xref: /openbmc/linux/fs/ocfs2/suballoc.c (revision f5b06569)
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * suballoc.c
5  *
6  * metadata alloc and free
7  * Inspired by ext3 block groups.
8  *
9  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
10  *
11  * This program is free software; you can redistribute it and/or
12  * modify it under the terms of the GNU General Public
13  * License as published by the Free Software Foundation; either
14  * version 2 of the License, or (at your option) any later version.
15  *
16  * This program is distributed in the hope that it will be useful,
17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
19  * General Public License for more details.
20  *
21  * You should have received a copy of the GNU General Public
22  * License along with this program; if not, write to the
23  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
24  * Boston, MA 021110-1307, USA.
25  */
26 
27 #include <linux/fs.h>
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
31 
32 #include <cluster/masklog.h>
33 
34 #include "ocfs2.h"
35 
36 #include "alloc.h"
37 #include "blockcheck.h"
38 #include "dlmglue.h"
39 #include "inode.h"
40 #include "journal.h"
41 #include "localalloc.h"
42 #include "suballoc.h"
43 #include "super.h"
44 #include "sysfile.h"
45 #include "uptodate.h"
46 #include "ocfs2_trace.h"
47 
48 #include "buffer_head_io.h"
49 
50 #define NOT_ALLOC_NEW_GROUP		0
51 #define ALLOC_NEW_GROUP			0x1
52 #define ALLOC_GROUPS_FROM_GLOBAL	0x2
53 
54 #define OCFS2_MAX_TO_STEAL		1024
55 
56 struct ocfs2_suballoc_result {
57 	u64		sr_bg_blkno;	/* The bg we allocated from.  Set
58 					   to 0 when a block group is
59 					   contiguous. */
60 	u64		sr_bg_stable_blkno; /*
61 					     * Doesn't change, always
62 					     * set to target block
63 					     * group descriptor
64 					     * block.
65 					     */
66 	u64		sr_blkno;	/* The first allocated block */
67 	unsigned int	sr_bit_offset;	/* The bit in the bg */
68 	unsigned int	sr_bits;	/* How many bits we claimed */
69 };
70 
71 static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
72 {
73 	if (res->sr_blkno == 0)
74 		return 0;
75 
76 	if (res->sr_bg_blkno)
77 		return res->sr_bg_blkno;
78 
79 	return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset);
80 }
81 
82 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg);
83 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe);
84 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
85 static int ocfs2_block_group_fill(handle_t *handle,
86 				  struct inode *alloc_inode,
87 				  struct buffer_head *bg_bh,
88 				  u64 group_blkno,
89 				  unsigned int group_clusters,
90 				  u16 my_chain,
91 				  struct ocfs2_chain_list *cl);
92 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
93 				   struct inode *alloc_inode,
94 				   struct buffer_head *bh,
95 				   u64 max_block,
96 				   u64 *last_alloc_group,
97 				   int flags);
98 
99 static int ocfs2_cluster_group_search(struct inode *inode,
100 				      struct buffer_head *group_bh,
101 				      u32 bits_wanted, u32 min_bits,
102 				      u64 max_block,
103 				      struct ocfs2_suballoc_result *res);
104 static int ocfs2_block_group_search(struct inode *inode,
105 				    struct buffer_head *group_bh,
106 				    u32 bits_wanted, u32 min_bits,
107 				    u64 max_block,
108 				    struct ocfs2_suballoc_result *res);
109 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
110 				     handle_t *handle,
111 				     u32 bits_wanted,
112 				     u32 min_bits,
113 				     struct ocfs2_suballoc_result *res);
114 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
115 					 int nr);
116 static int ocfs2_relink_block_group(handle_t *handle,
117 				    struct inode *alloc_inode,
118 				    struct buffer_head *fe_bh,
119 				    struct buffer_head *bg_bh,
120 				    struct buffer_head *prev_bg_bh,
121 				    u16 chain);
122 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
123 						     u32 wanted);
124 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
125 						   u64 bg_blkno,
126 						   u16 bg_bit_off);
127 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
128 						u64 data_blkno,
129 						u64 *bg_blkno,
130 						u16 *bg_bit_off);
131 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
132 					     u32 bits_wanted, u64 max_block,
133 					     int flags,
134 					     struct ocfs2_alloc_context **ac);
135 
136 void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
137 {
138 	struct inode *inode = ac->ac_inode;
139 
140 	if (inode) {
141 		if (ac->ac_which != OCFS2_AC_USE_LOCAL)
142 			ocfs2_inode_unlock(inode, 1);
143 
144 		inode_unlock(inode);
145 
146 		iput(inode);
147 		ac->ac_inode = NULL;
148 	}
149 	brelse(ac->ac_bh);
150 	ac->ac_bh = NULL;
151 	ac->ac_resv = NULL;
152 	kfree(ac->ac_find_loc_priv);
153 	ac->ac_find_loc_priv = NULL;
154 }
155 
156 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
157 {
158 	ocfs2_free_ac_resource(ac);
159 	kfree(ac);
160 }
161 
162 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
163 {
164 	return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
165 }
166 
167 #define do_error(fmt, ...)						\
168 do {									\
169 	if (resize)							\
170 		mlog(ML_ERROR, fmt, ##__VA_ARGS__);			\
171 	else								\
172 		return ocfs2_error(sb, fmt, ##__VA_ARGS__);		\
173 } while (0)
174 
175 static int ocfs2_validate_gd_self(struct super_block *sb,
176 				  struct buffer_head *bh,
177 				  int resize)
178 {
179 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
180 
181 	if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
182 		do_error("Group descriptor #%llu has bad signature %.*s\n",
183 			 (unsigned long long)bh->b_blocknr, 7,
184 			 gd->bg_signature);
185 	}
186 
187 	if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
188 		do_error("Group descriptor #%llu has an invalid bg_blkno of %llu\n",
189 			 (unsigned long long)bh->b_blocknr,
190 			 (unsigned long long)le64_to_cpu(gd->bg_blkno));
191 	}
192 
193 	if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
194 		do_error("Group descriptor #%llu has an invalid fs_generation of #%u\n",
195 			 (unsigned long long)bh->b_blocknr,
196 			 le32_to_cpu(gd->bg_generation));
197 	}
198 
199 	if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
200 		do_error("Group descriptor #%llu has bit count %u but claims that %u are free\n",
201 			 (unsigned long long)bh->b_blocknr,
202 			 le16_to_cpu(gd->bg_bits),
203 			 le16_to_cpu(gd->bg_free_bits_count));
204 	}
205 
206 	if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
207 		do_error("Group descriptor #%llu has bit count %u but max bitmap bits of %u\n",
208 			 (unsigned long long)bh->b_blocknr,
209 			 le16_to_cpu(gd->bg_bits),
210 			 8 * le16_to_cpu(gd->bg_size));
211 	}
212 
213 	return 0;
214 }
215 
216 static int ocfs2_validate_gd_parent(struct super_block *sb,
217 				    struct ocfs2_dinode *di,
218 				    struct buffer_head *bh,
219 				    int resize)
220 {
221 	unsigned int max_bits;
222 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
223 
224 	if (di->i_blkno != gd->bg_parent_dinode) {
225 		do_error("Group descriptor #%llu has bad parent pointer (%llu, expected %llu)\n",
226 			 (unsigned long long)bh->b_blocknr,
227 			 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
228 			 (unsigned long long)le64_to_cpu(di->i_blkno));
229 	}
230 
231 	max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
232 	if (le16_to_cpu(gd->bg_bits) > max_bits) {
233 		do_error("Group descriptor #%llu has bit count of %u\n",
234 			 (unsigned long long)bh->b_blocknr,
235 			 le16_to_cpu(gd->bg_bits));
236 	}
237 
238 	/* In resize, we may meet the case bg_chain == cl_next_free_rec. */
239 	if ((le16_to_cpu(gd->bg_chain) >
240 	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
241 	    ((le16_to_cpu(gd->bg_chain) ==
242 	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
243 		do_error("Group descriptor #%llu has bad chain %u\n",
244 			 (unsigned long long)bh->b_blocknr,
245 			 le16_to_cpu(gd->bg_chain));
246 	}
247 
248 	return 0;
249 }
250 
251 #undef do_error
252 
253 /*
254  * This version only prints errors.  It does not fail the filesystem, and
255  * exists only for resize.
256  */
257 int ocfs2_check_group_descriptor(struct super_block *sb,
258 				 struct ocfs2_dinode *di,
259 				 struct buffer_head *bh)
260 {
261 	int rc;
262 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
263 
264 	BUG_ON(!buffer_uptodate(bh));
265 
266 	/*
267 	 * If the ecc fails, we return the error but otherwise
268 	 * leave the filesystem running.  We know any error is
269 	 * local to this block.
270 	 */
271 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
272 	if (rc) {
273 		mlog(ML_ERROR,
274 		     "Checksum failed for group descriptor %llu\n",
275 		     (unsigned long long)bh->b_blocknr);
276 	} else
277 		rc = ocfs2_validate_gd_self(sb, bh, 1);
278 	if (!rc)
279 		rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
280 
281 	return rc;
282 }
283 
284 static int ocfs2_validate_group_descriptor(struct super_block *sb,
285 					   struct buffer_head *bh)
286 {
287 	int rc;
288 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
289 
290 	trace_ocfs2_validate_group_descriptor(
291 					(unsigned long long)bh->b_blocknr);
292 
293 	BUG_ON(!buffer_uptodate(bh));
294 
295 	/*
296 	 * If the ecc fails, we return the error but otherwise
297 	 * leave the filesystem running.  We know any error is
298 	 * local to this block.
299 	 */
300 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
301 	if (rc)
302 		return rc;
303 
304 	/*
305 	 * Errors after here are fatal.
