xref: /linux/fs/ocfs2/suballoc.c (revision 165f2d2858013253042809df082b8df7e34e86d7)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* -*- mode: c; c-basic-offset: 8; -*-
3  * vim: noexpandtab sw=8 ts=8 sts=0:
4  *
5  * suballoc.c
6  *
7  * metadata alloc and free
8  * Inspired by ext3 block groups.
9  *
10  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
11  */
12 
13 #include <linux/fs.h>
14 #include <linux/types.h>
15 #include <linux/slab.h>
16 #include <linux/highmem.h>
17 
18 #include <cluster/masklog.h>
19 
20 #include "ocfs2.h"
21 
22 #include "alloc.h"
23 #include "blockcheck.h"
24 #include "dlmglue.h"
25 #include "inode.h"
26 #include "journal.h"
27 #include "localalloc.h"
28 #include "suballoc.h"
29 #include "super.h"
30 #include "sysfile.h"
31 #include "uptodate.h"
32 #include "ocfs2_trace.h"
33 
34 #include "buffer_head_io.h"
35 
36 #define NOT_ALLOC_NEW_GROUP		0
37 #define ALLOC_NEW_GROUP			0x1
38 #define ALLOC_GROUPS_FROM_GLOBAL	0x2
39 
40 #define OCFS2_MAX_TO_STEAL		1024
41 
42 struct ocfs2_suballoc_result {
43 	u64		sr_bg_blkno;	/* The bg we allocated from.  Set
44 					   to 0 when a block group is
45 					   contiguous. */
46 	u64		sr_bg_stable_blkno; /*
47 					     * Doesn't change, always
48 					     * set to target block
49 					     * group descriptor
50 					     * block.
51 					     */
52 	u64		sr_blkno;	/* The first allocated block */
53 	unsigned int	sr_bit_offset;	/* The bit in the bg */
54 	unsigned int	sr_bits;	/* How many bits we claimed */
55 };
56 
57 static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
58 {
59 	if (res->sr_blkno == 0)
60 		return 0;
61 
62 	if (res->sr_bg_blkno)
63 		return res->sr_bg_blkno;
64 
65 	return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset);
66 }
67 
68 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
69 static int ocfs2_block_group_fill(handle_t *handle,
70 				  struct inode *alloc_inode,
71 				  struct buffer_head *bg_bh,
72 				  u64 group_blkno,
73 				  unsigned int group_clusters,
74 				  u16 my_chain,
75 				  struct ocfs2_chain_list *cl);
76 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
77 				   struct inode *alloc_inode,
78 				   struct buffer_head *bh,
79 				   u64 max_block,
80 				   u64 *last_alloc_group,
81 				   int flags);
82 
83 static int ocfs2_cluster_group_search(struct inode *inode,
84 				      struct buffer_head *group_bh,
85 				      u32 bits_wanted, u32 min_bits,
86 				      u64 max_block,
87 				      struct ocfs2_suballoc_result *res);
88 static int ocfs2_block_group_search(struct inode *inode,
89 				    struct buffer_head *group_bh,
90 				    u32 bits_wanted, u32 min_bits,
91 				    u64 max_block,
92 				    struct ocfs2_suballoc_result *res);
93 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
94 				     handle_t *handle,
95 				     u32 bits_wanted,
96 				     u32 min_bits,
97 				     struct ocfs2_suballoc_result *res);
98 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
99 					 int nr);
100 static int ocfs2_relink_block_group(handle_t *handle,
101 				    struct inode *alloc_inode,
102 				    struct buffer_head *fe_bh,
103 				    struct buffer_head *bg_bh,
104 				    struct buffer_head *prev_bg_bh,
105 				    u16 chain);
106 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
107 						     u32 wanted);
108 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
109 						   u64 bg_blkno,
110 						   u16 bg_bit_off);
111 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
112 						u64 data_blkno,
113 						u64 *bg_blkno,
114 						u16 *bg_bit_off);
115 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
116 					     u32 bits_wanted, u64 max_block,
117 					     int flags,
118 					     struct ocfs2_alloc_context **ac);
119 
120 void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
121 {
122 	struct inode *inode = ac->ac_inode;
123 
124 	if (inode) {
125 		if (ac->ac_which != OCFS2_AC_USE_LOCAL)
126 			ocfs2_inode_unlock(inode, 1);
127 
128 		inode_unlock(inode);
129 
130 		iput(inode);
131 		ac->ac_inode = NULL;
132 	}
133 	brelse(ac->ac_bh);
134 	ac->ac_bh = NULL;
135 	ac->ac_resv = NULL;
136 	kfree(ac->ac_find_loc_priv);
137 	ac->ac_find_loc_priv = NULL;
138 }
139 
140 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
141 {
142 	ocfs2_free_ac_resource(ac);
143 	kfree(ac);
144 }
145 
146 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
147 {
148 	return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
149 }
150 
151 #define do_error(fmt, ...)						\
152 do {									\
153 	if (resize)							\
154 		mlog(ML_ERROR, fmt, ##__VA_ARGS__);			\
155 	else								\
156 		return ocfs2_error(sb, fmt, ##__VA_ARGS__);		\
157 } while (0)
158 
159 static int ocfs2_validate_gd_self(struct super_block *sb,
160 				  struct buffer_head *bh,
161 				  int resize)
162 {
163 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
164 
165 	if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
166 		do_error("Group descriptor #%llu has bad signature %.*s\n",
167 			 (unsigned long long)bh->b_blocknr, 7,
168 			 gd->bg_signature);
169 	}
170 
171 	if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
172 		do_error("Group descriptor #%llu has an invalid bg_blkno of %llu\n",
173 			 (unsigned long long)bh->b_blocknr,
174 			 (unsigned long long)le64_to_cpu(gd->bg_blkno));
175 	}
176 
177 	if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
178 		do_error("Group descriptor #%llu has an invalid fs_generation of #%u\n",
179 			 (unsigned long long)bh->b_blocknr,
180 			 le32_to_cpu(gd->bg_generation));
181 	}
182 
183 	if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
184 		do_error("Group descriptor #%llu has bit count %u but claims that %u are free\n",
185 			 (unsigned long long)bh->b_blocknr,
186 			 le16_to_cpu(gd->bg_bits),
187 			 le16_to_cpu(gd->bg_free_bits_count));
188 	}
189 
190 	if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
191 		do_error("Group descriptor #%llu has bit count %u but max bitmap bits of %u\n",
192 			 (unsigned long long)bh->b_blocknr,
193 			 le16_to_cpu(gd->bg_bits),
194 			 8 * le16_to_cpu(gd->bg_size));
195 	}
196 
197 	return 0;
198 }
199 
200 static int ocfs2_validate_gd_parent(struct super_block *sb,
201 				    struct ocfs2_dinode *di,
202 				    struct buffer_head *bh,
203 				    int resize)
204 {
205 	unsigned int max_bits;
206 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
207 
208 	if (di->i_blkno != gd->bg_parent_dinode) {
209 		do_error("Group descriptor #%llu has bad parent pointer (%llu, expected %llu)\n",
210 			 (unsigned long long)bh->b_blocknr,
211 			 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
212 			 (unsigned long long)le64_to_cpu(di->i_blkno));
213 	}
214 
215 	max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
216 	if (le16_to_cpu(gd->bg_bits) > max_bits) {
217 		do_error("Group descriptor #%llu has bit count of %u\n",
218 			 (unsigned long long)bh->b_blocknr,
219 			 le16_to_cpu(gd->bg_bits));
220 	}
221 
222 	/* In resize, we may meet the case bg_chain == cl_next_free_rec. */
223 	if ((le16_to_cpu(gd->bg_chain) >
224 	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
225 	    ((le16_to_cpu(gd->bg_chain) ==
226 	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
227 		do_error("Group descriptor #%llu has bad chain %u\n",
228 			 (unsigned long long)bh->b_blocknr,
229 			 le16_to_cpu(gd->bg_chain));
230 	}
231 
232 	return 0;
233 }
234 
235 #undef do_error
236 
237 /*
238  * This version only prints errors.  It does not fail the filesystem, and
239  * exists only for resize.
240  */
241 int ocfs2_check_group_descriptor(struct super_block *sb,
242 				 struct ocfs2_dinode *di,
243 				 struct buffer_head *bh)
244 {
245 	int rc;
246 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
247 
248 	BUG_ON(!buffer_uptodate(bh));
249 
250 	/*
251 	 * If the ecc fails, we return the error but otherwise
252 	 * leave the filesystem running.  We know any error is
253 	 * local to this block.
254 	 */
255 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
256 	if (rc) {
257 		mlog(ML_ERROR,
258 		     "Checksum failed for group descriptor %llu\n",
259 		     (unsigned long long)bh->b_blocknr);
260 	} else
261 		rc = ocfs2_validate_gd_self(sb, bh, 1);
262 	if (!rc)
263 		rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
264 
265 	return rc;
266 }
267 
268 static int ocfs2_validate_group_descriptor(struct super_block *sb,
269 					   struct buffer_head *bh)
270 {
271 	int rc;
272 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
273 
274 	trace_ocfs2_validate_group_descriptor(
275 					(unsigned long long)bh->b_blocknr);
276 
277 	BUG_ON(!buffer_uptodate(bh));
278 
279 	/*
280 	 * If the ecc fails, we return the error but otherwise
281 	 * leave the filesystem running.  We know any error is
282 	 * local to this block.
283 	 */
284 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
285 	if (rc)
286 		return rc;
287 
288 	/*
289 	 * Errors after here are fatal.
