits)
1978			ac->ac_last_group = 0;
1979		else
1980			ac->ac_last_group = hint;
1981	}
1982
1983bail:
1984	if (status)
1985		mlog_errno(status);
1986	return status;
1987}
1988
1989int ocfs2_claim_metadata(handle_t *handle,
1990			 struct ocfs2_alloc_context *ac,
1991			 u32 bits_wanted,
1992			 u64 *suballoc_loc,
1993			 u16 *suballoc_bit_start,
1994			 unsigned int *num_bits,
1995			 u64 *blkno_start)
1996{
1997	int status;
1998	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
1999
2000	BUG_ON(!ac);
2001	BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
2002	BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
2003
2004	status = ocfs2_claim_suballoc_bits(ac,
2005					   handle,
2006					   bits_wanted,
2007					   1,
2008					   &res);
2009	if (status < 0) {
2010		mlog_errno(status);
2011		goto bail;
2012	}
2013	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2014
2015	*suballoc_loc = res.sr_bg_blkno;
2016	*suballoc_bit_start = res.sr_bit_offset;
2017	*blkno_start = res.sr_blkno;
2018	ac->ac_bits_given += res.sr_bits;
2019	*num_bits = res.sr_bits;
2020	status = 0;
2021bail:
2022	if (status)
2023		mlog_errno(status);
2024	return status;
2025}
2026
2027static void ocfs2_init_inode_ac_group(struct inode *dir,
2028				      struct buffer_head *parent_di_bh,
2029				      struct ocfs2_alloc_context *ac)
2030{
2031	struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2032	/*
2033	 * Try to allocate inodes from some specific group.
2034	 *
2035	 * If the parent dir has recorded the last group used in allocation,
2036	 * cool, use it. Otherwise if we try to allocate new inode from the
2037	 * same slot the parent dir belongs to, use the same chunk.
2038	 *
2039	 * We are very careful here to avoid the mistake of setting
2040	 * ac_last_group to a group descriptor from a different (unlocked) slot.
2041	 */
2042	if (OCFS2_I(dir)->ip_last_used_group &&
2043	    OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2044		ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2045	else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2046		if (di->i_suballoc_loc)
2047			ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2048		else
2049			ac->ac_last_group = ocfs2_which_suballoc_group(
2050					le64_to_cpu(di->i_blkno),
2051					le16_to_cpu(di->i_suballoc_bit));
2052	}
2053}
2054
2055static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2056					     struct ocfs2_alloc_context *ac)
2057{
2058	OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2059	OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2060}
2061
2062int ocfs2_find_new_inode_loc(struct inode *dir,
2063			     struct buffer_head *parent_fe_bh,
2064			     struct ocfs2_alloc_context *ac,
2065			     u64 *fe_blkno)
2066{
2067	int ret;
2068	handle_t *handle = NULL;
2069	struct ocfs2_suballoc_result *res;
2070
2071	BUG_ON(!ac);
2072	BUG_ON(ac->ac_bits_given != 0);
2073	BUG_ON(ac->ac_bits_wanted != 1);
2074	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2075
2076	res = kzalloc(sizeof(*res), GFP_NOFS);
2077	if (res == NULL) {
2078		ret = -ENOMEM;
2079		mlog_errno(ret);
2080		goto out;
2081	}
2082
2083	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2084
2085	/*
2086	 * The handle started here is for chain relink. Alternatively,
2087	 * we could just disable relink for these calls.
2088	 */
2089	handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2090	if (IS_ERR(handle)) {
2091		ret = PTR_ERR(handle);
2092		handle = NULL;
2093		mlog_errno(ret);
2094		goto out;
2095	}
2096
2097	/*
2098	 * This will instruct ocfs2_claim_suballoc_bits and
2099	 * ocfs2_search_one_group to search but save actual allocation
2100	 * for later.
