xref: /openbmc/linux/fs/ocfs2/suballoc.c (revision 6dfcd296)
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * suballoc.c
5  *
6  * metadata alloc and free
7  * Inspired by ext3 block groups.
8  *
9  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
10  *
11  * This program is free software; you can redistribute it and/or
12  * modify it under the terms of the GNU General Public
13  * License as published by the Free Software Foundation; either
14  * version 2 of the License, or (at your option) any later version.
15  *
16  * This program is distributed in the hope that it will be useful,
17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
19  * General Public License for more details.
20  *
21  * You should have received a copy of the GNU General Public
22  * License along with this program; if not, write to the
23  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
24  * Boston, MA 021110-1307, USA.
25  */
26 
27 #include <linux/fs.h>
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
31 
32 #include <cluster/masklog.h>
33 
34 #include "ocfs2.h"
35 
36 #include "alloc.h"
37 #include "blockcheck.h"
38 #include "dlmglue.h"
39 #include "inode.h"
40 #include "journal.h"
41 #include "localalloc.h"
42 #include "suballoc.h"
43 #include "super.h"
44 #include "sysfile.h"
45 #include "uptodate.h"
46 #include "ocfs2_trace.h"
47 
48 #include "buffer_head_io.h"
49 
50 #define NOT_ALLOC_NEW_GROUP		0
51 #define ALLOC_NEW_GROUP			0x1
52 #define ALLOC_GROUPS_FROM_GLOBAL	0x2
53 
54 #define OCFS2_MAX_TO_STEAL		1024
55 
56 struct ocfs2_suballoc_result {
57 	u64		sr_bg_blkno;	/* The bg we allocated from.  Set
58 					   to 0 when a block group is
59 					   contiguous. */
60 	u64		sr_bg_stable_blkno; /*
61 					     * Doesn't change, always
62 					     * set to target block
63 					     * group descriptor
64 					     * block.
65 					     */
66 	u64		sr_blkno;	/* The first allocated block */
67 	unsigned int	sr_bit_offset;	/* The bit in the bg */
68 	unsigned int	sr_bits;	/* How many bits we claimed */
69 };
70 
71 static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
72 {
73 	if (res->sr_blkno == 0)
74 		return 0;
75 
76 	if (res->sr_bg_blkno)
77 		return res->sr_bg_blkno;
78 
79 	return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset);
80 }
81 
82 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg);
83 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe);
84 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
85 static int ocfs2_block_group_fill(handle_t *handle,
86 				  struct inode *alloc_inode,
87 				  struct buffer_head *bg_bh,
88 				  u64 group_blkno,
89 				  unsigned int group_clusters,
90 				  u16 my_chain,
91 				  struct ocfs2_chain_list *cl);
92 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
93 				   struct inode *alloc_inode,
94 				   struct buffer_head *bh,
95 				   u64 max_block,
96 				   u64 *last_alloc_group,
97 				   int flags);
98 
99 static int ocfs2_cluster_group_search(struct inode *inode,
100 				      struct buffer_head *group_bh,
101 				      u32 bits_wanted, u32 min_bits,
102 				      u64 max_block,
103 				      struct ocfs2_suballoc_result *res);
104 static int ocfs2_block_group_search(struct inode *inode,
105 				    struct buffer_head *group_bh,
106 				    u32 bits_wanted, u32 min_bits,
107 				    u64 max_block,
108 				    struct ocfs2_suballoc_result *res);
109 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
110 				     handle_t *handle,
111 				     u32 bits_wanted,
112 				     u32 min_bits,
113 				     struct ocfs2_suballoc_result *res);
114 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
115 					 int nr);
116 static int ocfs2_relink_block_group(handle_t *handle,
117 				    struct inode *alloc_inode,
118 				    struct buffer_head *fe_bh,
119 				    struct buffer_head *bg_bh,
120 				    struct buffer_head *prev_bg_bh,
121 				    u16 chain);
122 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
123 						     u32 wanted);
124 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
125 						   u64 bg_blkno,
126 						   u16 bg_bit_off);
127 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
128 						u64 data_blkno,
129 						u64 *bg_blkno,
130 						u16 *bg_bit_off);
131 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
132 					     u32 bits_wanted, u64 max_block,
133 					     int flags,
134 					     struct ocfs2_alloc_context **ac);
135 
136 void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
137 {
138 	struct inode *inode = ac->ac_inode;
139 
140 	if (inode) {
141 		if (ac->ac_which != OCFS2_AC_USE_LOCAL)
142 			ocfs2_inode_unlock(inode, 1);
143 
144 		inode_unlock(inode);
145 
146 		iput(inode);
147 		ac->ac_inode = NULL;
148 	}
149 	brelse(ac->ac_bh);
150 	ac->ac_bh = NULL;
151 	ac->ac_resv = NULL;
152 	kfree(ac->ac_find_loc_priv);