290 	 */
291 
292 	return ocfs2_validate_gd_self(sb, bh, 0);
293 }
294 
295 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
296 				u64 gd_blkno, struct buffer_head **bh)
297 {
298 	int rc;
299 	struct buffer_head *tmp = *bh;
300 
301 	rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
302 			      ocfs2_validate_group_descriptor);
303 	if (rc)
304 		goto out;
305 
306 	rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
307 	if (rc) {
308 		brelse(tmp);
309 		goto out;
310 	}
311 
312 	/* If ocfs2_read_block() got us a new bh, pass it up. */
313 	if (!*bh)
314 		*bh = tmp;
315 
316 out:
317 	return rc;
318 }
319 
320 static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
321 					  struct ocfs2_group_desc *bg,
322 					  struct ocfs2_chain_list *cl,
323 					  u64 p_blkno, unsigned int clusters)
324 {
325 	struct ocfs2_extent_list *el = &bg->bg_list;
326 	struct ocfs2_extent_rec *rec;
327 
328 	BUG_ON(!ocfs2_supports_discontig_bg(osb));
329 	if (!el->l_next_free_rec)
330 		el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
331 	rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
332 	rec->e_blkno = cpu_to_le64(p_blkno);
333 	rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
334 				  le16_to_cpu(cl->cl_bpc));
335 	rec->e_leaf_clusters = cpu_to_le16(clusters);
336 	le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
337 	le16_add_cpu(&bg->bg_free_bits_count,
338 		     clusters * le16_to_cpu(cl->cl_bpc));
339 	le16_add_cpu(&el->l_next_free_rec, 1);
340 }
341 
342 static int ocfs2_block_group_fill(handle_t *handle,
343 				  struct inode *alloc_inode,
344 				  struct buffer_head *bg_bh,
345 				  u64 group_blkno,
346 				  unsigned int group_clusters,
347 				  u16 my_chain,
348 				  struct ocfs2_chain_list *cl)
349 {
350 	int status = 0;
351 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
352 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
353 	struct super_block * sb = alloc_inode->i_sb;
354 
355 	if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
356 		status = ocfs2_error(alloc_inode->i_sb,
357 				     "group block (%llu) != b_blocknr (%llu)\n",
358 				     (unsigned long long)group_blkno,
359 				     (unsigned long long) bg_bh->b_blocknr);
360 		goto bail;
361 	}
362 
363 	status = ocfs2_journal_access_gd(handle,
364 					 INODE_CACHE(alloc_inode),
365 					 bg_bh,
366 					 OCFS2_JOURNAL_ACCESS_CREATE);
367 	if (status < 0) {
368 		mlog_errno(status);
369 		goto bail;
370 	}
371 
372 	memset(bg, 0, sb->s_blocksize);
373 	strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
374 	bg->bg_generation = cpu_to_le32(osb->fs_generation);
375 	bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
376 						osb->s_feature_incompat));
377 	bg->bg_chain = cpu_to_le16(my_chain);
378 	bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
379 	bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
380 	bg->bg_blkno = cpu_to_le64(group_blkno);
381 	if (group_clusters == le16_to_cpu(cl->cl_cpg))
382 		bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
383 	else
384 		ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
385 					      group_clusters);
386 
387 	/* set the 1st bit in the bitmap to account for the descriptor block */
388 	ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
389 	bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
390 
391 	ocfs2_journal_dirty(handle, bg_bh);
392 
393 	/* There is no need to zero out or otherwise initialize the
394 	 * other blocks in a group - All valid FS metadata in a block
395 	 * group stores the superblock fs_generation value at
396 	 * allocation time. */
397 
398 bail:
399 	if (status)
400 		mlog_errno(status);
401 	return status;
402 }
403 
404 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
405 {
406 	u16 curr, best;
407 
408 	best = curr = 0;
409 	while (curr < le16_to_cpu(cl->cl_count)) {
410 		if (le32_to_cpu(cl->cl_recs[best].c_total) >
411 		    le32_to_cpu(cl->cl_recs[curr].c_total))
412 			best = curr;
413 		curr++;
414 	}
415 	return best;
416 }
417 
418 static struct buffer_head *
419 ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
420 			       struct inode *alloc_inode,
421 			       struct ocfs2_alloc_context *ac,
422 			       struct ocfs2_chain_list *cl)
423 {
424 	int status;
425 	u32 bit_off, num_bits;
426 	u64 bg_blkno;
427 	struct buffer_head *bg_bh;
428 	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
429 
430 	status = ocfs2_claim_clusters(handle, ac,
431 				      le16_to_cpu(cl->cl_cpg), &bit_off,
432 				      &num_bits);
433 	if (status < 0) {
434 		if (status != -ENOSPC)
435 			mlog_errno(status);
436 		goto bail;
437 	}
438 
439 	/* setup the group */
440 	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
441 	trace_ocfs2_block_group_alloc_contig(
442 	     (unsigned long long)bg_blkno, alloc_rec);
443 
444 	bg_bh = sb_getblk(osb->sb, bg_blkno);
445 	if (!bg_bh) {
446 		status = -ENOMEM;
447 		mlog_errno(status);
448 		goto bail;
449 	}
450 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
451 
452 	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
453 					bg_blkno, num_bits, alloc_rec, cl);
454 	if (status < 0) {
455 		brelse(bg_bh);
456 		mlog_errno(status);
457 	}
458 
459 bail:
460 	return status ? ERR_PTR(status) : bg_bh;
461 }
462 
463 static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
464 					handle_t *handle,
465 					struct ocfs2_alloc_context *ac,
466 					unsigned int min_bits,
467 					u32 *bit_off, u32 *num_bits)
468 {
469 	int status = 0;
470 
471 	while (min_bits) {
472 		status = ocfs2_claim_clusters(handle, ac, min_bits,
473 					      bit_off, num_bits);
474 		if (status != -ENOSPC)
475 			break;
476 
477 		min_bits >>= 1;
478 	}
479 
480 	return status;
481 }
482 
483 static int ocfs2_block_group_grow_discontig(handle_t *handle,
484 					    struct inode *alloc_inode,
485 					    struct buffer_head *bg_bh,
486 					    struct ocfs2_alloc_context *ac,
487 					    struct ocfs2_chain_list *cl,
488 					    unsigned int min_bits)
489 {
490 	int status;
491 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
492 	struct ocfs2_group_desc *bg =
493 		(struct ocfs2_group_desc *)bg_bh->b_data;
494 	unsigned int needed = le16_to_cpu(cl->cl_cpg) -
495 			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
496 	u32 p_cpos, clusters;
497 	u64 p_blkno;
498 	struct ocfs2_extent_list *el = &bg->bg_list;
499 
500 	status = ocfs2_journal_access_gd(handle,
501 					 INODE_CACHE(alloc_inode),
502 					 bg_bh,
503 					 OCFS2_JOURNAL_ACCESS_CREATE);
504 	if (status < 0) {
505 		mlog_errno(status);
506 		goto bail;
507 	}
508 
509 	while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
510 				le16_to_cpu(el->l_count))) {
511 		if (min_bits > needed)
512 			min_bits = needed;
513 		status = ocfs2_block_group_claim_bits(osb, handle, ac,
514 						      min_bits, &p_cpos,
515 						      &clusters);
516 		if (status < 0) {
517 			if (status != -ENOSPC)
518 				mlog_errno(status);
519 			goto bail;
520 		}
521 		p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
522 		ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
523 					      clusters);
524 
525 		min_bits = clusters;
526 		needed = le16_to_cpu(cl->cl_cpg) -
527 			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
528 	}
529 
530 	if (needed > 0) {
531 		/*
532 		 * We have used up all the extent rec but can't fill up
533 		 * the cpg. So bail out.
534 		 */
535 		status = -ENOSPC;
536 		goto bail;
537 	}
538 
539 	ocfs2_journal_dirty(handle, bg_bh);
540 
541 bail:
542 	return status;
543 }
544 
545 static void ocfs2_bg_alloc_cleanup(handle_t *handle,
546 				   struct ocfs2_alloc_context *cluster_ac,
547 				   struct inode *alloc_inode,
548 				   struct buffer_head *bg_bh)
549 {
550 	int i, ret;
551 	struct ocfs2_group_desc *bg;
552 	struct ocfs2_extent_list *el;
553 	struct ocfs2_extent_rec *rec;
554 
555 	if (!bg_bh)
556 		return;
557 
558 	bg = (struct ocfs2_group_desc *)bg_bh->b_data;
559 	el = &bg->bg_list;
560 	for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
561 		rec = &el->l_recs[i];
562 		ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
563 					  cluster_ac->ac_bh,
564 					  le64_to_cpu(rec->e_blkno),
565 					  le16_to_cpu(rec->e_leaf_clusters));
566 		if (ret)
567 			mlog_errno(ret);
568 		/* Try all the clusters to free */
569 	}
570 
571 	ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
572 	brelse(bg_bh);
573 }
574 
575 static struct buffer_head *
576 ocfs2_block_group_alloc_discontig(handle_t *handle,
577 				  struct inode *alloc_inode,
578 				  struct ocfs2_alloc_context *ac,
579 				  struct ocfs2_chain_list *cl)
580 {
581 	int status;
582 	u32 bit_off, num_bits;
583 	u64 bg_blkno;
584 	unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
585 	struct buffer_head *bg_bh = NULL;
586 	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
587 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
588 
589 	if (!ocfs2_supports_discontig_bg(osb)) {
590 		status = -ENOSPC;
591 		goto bail;
592 	}
593 
594 	status = ocfs2_extend_trans(handle,
595 				    ocfs2_calc_bg_discontig_credits(osb->sb));
596 	if (status) {
597 		mlog_errno(status);
598 		goto bail;
599 	}
600 
601 	/*
602 	 * We're going to be grabbing from multiple cluster groups.
603 	 * We don't have enough credits to relink them all, and the
604 	 * cluster groups will be staying in cache for the duration of
605 	 * this operation.
606 	 */
607 	ac->ac_disable_chain_relink = 1;
608 
609 	/* Claim the first region */
610 	status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
611 					      &bit_off, &num_bits);
612 	if (status < 0) {
613 		if (status != -ENOSPC)
614 			mlog_errno(status);
615 		goto bail;
616 	}
617 	min_bits = num_bits;
618 
619 	/* setup the group */
620 	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
621 	trace_ocfs2_block_group_alloc_discontig(
622 				(unsigned long long)bg_blkno, alloc_rec);
623 
624 	bg_bh = sb_getblk(osb->sb, bg_blkno);
625 	if (!bg_bh) {
626 		status = -ENOMEM;
627 		mlog_errno(status);
628 		goto bail;
629 	}
630 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
631 
632 	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
633 					bg_blkno, num_bits, alloc_rec, cl);
634 	if (status < 0) {
635 		mlog_errno(status);
636 		goto bail;
637 	}
638 
639 	status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
640 						  bg_bh, ac, cl, min_bits);
641 	if (status)
642 		mlog_errno(status);
643 
644 bail:
645 	if (status)
646 		ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
647 	return status ? ERR_PTR(status) : bg_bh;
648 }
649 
650 /*
651  * We expect the block group allocator to already be locked.
