xref: /linux/fs/ocfs2/suballoc.c (revision 38da32ee70b876f5b8bea7c4135eff46339c18f2)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * suballoc.c
4  *
5  * metadata alloc and free
6  * Inspired by ext3 block groups.
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  */
10 
11 #include <linux/fs.h>
12 #include <linux/types.h>
13 #include <linux/slab.h>
14 #include <linux/highmem.h>
15 
16 #include <cluster/masklog.h>
17 
18 #include "ocfs2.h"
19 
20 #include "alloc.h"
21 #include "blockcheck.h"
22 #include "dlmglue.h"
23 #include "inode.h"
24 #include "journal.h"
25 #include "localalloc.h"
26 #include "suballoc.h"
27 #include "super.h"
28 #include "sysfile.h"
29 #include "uptodate.h"
30 #include "ocfs2_trace.h"
31 
32 #include "buffer_head_io.h"
33 
34 #define NOT_ALLOC_NEW_GROUP		0
35 #define ALLOC_NEW_GROUP			0x1
36 #define ALLOC_GROUPS_FROM_GLOBAL	0x2
37 
38 #define OCFS2_MAX_TO_STEAL		1024
39 
40 struct ocfs2_suballoc_result {
41 	u64		sr_bg_blkno;	/* The bg we allocated from.  Set
42 					   to 0 when a block group is
43 					   contiguous. */
44 	u64		sr_bg_stable_blkno; /*
45 					     * Doesn't change, always
46 					     * set to target block
47 					     * group descriptor
48 					     * block.
49 					     */
50 	u64		sr_blkno;	/* The first allocated block */
51 	unsigned int	sr_bit_offset;	/* The bit in the bg */
52 	unsigned int	sr_bits;	/* How many bits we claimed */
53 	unsigned int	sr_max_contig_bits; /* The length for contiguous
54 					     * free bits, only available
55 					     * for cluster group
56 					     */
57 };
58 
59 static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
60 {
61 	if (res->sr_blkno == 0)
62 		return 0;
63 
64 	if (res->sr_bg_blkno)
65 		return res->sr_bg_blkno;
66 
67 	return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset);
68 }
69 
70 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
71 static int ocfs2_block_group_fill(handle_t *handle,
72 				  struct inode *alloc_inode,
73 				  struct buffer_head *bg_bh,
74 				  u64 group_blkno,
75 				  unsigned int group_clusters,
76 				  u16 my_chain,
77 				  struct ocfs2_chain_list *cl);
78 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
79 				   struct inode *alloc_inode,
80 				   struct buffer_head *bh,
81 				   u64 max_block,
82 				   u64 *last_alloc_group,
83 				   int flags);
84 
85 static int ocfs2_cluster_group_search(struct inode *inode,
86 				      struct buffer_head *group_bh,
87 				      u32 bits_wanted, u32 min_bits,
88 				      u64 max_block,
89 				      struct ocfs2_suballoc_result *res);
90 static int ocfs2_block_group_search(struct inode *inode,
91 				    struct buffer_head *group_bh,
92 				    u32 bits_wanted, u32 min_bits,
93 				    u64 max_block,
94 				    struct ocfs2_suballoc_result *res);
95 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
96 				     handle_t *handle,
97 				     u32 bits_wanted,
98 				     u32 min_bits,
99 				     struct ocfs2_suballoc_result *res);
100 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
101 					 int nr);
102 static int ocfs2_relink_block_group(handle_t *handle,
103 				    struct inode *alloc_inode,
104 				    struct buffer_head *fe_bh,
105 				    struct buffer_head *bg_bh,
106 				    struct buffer_head *prev_bg_bh,
107 				    u16 chain);
108 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
109 						     u32 wanted);
110 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
111 						   u64 bg_blkno,
112 						   u16 bg_bit_off);
113 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
114 						u64 data_blkno,
115 						u64 *bg_blkno,
116 						u16 *bg_bit_off);
117 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
118 					     u32 bits_wanted, u64 max_block,
119 					     int flags,
120 					     struct ocfs2_alloc_context **ac);
121 
122 void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
123 {
124 	struct inode *inode = ac->ac_inode;
125 
126 	if (inode) {
127 		if (ac->ac_which != OCFS2_AC_USE_LOCAL)
128 			ocfs2_inode_unlock(inode, 1);
129 
130 		inode_unlock(inode);
131 
132 		iput(inode);
133 		ac->ac_inode = NULL;
134 	}
135 	brelse(ac->ac_bh);
136 	ac->ac_bh = NULL;
137 	ac->ac_resv = NULL;
138 	kfree(ac->ac_find_loc_priv);
139 	ac->ac_find_loc_priv = NULL;
140 }
141 
142 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
143 {
144 	ocfs2_free_ac_resource(ac);
145 	kfree(ac);
146 }
147 
148 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
149 {
150 	return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
151 }
152 
153 #define do_error(fmt, ...)						\
154 do {									\
155 	if (resize)							\
156 		mlog(ML_ERROR, fmt, ##__VA_ARGS__);			\
157 	else								\
158 		return ocfs2_error(sb, fmt, ##__VA_ARGS__);		\
159 } while (0)
160 
161 static int ocfs2_validate_gd_self(struct super_block *sb,
162 				  struct buffer_head *bh,
163 				  int resize)
164 {
165 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
166 
167 	if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
168 		do_error("Group descriptor #%llu has bad signature %.*s\n",
169 			 (unsigned long long)bh->b_blocknr, 7,
170 			 gd->bg_signature);
171 	}
172 
173 	if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
174 		do_error("Group descriptor #%llu has an invalid bg_blkno of %llu\n",
175 			 (unsigned long long)bh->b_blocknr,
176 			 (unsigned long long)le64_to_cpu(gd->bg_blkno));
177 	}
178 
179 	if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
180 		do_error("Group descriptor #%llu has an invalid fs_generation of #%u\n",
181 			 (unsigned long long)bh->b_blocknr,
182 			 le32_to_cpu(gd->bg_generation));
183 	}
184 
185 	if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
186 		do_error("Group descriptor #%llu has bit count %u but claims that %u are free\n",
187 			 (unsigned long long)bh->b_blocknr,
188 			 le16_to_cpu(gd->bg_bits),
189 			 le16_to_cpu(gd->bg_free_bits_count));
190 	}
191 
192 	if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
193 		do_error("Group descriptor #%llu has bit count %u but max bitmap bits of %u\n",
194 			 (unsigned long long)bh->b_blocknr,
195 			 le16_to_cpu(gd->bg_bits),
196 			 8 * le16_to_cpu(gd->bg_size));
197 	}
198 
199 	return 0;
200 }
201 
202 static int ocfs2_validate_gd_parent(struct super_block *sb,
203 				    struct ocfs2_dinode *di,
204 				    struct buffer_head *bh,
205 				    int resize)
206 {
207 	unsigned int max_bits;
208 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
209 
210 	if (di->i_blkno != gd->bg_parent_dinode) {
211 		do_error("Group descriptor #%llu has bad parent pointer (%llu, expected %llu)\n",
212 			 (unsigned long long)bh->b_blocknr,
213 			 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
214 			 (unsigned long long)le64_to_cpu(di->i_blkno));
215 	}
216 
217 	max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
218 	if (le16_to_cpu(gd->bg_bits) > max_bits) {
219 		do_error("Group descriptor #%llu has bit count of %u\n",
220 			 (unsigned long long)bh->b_blocknr,
221 			 le16_to_cpu(gd->bg_bits));
222 	}
223 
224 	/* In resize, we may meet the case bg_chain == cl_next_free_rec. */
225 	if ((le16_to_cpu(gd->bg_chain) >
226 	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
227 	    ((le16_to_cpu(gd->bg_chain) ==
228 	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
229 		do_error("Group descriptor #%llu has bad chain %u\n",
230 			 (unsigned long long)bh->b_blocknr,
231 			 le16_to_cpu(gd->bg_chain));
232 	}
233 
234 	return 0;
235 }
236 
237 #undef do_error
238 
239 /*
240  * This version only prints errors.  It does not fail the filesystem, and
241  * exists only for resize.
242  */
243 int ocfs2_check_group_descriptor(struct super_block *sb,
244 				 struct ocfs2_dinode *di,
245 				 struct buffer_head *bh)
246 {
247 	int rc;
248 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
249 
250 	BUG_ON(!buffer_uptodate(bh));
251 
252 	/*
253 	 * If the ecc fails, we return the error but otherwise
254 	 * leave the filesystem running.  We know any error is
255 	 * local to this block.
256 	 */
257 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
258 	if (rc) {
259 		mlog(ML_ERROR,
260 		     "Checksum failed for group descriptor %llu\n",
261 		     (unsigned long long)bh->b_blocknr);
262 	} else
263 		rc = ocfs2_validate_gd_self(sb, bh, 1);
264 	if (!rc)
265 		rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
266 
267 	return rc;
268 }
269 
270 static int ocfs2_validate_group_descriptor(struct super_block *sb,
271 					   struct buffer_head *bh)
272 {
273 	int rc;
274 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
275 
276 	trace_ocfs2_validate_group_descriptor(
277 					(unsigned long long)bh->b_blocknr);
278 
279 	BUG_ON(!buffer_uptodate(bh));
280 
281 	/*
282 	 * If the ecc fails, we return the error but otherwise
283 	 * leave the filesystem running.  We know any error is
284 	 * local to this block.
285 	 */
286 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
287 	if (rc)
288 		return rc;
289 
290 	/*
291 	 * Errors after here are fatal.
