xref: /openbmc/linux/fs/ocfs2/suballoc.c (revision 82e6fdd6)
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 
1153 	return status;
1154 }
1155 
1156 /* Callers don't need to care which bitmap (local alloc or main) to
1157  * use so we figure it out for them, but unfortunately this clutters
1158  * things a bit. */
1159 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1160 					     u32 bits_wanted, u64 max_block,
1161 					     int flags,
1162 					     struct ocfs2_alloc_context **ac)
1163 {
1164 	int status, ret = 0;
1165 	int retried = 0;
1166 
1167 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1168 	if (!(*ac)) {
1169 		status = -ENOMEM;
1170 		mlog_errno(status);
1171 		goto bail;
1172 	}
1173 
1174 	(*ac)->ac_bits_wanted = bits_wanted;
1175 	(*ac)->ac_max_block = max_block;
1176 
1177 	status = -ENOSPC;
1178 	if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1179 	    ocfs2_alloc_should_use_local(osb, bits_wanted)) {
1180 		status = ocfs2_reserve_local_alloc_bits(osb,
1181 							bits_wanted,
1182 							*ac);
1183 		if ((status < 0) && (status != -ENOSPC)) {
1184 			mlog_errno(status);
1185 			goto bail;
1186 		}
1187 	}
1188 
1189 	if (status == -ENOSPC) {
1190 retry:
1191 		status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1192 		/* Retry if there is sufficient space cached in truncate log */
1193 		if (status == -ENOSPC && !retried) {
1194 			retried = 1;
1195 			ocfs2_inode_unlock((*ac)->ac_inode, 1);
1196 			inode_unlock((*ac)->ac_inode);
1197 
1198 			ret = ocfs2_try_to_free_truncate_log(osb, bits_wanted);
1199 			if (ret == 1) {
1200 				iput((*ac)->ac_inode);
1201 				(*ac)->ac_inode = NULL;
1202 				goto retry;
1203 			}
1204 
1205 			if (ret < 0)
1206 				mlog_errno(ret);
1207 
1208 			inode_lock((*ac)->ac_inode);
1209 			ret = ocfs2_inode_lock((*ac)->ac_inode, NULL, 1);
1210 			if (ret < 0) {
1211 				mlog_errno(ret);
1212 				inode_unlock((*ac)->ac_inode);
1213 				iput((*ac)->ac_inode);
1214 				(*ac)->ac_inode = NULL;
1215 				goto bail;
1216 			}
1217 		}
1218 		if (status < 0) {
1219 			if (status != -ENOSPC)
1220 				mlog_errno(status);
1221 			goto bail;
1222 		}
1223 	}
1224 
1225 	status = 0;
1226 bail:
1227 	if ((status < 0) && *ac) {
1228 		ocfs2_free_alloc_context(*ac);
1229 		*ac = NULL;
1230 	}
1231 
1232 	if (status)
1233 		mlog_errno(status);
1234 	return status;
1235 }
1236 
1237 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1238 			   u32 bits_wanted,
1239 			   struct ocfs2_alloc_context **ac)
1240 {
1241 	return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1242 						 ALLOC_NEW_GROUP, ac);
1243 }
1244 
1245 /*
1246  * More or less lifted from ext3. I'll leave their description below:
1247  *
1248  * "For ext3 allocations, we must not reuse any blocks which are
1249  * allocated in the bitmap buffer's "last committed data" copy.  This
1250  * prevents deletes from freeing up the page for reuse until we have
1251  * committed the delete transaction.
1252  *
1253  * If we didn't do this, then deleting something and reallocating it as
1254  * data would allow the old block to be overwritten before the
1255  * transaction committed (because we force data to disk before commit).
1256  * This would lead to corruption if we crashed between overwriting the
1257  * data and committing the delete.
1258  *
1259  * @@@ We may want to make this allocation behaviour conditional on
1260  * data-writes at some point, and disable it for metadata allocations or
1261  * sync-data inodes."
1262  *
1263  * Note: OCFS2 already does this differently for metadata vs data
1264  * allocations, as those bitmaps are separate and undo access is never
1265  * called on a metadata group descriptor.
1266  */
1267 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1268 					 int nr)
1269 {
1270 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1271 	int ret;
1272 
1273 	if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1274 		return 0;
1275 
1276 	if (!buffer_jbd(bg_bh))
1277 		return 1;
1278 
1279 	jbd_lock_bh_state(bg_bh);
1280 	bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
1281 	if (bg)
1282 		ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1283 	else
1284 		ret = 1;
1285 	jbd_unlock_bh_state(bg_bh);
1286 
1287 	return ret;
1288 }
1289 
1290 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1291 					     struct buffer_head *bg_bh,
1292 					     unsigned int bits_wanted,
1293 					     unsigned int total_bits,
1294 					     struct ocfs2_suballoc_result *res)
1295 {
1296 	void *bitmap;
1297 	u16 best_offset, best_size;
1298 	int offset, start, found, status = 0;
1299 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1300 
1301 	/* Callers got this descriptor from
1302 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1303 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1304 
1305 	found = start = best_offset = best_size = 0;
1306 	bitmap = bg->bg_bitmap;
1307 
1308 	while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1309 		if (offset == total_bits)
1310 			break;
1311 
1312 		if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1313 			/* We found a zero, but we can't use it as it
1314 			 * hasn't been put to disk yet! */
1315 			found = 0;
1316 			start = offset + 1;
1317 		} else if (offset == start) {
1318 			/* we found a zero */
1319 			found++;
1320 			/* move start to the next bit to test */
1321 			start++;
1322 		} else {
1323 			/* got a zero after some ones */
1324 			found = 1;
1325 			start = offset + 1;
1326 		}
1327 		if (found > best_size) {
1328 			best_size = found;
1329 			best_offset = start - found;
1330 		}
1331 		/* we got everything we needed */
1332 		if (found == bits_wanted) {
1333 			/* mlog(0, "Found it all!\n"); */
1334 			break;
1335 		}
1336 	}
1337 
1338 	if (best_size) {
1339 		res->sr_bit_offset = best_offset;
1340 		res->sr_bits = best_size;
1341 	} else {
1342 		status = -ENOSPC;
1343 		/* No error log here -- see the comment above
1344 		 * ocfs2_test_bg_bit_allocatable */
1345 	}
1346 
1347 	return status;
1348 }
1349 
1350 int ocfs2_block_group_set_bits(handle_t *handle,
1351 					     struct inode *alloc_inode,
1352 					     struct ocfs2_group_desc *bg,
1353 					     struct buffer_head *group_bh,
1354 					     unsigned int bit_off,
1355 					     unsigned int num_bits)
1356 {
1357 	int status;
1358 	void *bitmap = bg->bg_bitmap;
1359 	int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1360 
1361 	/* All callers get the descriptor via