652  */
653 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
654 				   struct inode *alloc_inode,
655 				   struct buffer_head *bh,
656 				   u64 max_block,
657 				   u64 *last_alloc_group,
658 				   int flags)
659 {
660 	int status, credits;
661 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
662 	struct ocfs2_chain_list *cl;
663 	struct ocfs2_alloc_context *ac = NULL;
664 	handle_t *handle = NULL;
665 	u16 alloc_rec;
666 	struct buffer_head *bg_bh = NULL;
667 	struct ocfs2_group_desc *bg;
668 
669 	BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
670 
671 	cl = &fe->id2.i_chain;
672 	status = ocfs2_reserve_clusters_with_limit(osb,
673 						   le16_to_cpu(cl->cl_cpg),
674 						   max_block, flags, &ac);
675 	if (status < 0) {
676 		if (status != -ENOSPC)
677 			mlog_errno(status);
678 		goto bail;
679 	}
680 
681 	credits = ocfs2_calc_group_alloc_credits(osb->sb,
682 						 le16_to_cpu(cl->cl_cpg));
683 	handle = ocfs2_start_trans(osb, credits);
684 	if (IS_ERR(handle)) {
685 		status = PTR_ERR(handle);
686 		handle = NULL;
687 		mlog_errno(status);
688 		goto bail;
689 	}
690 
691 	if (last_alloc_group && *last_alloc_group != 0) {
692 		trace_ocfs2_block_group_alloc(
693 				(unsigned long long)*last_alloc_group);
694 		ac->ac_last_group = *last_alloc_group;
695 	}
696 
697 	bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
698 					       ac, cl);
699 	if (PTR_ERR(bg_bh) == -ENOSPC)
700 		bg_bh = ocfs2_block_group_alloc_discontig(handle,
701 							  alloc_inode,
702 							  ac, cl);
703 	if (IS_ERR(bg_bh)) {
704 		status = PTR_ERR(bg_bh);
705 		bg_bh = NULL;
706 		if (status != -ENOSPC)
707 			mlog_errno(status);
708 		goto bail;
709 	}
710 	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
711 
712 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
713 					 bh, OCFS2_JOURNAL_ACCESS_WRITE);
714 	if (status < 0) {
715 		mlog_errno(status);
716 		goto bail;
717 	}
718 
719 	alloc_rec = le16_to_cpu(bg->bg_chain);
720 	le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
721 		     le16_to_cpu(bg->bg_free_bits_count));
722 	le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
723 		     le16_to_cpu(bg->bg_bits));
724 	cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
725 	if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
726 		le16_add_cpu(&cl->cl_next_free_rec, 1);
727 
728 	le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
729 					le16_to_cpu(bg->bg_free_bits_count));
730 	le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
731 	le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
732 
733 	ocfs2_journal_dirty(handle, bh);
734 
735 	spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
736 	OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
737 	fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
738 					     le32_to_cpu(fe->i_clusters)));
739 	spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
740 	i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
741 	alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
742 	ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
743 
744 	status = 0;
745 
746 	/* save the new last alloc group so that the caller can cache it. */
747 	if (last_alloc_group)
748 		*last_alloc_group = ac->ac_last_group;
749 
750 bail:
751 	if (handle)
752 		ocfs2_commit_trans(osb, handle);
753 
754 	if (ac)
755 		ocfs2_free_alloc_context(ac);
756 
757 	brelse(bg_bh);
758 
759 	if (status)
760 		mlog_errno(status);
761 	return status;
762 }
763 
764 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
765 				       struct ocfs2_alloc_context *ac,
766 				       int type,
767 				       u32 slot,
768 				       u64 *last_alloc_group,
769 				       int flags)
770 {
771 	int status;
772 	u32 bits_wanted = ac->ac_bits_wanted;
773 	struct inode *alloc_inode;
774 	struct buffer_head *bh = NULL;
775 	struct ocfs2_dinode *fe;
776 	u32 free_bits;
777 
778 	alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
779 	if (!alloc_inode) {
780 		mlog_errno(-EINVAL);
781 		return -EINVAL;
782 	}
783 
784 	inode_lock(alloc_inode);
785 
786 	status = ocfs2_inode_lock(alloc_inode, &bh, 1);
787 	if (status < 0) {
788 		inode_unlock(alloc_inode);
789 		iput(alloc_inode);
790 
791 		mlog_errno(status);
792 		return status;
793 	}
794 
795 	ac->ac_inode = alloc_inode;
796 	ac->ac_alloc_slot = slot;
797 
798 	fe = (struct ocfs2_dinode *) bh->b_data;
799 
800 	/* The bh was validated by the inode read inside
801 	 * ocfs2_inode_lock().  Any corruption is a code bug. */
802 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
803 
804 	if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
805 		status = ocfs2_error(alloc_inode->i_sb,
806 				     "Invalid chain allocator %llu\n",
807 				     (unsigned long long)le64_to_cpu(fe->i_blkno));
808 		goto bail;
809 	}
810 
811 	free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
812 		le32_to_cpu(fe->id1.bitmap1.i_used);
813 
814 	if (bits_wanted > free_bits) {
815 		/* cluster bitmap never grows */
816 		if (ocfs2_is_cluster_bitmap(alloc_inode)) {
817 			trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
818 								free_bits);
819 			status = -ENOSPC;
820 			goto bail;
821 		}
822 
823 		if (!(flags & ALLOC_NEW_GROUP)) {
824 			trace_ocfs2_reserve_suballoc_bits_no_new_group(
825 						slot, bits_wanted, free_bits);
826 			status = -ENOSPC;
827 			goto bail;
828 		}
829 
830 		status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
831 						 ac->ac_max_block,
832 						 last_alloc_group, flags);
833 		if (status < 0) {
834 			if (status != -ENOSPC)
835 				mlog_errno(status);
836 			goto bail;
837 		}
838 		atomic_inc(&osb->alloc_stats.bg_extends);
839 
840 		/* You should never ask for this much metadata */
841 		BUG_ON(bits_wanted >
842 		       (le32_to_cpu(fe->id1.bitmap1.i_total)
843 			- le32_to_cpu(fe->id1.bitmap1.i_used)));
844 	}
845 
846 	get_bh(bh);
847 	ac->ac_bh = bh;
848 bail:
849 	brelse(bh);
850 
851 	if (status)
852 		mlog_errno(status);
853 	return status;
854 }
855 
856 static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
857 {
858 	spin_lock(&osb->osb_lock);
859 	osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
860 	spin_unlock(&osb->osb_lock);
861 	atomic_set(&osb->s_num_inodes_stolen, 0);
862 }
863 
864 static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
865 {
866 	spin_lock(&osb->osb_lock);
867 	osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
868 	spin_unlock(&osb->osb_lock);
869 	atomic_set(&osb->s_num_meta_stolen, 0);
870 }
871 
872 void ocfs2_init_steal_slots(struct ocfs2_super *osb)
873 {
874 	ocfs2_init_inode_steal_slot(osb);
875 	ocfs2_init_meta_steal_slot(osb);
876 }
877 
878 static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
879 {
880 	spin_lock(&osb->osb_lock);
881 	if (type == INODE_ALLOC_SYSTEM_INODE)
882 		osb->s_inode_steal_slot = slot;
883 	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
884 		osb->s_meta_steal_slot = slot;
885 	spin_unlock(&osb->osb_lock);
886 }
887 
888 static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
889 {
890 	int slot = OCFS2_INVALID_SLOT;
891 
892 	spin_lock(&osb->osb_lock);
893 	if (type == INODE_ALLOC_SYSTEM_INODE)
894 		slot = osb->s_inode_steal_slot;
895 	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
896 		slot = osb->s_meta_steal_slot;
897 	spin_unlock(&osb->osb_lock);
898 
899 	return slot;
900 }
901 
902 static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
903 {
904 	return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
905 }
906 
907 static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
908 {
909 	return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
910 }
911 
912 static int ocfs2_steal_resource(struct ocfs2_super *osb,
913 				struct ocfs2_alloc_context *ac,
914 				int type)
915 {
916 	int i, status = -ENOSPC;
917 	int slot = __ocfs2_get_steal_slot(osb, type);
918 
919 	/* Start to steal resource from the first slot after ours. */
920 	if (slot == OCFS2_INVALID_SLOT)
921 		slot = osb->slot_num + 1;
922 
923 	for (i = 0; i < osb->max_slots; i++, slot++) {
924 		if (slot == osb->max_slots)
925 			slot = 0;
926 
927 		if (slot == osb->slot_num)
928 			continue;
929 
930 		status = ocfs2_reserve_suballoc_bits(osb, ac,
931 						     type,
932 						     (u32)slot, NULL,
933 						     NOT_ALLOC_NEW_GROUP);
934 		if (status >= 0) {
935 			__ocfs2_set_steal_slot(osb, slot, type);
936 			break;
937 		}
938 
939 		ocfs2_free_ac_resource(ac);
940 	}
941 
942 	return status;
943 }
944 
945 static int ocfs2_steal_inode(struct ocfs2_super *osb,
946 			     struct ocfs2_alloc_context *ac)
947 {
948 	return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
949 }
950 
951 static int ocfs2_steal_meta(struct ocfs2_super *osb,
952 			    struct ocfs2_alloc_context *ac)
953 {
954 	return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
955 }
956 
957 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
958 				      int blocks,
959 				      struct ocfs2_alloc_context **ac)
960 {
961 	int status;
962 	int slot = ocfs2_get_meta_steal_slot(osb);
963 
964 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
965 	if (!(*ac)) {
966 		status = -ENOMEM;
967 		mlog_errno(status);
968 		goto bail;
969 	}
970 
971 	(*ac)->ac_bits_wanted = blocks;
972 	(*ac)->ac_which = OCFS2_AC_USE_META;
973 	(*ac)->ac_group_search = ocfs2_block_group_search;
974 
975 	if (slot != OCFS2_INVALID_SLOT &&
976 		atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
977 		goto extent_steal;
978 
979 	atomic_set(&osb->s_num_meta_stolen, 0);
980 	status = ocfs2_reserve_suballoc_bits(osb, (*ac),
981 					     EXTENT_ALLOC_SYSTEM_INODE,
982 					     (u32)osb->slot_num, NULL,
983 					     ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
984 
985 
986 	if (status >= 0) {
987 		status = 0;
988 		if (slot != OCFS2_INVALID_SLOT)
989 			ocfs2_init_meta_steal_slot(osb);
990 		goto bail;
991 	} else if (status < 0 && status != -ENOSPC) {
992 		mlog_errno(status);
993 		goto bail;
994 	}
995 
996 	ocfs2_free_ac_resource(*ac);
997 
998 extent_steal:
999 	status = ocfs2_steal_meta(osb, *ac);
1000 	atomic_inc(&osb->s_num_meta_stolen);
1001 	if (status < 0) {
1002 		if (status != -ENOSPC)
1003 			mlog_errno(status);
1004 		goto bail;
1005 	}
1006 
1007 	status = 0;
1008 bail:
1009 	if ((status < 0) && *ac) {
1010 		ocfs2_free_alloc_context(*ac);
1011 		*ac = NULL;
1012 	}
1013 
1014 	if (status)
1015 		mlog_errno(status);
1016 	return status;
1017 }
1018 
1019 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1020 			       struct ocfs2_extent_list *root_el,
1021 			       struct ocfs2_alloc_context **ac)
1022 {
1023 	return ocfs2_reserve_new_metadata_blocks(osb,
1024 					ocfs2_extend_meta_needed(root_el),
1025 					ac);
1026 }
1027 
1028 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
1029 			    struct ocfs2_alloc_context **ac)
1030 {
1031 	int status;
1032 	int slot = ocfs2_get_inode_steal_slot(osb);
1033 	u64 alloc_group;
1034 
1035 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1036 	if (!(*ac)) {
1037 		status = -ENOMEM;
1038 		mlog_errno(status);
1039 		goto bail;
1040 	}
1041 
1042 	(*ac)->ac_bits_wanted = 1;
1043 	(*ac)->ac_which = OCFS2_AC_USE_INODE;
1044 
1045 	(*ac)->ac_group_search = ocfs2_block_group_search;
1046 
1047 	/*
1048 	 * stat(2) can't handle i_ino > 32bits, so we tell the
1049 	 * lower levels not to allocate us a block group past that
1050 	 * limit.  The 'inode64' mount option avoids this behavior.
1051 	 */
1052 	if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1053 		(*ac)->ac_max_block = (u32)~0U;
1054 
1055 	/*
1056 	 * slot is set when we successfully steal inode from other nodes.
1057 	 * It is reset in 3 places:
1058 	 * 1. when we flush the truncate log
1059 	 * 2. when we complete local alloc recovery.
1060 	 * 3. when we successfully allocate from our own slot.
1061 	 * After it is set, we will go on stealing inodes until we find the
1062 	 * need to check our slots to see whether there is some space for us.
1063 	 */
1064 	if (slot != OCFS2_INVALID_SLOT &&
1065 	    atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
1066 		goto inode_steal;
1067 
1068 	atomic_set(&osb->s_num_inodes_stolen, 0);
1069 	alloc_group = osb->osb_inode_alloc_group;
1070 	status = ocfs2_reserve_suballoc_bits(osb, *ac,
1071 					     INODE_ALLOC_SYSTEM_INODE,
1072 					     (u32)osb->slot_num,
1073 					     &alloc_group,
1074 					     ALLOC_NEW_GROUP |
1075 					     ALLOC_GROUPS_FROM_GLOBAL);
1076 	if (status >= 0) {
1077 		status = 0;
1078 
1079 		spin_lock(&osb->osb_lock);
1080 		osb->osb_inode_alloc_group = alloc_group;
1081 		spin_unlock(&osb->osb_lock);
1082 		trace_ocfs2_reserve_new_inode_new_group(
1083 			(unsigned long long)alloc_group);
1084 
1085 		/*
1086 		 * Some inodes must be freed by us, so try to allocate
1087 		 * from our own next time.