306 	 */
307 
308 	return ocfs2_validate_gd_self(sb, bh, 0);
309 }
310 
311 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
312 				u64 gd_blkno, struct buffer_head **bh)
313 {
314 	int rc;
315 	struct buffer_head *tmp = *bh;
316 
317 	rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
318 			      ocfs2_validate_group_descriptor);
319 	if (rc)
320 		goto out;
321 
322 	rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
323 	if (rc) {
324 		brelse(tmp);
325 		goto out;
326 	}
327 
328 	/* If ocfs2_read_block() got us a new bh, pass it up. */
329 	if (!*bh)
330 		*bh = tmp;
331 
332 out:
333 	return rc;
334 }
335 
336 static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
337 					  struct ocfs2_group_desc *bg,
338 					  struct ocfs2_chain_list *cl,
339 					  u64 p_blkno, unsigned int clusters)
340 {
341 	struct ocfs2_extent_list *el = &bg->bg_list;
342 	struct ocfs2_extent_rec *rec;
343 
344 	BUG_ON(!ocfs2_supports_discontig_bg(osb));
345 	if (!el->l_next_free_rec)
346 		el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
347 	rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
348 	rec->e_blkno = cpu_to_le64(p_blkno);
349 	rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
350 				  le16_to_cpu(cl->cl_bpc));
351 	rec->e_leaf_clusters = cpu_to_le16(clusters);
352 	le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
353 	le16_add_cpu(&bg->bg_free_bits_count,
354 		     clusters * le16_to_cpu(cl->cl_bpc));
355 	le16_add_cpu(&el->l_next_free_rec, 1);
356 }
357 
358 static int ocfs2_block_group_fill(handle_t *handle,
359 				  struct inode *alloc_inode,
360 				  struct buffer_head *bg_bh,
361 				  u64 group_blkno,
362 				  unsigned int group_clusters,
363 				  u16 my_chain,
364 				  struct ocfs2_chain_list *cl)
365 {
366 	int status = 0;
367 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
368 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
369 	struct super_block * sb = alloc_inode->i_sb;
370 
371 	if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
372 		status = ocfs2_error(alloc_inode->i_sb,
373 				     "group block (%llu) != b_blocknr (%llu)\n",
374 				     (unsigned long long)group_blkno,
375 				     (unsigned long long) bg_bh->b_blocknr);
376 		goto bail;
377 	}
378 
379 	status = ocfs2_journal_access_gd(handle,
380 					 INODE_CACHE(alloc_inode),
381 					 bg_bh,
382 					 OCFS2_JOURNAL_ACCESS_CREATE);
383 	if (status < 0) {
384 		mlog_errno(status);
385 		goto bail;
386 	}
387 
388 	memset(bg, 0, sb->s_blocksize);
389 	strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
390 	bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
391 	bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
392 						osb->s_feature_incompat));
393 	bg->bg_chain = cpu_to_le16(my_chain);
394 	bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
395 	bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
396 	bg->bg_blkno = cpu_to_le64(group_blkno);
397 	if (group_clusters == le16_to_cpu(cl->cl_cpg))
398 		bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
399 	else
400 		ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
401 					      group_clusters);
402 
403 	/* set the 1st bit in the bitmap to account for the descriptor block */
404 	ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
405 	bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
406 
407 	ocfs2_journal_dirty(handle, bg_bh);
408 
409 	/* There is no need to zero out or otherwise initialize the
410 	 * other blocks in a group - All valid FS metadata in a block
411 	 * group stores the superblock fs_generation value at
412 	 * allocation time. */
413 
414 bail:
415 	if (status)
416 		mlog_errno(status);
417 	return status;
418 }
419 
420 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
421 {
422 	u16 curr, best;
423 
424 	best = curr = 0;
425 	while (curr < le16_to_cpu(cl->cl_count)) {
426 		if (le32_to_cpu(cl->cl_recs[best].c_total) >
427 		    le32_to_cpu(cl->cl_recs[curr].c_total))
428 			best = curr;
429 		curr++;
430 	}
431 	return best;
432 }
433 
434 static struct buffer_head *
435 ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
436 			       struct inode *alloc_inode,
437 			       struct ocfs2_alloc_context *ac,
438 			       struct ocfs2_chain_list *cl)
439 {
440 	int status;
441 	u32 bit_off, num_bits;
442 	u64 bg_blkno;
443 	struct buffer_head *bg_bh;
444 	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
445 
446 	status = ocfs2_claim_clusters(handle, ac,
447 				      le16_to_cpu(cl->cl_cpg), &bit_off,
448 				      &num_bits);
449 	if (status < 0) {
450 		if (status != -ENOSPC)
451 			mlog_errno(status);
452 		goto bail;
453 	}
454 
455 	/* setup the group */
456 	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
457 	trace_ocfs2_block_group_alloc_contig(
458 	     (unsigned long long)bg_blkno, alloc_rec);
459 
460 	bg_bh = sb_getblk(osb->sb, bg_blkno);
461 	if (!bg_bh) {
462 		status = -ENOMEM;
463 		mlog_errno(status);
464 		goto bail;
465 	}
466 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
467 
468 	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
469 					bg_blkno, num_bits, alloc_rec, cl);
470 	if (status < 0) {
471 		brelse(bg_bh);
472 		mlog_errno(status);
473 	}
474 
475 bail:
476 	return status ? ERR_PTR(status) : bg_bh;
477 }
478 
479 static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
480 					handle_t *handle,
481 					struct ocfs2_alloc_context *ac,
482 					unsigned int min_bits,
483 					u32 *bit_off, u32 *num_bits)
484 {
485 	int status = 0;
486 
487 	while (min_bits) {
488 		status = ocfs2_claim_clusters(handle, ac, min_bits,
489 					      bit_off, num_bits);
490 		if (status != -ENOSPC)
491 			break;
492 
493 		min_bits >>= 1;
494 	}
495 
496 	return status;
497 }
498 
499 static int ocfs2_block_group_grow_discontig(handle_t *handle,
500 					    struct inode *alloc_inode,
501 					    struct buffer_head *bg_bh,
502 					    struct ocfs2_alloc_context *ac,
503 					    struct ocfs2_chain_list *cl,
504 					    unsigned int min_bits)
505 {
506 	int status;
507 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
508 	struct ocfs2_group_desc *bg =
509 		(struct ocfs2_group_desc *)bg_bh->b_data;
510 	unsigned int needed = le16_to_cpu(cl->cl_cpg) -
511 			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
512 	u32 p_cpos, clusters;
513 	u64 p_blkno;
514 	struct ocfs2_extent_list *el = &bg->bg_list;
515 
516 	status = ocfs2_journal_access_gd(handle,
517 					 INODE_CACHE(alloc_inode),
518 					 bg_bh,
519 					 OCFS2_JOURNAL_ACCESS_CREATE);
520 	if (status < 0) {
521 		mlog_errno(status);
522 		goto bail;
523 	}
524 
525 	while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
526 				le16_to_cpu(el->l_count))) {
527 		if (min_bits > needed)
528 			min_bits = needed;
529 		status = ocfs2_block_group_claim_bits(osb, handle, ac,
530 						      min_bits, &p_cpos,
531 						      &clusters);
532 		if (status < 0) {
533 			if (status != -ENOSPC)
534 				mlog_errno(status);
535 			goto bail;
536 		}
537 		p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
538 		ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
539 					      clusters);
540 
541 		min_bits = clusters;
542 		needed = le16_to_cpu(cl->cl_cpg) -
543 			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
544 	}
545 
546 	if (needed > 0) {
547 		/*
548 		 * We have used up all the extent rec but can't fill up
549 		 * the cpg. So bail out.
550 		 */
551 		status = -ENOSPC;
552 		goto bail;
553 	}
554 
555 	ocfs2_journal_dirty(handle, bg_bh);
556 
557 bail:
558 	return status;
559 }
560 
561 static void ocfs2_bg_alloc_cleanup(handle_t *handle,
562 				   struct ocfs2_alloc_context *cluster_ac,
563 				   struct inode *alloc_inode,
564 				   struct buffer_head *bg_bh)
565 {
566 	int i, ret;
567 	struct ocfs2_group_desc *bg;
568 	struct ocfs2_extent_list *el;
569 	struct ocfs2_extent_rec *rec;
570 
571 	if (!bg_bh)
572 		return;
573 
574 	bg = (struct ocfs2_group_desc *)bg_bh->b_data;
575 	el = &bg->bg_list;
576 	for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
577 		rec = &el->l_recs[i];
578 		ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
579 					  cluster_ac->ac_bh,
580 					  le64_to_cpu(rec->e_blkno),
581 					  le16_to_cpu(rec->e_leaf_clusters));
582 		if (ret)
583 			mlog_errno(ret);
584 		/* Try all the clusters to free */
585 	}
586 
587 	ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
588 	brelse(bg_bh);
589 }
590 
591 static struct buffer_head *
592 ocfs2_block_group_alloc_discontig(handle_t *handle,
593 				  struct inode *alloc_inode,
594 				  struct ocfs2_alloc_context *ac,
595 				  struct ocfs2_chain_list *cl)
596 {
597 	int status;
598 	u32 bit_off, num_bits;
599 	u64 bg_blkno;
600 	unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
601 	struct buffer_head *bg_bh = NULL;
602 	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
603 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
604 
605 	if (!ocfs2_supports_discontig_bg(osb)) {
606 		status = -ENOSPC;
607 		goto bail;
608 	}
609 
610 	status = ocfs2_extend_trans(handle,
611 				    ocfs2_calc_bg_discontig_credits(osb->sb));
612 	if (status) {
613 		mlog_errno(status);
614 		goto bail;
615 	}
616 
617 	/*
618 	 * We're going to be grabbing from multiple cluster groups.
619 	 * We don't have enough credits to relink them all, and the
620 	 * cluster groups will be staying in cache for the duration of
621 	 * this operation.
622 	 */
623 	ac->ac_disable_chain_relink = 1;
624 
625 	/* Claim the first region */
626 	status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
627 					      &bit_off, &num_bits);
628 	if (status < 0) {
629 		if (status != -ENOSPC)
630 			mlog_errno(status);
631 		goto bail;
632 	}
633 	min_bits = num_bits;
634 
635 	/* setup the group */
636 	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
637 	trace_ocfs2_block_group_alloc_discontig(
638 				(unsigned long long)bg_blkno, alloc_rec);
639 
640 	bg_bh = sb_getblk(osb->sb, bg_blkno);
641 	if (!bg_bh) {
642 		status = -ENOMEM;
643 		mlog_errno(status);
644 		goto bail;
645 	}
646 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
647 
648 	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
649 					bg_blkno, num_bits, alloc_rec, cl);
650 	if (status < 0) {
651 		mlog_errno(status);
652 		goto bail;
653 	}
654 
655 	status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
656 						  bg_bh, ac, cl, min_bits);
657 	if (status)
658 		mlog_errno(status);
659 
660 bail:
661 	if (status)
662 		ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
663 	return status ? ERR_PTR(status) : bg_bh;
664 }
665 
666 /*
667  * We expect the block group allocator to already be locked.