1362 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1363 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1364 	BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1365 
1366 	trace_ocfs2_block_group_set_bits(bit_off, num_bits);
1367 
1368 	if (ocfs2_is_cluster_bitmap(alloc_inode))
1369 		journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1370 
1371 	status = ocfs2_journal_access_gd(handle,
1372 					 INODE_CACHE(alloc_inode),
1373 					 group_bh,
1374 					 journal_type);
1375 	if (status < 0) {
1376 		mlog_errno(status);
1377 		goto bail;
1378 	}
1379 
1380 	le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1381 	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1382 		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
1383 				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
1384 				   le16_to_cpu(bg->bg_bits),
1385 				   le16_to_cpu(bg->bg_free_bits_count),
1386 				   num_bits);
1387 	}
1388 	while(num_bits--)
1389 		ocfs2_set_bit(bit_off++, bitmap);
1390 
1391 	ocfs2_journal_dirty(handle, group_bh);
1392 
1393 bail:
1394 	return status;
1395 }
1396 
1397 /* find the one with the most empty bits */
1398 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1399 {
1400 	u16 curr, best;
1401 
1402 	BUG_ON(!cl->cl_next_free_rec);
1403 
1404 	best = curr = 0;
1405 	while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1406 		if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1407 		    le32_to_cpu(cl->cl_recs[best].c_free))
1408 			best = curr;
1409 		curr++;
1410 	}
1411 
1412 	BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1413 	return best;
1414 }
1415 
1416 static int ocfs2_relink_block_group(handle_t *handle,
1417 				    struct inode *alloc_inode,
1418 				    struct buffer_head *fe_bh,
1419 				    struct buffer_head *bg_bh,
1420 				    struct buffer_head *prev_bg_bh,
1421 				    u16 chain)
1422 {
1423 	int status;
1424 	/* there is a really tiny chance the journal calls could fail,
1425 	 * but we wouldn't want inconsistent blocks in *any* case. */
1426 	u64 bg_ptr, prev_bg_ptr;
1427 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1428 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1429 	struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1430 
1431 	/* The caller got these descriptors from
1432 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1433 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1434 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1435 
1436 	trace_ocfs2_relink_block_group(
1437 		(unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1438 		(unsigned long long)le64_to_cpu(bg->bg_blkno),
1439 		(unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1440 
1441 	bg_ptr = le64_to_cpu(bg->bg_next_group);
1442 	prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1443 
1444 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1445 					 prev_bg_bh,
1446 					 OCFS2_JOURNAL_ACCESS_WRITE);
1447 	if (status < 0)
1448 		goto out;
1449 
1450 	prev_bg->bg_next_group = bg->bg_next_group;
1451 	ocfs2_journal_dirty(handle, prev_bg_bh);
1452 
1453 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1454 					 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1455 	if (status < 0)
1456 		goto out_rollback_prev_bg;
1457 
1458 	bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1459 	ocfs2_journal_dirty(handle, bg_bh);
1460 
1461 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1462 					 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1463 	if (status < 0)
1464 		goto out_rollback_bg;
1465 
1466 	fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1467 	ocfs2_journal_dirty(handle, fe_bh);
1468 
1469 out:
1470 	if (status < 0)
1471 		mlog_errno(status);
1472 	return status;
1473 
1474 out_rollback_bg:
1475 	bg->bg_next_group = cpu_to_le64(bg_ptr);
1476 out_rollback_prev_bg:
1477 	prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1478 	goto out;
1479 }
1480 
1481 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1482 						     u32 wanted)
1483 {
1484 	return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1485 }
1486 
1487 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1488  * value on error. */
1489 static int ocfs2_cluster_group_search(struct inode *inode,
1490 				      struct buffer_head *group_bh,
1491 				      u32 bits_wanted, u32 min_bits,
1492 				      u64 max_block,
1493 				      struct ocfs2_suballoc_result *res)
1494 {
1495 	int search = -ENOSPC;
1496 	int ret;
1497 	u64 blkoff;
1498 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1499 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1500 	unsigned int max_bits, gd_cluster_off;
1501 
1502 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1503 
1504 	if (gd->bg_free_bits_count) {
1505 		max_bits = le16_to_cpu(gd->bg_bits);
1506 
1507 		/* Tail groups in cluster bitmaps which aren't cpg
1508 		 * aligned are prone to partial extension by a failed
1509 		 * fs resize. If the file system resize never got to
1510 		 * update the dinode cluster count, then we don't want
1511 		 * to trust any clusters past it, regardless of what
1512 		 * the group descriptor says. */
1513 		gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1514 							  le64_to_cpu(gd->bg_blkno));
1515 		if ((gd_cluster_off + max_bits) >
1516 		    OCFS2_I(inode)->ip_clusters) {
1517 			max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1518 			trace_ocfs2_cluster_group_search_wrong_max_bits(
1519 				(unsigned long long)le64_to_cpu(gd->bg_blkno),
1520 				le16_to_cpu(gd->bg_bits),
1521 				OCFS2_I(inode)->ip_clusters, max_bits);
1522 		}
1523 
1524 		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1525 							group_bh, bits_wanted,
1526 							max_bits, res);
1527 		if (ret)
1528 			return ret;
1529 
1530 		if (max_block) {
1531 			blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1532 							  gd_cluster_off +
1533 							  res->sr_bit_offset +
1534 							  res->sr_bits);
1535 			trace_ocfs2_cluster_group_search_max_block(
1536 				(unsigned long long)blkoff,
1537 				(unsigned long long)max_block);
1538 			if (blkoff > max_block)
1539 				return -ENOSPC;
1540 		}
1541 
1542 		/* ocfs2_block_group_find_clear_bits() might
1543 		 * return success, but we still want to return
1544 		 * -ENOSPC unless it found the minimum number
1545 		 * of bits. */
1546 		if (min_bits <= res->sr_bits)
1547 			search = 0; /* success */
1548 		else if (res->sr_bits) {
1549 			/*
1550 			 * Don't show bits which we'll be returning
1551 			 * for allocation to the local alloc bitmap.