292 	 */
293 
294 	return ocfs2_validate_gd_self(sb, bh, 0);
295 }
296 
297 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
298 				u64 gd_blkno, struct buffer_head **bh)
299 {
300 	int rc;
301 	struct buffer_head *tmp = *bh;
302 
303 	rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
304 			      ocfs2_validate_group_descriptor);
305 	if (rc)
306 		goto out;
307 
308 	rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
309 	if (rc) {
310 		brelse(tmp);
311 		goto out;
312 	}
313 
314 	/* If ocfs2_read_block() got us a new bh, pass it up. */
315 	if (!*bh)
316 		*bh = tmp;
317 
318 out:
319 	return rc;
320 }
321 
322 static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
323 					  struct ocfs2_group_desc *bg,
324 					  struct ocfs2_chain_list *cl,
325 					  u64 p_blkno, unsigned int clusters)
326 {
327 	struct ocfs2_extent_list *el = &bg->bg_list;
328 	struct ocfs2_extent_rec *rec;
329 
330 	BUG_ON(!ocfs2_supports_discontig_bg(osb));
331 	if (!el->l_next_free_rec)
332 		el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
333 	rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
334 	rec->e_blkno = cpu_to_le64(p_blkno);
335 	rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
336 				  le16_to_cpu(cl->cl_bpc));
337 	rec->e_leaf_clusters = cpu_to_le16(clusters);
338 	le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
339 	le16_add_cpu(&bg->bg_free_bits_count,
340 		     clusters * le16_to_cpu(cl->cl_bpc));
341 	le16_add_cpu(&el->l_next_free_rec, 1);
342 }
343 
344 static int ocfs2_block_group_fill(handle_t *handle,
345 				  struct inode *alloc_inode,
346 				  struct buffer_head *bg_bh,
347 				  u64 group_blkno,
348 				  unsigned int group_clusters,
349 				  u16 my_chain,
350 				  struct ocfs2_chain_list *cl)
351 {
352 	int status = 0;
353 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
354 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
355 	struct super_block * sb = alloc_inode->i_sb;
356 
357 	if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
358 		status = ocfs2_error(alloc_inode->i_sb,
359 				     "group block (%llu) != b_blocknr (%llu)\n",
360 				     (unsigned long long)group_blkno,
361 				     (unsigned long long) bg_bh->b_blocknr);
362 		goto bail;
363 	}
364 
365 	status = ocfs2_journal_access_gd(handle,
366 					 INODE_CACHE(alloc_inode),
367 					 bg_bh,
368 					 OCFS2_JOURNAL_ACCESS_CREATE);
369 	if (status < 0) {
370 		mlog_errno(status);
371 		goto bail;
372 	}
373 
374 	memset(bg, 0, sb->s_blocksize);
375 	strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
376 	bg->bg_generation = cpu_to_le32(osb->fs_generation);
377 	bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
378 						osb->s_feature_incompat));
379 	bg->bg_chain = cpu_to_le16(my_chain);
380 	bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
381 	bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
382 	bg->bg_blkno = cpu_to_le64(group_blkno);
383 	if (group_clusters == le16_to_cpu(cl->cl_cpg))
384 		bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
385 	else
386 		ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
387 					      group_clusters);
388 
389 	/* set the 1st bit in the bitmap to account for the descriptor block */
390 	ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
391 	bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
392 
393 	ocfs2_journal_dirty(handle, bg_bh);
394 
395 	/* There is no need to zero out or otherwise initialize the
396 	 * other blocks in a group - All valid FS metadata in a block
397 	 * group stores the superblock fs_generation value at
398 	 * allocation time. */
399 
400 bail:
401 	if (status)
402 		mlog_errno(status);
403 	return status;
404 }
405 
406 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
407 {
408 	u16 curr, best;
409 
410 	best = curr = 0;
411 	while (curr < le16_to_cpu(cl->cl_count)) {
412 		if (le32_to_cpu(cl->cl_recs[best].c_total) >
413 		    le32_to_cpu(cl->cl_recs[curr].c_total))
414 			best = curr;
415 		curr++;
416 	}
417 	return best;
418 }
419 
420 static struct buffer_head *
421 ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
422 			       struct inode *alloc_inode,
423 			       struct ocfs2_alloc_context *ac,
424 			       struct ocfs2_chain_list *cl)
425 {
426 	int status;
427 	u32 bit_off, num_bits;
428 	u64 bg_blkno;
429 	struct buffer_head *bg_bh;
430 	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
431 
432 	status = ocfs2_claim_clusters(handle, ac,
433 				      le16_to_cpu(cl->cl_cpg), &bit_off,
434 				      &num_bits);
435 	if (status < 0) {
436 		if (status != -ENOSPC)
437 			mlog_errno(status);
438 		goto bail;
439 	}
440 
441 	/* setup the group */
442 	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
443 	trace_ocfs2_block_group_alloc_contig(
444 	     (unsigned long long)bg_blkno, alloc_rec);
445 
446 	bg_bh = sb_getblk(osb->sb, bg_blkno);
447 	if (!bg_bh) {
448 		status = -ENOMEM;
449 		mlog_errno(status);
450 		goto bail;
451 	}
452 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
453 
454 	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
455 					bg_blkno, num_bits, alloc_rec, cl);
456 	if (status < 0) {
457 		brelse(bg_bh);
458 		mlog_errno(status);
459 	}
460 
461 bail:
462 	return status ? ERR_PTR(status) : bg_bh;
463 }
464 
465 static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
466 					handle_t *handle,
467 					struct ocfs2_alloc_context *ac,
468 					unsigned int min_bits,
469 					u32 *bit_off, u32 *num_bits)
470 {
471 	int status = 0;
472 
473 	while (min_bits) {
474 		status = ocfs2_claim_clusters(handle, ac, min_bits,
475 					      bit_off, num_bits);
476 		if (status != -ENOSPC)
477 			break;
478 
479 		min_bits >>= 1;
480 	}
481 
482 	return status;
483 }
484 
485 static int ocfs2_block_group_grow_discontig(handle_t *handle,
486 					    struct inode *alloc_inode,
487 					    struct buffer_head *bg_bh,
488 					    struct ocfs2_alloc_context *ac,
489 					    struct ocfs2_chain_list *cl,
490 					    unsigned int min_bits)
491 {
492 	int status;
493 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
494 	struct ocfs2_group_desc *bg =
495 		(struct ocfs2_group_desc *)bg_bh->b_data;
496 	unsigned int needed = le16_to_cpu(cl->cl_cpg) -
497 			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
498 	u32 p_cpos, clusters;
499 	u64 p_blkno;
500 	struct ocfs2_extent_list *el = &bg->bg_list;
501 
502 	status = ocfs2_journal_access_gd(handle,
503 					 INODE_CACHE(alloc_inode),
504 					 bg_bh,
505 					 OCFS2_JOURNAL_ACCESS_CREATE);
506 	if (status < 0) {
507 		mlog_errno(status);
508 		goto bail;
509 	}
510 
511 	while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
512 				le16_to_cpu(el->l_count))) {
513 		if (min_bits > needed)
514 			min_bits = needed;
515 		status = ocfs2_block_group_claim_bits(osb, handle, ac,
516 						      min_bits, &p_cpos,
517 						      &clusters);
518 		if (status < 0) {
519 			if (status != -ENOSPC)
520 				mlog_errno(status);
521 			goto bail;
522 		}
523 		p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
524 		ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
525 					      clusters);
526 
527 		min_bits = clusters;
528 		needed = le16_to_cpu(cl->cl_cpg) -
529 			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
530 	}
531 
532 	if (needed > 0) {
533 		/*
534 		 * We have used up all the extent rec but can't fill up
535 		 * the cpg. So bail out.
536 		 */
537 		status = -ENOSPC;
538 		goto bail;
539 	}
540 
541 	ocfs2_journal_dirty(handle, bg_bh);
542 
543 bail:
544 	return status;
545 }
546 
547 static void ocfs2_bg_alloc_cleanup(handle_t *handle,
548 				   struct ocfs2_alloc_context *cluster_ac,
549 				   struct inode *alloc_inode,
550 				   struct buffer_head *bg_bh)
551 {
552 	int i, ret;
553 	struct ocfs2_group_desc *bg;
554 	struct ocfs2_extent_list *el;
555 	struct ocfs2_extent_rec *rec;
556 
557 	if (!bg_bh)
558 		return;
559 
560 	bg = (struct ocfs2_group_desc *)bg_bh->b_data;
561 	el = &bg->bg_list;
562 	for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
563 		rec = &el->l_recs[i];
564 		ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
565 					  cluster_ac->ac_bh,
566 					  le64_to_cpu(rec->e_blkno),
567 					  le16_to_cpu(rec->e_leaf_clusters));
568 		if (ret)
569 			mlog_errno(ret);
570 		/* Try all the clusters to free */
571 	}
572 
573 	ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
574 	brelse(bg_bh);
575 }
576 
577 static struct buffer_head *
578 ocfs2_block_group_alloc_discontig(handle_t *handle,
579 				  struct inode *alloc_inode,
580 				  struct ocfs2_alloc_context *ac,
581 				  struct ocfs2_chain_list *cl)
582 {
583 	int status;
584 	u32 bit_off, num_bits;
585 	u64 bg_blkno;
586 	unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
587 	struct buffer_head *bg_bh = NULL;
588 	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
589 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
590 
591 	if (!ocfs2_supports_discontig_bg(osb)) {
592 		status = -ENOSPC;
593 		goto bail;
594 	}
595 
596 	status = ocfs2_extend_trans(handle,
597 				    ocfs2_calc_bg_discontig_credits(osb->sb));
598 	if (status) {
599 		mlog_errno(status);
600 		goto bail;
601 	}
602 
603 	/*
604 	 * We're going to be grabbing from multiple cluster groups.
605 	 * We don't have enough credits to relink them all, and the
606 	 * cluster groups will be staying in cache for the duration of
607 	 * this operation.
608 	 */
609 	ac->ac_disable_chain_relink = 1;
610 
611 	/* Claim the first region */
612 	status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
613 					      &bit_off, &num_bits);
614 	if (status < 0) {
615 		if (status != -ENOSPC)
616 			mlog_errno(status);
617 		goto bail;
618 	}
619 	min_bits = num_bits;
620 
621 	/* setup the group */
622 	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
623 	trace_ocfs2_block_group_alloc_discontig(
624 				(unsigned long long)bg_blkno, alloc_rec);
625 
626 	bg_bh = sb_getblk(osb->sb, bg_blkno);
627 	if (!bg_bh) {
628 		status = -ENOMEM;
629 		mlog_errno(status);
630 		goto bail;
631 	}
632 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
633 
634 	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
635 					bg_blkno, num_bits, alloc_rec, cl);
636 	if (status < 0) {
637 		mlog_errno(status);
638 		goto bail;
639 	}
640 
641 	status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
642 						  bg_bh, ac, cl, min_bits);
643 	if (status)
644 		mlog_errno(status);
645 
646 bail:
647 	if (status)
648 		ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
649 	return status ? ERR_PTR(status) : bg_bh;
650 }
651 
652 /*
653  * We expect the block group allocator to already be locked.