153 	ac->ac_find_loc_priv = NULL;
154 }
155 
156 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
157 {
158 	ocfs2_free_ac_resource(ac);
159 	kfree(ac);
160 }
161 
162 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
163 {
164 	return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
165 }
166 
167 #define do_error(fmt, ...)						\
168 do {									\
169 	if (resize)							\
170 		mlog(ML_ERROR, fmt, ##__VA_ARGS__);			\
171 	else								\
172 		return ocfs2_error(sb, fmt, ##__VA_ARGS__);		\
173 } while (0)
174 
175 static int ocfs2_validate_gd_self(struct super_block *sb,
176 				  struct buffer_head *bh,
177 				  int resize)
178 {
179 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
180 
181 	if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
182 		do_error("Group descriptor #%llu has bad signature %.*s\n",
183 			 (unsigned long long)bh->b_blocknr, 7,
184 			 gd->bg_signature);
185 	}
186 
187 	if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
188 		do_error("Group descriptor #%llu has an invalid bg_blkno of %llu\n",
189 			 (unsigned long long)bh->b_blocknr,
190 			 (unsigned long long)le64_to_cpu(gd->bg_blkno));
191 	}
192 
193 	if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
194 		do_error("Group descriptor #%llu has an invalid fs_generation of #%u\n",
195 			 (unsigned long long)bh->b_blocknr,
196 			 le32_to_cpu(gd->bg_generation));
197 	}
198 
199 	if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
200 		do_error("Group descriptor #%llu has bit count %u but claims that %u are free\n",
201 			 (unsigned long long)bh->b_blocknr,
202 			 le16_to_cpu(gd->bg_bits),
203 			 le16_to_cpu(gd->bg_free_bits_count));
204 	}
205 
206 	if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
207 		do_error("Group descriptor #%llu has bit count %u but max bitmap bits of %u\n",
208 			 (unsigned long long)bh->b_blocknr,
209 			 le16_to_cpu(gd->bg_bits),
210 			 8 * le16_to_cpu(gd->bg_size));
211 	}
212 
213 	return 0;
214 }
215 
216 static int ocfs2_validate_gd_parent(struct super_block *sb,
217 				    struct ocfs2_dinode *di,
218 				    struct buffer_head *bh,
219 				    int resize)
220 {
221 	unsigned int max_bits;
222 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
223 
224 	if (di->i_blkno != gd->bg_parent_dinode) {
225 		do_error("Group descriptor #%llu has bad parent pointer (%llu, expected %llu)\n",
226 			 (unsigned long long)bh->b_blocknr,
227 			 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
228 			 (unsigned long long)le64_to_cpu(di->i_blkno));
229 	}
230 
231 	max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
232 	if (le16_to_cpu(gd->bg_bits) > max_bits) {
233 		do_error("Group descriptor #%llu has bit count of %u\n",
234 			 (unsigned long long)bh->b_blocknr,
235 			 le16_to_cpu(gd->bg_bits));
236 	}
237 
238 	/* In resize, we may meet the case bg_chain == cl_next_free_rec. */
239 	if ((le16_to_cpu(gd->bg_chain) >
240 	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
241 	    ((le16_to_cpu(gd->bg_chain) ==
242 	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
243 		do_error("Group descriptor #%llu has bad chain %u\n",
244 			 (unsigned long long)bh->b_blocknr,
245 			 le16_to_cpu(gd->bg_chain));
246 	}
247 
248 	return 0;
249 }
250 
251 #undef do_error
252 
253 /*
254  * This version only prints errors.  It does not fail the filesystem, and
255  * exists only for resize.
256  */
257 int ocfs2_check_group_descriptor(struct super_block *sb,
258 				 struct ocfs2_dinode *di,
259 				 struct buffer_head *bh)
260 {
261 	int rc;
262 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
263 
264 	BUG_ON(!buffer_uptodate(bh));
265 
266 	/*
267 	 * If the ecc fails, we return the error but otherwise
268 	 * leave the filesystem running.  We know any error is
269 	 * local to this block.
270 	 */
271 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
272 	if (rc) {
273 		mlog(ML_ERROR,
274 		     "Checksum failed for group descriptor %llu\n",
275 		     (unsigned long long)bh->b_blocknr);
276 	} else
277 		rc = ocfs2_validate_gd_self(sb, bh, 1);
278 	if (!rc)
279 		rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
280 
281 	return rc;
282 }
283 
284 static int ocfs2_validate_group_descriptor(struct super_block *sb,
285 					   struct buffer_head *bh)
286 {
287 	int rc;
288 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
289 
290 	trace_ocfs2_validate_group_descriptor(
291 					(unsigned long long)bh->b_blocknr);
292 
293 	BUG_ON(!buffer_uptodate(bh));
294 
295 	/*
296 	 * If the ecc fails, we return the error but otherwise
297 	 * leave the filesystem running.  We know any error is
298 	 * local to this block.
299 	 */
300 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
301 	if (rc)
302 		return rc;
303 
304 	/*
305 	 * Errors after here are fatal.