1552 			 */
1553 			ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
1554 		}
1555 	}
1556 
1557 	return search;
1558 }
1559 
1560 static int ocfs2_block_group_search(struct inode *inode,
1561 				    struct buffer_head *group_bh,
1562 				    u32 bits_wanted, u32 min_bits,
1563 				    u64 max_block,
1564 				    struct ocfs2_suballoc_result *res)
1565 {
1566 	int ret = -ENOSPC;
1567 	u64 blkoff;
1568 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1569 
1570 	BUG_ON(min_bits != 1);
1571 	BUG_ON(ocfs2_is_cluster_bitmap(inode));
1572 
1573 	if (bg->bg_free_bits_count) {
1574 		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1575 							group_bh, bits_wanted,
1576 							le16_to_cpu(bg->bg_bits),
1577 							res);
1578 		if (!ret && max_block) {
1579 			blkoff = le64_to_cpu(bg->bg_blkno) +
1580 				res->sr_bit_offset + res->sr_bits;
1581 			trace_ocfs2_block_group_search_max_block(
1582 				(unsigned long long)blkoff,
1583 				(unsigned long long)max_block);
1584 			if (blkoff > max_block)
1585 				ret = -ENOSPC;
1586 		}
1587 	}
1588 
1589 	return ret;
1590 }
1591 
1592 int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1593 				       handle_t *handle,
1594 				       struct buffer_head *di_bh,
1595 				       u32 num_bits,
1596 				       u16 chain)
1597 {
1598 	int ret;
1599 	u32 tmp_used;
1600 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1601 	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1602 
1603 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1604 				      OCFS2_JOURNAL_ACCESS_WRITE);
1605 	if (ret < 0) {
1606 		mlog_errno(ret);
1607 		goto out;
1608 	}
1609 
1610 	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1611 	di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1612 	le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1613 	ocfs2_journal_dirty(handle, di_bh);
1614 
1615 out:
1616 	return ret;
1617 }
1618 
1619 void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
1620 				       struct buffer_head *di_bh,
1621 				       u32 num_bits,
1622 				       u16 chain)
1623 {
1624 	u32 tmp_used;
1625 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1626 	struct ocfs2_chain_list *cl;
1627 
1628 	cl = (struct ocfs2_chain_list *)&di->id2.i_chain;
1629 	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1630 	di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits);
1631 	le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits);
1632 }
1633 
1634 static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1635 					 struct ocfs2_extent_rec *rec,
1636 					 struct ocfs2_chain_list *cl)
1637 {
1638 	unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1639 	unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1640 	unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1641 
1642 	if (res->sr_bit_offset < bitoff)
1643 		return 0;
1644 	if (res->sr_bit_offset >= (bitoff + bitcount))
1645 		return 0;
1646 	res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1647 		(res->sr_bit_offset - bitoff);
1648 	if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1649 		res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1650 	return 1;
1651 }
1652 
1653 static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1654 					  struct ocfs2_group_desc *bg,
1655 					  struct ocfs2_suballoc_result *res)
1656 {
1657 	int i;
1658 	u64 bg_blkno = res->sr_bg_blkno;  /* Save off */
1659 	struct ocfs2_extent_rec *rec;
1660 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1661 	struct ocfs2_chain_list *cl = &di->id2.i_chain;
1662 
1663 	if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1664 		res->sr_blkno = 0;
1665 		return;
1666 	}
1667 
1668 	res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1669 	res->sr_bg_blkno = 0;  /* Clear it for contig block groups */
1670 	if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1671 	    !bg->bg_list.l_next_free_rec)
1672 		return;
1673 
1674 	for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1675 		rec = &bg->bg_list.l_recs[i];
1676 		if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1677 			res->sr_bg_blkno = bg_blkno;  /* Restore */
1678 			break;
1679 		}
1680 	}
1681 }
1682 
1683 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1684 				  handle_t *handle,
1685 				  u32 bits_wanted,
1686 				  u32 min_bits,
1687 				  struct ocfs2_suballoc_result *res,
1688 				  u16 *bits_left)
1689 {
1690 	int ret;
1691 	struct buffer_head *group_bh = NULL;
1692 	struct ocfs2_group_desc *gd;
1693 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1694 	struct inode *alloc_inode = ac->ac_inode;
1695 
1696 	ret = ocfs2_read_group_descriptor(alloc_inode, di,
1697 					  res->sr_bg_blkno, &group_bh);
1698 	if (ret < 0) {
1699 		mlog_errno(ret);
1700 		return ret;
1701 	}
1702 
1703 	gd = (struct ocfs2_group_desc *) group_bh->b_data;
1704 	ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1705 				  ac->ac_max_block, res);
1706 	if (ret < 0) {
1707 		if (ret != -ENOSPC)
1708 			mlog_errno(ret);
1709 		goto out;
1710 	}
1711 
1712 	if (!ret)
1713 		ocfs2_bg_discontig_fix_result(ac, gd, res);
1714 
1715 	/*
1716 	 * sr_bg_blkno might have been changed by
1717 	 * ocfs2_bg_discontig_fix_result
1718 	 */
1719 	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1720 
1721 	if (ac->ac_find_loc_only)
1722 		goto out_loc_only;
1723 
1724 	ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1725 					       res->sr_bits,
1726 					       le16_to_cpu(gd->bg_chain));
1727 	if (ret < 0) {
1728 		mlog_errno(ret);
1729 		goto out;
1730 	}
1731 
1732 	ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1733 					 res->sr_bit_offset, res->sr_bits);
1734 	if (ret < 0) {
1735 		ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh,
1736 					       res->sr_bits,
1737 					       le16_to_cpu(gd->bg_chain));
1738 		mlog_errno(ret);
1739 	}
1740 
1741 out_loc_only:
1742 	*bits_left = le16_to_cpu(gd->bg_free_bits_count);
1743 
1744 out:
1745 	brelse(group_bh);
1746 
1747 	return ret;
1748 }
1749 
1750 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1751 			      handle_t *handle,
1752 			      u32 bits_wanted,
1753 			      u32 min_bits,
1754 			      struct ocfs2_suballoc_result *res,
1755 			      u16 *bits_left)
1756 {
1757 	int status;
1758 	u16 chain;
1759 	u64 next_group;
1760 	struct inode *alloc_inode = ac->ac_inode;
1761 	struct buffer_head *group_bh = NULL;
1762 	struct buffer_head *prev_group_bh = NULL;
1763 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1764 	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1765 	struct ocfs2_group_desc *bg;
1766 
1767 	chain = ac->ac_chain;
1768 	trace_ocfs2_search_chain_begin(
1769 		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1770 		bits_wanted, chain);
1771 
1772 	status = ocfs2_read_group_descriptor(alloc_inode, fe,
1773 					     le64_to_cpu(cl->cl_recs[chain].c_blkno),
1774 					     &group_bh);
1775 	if (status < 0) {
1776 		mlog_errno(status);
1777 		goto bail;
1778 	}
1779 	bg = (struct ocfs2_group_desc *) group_bh->b_data;
1780 
1781 	status = -ENOSPC;
1782 	/* for now, the chain search is a bit simplistic. We just use
1783 	 * the 1st group with any empty bits. */
1784 	while ((status = ac->ac_group_search(alloc_inode, group_bh,
1785 					     bits_wanted, min_bits,
1786 					     ac->ac_max_block,
1787 					     res)) == -ENOSPC) {
1788 		if (!bg->bg_next_group)
1789 			break;
1790 
1791 		brelse(prev_group_bh);
1792 		prev_group_bh = NULL;
1793 
1794 		next_group = le64_to_cpu(bg->bg_next_group);
1795 		prev_group_bh = group_bh;
1796 		group_bh = NULL;
1797 		status = ocfs2_read_group_descriptor(alloc_inode, fe,
1798 						     next_group, &group_bh);
1799 		if (status < 0) {
1800 			mlog_errno(status);
1801 			goto bail;
1802 		}
1803 		bg = (struct ocfs2_group_desc *) group_bh->b_data;
1804 	}
1805 	if (status < 0) {
1806 		if (status != -ENOSPC)
1807 			mlog_errno(status);
1808 		goto bail;
1809 	}
1810 
1811 	trace_ocfs2_search_chain_succ(
1812 		(unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1813 
1814 	res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1815 
1816 	BUG_ON(res->sr_bits == 0);
1817 	if (!status)
1818 		ocfs2_bg_discontig_fix_result(ac, bg, res);
1819 
1820 	/*
1821 	 * sr_bg_blkno might have been changed by
1822 	 * ocfs2_bg_discontig_fix_result
1823 	 */
1824 	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1825 
1826 	/*
1827 	 * Keep track of previous block descriptor read. When
1828 	 * we find a target, if we have read more than X
1829 	 * number of descriptors, and the target is reasonably
1830 	 * empty, relink him to top of his chain.
1831 	 *
1832 	 * We've read 0 extra blocks and only send one more to
1833 	 * the transaction, yet the next guy to search has a
1834 	 * much easier time.