668  */
669 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
670 				   struct inode *alloc_inode,
671 				   struct buffer_head *bh,
672 				   u64 max_block,
673 				   u64 *last_alloc_group,
674 				   int flags)
675 {
676 	int status, credits;
677 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
678 	struct ocfs2_chain_list *cl;
679 	struct ocfs2_alloc_context *ac = NULL;
680 	handle_t *handle = NULL;
681 	u16 alloc_rec;
682 	struct buffer_head *bg_bh = NULL;
683 	struct ocfs2_group_desc *bg;
684 
685 	BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
686 
687 	cl = &fe->id2.i_chain;
688 	status = ocfs2_reserve_clusters_with_limit(osb,
689 						   le16_to_cpu(cl->cl_cpg),
690 						   max_block, flags, &ac);
691 	if (status < 0) {
692 		if (status != -ENOSPC)
693 			mlog_errno(status);
694 		goto bail;
695 	}
696 
697 	credits = ocfs2_calc_group_alloc_credits(osb->sb,
698 						 le16_to_cpu(cl->cl_cpg));
699 	handle = ocfs2_start_trans(osb, credits);
700 	if (IS_ERR(handle)) {
701 		status = PTR_ERR(handle);
702 		handle = NULL;
703 		mlog_errno(status);
704 		goto bail;
705 	}
706 
707 	if (last_alloc_group && *last_alloc_group != 0) {
708 		trace_ocfs2_block_group_alloc(
709 				(unsigned long long)*last_alloc_group);
710 		ac->ac_last_group = *last_alloc_group;
711 	}
712 
713 	bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
714 					       ac, cl);
715 	if (IS_ERR(bg_bh) && (PTR_ERR(bg_bh) == -ENOSPC))
716 		bg_bh = ocfs2_block_group_alloc_discontig(handle,
717 							  alloc_inode,
718 							  ac, cl);
719 	if (IS_ERR(bg_bh)) {
720 		status = PTR_ERR(bg_bh);
721 		bg_bh = NULL;
722 		if (status != -ENOSPC)
723 			mlog_errno(status);
724 		goto bail;
725 	}
726 	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
727 
728 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
729 					 bh, OCFS2_JOURNAL_ACCESS_WRITE);
730 	if (status < 0) {
731 		mlog_errno(status);
732 		goto bail;
733 	}
734 
735 	alloc_rec = le16_to_cpu(bg->bg_chain);
736 	le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
737 		     le16_to_cpu(bg->bg_free_bits_count));
738 	le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
739 		     le16_to_cpu(bg->bg_bits));
740 	cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
741 	if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
742 		le16_add_cpu(&cl->cl_next_free_rec, 1);
743 
744 	le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
745 					le16_to_cpu(bg->bg_free_bits_count));
746 	le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
747 	le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
748 
749 	ocfs2_journal_dirty(handle, bh);
750 
751 	spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
752 	OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
753 	fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
754 					     le32_to_cpu(fe->i_clusters)));
755 	spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
756 	i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
757 	alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
758 	ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
759 
760 	status = 0;
761 
762 	/* save the new last alloc group so that the caller can cache it. */
763 	if (last_alloc_group)
764 		*last_alloc_group = ac->ac_last_group;
765 
766 bail:
767 	if (handle)
768 		ocfs2_commit_trans(osb, handle);
769 
770 	if (ac)
771 		ocfs2_free_alloc_context(ac);
772 
773 	brelse(bg_bh);
774 
775 	if (status)
776 		mlog_errno(status);
777 	return status;
778 }
779 
780 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
781 				       struct ocfs2_alloc_context *ac,
782 				       int type,
783 				       u32 slot,
784 				       u64 *last_alloc_group,
785 				       int flags)
786 {
787 	int status;
788 	u32 bits_wanted = ac->ac_bits_wanted;
789 	struct inode *alloc_inode;
790 	struct buffer_head *bh = NULL;
791 	struct ocfs2_dinode *fe;
792 	u32 free_bits;
793 
794 	alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
795 	if (!alloc_inode) {
796 		mlog_errno(-EINVAL);
797 		return -EINVAL;
798 	}
799 
800 	inode_lock(alloc_inode);
801 
802 	status = ocfs2_inode_lock(alloc_inode, &bh, 1);
803 	if (status < 0) {
804 		inode_unlock(alloc_inode);
805 		iput(alloc_inode);
806 
807 		mlog_errno(status);
808 		return status;
809 	}
810 
811 	ac->ac_inode = alloc_inode;
812 	ac->ac_alloc_slot = slot;
813 
814 	fe = (struct ocfs2_dinode *) bh->b_data;
815 
816 	/* The bh was validated by the inode read inside
817 	 * ocfs2_inode_lock().  Any corruption is a code bug. */
818 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
819 
820 	if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
821 		status = ocfs2_error(alloc_inode->i_sb,
822 				     "Invalid chain allocator %llu\n",
823 				     (unsigned long long)le64_to_cpu(fe->i_blkno));
824 		goto bail;
825 	}
826 
827 	free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
828 		le32_to_cpu(fe->id1.bitmap1.i_used);
829 
830 	if (bits_wanted > free_bits) {
831 		/* cluster bitmap never grows */
832 		if (ocfs2_is_cluster_bitmap(alloc_inode)) {
833 			trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
834 								free_bits);
835 			status = -ENOSPC;
836 			goto bail;
837 		}
838 
839 		if (!(flags & ALLOC_NEW_GROUP)) {
840 			trace_ocfs2_reserve_suballoc_bits_no_new_group(
841 						slot, bits_wanted, free_bits);
842 			status = -ENOSPC;
843 			goto bail;
844 		}
845 
846 		status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
847 						 ac->ac_max_block,
848 						 last_alloc_group, flags);
849 		if (status < 0) {
850 			if (status != -ENOSPC)
851 				mlog_errno(status);
852 			goto bail;
853 		}
854 		atomic_inc(&osb->alloc_stats.bg_extends);
855 
856 		/* You should never ask for this much metadata */
857 		BUG_ON(bits_wanted >
858 		       (le32_to_cpu(fe->id1.bitmap1.i_total)
859 			- le32_to_cpu(fe->id1.bitmap1.i_used)));
860 	}
861 
862 	get_bh(bh);
863 	ac->ac_bh = bh;
864 bail:
865 	brelse(bh);
866 
867 	if (status)
868 		mlog_errno(status);
869 	return status;
870 }
871 
872 static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
873 {
874 	spin_lock(&osb->osb_lock);
875 	osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
876 	spin_unlock(&osb->osb_lock);
877 	atomic_set(&osb->s_num_inodes_stolen, 0);
878 }
879 
880 static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
881 {
882 	spin_lock(&osb->osb_lock);
883 	osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
884 	spin_unlock(&osb->osb_lock);
885 	atomic_set(&osb->s_num_meta_stolen, 0);
886 }
887 
888 void ocfs2_init_steal_slots(struct ocfs2_super *osb)
889 {
890 	ocfs2_init_inode_steal_slot(osb);
891 	ocfs2_init_meta_steal_slot(osb);
892 }
893 
894 static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
895 {
896 	spin_lock(&osb->osb_lock);
897 	if (type == INODE_ALLOC_SYSTEM_INODE)
898 		osb->s_inode_steal_slot = slot;
899 	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
900 		osb->s_meta_steal_slot = slot;
901 	spin_unlock(&osb->osb_lock);
902 }
903 
904 static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
905 {
906 	int slot = OCFS2_INVALID_SLOT;
907 
908 	spin_lock(&osb->osb_lock);
909 	if (type == INODE_ALLOC_SYSTEM_INODE)
910 		slot = osb->s_inode_steal_slot;
911 	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
912 		slot = osb->s_meta_steal_slot;
913 	spin_unlock(&osb->osb_lock);
914 
915 	return slot;
916 }
917 
918 static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
919 {
920 	return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
921 }
922 
923 static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
924 {
925 	return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
926 }
927 
928 static int ocfs2_steal_resource(struct ocfs2_super *osb,
929 				struct ocfs2_alloc_context *ac,
930 				int type)
931 {
932 	int i, status = -ENOSPC;
933 	int slot = __ocfs2_get_steal_slot(osb, type);
934 
935 	/* Start to steal resource from the first slot after ours. */
936 	if (slot == OCFS2_INVALID_SLOT)
937 		slot = osb->slot_num + 1;
938 
939 	for (i = 0; i < osb->max_slots; i++, slot++) {
940 		if (slot == osb->max_slots)
941 			slot = 0;
942 
943 		if (slot == osb->slot_num)
944 			continue;
945 
946 		status = ocfs2_reserve_suballoc_bits(osb, ac,
947 						     type,
948 						     (u32)slot, NULL,
949 						     NOT_ALLOC_NEW_GROUP);
950 		if (status >= 0) {
951 			__ocfs2_set_steal_slot(osb, slot, type);
952 			break;
953 		}
954 
955 		ocfs2_free_ac_resource(ac);
956 	}
957 
958 	return status;
959 }
960 
961 static int ocfs2_steal_inode(struct ocfs2_super *osb,
962 			     struct ocfs2_alloc_context *ac)
963 {
964 	return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
965 }
966 
967 static int ocfs2_steal_meta(struct ocfs2_super *osb,
968 			    struct ocfs2_alloc_context *ac)
969 {
970 	return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
971 }
972 
973 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
974 				      int blocks,
975 				      struct ocfs2_alloc_context **ac)
976 {
977 	int status;
978 	int slot = ocfs2_get_meta_steal_slot(osb);
979 
980 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
981 	if (!(*ac)) {
982 		status = -ENOMEM;
983 		mlog_errno(status);
984 		goto bail;
985 	}
986 
987 	(*ac)->ac_bits_wanted = blocks;
988 	(*ac)->ac_which = OCFS2_AC_USE_META;
989 	(*ac)->ac_group_search = ocfs2_block_group_search;
990 
991 	if (slot != OCFS2_INVALID_SLOT &&
992 		atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
993 		goto extent_steal;
994 
995 	atomic_set(&osb->s_num_meta_stolen, 0);
996 	status = ocfs2_reserve_suballoc_bits(osb, (*ac),
997 					     EXTENT_ALLOC_SYSTEM_INODE,
998 					     (u32)osb->slot_num, NULL,
999 					     ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
1000 
1001 
1002 	if (status >= 0) {
1003 		status = 0;
1004 		if (slot != OCFS2_INVALID_SLOT)
1005 			ocfs2_init_meta_steal_slot(osb);
1006 		goto bail;
1007 	} else if (status < 0 && status != -ENOSPC) {
1008 		mlog_errno(status);
1009 		goto bail;
1010 	}
1011 
1012 	ocfs2_free_ac_resource(*ac);
1013 
1014 extent_steal:
1015 	status = ocfs2_steal_meta(osb, *ac);
1016 	atomic_inc(&osb->s_num_meta_stolen);
1017 	if (status < 0) {
1018 		if (status != -ENOSPC)
1019 			mlog_errno(status);
1020 		goto bail;
1021 	}
1022 
1023 	status = 0;
1024 bail:
1025 	if ((status < 0) && *ac) {
1026 		ocfs2_free_alloc_context(*ac);
1027 		*ac = NULL;
1028 	}
1029 
1030 	if (status)
1031 		mlog_errno(status);
1032 	return status;
1033 }
1034 
1035 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1036 			       struct ocfs2_extent_list *root_el,
1037 			       struct ocfs2_alloc_context **ac)
1038 {
1039 	return ocfs2_reserve_new_metadata_blocks(osb,
1040 					ocfs2_extend_meta_needed(root_el),
1041 					ac);
1042 }
1043 
1044 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
1045 			    struct ocfs2_alloc_context **ac)
1046 {
1047 	int status;
1048 	int slot = ocfs2_get_inode_steal_slot(osb);
1049 	u64 alloc_group;
1050 
1051 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1052 	if (!(*ac)) {
1053 		status = -ENOMEM;
1054 		mlog_errno(status);
1055 		goto bail;
1056 	}
1057 
1058 	(*ac)->ac_bits_wanted = 1;
1059 	(*ac)->ac_which = OCFS2_AC_USE_INODE;
1060 
1061 	(*ac)->ac_group_search = ocfs2_block_group_search;
1062 
1063 	/*
1064 	 * stat(2) can't handle i_ino > 32bits, so we tell the
1065 	 * lower levels not to allocate us a block group past that
1066 	 * limit.  The 'inode64' mount option avoids this behavior.
1067 	 */
1068 	if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1069 		(*ac)->ac_max_block = (u32)~0U;
1070 
1071 	/*
1072 	 * slot is set when we successfully steal inode from other nodes.
1073 	 * It is reset in 3 places:
1074 	 * 1. when we flush the truncate log
1075 	 * 2. when we complete local alloc recovery.
1076 	 * 3. when we successfully allocate from our own slot.
1077 	 * After it is set, we will go on stealing inodes until we find the
1078 	 * need to check our slots to see whether there is some space for us.
1079 	 */
1080 	if (slot != OCFS2_INVALID_SLOT &&
1081 	    atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
1082 		goto inode_steal;
1083 
1084 	atomic_set(&osb->s_num_inodes_stolen, 0);
1085 	alloc_group = osb->osb_inode_alloc_group;
1086 	status = ocfs2_reserve_suballoc_bits(osb, *ac,
1087 					     INODE_ALLOC_SYSTEM_INODE,
1088 					     (u32)osb->slot_num,
1089 					     &alloc_group,
1090 					     ALLOC_NEW_GROUP |
1091 					     ALLOC_GROUPS_FROM_GLOBAL);
1092 	if (status >= 0) {
1093 		status = 0;
1094 
1095 		spin_lock(&osb->osb_lock);
1096 		osb->osb_inode_alloc_group = alloc_group;
1097 		spin_unlock(&osb->osb_lock);
1098 		trace_ocfs2_reserve_new_inode_new_group(
1099 			(unsigned long long)alloc_group);
1100 
1101 		/*
1102 		 * Some inodes must be freed by us, so try to allocate
1103 		 * from our own next time.