1088 		 */
1089 		if (slot != OCFS2_INVALID_SLOT)
1090 			ocfs2_init_inode_steal_slot(osb);
1091 		goto bail;
1092 	} else if (status < 0 && status != -ENOSPC) {
1093 		mlog_errno(status);
1094 		goto bail;
1095 	}
1096 
1097 	ocfs2_free_ac_resource(*ac);
1098 
1099 inode_steal:
1100 	status = ocfs2_steal_inode(osb, *ac);
1101 	atomic_inc(&osb->s_num_inodes_stolen);
1102 	if (status < 0) {
1103 		if (status != -ENOSPC)
1104 			mlog_errno(status);
1105 		goto bail;
1106 	}
1107 
1108 	status = 0;
1109 bail:
1110 	if ((status < 0) && *ac) {
1111 		ocfs2_free_alloc_context(*ac);
1112 		*ac = NULL;
1113 	}
1114 
1115 	if (status)
1116 		mlog_errno(status);
1117 	return status;
1118 }
1119 
1120 /* local alloc code has to do the same thing, so rather than do this
1121  * twice.. */
1122 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1123 				      struct ocfs2_alloc_context *ac)
1124 {
1125 	int status;
1126 
1127 	ac->ac_which = OCFS2_AC_USE_MAIN;
1128 	ac->ac_group_search = ocfs2_cluster_group_search;
1129 
1130 	status = ocfs2_reserve_suballoc_bits(osb, ac,
1131 					     GLOBAL_BITMAP_SYSTEM_INODE,
1132 					     OCFS2_INVALID_SLOT, NULL,
1133 					     ALLOC_NEW_GROUP);
1134 	if (status < 0 && status != -ENOSPC)
1135 		mlog_errno(status);
1136 
1137 	return status;
1138 }
1139 
1140 /* Callers don't need to care which bitmap (local alloc or main) to
1141  * use so we figure it out for them, but unfortunately this clutters
1142  * things a bit. */
1143 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1144 					     u32 bits_wanted, u64 max_block,
1145 					     int flags,
1146 					     struct ocfs2_alloc_context **ac)
1147 {
1148 	int status, ret = 0;
1149 	int retried = 0;
1150 
1151 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1152 	if (!(*ac)) {
1153 		status = -ENOMEM;
1154 		mlog_errno(status);
1155 		goto bail;
1156 	}
1157 
1158 	(*ac)->ac_bits_wanted = bits_wanted;
1159 	(*ac)->ac_max_block = max_block;
1160 
1161 	status = -ENOSPC;
1162 	if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1163 	    ocfs2_alloc_should_use_local(osb, bits_wanted)) {
1164 		status = ocfs2_reserve_local_alloc_bits(osb,
1165 							bits_wanted,
1166 							*ac);
1167 		if ((status < 0) && (status != -ENOSPC)) {
1168 			mlog_errno(status);
1169 			goto bail;
1170 		}
1171 	}
1172 
1173 	if (status == -ENOSPC) {
1174 retry:
1175 		status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1176 		/* Retry if there is sufficient space cached in truncate log */
1177 		if (status == -ENOSPC && !retried) {
1178 			retried = 1;
1179 			ocfs2_inode_unlock((*ac)->ac_inode, 1);
1180 			inode_unlock((*ac)->ac_inode);
1181 
1182 			ret = ocfs2_try_to_free_truncate_log(osb, bits_wanted);
1183 			if (ret == 1) {
1184 				iput((*ac)->ac_inode);
1185 				(*ac)->ac_inode = NULL;
1186 				goto retry;
1187 			}
1188 
1189 			if (ret < 0)
1190 				mlog_errno(ret);
1191 
1192 			inode_lock((*ac)->ac_inode);
1193 			ret = ocfs2_inode_lock((*ac)->ac_inode, NULL, 1);
1194 			if (ret < 0) {
1195 				mlog_errno(ret);
1196 				inode_unlock((*ac)->ac_inode);
1197 				iput((*ac)->ac_inode);
1198 				(*ac)->ac_inode = NULL;
1199 				goto bail;
1200 			}
1201 		}
1202 		if (status < 0) {
1203 			if (status != -ENOSPC)
1204 				mlog_errno(status);
1205 			goto bail;
1206 		}
1207 	}
1208 
1209 	status = 0;
1210 bail:
1211 	if ((status < 0) && *ac) {
1212 		ocfs2_free_alloc_context(*ac);
1213 		*ac = NULL;
1214 	}
1215 
1216 	if (status)
1217 		mlog_errno(status);
1218 	return status;
1219 }
1220 
1221 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1222 			   u32 bits_wanted,
1223 			   struct ocfs2_alloc_context **ac)
1224 {
1225 	return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1226 						 ALLOC_NEW_GROUP, ac);
1227 }
1228 
1229 /*
1230  * More or less lifted from ext3. I'll leave their description below:
1231  *
1232  * "For ext3 allocations, we must not reuse any blocks which are
1233  * allocated in the bitmap buffer's "last committed data" copy.  This
1234  * prevents deletes from freeing up the page for reuse until we have
1235  * committed the delete transaction.
1236  *
1237  * If we didn't do this, then deleting something and reallocating it as
1238  * data would allow the old block to be overwritten before the
1239  * transaction committed (because we force data to disk before commit).
1240  * This would lead to corruption if we crashed between overwriting the
1241  * data and committing the delete.
1242  *
1243  * @@@ We may want to make this allocation behaviour conditional on
1244  * data-writes at some point, and disable it for metadata allocations or
1245  * sync-data inodes."
1246  *
1247  * Note: OCFS2 already does this differently for metadata vs data
1248  * allocations, as those bitmaps are separate and undo access is never
1249  * called on a metadata group descriptor.
1250  */
1251 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1252 					 int nr)
1253 {
1254 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1255 	struct journal_head *jh;
1256 	int ret;
1257 
1258 	if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1259 		return 0;
1260 
1261 	if (!buffer_jbd(bg_bh))
1262 		return 1;
1263 
1264 	jh = bh2jh(bg_bh);
1265 	spin_lock(&jh->b_state_lock);
1266 	bg = (struct ocfs2_group_desc *) jh->b_committed_data;
1267 	if (bg)
1268 		ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1269 	else
1270 		ret = 1;
1271 	spin_unlock(&jh->b_state_lock);
1272 
1273 	return ret;
1274 }
1275 
1276 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1277 					     struct buffer_head *bg_bh,
1278 					     unsigned int bits_wanted,
1279 					     unsigned int total_bits,
1280 					     struct ocfs2_suballoc_result *res)
1281 {
1282 	void *bitmap;
1283 	u16 best_offset, best_size;
1284 	int offset, start, found, status = 0;
1285 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1286 
1287 	/* Callers got this descriptor from
1288 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1289 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1290 
1291 	found = start = best_offset = best_size = 0;
1292 	bitmap = bg->bg_bitmap;
1293 
1294 	while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1295 		if (offset == total_bits)
1296 			break;
1297 
1298 		if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1299 			/* We found a zero, but we can't use it as it
1300 			 * hasn't been put to disk yet! */
1301 			found = 0;
1302 			start = offset + 1;
1303 		} else if (offset == start) {
1304 			/* we found a zero */
1305 			found++;
1306 			/* move start to the next bit to test */
1307 			start++;
1308 		} else {
1309 			/* got a zero after some ones */
1310 			found = 1;
1311 			start = offset + 1;
1312 		}
1313 		if (found > best_size) {
1314 			best_size = found;
1315 			best_offset = start - found;
1316 		}
1317 		/* we got everything we needed */
1318 		if (found == bits_wanted) {
1319 			/* mlog(0, "Found it all!\n"); */
1320 			break;
1321 		}
1322 	}
1323 
1324 	if (best_size) {
1325 		res->sr_bit_offset = best_offset;
1326 		res->sr_bits = best_size;
1327 	} else {
1328 		status = -ENOSPC;
1329 		/* No error log here -- see the comment above
1330 		 * ocfs2_test_bg_bit_allocatable */
1331 	}
1332 
1333 	return status;
1334 }
1335 
1336 int ocfs2_block_group_set_bits(handle_t *handle,
1337 					     struct inode *alloc_inode,
1338 					     struct ocfs2_group_desc *bg,
1339 					     struct buffer_head *group_bh,
1340 					     unsigned int bit_off,
1341 					     unsigned int num_bits)
1342 {
1343 	int status;
1344 	void *bitmap = bg->bg_bitmap;
1345 	int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1346 
1347 	/* All callers get the descriptor via
1348 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1349 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1350 	BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1351 
1352 	trace_ocfs2_block_group_set_bits(bit_off, num_bits);
1353 
1354 	if (ocfs2_is_cluster_bitmap(alloc_inode))
1355 		journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1356 
1357 	status = ocfs2_journal_access_gd(handle,
1358 					 INODE_CACHE(alloc_inode),
1359 					 group_bh,
1360 					 journal_type);
1361 	if (status < 0) {
1362 		mlog_errno(status);
1363 		goto bail;
1364 	}
1365 
1366 	le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1367 	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1368 		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
1369 				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
1370 				   le16_to_cpu(bg->bg_bits),
1371 				   le16_to_cpu(bg->bg_free_bits_count),
1372 				   num_bits);
1373 	}
1374 	while(num_bits--)
1375 		ocfs2_set_bit(bit_off++, bitmap);
1376 
1377 	ocfs2_journal_dirty(handle, group_bh);
1378 
1379 bail:
1380 	return status;
1381 }
1382 
1383 /* find the one with the most empty bits */
1384 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1385 {
1386 	u16 curr, best;
1387 
1388 	BUG_ON(!cl->cl_next_free_rec);
1389 
1390 	best = curr = 0;
1391 	while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1392 		if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1393 		    le32_to_cpu(cl->cl_recs[best].c_free))
1394 			best = curr;
1395 		curr++;
1396 	}
1397 
1398 	BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1399 	return best;
1400 }
1401 
1402 static int ocfs2_relink_block_group(handle_t *handle,
1403 				    struct inode *alloc_inode,
1404 				    struct buffer_head *fe_bh,
1405 				    struct buffer_head *bg_bh,
1406 				    struct buffer_head *prev_bg_bh,
1407 				    u16 chain)
1408 {
1409 	int status;
1410 	/* there is a really tiny chance the journal calls could fail,
1411 	 * but we wouldn't want inconsistent blocks in *any* case. */
1412 	u64 bg_ptr, prev_bg_ptr;
1413 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1414 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1415 	struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1416 
1417 	/* The caller got these descriptors from
1418 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1419 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1420 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1421 
1422 	trace_ocfs2_relink_block_group(
1423 		(unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1424 		(unsigned long long)le64_to_cpu(bg->bg_blkno),
1425 		(unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1426 
1427 	bg_ptr = le64_to_cpu(bg->bg_next_group);
1428 	prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1429 
1430 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1431 					 prev_bg_bh,
1432 					 OCFS2_JOURNAL_ACCESS_WRITE);
1433 	if (status < 0)
1434 		goto out;
1435 
1436 	prev_bg->bg_next_group = bg->bg_next_group;
1437 	ocfs2_journal_dirty(handle, prev_bg_bh);
1438 
1439 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1440 					 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1441 	if (status < 0)
1442 		goto out_rollback_prev_bg;
1443 
1444 	bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1445 	ocfs2_journal_dirty(handle, bg_bh);
1446 
1447 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1448 					 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1449 	if (status < 0)
1450 		goto out_rollback_bg;
1451 
1452 	fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1453 	ocfs2_journal_dirty(handle, fe_bh);
1454 
1455 out:
1456 	if (status < 0)
1457 		mlog_errno(status);
1458 	return status;
1459 
1460 out_rollback_bg:
1461 	bg->bg_next_group = cpu_to_le64(bg_ptr);
1462 out_rollback_prev_bg:
1463 	prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1464 	goto out;
1465 }
1466 
1467 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1468 						     u32 wanted)
1469 {
1470 	return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1471 }
1472 
1473 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1474  * value on error. */
1475 static int ocfs2_cluster_group_search(struct inode *inode,
1476 				      struct buffer_head *group_bh,
1477 				      u32 bits_wanted, u32 min_bits,
1478 				      u64 max_block,
1479 				      struct ocfs2_suballoc_result *res)
1480 {
1481 	int search = -ENOSPC;
1482 	int ret;
1483 	u64 blkoff;
1484 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1485 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1486 	unsigned int max_bits, gd_cluster_off;
1487 
1488 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1489 
1490 	if (gd->bg_free_bits_count) {
1491 		max_bits = le16_to_cpu(gd->bg_bits);
1492 
1493 		/* Tail groups in cluster bitmaps which aren't cpg
1494 		 * aligned are prone to partial extension by a failed
1495 		 * fs resize. If the file system resize never got to
1496 		 * update the dinode cluster count, then we don't want
1497 		 * to trust any clusters past it, regardless of what
1498 		 * the group descriptor says. */
1499 		gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1500 							  le64_to_cpu(gd->bg_blkno));
1501 		if ((gd_cluster_off + max_bits) >
1502 		    OCFS2_I(inode)->ip_clusters) {
1503 			max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1504 			trace_ocfs2_cluster_group_search_wrong_max_bits(
1505 				(unsigned long long)le64_to_cpu(gd->bg_blkno),
1506 				le16_to_cpu(gd->bg_bits),
1507 				OCFS2_I(inode)->ip_clusters, max_bits);
1508 		}
1509 
1510 		ret = ocfs2_block_group_find_clear_bits(osb,
1511 							group_bh, bits_wanted,
1512 							max_bits, res);
1513 		if (ret)
1514 			return ret;
1515 
1516 		if (max_block) {
1517 			blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1518 							  gd_cluster_off +
1519 							  res->sr_bit_offset +
1520 							  res->sr_bits);
1521 			trace_ocfs2_cluster_group_search_max_block(
1522 				(unsigned long long)blkoff,
1523 				(unsigned long long)max_block);
1524 			if (blkoff > max_block)
1525 				return -ENOSPC;
1526 		}
1527 
1528 		/* ocfs2_block_group_find_clear_bits() might
1529 		 * return success, but we still want to return
1530 		 * -ENOSPC unless it found the minimum number
1531 		 * of bits. */
1532 		if (min_bits <= res->sr_bits)
1533 			search = 0; /* success */
1534 		else if (res->sr_bits) {
1535 			/*
1536 			 * Don't show bits which we'll be returning
1537 			 * for allocation to the local alloc bitmap.