654  */
655 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
656 				   struct inode *alloc_inode,
657 				   struct buffer_head *bh,
658 				   u64 max_block,
659 				   u64 *last_alloc_group,
660 				   int flags)
661 {
662 	int status, credits;
663 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
664 	struct ocfs2_chain_list *cl;
665 	struct ocfs2_alloc_context *ac = NULL;
666 	handle_t *handle = NULL;
667 	u16 alloc_rec;
668 	struct buffer_head *bg_bh = NULL;
669 	struct ocfs2_group_desc *bg;
670 
671 	BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
672 
673 	cl = &fe->id2.i_chain;
674 	status = ocfs2_reserve_clusters_with_limit(osb,
675 						   le16_to_cpu(cl->cl_cpg),
676 						   max_block, flags, &ac);
677 	if (status < 0) {
678 		if (status != -ENOSPC)
679 			mlog_errno(status);
680 		goto bail;
681 	}
682 
683 	credits = ocfs2_calc_group_alloc_credits(osb->sb,
684 						 le16_to_cpu(cl->cl_cpg));
685 	handle = ocfs2_start_trans(osb, credits);
686 	if (IS_ERR(handle)) {
687 		status = PTR_ERR(handle);
688 		handle = NULL;
689 		mlog_errno(status);
690 		goto bail;
691 	}
692 
693 	if (last_alloc_group && *last_alloc_group != 0) {
694 		trace_ocfs2_block_group_alloc(
695 				(unsigned long long)*last_alloc_group);
696 		ac->ac_last_group = *last_alloc_group;
697 	}
698 
699 	bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
700 					       ac, cl);
701 	if (PTR_ERR(bg_bh) == -ENOSPC)
702 		bg_bh = ocfs2_block_group_alloc_discontig(handle,
703 							  alloc_inode,
704 							  ac, cl);
705 	if (IS_ERR(bg_bh)) {
706 		status = PTR_ERR(bg_bh);
707 		bg_bh = NULL;
708 		if (status != -ENOSPC)
709 			mlog_errno(status);
710 		goto bail;
711 	}
712 	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
713 
714 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
715 					 bh, OCFS2_JOURNAL_ACCESS_WRITE);
716 	if (status < 0) {
717 		mlog_errno(status);
718 		goto bail;
719 	}
720 
721 	alloc_rec = le16_to_cpu(bg->bg_chain);
722 	le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
723 		     le16_to_cpu(bg->bg_free_bits_count));
724 	le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
725 		     le16_to_cpu(bg->bg_bits));
726 	cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
727 	if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
728 		le16_add_cpu(&cl->cl_next_free_rec, 1);
729 
730 	le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
731 					le16_to_cpu(bg->bg_free_bits_count));
732 	le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
733 	le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
734 
735 	ocfs2_journal_dirty(handle, bh);
736 
737 	spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
738 	OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
739 	fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
740 					     le32_to_cpu(fe->i_clusters)));
741 	spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
742 	i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
743 	alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
744 	ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
745 
746 	status = 0;
747 
748 	/* save the new last alloc group so that the caller can cache it. */
749 	if (last_alloc_group)
750 		*last_alloc_group = ac->ac_last_group;
751 
752 bail:
753 	if (handle)
754 		ocfs2_commit_trans(osb, handle);
755 
756 	if (ac)
757 		ocfs2_free_alloc_context(ac);
758 
759 	brelse(bg_bh);
760 
761 	if (status)
762 		mlog_errno(status);
763 	return status;
764 }
765 
766 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
767 				       struct ocfs2_alloc_context *ac,
768 				       int type,
769 				       u32 slot,
770 				       u64 *last_alloc_group,
771 				       int flags)
772 {
773 	int status;
774 	u32 bits_wanted = ac->ac_bits_wanted;
775 	struct inode *alloc_inode;
776 	struct buffer_head *bh = NULL;
777 	struct ocfs2_dinode *fe;
778 	u32 free_bits;
779 
780 	alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
781 	if (!alloc_inode) {
782 		mlog_errno(-EINVAL);
783 		return -EINVAL;
784 	}
785 
786 	inode_lock(alloc_inode);
787 
788 	status = ocfs2_inode_lock(alloc_inode, &bh, 1);
789 	if (status < 0) {
790 		inode_unlock(alloc_inode);
791 		iput(alloc_inode);
792 
793 		mlog_errno(status);
794 		return status;
795 	}
796 
797 	ac->ac_inode = alloc_inode;
798 	ac->ac_alloc_slot = slot;
799 
800 	fe = (struct ocfs2_dinode *) bh->b_data;
801 
802 	/* The bh was validated by the inode read inside
803 	 * ocfs2_inode_lock().  Any corruption is a code bug. */
804 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
805 
806 	if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
807 		status = ocfs2_error(alloc_inode->i_sb,
808 				     "Invalid chain allocator %llu\n",
809 				     (unsigned long long)le64_to_cpu(fe->i_blkno));
810 		goto bail;
811 	}
812 
813 	free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
814 		le32_to_cpu(fe->id1.bitmap1.i_used);
815 
816 	if (bits_wanted > free_bits) {
817 		/* cluster bitmap never grows */
818 		if (ocfs2_is_cluster_bitmap(alloc_inode)) {
819 			trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
820 								free_bits);
821 			status = -ENOSPC;
822 			goto bail;
823 		}
824 
825 		if (!(flags & ALLOC_NEW_GROUP)) {
826 			trace_ocfs2_reserve_suballoc_bits_no_new_group(
827 						slot, bits_wanted, free_bits);
828 			status = -ENOSPC;
829 			goto bail;
830 		}
831 
832 		status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
833 						 ac->ac_max_block,
834 						 last_alloc_group, flags);
835 		if (status < 0) {
836 			if (status != -ENOSPC)
837 				mlog_errno(status);
838 			goto bail;
839 		}
840 		atomic_inc(&osb->alloc_stats.bg_extends);
841 
842 		/* You should never ask for this much metadata */
843 		BUG_ON(bits_wanted >
844 		       (le32_to_cpu(fe->id1.bitmap1.i_total)
845 			- le32_to_cpu(fe->id1.bitmap1.i_used)));
846 	}
847 
848 	get_bh(bh);
849 	ac->ac_bh = bh;
850 bail:
851 	brelse(bh);
852 
853 	if (status)
854 		mlog_errno(status);
855 	return status;
856 }
857 
858 static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
859 {
860 	spin_lock(&osb->osb_lock);
861 	osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
862 	spin_unlock(&osb->osb_lock);
863 	atomic_set(&osb->s_num_inodes_stolen, 0);
864 }
865 
866 static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
867 {
868 	spin_lock(&osb->osb_lock);
869 	osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
870 	spin_unlock(&osb->osb_lock);
871 	atomic_set(&osb->s_num_meta_stolen, 0);
872 }
873 
874 void ocfs2_init_steal_slots(struct ocfs2_super *osb)
875 {
876 	ocfs2_init_inode_steal_slot(osb);
877 	ocfs2_init_meta_steal_slot(osb);
878 }
879 
880 static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
881 {
882 	spin_lock(&osb->osb_lock);
883 	if (type == INODE_ALLOC_SYSTEM_INODE)
884 		osb->s_inode_steal_slot = (u16)slot;
885 	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
886 		osb->s_meta_steal_slot = (u16)slot;
887 	spin_unlock(&osb->osb_lock);
888 }
889 
890 static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
891 {
892 	int slot = OCFS2_INVALID_SLOT;
893 
894 	spin_lock(&osb->osb_lock);
895 	if (type == INODE_ALLOC_SYSTEM_INODE)
896 		slot = osb->s_inode_steal_slot;
897 	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
898 		slot = osb->s_meta_steal_slot;
899 	spin_unlock(&osb->osb_lock);
900 
901 	return slot;
902 }
903 
904 static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
905 {
906 	return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
907 }
908 
909 static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
910 {
911 	return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
912 }
913 
914 static int ocfs2_steal_resource(struct ocfs2_super *osb,
915 				struct ocfs2_alloc_context *ac,
916 				int type)
917 {
918 	int i, status = -ENOSPC;
919 	int slot = __ocfs2_get_steal_slot(osb, type);
920 
921 	/* Start to steal resource from the first slot after ours. */
922 	if (slot == OCFS2_INVALID_SLOT)
923 		slot = osb->slot_num + 1;
924 
925 	for (i = 0; i < osb->max_slots; i++, slot++) {
926 		if (slot == osb->max_slots)
927 			slot = 0;
928 
929 		if (slot == osb->slot_num)
930 			continue;
931 
932 		status = ocfs2_reserve_suballoc_bits(osb, ac,
933 						     type,
934 						     (u32)slot, NULL,
935 						     NOT_ALLOC_NEW_GROUP);
936 		if (status >= 0) {
937 			__ocfs2_set_steal_slot(osb, slot, type);
938 			break;
939 		}
940 
941 		ocfs2_free_ac_resource(ac);
942 	}
943 
944 	return status;
945 }
946 
947 static int ocfs2_steal_inode(struct ocfs2_super *osb,
948 			     struct ocfs2_alloc_context *ac)
949 {
950 	return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
951 }
952 
953 static int ocfs2_steal_meta(struct ocfs2_super *osb,
954 			    struct ocfs2_alloc_context *ac)
955 {
956 	return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
957 }
958 
959 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
960 				      int blocks,
961 				      struct ocfs2_alloc_context **ac)
962 {
963 	int status;
964 	int slot = ocfs2_get_meta_steal_slot(osb);
965 
966 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
967 	if (!(*ac)) {
968 		status = -ENOMEM;
969 		mlog_errno(status);
970 		goto bail;
971 	}
972 
973 	(*ac)->ac_bits_wanted = blocks;
974 	(*ac)->ac_which = OCFS2_AC_USE_META;
975 	(*ac)->ac_group_search = ocfs2_block_group_search;
976 
977 	if (slot != OCFS2_INVALID_SLOT &&
978 		atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
979 		goto extent_steal;
980 
981 	atomic_set(&osb->s_num_meta_stolen, 0);
982 	status = ocfs2_reserve_suballoc_bits(osb, (*ac),
983 					     EXTENT_ALLOC_SYSTEM_INODE,
984 					     (u32)osb->slot_num, NULL,
985 					     ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
986 
987 
988 	if (status >= 0) {
989 		status = 0;
990 		if (slot != OCFS2_INVALID_SLOT)
991 			ocfs2_init_meta_steal_slot(osb);
992 		goto bail;
993 	} else if (status < 0 && status != -ENOSPC) {
994 		mlog_errno(status);
995 		goto bail;
996 	}
997 
998 	ocfs2_free_ac_resource(*ac);
999 
1000 extent_steal:
1001 	status = ocfs2_steal_meta(osb, *ac);
1002 	atomic_inc(&osb->s_num_meta_stolen);
1003 	if (status < 0) {
1004 		if (status != -ENOSPC)
1005 			mlog_errno(status);
1006 		goto bail;
1007 	}
1008 
1009 	status = 0;
1010 bail:
1011 	if ((status < 0) && *ac) {
1012 		ocfs2_free_alloc_context(*ac);
1013 		*ac = NULL;
1014 	}
1015 
1016 	if (status)
1017 		mlog_errno(status);
1018 	return status;
1019 }
1020 
1021 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1022 			       struct ocfs2_extent_list *root_el,
1023 			       struct ocfs2_alloc_context **ac)
1024 {
1025 	return ocfs2_reserve_new_metadata_blocks(osb,
1026 					ocfs2_extend_meta_needed(root_el),
1027 					ac);
1028 }
1029 
1030 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
1031 			    struct ocfs2_alloc_context **ac)
1032 {
1033 	int status;
1034 	int slot = ocfs2_get_inode_steal_slot(osb);
1035 	u64 alloc_group;
1036 
1037 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1038 	if (!(*ac)) {
1039 		status = -ENOMEM;
1040 		mlog_errno(status);
1041 		goto bail;
1042 	}
1043 
1044 	(*ac)->ac_bits_wanted = 1;
1045 	(*ac)->ac_which = OCFS2_AC_USE_INODE;
1046 
1047 	(*ac)->ac_group_search = ocfs2_block_group_search;
1048 
1049 	/*
1050 	 * stat(2) can't handle i_ino > 32bits, so we tell the
1051 	 * lower levels not to allocate us a block group past that
1052 	 * limit.  The 'inode64' mount option avoids this behavior.
1053 	 */
1054 	if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1055 		(*ac)->ac_max_block = (u32)~0U;
1056 
1057 	/*
1058 	 * slot is set when we successfully steal inode from other nodes.
1059 	 * It is reset in 3 places:
1060 	 * 1. when we flush the truncate log
1061 	 * 2. when we complete local alloc recovery.
1062 	 * 3. when we successfully allocate from our own slot.
1063 	 * After it is set, we will go on stealing inodes until we find the
1064 	 * need to check our slots to see whether there is some space for us.
1065 	 */
1066 	if (slot != OCFS2_INVALID_SLOT &&
1067 	    atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
1068 		goto inode_steal;
1069 
1070 	atomic_set(&osb->s_num_inodes_stolen, 0);
1071 	alloc_group = osb->osb_inode_alloc_group;
1072 	status = ocfs2_reserve_suballoc_bits(osb, *ac,
1073 					     INODE_ALLOC_SYSTEM_INODE,
1074 					     (u32)osb->slot_num,
1075 					     &alloc_group,
1076 					     ALLOC_NEW_GROUP |
1077 					     ALLOC_GROUPS_FROM_GLOBAL);
1078 	if (status >= 0) {
1079 		status = 0;
1080 
1081 		spin_lock(&osb->osb_lock);
1082 		osb->osb_inode_alloc_group = alloc_group;
1083 		spin_unlock(&osb->osb_lock);
1084 		trace_ocfs2_reserve_new_inode_new_group(
1085 			(unsigned long long)alloc_group);
1086 
1087 		/*
1088 		 * Some inodes must be freed by us, so try to allocate
1089 		 * from our own next time.