1104 		 */
1105 		if (slot != OCFS2_INVALID_SLOT)
1106 			ocfs2_init_inode_steal_slot(osb);
1107 		goto bail;
1108 	} else if (status < 0 && status != -ENOSPC) {
1109 		mlog_errno(status);
1110 		goto bail;
1111 	}
1112 
1113 	ocfs2_free_ac_resource(*ac);
1114 
1115 inode_steal:
1116 	status = ocfs2_steal_inode(osb, *ac);
1117 	atomic_inc(&osb->s_num_inodes_stolen);
1118 	if (status < 0) {
1119 		if (status != -ENOSPC)
1120 			mlog_errno(status);
1121 		goto bail;
1122 	}
1123 
1124 	status = 0;
1125 bail:
1126 	if ((status < 0) && *ac) {
1127 		ocfs2_free_alloc_context(*ac);
1128 		*ac = NULL;
1129 	}
1130 
1131 	if (status)
1132 		mlog_errno(status);
1133 	return status;
1134 }
1135 
1136 /* local alloc code has to do the same thing, so rather than do this
1137  * twice.. */
1138 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1139 				      struct ocfs2_alloc_context *ac)
1140 {
1141 	int status;
1142 
1143 	ac->ac_which = OCFS2_AC_USE_MAIN;
1144 	ac->ac_group_search = ocfs2_cluster_group_search;
1145 
1146 	status = ocfs2_reserve_suballoc_bits(osb, ac,
1147 					     GLOBAL_BITMAP_SYSTEM_INODE,
1148 					     OCFS2_INVALID_SLOT, NULL,
1149 					     ALLOC_NEW_GROUP);
1150 	if (status < 0 && status != -ENOSPC) {
1151 		mlog_errno(status);
1152 		goto bail;
1153 	}
1154 
1155 bail:
1156 	return status;
1157 }
1158 
1159 /* Callers don't need to care which bitmap (local alloc or main) to
1160  * use so we figure it out for them, but unfortunately this clutters
1161  * things a bit. */
1162 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1163 					     u32 bits_wanted, u64 max_block,
1164 					     int flags,
1165 					     struct ocfs2_alloc_context **ac)
1166 {
1167 	int status, ret = 0;
1168 	int retried = 0;
1169 
1170 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1171 	if (!(*ac)) {
1172 		status = -ENOMEM;
1173 		mlog_errno(status);
1174 		goto bail;
1175 	}
1176 
1177 	(*ac)->ac_bits_wanted = bits_wanted;
1178 	(*ac)->ac_max_block = max_block;
1179 
1180 	status = -ENOSPC;
1181 	if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1182 	    ocfs2_alloc_should_use_local(osb, bits_wanted)) {
1183 		status = ocfs2_reserve_local_alloc_bits(osb,
1184 							bits_wanted,
1185 							*ac);
1186 		if ((status < 0) && (status != -ENOSPC)) {
1187 			mlog_errno(status);
1188 			goto bail;
1189 		}
1190 	}
1191 
1192 	if (status == -ENOSPC) {
1193 retry:
1194 		status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1195 		/* Retry if there is sufficient space cached in truncate log */
1196 		if (status == -ENOSPC && !retried) {
1197 			retried = 1;
1198 			ocfs2_inode_unlock((*ac)->ac_inode, 1);
1199 			inode_unlock((*ac)->ac_inode);
1200 
1201 			ret = ocfs2_try_to_free_truncate_log(osb, bits_wanted);
1202 			if (ret == 1)
1203 				goto retry;
1204 
1205 			if (ret < 0)
1206 				mlog_errno(ret);
1207 
1208 			inode_lock((*ac)->ac_inode);
1209 			ocfs2_inode_lock((*ac)->ac_inode, NULL, 1);
1210 		}
1211 		if (status < 0) {
1212 			if (status != -ENOSPC)
1213 				mlog_errno(status);
1214 			goto bail;
1215 		}
1216 	}
1217 
1218 	status = 0;
1219 bail:
1220 	if ((status < 0) && *ac) {
1221 		ocfs2_free_alloc_context(*ac);
1222 		*ac = NULL;
1223 	}
1224 
1225 	if (status)
1226 		mlog_errno(status);
1227 	return status;
1228 }
1229 
1230 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1231 			   u32 bits_wanted,
1232 			   struct ocfs2_alloc_context **ac)
1233 {
1234 	return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1235 						 ALLOC_NEW_GROUP, ac);
1236 }
1237 
1238 /*
1239  * More or less lifted from ext3. I'll leave their description below:
1240  *
1241  * "For ext3 allocations, we must not reuse any blocks which are
1242  * allocated in the bitmap buffer's "last committed data" copy.  This
1243  * prevents deletes from freeing up the page for reuse until we have
1244  * committed the delete transaction.
1245  *
1246  * If we didn't do this, then deleting something and reallocating it as
1247  * data would allow the old block to be overwritten before the
1248  * transaction committed (because we force data to disk before commit).
1249  * This would lead to corruption if we crashed between overwriting the
1250  * data and committing the delete.
1251  *
1252  * @@@ We may want to make this allocation behaviour conditional on
1253  * data-writes at some point, and disable it for metadata allocations or
1254  * sync-data inodes."
1255  *
1256  * Note: OCFS2 already does this differently for metadata vs data
1257  * allocations, as those bitmaps are separate and undo access is never
1258  * called on a metadata group descriptor.
1259  */
1260 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1261 					 int nr)
1262 {
1263 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1264 	int ret;
1265 
1266 	if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1267 		return 0;
1268 
1269 	if (!buffer_jbd(bg_bh))
1270 		return 1;
1271 
1272 	jbd_lock_bh_state(bg_bh);
1273 	bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
1274 	if (bg)
1275 		ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1276 	else
1277 		ret = 1;
1278 	jbd_unlock_bh_state(bg_bh);
1279 
1280 	return ret;
1281 }
1282 
1283 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1284 					     struct buffer_head *bg_bh,
1285 					     unsigned int bits_wanted,
1286 					     unsigned int total_bits,
1287 					     struct ocfs2_suballoc_result *res)
1288 {
1289 	void *bitmap;
1290 	u16 best_offset, best_size;
1291 	int offset, start, found, status = 0;
1292 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1293 
1294 	/* Callers got this descriptor from
1295 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1296 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1297 
1298 	found = start = best_offset = best_size = 0;
1299 	bitmap = bg->bg_bitmap;
1300 
1301 	while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1302 		if (offset == total_bits)
1303 			break;
1304 
1305 		if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1306 			/* We found a zero, but we can't use it as it
1307 			 * hasn't been put to disk yet! */
1308 			found = 0;
1309 			start = offset + 1;
1310 		} else if (offset == start) {
1311 			/* we found a zero */
1312 			found++;
1313 			/* move start to the next bit to test */
1314 			start++;
1315 		} else {
1316 			/* got a zero after some ones */
1317 			found = 1;
1318 			start = offset + 1;
1319 		}
1320 		if (found > best_size) {
1321 			best_size = found;
1322 			best_offset = start - found;
1323 		}
1324 		/* we got everything we needed */
1325 		if (found == bits_wanted) {
1326 			/* mlog(0, "Found it all!\n"); */
1327 			break;
1328 		}
1329 	}
1330 
1331 	if (best_size) {
1332 		res->sr_bit_offset = best_offset;
1333 		res->sr_bits = best_size;
1334 	} else {
1335 		status = -ENOSPC;
1336 		/* No error log here -- see the comment above
1337 		 * ocfs2_test_bg_bit_allocatable */
1338 	}
1339 
1340 	return status;
1341 }
1342 
1343 int ocfs2_block_group_set_bits(handle_t *handle,
1344 					     struct inode *alloc_inode,
1345 					     struct ocfs2_group_desc *bg,
1346 					     struct buffer_head *group_bh,
1347 					     unsigned int bit_off,
1348 					     unsigned int num_bits)
1349 {
1350 	int status;
1351 	void *bitmap = bg->bg_bitmap;
1352 	int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1353 
1354 	/* All callers get the descriptor via
1355 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1356 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1357 	BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1358 
1359 	trace_ocfs2_block_group_set_bits(bit_off, num_bits);
1360 
1361 	if (ocfs2_is_cluster_bitmap(alloc_inode))
1362 		journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1363 
1364 	status = ocfs2_journal_access_gd(handle,
1365 					 INODE_CACHE(alloc_inode),
1366 					 group_bh,
1367 					 journal_type);
1368 	if (status < 0) {
1369 		mlog_errno(status);
1370 		goto bail;
1371 	}
1372 
1373 	le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1374 	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1375 		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
1376 				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
1377 				   le16_to_cpu(bg->bg_bits),
1378 				   le16_to_cpu(bg->bg_free_bits_count),
1379 				   num_bits);
1380 	}
1381 	while(num_bits--)
1382 		ocfs2_set_bit(bit_off++, bitmap);
1383 
1384 	ocfs2_journal_dirty(handle, group_bh);
1385 
1386 bail:
1387 	return status;
1388 }
1389 
1390 /* find the one with the most empty bits */
1391 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1392 {
1393 	u16 curr, best;
1394 
1395 	BUG_ON(!cl->cl_next_free_rec);
1396 
1397 	best = curr = 0;
1398 	while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1399 		if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1400 		    le32_to_cpu(cl->cl_recs[best].c_free))
1401 			best = curr;
1402 		curr++;
1403 	}
1404 
1405 	BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1406 	return best;
1407 }
1408 
1409 static int ocfs2_relink_block_group(handle_t *handle,
1410 				    struct inode *alloc_inode,
1411 				    struct buffer_head *fe_bh,
1412 				    struct buffer_head *bg_bh,
1413 				    struct buffer_head *prev_bg_bh,
1414 				    u16 chain)
1415 {
1416 	int status;
1417 	/* there is a really tiny chance the journal calls could fail,
1418 	 * but we wouldn't want inconsistent blocks in *any* case. */
1419 	u64 bg_ptr, prev_bg_ptr;
1420 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1421 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1422 	struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1423 
1424 	/* The caller got these descriptors from
1425 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1426 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1427 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1428 
1429 	trace_ocfs2_relink_block_group(
1430 		(unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1431 		(unsigned long long)le64_to_cpu(bg->bg_blkno),
1432 		(unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1433 
1434 	bg_ptr = le64_to_cpu(bg->bg_next_group);
1435 	prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1436 
1437 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1438 					 prev_bg_bh,
1439 					 OCFS2_JOURNAL_ACCESS_WRITE);
1440 	if (status < 0)
1441 		goto out;
1442 
1443 	prev_bg->bg_next_group = bg->bg_next_group;
1444 	ocfs2_journal_dirty(handle, prev_bg_bh);
1445 
1446 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1447 					 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1448 	if (status < 0)
1449 		goto out_rollback_prev_bg;
1450 
1451 	bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1452 	ocfs2_journal_dirty(handle, bg_bh);
1453 
1454 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1455 					 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1456 	if (status < 0)
1457 		goto out_rollback_bg;
1458 
1459 	fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1460 	ocfs2_journal_dirty(handle, fe_bh);
1461 
1462 out:
1463 	if (status < 0)
1464 		mlog_errno(status);
1465 	return status;
1466 
1467 out_rollback_bg:
1468 	bg->bg_next_group = cpu_to_le64(bg_ptr);
1469 out_rollback_prev_bg:
1470 	prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1471 	goto out;
1472 }
1473 
1474 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1475 						     u32 wanted)
1476 {
1477 	return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1478 }
1479 
1480 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1481  * value on error. */
1482 static int ocfs2_cluster_group_search(struct inode *inode,
1483 				      struct buffer_head *group_bh,
1484 				      u32 bits_wanted, u32 min_bits,
1485 				      u64 max_block,
1486 				      struct ocfs2_suballoc_result *res)
1487 {
1488 	int search = -ENOSPC;
1489 	int ret;
1490 	u64 blkoff;
1491 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1492 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1493 	unsigned int max_bits, gd_cluster_off;
1494 
1495 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1496 
1497 	if (gd->bg_free_bits_count) {
1498 		max_bits = le16_to_cpu(gd->bg_bits);
1499 
1500 		/* Tail groups in cluster bitmaps which aren't cpg
1501 		 * aligned are prone to partial extension by a failed
1502 		 * fs resize. If the file system resize never got to
1503 		 * update the dinode cluster count, then we don't want
1504 		 * to trust any clusters past it, regardless of what
1505 		 * the group descriptor says. */
1506 		gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1507 							  le64_to_cpu(gd->bg_blkno));
1508 		if ((gd_cluster_off + max_bits) >
1509 		    OCFS2_I(inode)->ip_clusters) {
1510 			max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1511 			trace_ocfs2_cluster_group_search_wrong_max_bits(
1512 				(unsigned long long)le64_to_cpu(gd->bg_blkno),
1513 				le16_to_cpu(gd->bg_bits),
1514 				OCFS2_I(inode)->ip_clusters, max_bits);
1515 		}
1516 
1517 		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1518 							group_bh, bits_wanted,
1519 							max_bits, res);
1520 		if (ret)
1521 			return ret;
1522 
1523 		if (max_block) {
1524 			blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1525 							  gd_cluster_off +
1526 							  res->sr_bit_offset +
1527 							  res->sr_bits);
1528 			trace_ocfs2_cluster_group_search_max_block(
1529 				(unsigned long long)blkoff,
1530 				(unsigned long long)max_block);
1531 			if (blkoff > max_block)
1532 				return -ENOSPC;
1533 		}
1534 
1535 		/* ocfs2_block_group_find_clear_bits() might
1536 		 * return success, but we still want to return
1537 		 * -ENOSPC unless it found the minimum number
1538 		 * of bits. */
1539 		if (min_bits <= res->sr_bits)
1540 			search = 0; /* success */
1541 		else if (res->sr_bits) {
1542 			/*
1543 			 * Don't show bits which we'll be returning
1544 			 * for allocation to the local alloc bitmap.