2101	 */
2102	ac->ac_find_loc_only = 1;
2103
2104	ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2105	if (ret < 0) {
2106		mlog_errno(ret);
2107		goto out;
2108	}
2109
2110	ac->ac_find_loc_priv = res;
2111	*fe_blkno = res->sr_blkno;
2112	ocfs2_update_inode_fsync_trans(handle, dir, 0);
2113out:
2114	if (handle)
2115		ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2116
2117	if (ret)
2118		kfree(res);
2119
2120	return ret;
2121}
2122
2123int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2124				 struct inode *dir,
2125				 struct ocfs2_alloc_context *ac,
2126				 u64 *suballoc_loc,
2127				 u16 *suballoc_bit,
2128				 u64 di_blkno)
2129{
2130	int ret;
2131	u16 chain;
2132	struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2133	struct buffer_head *bg_bh = NULL;
2134	struct ocfs2_group_desc *bg;
2135	struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2136
2137	/*
2138	 * Since di_blkno is being passed back in, we check for any
2139	 * inconsistencies which may have happened between
2140	 * calls. These are code bugs as di_blkno is not expected to
2141	 * change once returned from ocfs2_find_new_inode_loc()
2142	 */
2143	BUG_ON(res->sr_blkno != di_blkno);
2144
2145	ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2146					  res->sr_bg_stable_blkno, &bg_bh);
2147	if (ret) {
2148		mlog_errno(ret);
2149		goto out;
2150	}
2151
2152	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2153	chain = le16_to_cpu(bg->bg_chain);
2154
2155	ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2156					       ac->ac_bh, res->sr_bits,
2157					       chain);
2158	if (ret) {
2159		mlog_errno(ret);
2160		goto out;
2161	}
2162
2163	ret = ocfs2_block_group_set_bits(handle,
2164					 ac->ac_inode,
2165					 bg,
2166					 bg_bh,
2167					 res->sr_bit_offset,
2168					 res->sr_bits);
2169	if (ret < 0) {
2170		ocfs2_rollback_alloc_dinode_counts(ac->ac_inode,
2171					       ac->ac_bh, res->sr_bits, chain);
2172		mlog_errno(ret);
2173		goto out;
2174	}
2175
2176	trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2177					   res->sr_bits);
2178
2179	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2180
2181	BUG_ON(res->sr_bits != 1);
2182
2183	*suballoc_loc = res->sr_bg_blkno;
2184	*suballoc_bit = res->sr_bit_offset;
2185	ac->ac_bits_given++;
2186	ocfs2_save_inode_ac_group(dir, ac);
2187
2188out:
2189	brelse(bg_bh);
2190
2191	return ret;
2192}
2193
2194int ocfs2_claim_new_inode(handle_t *handle,
2195			  struct inode *dir,
2196			  struct buffer_head *parent_fe_bh,
2197			  struct ocfs2_alloc_context *ac,
2198			  u64 *suballoc_loc,
2199			  u16 *suballoc_bit,
2200			  u64 *fe_blkno)
2201{
2202	int status;
2203	struct ocfs2_suballoc_result res;
2204
2205	BUG_ON(!ac);
2206	BUG_ON(ac->ac_bits_given != 0);
2207	BUG_ON(ac->ac_bits_wanted != 1);
2208	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2209
2210	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2211
2212	status = ocfs2_claim_suballoc_bits(ac,
2213					   handle,
2214					   1,
2215					   1,
2216					   &res);
2217	if (status < 0) {
2218		mlog_errno(status);
2219		goto bail;
2220	}
2221	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2222
2223	BUG_ON(res.sr_bits != 1);
2224
2225	*suballoc_loc = res.sr_bg_blkno;
2226	*suballoc_bit = res.sr_bit_offset;
2227	*fe_blkno = res.sr_blkno;
2228	ac->ac_bits_given++;
2229	ocfs2_save_inode_ac_group(dir, ac);
2230	status = 0;
2231bail:
2232	if (status)
2233		mlog_errno(status);
2234	return status;
2235}
2236
2237/* translate a group desc. blkno and it's bitmap offset into
2238 * disk cluster offset. */
2239static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2240						   u64 bg_blkno,
2241						   u16 bg_bit_off)
2242{
2243	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2244	u32 cluster = 0;
2245
2246	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2247
2248	if (bg_blkno != osb->first_cluster_group_blkno)
2249		cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2250	cluster += (u32) bg_bit_off;
2251	return cluster;
2252}
2253
2254/* given a cluster offset, calculate which block group it belongs to
2255 * and return that block offset. */
2256u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2257{
2258	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2259	u32 group_no;
2260
2261	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2262
2263	group_no = cluster / osb->bitmap_cpg;
2264	if (!group_no)
2265		return osb->first_cluster_group_blkno;
2266	return ocfs2_clusters_to_blocks(inode->i_sb,
2267					group_no * osb->bitmap_cpg);
2268}
2269
2270/* given the block number of a cluster start, calculate which cluster
2271 * group and descriptor bitmap offset that corresponds to. */
2272static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2273						u64 data_blkno,
2274						u64 *bg_blkno,
2275						u16 *bg_bit_off)
2276{
2277	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2278	u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2279
2280	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2281
2282	*bg_blkno = ocfs2_which_cluster_group(inode,
2283					      data_cluster);
2284
2285	if (*bg_blkno == osb->first_cluster_group_blkno)
2286		*bg_bit_off = (u16) data_cluster;
2287	else
2288		*bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2289							     data_blkno - *bg_blkno);
2290}
2291
2292/*
2293 * min_bits - minimum contiguous chunk from this total allocation we
2294 * can handle. set to what we asked for originally for a full
2295 * contig. allocation, set to '1' to indicate we can deal with extents
2296 * of any size.