1538 			 */
1539 			ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
1540 		}
1541 	}
1542 
1543 	return search;
1544 }
1545 
1546 static int ocfs2_block_group_search(struct inode *inode,
1547 				    struct buffer_head *group_bh,
1548 				    u32 bits_wanted, u32 min_bits,
1549 				    u64 max_block,
1550 				    struct ocfs2_suballoc_result *res)
1551 {
1552 	int ret = -ENOSPC;
1553 	u64 blkoff;
1554 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1555 
1556 	BUG_ON(min_bits != 1);
1557 	BUG_ON(ocfs2_is_cluster_bitmap(inode));
1558 
1559 	if (bg->bg_free_bits_count) {
1560 		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1561 							group_bh, bits_wanted,
1562 							le16_to_cpu(bg->bg_bits),
1563 							res);
1564 		if (!ret && max_block) {
1565 			blkoff = le64_to_cpu(bg->bg_blkno) +
1566 				res->sr_bit_offset + res->sr_bits;
1567 			trace_ocfs2_block_group_search_max_block(
1568 				(unsigned long long)blkoff,
1569 				(unsigned long long)max_block);
1570 			if (blkoff > max_block)
1571 				ret = -ENOSPC;
1572 		}
1573 	}
1574 
1575 	return ret;
1576 }
1577 
1578 int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1579 				       handle_t *handle,
1580 				       struct buffer_head *di_bh,
1581 				       u32 num_bits,
1582 				       u16 chain)
1583 {
1584 	int ret;
1585 	u32 tmp_used;
1586 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1587 	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1588 
1589 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1590 				      OCFS2_JOURNAL_ACCESS_WRITE);
1591 	if (ret < 0) {
1592 		mlog_errno(ret);
1593 		goto out;
1594 	}
1595 
1596 	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1597 	di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1598 	le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1599 	ocfs2_journal_dirty(handle, di_bh);
1600 
1601 out:
1602 	return ret;
1603 }
1604 
1605 void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
1606 				       struct buffer_head *di_bh,
1607 				       u32 num_bits,
1608 				       u16 chain)
1609 {
1610 	u32 tmp_used;
1611 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1612 	struct ocfs2_chain_list *cl;
1613 
1614 	cl = (struct ocfs2_chain_list *)&di->id2.i_chain;
1615 	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1616 	di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits);
1617 	le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits);
1618 }
1619 
1620 static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1621 					 struct ocfs2_extent_rec *rec,
1622 					 struct ocfs2_chain_list *cl)
1623 {
1624 	unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1625 	unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1626 	unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1627 
1628 	if (res->sr_bit_offset < bitoff)
1629 		return 0;
1630 	if (res->sr_bit_offset >= (bitoff + bitcount))
1631 		return 0;
1632 	res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1633 		(res->sr_bit_offset - bitoff);
1634 	if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1635 		res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1636 	return 1;
1637 }
1638 
1639 static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1640 					  struct ocfs2_group_desc *bg,
1641 					  struct ocfs2_suballoc_result *res)
1642 {
1643 	int i;
1644 	u64 bg_blkno = res->sr_bg_blkno;  /* Save off */
1645 	struct ocfs2_extent_rec *rec;
1646 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1647 	struct ocfs2_chain_list *cl = &di->id2.i_chain;
1648 
1649 	if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1650 		res->sr_blkno = 0;
1651 		return;
1652 	}
1653 
1654 	res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1655 	res->sr_bg_blkno = 0;  /* Clear it for contig block groups */
1656 	if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1657 	    !bg->bg_list.l_next_free_rec)
1658 		return;
1659 
1660 	for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1661 		rec = &bg->bg_list.l_recs[i];
1662 		if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1663 			res->sr_bg_blkno = bg_blkno;  /* Restore */
1664 			break;
1665 		}
1666 	}
1667 }
1668 
1669 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1670 				  handle_t *handle,
1671 				  u32 bits_wanted,
1672 				  u32 min_bits,
1673 				  struct ocfs2_suballoc_result *res,
1674 				  u16 *bits_left)
1675 {
1676 	int ret;
1677 	struct buffer_head *group_bh = NULL;
1678 	struct ocfs2_group_desc *gd;
1679 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1680 	struct inode *alloc_inode = ac->ac_inode;
1681 
1682 	ret = ocfs2_read_group_descriptor(alloc_inode, di,
1683 					  res->sr_bg_blkno, &group_bh);
1684 	if (ret < 0) {
1685 		mlog_errno(ret);
1686 		return ret;
1687 	}
1688 
1689 	gd = (struct ocfs2_group_desc *) group_bh->b_data;
1690 	ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1691 				  ac->ac_max_block, res);
1692 	if (ret < 0) {
1693 		if (ret != -ENOSPC)
1694 			mlog_errno(ret);
1695 		goto out;
1696 	}
1697 
1698 	if (!ret)
1699 		ocfs2_bg_discontig_fix_result(ac, gd, res);
1700 
1701 	/*
1702 	 * sr_bg_blkno might have been changed by
1703 	 * ocfs2_bg_discontig_fix_result
1704 	 */
1705 	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1706 
1707 	if (ac->ac_find_loc_only)
1708 		goto out_loc_only;
1709 
1710 	ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1711 					       res->sr_bits,
1712 					       le16_to_cpu(gd->bg_chain));
1713 	if (ret < 0) {
1714 		mlog_errno(ret);
1715 		goto out;
1716 	}
1717 
1718 	ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1719 					 res->sr_bit_offset, res->sr_bits);
1720 	if (ret < 0) {
1721 		ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh,
1722 					       res->sr_bits,
1723 					       le16_to_cpu(gd->bg_chain));
1724 		mlog_errno(ret);
1725 	}
1726 
1727 out_loc_only:
1728 	*bits_left = le16_to_cpu(gd->bg_free_bits_count);
1729 
1730 out:
1731 	brelse(group_bh);
1732 
1733 	return ret;
1734 }
1735 
1736 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1737 			      handle_t *handle,
1738 			      u32 bits_wanted,
1739 			      u32 min_bits,
1740 			      struct ocfs2_suballoc_result *res,
1741 			      u16 *bits_left)
1742 {
1743 	int status;
1744 	u16 chain;
1745 	u64 next_group;
1746 	struct inode *alloc_inode = ac->ac_inode;
1747 	struct buffer_head *group_bh = NULL;
1748 	struct buffer_head *prev_group_bh = NULL;
1749 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1750 	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1751 	struct ocfs2_group_desc *bg;
1752 
1753 	chain = ac->ac_chain;
1754 	trace_ocfs2_search_chain_begin(
1755 		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1756 		bits_wanted, chain);
1757 
1758 	status = ocfs2_read_group_descriptor(alloc_inode, fe,
1759 					     le64_to_cpu(cl->cl_recs[chain].c_blkno),
1760 					     &group_bh);
1761 	if (status < 0) {
1762 		mlog_errno(status);
1763 		goto bail;
1764 	}
1765 	bg = (struct ocfs2_group_desc *) group_bh->b_data;
1766 
1767 	status = -ENOSPC;
1768 	/* for now, the chain search is a bit simplistic. We just use
1769 	 * the 1st group with any empty bits. */
1770 	while ((status = ac->ac_group_search(alloc_inode, group_bh,
1771 					     bits_wanted, min_bits,
1772 					     ac->ac_max_block,
1773 					     res)) == -ENOSPC) {
1774 		if (!bg->bg_next_group)
1775 			break;
1776 
1777 		brelse(prev_group_bh);
1778 		prev_group_bh = NULL;
1779 
1780 		next_group = le64_to_cpu(bg->bg_next_group);
1781 		prev_group_bh = group_bh;
1782 		group_bh = NULL;
1783 		status = ocfs2_read_group_descriptor(alloc_inode, fe,
1784 						     next_group, &group_bh);
1785 		if (status < 0) {
1786 			mlog_errno(status);
1787 			goto bail;
1788 		}
1789 		bg = (struct ocfs2_group_desc *) group_bh->b_data;
1790 	}
1791 	if (status < 0) {
1792 		if (status != -ENOSPC)
1793 			mlog_errno(status);
1794 		goto bail;
1795 	}
1796 
1797 	trace_ocfs2_search_chain_succ(
1798 		(unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1799 
1800 	res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1801 
1802 	BUG_ON(res->sr_bits == 0);
1803 	if (!status)
1804 		ocfs2_bg_discontig_fix_result(ac, bg, res);
1805 
1806 	/*
1807 	 * sr_bg_blkno might have been changed by
1808 	 * ocfs2_bg_discontig_fix_result
1809 	 */
1810 	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1811 
1812 	/*
1813 	 * Keep track of previous block descriptor read. When
1814 	 * we find a target, if we have read more than X
1815 	 * number of descriptors, and the target is reasonably
1816 	 * empty, relink him to top of his chain.