1545 			 */
1546 			ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
1547 		}
1548 	}
1549 
1550 	return search;
1551 }
1552 
1553 static int ocfs2_block_group_search(struct inode *inode,
1554 				    struct buffer_head *group_bh,
1555 				    u32 bits_wanted, u32 min_bits,
1556 				    u64 max_block,
1557 				    struct ocfs2_suballoc_result *res)
1558 {
1559 	int ret = -ENOSPC;
1560 	u64 blkoff;
1561 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1562 
1563 	BUG_ON(min_bits != 1);
1564 	BUG_ON(ocfs2_is_cluster_bitmap(inode));
1565 
1566 	if (bg->bg_free_bits_count) {
1567 		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1568 							group_bh, bits_wanted,
1569 							le16_to_cpu(bg->bg_bits),
1570 							res);
1571 		if (!ret && max_block) {
1572 			blkoff = le64_to_cpu(bg->bg_blkno) +
1573 				res->sr_bit_offset + res->sr_bits;
1574 			trace_ocfs2_block_group_search_max_block(
1575 				(unsigned long long)blkoff,
1576 				(unsigned long long)max_block);
1577 			if (blkoff > max_block)
1578 				ret = -ENOSPC;
1579 		}
1580 	}
1581 
1582 	return ret;
1583 }
1584 
1585 int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1586 				       handle_t *handle,
1587 				       struct buffer_head *di_bh,
1588 				       u32 num_bits,
1589 				       u16 chain)
1590 {
1591 	int ret;
1592 	u32 tmp_used;
1593 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1594 	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1595 
1596 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1597 				      OCFS2_JOURNAL_ACCESS_WRITE);
1598 	if (ret < 0) {
1599 		mlog_errno(ret);
1600 		goto out;
1601 	}
1602 
1603 	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1604 	di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1605 	le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1606 	ocfs2_journal_dirty(handle, di_bh);
1607 
1608 out:
1609 	return ret;
1610 }
1611 
1612 void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
1613 				       struct buffer_head *di_bh,
1614 				       u32 num_bits,
1615 				       u16 chain)
1616 {
1617 	u32 tmp_used;
1618 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1619 	struct ocfs2_chain_list *cl;
1620 
1621 	cl = (struct ocfs2_chain_list *)&di->id2.i_chain;
1622 	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1623 	di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits);
1624 	le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits);
1625 }
1626 
1627 static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1628 					 struct ocfs2_extent_rec *rec,
1629 					 struct ocfs2_chain_list *cl)
1630 {
1631 	unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1632 	unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1633 	unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1634 
1635 	if (res->sr_bit_offset < bitoff)
1636 		return 0;
1637 	if (res->sr_bit_offset >= (bitoff + bitcount))
1638 		return 0;
1639 	res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1640 		(res->sr_bit_offset - bitoff);
1641 	if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1642 		res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1643 	return 1;
1644 }
1645 
1646 static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1647 					  struct ocfs2_group_desc *bg,
1648 					  struct ocfs2_suballoc_result *res)
1649 {
1650 	int i;
1651 	u64 bg_blkno = res->sr_bg_blkno;  /* Save off */
1652 	struct ocfs2_extent_rec *rec;
1653 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1654 	struct ocfs2_chain_list *cl = &di->id2.i_chain;
1655 
1656 	if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1657 		res->sr_blkno = 0;
1658 		return;
1659 	}
1660 
1661 	res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1662 	res->sr_bg_blkno = 0;  /* Clear it for contig block groups */
1663 	if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1664 	    !bg->bg_list.l_next_free_rec)
1665 		return;
1666 
1667 	for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1668 		rec = &bg->bg_list.l_recs[i];
1669 		if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1670 			res->sr_bg_blkno = bg_blkno;  /* Restore */
1671 			break;
1672 		}
1673 	}
1674 }
1675 
1676 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1677 				  handle_t *handle,
1678 				  u32 bits_wanted,
1679 				  u32 min_bits,
1680 				  struct ocfs2_suballoc_result *res,
1681 				  u16 *bits_left)
1682 {
1683 	int ret;
1684 	struct buffer_head *group_bh = NULL;
1685 	struct ocfs2_group_desc *gd;
1686 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1687 	struct inode *alloc_inode = ac->ac_inode;
1688 
1689 	ret = ocfs2_read_group_descriptor(alloc_inode, di,
1690 					  res->sr_bg_blkno, &group_bh);
1691 	if (ret < 0) {
1692 		mlog_errno(ret);
1693 		return ret;
1694 	}
1695 
1696 	gd = (struct ocfs2_group_desc *) group_bh->b_data;
1697 	ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1698 				  ac->ac_max_block, res);
1699 	if (ret < 0) {
1700 		if (ret != -ENOSPC)
1701 			mlog_errno(ret);
1702 		goto out;
1703 	}
1704 
1705 	if (!ret)
1706 		ocfs2_bg_discontig_fix_result(ac, gd, res);
1707 
1708 	/*
1709 	 * sr_bg_blkno might have been changed by
1710 	 * ocfs2_bg_discontig_fix_result
1711 	 */
1712 	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1713 
1714 	if (ac->ac_find_loc_only)
1715 		goto out_loc_only;
1716 
1717 	ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1718 					       res->sr_bits,
1719 					       le16_to_cpu(gd->bg_chain));
1720 	if (ret < 0) {
1721 		mlog_errno(ret);
1722 		goto out;
1723 	}
1724 
1725 	ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1726 					 res->sr_bit_offset, res->sr_bits);
1727 	if (ret < 0) {
1728 		ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh,
1729 					       res->sr_bits,
1730 					       le16_to_cpu(gd->bg_chain));
1731 		mlog_errno(ret);
1732 	}
1733 
1734 out_loc_only:
1735 	*bits_left = le16_to_cpu(gd->bg_free_bits_count);
1736 
1737 out:
1738 	brelse(group_bh);
1739 
1740 	return ret;
1741 }
1742 
1743 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1744 			      handle_t *handle,
1745 			      u32 bits_wanted,
1746 			      u32 min_bits,
1747 			      struct ocfs2_suballoc_result *res,
1748 			      u16 *bits_left)
1749 {
1750 	int status;
1751 	u16 chain;
1752 	u64 next_group;
1753 	struct inode *alloc_inode = ac->ac_inode;
1754 	struct buffer_head *group_bh = NULL;
1755 	struct buffer_head *prev_group_bh = NULL;
1756 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1757 	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1758 	struct ocfs2_group_desc *bg;
1759 
1760 	chain = ac->ac_chain;
1761 	trace_ocfs2_search_chain_begin(
1762 		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1763 		bits_wanted, chain);
1764 
1765 	status = ocfs2_read_group_descriptor(alloc_inode, fe,
1766 					     le64_to_cpu(cl->cl_recs[chain].c_blkno),
1767 					     &group_bh);
1768 	if (status < 0) {
1769 		mlog_errno(status);
1770 		goto bail;
1771 	}
1772 	bg = (struct ocfs2_group_desc *) group_bh->b_data;
1773 
1774 	status = -ENOSPC;
1775 	/* for now, the chain search is a bit simplistic. We just use
1776 	 * the 1st group with any empty bits. */
1777 	while ((status = ac->ac_group_search(alloc_inode, group_bh,
1778 					     bits_wanted, min_bits,
1779 					     ac->ac_max_block,
1780 					     res)) == -ENOSPC) {
1781 		if (!bg->bg_next_group)
1782 			break;
1783 
1784 		brelse(prev_group_bh);
1785 		prev_group_bh = NULL;
1786 
1787 		next_group = le64_to_cpu(bg->bg_next_group);
1788 		prev_group_bh = group_bh;
1789 		group_bh = NULL;
1790 		status = ocfs2_read_group_descriptor(alloc_inode, fe,
1791 						     next_group, &group_bh);
1792 		if (status < 0) {
1793 			mlog_errno(status);
1794 			goto bail;
1795 		}
1796 		bg = (struct ocfs2_group_desc *) group_bh->b_data;
1797 	}
1798 	if (status < 0) {
1799 		if (status != -ENOSPC)
1800 			mlog_errno(status);
1801 		goto bail;
1802 	}
1803 
1804 	trace_ocfs2_search_chain_succ(
1805 		(unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1806 
1807 	res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1808 
1809 	BUG_ON(res->sr_bits == 0);
1810 	if (!status)
1811 		ocfs2_bg_discontig_fix_result(ac, bg, res);
1812 
1813 	/*
1814 	 * sr_bg_blkno might have been changed by
1815 	 * ocfs2_bg_discontig_fix_result
1816 	 */
1817 	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1818 
1819 	/*
1820 	 * Keep track of previous block descriptor read. When
1821 	 * we find a target, if we have read more than X
1822 	 * number of descriptors, and the target is reasonably
1823 	 * empty, relink him to top of his chain.
1824 	 *
1825 	 * We've read 0 extra blocks and only send one more to
1826 	 * the transaction, yet the next guy to search has a
1827 	 * much easier time.
1828 	 *
1829 	 * Do this *after* figuring out how many bits we're taking out
1830 	 * of our target group.