2297 */
2298int __ocfs2_claim_clusters(handle_t *handle,
2299			   struct ocfs2_alloc_context *ac,
2300			   u32 min_clusters,
2301			   u32 max_clusters,
2302			   u32 *cluster_start,
2303			   u32 *num_clusters)
2304{
2305	int status;
2306	unsigned int bits_wanted = max_clusters;
2307	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2308	struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2309
2310	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2311
2312	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2313	       && ac->ac_which != OCFS2_AC_USE_MAIN);
2314
2315	if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2316		WARN_ON(min_clusters > 1);
2317
2318		status = ocfs2_claim_local_alloc_bits(osb,
2319						      handle,
2320						      ac,
2321						      bits_wanted,
2322						      cluster_start,
2323						      num_clusters);
2324		if (!status)
2325			atomic_inc(&osb->alloc_stats.local_data);
2326	} else {
2327		if (min_clusters > (osb->bitmap_cpg - 1)) {
2328			/* The only paths asking for contiguousness
2329			 * should know about this already. */
2330			mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2331			     "group bitmap size %u!\n", min_clusters,
2332			     osb->bitmap_cpg);
2333			status = -ENOSPC;
2334			goto bail;
2335		}
2336		/* clamp the current request down to a realistic size. */
2337		if (bits_wanted > (osb->bitmap_cpg - 1))
2338			bits_wanted = osb->bitmap_cpg - 1;
2339
2340		status = ocfs2_claim_suballoc_bits(ac,
2341						   handle,
2342						   bits_wanted,
2343						   min_clusters,
2344						   &res);
2345		if (!status) {
2346			BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2347			*cluster_start =
2348				ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2349								 res.sr_bg_blkno,
2350								 res.sr_bit_offset);
2351			atomic_inc(&osb->alloc_stats.bitmap_data);
2352			*num_clusters = res.sr_bits;
2353		}
2354	}
2355	if (status < 0) {
2356		if (status != -ENOSPC)
2357			mlog_errno(status);
2358		goto bail;
2359	}
2360
2361	ac->ac_bits_given += *num_clusters;
2362
2363bail:
2364	if (status)
2365		mlog_errno(status);
2366	return status;
2367}
2368
2369int ocfs2_claim_clusters(handle_t *handle,
2370			 struct ocfs2_alloc_context *ac,
2371			 u32 min_clusters,
2372			 u32 *cluster_start,
2373			 u32 *num_clusters)
2374{
2375	unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2376
2377	return __ocfs2_claim_clusters(handle, ac, min_clusters,
2378				      bits_wanted, cluster_start, num_clusters);
2379}
2380
2381static int ocfs2_block_group_clear_bits(handle_t *handle,
2382					struct inode *alloc_inode,
2383					struct ocfs2_group_desc *bg,
2384					struct buffer_head *group_bh,
2385					unsigned int bit_off,
2386					unsigned int num_bits,
2387					void (*undo_fn)(unsigned int bit,
2388							unsigned long *bmap))
2389{
2390	int status;
2391	unsigned int tmp;
2392	struct ocfs2_group_desc *undo_bg = NULL;
2393	struct journal_head *jh;
2394
2395	/* The caller got this descriptor from
2396	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
2397	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2398
2399	trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2400
2401	BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2402	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2403					 group_bh,
2404					 undo_fn ?