1835 	 *
1836 	 * Do this *after* figuring out how many bits we're taking out
1837 	 * of our target group.
1838 	 */
1839 	if (!ac->ac_disable_chain_relink &&
1840 	    (prev_group_bh) &&
1841 	    (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1842 		status = ocfs2_relink_block_group(handle, alloc_inode,
1843 						  ac->ac_bh, group_bh,
1844 						  prev_group_bh, chain);
1845 		if (status < 0) {
1846 			mlog_errno(status);
1847 			goto bail;
1848 		}
1849 	}
1850 
1851 	if (ac->ac_find_loc_only)
1852 		goto out_loc_only;
1853 
1854 	status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1855 						  ac->ac_bh, res->sr_bits,
1856 						  chain);
1857 	if (status) {
1858 		mlog_errno(status);
1859 		goto bail;
1860 	}
1861 
1862 	status = ocfs2_block_group_set_bits(handle,
1863 					    alloc_inode,
1864 					    bg,
1865 					    group_bh,
1866 					    res->sr_bit_offset,
1867 					    res->sr_bits);
1868 	if (status < 0) {
1869 		ocfs2_rollback_alloc_dinode_counts(alloc_inode,
1870 					ac->ac_bh, res->sr_bits, chain);
1871 		mlog_errno(status);
1872 		goto bail;
1873 	}
1874 
1875 	trace_ocfs2_search_chain_end(
1876 			(unsigned long long)le64_to_cpu(fe->i_blkno),
1877 			res->sr_bits);
1878 
1879 out_loc_only:
1880 	*bits_left = le16_to_cpu(bg->bg_free_bits_count);
1881 bail:
1882 	brelse(group_bh);
1883 	brelse(prev_group_bh);
1884 
1885 	if (status)
1886 		mlog_errno(status);
1887 	return status;
1888 }
1889 
1890 /* will give out up to bits_wanted contiguous bits. */
1891 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1892 				     handle_t *handle,
1893 				     u32 bits_wanted,
1894 				     u32 min_bits,
1895 				     struct ocfs2_suballoc_result *res)
1896 {
1897 	int status;
1898 	u16 victim, i;
1899 	u16 bits_left = 0;
1900 	u64 hint = ac->ac_last_group;
1901 	struct ocfs2_chain_list *cl;
1902 	struct ocfs2_dinode *fe;
1903 
1904 	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1905 	BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1906 	BUG_ON(!ac->ac_bh);
1907 
1908 	fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1909 
1910 	/* The bh was validated by the inode read during
1911 	 * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1912 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1913 
1914 	if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1915 	    le32_to_cpu(fe->id1.bitmap1.i_total)) {
1916 		status = ocfs2_error(ac->ac_inode->i_sb,
1917 				     "Chain allocator dinode %llu has %u used bits but only %u total\n",
1918 				     (unsigned long long)le64_to_cpu(fe->i_blkno),
1919 				     le32_to_cpu(fe->id1.bitmap1.i_used),
1920 				     le32_to_cpu(fe->id1.bitmap1.i_total));
1921 		goto bail;
1922 	}
1923 
1924 	res->sr_bg_blkno = hint;
1925 	if (res->sr_bg_blkno) {
1926 		/* Attempt to short-circuit the usual search mechanism
1927 		 * by jumping straight to the most recently used
1928 		 * allocation group. This helps us maintain some
1929 		 * contiguousness across allocations. */
1930 		status = ocfs2_search_one_group(ac, handle, bits_wanted,
1931 						min_bits, res, &bits_left);
1932 		if (!status)
1933 			goto set_hint;
1934 		if (status < 0 && status != -ENOSPC) {
1935 			mlog_errno(status);
1936 			goto bail;
1937 		}
1938 	}
1939 
1940 	cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1941 
1942 	victim = ocfs2_find_victim_chain(cl);
1943 	ac->ac_chain = victim;
1944 
1945 	status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1946 				    res, &bits_left);
1947 	if (!status) {
1948 		if (ocfs2_is_cluster_bitmap(ac->ac_inode))
1949 			hint = res->sr_bg_blkno;
1950 		else
1951 			hint = ocfs2_group_from_res(res);
1952 		goto set_hint;
1953 	}
1954 	if (status < 0 && status != -ENOSPC) {
1955 		mlog_errno(status);
1956 		goto bail;
1957 	}
1958 
1959 	trace_ocfs2_claim_suballoc_bits(victim);
1960 
1961 	/* If we didn't pick a good victim, then just default to
1962 	 * searching each chain in order. Don't allow chain relinking
1963 	 * because we only calculate enough journal credits for one
1964 	 * relink per alloc. */
1965 	ac->ac_disable_chain_relink = 1;
1966 	for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1967 		if (i == victim)
1968 			continue;
1969 		if (!cl->cl_recs[i].c_free)
1970 			continue;
1971 
1972 		ac->ac_chain = i;
1973 		status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1974 					    res, &bits_left);
1975 		if (!status) {
1976 			hint = ocfs2_group_from_res(res);
1977 			break;
1978 		}
1979 		if (status < 0 && status != -ENOSPC) {
1980 			mlog_errno(status);
1981 			goto bail;
1982 		}
1983 	}
1984 
1985 set_hint:
1986 	if (status != -ENOSPC) {
1987 		/* If the next search of this group is not likely to
1988 		 * yield a suitable extent, then we reset the last
1989 		 * group hint so as to not waste a disk read */
1990 		if (bits_left < min_bits)
1991 			ac->ac_last_group = 0;
1992 		else
1993 			ac->ac_last_group = hint;
1994 	}
1995 
1996 bail:
1997 	if (status)
1998 		mlog_errno(status);
1999 	return status;
2000 }
2001 
2002 int ocfs2_claim_metadata(handle_t *handle,
2003 			 struct ocfs2_alloc_context *ac,
2004 			 u32 bits_wanted,
2005 			 u64 *suballoc_loc,
2006 			 u16 *suballoc_bit_start,
2007 			 unsigned int *num_bits,
2008 			 u64 *blkno_start)
2009 {
2010 	int status;
2011 	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2012 
2013 	BUG_ON(!ac);
2014 	BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
2015 	BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
2016 
2017 	status = ocfs2_claim_suballoc_bits(ac,
2018 					   handle,
2019 					   bits_wanted,
2020 					   1,
2021 					   &res);
2022 	if (status < 0) {
2023 		mlog_errno(status);
2024 		goto bail;
2025 	}
2026 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2027 
2028 	*suballoc_loc = res.sr_bg_blkno;
2029 	*suballoc_bit_start = res.sr_bit_offset;
2030 	*blkno_start = res.sr_blkno;
2031 	ac->ac_bits_given += res.sr_bits;
2032 	*num_bits = res.sr_bits;
2033 	status = 0;
2034 bail:
2035 	if (status)
2036 		mlog_errno(status);
2037 	return status;
2038 }
2039 
2040 static void ocfs2_init_inode_ac_group(struct inode *dir,
2041 				      struct buffer_head *parent_di_bh,
2042 				      struct ocfs2_alloc_context *ac)
2043 {
2044 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2045 	/*
2046 	 * Try to allocate inodes from some specific group.