306 	 */
307 
308 	return ocfs2_validate_gd_self(sb, bh, 0);
309 }
310 
311 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
312 				u64 gd_blkno, struct buffer_head **bh)
313 {
314 	int rc;
315 	struct buffer_head *tmp = *bh;
316 
317 	rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
318 			      ocfs2_validate_group_descriptor);
319 	if (rc)
320 		goto out;
321 
322 	rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
323 	if (rc) {
324 		brelse(tmp);
325 		goto out;
326 	}
327 
328 	/* If ocfs2_read_block() got us a new bh, pass it up. */
329 	if (!*bh)
330 		*bh = tmp;
331 
332 out:
333 	return rc;
334 }
335 
336 static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
337 					  struct ocfs2_group_desc *bg,
338 					  struct ocfs2_chain_list *cl,
339 					  u64 p_blkno, unsigned int clusters)
340 {
341 	struct ocfs2_extent_list *el = &bg->bg_list;
342 	struct ocfs2_extent_rec *rec;
343 
344 	BUG_ON(!ocfs2_supports_discontig_bg(osb));
345 	if (!el->l_next_free_rec)
346 		el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
347 	rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
348 	rec->e_blkno = cpu_to_le64(p_blkno);
349 	rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
350 				  le16_to_cpu(cl->cl_bpc));
351 	rec->e_leaf_clusters = cpu_to_le16(clusters);
352 	le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
353 	le16_add_cpu(&bg->bg_free_bits_count,
354 		     clusters * le16_to_cpu(cl->cl_bpc));
355 	le16_add_cpu(&el->l_next_free_rec, 1);
356 }
357 
358 static int ocfs2_block_group_fill(handle_t *handle,
359 				  struct inode *alloc_inode,
360 				  struct buffer_head *bg_bh,
361 				  u64 group_blkno,
362 				  unsigned int group_clusters,
363 				  u16 my_chain,
364 				  struct ocfs2_chain_list *cl)
365 {
366 	int status = 0;
367 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
368 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
369 	struct super_block * sb = alloc_inode->i_sb;
370 
371 	if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
372 		status = ocfs2_error(alloc_inode->i_sb,
373 				     "group block (%llu) != b_blocknr (%llu)\n",
374 				     (unsigned long long)group_blkno,
375 				     (unsigned long long) bg_bh->b_blocknr);
376 		goto bail;
377 	}
378 
379 	status = ocfs2_journal_access_gd(handle,
380 					 INODE_CACHE(alloc_inode),
381 					 bg_bh,
382 					 OCFS2_JOURNAL_ACCESS_CREATE);
383 	if (status < 0) {
384 		mlog_errno(status);
385 		goto bail;
386 	}
387 
388 	memset(bg, 0, sb->s_blocksize);
389 	strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
390 	bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
391 	bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
392 						osb->s_feature_incompat));
393 	bg->bg_chain = cpu_to_le16(my_chain);
394 	bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
395 	bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
396 	bg->bg_blkno = cpu_to_le64(group_blkno);
397 	if (group_clusters == le16_to_cpu(cl->cl_cpg))
398 		bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
399 	else
400 		ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
401 					      group_clusters);
402 
403 	/* set the 1st bit in the bitmap to account for the descriptor block */
404 	ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
405 	bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
406 
407 	ocfs2_journal_dirty(handle, bg_bh);
408 
409 	/* There is no need to zero out or otherwise initialize the
410 	 * other blocks in a group - All valid FS metadata in a block
411 	 * group stores the superblock fs_generation value at
412 	 * allocation time. */
413 
414 bail:
415 	if (status)
416 		mlog_errno(status);
417 	return status;
418 }
419 
420 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
421 {
422 	u16 curr, best;
423 
424 	best = curr = 0;
425 	while (curr < le16_to_cpu(cl->cl_count)) {
426 		if (le32_to_cpu(cl->cl_recs[best].c_total) >
427 		    le32_to_cpu(cl->cl_recs[curr].c_total))
428 			best = curr;
429 		curr++;
430 	}
431 	return best;
432 }
433 
434 static struct buffer_head *
435 ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
436 			       struct inode *alloc_inode,
437 			       struct ocfs2_alloc_context *ac,
438 			       struct ocfs2_chain_list *cl)
439 {
440 	int status;
441 	u32 bit_off, num_bits;
442 	u64 bg_blkno;
443 	struct buffer_head *bg_bh;
444 	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
445 
446 	status = ocfs2_claim_clusters(handle, ac,
447 				      le16_to_cpu(cl->cl_cpg), &bit_off,
448 				      &num_bits);
449 	if (status < 0) {
450 		if (status != -ENOSPC)
451 			mlog_errno(status);
452 		goto bail;
453 	}
454 
455 	/* setup the group */
456 	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
457 	trace_ocfs2_block_group_alloc_contig(
458 	     (unsigned long long)bg_blkno, alloc_rec);
459 
460 	bg_bh = sb_getblk(osb->sb, bg_blkno);
461 	if (!bg_bh) {
462 		status = -ENOMEM;
463 		mlog_errno(status);
464 		goto bail;
465 	}
466 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
467 
468 	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
469 					bg_blkno, num_bits, alloc_rec, cl);
470 	if (status < 0) {
471 		brelse(bg_bh);
472 		mlog_errno(status);
473 	}
474 
475 bail:
476 	return status ? ERR_PTR(status) : bg_bh;
477 }
478 
479 static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