2405					 OCFS2_JOURNAL_ACCESS_UNDO :
2406					 OCFS2_JOURNAL_ACCESS_WRITE);
2407	if (status < 0) {
2408		mlog_errno(status);
2409		goto bail;
2410	}
2411
2412	jh = bh2jh(group_bh);
2413	if (undo_fn) {
2414		spin_lock(&jh->b_state_lock);
2415		undo_bg = (struct ocfs2_group_desc *) jh->b_committed_data;
2416		BUG_ON(!undo_bg);
2417	}
2418
2419	tmp = num_bits;
2420	while(tmp--) {
2421		ocfs2_clear_bit((bit_off + tmp),
2422				(unsigned long *) bg->bg_bitmap);
2423		if (undo_fn)
2424			undo_fn(bit_off + tmp,
2425				(unsigned long *) undo_bg->bg_bitmap);
2426	}
2427	le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2428	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2429		if (undo_fn)
2430			spin_unlock(&jh->b_state_lock);
2431		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
2432				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
2433				   le16_to_cpu(bg->bg_bits),
2434				   le16_to_cpu(bg->bg_free_bits_count),
2435				   num_bits);
2436	}
2437
2438	if (undo_fn)
2439		spin_unlock(&jh->b_state_lock);
2440
2441	ocfs2_journal_dirty(handle, group_bh);
2442bail:
2443	return status;
2444}
2445
2446/*
2447 * expects the suballoc inode to already be locked.
2448 */
2449static int _ocfs2_free_suballoc_bits(handle_t *handle,
2450				     struct inode *alloc_inode,
2451				     struct buffer_head *alloc_bh,
2452				     unsigned int start_bit,
2453				     u64 bg_blkno,
2454				     unsigned int count,
2455				     void (*undo_fn)(unsigned int bit,
2456						     unsigned long *bitmap))
2457{
2458	int status = 0;
2459	u32 tmp_used;
2460	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2461	struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2462	struct buffer_head *group_bh = NULL;
2463	struct ocfs2_group_desc *group;
2464
2465	/* The alloc_bh comes from ocfs2_free_dinode() or
2466	 * ocfs2_free_clusters().  The callers have all locked the
2467	 * allocator and gotten alloc_bh from the lock call.  This
2468	 * validates the dinode buffer.  Any corruption that has happened
2469	 * is a code bug. */
2470	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2471	BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2472
2473	trace_ocfs2_free_suballoc_bits(
2474		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2475		(unsigned long long)bg_blkno,
2476		start_bit, count);
2477
2478	status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2479					     &group_bh);
2480	if (status < 0) {
2481		mlog_errno(status);
2482		goto bail;
2483	}
2484	group = (struct ocfs2_group_desc *) group_bh->b_data;
2485
2486	BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2487
2488	status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2489					      group, group_bh,
2490					      start_bit, count, undo_fn);
2491	if (status < 0) {
2492		mlog_errno(status);
2493		goto bail;
2494	}
2495
2496	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2497					 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2498	if (status < 0) {
2499		mlog_errno(status);
2500		ocfs2_block_group_set_bits(handle, alloc_inode, group, group_bh,
2501				start_bit, count);
2502		goto bail;
2503	}
2504
2505	le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2506		     count);
2507	tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2508	fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2509	ocfs2_journal_dirty(handle, alloc_bh);
2510
2511bail:
2512	brelse(group_bh);
2513	return status;
2514}
2515
2516int ocfs2_free_suballoc_bits(handle_t *handle,
2517			     struct inode *alloc_inode,
2518			     struct buffer_head *alloc_bh,
2519			     unsigned int start_bit,
2520			     u64 bg_blkno,
2521			     unsigned int count)
2522{
2523	return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2524					 start_bit, bg_blkno, count, NULL);
2525}
2526
2527int ocfs2_free_dinode(handle_t *handle,
2528		      struct inode *inode_alloc_inode,
2529		      struct buffer_head *inode_alloc_bh,
2530		      struct ocfs2_dinode *di)
2531{
2532	u64 blk = le64_to_cpu(di->i_blkno);
2533	u16 bit = le16_to_cpu(di->i_suballoc_bit);
2534	u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2535
2536	if (di->i_suballoc_loc)
2537		bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2538	return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2539					inode_alloc_bh, bit, bg_blkno, 1);
2540}
2541
2542static int _ocfs2_free_clusters(handle_t *handle,
2543				struct inode *bitmap_inode,
2544				struct buffer_head *bitmap_bh,
2545				u64 start_blk,
2546				unsigned int num_clusters,
2547				void (*undo_fn)(unsigned int bit,
2548						unsigned long *bitmap))
2549{
2550	int status;
2551	u16 bg_start_bit;
2552	u64 bg_blkno;
2553
2554	/* You can't ever have a contiguous set of clusters
2555	 * bigger than a block group bitmap so we never have to worry
2556	 * about looping on them.