2047 	 *
2048 	 * If the parent dir has recorded the last group used in allocation,
2049 	 * cool, use it. Otherwise if we try to allocate new inode from the
2050 	 * same slot the parent dir belongs to, use the same chunk.
2051 	 *
2052 	 * We are very careful here to avoid the mistake of setting
2053 	 * ac_last_group to a group descriptor from a different (unlocked) slot.
2054 	 */
2055 	if (OCFS2_I(dir)->ip_last_used_group &&
2056 	    OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2057 		ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2058 	else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2059 		if (di->i_suballoc_loc)
2060 			ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2061 		else
2062 			ac->ac_last_group = ocfs2_which_suballoc_group(
2063 					le64_to_cpu(di->i_blkno),
2064 					le16_to_cpu(di->i_suballoc_bit));
2065 	}
2066 }
2067 
2068 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2069 					     struct ocfs2_alloc_context *ac)
2070 {
2071 	OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2072 	OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2073 }
2074 
2075 int ocfs2_find_new_inode_loc(struct inode *dir,
2076 			     struct buffer_head *parent_fe_bh,
2077 			     struct ocfs2_alloc_context *ac,
2078 			     u64 *fe_blkno)
2079 {
2080 	int ret;
2081 	handle_t *handle = NULL;
2082 	struct ocfs2_suballoc_result *res;
2083 
2084 	BUG_ON(!ac);
2085 	BUG_ON(ac->ac_bits_given != 0);
2086 	BUG_ON(ac->ac_bits_wanted != 1);
2087 	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2088 
2089 	res = kzalloc(sizeof(*res), GFP_NOFS);
2090 	if (res == NULL) {
2091 		ret = -ENOMEM;
2092 		mlog_errno(ret);
2093 		goto out;
2094 	}
2095 
2096 	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2097 
2098 	/*
2099 	 * The handle started here is for chain relink. Alternatively,
2100 	 * we could just disable relink for these calls.
2101 	 */
2102 	handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2103 	if (IS_ERR(handle)) {
2104 		ret = PTR_ERR(handle);
2105 		handle = NULL;
2106 		mlog_errno(ret);
2107 		goto out;
2108 	}
2109 
2110 	/*
2111 	 * This will instruct ocfs2_claim_suballoc_bits and
2112 	 * ocfs2_search_one_group to search but save actual allocation
2113 	 * for later.
2114 	 */
2115 	ac->ac_find_loc_only = 1;
2116 
2117 	ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2118 	if (ret < 0) {
2119 		mlog_errno(ret);
2120 		goto out;
2121 	}
2122 
2123 	ac->ac_find_loc_priv = res;
2124 	*fe_blkno = res->sr_blkno;
2125 	ocfs2_update_inode_fsync_trans(handle, dir, 0);
2126 out:
2127 	if (handle)
2128 		ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2129 
2130 	if (ret)
2131 		kfree(res);
2132 
2133 	return ret;
2134 }
2135 
2136 int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2137 				 struct inode *dir,
2138 				 struct ocfs2_alloc_context *ac,
2139 				 u64 *suballoc_loc,
2140 				 u16 *suballoc_bit,
2141 				 u64 di_blkno)
2142 {
2143 	int ret;
2144 	u16 chain;
2145 	struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2146 	struct buffer_head *bg_bh = NULL;
2147 	struct ocfs2_group_desc *bg;
2148 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2149 
2150 	/*
2151 	 * Since di_blkno is being passed back in, we check for any
2152 	 * inconsistencies which may have happened between
2153 	 * calls. These are code bugs as di_blkno is not expected to
2154 	 * change once returned from ocfs2_find_new_inode_loc()
2155 	 */
2156 	BUG_ON(res->sr_blkno != di_blkno);
2157 
2158 	ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2159 					  res->sr_bg_stable_blkno, &bg_bh);
2160 	if (ret) {
2161 		mlog_errno(ret);
2162 		goto out;
2163 	}
2164 
2165 	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2166 	chain = le16_to_cpu(bg->bg_chain);
2167 
2168 	ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2169 					       ac->ac_bh, res->sr_bits,
2170 					       chain);
2171 	if (ret) {
2172 		mlog_errno(ret);
2173 		goto out;
2174 	}
2175 
2176 	ret = ocfs2_block_group_set_bits(handle,
2177 					 ac->ac_inode,
2178 					 bg,
2179 					 bg_bh,
2180 					 res->sr_bit_offset,
2181 					 res->sr_bits);
2182 	if (ret < 0) {
2183 		ocfs2_rollback_alloc_dinode_counts(ac->ac_inode,
2184 					       ac->ac_bh, res->sr_bits, chain);
2185 		mlog_errno(ret);
2186 		goto out;
2187 	}
2188 
2189 	trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2190 					   res->sr_bits);
2191 
2192 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2193 
2194 	BUG_ON(res->sr_bits != 1);
2195 
2196 	*suballoc_loc = res->sr_bg_blkno;
2197 	*suballoc_bit = res->sr_bit_offset;
2198 	ac->ac_bits_given++;
2199 	ocfs2_save_inode_ac_group(dir, ac);
2200 
2201 out:
2202 	brelse(bg_bh);
2203 
2204 	return ret;
2205 }
2206 
2207 int ocfs2_claim_new_inode(handle_t *handle,
2208 			  struct inode *dir,
2209 			  struct buffer_head *parent_fe_bh,
2210 			  struct ocfs2_alloc_context *ac,
2211 			  u64 *suballoc_loc,
2212 			  u16 *suballoc_bit,
2213 			  u64 *fe_blkno)
2214 {
2215 	int status;
2216 	struct ocfs2_suballoc_result res;
2217 
2218 	BUG_ON(!ac);
2219 	BUG_ON(ac->ac_bits_given != 0);
2220 	BUG_ON(ac->ac_bits_wanted != 1);
2221 	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2222 
2223 	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2224 
2225 	status = ocfs2_claim_suballoc_bits(ac,
2226 					   handle,
2227 					   1,
2228 					   1,
2229 					   &res);
2230 	if (status < 0) {
2231 		mlog_errno(status);
2232 		goto bail;
2233 	}
2234 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2235 
2236 	BUG_ON(res.sr_bits != 1);
2237 
2238 	*suballoc_loc = res.sr_bg_blkno;
2239 	*suballoc_bit = res.sr_bit_offset;
2240 	*fe_blkno = res.sr_blkno;
2241 	ac->ac_bits_given++;
2242 	ocfs2_save_inode_ac_group(dir, ac);
2243 	status = 0;
2244 bail:
2245 	if (status)
2246 		mlog_errno(status);
2247 	return status;
2248 }
2249 
2250 /* translate a group desc. blkno and it's bitmap offset into
2251  * disk cluster offset. */
2252 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2253 						   u64 bg_blkno,
2254 						   u16 bg_bit_off)
2255 {
2256 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2257 	u32 cluster = 0;
2258 
2259 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2260 
2261 	if (bg_blkno != osb->first_cluster_group_blkno)
2262 		cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2263 	cluster += (u32) bg_bit_off;
2264 	return cluster;
2265 }
2266 
2267 /* given a cluster offset, calculate which block group it belongs to
2268  * and return that block offset. */
2269 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2270 {
2271 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2272 	u32 group_no;
2273 
2274 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2275 
2276 	group_no = cluster / osb->bitmap_cpg;
2277 	if (!group_no)
2278 		return osb->first_cluster_group_blkno;
2279 	return ocfs2_clusters_to_blocks(inode->i_sb,
2280 					group_no * osb->bitmap_cpg);
2281 }
2282 
2283 /* given the block number of a cluster start, calculate which cluster
2284  * group and descriptor bitmap offset that corresponds to. */
2285 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2286 						u64 data_blkno,
2287 						u64 *bg_blkno,
2288 						u16 *bg_bit_off)
2289 {
2290 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2291 	u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2292 
2293 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2294 
2295 	*bg_blkno = ocfs2_which_cluster_group(inode,
2296 					      data_cluster);
2297 
2298 	if (*bg_blkno == osb->first_cluster_group_blkno)
2299 		*bg_bit_off = (u16) data_cluster;
2300 	else
2301 		*bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2302 							     data_blkno - *bg_blkno);
2303 }
2304 
2305 /*
2306  * min_bits - minimum contiguous chunk from this total allocation we
2307  * can handle. set to what we asked for originally for a full
2308  * contig. allocation, set to '1' to indicate we can deal with extents
2309  * of any size.