480 					handle_t *handle,
481 					struct ocfs2_alloc_context *ac,
482 					unsigned int min_bits,
483 					u32 *bit_off, u32 *num_bits)
484 {
485 	int status = 0;
486 
487 	while (min_bits) {
488 		status = ocfs2_claim_clusters(handle, ac, min_bits,
489 					      bit_off, num_bits);
490 		if (status != -ENOSPC)
491 			break;
492 
493 		min_bits >>= 1;
494 	}
495 
496 	return status;
497 }
498 
499 static int ocfs2_block_group_grow_discontig(handle_t *handle,
500 					    struct inode *alloc_inode,
501 					    struct buffer_head *bg_bh,
502 					    struct ocfs2_alloc_context *ac,
503 					    struct ocfs2_chain_list *cl,
504 					    unsigned int min_bits)
505 {
506 	int status;
507 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
508 	struct ocfs2_group_desc *bg =
509 		(struct ocfs2_group_desc *)bg_bh->b_data;
510 	unsigned int needed = le16_to_cpu(cl->cl_cpg) -
511 			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
512 	u32 p_cpos, clusters;
513 	u64 p_blkno;
514 	struct ocfs2_extent_list *el = &bg->bg_list;
515 
516 	status = ocfs2_journal_access_gd(handle,
517 					 INODE_CACHE(alloc_inode),
518 					 bg_bh,
519 					 OCFS2_JOURNAL_ACCESS_CREATE);
520 	if (status < 0) {
521 		mlog_errno(status);
522 		goto bail;
523 	}
524 
525 	while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
526 				le16_to_cpu(el->l_count))) {
527 		if (min_bits > needed)
528 			min_bits = needed;
529 		status = ocfs2_block_group_claim_bits(osb, handle, ac,
530 						      min_bits, &p_cpos,
531 						      &clusters);
532 		if (status < 0) {
533 			if (status != -ENOSPC)
534 				mlog_errno(status);
535 			goto bail;
536 		}
537 		p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
538 		ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
539 					      clusters);
540 
541 		min_bits = clusters;
542 		needed = le16_to_cpu(cl->cl_cpg) -
543 			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
544 	}
545 
546 	if (needed > 0) {
547 		/*
548 		 * We have used up all the extent rec but can't fill up
549 		 * the cpg. So bail out.
550 		 */
551 		status = -ENOSPC;
552 		goto bail;
553 	}
554 
555 	ocfs2_journal_dirty(handle, bg_bh);
556 
557 bail:
558 	return status;
559 }
560 
561 static void ocfs2_bg_alloc_cleanup(handle_t *handle,
562 				   struct ocfs2_alloc_context *cluster_ac,
563 				   struct inode *alloc_inode,
564 				   struct buffer_head *bg_bh)
565 {
566 	int i, ret;
567 	struct ocfs2_group_desc *bg;
568 	struct ocfs2_extent_list *el;
569 	struct ocfs2_extent_rec *rec;
570 
571 	if (!bg_bh)
572 		return;
573 
574 	bg = (struct ocfs2_group_desc *)bg_bh->b_data;
575 	el = &bg->bg_list;
576 	for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
577 		rec = &el->l_recs[i];
578 		ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
579 					  cluster_ac->ac_bh,
580 					  le64_to_cpu(rec->e_blkno),
581 					  le16_to_cpu(rec->e_leaf_clusters));
582 		if (ret)
583 			mlog_errno(ret);
584 		/* Try all the clusters to free */
585 	}
586 
587 	ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
588 	brelse(bg_bh);
589 }
590 
591 static struct buffer_head *
592 ocfs2_block_group_alloc_discontig(handle_t *handle,
593 				  struct inode *alloc_inode,
594 				  struct ocfs2_alloc_context *ac,
595 				  struct ocfs2_chain_list *cl)
596 {
597 	int status;
598 	u32 bit_off, num_bits;
599 	u64 bg_blkno;
600 	unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
601 	struct buffer_head *bg_bh = NULL;
602 	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
603 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
604 
605 	if (!ocfs2_supports_discontig_bg(osb)) {
606 		status = -ENOSPC;
607 		goto bail;
608 	}
609 
610 	status = ocfs2_extend_trans(handle,
611 				    ocfs2_calc_bg_discontig_credits(osb->sb));
612 	if (status) {
613 		mlog_errno(status);
614 		goto bail;
615 	}
616 
617 	/*
618 	 * We're going to be grabbing from multiple cluster groups.
619 	 * We don't have enough credits to relink them all, and the
620 	 * cluster groups will be staying in cache for the duration of
621 	 * this operation.
622 	 */
623 	ac->ac_disable_chain_relink = 1;
624 
625 	/* Claim the first region */
626 	status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
627 					      &bit_off, &num_bits);
628 	if (status < 0) {
629 		if (status != -ENOSPC)
630 			mlog_errno(status);
631 		goto bail;
632 	}
633 	min_bits = num_bits;
634 
635 	/* setup the group */
636 	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
637 	trace_ocfs2_block_group_alloc_discontig(
638 				(unsigned long long)bg_blkno, alloc_rec);
639 
640 	bg_bh = sb_getblk(osb->sb, bg_blkno);
641 	if (!bg_bh) {
642 		status = -ENOMEM;
643 		mlog_errno(status);
644 		goto bail;
645 	}
646 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
647 
648 	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
649 					bg_blkno, num_bits, alloc_rec, cl);
650 	if (status < 0) {
651 		mlog_errno(status);
652 		goto bail;
653 	}
654 
655 	status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
656 						  bg_bh, ac, cl, min_bits);
657 	if (status)
658 		mlog_errno(status);
659 
660 bail:
661 	if (status)
662 		ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
663 	return status ? ERR_PTR(status) : bg_bh;
664 }
665 
666 /*
667  * We expect the block group allocator to already be locked.