1090 		 */
1091 		if (slot != OCFS2_INVALID_SLOT)
1092 			ocfs2_init_inode_steal_slot(osb);
1093 		goto bail;
1094 	} else if (status < 0 && status != -ENOSPC) {
1095 		mlog_errno(status);
1096 		goto bail;
1097 	}
1098 
1099 	ocfs2_free_ac_resource(*ac);
1100 
1101 inode_steal:
1102 	status = ocfs2_steal_inode(osb, *ac);
1103 	atomic_inc(&osb->s_num_inodes_stolen);
1104 	if (status < 0) {
1105 		if (status != -ENOSPC)
1106 			mlog_errno(status);
1107 		goto bail;
1108 	}
1109 
1110 	status = 0;
1111 bail:
1112 	if ((status < 0) && *ac) {
1113 		ocfs2_free_alloc_context(*ac);
1114 		*ac = NULL;
1115 	}
1116 
1117 	if (status)
1118 		mlog_errno(status);
1119 	return status;
1120 }
1121 
1122 /* local alloc code has to do the same thing, so rather than do this
1123  * twice.. */
1124 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1125 				      struct ocfs2_alloc_context *ac)
1126 {
1127 	int status;
1128 
1129 	ac->ac_which = OCFS2_AC_USE_MAIN;
1130 	ac->ac_group_search = ocfs2_cluster_group_search;
1131 
1132 	status = ocfs2_reserve_suballoc_bits(osb, ac,
1133 					     GLOBAL_BITMAP_SYSTEM_INODE,
1134 					     OCFS2_INVALID_SLOT, NULL,
1135 					     ALLOC_NEW_GROUP);
1136 	if (status < 0 && status != -ENOSPC)
1137 		mlog_errno(status);
1138 
1139 	return status;
1140 }
1141 
1142 /* Callers don't need to care which bitmap (local alloc or main) to
1143  * use so we figure it out for them, but unfortunately this clutters
1144  * things a bit. */
1145 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1146 					     u32 bits_wanted, u64 max_block,
1147 					     int flags,
1148 					     struct ocfs2_alloc_context **ac)
1149 {
1150 	int status, ret = 0;
1151 	int retried = 0;
1152 
1153 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1154 	if (!(*ac)) {
1155 		status = -ENOMEM;
1156 		mlog_errno(status);
1157 		goto bail;
1158 	}
1159 
1160 	(*ac)->ac_bits_wanted = bits_wanted;
1161 	(*ac)->ac_max_block = max_block;
1162 
1163 	status = -ENOSPC;
1164 	if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1165 	    ocfs2_alloc_should_use_local(osb, bits_wanted)) {
1166 		status = ocfs2_reserve_local_alloc_bits(osb,
1167 							bits_wanted,
1168 							*ac);
1169 		if ((status < 0) && (status != -ENOSPC)) {
1170 			mlog_errno(status);
1171 			goto bail;
1172 		}
1173 	}
1174 
1175 	if (status == -ENOSPC) {
1176 retry:
1177 		status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1178 		/* Retry if there is sufficient space cached in truncate log */
1179 		if (status == -ENOSPC && !retried) {
1180 			retried = 1;
1181 			ocfs2_inode_unlock((*ac)->ac_inode, 1);
1182 			inode_unlock((*ac)->ac_inode);
1183 
1184 			ret = ocfs2_try_to_free_truncate_log(osb, bits_wanted);
1185 			if (ret == 1) {
1186 				iput((*ac)->ac_inode);
1187 				(*ac)->ac_inode = NULL;
1188 				goto retry;
1189 			}
1190 
1191 			if (ret < 0)
1192 				mlog_errno(ret);
1193 
1194 			inode_lock((*ac)->ac_inode);
1195 			ret = ocfs2_inode_lock((*ac)->ac_inode, NULL, 1);
1196 			if (ret < 0) {
1197 				mlog_errno(ret);
1198 				inode_unlock((*ac)->ac_inode);
1199 				iput((*ac)->ac_inode);
1200 				(*ac)->ac_inode = NULL;
1201 				goto bail;
1202 			}
1203 		}
1204 		if (status < 0) {
1205 			if (status != -ENOSPC)
1206 				mlog_errno(status);
1207 			goto bail;
1208 		}
1209 	}
1210 
1211 	status = 0;
1212 bail:
1213 	if ((status < 0) && *ac) {
1214 		ocfs2_free_alloc_context(*ac);
1215 		*ac = NULL;
1216 	}
1217 
1218 	if (status)
1219 		mlog_errno(status);
1220 	return status;
1221 }
1222 
1223 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1224 			   u32 bits_wanted,
1225 			   struct ocfs2_alloc_context **ac)
1226 {
1227 	return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1228 						 ALLOC_NEW_GROUP, ac);
1229 }
1230 
1231 /*
1232  * More or less lifted from ext3. I'll leave their description below:
1233  *
1234  * "For ext3 allocations, we must not reuse any blocks which are
1235  * allocated in the bitmap buffer's "last committed data" copy.  This
1236  * prevents deletes from freeing up the page for reuse until we have
1237  * committed the delete transaction.
1238  *
1239  * If we didn't do this, then deleting something and reallocating it as
1240  * data would allow the old block to be overwritten before the
1241  * transaction committed (because we force data to disk before commit).
1242  * This would lead to corruption if we crashed between overwriting the
1243  * data and committing the delete.
1244  *
1245  * @@@ We may want to make this allocation behaviour conditional on
1246  * data-writes at some point, and disable it for metadata allocations or
1247  * sync-data inodes."
1248  *
1249  * Note: OCFS2 already does this differently for metadata vs data
1250  * allocations, as those bitmaps are separate and undo access is never
1251  * called on a metadata group descriptor.
1252  */
1253 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1254 					 int nr)
1255 {
1256 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1257 	struct journal_head *jh;
1258 	int ret;
1259 
1260 	if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1261 		return 0;
1262 
1263 	jh = jbd2_journal_grab_journal_head(bg_bh);
1264 	if (!jh)
1265 		return 1;
1266 
1267 	spin_lock(&jh->b_state_lock);
1268 	bg = (struct ocfs2_group_desc *) jh->b_committed_data;
1269 	if (bg)
1270 		ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1271 	else
1272 		ret = 1;
1273 	spin_unlock(&jh->b_state_lock);
1274 	jbd2_journal_put_journal_head(jh);
1275 
1276 	return ret;
1277 }
1278 
1279 u16 ocfs2_find_max_contig_free_bits(void *bitmap,
1280 			 u16 total_bits, u16 start)
1281 {
1282 	u16 offset, free_bits;
1283 	u16 contig_bits = 0;
1284 
1285 	while (start < total_bits) {
1286 		offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start);
1287 		if (offset == total_bits)
1288 			break;
1289 
1290 		start = ocfs2_find_next_bit(bitmap, total_bits, offset);
1291 		free_bits = start - offset;
1292 		if (contig_bits < free_bits)
1293 			contig_bits = free_bits;
1294 	}
1295 
1296 	return contig_bits;
1297 }
1298 
1299 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1300 					     struct buffer_head *bg_bh,
1301 					     unsigned int bits_wanted,
1302 					     unsigned int total_bits,
1303 					     struct ocfs2_suballoc_result *res)
1304 {
1305 	void *bitmap;
1306 	u16 best_offset, best_size;
1307 	u16 prev_best_size = 0;
1308 	int offset, start, found, status = 0;
1309 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1310 
1311 	/* Callers got this descriptor from
1312 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1313 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1314 
1315 	found = start = best_offset = best_size = 0;
1316 	bitmap = bg->bg_bitmap;
1317 
1318 	while ((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) <
1319 	       total_bits) {
1320 		if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1321 			/* We found a zero, but we can't use it as it
1322 			 * hasn't been put to disk yet! */
1323 			found = 0;
1324 			start = offset + 1;
1325 		} else if (offset == start) {
1326 			/* we found a zero */
1327 			found++;
1328 			/* move start to the next bit to test */
1329 			start++;
1330 		} else {
1331 			/* got a zero after some ones */
1332 			found = 1;
1333 			start = offset + 1;
1334 			prev_best_size = best_size;
1335 		}
1336 		if (found > best_size) {
1337 			best_size = found;
1338 			best_offset = start - found;
1339 		}
1340 		/* we got everything we needed */
1341 		if (found == bits_wanted) {
1342 			/* mlog(0, "Found it all!\n"); */
1343 			break;
1344 		}
1345 	}
1346 
1347 	/* best_size will be allocated, we save prev_best_size */
1348 	res->sr_max_contig_bits = prev_best_size;
1349 	if (best_size) {
1350 		res->sr_bit_offset = best_offset;
1351 		res->sr_bits = best_size;
1352 	} else {
1353 		status = -ENOSPC;
1354 		/* No error log here -- see the comment above
1355 		 * ocfs2_test_bg_bit_allocatable */
1356 	}
1357 
1358 	return status;
1359 }
1360 
1361 int ocfs2_block_group_set_bits(handle_t *handle,
1362 					     struct inode *alloc_inode,
1363 					     struct ocfs2_group_desc *bg,
1364 					     struct buffer_head *group_bh,
1365 					     unsigned int bit_off,
1366 					     unsigned int num_bits,
1367 					     unsigned int max_contig_bits,
1368 					     int fastpath)
1369 {
1370 	int status;
1371 	void *bitmap = bg->bg_bitmap;
1372 	int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1373 	unsigned int start = bit_off + num_bits;
1374 	u16 contig_bits;
1375 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
1376 
1377 	/* All callers get the descriptor via
1378 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1379 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1380 	BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1381 
1382 	trace_ocfs2_block_group_set_bits(bit_off, num_bits);
1383 
1384 	if (ocfs2_is_cluster_bitmap(alloc_inode))
1385 		journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1386 
1387 	status = ocfs2_journal_access_gd(handle,
1388 					 INODE_CACHE(alloc_inode),
1389 					 group_bh,
1390 					 journal_type);
1391 	if (status < 0) {
1392 		mlog_errno(status);
1393 		goto bail;
1394 	}
1395 
1396 	le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1397 	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1398 		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
1399 				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
1400 				   le16_to_cpu(bg->bg_bits),
1401 				   le16_to_cpu(bg->bg_free_bits_count),
1402 				   num_bits);
1403 	}
1404 	while(num_bits--)
1405 		ocfs2_set_bit(bit_off++, bitmap);
1406 
1407 	/*
1408 	 * this is optimize path, caller set old contig value
1409 	 * in max_contig_bits to bypass finding action.
1410 	 */
1411 	if (fastpath) {
1412 		bg->bg_contig_free_bits = cpu_to_le16(max_contig_bits);
1413 	} else if (ocfs2_is_cluster_bitmap(alloc_inode)) {
1414 		/*
1415 		 * Usually, the block group bitmap allocates only 1 bit
1416 		 * at a time, while the cluster group allocates n bits
1417 		 * each time. Therefore, we only save the contig bits for
1418 		 * the cluster group.