1831 	 */
1832 	if (!ac->ac_disable_chain_relink &&
1833 	    (prev_group_bh) &&
1834 	    (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1835 		status = ocfs2_relink_block_group(handle, alloc_inode,
1836 						  ac->ac_bh, group_bh,
1837 						  prev_group_bh, chain);
1838 		if (status < 0) {
1839 			mlog_errno(status);
1840 			goto bail;
1841 		}
1842 	}
1843 
1844 	if (ac->ac_find_loc_only)
1845 		goto out_loc_only;
1846 
1847 	status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1848 						  ac->ac_bh, res->sr_bits,
1849 						  chain);
1850 	if (status) {
1851 		mlog_errno(status);
1852 		goto bail;
1853 	}
1854 
1855 	status = ocfs2_block_group_set_bits(handle,
1856 					    alloc_inode,
1857 					    bg,
1858 					    group_bh,
1859 					    res->sr_bit_offset,
1860 					    res->sr_bits);
1861 	if (status < 0) {
1862 		ocfs2_rollback_alloc_dinode_counts(alloc_inode,
1863 					ac->ac_bh, res->sr_bits, chain);
1864 		mlog_errno(status);
1865 		goto bail;
1866 	}
1867 
1868 	trace_ocfs2_search_chain_end(
1869 			(unsigned long long)le64_to_cpu(fe->i_blkno),
1870 			res->sr_bits);
1871 
1872 out_loc_only:
1873 	*bits_left = le16_to_cpu(bg->bg_free_bits_count);
1874 bail:
1875 	brelse(group_bh);
1876 	brelse(prev_group_bh);
1877 
1878 	if (status)
1879 		mlog_errno(status);
1880 	return status;
1881 }
1882 
1883 /* will give out up to bits_wanted contiguous bits. */
1884 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1885 				     handle_t *handle,
1886 				     u32 bits_wanted,
1887 				     u32 min_bits,
1888 				     struct ocfs2_suballoc_result *res)
1889 {
1890 	int status;
1891 	u16 victim, i;
1892 	u16 bits_left = 0;
1893 	u64 hint = ac->ac_last_group;
1894 	struct ocfs2_chain_list *cl;
1895 	struct ocfs2_dinode *fe;
1896 
1897 	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1898 	BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1899 	BUG_ON(!ac->ac_bh);
1900 
1901 	fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1902 
1903 	/* The bh was validated by the inode read during
1904 	 * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1905 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1906 
1907 	if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1908 	    le32_to_cpu(fe->id1.bitmap1.i_total)) {
1909 		status = ocfs2_error(ac->ac_inode->i_sb,
1910 				     "Chain allocator dinode %llu has %u used bits but only %u total\n",
1911 				     (unsigned long long)le64_to_cpu(fe->i_blkno),
1912 				     le32_to_cpu(fe->id1.bitmap1.i_used),
1913 				     le32_to_cpu(fe->id1.bitmap1.i_total));
1914 		goto bail;
1915 	}
1916 
1917 	res->sr_bg_blkno = hint;
1918 	if (res->sr_bg_blkno) {
1919 		/* Attempt to short-circuit the usual search mechanism
1920 		 * by jumping straight to the most recently used
1921 		 * allocation group. This helps us maintain some
1922 		 * contiguousness across allocations. */
1923 		status = ocfs2_search_one_group(ac, handle, bits_wanted,
1924 						min_bits, res, &bits_left);
1925 		if (!status)
1926 			goto set_hint;
1927 		if (status < 0 && status != -ENOSPC) {
1928 			mlog_errno(status);
1929 			goto bail;
1930 		}
1931 	}
1932 
1933 	cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1934 
1935 	victim = ocfs2_find_victim_chain(cl);
1936 	ac->ac_chain = victim;
1937 
1938 	status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1939 				    res, &bits_left);
1940 	if (!status) {
1941 		if (ocfs2_is_cluster_bitmap(ac->ac_inode))
1942 			hint = res->sr_bg_blkno;
1943 		else
1944 			hint = ocfs2_group_from_res(res);
1945 		goto set_hint;
1946 	}
1947 	if (status < 0 && status != -ENOSPC) {
1948 		mlog_errno(status);
1949 		goto bail;
1950 	}
1951 
1952 	trace_ocfs2_claim_suballoc_bits(victim);
1953 
1954 	/* If we didn't pick a good victim, then just default to
1955 	 * searching each chain in order. Don't allow chain relinking
1956 	 * because we only calculate enough journal credits for one
1957 	 * relink per alloc. */
1958 	ac->ac_disable_chain_relink = 1;
1959 	for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1960 		if (i == victim)
1961 			continue;
1962 		if (!cl->cl_recs[i].c_free)
1963 			continue;
1964 
1965 		ac->ac_chain = i;
1966 		status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1967 					    res, &bits_left);
1968 		if (!status) {
1969 			hint = ocfs2_group_from_res(res);
1970 			break;
1971 		}
1972 		if (status < 0 && status != -ENOSPC) {
1973 			mlog_errno(status);
1974 			goto bail;
1975 		}
1976 	}
1977 
1978 set_hint:
1979 	if (status != -ENOSPC) {
1980 		/* If the next search of this group is not likely to
1981 		 * yield a suitable extent, then we reset the last
1982 		 * group hint so as to not waste a disk read */
1983 		if (bits_left < min_bits)
1984 			ac->ac_last_group = 0;
1985 		else
1986 			ac->ac_last_group = hint;
1987 	}
1988 
1989 bail:
1990 	if (status)
1991 		mlog_errno(status);
1992 	return status;
1993 }
1994 
1995 int ocfs2_claim_metadata(handle_t *handle,
1996 			 struct ocfs2_alloc_context *ac,
1997 			 u32 bits_wanted,
1998 			 u64 *suballoc_loc,
1999 			 u16 *suballoc_bit_start,
2000 			 unsigned int *num_bits,
2001 			 u64 *blkno_start)
2002 {
2003 	int status;
2004 	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2005 
2006 	BUG_ON(!ac);
2007 	BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
2008 	BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
2009 
2010 	status = ocfs2_claim_suballoc_bits(ac,
2011 					   handle,
2012 					   bits_wanted,
2013 					   1,
2014 					   &res);
2015 	if (status < 0) {
2016 		mlog_errno(status);
2017 		goto bail;
2018 	}
2019 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2020 
2021 	*suballoc_loc = res.sr_bg_blkno;
2022 	*suballoc_bit_start = res.sr_bit_offset;
2023 	*blkno_start = res.sr_blkno;
2024 	ac->ac_bits_given += res.sr_bits;
2025 	*num_bits = res.sr_bits;
2026 	status = 0;
2027 bail:
2028 	if (status)
2029 		mlog_errno(status);
2030 	return status;
2031 }
2032 
2033 static void ocfs2_init_inode_ac_group(struct inode *dir,
2034 				      struct buffer_head *parent_di_bh,
2035 				      struct ocfs2_alloc_context *ac)
2036 {
2037 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2038 	/*
2039 	 * Try to allocate inodes from some specific group.
2040 	 *
2041 	 * If the parent dir has recorded the last group used in allocation,
2042 	 * cool, use it. Otherwise if we try to allocate new inode from the
2043 	 * same slot the parent dir belongs to, use the same chunk.
2044 	 *
2045 	 * We are very careful here to avoid the mistake of setting
2046 	 * ac_last_group to a group descriptor from a different (unlocked) slot.
2047 	 */
2048 	if (OCFS2_I(dir)->ip_last_used_group &&
2049 	    OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2050 		ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2051 	else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2052 		if (di->i_suballoc_loc)
2053 			ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2054 		else
2055 			ac->ac_last_group = ocfs2_which_suballoc_group(
2056 					le64_to_cpu(di->i_blkno),
2057 					le16_to_cpu(di->i_suballoc_bit));
2058 	}
2059 }
2060 
2061 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2062 					     struct ocfs2_alloc_context *ac)
2063 {
2064 	OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2065 	OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2066 }
2067 
2068 int ocfs2_find_new_inode_loc(struct inode *dir,
2069 			     struct buffer_head *parent_fe_bh,
2070 			     struct ocfs2_alloc_context *ac,
2071 			     u64 *fe_blkno)
2072 {
2073 	int ret;
2074 	handle_t *handle = NULL;
2075 	struct ocfs2_suballoc_result *res;
2076 
2077 	BUG_ON(!ac);
2078 	BUG_ON(ac->ac_bits_given != 0);
2079 	BUG_ON(ac->ac_bits_wanted != 1);
2080 	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2081 
2082 	res = kzalloc(sizeof(*res), GFP_NOFS);
2083 	if (res == NULL) {
2084 		ret = -ENOMEM;
2085 		mlog_errno(ret);
2086 		goto out;
2087 	}
2088 
2089 	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2090 
2091 	/*
2092 	 * The handle started here is for chain relink. Alternatively,
2093 	 * we could just disable relink for these calls.
2094 	 */
2095 	handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2096 	if (IS_ERR(handle)) {
2097 		ret = PTR_ERR(handle);
2098 		handle = NULL;
2099 		mlog_errno(ret);
2100 		goto out;
2101 	}
2102 
2103 	/*
2104 	 * This will instruct ocfs2_claim_suballoc_bits and
2105 	 * ocfs2_search_one_group to search but save actual allocation
2106 	 * for later.