668  */
669 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
670 				   struct inode *alloc_inode,
671 				   struct buffer_head *bh,
672 				   u64 max_block,
673 				   u64 *last_alloc_group,
674 				   int flags)
675 {
676 	int status, credits;
677 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
678 	struct ocfs2_chain_list *cl;
679 	struct ocfs2_alloc_context *ac = NULL;
680 	handle_t *handle = NULL;
681 	u16 alloc_rec;
682 	struct buffer_head *bg_bh = NULL;
683 	struct ocfs2_group_desc *bg;
684 
685 	BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
686 
687 	cl = &fe->id2.i_chain;
688 	status = ocfs2_reserve_clusters_with_limit(osb,
689 						   le16_to_cpu(cl->cl_cpg),
690 						   max_block, flags, &ac);
691 	if (status < 0) {
692 		if (status != -ENOSPC)
693 			mlog_errno(status);
694 		goto bail;
695 	}
696 
697 	credits = ocfs2_calc_group_alloc_credits(osb->sb,
698 						 le16_to_cpu(cl->cl_cpg));
699 	handle = ocfs2_start_trans(osb, credits);
700 	if (IS_ERR(handle)) {
701 		status = PTR_ERR(handle);
702 		handle = NULL;
703 		mlog_errno(status);
704 		goto bail;
705 	}
706 
707 	if (last_alloc_group && *last_alloc_group != 0) {
708 		trace_ocfs2_block_group_alloc(
709 				(unsigned long long)*last_alloc_group);
710 		ac->ac_last_group = *last_alloc_group;
711 	}
712 
713 	bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
714 					       ac, cl);
715 	if (IS_ERR(bg_bh) && (PTR_ERR(bg_bh) == -ENOSPC))
716 		bg_bh = ocfs2_block_group_alloc_discontig(handle,
717 							  alloc_inode,
718 							  ac, cl);
719 	if (IS_ERR(bg_bh)) {
720 		status = PTR_ERR(bg_bh);
721 		bg_bh = NULL;
722 		if (status != -ENOSPC)
723 			mlog_errno(status);
724 		goto bail;
725 	}
726 	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
727 
728 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
729 					 bh, OCFS2_JOURNAL_ACCESS_WRITE);
730 	if (status < 0) {
731 		mlog_errno(status);
732 		goto bail;
733 	}
734 
735 	alloc_rec = le16_to_cpu(bg->bg_chain);
736 	le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
737 		     le16_to_cpu(bg->bg_free_bits_count));
738 	le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
739 		     le16_to_cpu(bg->bg_bits));
740 	cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
741 	if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
742 		le16_add_cpu(&cl->cl_next_free_rec, 1);
743 
744 	le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
745 					le16_to_cpu(bg->bg_free_bits_count));
746 	le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
747 	le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
748 
749 	ocfs2_journal_dirty(handle, bh);
750 
751 	spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
752 	OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
753 	fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
754 					     le32_to_cpu(fe->i_clusters)));
755 	spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
756 	i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
757 	alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
758 	ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
759 
760 	status = 0;
761 
762 	/* save the new last alloc group so that the caller can cache it. */
763 	if (last_alloc_group)
764 		*last_alloc_group = ac->ac_last_group;
765 
766 bail:
767 	if (handle)
768 		ocfs2_commit_trans(osb, handle);
769 
770 	if (ac)
771 		ocfs2_free_alloc_context(ac);
772 
773 	brelse(bg_bh);
774 
775 	if (status)
776 		mlog_errno(status);
777 	return status;
778 }
779 
780 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
781 				       struct ocfs2_alloc_context *ac,
782 				       int type,
783 				       u32 slot,
784 				       u64 *last_alloc_group,
785 				       int flags)
786 {
787 	int status;
788 	u32 bits_wanted = ac->ac_bits_wanted;
789 	struct inode *alloc_inode;
790 	struct buffer_head *bh = NULL;
791 	struct ocfs2_dinode *fe;
792 	u32 free_bits;
793 
794 	alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
795 	if (!alloc_inode) {
796 		mlog_errno(-EINVAL);
797 		return -EINVAL;
798 	}
799 
800 	inode_lock(alloc_inode);
801 
802 	status = ocfs2_inode_lock(alloc_inode, &bh, 1);
803 	if (status < 0) {
804 		inode_unlock(alloc_inode);
805 		iput(alloc_inode);
806 
807 		mlog_errno(status);
808 		return status;