2310  */
2311 int __ocfs2_claim_clusters(handle_t *handle,
2312 			   struct ocfs2_alloc_context *ac,
2313 			   u32 min_clusters,
2314 			   u32 max_clusters,
2315 			   u32 *cluster_start,
2316 			   u32 *num_clusters)
2317 {
2318 	int status;
2319 	unsigned int bits_wanted = max_clusters;
2320 	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2321 	struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2322 
2323 	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2324 
2325 	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2326 	       && ac->ac_which != OCFS2_AC_USE_MAIN);
2327 
2328 	if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2329 		WARN_ON(min_clusters > 1);
2330 
2331 		status = ocfs2_claim_local_alloc_bits(osb,
2332 						      handle,
2333 						      ac,
2334 						      bits_wanted,
2335 						      cluster_start,
2336 						      num_clusters);
2337 		if (!status)
2338 			atomic_inc(&osb->alloc_stats.local_data);
2339 	} else {
2340 		if (min_clusters > (osb->bitmap_cpg - 1)) {
2341 			/* The only paths asking for contiguousness
2342 			 * should know about this already. */
2343 			mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2344 			     "group bitmap size %u!\n", min_clusters,
2345 			     osb->bitmap_cpg);
2346 			status = -ENOSPC;
2347 			goto bail;
2348 		}
2349 		/* clamp the current request down to a realistic size. */
2350 		if (bits_wanted > (osb->bitmap_cpg - 1))
2351 			bits_wanted = osb->bitmap_cpg - 1;
2352 
2353 		status = ocfs2_claim_suballoc_bits(ac,
2354 						   handle,
2355 						   bits_wanted,
2356 						   min_clusters,
2357 						   &res);
2358 		if (!status) {
2359 			BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2360 			*cluster_start =
2361 				ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2362 								 res.sr_bg_blkno,
2363 								 res.sr_bit_offset);
2364 			atomic_inc(&osb->alloc_stats.bitmap_data);
2365 			*num_clusters = res.sr_bits;
2366 		}
2367 	}
2368 	if (status < 0) {
2369 		if (status != -ENOSPC)
2370 			mlog_errno(status);
2371 		goto bail;
2372 	}
2373 
2374 	ac->ac_bits_given += *num_clusters;
2375 
2376 bail:
2377 	if (status)
2378 		mlog_errno(status);
2379 	return status;
2380 }
2381 
2382 int ocfs2_claim_clusters(handle_t *handle,
2383 			 struct ocfs2_alloc_context *ac,
2384 			 u32 min_clusters,
2385 			 u32 *cluster_start,
2386 			 u32 *num_clusters)
2387 {
2388 	unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2389 
2390 	return __ocfs2_claim_clusters(handle, ac, min_clusters,
2391 				      bits_wanted, cluster_start, num_clusters);
2392 }
2393 
2394 static int ocfs2_block_group_clear_bits(handle_t *handle,
2395 					struct inode *alloc_inode,
2396 					struct ocfs2_group_desc *bg,
2397 					struct buffer_head *group_bh,
2398 					unsigned int bit_off,
2399 					unsigned int num_bits,
2400 					void (*undo_fn)(unsigned int bit,
2401 							unsigned long *bmap))
2402 {
2403 	int status;
2404 	unsigned int tmp;
2405 	struct ocfs2_group_desc *undo_bg = NULL;
2406 
2407 	/* The caller got this descriptor from
2408 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
2409 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2410 
2411 	trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2412 
2413 	BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2414 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2415 					 group_bh,
2416 					 undo_fn ?
2417 					 OCFS2_JOURNAL_ACCESS_UNDO :
2418 					 OCFS2_JOURNAL_ACCESS_WRITE);
2419 	if (status < 0) {
2420 		mlog_errno(status);
2421 		goto bail;
2422 	}
2423 
2424 	if (undo_fn) {
2425 		jbd_lock_bh_state(group_bh);
2426 		undo_bg = (struct ocfs2_group_desc *)
2427 					bh2jh(group_bh)->b_committed_data;
2428 		BUG_ON(!undo_bg);
2429 	}
2430 
2431 	tmp = num_bits;
2432 	while(tmp--) {
2433 		ocfs2_clear_bit((bit_off + tmp),
2434 				(unsigned long *) bg->bg_bitmap);
2435 		if (undo_fn)
2436 			undo_fn(bit_off + tmp,
2437 				(unsigned long *) undo_bg->bg_bitmap);
2438 	}
2439 	le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2440 	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2441 		if (undo_fn)
2442 			jbd_unlock_bh_state(group_bh);
2443 		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
2444 				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
2445 				   le16_to_cpu(bg->bg_bits),
2446 				   le16_to_cpu(bg->bg_free_bits_count),
2447 				   num_bits);
2448 	}
2449 
2450 	if (undo_fn)
2451 		jbd_unlock_bh_state(group_bh);
2452 
2453 	ocfs2_journal_dirty(handle, group_bh);
2454 bail:
2455 	return status;
2456 }
2457 
2458 /*
2459  * expects the suballoc inode to already be locked.