1419 		 */
1420 		contig_bits = ocfs2_find_max_contig_free_bits(bitmap,
1421 				    le16_to_cpu(bg->bg_bits), start);
1422 		if (contig_bits > max_contig_bits)
1423 			max_contig_bits = contig_bits;
1424 		bg->bg_contig_free_bits = cpu_to_le16(max_contig_bits);
1425 		ocfs2_local_alloc_seen_free_bits(osb, max_contig_bits);
1426 	} else {
1427 		bg->bg_contig_free_bits = 0;
1428 	}
1429 
1430 	ocfs2_journal_dirty(handle, group_bh);
1431 
1432 bail:
1433 	return status;
1434 }
1435 
1436 /* find the one with the most empty bits */
1437 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1438 {
1439 	u16 curr, best;
1440 
1441 	BUG_ON(!cl->cl_next_free_rec);
1442 
1443 	best = curr = 0;
1444 	while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1445 		if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1446 		    le32_to_cpu(cl->cl_recs[best].c_free))
1447 			best = curr;
1448 		curr++;
1449 	}
1450 
1451 	BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1452 	return best;
1453 }
1454 
1455 static int ocfs2_relink_block_group(handle_t *handle,
1456 				    struct inode *alloc_inode,
1457 				    struct buffer_head *fe_bh,
1458 				    struct buffer_head *bg_bh,
1459 				    struct buffer_head *prev_bg_bh,
1460 				    u16 chain)
1461 {
1462 	int status;
1463 	/* there is a really tiny chance the journal calls could fail,
1464 	 * but we wouldn't want inconsistent blocks in *any* case. */
1465 	u64 bg_ptr, prev_bg_ptr;
1466 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1467 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1468 	struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1469 
1470 	/* The caller got these descriptors from
1471 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1472 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1473 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1474 
1475 	trace_ocfs2_relink_block_group(
1476 		(unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1477 		(unsigned long long)le64_to_cpu(bg->bg_blkno),
1478 		(unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1479 
1480 	bg_ptr = le64_to_cpu(bg->bg_next_group);
1481 	prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1482 
1483 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1484 					 prev_bg_bh,
1485 					 OCFS2_JOURNAL_ACCESS_WRITE);
1486 	if (status < 0)
1487 		goto out;
1488 
1489 	prev_bg->bg_next_group = bg->bg_next_group;
1490 	ocfs2_journal_dirty(handle, prev_bg_bh);
1491 
1492 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1493 					 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1494 	if (status < 0)
1495 		goto out_rollback_prev_bg;
1496 
1497 	bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1498 	ocfs2_journal_dirty(handle, bg_bh);
1499 
1500 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1501 					 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1502 	if (status < 0)
1503 		goto out_rollback_bg;
1504 
1505 	fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1506 	ocfs2_journal_dirty(handle, fe_bh);
1507 
1508 out:
1509 	if (status < 0)
1510 		mlog_errno(status);
1511 	return status;
1512 
1513 out_rollback_bg:
1514 	bg->bg_next_group = cpu_to_le64(bg_ptr);
1515 out_rollback_prev_bg:
1516 	prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1517 	goto out;
1518 }
1519 
1520 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1521 						     u32 wanted)
1522 {
1523 	return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1524 }
1525 
1526 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1527  * value on error. */
1528 static int ocfs2_cluster_group_search(struct inode *inode,
1529 				      struct buffer_head *group_bh,
1530 				      u32 bits_wanted, u32 min_bits,
1531 				      u64 max_block,
1532 				      struct ocfs2_suballoc_result *res)
1533 {
1534 	int search = -ENOSPC;
1535 	int ret;
1536 	u64 blkoff;
1537 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1538 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1539 	unsigned int max_bits, gd_cluster_off;
1540 
1541 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1542 
1543 	if (le16_to_cpu(gd->bg_contig_free_bits) &&
1544 	    le16_to_cpu(gd->bg_contig_free_bits) < bits_wanted)
1545 		return -ENOSPC;
1546 
1547 	/* ->bg_contig_free_bits may un-initialized, so compare again */
1548 	if (le16_to_cpu(gd->bg_free_bits_count) >= bits_wanted) {
1549 		max_bits = le16_to_cpu(gd->bg_bits);
1550 
1551 		/* Tail groups in cluster bitmaps which aren't cpg
1552 		 * aligned are prone to partial extension by a failed
1553 		 * fs resize. If the file system resize never got to
1554 		 * update the dinode cluster count, then we don't want
1555 		 * to trust any clusters past it, regardless of what
1556 		 * the group descriptor says. */
1557 		gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1558 							  le64_to_cpu(gd->bg_blkno));
1559 		if ((gd_cluster_off + max_bits) >
1560 		    OCFS2_I(inode)->ip_clusters) {
1561 			max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1562 			trace_ocfs2_cluster_group_search_wrong_max_bits(
1563 				(unsigned long long)le64_to_cpu(gd->bg_blkno),
1564 				le16_to_cpu(gd->bg_bits),
1565 				OCFS2_I(inode)->ip_clusters, max_bits);
1566 		}
1567 
1568 		ret = ocfs2_block_group_find_clear_bits(osb,
1569 							group_bh, bits_wanted,
1570 							max_bits, res);
1571 		if (ret)
1572 			return ret;
1573 
1574 		if (max_block) {
1575 			blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1576 							  gd_cluster_off +
1577 							  res->sr_bit_offset +
1578 							  res->sr_bits);
1579 			trace_ocfs2_cluster_group_search_max_block(
1580 				(unsigned long long)blkoff,
1581 				(unsigned long long)max_block);
1582 			if (blkoff > max_block)
1583 				return -ENOSPC;
1584 		}
1585 
1586 		/* ocfs2_block_group_find_clear_bits() might
1587 		 * return success, but we still want to return
1588 		 * -ENOSPC unless it found the minimum number
1589 		 * of bits. */
1590 		if (min_bits <= res->sr_bits)
1591 			search = 0; /* success */
1592 	}
1593 
1594 	return search;
1595 }
1596 
1597 static int ocfs2_block_group_search(struct inode *inode,
1598 				    struct buffer_head *group_bh,
1599 				    u32 bits_wanted, u32 min_bits,
1600 				    u64 max_block,
1601 				    struct ocfs2_suballoc_result *res)
1602 {
1603 	int ret = -ENOSPC;
1604 	u64 blkoff;
1605 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1606 
1607 	BUG_ON(min_bits != 1);
1608 	BUG_ON(ocfs2_is_cluster_bitmap(inode));
1609 
1610 	if (le16_to_cpu(bg->bg_free_bits_count) >= bits_wanted) {
1611 		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1612 							group_bh, bits_wanted,
1613 							le16_to_cpu(bg->bg_bits),
1614 							res);
1615 		if (!ret && max_block) {
1616 			blkoff = le64_to_cpu(bg->bg_blkno) +
1617 				res->sr_bit_offset + res->sr_bits;
1618 			trace_ocfs2_block_group_search_max_block(
1619 				(unsigned long long)blkoff,
1620 				(unsigned long long)max_block);
1621 			if (blkoff > max_block)
1622 				ret = -ENOSPC;
1623 		}
1624 	}
1625 
1626 	return ret;
1627 }
1628 
1629 int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1630 				       handle_t *handle,
1631 				       struct buffer_head *di_bh,
1632 				       u32 num_bits,
1633 				       u16 chain)
1634 {
1635 	int ret;
1636 	u32 tmp_used;
1637 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1638 	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1639 
1640 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1641 				      OCFS2_JOURNAL_ACCESS_WRITE);
1642 	if (ret < 0) {
1643 		mlog_errno(ret);
1644 		goto out;
1645 	}
1646 
1647 	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1648 	di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1649 	le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1650 	ocfs2_journal_dirty(handle, di_bh);
1651 
1652 out:
1653 	return ret;
1654 }
1655 
1656 void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
1657 				       struct buffer_head *di_bh,
1658 				       u32 num_bits,
1659 				       u16 chain)
1660 {
1661 	u32 tmp_used;
1662 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1663 	struct ocfs2_chain_list *cl;
1664 
1665 	cl = (struct ocfs2_chain_list *)&di->id2.i_chain;
1666 	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1667 	di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits);
1668 	le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits);
1669 }
1670 
1671 static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1672 					 struct ocfs2_extent_rec *rec,
1673 					 struct ocfs2_chain_list *cl)
1674 {
1675 	unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1676 	unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1677 	unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1678 
1679 	if (res->sr_bit_offset < bitoff)
1680 		return 0;
1681 	if (res->sr_bit_offset >= (bitoff + bitcount))
1682 		return 0;
1683 	res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1684 		(res->sr_bit_offset - bitoff);
1685 	if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1686 		res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1687 	return 1;
1688 }
1689 
1690 static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1691 					  struct ocfs2_group_desc *bg,
1692 					  struct ocfs2_suballoc_result *res)
1693 {
1694 	int i;
1695 	u64 bg_blkno = res->sr_bg_blkno;  /* Save off */
1696 	struct ocfs2_extent_rec *rec;
1697 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1698 	struct ocfs2_chain_list *cl = &di->id2.i_chain;
1699 
1700 	if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1701 		res->sr_blkno = 0;
1702 		return;
1703 	}
1704 
1705 	res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1706 	res->sr_bg_blkno = 0;  /* Clear it for contig block groups */
1707 	if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1708 	    !bg->bg_list.l_next_free_rec)
1709 		return;
1710 
1711 	for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1712 		rec = &bg->bg_list.l_recs[i];
1713 		if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1714 			res->sr_bg_blkno = bg_blkno;  /* Restore */
1715 			break;
1716 		}
1717 	}
1718 }
1719 
1720 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1721 				  handle_t *handle,
1722 				  u32 bits_wanted,
1723 				  u32 min_bits,
1724 				  struct ocfs2_suballoc_result *res,
1725 				  u16 *bits_left)
1726 {
1727 	int ret;
1728 	struct buffer_head *group_bh = NULL;
1729 	struct ocfs2_group_desc *gd;
1730 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1731 	struct inode *alloc_inode = ac->ac_inode;
1732 
1733 	ret = ocfs2_read_group_descriptor(alloc_inode, di,
1734 					  res->sr_bg_blkno, &group_bh);
1735 	if (ret < 0) {
1736 		mlog_errno(ret);
1737 		return ret;
1738 	}
1739 
1740 	gd = (struct ocfs2_group_desc *) group_bh->b_data;
1741 	ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1742 				  ac->ac_max_block, res);
1743 	if (ret < 0) {
1744 		if (ret != -ENOSPC)
1745 			mlog_errno(ret);
1746 		goto out;
1747 	}
1748 
1749 	if (!ret)
1750 		ocfs2_bg_discontig_fix_result(ac, gd, res);
1751 
1752 	/*
1753 	 * sr_bg_blkno might have been changed by
1754 	 * ocfs2_bg_discontig_fix_result
1755 	 */
1756 	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1757 
1758 	if (ac->ac_find_loc_only)
1759 		goto out_loc_only;
1760 
1761 	ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1762 					       res->sr_bits,
1763 					       le16_to_cpu(gd->bg_chain));
1764 	if (ret < 0) {
1765 		mlog_errno(ret);
1766 		goto out;
1767 	}
1768 
1769 	ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1770 					 res->sr_bit_offset, res->sr_bits,
1771 					 res->sr_max_contig_bits, 0);
1772 	if (ret < 0) {
1773 		ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh,
1774 					       res->sr_bits,
1775 					       le16_to_cpu(gd->bg_chain));
1776 		mlog_errno(ret);
1777 	}
1778 
1779 out_loc_only:
1780 	*bits_left = le16_to_cpu(gd->bg_free_bits_count);
1781 
1782 out:
1783 	brelse(group_bh);
1784 
1785 	return ret;
1786 }
1787 
1788 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1789 			      handle_t *handle,
1790 			      u32 bits_wanted,
1791 			      u32 min_bits,
1792 			      struct ocfs2_suballoc_result *res,
1793 			      u16 *bits_left)
1794 {
1795 	int status;
1796 	u16 chain;
1797 	u64 next_group;
1798 	struct inode *alloc_inode = ac->ac_inode;
1799 	struct buffer_head *group_bh = NULL;
1800 	struct buffer_head *prev_group_bh = NULL;
1801 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1802 	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1803 	struct ocfs2_group_desc *bg;
1804 
1805 	chain = ac->ac_chain;
1806 	trace_ocfs2_search_chain_begin(
1807 		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1808 		bits_wanted, chain);
1809 
1810 	status = ocfs2_read_group_descriptor(alloc_inode, fe,
1811 					     le64_to_cpu(cl->cl_recs[chain].c_blkno),
1812 					     &group_bh);
1813 	if (status < 0) {
1814 		mlog_errno(status);
1815 		goto bail;
1816 	}
1817 	bg = (struct ocfs2_group_desc *) group_bh->b_data;
1818 
1819 	status = -ENOSPC;
1820 	/* for now, the chain search is a bit simplistic. We just use
1821 	 * the 1st group with any empty bits. */
1822 	while ((status = ac->ac_group_search(alloc_inode, group_bh,
1823 					     bits_wanted, min_bits,
1824 					     ac->ac_max_block,
1825 					     res)) == -ENOSPC) {
1826 		if (!bg->bg_next_group)
1827 			break;
1828 
1829 		brelse(prev_group_bh);
1830 		prev_group_bh = NULL;
1831 
1832 		next_group = le64_to_cpu(bg->bg_next_group);
1833 		prev_group_bh = group_bh;
1834 		group_bh = NULL;
1835 		status = ocfs2_read_group_descriptor(alloc_inode, fe,
1836 						     next_group, &group_bh);
1837 		if (status < 0) {
1838 			mlog_errno(status);
1839 			goto bail;
1840 		}
1841 		bg = (struct ocfs2_group_desc *) group_bh->b_data;
1842 	}
1843 	if (status < 0) {
1844 		if (status != -ENOSPC)
1845 			mlog_errno(status);
1846 		goto bail;
1847 	}
1848 
1849 	trace_ocfs2_search_chain_succ(
1850 		(unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1851 
1852 	res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1853 
1854 	BUG_ON(res->sr_bits == 0);
1855 	if (!status)
1856 		ocfs2_bg_discontig_fix_result(ac, bg, res);
1857 
1858 	/*
1859 	 * sr_bg_blkno might have been changed by
1860 	 * ocfs2_bg_discontig_fix_result
1861 	 */
1862 	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1863 
1864 	/*
1865 	 * Keep track of previous block descriptor read. When
1866 	 * we find a target, if we have read more than X
1867 	 * number of descriptors, and the target is reasonably
1868 	 * empty, relink him to top of his chain.