2557	 * This is expensive. We can safely remove once this stuff has
2558	 * gotten tested really well. */
2559	BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb,
2560				ocfs2_blocks_to_clusters(bitmap_inode->i_sb,
2561							 start_blk)));
2562
2563
2564	ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2565				     &bg_start_bit);
2566
2567	trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2568			(unsigned long long)start_blk,
2569			bg_start_bit, num_clusters);
2570
2571	status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2572					   bg_start_bit, bg_blkno,
2573					   num_clusters, undo_fn);
2574	if (status < 0) {
2575		mlog_errno(status);
2576		goto out;
2577	}
2578
2579	ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2580					 num_clusters);
2581
2582out:
2583	return status;
2584}
2585
2586int ocfs2_free_clusters(handle_t *handle,
2587			struct inode *bitmap_inode,
2588			struct buffer_head *bitmap_bh,
2589			u64 start_blk,
2590			unsigned int num_clusters)
2591{
2592	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2593				    start_blk, num_clusters,
2594				    _ocfs2_set_bit);
2595}
2596
2597/*
2598 * Give never-used clusters back to the global bitmap.  We don't need
2599 * to protect these bits in the undo buffer.
2600 */
2601int ocfs2_release_clusters(handle_t *handle,
2602			   struct inode *bitmap_inode,
2603			   struct buffer_head *bitmap_bh,
2604			   u64 start_blk,
2605			   unsigned int num_clusters)
2606{
2607	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2608				    start_blk, num_clusters,
2609				    _ocfs2_clear_bit);
2610}
2611
2612/*
2613 * For a given allocation, determine which allocators will need to be
2614 * accessed, and lock them, reserving the appropriate number of bits.
2615 *
2616 * Sparse file systems call this from ocfs2_write_begin_nolock()
2617 * and ocfs2_allocate_unwritten_extents().
2618 *
2619 * File systems which don't support holes call this from
2620 * ocfs2_extend_allocation().
2621 */
2622int ocfs2_lock_allocators(struct inode *inode,
2623			  struct ocfs2_extent_tree *et,
2624			  u32 clusters_to_add, u32 extents_to_split,
2625			  struct ocfs2_alloc_context **data_ac,
2626			  struct ocfs2_alloc_context **meta_ac)
2627{
2628	int ret = 0, num_free_extents;
2629	unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2630	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2631
2632	*meta_ac = NULL;
2633	if (data_ac)
2634		*data_ac = NULL;
2635
2636	BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2637
2638	num_free_extents = ocfs2_num_free_extents(et);
2639	if (num_free_extents < 0) {
2640		ret = num_free_extents;
2641		mlog_errno(ret);
2642		goto out;
2643	}
2644
2645	/*
2646	 * Sparse allocation file systems need to be more conservative
2647	 * with reserving room for expansion - the actual allocation
2648	 * happens while we've got a journal handle open so re-taking
2649	 * a cluster lock (because we ran out of room for another
2650	 * extent) will violate ordering rules.
2651	 *
2652	 * Most of the time we'll only be seeing this 1 cluster at a time
2653	 * anyway.
2654	 *
2655	 * Always lock for any unwritten extents - we might want to
2656	 * add blocks during a split.
2657	 */
2658	if (!num_free_extents ||
2659	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2660		ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2661		if (ret < 0) {
2662			if (ret != -ENOSPC)
2663				mlog_errno(ret);
2664			goto out;
2665		}
2666	}
2667
2668	if (clusters_to_add == 0)
2669		goto out;
2670
2671	ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2672	if (ret < 0) {
2673		if (ret != -ENOSPC)
2674			mlog_errno(ret);
2675		goto out;
2676	}
2677
2678out:
2679	if (ret) {
2680		if (*meta_ac) {
2681			ocfs2_free_alloc_context(*meta_ac);
2682			*meta_ac = NULL;
2683		}
2684
2685		/*
2686		 * We cannot have an error and a non null *data_ac.
2687		 */
2688	}
2689
2690	return ret;
2691}
2692
2693/*
2694 * Read the inode specified by blkno to get suballoc_slot and
2695 * suballoc_bit.