2460  */
2461 static int _ocfs2_free_suballoc_bits(handle_t *handle,
2462 				     struct inode *alloc_inode,
2463 				     struct buffer_head *alloc_bh,
2464 				     unsigned int start_bit,
2465 				     u64 bg_blkno,
2466 				     unsigned int count,
2467 				     void (*undo_fn)(unsigned int bit,
2468 						     unsigned long *bitmap))
2469 {
2470 	int status = 0;
2471 	u32 tmp_used;
2472 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2473 	struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2474 	struct buffer_head *group_bh = NULL;
2475 	struct ocfs2_group_desc *group;
2476 
2477 	/* The alloc_bh comes from ocfs2_free_dinode() or
2478 	 * ocfs2_free_clusters().  The callers have all locked the
2479 	 * allocator and gotten alloc_bh from the lock call.  This
2480 	 * validates the dinode buffer.  Any corruption that has happened
2481 	 * is a code bug. */
2482 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2483 	BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2484 
2485 	trace_ocfs2_free_suballoc_bits(
2486 		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2487 		(unsigned long long)bg_blkno,
2488 		start_bit, count);
2489 
2490 	status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2491 					     &group_bh);
2492 	if (status < 0) {
2493 		mlog_errno(status);
2494 		goto bail;
2495 	}
2496 	group = (struct ocfs2_group_desc *) group_bh->b_data;
2497 
2498 	BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2499 
2500 	status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2501 					      group, group_bh,
2502 					      start_bit, count, undo_fn);
2503 	if (status < 0) {
2504 		mlog_errno(status);
2505 		goto bail;
2506 	}
2507 
2508 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2509 					 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2510 	if (status < 0) {
2511 		mlog_errno(status);
2512 		ocfs2_block_group_set_bits(handle, alloc_inode, group, group_bh,
2513 				start_bit, count);
2514 		goto bail;
2515 	}
2516 
2517 	le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2518 		     count);
2519 	tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2520 	fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2521 	ocfs2_journal_dirty(handle, alloc_bh);
2522 
2523 bail:
2524 	brelse(group_bh);
2525 
2526 	if (status)
2527 		mlog_errno(status);
2528 	return status;
2529 }
2530 
2531 int ocfs2_free_suballoc_bits(handle_t *handle,
2532 			     struct inode *alloc_inode,
2533 			     struct buffer_head *alloc_bh,
2534 			     unsigned int start_bit,
2535 			     u64 bg_blkno,
2536 			     unsigned int count)
2537 {
2538 	return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2539 					 start_bit, bg_blkno, count, NULL);
2540 }
2541 
2542 int ocfs2_free_dinode(handle_t *handle,
2543 		      struct inode *inode_alloc_inode,
2544 		      struct buffer_head *inode_alloc_bh,
2545 		      struct ocfs2_dinode *di)
2546 {
2547 	u64 blk = le64_to_cpu(di->i_blkno);
2548 	u16 bit = le16_to_cpu(di->i_suballoc_bit);
2549 	u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2550 
2551 	if (di->i_suballoc_loc)
2552 		bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2553 	return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2554 					inode_alloc_bh, bit, bg_blkno, 1);
2555 }
2556 
2557 static int _ocfs2_free_clusters(handle_t *handle,
2558 				struct inode *bitmap_inode,
2559 				struct buffer_head *bitmap_bh,
2560 				u64 start_blk,
2561 				unsigned int num_clusters,
2562 				void (*undo_fn)(unsigned int bit,
2563 						unsigned long *bitmap))
2564 {
2565 	int status;
2566 	u16 bg_start_bit;
2567 	u64 bg_blkno;
2568 
2569 	/* You can't ever have a contiguous set of clusters
2570 	 * bigger than a block group bitmap so we never have to worry
2571 	 * about looping on them.
2572 	 * This is expensive. We can safely remove once this stuff has
2573 	 * gotten tested really well. */
2574 	BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb,
2575 				ocfs2_blocks_to_clusters(bitmap_inode->i_sb,
2576 							 start_blk)));
2577 
2578 
2579 	ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2580 				     &bg_start_bit);
2581 
2582 	trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2583 			(unsigned long long)start_blk,
2584 			bg_start_bit, num_clusters);
2585 
2586 	status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2587 					   bg_start_bit, bg_blkno,
2588 					   num_clusters, undo_fn);
2589 	if (status < 0) {
2590 		mlog_errno(status);
2591 		goto out;
2592 	}
2593 
2594 	ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2595 					 num_clusters);
2596 
2597 out:
2598 	if (status)
2599 		mlog_errno(status);
2600 	return status;
2601 }
2602 
2603 int ocfs2_free_clusters(handle_t *handle,
2604 			struct inode *bitmap_inode,
2605 			struct buffer_head *bitmap_bh,
2606 			u64 start_blk,
2607 			unsigned int num_clusters)
2608 {
2609 	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2610 				    start_blk, num_clusters,
2611 				    _ocfs2_set_bit);
2612 }
2613 
2614 /*
2615  * Give never-used clusters back to the global bitmap.  We don't need
2616  * to protect these bits in the undo buffer.
2617  */
2618 int ocfs2_release_clusters(handle_t *handle,
2619 			   struct inode *bitmap_inode,
2620 			   struct buffer_head *bitmap_bh,
2621 			   u64 start_blk,
2622 			   unsigned int num_clusters)
2623 {
2624 	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2625 				    start_blk, num_clusters,
2626 				    _ocfs2_clear_bit);
2627 }
2628 
2629 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
2630 {
2631 	printk("Block Group:\n");
2632 	printk("bg_signature:       %s\n", bg->bg_signature);
2633 	printk("bg_size:            %u\n", bg->bg_size);
2634 	printk("bg_bits:            %u\n", bg->bg_bits);
2635 	printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
2636 	printk("bg_chain:           %u\n", bg->bg_chain);
2637 	printk("bg_generation:      %u\n", le32_to_cpu(bg->bg_generation));
2638 	printk("bg_next_group:      %llu\n",
2639 	       (unsigned long long)bg->bg_next_group);
2640 	printk("bg_parent_dinode:   %llu\n",
2641 	       (unsigned long long)bg->bg_parent_dinode);
2642 	printk("bg_blkno:           %llu\n",
2643 	       (unsigned long long)bg->bg_blkno);
2644 }
2645 
2646 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2647 {
2648 	int i;
2649 
2650 	printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
2651 	printk("i_signature:                  %s\n", fe->i_signature);
2652 	printk("i_size:                       %llu\n",
2653 	       (unsigned long long)fe->i_size);
2654 	printk("i_clusters:                   %u\n", fe->i_clusters);
2655 	printk("i_generation:                 %u\n",
2656 	       le32_to_cpu(fe->i_generation));
2657 	printk("id1.bitmap1.i_used:           %u\n",
2658 	       le32_to_cpu(fe->id1.bitmap1.i_used));
2659 	printk("id1.bitmap1.i_total:          %u\n",
2660 	       le32_to_cpu(fe->id1.bitmap1.i_total));
2661 	printk("id2.i_chain.cl_cpg:           %u\n", fe->id2.i_chain.cl_cpg);
2662 	printk("id2.i_chain.cl_bpc:           %u\n", fe->id2.i_chain.cl_bpc);
2663 	printk("id2.i_chain.cl_count:         %u\n", fe->id2.i_chain.cl_count);
2664 	printk("id2.i_chain.cl_next_free_rec: %u\n",
2665 	       fe->id2.i_chain.cl_next_free_rec);
2666 	for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2667 		printk("fe->id2.i_chain.cl_recs[%d].c_free:  %u\n", i,
2668 		       fe->id2.i_chain.cl_recs[i].c_free);
2669 		printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2670 		       fe->id2.i_chain.cl_recs[i].c_total);
2671 		printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2672 		       (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
2673 	}
2674 }
2675 
2676 /*
2677  * For a given allocation, determine which allocators will need to be
2678  * accessed, and lock them, reserving the appropriate number of bits.
2679  *
2680  * Sparse file systems call this from ocfs2_write_begin_nolock()
2681  * and ocfs2_allocate_unwritten_extents().
2682  *
2683  * File systems which don't support holes call this from
2684  * ocfs2_extend_allocation().