1817 	 *
1818 	 * We've read 0 extra blocks and only send one more to
1819 	 * the transaction, yet the next guy to search has a
1820 	 * much easier time.
1821 	 *
1822 	 * Do this *after* figuring out how many bits we're taking out
1823 	 * of our target group.
1824 	 */
1825 	if (!ac->ac_disable_chain_relink &&
1826 	    (prev_group_bh) &&
1827 	    (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1828 		status = ocfs2_relink_block_group(handle, alloc_inode,
1829 						  ac->ac_bh, group_bh,
1830 						  prev_group_bh, chain);
1831 		if (status < 0) {
1832 			mlog_errno(status);
1833 			goto bail;
1834 		}
1835 	}
1836 
1837 	if (ac->ac_find_loc_only)
1838 		goto out_loc_only;
1839 
1840 	status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1841 						  ac->ac_bh, res->sr_bits,
1842 						  chain);
1843 	if (status) {
1844 		mlog_errno(status);
1845 		goto bail;
1846 	}
1847 
1848 	status = ocfs2_block_group_set_bits(handle,
1849 					    alloc_inode,
1850 					    bg,
1851 					    group_bh,
1852 					    res->sr_bit_offset,
1853 					    res->sr_bits);
1854 	if (status < 0) {
1855 		ocfs2_rollback_alloc_dinode_counts(alloc_inode,
1856 					ac->ac_bh, res->sr_bits, chain);
1857 		mlog_errno(status);
1858 		goto bail;
1859 	}
1860 
1861 	trace_ocfs2_search_chain_end(
1862 			(unsigned long long)le64_to_cpu(fe->i_blkno),
1863 			res->sr_bits);
1864 
1865 out_loc_only:
1866 	*bits_left = le16_to_cpu(bg->bg_free_bits_count);
1867 bail:
1868 	brelse(group_bh);
1869 	brelse(prev_group_bh);
1870 
1871 	if (status)
1872 		mlog_errno(status);
1873 	return status;
1874 }
1875 
1876 /* will give out up to bits_wanted contiguous bits. */
1877 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1878 				     handle_t *handle,
1879 				     u32 bits_wanted,
1880 				     u32 min_bits,
1881 				     struct ocfs2_suballoc_result *res)
1882 {
1883 	int status;
1884 	u16 victim, i;
1885 	u16 bits_left = 0;
1886 	u64 hint = ac->ac_last_group;
1887 	struct ocfs2_chain_list *cl;
1888 	struct ocfs2_dinode *fe;
1889 
1890 	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1891 	BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1892 	BUG_ON(!ac->ac_bh);
1893 
1894 	fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1895 
1896 	/* The bh was validated by the inode read during
1897 	 * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1898 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1899 
1900 	if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1901 	    le32_to_cpu(fe->id1.bitmap1.i_total)) {
1902 		status = ocfs2_error(ac->ac_inode->i_sb,
1903 				     "Chain allocator dinode %llu has %u used bits but only %u total\n",
1904 				     (unsigned long long)le64_to_cpu(fe->i_blkno),
1905 				     le32_to_cpu(fe->id1.bitmap1.i_used),
1906 				     le32_to_cpu(fe->id1.bitmap1.i_total));
1907 		goto bail;
1908 	}
1909 
1910 	res->sr_bg_blkno = hint;
1911 	if (res->sr_bg_blkno) {
1912 		/* Attempt to short-circuit the usual search mechanism
1913 		 * by jumping straight to the most recently used
1914 		 * allocation group. This helps us maintain some
1915 		 * contiguousness across allocations. */
1916 		status = ocfs2_search_one_group(ac, handle, bits_wanted,
1917 						min_bits, res, &bits_left);
1918 		if (!status)
1919 			goto set_hint;
1920 		if (status < 0 && status != -ENOSPC) {
1921 			mlog_errno(status);
1922 			goto bail;
1923 		}
1924 	}
1925 
1926 	cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1927 
1928 	victim = ocfs2_find_victim_chain(cl);
1929 	ac->ac_chain = victim;
1930 
1931 	status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1932 				    res, &bits_left);
1933 	if (!status) {
1934 		if (ocfs2_is_cluster_bitmap(ac->ac_inode))
1935 			hint = res->sr_bg_blkno;
1936 		else
1937 			hint = ocfs2_group_from_res(res);
1938 		goto set_hint;
1939 	}
1940 	if (status < 0 && status != -ENOSPC) {
1941 		mlog_errno(status);
1942 		goto bail;
1943 	}
1944 
1945 	trace_ocfs2_claim_suballoc_bits(victim);
1946 
1947 	/* If we didn't pick a good victim, then just default to
1948 	 * searching each chain in order. Don't allow chain relinking
1949 	 * because we only calculate enough journal credits for one
1950 	 * relink per alloc. */
1951 	ac->ac_disable_chain_relink = 1;
1952 	for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1953 		if (i == victim)
1954 			continue;
1955 		if (!cl->cl_recs[i].c_free)
1956 			continue;
1957 
1958 		ac->ac_chain = i;
1959 		status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1960 					    res, &bits_left);
1961 		if (!status) {
1962 			hint = ocfs2_group_from_res(res);
1963 			break;
1964 		}
1965 		if (status < 0 && status != -ENOSPC) {
1966 			mlog_errno(status);
1967 			goto bail;
1968 		}
1969 	}
1970 
1971 set_hint:
1972 	if (status != -ENOSPC) {
1973 		/* If the next search of this group is not likely to
1974 		 * yield a suitable extent, then we reset the last
1975 		 * group hint so as to not waste a disk read */
1976 		if (bits_left < min_bits)
1977 			ac->ac_last_group = 0;
1978 		else
1979 			ac->ac_last_group = hint;
1980 	}
1981 
1982 bail:
1983 	if (status)
1984 		mlog_errno(status);
1985 	return status;
1986 }
1987 
1988 int ocfs2_claim_metadata(handle_t *handle,
1989 			 struct ocfs2_alloc_context *ac,
1990 			 u32 bits_wanted,
1991 			 u64 *suballoc_loc,
1992 			 u16 *suballoc_bit_start,
1993 			 unsigned int *num_bits,
1994 			 u64 *blkno_start)
1995 {
1996 	int status;
1997 	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
1998 
1999 	BUG_ON(!ac);
2000 	BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
2001 	BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
2002 
2003 	status = ocfs2_claim_suballoc_bits(ac,
2004 					   handle,
2005 					   bits_wanted,
2006 					   1,
2007 					   &res);
2008 	if (status < 0) {
2009 		mlog_errno(status);
2010 		goto bail;
2011 	}
2012 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2013 
2014 	*suballoc_loc = res.sr_bg_blkno;
2015 	*suballoc_bit_start = res.sr_bit_offset;
2016 	*blkno_start = res.sr_blkno;
2017 	ac->ac_bits_given += res.sr_bits;
2018 	*num_bits = res.sr_bits;
2019 	status = 0;
2020 bail:
2021 	if (status)
2022 		mlog_errno(status);
2023 	return status;
2024 }
2025 
2026 static void ocfs2_init_inode_ac_group(struct inode *dir,
2027 				      struct buffer_head *parent_di_bh,
2028 				      struct ocfs2_alloc_context *ac)
2029 {
2030 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2031 	/*
2032 	 * Try to allocate inodes from some specific group.
2033 	 *
2034 	 * If the parent dir has recorded the last group used in allocation,
2035 	 * cool, use it. Otherwise if we try to allocate new inode from the
2036 	 * same slot the parent dir belongs to, use the same chunk.
2037 	 *
2038 	 * We are very careful here to avoid the mistake of setting
2039 	 * ac_last_group to a group descriptor from a different (unlocked) slot.
2040 	 */
2041 	if (OCFS2_I(dir)->ip_last_used_group &&
2042 	    OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2043 		ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2044 	else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2045 		if (di->i_suballoc_loc)
2046 			ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2047 		else
2048 			ac->ac_last_group = ocfs2_which_suballoc_group(
2049 					le64_to_cpu(di->i_blkno),
2050 					le16_to_cpu(di->i_suballoc_bit));
2051 	}
2052 }
2053 
2054 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2055 					     struct ocfs2_alloc_context *ac)
2056 {
2057 	OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2058 	OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2059 }
2060 
2061 int ocfs2_find_new_inode_loc(struct inode *dir,
2062 			     struct buffer_head *parent_fe_bh,
2063 			     struct ocfs2_alloc_context *ac,
2064 			     u64 *fe_blkno)
2065 {
2066 	int ret;
2067 	handle_t *handle = NULL;
2068 	struct ocfs2_suballoc_result *res;
2069 
2070 	BUG_ON(!ac);
2071 	BUG_ON(ac->ac_bits_given != 0);
2072 	BUG_ON(ac->ac_bits_wanted != 1);
2073 	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2074 
2075 	res = kzalloc(sizeof(*res), GFP_NOFS);
2076 	if (res == NULL) {
2077 		ret = -ENOMEM;
2078 		mlog_errno(ret);
2079 		goto out;
2080 	}
2081 
2082 	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2083 
2084 	/*
2085 	 * The handle started here is for chain relink. Alternatively,
2086 	 * we could just disable relink for these calls.
2087 	 */
2088 	handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2089 	if (IS_ERR(handle)) {
2090 		ret = PTR_ERR(handle);
2091 		handle = NULL;
2092 		mlog_errno(ret);
2093 		goto out;
2094 	}
2095 
2096 	/*
2097 	 * This will instruct ocfs2_claim_suballoc_bits and
2098 	 * ocfs2_search_one_group to search but save actual allocation
2099 	 * for later.