809 	}
810 
811 	ac->ac_inode = alloc_inode;
812 	ac->ac_alloc_slot = slot;
813 
814 	fe = (struct ocfs2_dinode *) bh->b_data;
815 
816 	/* The bh was validated by the inode read inside
817 	 * ocfs2_inode_lock().  Any corruption is a code bug. */
818 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
819 
820 	if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
821 		status = ocfs2_error(alloc_inode->i_sb,
822 				     "Invalid chain allocator %llu\n",
823 				     (unsigned long long)le64_to_cpu(fe->i_blkno));
824 		goto bail;
825 	}
826 
827 	free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
828 		le32_to_cpu(fe->id1.bitmap1.i_used);
829 
830 	if (bits_wanted > free_bits) {
831 		/* cluster bitmap never grows */
832 		if (ocfs2_is_cluster_bitmap(alloc_inode)) {
833 			trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
834 								free_bits);
835 			status = -ENOSPC;
836 			goto bail;
837 		}
838 
839 		if (!(flags & ALLOC_NEW_GROUP)) {
840 			trace_ocfs2_reserve_suballoc_bits_no_new_group(
841 						slot, bits_wanted, free_bits);
842 			status = -ENOSPC;
843 			goto bail;
844 		}
845 
846 		status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
847 						 ac->ac_max_block,
848 						 last_alloc_group, flags);
849 		if (status < 0) {
850 			if (status != -ENOSPC)
851 				mlog_errno(status);
852 			goto bail;
853 		}
854 		atomic_inc(&osb->alloc_stats.bg_extends);
855 
856 		/* You should never ask for this much metadata */
857 		BUG_ON(bits_wanted >
858 		       (le32_to_cpu(fe->id1.bitmap1.i_total)
859 			- le32_to_cpu(fe->id1.bitmap1.i_used)));
860 	}
861 
862 	get_bh(bh);
863 	ac->ac_bh = bh;
864 bail:
865 	brelse(bh);
866 
867 	if (status)
868 		mlog_errno(status);
869 	return status;
870 }
871 
872 static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
873 {
874 	spin_lock(&osb->osb_lock);
875 	osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
876 	spin_unlock(&osb->osb_lock);
877 	atomic_set(&osb->s_num_inodes_stolen, 0);
878 }
879 
880 static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
881 {
882 	spin_lock(&osb->osb_lock);
883 	osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
884 	spin_unlock(&osb->osb_lock);
885 	atomic_set(&osb->s_num_meta_stolen, 0);
886 }
887 
888 void ocfs2_init_steal_slots(struct ocfs2_super *osb)
889 {
890 	ocfs2_init_inode_steal_slot(osb);
891 	ocfs2_init_meta_steal_slot(osb);
892 }
893 
894 static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
895 {
896 	spin_lock(&osb->osb_lock);
897 	if (type == INODE_ALLOC_SYSTEM_INODE)
898 		osb->s_inode_steal_slot = slot;
899 	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
900 		osb->s_meta_steal_slot = slot;
901 	spin_unlock(&osb->osb_lock);
902 }
903 
904 static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
905 {
906 	int slot = OCFS2_INVALID_SLOT;
907 
908 	spin_lock(&osb->osb_lock);
909 	if (type == INODE_ALLOC_SYSTEM_INODE)
910 		slot = osb->s_inode_steal_slot;
911 	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
912 		slot = osb->s_meta_steal_slot;
913 	spin_unlock(&osb->osb_lock);
914 
915 	return slot;
916 }
917 
918 static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
919 {
920 	return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
921 }
922 
923 static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
924 {
925 	return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
926 }
927 
928 static int ocfs2_steal_resource(struct ocfs2_super *osb,
929 				struct ocfs2_alloc_context *ac,
930 				int type)
931 {
932 	int i, status = -ENOSPC;
933 	int slot = __ocfs2_get_steal_slot(osb, type);
934 
935 	/* Start to steal resource from the first slot after ours. */
936 	if (slot == OCFS2_INVALID_SLOT)
937 		slot = osb->slot_num + 1;
938 
939 	for (i = 0; i < osb->max_slots; i++, slot++) {
940 		if (slot == osb->max_slots)
941 			slot = 0;
942 
943 		if (slot == osb->slot_num)
944 			continue;
945 
946 		status = ocfs2_reserve_suballoc_bits(osb, ac,
947 						     type,
948 						     (u32)slot, NULL,
949 						     NOT_ALLOC_NEW_GROUP);
950 		if (status >= 0) {
951 			__ocfs2_set_steal_slot(osb, slot, type);
952 			break;
953 		}
954 
955 		ocfs2_free_ac_resource(ac);
956 	}
957 
958 	return status;
959 }
960 
961 static int ocfs2_steal_inode(struct ocfs2_super *osb,