2685  */
2686 int ocfs2_lock_allocators(struct inode *inode,
2687 			  struct ocfs2_extent_tree *et,
2688 			  u32 clusters_to_add, u32 extents_to_split,
2689 			  struct ocfs2_alloc_context **data_ac,
2690 			  struct ocfs2_alloc_context **meta_ac)
2691 {
2692 	int ret = 0, num_free_extents;
2693 	unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2694 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2695 
2696 	*meta_ac = NULL;
2697 	if (data_ac)
2698 		*data_ac = NULL;
2699 
2700 	BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2701 
2702 	num_free_extents = ocfs2_num_free_extents(et);
2703 	if (num_free_extents < 0) {
2704 		ret = num_free_extents;
2705 		mlog_errno(ret);
2706 		goto out;
2707 	}
2708 
2709 	/*
2710 	 * Sparse allocation file systems need to be more conservative
2711 	 * with reserving room for expansion - the actual allocation
2712 	 * happens while we've got a journal handle open so re-taking
2713 	 * a cluster lock (because we ran out of room for another
2714 	 * extent) will violate ordering rules.
2715 	 *
2716 	 * Most of the time we'll only be seeing this 1 cluster at a time
2717 	 * anyway.
2718 	 *
2719 	 * Always lock for any unwritten extents - we might want to
2720 	 * add blocks during a split.
2721 	 */
2722 	if (!num_free_extents ||
2723 	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2724 		ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2725 		if (ret < 0) {
2726 			if (ret != -ENOSPC)
2727 				mlog_errno(ret);
2728 			goto out;
2729 		}
2730 	}
2731 
2732 	if (clusters_to_add == 0)
2733 		goto out;
2734 
2735 	ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2736 	if (ret < 0) {
2737 		if (ret != -ENOSPC)
2738 			mlog_errno(ret);
2739 		goto out;
2740 	}
2741 
2742 out:
2743 	if (ret) {
2744 		if (*meta_ac) {
2745 			ocfs2_free_alloc_context(*meta_ac);
2746 			*meta_ac = NULL;
2747 		}
2748 
2749 		/*
2750 		 * We cannot have an error and a non null *data_ac.
2751 		 */
2752 	}
2753 
2754 	return ret;
2755 }
2756 
2757 /*
2758  * Read the inode specified by blkno to get suballoc_slot and
2759  * suballoc_bit.
2760  */
2761 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2762 				       u16 *suballoc_slot, u64 *group_blkno,
2763 				       u16 *suballoc_bit)
2764 {
2765 	int status;
2766 	struct buffer_head *inode_bh = NULL;
2767 	struct ocfs2_dinode *inode_fe;
2768 
2769 	trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2770 
2771 	/* dirty read disk */
2772 	status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2773 	if (status < 0) {
2774 		mlog(ML_ERROR, "read block %llu failed %d\n",
2775 		     (unsigned long long)blkno, status);
2776 		goto bail;
2777 	}
2778 
2779 	inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2780 	if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2781 		mlog(ML_ERROR, "invalid inode %llu requested\n",
2782 		     (unsigned long long)blkno);
2783 		status = -EINVAL;
2784 		goto bail;
2785 	}
2786 
2787 	if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2788 	    (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2789 		mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2790 		     (unsigned long long)blkno,
2791 		     (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2792 		status = -EINVAL;
2793 		goto bail;
2794 	}
2795 
2796 	if (suballoc_slot)
2797 		*suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2798 	if (suballoc_bit)
2799 		*suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2800 	if (group_blkno)
2801 		*group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2802 
2803 bail:
2804 	brelse(inode_bh);
2805 
2806 	if (status)
2807 		mlog_errno(status);
2808 	return status;
2809 }
2810 
2811 /*
2812  * test whether bit is SET in allocator bitmap or not.  on success, 0
2813  * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
2814  * is returned and *res is meaningless.  Call this after you have
2815  * cluster locked against suballoc, or you may get a result based on
2816  * non-up2date contents
2817  */
2818 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2819 				   struct inode *suballoc,
2820 				   struct buffer_head *alloc_bh,
2821 				   u64 group_blkno, u64 blkno,
2822 				   u16 bit, int *res)
2823 {
2824 	struct ocfs2_dinode *alloc_di;
2825 	struct ocfs2_group_desc *group;
2826 	struct buffer_head *group_bh = NULL;
2827 	u64 bg_blkno;
2828 	int status;
2829 
2830 	trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2831 				      (unsigned int)bit);
2832 
2833 	alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2834 	if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2835 		mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2836 		     (unsigned int)bit,
2837 		     ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2838 		status = -EINVAL;
2839 		goto bail;
2840 	}
2841 
2842 	bg_blkno = group_blkno ? group_blkno :
2843 		   ocfs2_which_suballoc_group(blkno, bit);
2844 	status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2845 					     &group_bh);
2846 	if (status < 0) {
2847 		mlog(ML_ERROR, "read group %llu failed %d\n",
2848 		     (unsigned long long)bg_blkno, status);
2849 		goto bail;
2850 	}
2851 
2852 	group = (struct ocfs2_group_desc *) group_bh->b_data;
2853 	*res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2854 
2855 bail:
2856 	brelse(group_bh);
2857 
2858 	if (status)
2859 		mlog_errno(status);
2860 	return status;
2861 }
2862 
2863 /*
2864  * Test if the bit representing this inode (blkno) is set in the
2865  * suballocator.
2866  *
2867  * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2868  *
2869  * In the event of failure, a negative value is returned and *res is
2870  * meaningless.
2871  *
2872  * Callers must make sure to hold nfs_sync_lock to prevent
2873  * ocfs2_delete_inode() on another node from accessing the same
2874  * suballocator concurrently.
2875  */
2876 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2877 {
2878 	int status;
2879 	u64 group_blkno = 0;
2880 	u16 suballoc_bit = 0, suballoc_slot = 0;
2881 	struct inode *inode_alloc_inode;
2882 	struct buffer_head *alloc_bh = NULL;
2883 
2884 	trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2885 
2886 	status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2887 					     &group_blkno, &suballoc_bit);
2888 	if (status < 0) {
2889 		mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2890 		goto bail;
2891 	}
2892 
2893 	inode_alloc_inode =
2894 		ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2895 					    suballoc_slot);
2896 	if (!inode_alloc_inode) {
2897 		/* the error code could be inaccurate, but we are not able to
2898 		 * get the correct one. */
2899 		status = -EINVAL;
2900 		mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2901 		     (u32)suballoc_slot);
2902 		goto bail;
2903 	}
2904 
2905 	inode_lock(inode_alloc_inode);
2906 	status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2907 	if (status < 0) {
2908 		inode_unlock(inode_alloc_inode);
2909 		iput(inode_alloc_inode);
2910 		mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2911 		     (u32)suballoc_slot, status);
2912 		goto bail;
2913 	}
2914 
2915 	status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2916 					 group_blkno, blkno, suballoc_bit, res);
2917 	if (status < 0)
2918 		mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2919 
2920 	ocfs2_inode_unlock(inode_alloc_inode, 0);
2921 	inode_unlock(inode_alloc_inode);
2922 
2923 	iput(inode_alloc_inode);
2924 	brelse(alloc_bh);
2925 bail:
2926 	if (status)
2927 		mlog_errno(status);
2928 	return status;
2929 }
2930