2100 	 */
2101 	ac->ac_find_loc_only = 1;
2102 
2103 	ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2104 	if (ret < 0) {
2105 		mlog_errno(ret);
2106 		goto out;
2107 	}
2108 
2109 	ac->ac_find_loc_priv = res;
2110 	*fe_blkno = res->sr_blkno;
2111 	ocfs2_update_inode_fsync_trans(handle, dir, 0);
2112 out:
2113 	if (handle)
2114 		ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2115 
2116 	if (ret)
2117 		kfree(res);
2118 
2119 	return ret;
2120 }
2121 
2122 int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2123 				 struct inode *dir,
2124 				 struct ocfs2_alloc_context *ac,
2125 				 u64 *suballoc_loc,
2126 				 u16 *suballoc_bit,
2127 				 u64 di_blkno)
2128 {
2129 	int ret;
2130 	u16 chain;
2131 	struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2132 	struct buffer_head *bg_bh = NULL;
2133 	struct ocfs2_group_desc *bg;
2134 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2135 
2136 	/*
2137 	 * Since di_blkno is being passed back in, we check for any
2138 	 * inconsistencies which may have happened between
2139 	 * calls. These are code bugs as di_blkno is not expected to
2140 	 * change once returned from ocfs2_find_new_inode_loc()
2141 	 */
2142 	BUG_ON(res->sr_blkno != di_blkno);
2143 
2144 	ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2145 					  res->sr_bg_stable_blkno, &bg_bh);
2146 	if (ret) {
2147 		mlog_errno(ret);
2148 		goto out;
2149 	}
2150 
2151 	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2152 	chain = le16_to_cpu(bg->bg_chain);
2153 
2154 	ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2155 					       ac->ac_bh, res->sr_bits,
2156 					       chain);
2157 	if (ret) {
2158 		mlog_errno(ret);
2159 		goto out;
2160 	}
2161 
2162 	ret = ocfs2_block_group_set_bits(handle,
2163 					 ac->ac_inode,
2164 					 bg,
2165 					 bg_bh,
2166 					 res->sr_bit_offset,
2167 					 res->sr_bits);
2168 	if (ret < 0) {
2169 		ocfs2_rollback_alloc_dinode_counts(ac->ac_inode,
2170 					       ac->ac_bh, res->sr_bits, chain);
2171 		mlog_errno(ret);
2172 		goto out;
2173 	}
2174 
2175 	trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2176 					   res->sr_bits);
2177 
2178 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2179 
2180 	BUG_ON(res->sr_bits != 1);
2181 
2182 	*suballoc_loc = res->sr_bg_blkno;
2183 	*suballoc_bit = res->sr_bit_offset;
2184 	ac->ac_bits_given++;
2185 	ocfs2_save_inode_ac_group(dir, ac);
2186 
2187 out:
2188 	brelse(bg_bh);
2189 
2190 	return ret;
2191 }
2192 
2193 int ocfs2_claim_new_inode(handle_t *handle,
2194 			  struct inode *dir,
2195 			  struct buffer_head *parent_fe_bh,
2196 			  struct ocfs2_alloc_context *ac,
2197 			  u64 *suballoc_loc,
2198 			  u16 *suballoc_bit,
2199 			  u64 *fe_blkno)
2200 {
2201 	int status;
2202 	struct ocfs2_suballoc_result res;
2203 
2204 	BUG_ON(!ac);
2205 	BUG_ON(ac->ac_bits_given != 0);
2206 	BUG_ON(ac->ac_bits_wanted != 1);
2207 	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2208 
2209 	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2210 
2211 	status = ocfs2_claim_suballoc_bits(ac,
2212 					   handle,
2213 					   1,
2214 					   1,
2215 					   &res);
2216 	if (status < 0) {
2217 		mlog_errno(status);
2218 		goto bail;
2219 	}
2220 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2221 
2222 	BUG_ON(res.sr_bits != 1);
2223 
2224 	*suballoc_loc = res.sr_bg_blkno;
2225 	*suballoc_bit = res.sr_bit_offset;
2226 	*fe_blkno = res.sr_blkno;
2227 	ac->ac_bits_given++;
2228 	ocfs2_save_inode_ac_group(dir, ac);
2229 	status = 0;
2230 bail:
2231 	if (status)
2232 		mlog_errno(status);
2233 	return status;
2234 }
2235 
2236 /* translate a group desc. blkno and it's bitmap offset into
2237  * disk cluster offset. */
2238 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2239 						   u64 bg_blkno,
2240 						   u16 bg_bit_off)
2241 {
2242 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2243 	u32 cluster = 0;
2244 
2245 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2246 
2247 	if (bg_blkno != osb->first_cluster_group_blkno)
2248 		cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2249 	cluster += (u32) bg_bit_off;
2250 	return cluster;
2251 }
2252 
2253 /* given a cluster offset, calculate which block group it belongs to
2254  * and return that block offset. */
2255 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2256 {
2257 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2258 	u32 group_no;
2259 
2260 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2261 
2262 	group_no = cluster / osb->bitmap_cpg;
2263 	if (!group_no)
2264 		return osb->first_cluster_group_blkno;
2265 	return ocfs2_clusters_to_blocks(inode->i_sb,
2266 					group_no * osb->bitmap_cpg);
2267 }
2268 
2269 /* given the block number of a cluster start, calculate which cluster
2270  * group and descriptor bitmap offset that corresponds to. */
2271 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2272 						u64 data_blkno,
2273 						u64 *bg_blkno,
2274 						u16 *bg_bit_off)
2275 {
2276 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2277 	u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2278 
2279 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2280 
2281 	*bg_blkno = ocfs2_which_cluster_group(inode,
2282 					      data_cluster);
2283 
2284 	if (*bg_blkno == osb->first_cluster_group_blkno)
2285 		*bg_bit_off = (u16) data_cluster;
2286 	else
2287 		*bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2288 							     data_blkno - *bg_blkno);
2289 }
2290 
2291 /*
2292  * min_bits - minimum contiguous chunk from this total allocation we
2293  * can handle. set to what we asked for originally for a full
2294  * contig. allocation, set to '1' to indicate we can deal with extents
2295  * of any size.
2296  */
2297 int __ocfs2_claim_clusters(handle_t *handle,
2298 			   struct ocfs2_alloc_context *ac,
2299 			   u32 min_clusters,
2300 			   u32 max_clusters,
2301 			   u32 *cluster_start,
2302 			   u32 *num_clusters)
2303 {
2304 	int status;
2305 	unsigned int bits_wanted = max_clusters;
2306 	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2307 	struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2308 
2309 	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2310 
2311 	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2312 	       && ac->ac_which != OCFS2_AC_USE_MAIN);
2313 
2314 	if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2315 		WARN_ON(min_clusters > 1);
2316 
2317 		status = ocfs2_claim_local_alloc_bits(osb,
2318 						      handle,
2319 						      ac,
2320 						      bits_wanted,
2321 						      cluster_start,
2322 						      num_clusters);
2323 		if (!status)
2324 			atomic_inc(&osb->alloc_stats.local_data);
2325 	} else {
2326 		if (min_clusters > (osb->bitmap_cpg - 1)) {
2327 			/* The only paths asking for contiguousness
2328 			 * should know about this already. */
2329 			mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2330 			     "group bitmap size %u!\n", min_clusters,
2331 			     osb->bitmap_cpg);
2332 			status = -ENOSPC;
2333 			goto bail;
2334 		}
2335 		/* clamp the current request down to a realistic size. */
2336 		if (bits_wanted > (osb->bitmap_cpg - 1))
2337 			bits_wanted = osb->bitmap_cpg - 1;
2338 
2339 		status = ocfs2_claim_suballoc_bits(ac,
2340 						   handle,
2341 						   bits_wanted,
2342 						   min_clusters,
2343 						   &res);
2344 		if (!status) {
2345 			BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2346 			*cluster_start =
2347 				ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2348 								 res.sr_bg_blkno,
2349 								 res.sr_bit_offset);
2350 			atomic_inc(&osb->alloc_stats.bitmap_data);
2351 			*num_clusters = res.sr_bits;
2352 		}
2353 	}
2354 	if (status < 0) {
2355 		if (status != -ENOSPC)
2356 			mlog_errno(status);
2357 		goto bail;
2358 	}
2359 
2360 	ac->ac_bits_given += *num_clusters;
2361 
2362 bail:
2363 	if (status)
2364 		mlog_errno(status);
2365 	return status;
2366 }
2367 
2368 int ocfs2_claim_clusters(handle_t *handle,
2369 			 struct ocfs2_alloc_context *ac,
2370 			 u32 min_clusters,
2371 			 u32 *cluster_start,
2372 			 u32 *num_clusters)
2373 {
2374 	unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2375 
2376 	return __ocfs2_claim_clusters(handle, ac, min_clusters,
2377 				      bits_wanted, cluster_start, num_clusters);
2378 }
2379 
2380 static int ocfs2_block_group_clear_bits(handle_t *handle,
2381 					struct inode *alloc_inode,
2382 					struct ocfs2_group_desc *bg,
2383 					struct buffer_head *group_bh,
2384 					unsigned int bit_off,
2385 					unsigned int num_bits,
2386 					void (*undo_fn)(unsigned int bit,
2387 							unsigned long *bmap))
2388 {
2389 	int status;
2390 	unsigned int tmp;
2391 	struct ocfs2_group_desc *undo_bg = NULL;
2392 	struct journal_head *jh;
2393 
2394 	/* The caller got this descriptor from
2395 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
2396 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2397 
2398 	trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2399 
2400 	BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2401 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2402 					 group_bh,
2403 					 undo_fn ?
2404 					 OCFS2_JOURNAL_ACCESS_UNDO :
2405 					 OCFS2_JOURNAL_ACCESS_WRITE);
2406 	if (status < 0) {
2407 		mlog_errno(status);
2408 		goto bail;
2409 	}
2410 
2411 	jh = bh2jh(group_bh);
2412 	if (undo_fn) {
2413 		spin_lock(&jh->b_state_lock);
2414 		undo_bg = (struct ocfs2_group_desc *) jh->b_committed_data;
2415 		BUG_ON(!undo_bg);
2416 	}
2417 
2418 	tmp = num_bits;
2419 	while(tmp--) {
2420 		ocfs2_clear_bit((bit_off + tmp),
2421 				(unsigned long *) bg->bg_bitmap);
2422 		if (undo_fn)
2423 			undo_fn(bit_off + tmp,
2424 				(unsigned long *) undo_bg->bg_bitmap);
2425 	}
2426 	le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2427 	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2428 		if (undo_fn)
2429 			spin_unlock(&jh->b_state_lock);
2430 		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
2431 				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
2432 				   le16_to_cpu(bg->bg_bits),
2433 				   le16_to_cpu(bg->bg_free_bits_count),
2434 				   num_bits);
2435 	}
2436 
2437 	if (undo_fn)
2438 		spin_unlock(&jh->b_state_lock);
2439 
2440 	ocfs2_journal_dirty(handle, group_bh);
2441 bail:
2442 	return status;
2443 }
2444 
2445 /*
2446  * expects the suballoc inode to already be locked.