962 			     struct ocfs2_alloc_context *ac)
963 {
964 	return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
965 }
966 
967 static int ocfs2_steal_meta(struct ocfs2_super *osb,
968 			    struct ocfs2_alloc_context *ac)
969 {
970 	return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
971 }
972 
973 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
974 				      int blocks,
975 				      struct ocfs2_alloc_context **ac)
976 {
977 	int status;
978 	int slot = ocfs2_get_meta_steal_slot(osb);
979 
980 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
981 	if (!(*ac)) {
982 		status = -ENOMEM;
983 		mlog_errno(status);
984 		goto bail;
985 	}
986 
987 	(*ac)->ac_bits_wanted = blocks;
988 	(*ac)->ac_which = OCFS2_AC_USE_META;
989 	(*ac)->ac_group_search = ocfs2_block_group_search;
990 
991 	if (slot != OCFS2_INVALID_SLOT &&
992 		atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
993 		goto extent_steal;
994 
995 	atomic_set(&osb->s_num_meta_stolen, 0);
996 	status = ocfs2_reserve_suballoc_bits(osb, (*ac),
997 					     EXTENT_ALLOC_SYSTEM_INODE,
998 					     (u32)osb->slot_num, NULL,
999 					     ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
1000 
1001 
1002 	if (status >= 0) {
1003 		status = 0;
1004 		if (slot != OCFS2_INVALID_SLOT)
1005 			ocfs2_init_meta_steal_slot(osb);
1006 		goto bail;
1007 	} else if (status < 0 && status != -ENOSPC) {
1008 		mlog_errno(status);
1009 		goto bail;
1010 	}
1011 
1012 	ocfs2_free_ac_resource(*ac);
1013 
1014 extent_steal:
1015 	status = ocfs2_steal_meta(osb, *ac);
1016 	atomic_inc(&osb->s_num_meta_stolen);
1017 	if (status < 0) {
1018 		if (status != -ENOSPC)
1019 			mlog_errno(status);
1020 		goto bail;
1021 	}
1022 
1023 	status = 0;
1024 bail:
1025 	if ((status < 0) && *ac) {
1026 		ocfs2_free_alloc_context(*ac);
1027 		*ac = NULL;
1028 	}
1029 
1030 	if (status)
1031 		mlog_errno(status);
1032 	return status;
1033 }
1034 
1035 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1036 			       struct ocfs2_extent_list *root_el,
1037 			       struct ocfs2_alloc_context **ac)
1038 {
1039 	return ocfs2_reserve_new_metadata_blocks(osb,
1040 					ocfs2_extend_meta_needed(root_el),
1041 					ac);
1042 }
1043 
1044 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
1045 			    struct ocfs2_alloc_context **ac)
1046 {
1047 	int status;
1048 	int slot = ocfs2_get_inode_steal_slot(osb);
1049 	u64 alloc_group;
1050 
1051 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1052 	if (!(*ac)) {
1053 		status = -ENOMEM;
1054 		mlog_errno(status);
1055 		goto bail;
1056 	}
1057 
1058 	(*ac)->ac_bits_wanted = 1;
1059 	(*ac)->ac_which = OCFS2_AC_USE_INODE;
1060 
1061 	(*ac)->ac_group_search = ocfs2_block_group_search;
1062 
1063 	/*
1064 	 * stat(2) can't handle i_ino > 32bits, so we tell the
1065 	 * lower levels not to allocate us a block group past that
1066 	 * limit.  The 'inode64' mount option avoids this behavior.
1067 	 */
1068 	if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1069 		(*ac)->ac_max_block = (u32)~0U;
1070 
1071 	/*
1072 	 * slot is set when we successfully steal inode from other nodes.
1073 	 * It is reset in 3 places:
1074 	 * 1. when we flush the truncate log
1075 	 * 2. when we complete local alloc recovery.
1076 	 * 3. when we successfully allocate from our own slot.
1077 	 * After it is set, we will go on stealing inodes until we find the
1078 	 * need to check our slots to see whether there is some space for us.
1079 	 */
1080 	if (slot != OCFS2_INVALID_SLOT &&
1081 	    atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
1082 		goto inode_steal;
1083 
1084 	atomic_set(&osb->s_num_inodes_stolen, 0);
1085 	alloc_group = osb->osb_inode_alloc_group;
1086 	status = ocfs2_reserve_suballoc_bits(osb, *ac,
1087 					     INODE_ALLOC_SYSTEM_INODE,
1088 					     (u32)osb->slot_num,
1089 					     &alloc_group,
1090 					     ALLOC_NEW_GROUP |
1091 					     ALLOC_GROUPS_FROM_GLOBAL);
1092 	if (status >= 0) {
1093 		status = 0;
1094 
1095 		spin_lock(&osb->osb_lock);
1096 		osb->osb_inode_alloc_group = alloc_group;
1097 		spin_unlock(&osb->osb_lock);
1098 		trace_ocfs2_reserve_new_inode_new_group(
1099 			(unsigned long long)alloc_group);
1100 
1101 		/*
1102 		 * Some inodes must be freed by us, so try to allocate
1103 		 * from our own next time.