1869 	 *
1870 	 * We've read 0 extra blocks and only send one more to
1871 	 * the transaction, yet the next guy to search has a
1872 	 * much easier time.
1873 	 *
1874 	 * Do this *after* figuring out how many bits we're taking out
1875 	 * of our target group.
1876 	 */
1877 	if (!ac->ac_disable_chain_relink &&
1878 	    (prev_group_bh) &&
1879 	    (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1880 		status = ocfs2_relink_block_group(handle, alloc_inode,
1881 						  ac->ac_bh, group_bh,
1882 						  prev_group_bh, chain);
1883 		if (status < 0) {
1884 			mlog_errno(status);
1885 			goto bail;
1886 		}
1887 	}
1888 
1889 	if (ac->ac_find_loc_only)
1890 		goto out_loc_only;
1891 
1892 	status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1893 						  ac->ac_bh, res->sr_bits,
1894 						  chain);
1895 	if (status) {
1896 		mlog_errno(status);
1897 		goto bail;
1898 	}
1899 
1900 	status = ocfs2_block_group_set_bits(handle,
1901 					    alloc_inode,
1902 					    bg,
1903 					    group_bh,
1904 					    res->sr_bit_offset,
1905 					    res->sr_bits,
1906 					    res->sr_max_contig_bits,
1907 					    0);
1908 	if (status < 0) {
1909 		ocfs2_rollback_alloc_dinode_counts(alloc_inode,
1910 					ac->ac_bh, res->sr_bits, chain);
1911 		mlog_errno(status);
1912 		goto bail;
1913 	}
1914 
1915 	trace_ocfs2_search_chain_end(
1916 			(unsigned long long)le64_to_cpu(fe->i_blkno),
1917 			res->sr_bits);
1918 
1919 out_loc_only:
1920 	*bits_left = le16_to_cpu(bg->bg_free_bits_count);
1921 bail:
1922 	brelse(group_bh);
1923 	brelse(prev_group_bh);
1924 
1925 	if (status)
1926 		mlog_errno(status);
1927 	return status;
1928 }
1929 
1930 /* will give out up to bits_wanted contiguous bits. */
1931 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1932 				     handle_t *handle,
1933 				     u32 bits_wanted,
1934 				     u32 min_bits,
1935 				     struct ocfs2_suballoc_result *res)
1936 {
1937 	int status;
1938 	u16 victim, i;
1939 	u16 bits_left = 0;
1940 	u64 hint = ac->ac_last_group;
1941 	struct ocfs2_chain_list *cl;
1942 	struct ocfs2_dinode *fe;
1943 
1944 	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1945 	BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1946 	BUG_ON(!ac->ac_bh);
1947 
1948 	fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1949 
1950 	/* The bh was validated by the inode read during
1951 	 * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1952 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1953 
1954 	if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1955 	    le32_to_cpu(fe->id1.bitmap1.i_total)) {
1956 		status = ocfs2_error(ac->ac_inode->i_sb,
1957 				     "Chain allocator dinode %llu has %u used bits but only %u total\n",
1958 				     (unsigned long long)le64_to_cpu(fe->i_blkno),
1959 				     le32_to_cpu(fe->id1.bitmap1.i_used),
1960 				     le32_to_cpu(fe->id1.bitmap1.i_total));
1961 		goto bail;
1962 	}
1963 
1964 	res->sr_bg_blkno = hint;
1965 	if (res->sr_bg_blkno) {
1966 		/* Attempt to short-circuit the usual search mechanism
1967 		 * by jumping straight to the most recently used
1968 		 * allocation group. This helps us maintain some
1969 		 * contiguousness across allocations. */
1970 		status = ocfs2_search_one_group(ac, handle, bits_wanted,
1971 						min_bits, res, &bits_left);
1972 		if (!status)
1973 			goto set_hint;
1974 		if (status < 0 && status != -ENOSPC) {
1975 			mlog_errno(status);
1976 			goto bail;
1977 		}
1978 	}
1979 
1980 	cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1981 
1982 	victim = ocfs2_find_victim_chain(cl);
1983 	ac->ac_chain = victim;
1984 
1985 	status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1986 				    res, &bits_left);
1987 	if (!status) {
1988 		if (ocfs2_is_cluster_bitmap(ac->ac_inode))
1989 			hint = res->sr_bg_blkno;
1990 		else
1991 			hint = ocfs2_group_from_res(res);
1992 		goto set_hint;
1993 	}
1994 	if (status < 0 && status != -ENOSPC) {
1995 		mlog_errno(status);
1996 		goto bail;
1997 	}
1998 
1999 	trace_ocfs2_claim_suballoc_bits(victim);
2000 
2001 	/* If we didn't pick a good victim, then just default to
2002 	 * searching each chain in order. Don't allow chain relinking
2003 	 * because we only calculate enough journal credits for one
2004 	 * relink per alloc. */
2005 	ac->ac_disable_chain_relink = 1;
2006 	for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
2007 		if (i == victim)
2008 			continue;
2009 		if (le32_to_cpu(cl->cl_recs[i].c_free) < bits_wanted)
2010 			continue;
2011 
2012 		ac->ac_chain = i;
2013 		status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
2014 					    res, &bits_left);
2015 		if (!status) {
2016 			hint = ocfs2_group_from_res(res);
2017 			break;
2018 		}
2019 		if (status < 0 && status != -ENOSPC) {
2020 			mlog_errno(status);
2021 			goto bail;
2022 		}
2023 	}
2024 
2025 set_hint:
2026 	if (status != -ENOSPC) {
2027 		/* If the next search of this group is not likely to
2028 		 * yield a suitable extent, then we reset the last
2029 		 * group hint so as to not waste a disk read */
2030 		if (bits_left < min_bits)
2031 			ac->ac_last_group = 0;
2032 		else
2033 			ac->ac_last_group = hint;
2034 	}
2035 
2036 bail:
2037 	if (status)
2038 		mlog_errno(status);
2039 	return status;
2040 }
2041 
2042 int ocfs2_claim_metadata(handle_t *handle,
2043 			 struct ocfs2_alloc_context *ac,
2044 			 u32 bits_wanted,
2045 			 u64 *suballoc_loc,
2046 			 u16 *suballoc_bit_start,
2047 			 unsigned int *num_bits,
2048 			 u64 *blkno_start)
2049 {
2050 	int status;
2051 	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2052 
2053 	BUG_ON(!ac);
2054 	BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
2055 	BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
2056 
2057 	status = ocfs2_claim_suballoc_bits(ac,
2058 					   handle,
2059 					   bits_wanted,
2060 					   1,
2061 					   &res);
2062 	if (status < 0) {
2063 		mlog_errno(status);
2064 		goto bail;
2065 	}
2066 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2067 
2068 	*suballoc_loc = res.sr_bg_blkno;
2069 	*suballoc_bit_start = res.sr_bit_offset;
2070 	*blkno_start = res.sr_blkno;
2071 	ac->ac_bits_given += res.sr_bits;
2072 	*num_bits = res.sr_bits;
2073 	status = 0;
2074 bail:
2075 	if (status)
2076 		mlog_errno(status);
2077 	return status;
2078 }
2079 
2080 static void ocfs2_init_inode_ac_group(struct inode *dir,
2081 				      struct buffer_head *parent_di_bh,
2082 				      struct ocfs2_alloc_context *ac)
2083 {
2084 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2085 	/*
2086 	 * Try to allocate inodes from some specific group.
2087 	 *
2088 	 * If the parent dir has recorded the last group used in allocation,
2089 	 * cool, use it. Otherwise if we try to allocate new inode from the
2090 	 * same slot the parent dir belongs to, use the same chunk.
2091 	 *
2092 	 * We are very careful here to avoid the mistake of setting
2093 	 * ac_last_group to a group descriptor from a different (unlocked) slot.
2094 	 */
2095 	if (OCFS2_I(dir)->ip_last_used_group &&
2096 	    OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2097 		ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2098 	else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2099 		if (di->i_suballoc_loc)
2100 			ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2101 		else
2102 			ac->ac_last_group = ocfs2_which_suballoc_group(
2103 					le64_to_cpu(di->i_blkno),
2104 					le16_to_cpu(di->i_suballoc_bit));
2105 	}
2106 }
2107 
2108 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2109 					     struct ocfs2_alloc_context *ac)
2110 {
2111 	OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2112 	OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2113 }
2114 
2115 int ocfs2_find_new_inode_loc(struct inode *dir,
2116 			     struct buffer_head *parent_fe_bh,
2117 			     struct ocfs2_alloc_context *ac,
2118 			     u64 *fe_blkno)
2119 {
2120 	int ret;
2121 	handle_t *handle = NULL;
2122 	struct ocfs2_suballoc_result *res;
2123 
2124 	BUG_ON(!ac);
2125 	BUG_ON(ac->ac_bits_given != 0);
2126 	BUG_ON(ac->ac_bits_wanted != 1);
2127 	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2128 
2129 	res = kzalloc(sizeof(*res), GFP_NOFS);
2130 	if (res == NULL) {
2131 		ret = -ENOMEM;
2132 		mlog_errno(ret);
2133 		goto out;
2134 	}
2135 
2136 	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2137 
2138 	/*
2139 	 * The handle started here is for chain relink. Alternatively,
2140 	 * we could just disable relink for these calls.
2141 	 */
2142 	handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2143 	if (IS_ERR(handle)) {
2144 		ret = PTR_ERR(handle);
2145 		handle = NULL;
2146 		mlog_errno(ret);
2147 		goto out;
2148 	}
2149 
2150 	/*
2151 	 * This will instruct ocfs2_claim_suballoc_bits and
2152 	 * ocfs2_search_one_group to search but save actual allocation
2153 	 * for later.