2447  */
2448 static int _ocfs2_free_suballoc_bits(handle_t *handle,
2449 				     struct inode *alloc_inode,
2450 				     struct buffer_head *alloc_bh,
2451 				     unsigned int start_bit,
2452 				     u64 bg_blkno,
2453 				     unsigned int count,
2454 				     void (*undo_fn)(unsigned int bit,
2455 						     unsigned long *bitmap))
2456 {
2457 	int status = 0;
2458 	u32 tmp_used;
2459 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2460 	struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2461 	struct buffer_head *group_bh = NULL;
2462 	struct ocfs2_group_desc *group;
2463 
2464 	/* The alloc_bh comes from ocfs2_free_dinode() or
2465 	 * ocfs2_free_clusters().  The callers have all locked the
2466 	 * allocator and gotten alloc_bh from the lock call.  This
2467 	 * validates the dinode buffer.  Any corruption that has happened
2468 	 * is a code bug. */
2469 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2470 	BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2471 
2472 	trace_ocfs2_free_suballoc_bits(
2473 		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2474 		(unsigned long long)bg_blkno,
2475 		start_bit, count);
2476 
2477 	status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2478 					     &group_bh);
2479 	if (status < 0) {
2480 		mlog_errno(status);
2481 		goto bail;
2482 	}
2483 	group = (struct ocfs2_group_desc *) group_bh->b_data;
2484 
2485 	BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2486 
2487 	status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2488 					      group, group_bh,
2489 					      start_bit, count, undo_fn);
2490 	if (status < 0) {
2491 		mlog_errno(status);
2492 		goto bail;
2493 	}
2494 
2495 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2496 					 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2497 	if (status < 0) {
2498 		mlog_errno(status);
2499 		ocfs2_block_group_set_bits(handle, alloc_inode, group, group_bh,
2500 				start_bit, count);
2501 		goto bail;
2502 	}
2503 
2504 	le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2505 		     count);
2506 	tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2507 	fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2508 	ocfs2_journal_dirty(handle, alloc_bh);
2509 
2510 bail:
2511 	brelse(group_bh);
2512 	return status;
2513 }
2514 
2515 int ocfs2_free_suballoc_bits(handle_t *handle,
2516 			     struct inode *alloc_inode,
2517 			     struct buffer_head *alloc_bh,
2518 			     unsigned int start_bit,
2519 			     u64 bg_blkno,
2520 			     unsigned int count)
2521 {
2522 	return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2523 					 start_bit, bg_blkno, count, NULL);
2524 }
2525 
2526 int ocfs2_free_dinode(handle_t *handle,
2527 		      struct inode *inode_alloc_inode,
2528 		      struct buffer_head *inode_alloc_bh,
2529 		      struct ocfs2_dinode *di)
2530 {
2531 	u64 blk = le64_to_cpu(di->i_blkno);
2532 	u16 bit = le16_to_cpu(di->i_suballoc_bit);
2533 	u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2534 
2535 	if (di->i_suballoc_loc)
2536 		bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2537 	return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2538 					inode_alloc_bh, bit, bg_blkno, 1);
2539 }
2540 
2541 static int _ocfs2_free_clusters(handle_t *handle,
2542 				struct inode *bitmap_inode,
2543 				struct buffer_head *bitmap_bh,
2544 				u64 start_blk,
2545 				unsigned int num_clusters,
2546 				void (*undo_fn)(unsigned int bit,
2547 						unsigned long *bitmap))
2548 {
2549 	int status;
2550 	u16 bg_start_bit;
2551 	u64 bg_blkno;
2552 
2553 	/* You can't ever have a contiguous set of clusters
2554 	 * bigger than a block group bitmap so we never have to worry
2555 	 * about looping on them.
2556 	 * This is expensive. We can safely remove once this stuff has
2557 	 * gotten tested really well. */
2558 	BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb,
2559 				ocfs2_blocks_to_clusters(bitmap_inode->i_sb,
2560 							 start_blk)));
2561 
2562 
2563 	ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2564 				     &bg_start_bit);
2565 
2566 	trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2567 			(unsigned long long)start_blk,
2568 			bg_start_bit, num_clusters);
2569 
2570 	status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2571 					   bg_start_bit, bg_blkno,
2572 					   num_clusters, undo_fn);
2573 	if (status < 0) {
2574 		mlog_errno(status);
2575 		goto out;
2576 	}
2577 
2578 	ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2579 					 num_clusters);
2580 
2581 out:
2582 	return status;
2583 }
2584 
2585 int ocfs2_free_clusters(handle_t *handle,
2586 			struct inode *bitmap_inode,
2587 			struct buffer_head *bitmap_bh,
2588 			u64 start_blk,
2589 			unsigned int num_clusters)
2590 {
2591 	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2592 				    start_blk, num_clusters,
2593 				    _ocfs2_set_bit);
2594 }
2595 
2596 /*
2597  * Give never-used clusters back to the global bitmap.  We don't need
2598  * to protect these bits in the undo buffer.
2599  */
2600 int ocfs2_release_clusters(handle_t *handle,
2601 			   struct inode *bitmap_inode,
2602 			   struct buffer_head *bitmap_bh,
2603 			   u64 start_blk,
2604 			   unsigned int num_clusters)
2605 {
2606 	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2607 				    start_blk, num_clusters,
2608 				    _ocfs2_clear_bit);
2609 }
2610 
2611 /*
2612  * For a given allocation, determine which allocators will need to be
2613  * accessed, and lock them, reserving the appropriate number of bits.
2614  *
2615  * Sparse file systems call this from ocfs2_write_begin_nolock()
2616  * and ocfs2_allocate_unwritten_extents().
2617  *
2618  * File systems which don't support holes call this from
2619  * ocfs2_extend_allocation().
2620  */
2621 int ocfs2_lock_allocators(struct inode *inode,
2622 			  struct ocfs2_extent_tree *et,
2623 			  u32 clusters_to_add, u32 extents_to_split,
2624 			  struct ocfs2_alloc_context **data_ac,
2625 			  struct ocfs2_alloc_context **meta_ac)
2626 {
2627 	int ret = 0, num_free_extents;
2628 	unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2629 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2630 
2631 	*meta_ac = NULL;
2632 	if (data_ac)
2633 		*data_ac = NULL;
2634 
2635 	BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2636 
2637 	num_free_extents = ocfs2_num_free_extents(et);
2638 	if (num_free_extents < 0) {
2639 		ret = num_free_extents;
2640 		mlog_errno(ret);
2641 		goto out;
2642 	}
2643 
2644 	/*
2645 	 * Sparse allocation file systems need to be more conservative
2646 	 * with reserving room for expansion - the actual allocation
2647 	 * happens while we've got a journal handle open so re-taking
2648 	 * a cluster lock (because we ran out of room for another
2649 	 * extent) will violate ordering rules.
2650 	 *
2651 	 * Most of the time we'll only be seeing this 1 cluster at a time
2652 	 * anyway.
2653 	 *
2654 	 * Always lock for any unwritten extents - we might want to
2655 	 * add blocks during a split.
2656 	 */
2657 	if (!num_free_extents ||
2658 	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2659 		ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2660 		if (ret < 0) {
2661 			if (ret != -ENOSPC)
2662 				mlog_errno(ret);
2663 			goto out;
2664 		}
2665 	}
2666 
2667 	if (clusters_to_add == 0)
2668 		goto out;
2669 
2670 	ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2671 	if (ret < 0) {
2672 		if (ret != -ENOSPC)
2673 			mlog_errno(ret);
2674 		goto out;
2675 	}
2676 
2677 out:
2678 	if (ret) {
2679 		if (*meta_ac) {
2680 			ocfs2_free_alloc_context(*meta_ac);
2681 			*meta_ac = NULL;
2682 		}
2683 
2684 		/*
2685 		 * We cannot have an error and a non null *data_ac.
2686 		 */
2687 	}
2688 
2689 	return ret;
2690 }
2691 
2692 /*
2693  * Read the inode specified by blkno to get suballoc_slot and
2694  * suballoc_bit.
2695  */
2696 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2697 				       u16 *suballoc_slot, u64 *group_blkno,
2698 				       u16 *suballoc_bit)
2699 {
2700 	int status;
2701 	struct buffer_head *inode_bh = NULL;
2702 	struct ocfs2_dinode *inode_fe;
2703 
2704 	trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2705 
2706 	/* dirty read disk */
2707 	status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2708 	if (status < 0) {
2709 		mlog(ML_ERROR, "read block %llu failed %d\n",
2710 		     (unsigned long long)blkno, status);
2711 		goto bail;
2712 	}
2713 
2714 	inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2715 	if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2716 		mlog(ML_ERROR, "invalid inode %llu requested\n",
2717 		     (unsigned long long)blkno);
2718 		status = -EINVAL;
2719 		goto bail;
2720 	}
2721 
2722 	if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2723 	    (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2724 		mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2725 		     (unsigned long long)blkno,
2726 		     (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2727 		status = -EINVAL;
2728 		goto bail;
2729 	}
2730 
2731 	if (suballoc_slot)
2732 		*suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2733 	if (suballoc_bit)
2734 		*suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2735 	if (group_blkno)
2736 		*group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2737 
2738 bail:
2739 	brelse(inode_bh);
2740 
2741 	if (status)
2742 		mlog_errno(status);
2743 	return status;
2744 }
2745 
2746 /*
2747  * test whether bit is SET in allocator bitmap or not.  on success, 0
2748  * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
2749  * is returned and *res is meaningless.  Call this after you have
2750  * cluster locked against suballoc, or you may get a result based on
2751  * non-up2date contents
2752  */
2753 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2754 				   struct inode *suballoc,
2755 				   struct buffer_head *alloc_bh,
2756 				   u64 group_blkno, u64 blkno,
2757 				   u16 bit, int *res)
2758 {
2759 	struct ocfs2_dinode *alloc_di;
2760 	struct ocfs2_group_desc *group;
2761 	struct buffer_head *group_bh = NULL;
2762 	u64 bg_blkno;
2763 	int status;
2764 
2765 	trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2766 				      (unsigned int)bit);
2767 
2768 	alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2769 	if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2770 		mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2771 		     (unsigned int)bit,
2772 		     ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2773 		status = -EINVAL;
2774 		goto bail;
2775 	}
2776 
2777 	bg_blkno = group_blkno ? group_blkno :
2778 		   ocfs2_which_suballoc_group(blkno, bit);
2779 	status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2780 					     &group_bh);
2781 	if (status < 0) {
2782 		mlog(ML_ERROR, "read group %llu failed %d\n",
2783 		     (unsigned long long)bg_blkno, status);
2784 		goto bail;
2785 	}
2786 
2787 	group = (struct ocfs2_group_desc *) group_bh->b_data;
2788 	*res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2789 
2790 bail:
2791 	brelse(group_bh);
2792 
2793 	if (status)
2794 		mlog_errno(status);
2795 	return status;
2796 }
2797 
2798 /*
2799  * Test if the bit representing this inode (blkno) is set in the
2800  * suballocator.
2801  *
2802  * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2803  *
2804  * In the event of failure, a negative value is returned and *res is
2805  * meaningless.
2806  *
2807  * Callers must make sure to hold nfs_sync_lock to prevent
2808  * ocfs2_delete_inode() on another node from accessing the same
2809  * suballocator concurrently.
2810  */
2811 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2812 {
2813 	int status;
2814 	u64 group_blkno = 0;
2815 	u16 suballoc_bit = 0, suballoc_slot = 0;
2816 	struct inode *inode_alloc_inode;
2817 	struct buffer_head *alloc_bh = NULL;
2818 
2819 	trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2820 
2821 	status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2822 					     &group_blkno, &suballoc_bit);
2823 	if (status < 0) {
2824 		mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2825 		goto bail;
2826 	}
2827 
2828 	inode_alloc_inode =
2829 		ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2830 					    suballoc_slot);
2831 	if (!inode_alloc_inode) {
2832 		/* the error code could be inaccurate, but we are not able to
2833 		 * get the correct one. */
2834 		status = -EINVAL;
2835 		mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2836 		     (u32)suballoc_slot);
2837 		goto bail;
2838 	}
2839 
2840 	inode_lock(inode_alloc_inode);
2841 	status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2842 	if (status < 0) {
2843 		inode_unlock(inode_alloc_inode);
2844 		iput(inode_alloc_inode);
2845 		mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2846 		     (u32)suballoc_slot, status);
2847 		goto bail;
2848 	}
2849 
2850 	status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2851 					 group_blkno, blkno, suballoc_bit, res);
2852 	if (status < 0)
2853 		mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2854 
2855 	ocfs2_inode_unlock(inode_alloc_inode, 0);
2856 	inode_unlock(inode_alloc_inode);
2857 
2858 	iput(inode_alloc_inode);
2859 	brelse(alloc_bh);
2860 bail:
2861 	if (status)
2862 		mlog_errno(status);
2863 	return status;
2864 }
2865