2696 */
2697static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2698				       u16 *suballoc_slot, u64 *group_blkno,
2699				       u16 *suballoc_bit)
2700{
2701	int status;
2702	struct buffer_head *inode_bh = NULL;
2703	struct ocfs2_dinode *inode_fe;
2704
2705	trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2706
2707	/* dirty read disk */
2708	status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2709	if (status < 0) {
2710		mlog(ML_ERROR, "read block %llu failed %d\n",
2711		     (unsigned long long)blkno, status);
2712		goto bail;
2713	}
2714
2715	inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2716	if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2717		mlog(ML_ERROR, "invalid inode %llu requested\n",
2718		     (unsigned long long)blkno);
2719		status = -EINVAL;
2720		goto bail;
2721	}
2722
2723	if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2724	    (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2725		mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2726		     (unsigned long long)blkno,
2727		     (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2728		status = -EINVAL;
2729		goto bail;
2730	}
2731
2732	if (suballoc_slot)
2733		*suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2734	if (suballoc_bit)
2735		*suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2736	if (group_blkno)
2737		*group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2738
2739bail:
2740	brelse(inode_bh);
2741
2742	if (status)
2743		mlog_errno(status);
2744	return status;
2745}
2746
2747/*
2748 * test whether bit is SET in allocator bitmap or not.  on success, 0
2749 * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
2750 * is returned and *res is meaningless.  Call this after you have
2751 * cluster locked against suballoc, or you may get a result based on
2752 * non-up2date contents
2753 */
2754static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2755				   struct inode *suballoc,
2756				   struct buffer_head *alloc_bh,
2757				   u64 group_blkno, u64 blkno,
2758				   u16 bit, int *res)
2759{
2760	struct ocfs2_dinode *alloc_di;
2761	struct ocfs2_group_desc *group;
2762	struct buffer_head *group_bh = NULL;
2763	u64 bg_blkno;
2764	int status;
2765
2766	trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2767				      (unsigned int)bit);
2768
2769	alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2770	if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2771		mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2772		     (unsigned int)bit,
2773		     ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2774		status = -EINVAL;
2775		goto bail;
2776	}
2777
2778	bg_blkno = group_blkno ? group_blkno :
2779		   ocfs2_which_suballoc_group(blkno, bit);
2780	status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2781					     &group_bh);
2782	if (status < 0) {
2783		mlog(ML_ERROR, "read group %llu failed %d\n",
2784		     (unsigned long long)bg_blkno, status);
2785		goto bail;
2786	}
2787
2788	group = (struct ocfs2_group_desc *) group_bh->b_data;
2789	*res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2790
2791bail:
2792	brelse(group_bh);
2793
2794	if (status)
2795		mlog_errno(status);
2796	return status;
2797}
2798
2799/*
2800 * Test if the bit representing this inode (blkno) is set in the
2801 * suballocator.
2802 *
2803 * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2804 *
2805 * In the event of failure, a negative value is returned and *res is
2806 * meaningless.
2807 *
2808 * Callers must make sure to hold nfs_sync_lock to prevent
2809 * ocfs2_delete_inode() on another node from accessing the same
2810 * suballocator concurrently.
2811 */
2812int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2813{
2814	int status;
2815	u64 group_blkno = 0;
2816	u16 suballoc_bit = 0, suballoc_slot = 0;
2817	struct inode *inode_alloc_inode;
2818	struct buffer_head *alloc_bh = NULL;
2819
2820	trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2821
2822	status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2823					     &group_blkno, &suballoc_bit);
2824	if (status < 0) {
2825		mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2826		goto bail;
2827	}
2828
2829	if (suballoc_slot == (u16)OCFS2_INVALID_SLOT)
2830		inode_alloc_inode = ocfs2_get_system_file_inode(osb,
2831			GLOBAL_INODE_ALLOC_SYSTEM_INODE, suballoc_slot);
2832	else
2833		inode_alloc_inode = ocfs2_get_system_file_inode(osb,
2834			INODE_ALLOC_SYSTEM_INODE, suballoc_slot);
2835	if (!inode_alloc_inode) {
2836		/* the error code could be inaccurate, but we are not able to
2837		 * get the correct one. */
2838		status = -EINVAL;
2839		mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2840		     (u32)suballoc_slot);
2841		goto bail;
2842	}
2843
2844	inode_lock(inode_alloc_inode);
2845	status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2846	if (status < 0) {
2847		inode_unlock(inode_alloc_inode);
2848		iput(inode_alloc_inode);
2849		mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2850		     (u32)suballoc_slot, status);
2851		goto bail;
2852	}
2853
2854	status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2855					 group_blkno, blkno, suballoc_bit, res);
2856	if (status < 0)
2857		mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2858
2859	ocfs2_inode_unlock(inode_alloc_inode, 0);
2860	inode_unlock(inode_alloc_inode);
2861
2862	iput(inode_alloc_inode);
2863	brelse(alloc_bh);
2864bail:
2865	if (status)
2866		mlog_errno(status);
2867	return status;
2868}
2869