2154 	 */
2155 	ac->ac_find_loc_only = 1;
2156 
2157 	ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2158 	if (ret < 0) {
2159 		mlog_errno(ret);
2160 		goto out;
2161 	}
2162 
2163 	ac->ac_find_loc_priv = res;
2164 	*fe_blkno = res->sr_blkno;
2165 	ocfs2_update_inode_fsync_trans(handle, dir, 0);
2166 out:
2167 	if (handle)
2168 		ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2169 
2170 	if (ret)
2171 		kfree(res);
2172 
2173 	return ret;
2174 }
2175 
2176 int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2177 				 struct inode *dir,
2178 				 struct ocfs2_alloc_context *ac,
2179 				 u64 *suballoc_loc,
2180 				 u16 *suballoc_bit,
2181 				 u64 di_blkno)
2182 {
2183 	int ret;
2184 	u16 chain;
2185 	struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2186 	struct buffer_head *bg_bh = NULL;
2187 	struct ocfs2_group_desc *bg;
2188 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2189 
2190 	/*
2191 	 * Since di_blkno is being passed back in, we check for any
2192 	 * inconsistencies which may have happened between
2193 	 * calls. These are code bugs as di_blkno is not expected to
2194 	 * change once returned from ocfs2_find_new_inode_loc()
2195 	 */
2196 	BUG_ON(res->sr_blkno != di_blkno);
2197 
2198 	ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2199 					  res->sr_bg_stable_blkno, &bg_bh);
2200 	if (ret) {
2201 		mlog_errno(ret);
2202 		goto out;
2203 	}
2204 
2205 	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2206 	chain = le16_to_cpu(bg->bg_chain);
2207 
2208 	ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2209 					       ac->ac_bh, res->sr_bits,
2210 					       chain);
2211 	if (ret) {
2212 		mlog_errno(ret);
2213 		goto out;
2214 	}
2215 
2216 	ret = ocfs2_block_group_set_bits(handle,
2217 					 ac->ac_inode,
2218 					 bg,
2219 					 bg_bh,
2220 					 res->sr_bit_offset,
2221 					 res->sr_bits,
2222 					 res->sr_max_contig_bits,
2223 					 0);
2224 	if (ret < 0) {
2225 		ocfs2_rollback_alloc_dinode_counts(ac->ac_inode,
2226 					       ac->ac_bh, res->sr_bits, chain);
2227 		mlog_errno(ret);
2228 		goto out;
2229 	}
2230 
2231 	trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2232 					   res->sr_bits);
2233 
2234 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2235 
2236 	BUG_ON(res->sr_bits != 1);
2237 
2238 	*suballoc_loc = res->sr_bg_blkno;
2239 	*suballoc_bit = res->sr_bit_offset;
2240 	ac->ac_bits_given++;
2241 	ocfs2_save_inode_ac_group(dir, ac);
2242 
2243 out:
2244 	brelse(bg_bh);
2245 
2246 	return ret;
2247 }
2248 
2249 int ocfs2_claim_new_inode(handle_t *handle,
2250 			  struct inode *dir,
2251 			  struct buffer_head *parent_fe_bh,
2252 			  struct ocfs2_alloc_context *ac,
2253 			  u64 *suballoc_loc,
2254 			  u16 *suballoc_bit,
2255 			  u64 *fe_blkno)
2256 {
2257 	int status;
2258 	struct ocfs2_suballoc_result res;
2259 
2260 	BUG_ON(!ac);
2261 	BUG_ON(ac->ac_bits_given != 0);
2262 	BUG_ON(ac->ac_bits_wanted != 1);
2263 	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2264 
2265 	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2266 
2267 	status = ocfs2_claim_suballoc_bits(ac,
2268 					   handle,
2269 					   1,
2270 					   1,
2271 					   &res);
2272 	if (status < 0) {
2273 		mlog_errno(status);
2274 		goto bail;
2275 	}
2276 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2277 
2278 	BUG_ON(res.sr_bits != 1);
2279 
2280 	*suballoc_loc = res.sr_bg_blkno;
2281 	*suballoc_bit = res.sr_bit_offset;
2282 	*fe_blkno = res.sr_blkno;
2283 	ac->ac_bits_given++;
2284 	ocfs2_save_inode_ac_group(dir, ac);
2285 	status = 0;
2286 bail:
2287 	if (status)
2288 		mlog_errno(status);
2289 	return status;
2290 }
2291 
2292 /* translate a group desc. blkno and it's bitmap offset into
2293  * disk cluster offset. */
2294 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2295 						   u64 bg_blkno,
2296 						   u16 bg_bit_off)
2297 {
2298 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2299 	u32 cluster = 0;
2300 
2301 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2302 
2303 	if (bg_blkno != osb->first_cluster_group_blkno)
2304 		cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2305 	cluster += (u32) bg_bit_off;
2306 	return cluster;
2307 }
2308 
2309 /* given a cluster offset, calculate which block group it belongs to
2310  * and return that block offset. */
2311 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2312 {
2313 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2314 	u32 group_no;
2315 
2316 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2317 
2318 	group_no = cluster / osb->bitmap_cpg;
2319 	if (!group_no)
2320 		return osb->first_cluster_group_blkno;
2321 	return ocfs2_clusters_to_blocks(inode->i_sb,
2322 					group_no * osb->bitmap_cpg);
2323 }
2324 
2325 /* given the block number of a cluster start, calculate which cluster
2326  * group and descriptor bitmap offset that corresponds to. */
2327 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2328 						u64 data_blkno,
2329 						u64 *bg_blkno,
2330 						u16 *bg_bit_off)
2331 {
2332 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2333 	u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2334 
2335 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2336 
2337 	*bg_blkno = ocfs2_which_cluster_group(inode,
2338 					      data_cluster);
2339 
2340 	if (*bg_blkno == osb->first_cluster_group_blkno)
2341 		*bg_bit_off = (u16) data_cluster;
2342 	else
2343 		*bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2344 							     data_blkno - *bg_blkno);
2345 }
2346 
2347 /*
2348  * min_bits - minimum contiguous chunk from this total allocation we
2349  * can handle. set to what we asked for originally for a full
2350  * contig. allocation, set to '1' to indicate we can deal with extents
2351  * of any size.
2352  */
2353 int __ocfs2_claim_clusters(handle_t *handle,
2354 			   struct ocfs2_alloc_context *ac,
2355 			   u32 min_clusters,
2356 			   u32 max_clusters,
2357 			   u32 *cluster_start,
2358 			   u32 *num_clusters)
2359 {
2360 	int status;
2361 	unsigned int bits_wanted = max_clusters;
2362 	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2363 	struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2364 
2365 	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2366 
2367 	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2368 	       && ac->ac_which != OCFS2_AC_USE_MAIN);
2369 
2370 	if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2371 		WARN_ON(min_clusters > 1);
2372 
2373 		status = ocfs2_claim_local_alloc_bits(osb,
2374 						      handle,
2375 						      ac,
2376 						      bits_wanted,
2377 						      cluster_start,
2378 						      num_clusters);
2379 		if (!status)
2380 			atomic_inc(&osb->alloc_stats.local_data);
2381 	} else {
2382 		if (min_clusters > (osb->bitmap_cpg - 1)) {
2383 			/* The only paths asking for contiguousness
2384 			 * should know about this already. */
2385 			mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2386 			     "group bitmap size %u!\n", min_clusters,
2387 			     osb->bitmap_cpg);
2388 			status = -ENOSPC;
2389 			goto bail;
2390 		}
2391 		/* clamp the current request down to a realistic size. */
2392 		if (bits_wanted > (osb->bitmap_cpg - 1))
2393 			bits_wanted = osb->bitmap_cpg - 1;
2394 
2395 		status = ocfs2_claim_suballoc_bits(ac,
2396 						   handle,
2397 						   bits_wanted,
2398 						   min_clusters,
2399 						   &res);
2400 		if (!status) {
2401 			BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2402 			*cluster_start =
2403 				ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2404 								 res.sr_bg_blkno,
2405 								 res.sr_bit_offset);
2406 			atomic_inc(&osb->alloc_stats.bitmap_data);
2407 			*num_clusters = res.sr_bits;
2408 		}
2409 	}
2410 	if (status < 0) {
2411 		if (status != -ENOSPC)
2412 			mlog_errno(status);
2413 		goto bail;
2414 	}
2415 
2416 	ac->ac_bits_given += *num_clusters;
2417 
2418 bail:
2419 	if (status)
2420 		mlog_errno(status);
2421 	return status;
2422 }
2423 
2424 int ocfs2_claim_clusters(handle_t *handle,
2425 			 struct ocfs2_alloc_context *ac,
2426 			 u32 min_clusters,
2427 			 u32 *cluster_start,
2428 			 u32 *num_clusters)
2429 {
2430 	unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2431 
2432 	return __ocfs2_claim_clusters(handle, ac, min_clusters,
2433 				      bits_wanted, cluster_start, num_clusters);
2434 }
2435 
2436 static int ocfs2_block_group_clear_bits(handle_t *handle,
2437 					struct inode *alloc_inode,
2438 					struct ocfs2_group_desc *bg,
2439 					struct buffer_head *group_bh,
2440 					unsigned int bit_off,
2441 					unsigned int num_bits,
2442 					unsigned int max_contig_bits,
2443 					void (*undo_fn)(unsigned int bit,
2444 							unsigned long *bmap))
2445 {
2446 	int status;
2447 	unsigned int tmp;
2448 	u16 contig_bits;
2449 	struct ocfs2_group_desc *undo_bg = NULL;
2450 	struct journal_head *jh;
2451 
2452 	/* The caller got this descriptor from
2453 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
2454 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2455 
2456 	trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2457 
2458 	BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2459 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2460 					 group_bh,
2461 					 undo_fn ?
2462 					 OCFS2_JOURNAL_ACCESS_UNDO :
2463 					 OCFS2_JOURNAL_ACCESS_WRITE);
2464 	if (status < 0) {
2465 		mlog_errno(status);
2466 		goto bail;
2467 	}
2468 
2469 	jh = bh2jh(group_bh);
2470 	if (undo_fn) {
2471 		spin_lock(&jh->b_state_lock);
2472 		undo_bg = (struct ocfs2_group_desc *) jh->b_committed_data;
2473 		BUG_ON(!undo_bg);
2474 	}
2475 
2476 	tmp = num_bits;
2477 	while(tmp--) {
2478 		ocfs2_clear_bit((bit_off + tmp),
2479 				(unsigned long *) bg->bg_bitmap);
2480 		if (undo_fn)
2481 			undo_fn(bit_off + tmp,
2482 				(unsigned long *) undo_bg->bg_bitmap);
2483 	}
2484 	le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2485 	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2486 		if (undo_fn)
2487 			spin_unlock(&jh->b_state_lock);
2488 		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
2489 				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
2490 				   le16_to_cpu(bg->bg_bits),
2491 				   le16_to_cpu(bg->bg_free_bits_count),
2492 				   num_bits);
2493 	}
2494 
2495 	/*
2496 	 * TODO: even 'num_bits == 1' (the worst case, release 1 cluster),
2497 	 * we still need to rescan whole bitmap.
2498 	 */
2499 	if (ocfs2_is_cluster_bitmap(alloc_inode)) {
2500 		contig_bits = ocfs2_find_max_contig_free_bits(bg->bg_bitmap,
2501 				    le16_to_cpu(bg->bg_bits), 0);
2502 		if (contig_bits > max_contig_bits)
2503 			max_contig_bits = contig_bits;
2504 		bg->bg_contig_free_bits = cpu_to_le16(max_contig_bits);
2505 	} else {
2506 		bg->bg_contig_free_bits = 0;
2507 	}
2508 
2509 	if (undo_fn)
2510 		spin_unlock(&jh->b_state_lock);
2511 
2512 	ocfs2_journal_dirty(handle, group_bh);
2513 bail:
2514 	return status;
2515 }
2516 
2517 /*
2518  * expects the suballoc inode to already be locked.