2107 	 */
2108 	ac->ac_find_loc_only = 1;
2109 
2110 	ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2111 	if (ret < 0) {
2112 		mlog_errno(ret);
2113 		goto out;
2114 	}
2115 
2116 	ac->ac_find_loc_priv = res;
2117 	*fe_blkno = res->sr_blkno;
2118 	ocfs2_update_inode_fsync_trans(handle, dir, 0);
2119 out:
2120 	if (handle)
2121 		ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2122 
2123 	if (ret)
2124 		kfree(res);
2125 
2126 	return ret;
2127 }
2128 
2129 int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2130 				 struct inode *dir,
2131 				 struct ocfs2_alloc_context *ac,
2132 				 u64 *suballoc_loc,
2133 				 u16 *suballoc_bit,
2134 				 u64 di_blkno)
2135 {
2136 	int ret;
2137 	u16 chain;
2138 	struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2139 	struct buffer_head *bg_bh = NULL;
2140 	struct ocfs2_group_desc *bg;
2141 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2142 
2143 	/*
2144 	 * Since di_blkno is being passed back in, we check for any
2145 	 * inconsistencies which may have happened between
2146 	 * calls. These are code bugs as di_blkno is not expected to
2147 	 * change once returned from ocfs2_find_new_inode_loc()
2148 	 */
2149 	BUG_ON(res->sr_blkno != di_blkno);
2150 
2151 	ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2152 					  res->sr_bg_stable_blkno, &bg_bh);
2153 	if (ret) {
2154 		mlog_errno(ret);
2155 		goto out;
2156 	}
2157 
2158 	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2159 	chain = le16_to_cpu(bg->bg_chain);
2160 
2161 	ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2162 					       ac->ac_bh, res->sr_bits,
2163 					       chain);
2164 	if (ret) {
2165 		mlog_errno(ret);
2166 		goto out;
2167 	}
2168 
2169 	ret = ocfs2_block_group_set_bits(handle,
2170 					 ac->ac_inode,
2171 					 bg,
2172 					 bg_bh,
2173 					 res->sr_bit_offset,
2174 					 res->sr_bits);
2175 	if (ret < 0) {
2176 		ocfs2_rollback_alloc_dinode_counts(ac->ac_inode,
2177 					       ac->ac_bh, res->sr_bits, chain);
2178 		mlog_errno(ret);
2179 		goto out;
2180 	}
2181 
2182 	trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2183 					   res->sr_bits);
2184 
2185 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2186 
2187 	BUG_ON(res->sr_bits != 1);
2188 
2189 	*suballoc_loc = res->sr_bg_blkno;
2190 	*suballoc_bit = res->sr_bit_offset;
2191 	ac->ac_bits_given++;
2192 	ocfs2_save_inode_ac_group(dir, ac);
2193 
2194 out:
2195 	brelse(bg_bh);
2196 
2197 	return ret;
2198 }
2199 
2200 int ocfs2_claim_new_inode(handle_t *handle,
2201 			  struct inode *dir,
2202 			  struct buffer_head *parent_fe_bh,
2203 			  struct ocfs2_alloc_context *ac,
2204 			  u64 *suballoc_loc,
2205 			  u16 *suballoc_bit,
2206 			  u64 *fe_blkno)
2207 {
2208 	int status;
2209 	struct ocfs2_suballoc_result res;
2210 
2211 	BUG_ON(!ac);
2212 	BUG_ON(ac->ac_bits_given != 0);
2213 	BUG_ON(ac->ac_bits_wanted != 1);
2214 	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2215 
2216 	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2217 
2218 	status = ocfs2_claim_suballoc_bits(ac,
2219 					   handle,
2220 					   1,
2221 					   1,
2222 					   &res);
2223 	if (status < 0) {
2224 		mlog_errno(status);
2225 		goto bail;
2226 	}
2227 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2228 
2229 	BUG_ON(res.sr_bits != 1);
2230 
2231 	*suballoc_loc = res.sr_bg_blkno;
2232 	*suballoc_bit = res.sr_bit_offset;
2233 	*fe_blkno = res.sr_blkno;
2234 	ac->ac_bits_given++;
2235 	ocfs2_save_inode_ac_group(dir, ac);
2236 	status = 0;
2237 bail:
2238 	if (status)
2239 		mlog_errno(status);
2240 	return status;
2241 }
2242 
2243 /* translate a group desc. blkno and it's bitmap offset into
2244  * disk cluster offset. */
2245 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2246 						   u64 bg_blkno,
2247 						   u16 bg_bit_off)
2248 {
2249 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2250 	u32 cluster = 0;
2251 
2252 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2253 
2254 	if (bg_blkno != osb->first_cluster_group_blkno)
2255 		cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2256 	cluster += (u32) bg_bit_off;
2257 	return cluster;
2258 }
2259 
2260 /* given a cluster offset, calculate which block group it belongs to
2261  * and return that block offset. */
2262 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2263 {
2264 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2265 	u32 group_no;
2266 
2267 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2268 
2269 	group_no = cluster / osb->bitmap_cpg;
2270 	if (!group_no)
2271 		return osb->first_cluster_group_blkno;
2272 	return ocfs2_clusters_to_blocks(inode->i_sb,
2273 					group_no * osb->bitmap_cpg);
2274 }
2275 
2276 /* given the block number of a cluster start, calculate which cluster
2277  * group and descriptor bitmap offset that corresponds to. */
2278 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2279 						u64 data_blkno,
2280 						u64 *bg_blkno,
2281 						u16 *bg_bit_off)
2282 {
2283 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2284 	u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2285 
2286 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2287 
2288 	*bg_blkno = ocfs2_which_cluster_group(inode,
2289 					      data_cluster);
2290 
2291 	if (*bg_blkno == osb->first_cluster_group_blkno)
2292 		*bg_bit_off = (u16) data_cluster;
2293 	else
2294 		*bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2295 							     data_blkno - *bg_blkno);
2296 }
2297 
2298 /*
2299  * min_bits - minimum contiguous chunk from this total allocation we
2300  * can handle. set to what we asked for originally for a full
2301  * contig. allocation, set to '1' to indicate we can deal with extents
2302  * of any size.
2303  */
2304 int __ocfs2_claim_clusters(handle_t *handle,
2305 			   struct ocfs2_alloc_context *ac,
2306 			   u32 min_clusters,
2307 			   u32 max_clusters,
2308 			   u32 *cluster_start,
2309 			   u32 *num_clusters)
2310 {
2311 	int status;
2312 	unsigned int bits_wanted = max_clusters;
2313 	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2314 	struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2315 
2316 	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2317 
2318 	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2319 	       && ac->ac_which != OCFS2_AC_USE_MAIN);
2320 
2321 	if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2322 		WARN_ON(min_clusters > 1);
2323 
2324 		status = ocfs2_claim_local_alloc_bits(osb,
2325 						      handle,
2326 						      ac,
2327 						      bits_wanted,
2328 						      cluster_start,
2329 						      num_clusters);
2330 		if (!status)
2331 			atomic_inc(&osb->alloc_stats.local_data);
2332 	} else {
2333 		if (min_clusters > (osb->bitmap_cpg - 1)) {
2334 			/* The only paths asking for contiguousness
2335 			 * should know about this already. */
2336 			mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2337 			     "group bitmap size %u!\n", min_clusters,
2338 			     osb->bitmap_cpg);
2339 			status = -ENOSPC;
2340 			goto bail;
2341 		}
2342 		/* clamp the current request down to a realistic size. */
2343 		if (bits_wanted > (osb->bitmap_cpg - 1))
2344 			bits_wanted = osb->bitmap_cpg - 1;
2345 
2346 		status = ocfs2_claim_suballoc_bits(ac,
2347 						   handle,
2348 						   bits_wanted,
2349 						   min_clusters,
2350 						   &res);
2351 		if (!status) {
2352 			BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2353 			*cluster_start =
2354 				ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2355 								 res.sr_bg_blkno,
2356 								 res.sr_bit_offset);
2357 			atomic_inc(&osb->alloc_stats.bitmap_data);
2358 			*num_clusters = res.sr_bits;
2359 		}
2360 	}
2361 	if (status < 0) {
2362 		if (status != -ENOSPC)
2363 			mlog_errno(status);
2364 		goto bail;
2365 	}
2366 
2367 	ac->ac_bits_given += *num_clusters;
2368 
2369 bail:
2370 	if (status)
2371 		mlog_errno(status);
2372 	return status;
2373 }
2374 
2375 int ocfs2_claim_clusters(handle_t *handle,
2376 			 struct ocfs2_alloc_context *ac,
2377 			 u32 min_clusters,
2378 			 u32 *cluster_start,
2379 			 u32 *num_clusters)
2380 {
2381 	unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2382 
2383 	return __ocfs2_claim_clusters(handle, ac, min_clusters,
2384 				      bits_wanted, cluster_start, num_clusters);
2385 }
2386 
2387 static int ocfs2_block_group_clear_bits(handle_t *handle,
2388 					struct inode *alloc_inode,
2389 					struct ocfs2_group_desc *bg,
2390 					struct buffer_head *group_bh,
2391 					unsigned int bit_off,
2392 					unsigned int num_bits,
2393 					void (*undo_fn)(unsigned int bit,
2394 							unsigned long *bmap))
2395 {
2396 	int status;
2397 	unsigned int tmp;
2398 	struct ocfs2_group_desc *undo_bg = NULL;
2399 
2400 	/* The caller got this descriptor from
2401 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
2402 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2403 
2404 	trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2405 
2406 	BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2407 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2408 					 group_bh,
2409 					 undo_fn ?
2410 					 OCFS2_JOURNAL_ACCESS_UNDO :
2411 					 OCFS2_JOURNAL_ACCESS_WRITE);
2412 	if (status < 0) {
2413 		mlog_errno(status);
2414 		goto bail;
2415 	}
2416 
2417 	if (undo_fn) {
2418 		jbd_lock_bh_state(group_bh);
2419 		undo_bg = (struct ocfs2_group_desc *)
2420 					bh2jh(group_bh)->b_committed_data;
2421 		BUG_ON(!undo_bg);
2422 	}
2423 
2424 	tmp = num_bits;
2425 	while(tmp--) {
2426 		ocfs2_clear_bit((bit_off + tmp),
2427 				(unsigned long *) bg->bg_bitmap);
2428 		if (undo_fn)
2429 			undo_fn(bit_off + tmp,
2430 				(unsigned long *) undo_bg->bg_bitmap);
2431 	}
2432 	le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2433 	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2434 		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
2435 				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
2436 				   le16_to_cpu(bg->bg_bits),
2437 				   le16_to_cpu(bg->bg_free_bits_count),
2438 				   num_bits);
2439 	}
2440 
2441 	if (undo_fn)
2442 		jbd_unlock_bh_state(group_bh);
2443 
2444 	ocfs2_journal_dirty(handle, group_bh);
2445 bail:
2446 	return status;
2447 }
2448 
2449 /*
2450  * expects the suballoc inode to already be locked.
2451  */
2452 static int _ocfs2_free_suballoc_bits(handle_t *handle,
2453 				     struct inode *alloc_inode,
2454 				     struct buffer_head *alloc_bh,
2455 				     unsigned int start_bit,
2456 				     u64 bg_blkno,
2457 				     unsigned int count,
2458 				     void (*undo_fn)(unsigned int bit,
2459 						     unsigned long *bitmap))
2460 {
2461 	int status = 0;
2462 	u32 tmp_used;
2463 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2464 	struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2465 	struct buffer_head *group_bh = NULL;
2466 	struct ocfs2_group_desc *group;
2467 
2468 	/* The alloc_bh comes from ocfs2_free_dinode() or
2469 	 * ocfs2_free_clusters().  The callers have all locked the
2470 	 * allocator and gotten alloc_bh from the lock call.  This
2471 	 * validates the dinode buffer.  Any corruption that has happened
2472 	 * is a code bug. */
2473 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2474 	BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2475 
2476 	trace_ocfs2_free_suballoc_bits(
2477 		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2478 		(unsigned long long)bg_blkno,
2479 		start_bit, count);
2480 
2481 	status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2482 					     &group_bh);
2483 	if (status < 0) {
2484 		mlog_errno(status);
2485 		goto bail;
2486 	}
2487 	group = (struct ocfs2_group_desc *) group_bh->b_data;
2488 
2489 	BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2490 
2491 	status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2492 					      group, group_bh,
2493 					      start_bit, count, undo_fn);
2494 	if (status < 0) {
2495 		mlog_errno(status);
2496 		goto bail;
2497 	}
2498 
2499 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2500 					 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2501 	if (status < 0) {
2502 		mlog_errno(status);
2503 		ocfs2_block_group_set_bits(handle, alloc_inode, group, group_bh,
2504 				start_bit, count);
2505 		goto bail;
2506 	}
2507 
2508 	le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2509 		     count);
2510 	tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2511 	fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2512 	ocfs2_journal_dirty(handle, alloc_bh);
2513 
2514 bail:
2515 	brelse(group_bh);
2516 
2517 	if (status)
2518 		mlog_errno(status);
2519 	return status;
2520 }
2521 
2522 int ocfs2_free_suballoc_bits(handle_t *handle,
2523 			     struct inode *alloc_inode,
2524 			     struct buffer_head *alloc_bh,
2525 			     unsigned int start_bit,
2526 			     u64 bg_blkno,
2527 			     unsigned int count)
2528 {
2529 	return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2530 					 start_bit, bg_blkno, count, NULL);
2531 }
2532 
2533 int ocfs2_free_dinode(handle_t *handle,
2534 		      struct inode *inode_alloc_inode,
2535 		      struct buffer_head *inode_alloc_bh,
2536 		      struct ocfs2_dinode *di)
2537 {
2538 	u64 blk = le64_to_cpu(di->i_blkno);
2539 	u16 bit = le16_to_cpu(di->i_suballoc_bit);
2540 	u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2541 
2542 	if (di->i_suballoc_loc)
2543 		bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2544 	return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2545 					inode_alloc_bh, bit, bg_blkno, 1);
2546 }
2547 
2548 static int _ocfs2_free_clusters(handle_t *handle,
2549 				struct inode *bitmap_inode,
2550 				struct buffer_head *bitmap_bh,
2551 				u64 start_blk,
2552 				unsigned int num_clusters,
2553 				void (*undo_fn)(unsigned int bit,
2554 						unsigned long *bitmap))
2555 {
2556 	int status;
2557 	u16 bg_start_bit;
2558 	u64 bg_blkno;
2559 	struct ocfs2_dinode *fe;
2560 
2561 	/* You can't ever have a contiguous set of clusters
2562 	 * bigger than a block group bitmap so we never have to worry
2563 	 * about looping on them.