2519  */
2520 static int _ocfs2_free_suballoc_bits(handle_t *handle,
2521 				     struct inode *alloc_inode,
2522 				     struct buffer_head *alloc_bh,
2523 				     unsigned int start_bit,
2524 				     u64 bg_blkno,
2525 				     unsigned int count,
2526 				     void (*undo_fn)(unsigned int bit,
2527 						     unsigned long *bitmap))
2528 {
2529 	int status = 0;
2530 	u32 tmp_used;
2531 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2532 	struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2533 	struct buffer_head *group_bh = NULL;
2534 	struct ocfs2_group_desc *group;
2535 	__le16 old_bg_contig_free_bits = 0;
2536 
2537 	/* The alloc_bh comes from ocfs2_free_dinode() or
2538 	 * ocfs2_free_clusters().  The callers have all locked the
2539 	 * allocator and gotten alloc_bh from the lock call.  This
2540 	 * validates the dinode buffer.  Any corruption that has happened
2541 	 * is a code bug. */
2542 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2543 	BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2544 
2545 	trace_ocfs2_free_suballoc_bits(
2546 		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2547 		(unsigned long long)bg_blkno,
2548 		start_bit, count);
2549 
2550 	status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2551 					     &group_bh);
2552 	if (status < 0) {
2553 		mlog_errno(status);
2554 		goto bail;
2555 	}
2556 	group = (struct ocfs2_group_desc *) group_bh->b_data;
2557 
2558 	BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2559 
2560 	if (ocfs2_is_cluster_bitmap(alloc_inode))
2561 		old_bg_contig_free_bits = group->bg_contig_free_bits;
2562 	status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2563 					      group, group_bh,
2564 					      start_bit, count, 0, undo_fn);
2565 	if (status < 0) {
2566 		mlog_errno(status);
2567 		goto bail;
2568 	}
2569 
2570 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2571 					 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2572 	if (status < 0) {
2573 		mlog_errno(status);
2574 		ocfs2_block_group_set_bits(handle, alloc_inode, group, group_bh,
2575 				start_bit, count,
2576 				le16_to_cpu(old_bg_contig_free_bits), 1);
2577 		goto bail;
2578 	}
2579 
2580 	le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2581 		     count);
2582 	tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2583 	fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2584 	ocfs2_journal_dirty(handle, alloc_bh);
2585 
2586 bail:
2587 	brelse(group_bh);
2588 	return status;
2589 }
2590 
2591 int ocfs2_free_suballoc_bits(handle_t *handle,
2592 			     struct inode *alloc_inode,
2593 			     struct buffer_head *alloc_bh,
2594 			     unsigned int start_bit,
2595 			     u64 bg_blkno,
2596 			     unsigned int count)
2597 {
2598 	return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2599 					 start_bit, bg_blkno, count, NULL);
2600 }
2601 
2602 int ocfs2_free_dinode(handle_t *handle,
2603 		      struct inode *inode_alloc_inode,
2604 		      struct buffer_head *inode_alloc_bh,
2605 		      struct ocfs2_dinode *di)
2606 {
2607 	u64 blk = le64_to_cpu(di->i_blkno);
2608 	u16 bit = le16_to_cpu(di->i_suballoc_bit);
2609 	u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2610 
2611 	if (di->i_suballoc_loc)
2612 		bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2613 	return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2614 					inode_alloc_bh, bit, bg_blkno, 1);
2615 }
2616 
2617 static int _ocfs2_free_clusters(handle_t *handle,
2618 				struct inode *bitmap_inode,
2619 				struct buffer_head *bitmap_bh,
2620 				u64 start_blk,
2621 				unsigned int num_clusters,
2622 				void (*undo_fn)(unsigned int bit,
2623 						unsigned long *bitmap))
2624 {
2625 	int status;
2626 	u16 bg_start_bit;
2627 	u64 bg_blkno;
2628 
2629 	/* You can't ever have a contiguous set of clusters
2630 	 * bigger than a block group bitmap so we never have to worry
2631 	 * about looping on them.
2632 	 * This is expensive. We can safely remove once this stuff has
2633 	 * gotten tested really well. */
2634 	BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb,
2635 				ocfs2_blocks_to_clusters(bitmap_inode->i_sb,
2636 							 start_blk)));
2637 
2638 
2639 	ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2640 				     &bg_start_bit);
2641 
2642 	trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2643 			(unsigned long long)start_blk,
2644 			bg_start_bit, num_clusters);
2645 
2646 	status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2647 					   bg_start_bit, bg_blkno,
2648 					   num_clusters, undo_fn);
2649 	if (status < 0) {
2650 		mlog_errno(status);
2651 		goto out;
2652 	}
2653 
2654 	ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2655 					 num_clusters);
2656 
2657 out:
2658 	return status;
2659 }
2660 
2661 int ocfs2_free_clusters(handle_t *handle,
2662 			struct inode *bitmap_inode,
2663 			struct buffer_head *bitmap_bh,
2664 			u64 start_blk,
2665 			unsigned int num_clusters)
2666 {
2667 	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2668 				    start_blk, num_clusters,
2669 				    _ocfs2_set_bit);
2670 }
2671 
2672 /*
2673  * Give never-used clusters back to the global bitmap.  We don't need
2674  * to protect these bits in the undo buffer.
2675  */
2676 int ocfs2_release_clusters(handle_t *handle,
2677 			   struct inode *bitmap_inode,
2678 			   struct buffer_head *bitmap_bh,
2679 			   u64 start_blk,
2680 			   unsigned int num_clusters)
2681 {
2682 	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2683 				    start_blk, num_clusters,
2684 				    _ocfs2_clear_bit);
2685 }
2686 
2687 /*
2688  * For a given allocation, determine which allocators will need to be
2689  * accessed, and lock them, reserving the appropriate number of bits.
2690  *
2691  * Sparse file systems call this from ocfs2_write_begin_nolock()
2692  * and ocfs2_allocate_unwritten_extents().
2693  *
2694  * File systems which don't support holes call this from
2695  * ocfs2_extend_allocation().
2696  */
2697 int ocfs2_lock_allocators(struct inode *inode,
2698 			  struct ocfs2_extent_tree *et,
2699 			  u32 clusters_to_add, u32 extents_to_split,
2700 			  struct ocfs2_alloc_context **data_ac,
2701 			  struct ocfs2_alloc_context **meta_ac)
2702 {
2703 	int ret = 0, num_free_extents;
2704 	unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2705 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2706 
2707 	*meta_ac = NULL;
2708 	if (data_ac)
2709 		*data_ac = NULL;
2710 
2711 	BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2712 
2713 	num_free_extents = ocfs2_num_free_extents(et);
2714 	if (num_free_extents < 0) {
2715 		ret = num_free_extents;
2716 		mlog_errno(ret);
2717 		goto out;
2718 	}
2719 
2720 	/*
2721 	 * Sparse allocation file systems need to be more conservative
2722 	 * with reserving room for expansion - the actual allocation
2723 	 * happens while we've got a journal handle open so re-taking
2724 	 * a cluster lock (because we ran out of room for another
2725 	 * extent) will violate ordering rules.
2726 	 *
2727 	 * Most of the time we'll only be seeing this 1 cluster at a time
2728 	 * anyway.
2729 	 *
2730 	 * Always lock for any unwritten extents - we might want to
2731 	 * add blocks during a split.
2732 	 */
2733 	if (!num_free_extents ||
2734 	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2735 		ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2736 		if (ret < 0) {
2737 			if (ret != -ENOSPC)
2738 				mlog_errno(ret);
2739 			goto out;
2740 		}
2741 	}
2742 
2743 	if (clusters_to_add == 0)
2744 		goto out;
2745 
2746 	ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2747 	if (ret < 0) {
2748 		if (ret != -ENOSPC)
2749 			mlog_errno(ret);
2750 		goto out;
2751 	}
2752 
2753 out:
2754 	if (ret) {
2755 		if (*meta_ac) {
2756 			ocfs2_free_alloc_context(*meta_ac);
2757 			*meta_ac = NULL;
2758 		}
2759 
2760 		/*
2761 		 * We cannot have an error and a non null *data_ac.
2762 		 */
2763 	}
2764 
2765 	return ret;
2766 }
2767 
2768 /*
2769  * Read the inode specified by blkno to get suballoc_slot and
2770  * suballoc_bit.
2771  */
2772 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2773 				       u16 *suballoc_slot, u64 *group_blkno,
2774 				       u16 *suballoc_bit)
2775 {
2776 	int status;
2777 	struct buffer_head *inode_bh = NULL;
2778 	struct ocfs2_dinode *inode_fe;
2779 
2780 	trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2781 
2782 	/* dirty read disk */
2783 	status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2784 	if (status < 0) {
2785 		mlog(ML_ERROR, "read block %llu failed %d\n",
2786 		     (unsigned long long)blkno, status);
2787 		goto bail;
2788 	}
2789 
2790 	inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2791 	if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2792 		mlog(ML_ERROR, "invalid inode %llu requested\n",
2793 		     (unsigned long long)blkno);
2794 		status = -EINVAL;
2795 		goto bail;
2796 	}
2797 
2798 	if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2799 	    (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2800 		mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2801 		     (unsigned long long)blkno,
2802 		     (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2803 		status = -EINVAL;
2804 		goto bail;
2805 	}
2806 
2807 	if (suballoc_slot)
2808 		*suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2809 	if (suballoc_bit)
2810 		*suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2811 	if (group_blkno)
2812 		*group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2813 
2814 bail:
2815 	brelse(inode_bh);
2816 
2817 	if (status)
2818 		mlog_errno(status);
2819 	return status;
2820 }
2821 
2822 /*
2823  * test whether bit is SET in allocator bitmap or not.  on success, 0
2824  * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
2825  * is returned and *res is meaningless.  Call this after you have
2826  * cluster locked against suballoc, or you may get a result based on
2827  * non-up2date contents
2828  */
2829 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2830 				   struct inode *suballoc,
2831 				   struct buffer_head *alloc_bh,
2832 				   u64 group_blkno, u64 blkno,
2833 				   u16 bit, int *res)
2834 {
2835 	struct ocfs2_dinode *alloc_di;
2836 	struct ocfs2_group_desc *group;
2837 	struct buffer_head *group_bh = NULL;
2838 	u64 bg_blkno;
2839 	int status;
2840 
2841 	trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2842 				      (unsigned int)bit);
2843 
2844 	alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2845 	if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2846 		mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2847 		     (unsigned int)bit,
2848 		     ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2849 		status = -EINVAL;
2850 		goto bail;
2851 	}
2852 
2853 	bg_blkno = group_blkno ? group_blkno :
2854 		   ocfs2_which_suballoc_group(blkno, bit);
2855 	status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2856 					     &group_bh);
2857 	if (status < 0) {
2858 		mlog(ML_ERROR, "read group %llu failed %d\n",
2859 		     (unsigned long long)bg_blkno, status);
2860 		goto bail;
2861 	}
2862 
2863 	group = (struct ocfs2_group_desc *) group_bh->b_data;
2864 	*res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2865 
2866 bail:
2867 	brelse(group_bh);
2868 
2869 	if (status)
2870 		mlog_errno(status);
2871 	return status;
2872 }
2873 
2874 /*
2875  * Test if the bit representing this inode (blkno) is set in the
2876  * suballocator.
2877  *
2878  * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2879  *
2880  * In the event of failure, a negative value is returned and *res is
2881  * meaningless.
2882  *
2883  * Callers must make sure to hold nfs_sync_lock to prevent
2884  * ocfs2_delete_inode() on another node from accessing the same
2885  * suballocator concurrently.
2886  */
2887 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2888 {
2889 	int status;
2890 	u64 group_blkno = 0;
2891 	u16 suballoc_bit = 0, suballoc_slot = 0;
2892 	struct inode *inode_alloc_inode;
2893 	struct buffer_head *alloc_bh = NULL;
2894 
2895 	trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2896 
2897 	status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2898 					     &group_blkno, &suballoc_bit);
2899 	if (status < 0) {
2900 		mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2901 		goto bail;
2902 	}
2903 
2904 	if (suballoc_slot == (u16)OCFS2_INVALID_SLOT)
2905 		inode_alloc_inode = ocfs2_get_system_file_inode(osb,
2906 			GLOBAL_INODE_ALLOC_SYSTEM_INODE, suballoc_slot);
2907 	else
2908 		inode_alloc_inode = ocfs2_get_system_file_inode(osb,
2909 			INODE_ALLOC_SYSTEM_INODE, suballoc_slot);
2910 	if (!inode_alloc_inode) {
2911 		/* the error code could be inaccurate, but we are not able to
2912 		 * get the correct one. */
2913 		status = -EINVAL;
2914 		mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2915 		     (u32)suballoc_slot);
2916 		goto bail;
2917 	}
2918 
2919 	inode_lock(inode_alloc_inode);
2920 	status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2921 	if (status < 0) {
2922 		inode_unlock(inode_alloc_inode);
2923 		iput(inode_alloc_inode);
2924 		mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2925 		     (u32)suballoc_slot, status);
2926 		goto bail;
2927 	}
2928 
2929 	status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2930 					 group_blkno, blkno, suballoc_bit, res);
2931 	if (status < 0)
2932 		mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2933 
2934 	ocfs2_inode_unlock(inode_alloc_inode, 0);
2935 	inode_unlock(inode_alloc_inode);
2936 
2937 	iput(inode_alloc_inode);
2938 	brelse(alloc_bh);
2939 bail:
2940 	if (status)
2941 		mlog_errno(status);
2942 	return status;
2943 }
2944