2564 	 * This is expensive. We can safely remove once this stuff has
2565 	 * gotten tested really well. */
2566 	BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
2567 
2568 	fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
2569 
2570 	ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2571 				     &bg_start_bit);
2572 
2573 	trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2574 			(unsigned long long)start_blk,
2575 			bg_start_bit, num_clusters);
2576 
2577 	status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2578 					   bg_start_bit, bg_blkno,
2579 					   num_clusters, undo_fn);
2580 	if (status < 0) {
2581 		mlog_errno(status);
2582 		goto out;
2583 	}
2584 
2585 	ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2586 					 num_clusters);
2587 
2588 out:
2589 	if (status)
2590 		mlog_errno(status);
2591 	return status;
2592 }
2593 
2594 int ocfs2_free_clusters(handle_t *handle,
2595 			struct inode *bitmap_inode,
2596 			struct buffer_head *bitmap_bh,
2597 			u64 start_blk,
2598 			unsigned int num_clusters)
2599 {
2600 	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2601 				    start_blk, num_clusters,
2602 				    _ocfs2_set_bit);
2603 }
2604 
2605 /*
2606  * Give never-used clusters back to the global bitmap.  We don't need
2607  * to protect these bits in the undo buffer.
2608  */
2609 int ocfs2_release_clusters(handle_t *handle,
2610 			   struct inode *bitmap_inode,
2611 			   struct buffer_head *bitmap_bh,
2612 			   u64 start_blk,
2613 			   unsigned int num_clusters)
2614 {
2615 	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2616 				    start_blk, num_clusters,
2617 				    _ocfs2_clear_bit);
2618 }
2619 
2620 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
2621 {
2622 	printk("Block Group:\n");
2623 	printk("bg_signature:       %s\n", bg->bg_signature);
2624 	printk("bg_size:            %u\n", bg->bg_size);
2625 	printk("bg_bits:            %u\n", bg->bg_bits);
2626 	printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
2627 	printk("bg_chain:           %u\n", bg->bg_chain);
2628 	printk("bg_generation:      %u\n", le32_to_cpu(bg->bg_generation));
2629 	printk("bg_next_group:      %llu\n",
2630 	       (unsigned long long)bg->bg_next_group);
2631 	printk("bg_parent_dinode:   %llu\n",
2632 	       (unsigned long long)bg->bg_parent_dinode);
2633 	printk("bg_blkno:           %llu\n",
2634 	       (unsigned long long)bg->bg_blkno);
2635 }
2636 
2637 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2638 {
2639 	int i;
2640 
2641 	printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
2642 	printk("i_signature:                  %s\n", fe->i_signature);
2643 	printk("i_size:                       %llu\n",
2644 	       (unsigned long long)fe->i_size);
2645 	printk("i_clusters:                   %u\n", fe->i_clusters);
2646 	printk("i_generation:                 %u\n",
2647 	       le32_to_cpu(fe->i_generation));
2648 	printk("id1.bitmap1.i_used:           %u\n",
2649 	       le32_to_cpu(fe->id1.bitmap1.i_used));
2650 	printk("id1.bitmap1.i_total:          %u\n",
2651 	       le32_to_cpu(fe->id1.bitmap1.i_total));
2652 	printk("id2.i_chain.cl_cpg:           %u\n", fe->id2.i_chain.cl_cpg);
2653 	printk("id2.i_chain.cl_bpc:           %u\n", fe->id2.i_chain.cl_bpc);
2654 	printk("id2.i_chain.cl_count:         %u\n", fe->id2.i_chain.cl_count);
2655 	printk("id2.i_chain.cl_next_free_rec: %u\n",
2656 	       fe->id2.i_chain.cl_next_free_rec);
2657 	for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2658 		printk("fe->id2.i_chain.cl_recs[%d].c_free:  %u\n", i,
2659 		       fe->id2.i_chain.cl_recs[i].c_free);
2660 		printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2661 		       fe->id2.i_chain.cl_recs[i].c_total);
2662 		printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2663 		       (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
2664 	}
2665 }
2666 
2667 /*
2668  * For a given allocation, determine which allocators will need to be
2669  * accessed, and lock them, reserving the appropriate number of bits.
2670  *
2671  * Sparse file systems call this from ocfs2_write_begin_nolock()
2672  * and ocfs2_allocate_unwritten_extents().
2673  *
2674  * File systems which don't support holes call this from
2675  * ocfs2_extend_allocation().
2676  */
2677 int ocfs2_lock_allocators(struct inode *inode,
2678 			  struct ocfs2_extent_tree *et,
2679 			  u32 clusters_to_add, u32 extents_to_split,
2680 			  struct ocfs2_alloc_context **data_ac,
2681 			  struct ocfs2_alloc_context **meta_ac)
2682 {
2683 	int ret = 0, num_free_extents;
2684 	unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2685 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2686 
2687 	*meta_ac = NULL;
2688 	if (data_ac)
2689 		*data_ac = NULL;
2690 
2691 	BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2692 
2693 	num_free_extents = ocfs2_num_free_extents(osb, et);
2694 	if (num_free_extents < 0) {
2695 		ret = num_free_extents;
2696 		mlog_errno(ret);
2697 		goto out;
2698 	}
2699 
2700 	/*
2701 	 * Sparse allocation file systems need to be more conservative
2702 	 * with reserving room for expansion - the actual allocation
2703 	 * happens while we've got a journal handle open so re-taking
2704 	 * a cluster lock (because we ran out of room for another
2705 	 * extent) will violate ordering rules.
2706 	 *
2707 	 * Most of the time we'll only be seeing this 1 cluster at a time
2708 	 * anyway.
2709 	 *
2710 	 * Always lock for any unwritten extents - we might want to
2711 	 * add blocks during a split.
2712 	 */
2713 	if (!num_free_extents ||
2714 	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2715 		ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2716 		if (ret < 0) {
2717 			if (ret != -ENOSPC)
2718 				mlog_errno(ret);
2719 			goto out;
2720 		}
2721 	}
2722 
2723 	if (clusters_to_add == 0)
2724 		goto out;
2725 
2726 	ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2727 	if (ret < 0) {
2728 		if (ret != -ENOSPC)
2729 			mlog_errno(ret);
2730 		goto out;
2731 	}
2732 
2733 out:
2734 	if (ret) {
2735 		if (*meta_ac) {
2736 			ocfs2_free_alloc_context(*meta_ac);
2737 			*meta_ac = NULL;
2738 		}
2739 
2740 		/*
2741 		 * We cannot have an error and a non null *data_ac.
2742 		 */
2743 	}
2744 
2745 	return ret;
2746 }
2747 
2748 /*
2749  * Read the inode specified by blkno to get suballoc_slot and
2750  * suballoc_bit.
2751  */
2752 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2753 				       u16 *suballoc_slot, u64 *group_blkno,
2754 				       u16 *suballoc_bit)
2755 {
2756 	int status;
2757 	struct buffer_head *inode_bh = NULL;
2758 	struct ocfs2_dinode *inode_fe;
2759 
2760 	trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2761 
2762 	/* dirty read disk */
2763 	status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2764 	if (status < 0) {
2765 		mlog(ML_ERROR, "read block %llu failed %d\n",
2766 		     (unsigned long long)blkno, status);
2767 		goto bail;
2768 	}
2769 
2770 	inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2771 	if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2772 		mlog(ML_ERROR, "invalid inode %llu requested\n",
2773 		     (unsigned long long)blkno);
2774 		status = -EINVAL;
2775 		goto bail;
2776 	}
2777 
2778 	if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2779 	    (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2780 		mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2781 		     (unsigned long long)blkno,
2782 		     (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2783 		status = -EINVAL;
2784 		goto bail;
2785 	}
2786 
2787 	if (suballoc_slot)
2788 		*suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2789 	if (suballoc_bit)
2790 		*suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2791 	if (group_blkno)
2792 		*group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2793 
2794 bail:
2795 	brelse(inode_bh);
2796 
2797 	if (status)
2798 		mlog_errno(status);
2799 	return status;
2800 }
2801 
2802 /*
2803  * test whether bit is SET in allocator bitmap or not.  on success, 0
2804  * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
2805  * is returned and *res is meaningless.  Call this after you have
2806  * cluster locked against suballoc, or you may get a result based on
2807  * non-up2date contents
2808  */
2809 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2810 				   struct inode *suballoc,
2811 				   struct buffer_head *alloc_bh,
2812 				   u64 group_blkno, u64 blkno,
2813 				   u16 bit, int *res)
2814 {
2815 	struct ocfs2_dinode *alloc_di;
2816 	struct ocfs2_group_desc *group;
2817 	struct buffer_head *group_bh = NULL;
2818 	u64 bg_blkno;
2819 	int status;
2820 
2821 	trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2822 				      (unsigned int)bit);
2823 
2824 	alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2825 	if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2826 		mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2827 		     (unsigned int)bit,
2828 		     ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2829 		status = -EINVAL;
2830 		goto bail;
2831 	}
2832 
2833 	bg_blkno = group_blkno ? group_blkno :
2834 		   ocfs2_which_suballoc_group(blkno, bit);
2835 	status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2836 					     &group_bh);
2837 	if (status < 0) {
2838 		mlog(ML_ERROR, "read group %llu failed %d\n",
2839 		     (unsigned long long)bg_blkno, status);
2840 		goto bail;
2841 	}
2842 
2843 	group = (struct ocfs2_group_desc *) group_bh->b_data;
2844 	*res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2845 
2846 bail:
2847 	brelse(group_bh);
2848 
2849 	if (status)
2850 		mlog_errno(status);
2851 	return status;
2852 }
2853 
2854 /*
2855  * Test if the bit representing this inode (blkno) is set in the
2856  * suballocator.
2857  *
2858  * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2859  *
2860  * In the event of failure, a negative value is returned and *res is
2861  * meaningless.
2862  *
2863  * Callers must make sure to hold nfs_sync_lock to prevent
2864  * ocfs2_delete_inode() on another node from accessing the same
2865  * suballocator concurrently.
2866  */
2867 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2868 {
2869 	int status;
2870 	u64 group_blkno = 0;
2871 	u16 suballoc_bit = 0, suballoc_slot = 0;
2872 	struct inode *inode_alloc_inode;
2873 	struct buffer_head *alloc_bh = NULL;
2874 
2875 	trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2876 
2877 	status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2878 					     &group_blkno, &suballoc_bit);
2879 	if (status < 0) {
2880 		mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2881 		goto bail;
2882 	}
2883 
2884 	inode_alloc_inode =
2885 		ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2886 					    suballoc_slot);
2887 	if (!inode_alloc_inode) {
2888 		/* the error code could be inaccurate, but we are not able to
2889 		 * get the correct one. */
2890 		status = -EINVAL;
2891 		mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2892 		     (u32)suballoc_slot);
2893 		goto bail;
2894 	}
2895 
2896 	inode_lock(inode_alloc_inode);
2897 	status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2898 	if (status < 0) {
2899 		inode_unlock(inode_alloc_inode);
2900 		iput(inode_alloc_inode);
2901 		mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2902 		     (u32)suballoc_slot, status);
2903 		goto bail;
2904 	}
2905 
2906 	status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2907 					 group_blkno, blkno, suballoc_bit, res);
2908 	if (status < 0)
2909 		mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2910 
2911 	ocfs2_inode_unlock(inode_alloc_inode, 0);
2912 	inode_unlock(inode_alloc_inode);
2913 
2914 	iput(inode_alloc_inode);
2915 	brelse(alloc_bh);
2916 bail:
2917 	if (status)
2918 		mlog_errno(status);
2919 	return status;
2920 }
2921