1104 		 */
1105 		if (slot != OCFS2_INVALID_SLOT)
1106 			ocfs2_init_inode_steal_slot(osb);
1107 		goto bail;
1108 	} else if (status < 0 && status != -ENOSPC) {
1109 		mlog_errno(status);
1110 		goto bail;
1111 	}
1112 
1113 	ocfs2_free_ac_resource(*ac);
1114 
1115 inode_steal:
1116 	status = ocfs2_steal_inode(osb, *ac);
1117 	atomic_inc(&osb->s_num_inodes_stolen);
1118 	if (status < 0) {
1119 		if (status != -ENOSPC)
1120 			mlog_errno(status);
1121 		goto bail;
1122 	}
1123 
1124 	status = 0;
1125 bail:
1126 	if ((status < 0) && *ac) {
1127 		ocfs2_free_alloc_context(*ac);
1128 		*ac = NULL;
1129 	}
1130 
1131 	if (status)
1132 		mlog_errno(status);
1133 	return status;
1134 }
1135 
1136 /* local alloc code has to do the same thing, so rather than do this
1137  * twice.. */
1138 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1139 				      struct ocfs2_alloc_context *ac)
1140 {
1141 	int status;
1142 
1143 	ac->ac_which = OCFS2_AC_USE_MAIN;
1144 	ac->ac_group_search = ocfs2_cluster_group_search;
1145 
1146 	status = ocfs2_reserve_suballoc_bits(osb, ac,
1147 					     GLOBAL_BITMAP_SYSTEM_INODE,
1148 					     OCFS2_INVALID_SLOT, NULL,
1149 					     ALLOC_NEW_GROUP);
1150 	if (status < 0 && status != -ENOSPC) {
1151 		mlog_errno(status);
1152 		goto bail;
1153 	}
1154 
1155 bail:
1156 	return status;
1157 }
1158 
1159 /* Callers don't need to care which bitmap (local alloc or main) to
1160  * use so we figure it out for them, but unfortunately this clutters
1161  * things a bit. */
1162 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1163 					     u32 bits_wanted, u64 max_block,
1164 					     int flags,
1165 					     struct ocfs2_alloc_context **ac)
1166 {
1167 	int status, ret = 0;
1168 	int retried = 0;
1169 
1170 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1171 	if (!(*ac)) {
1172 		status = -ENOMEM;
1173 		mlog_errno(status);
1174 		goto bail;
1175 	}
1176 
1177 	(*ac)->ac_bits_wanted = bits_wanted;
1178 	(*ac)->ac_max_block = max_block;
1179 
1180 	status = -ENOSPC;
1181 	if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1182 	    ocfs2_alloc_should_use_local(osb, bits_wanted)) {
1183 		status = ocfs2_reserve_local_alloc_bits(osb,
1184 							bits_wanted,
1185 							*ac);
1186 		if ((status < 0) && (status != -ENOSPC)) {
1187 			mlog_errno(status);
1188 			goto bail;
1189 		}
1190 	}
1191 
1192 	if (status == -ENOSPC) {
1193 retry:
1194 		status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1195 		/* Retry if there is sufficient space cached in truncate log */
1196 		if (status == -ENOSPC && !retried) {
1197 			retried = 1;
1198 			ocfs2_inode_unlock((*ac)->ac_inode, 1);
1199 			inode_unlock((*ac)->ac_inode);
1200 
1201 			ret = ocfs2_try_to_free_truncate_log(osb, bits_wanted);
1202 			if (ret == 1) {
1203 				iput((*ac)->ac_inode);
1204 				(*ac)->ac_inode = NULL;
1205 				goto retry;
1206 			}
1207 
1208 			if (ret < 0)
1209 				mlog_errno(ret);
1210 
1211 			inode_lock((*ac)->ac_inode);
1212 			ret = ocfs2_inode_lock((*ac)->ac_inode, NULL, 1);
1213 			if (ret < 0) {
1214 				mlog_errno(ret);
1215 				inode_unlock((*ac)->ac_inode);
1216 				iput((*ac)->ac_inode);
1217 				(*ac)->ac_inode = NULL;
1218 				goto bail;
1219 			}
1220 		}
1221 		if (status < 0) {
1222 			if (status != -ENOSPC)
1223 				mlog_errno(status);
1224 			goto bail;
1225 		}
1226 	}
1227 
1228 	status = 0;
1229 bail:
1230 	if ((status < 0) && *ac) {
1231 		ocfs2_free_alloc_context(*ac);
1232 		*ac = NULL;
1233 	}
1234 
1235 	if (status)
1236 		mlog_errno(status);
1237 	return status;
1238 }
1239 
1240 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1241 			   u32 bits_wanted,
1242 			   struct ocfs2_alloc_context **ac)
1243 {
1244 	return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1245 						 ALLOC_NEW_GROUP, ac);
1246 }
1247 
1248 /*
1249  * More or less lifted from ext3. I'll leave their description below:
1250  *
1251  * "For ext3 allocations, we must not reuse any blocks which are
1252  * allocated in the bitmap buffer's "last committed data" copy.  This
1253  * prevents deletes from freeing up the page for reuse until we have
1254  * committed the delete transaction.
1255  *
1256  * If we didn't do this, then deleting something and reallocating it as
1257  * data would allow the old block to be overwritten before the
1258  * transaction committed (because we force data to disk before commit).
1259  * This would lead to corruption if we crashed between overwriting the
1260  * data and committing the delete.
1261  *
1262  * @@@ We may want to make this allocation behaviour conditional on
1263  * data-writes at some point, and disable it for metadata allocations or
1264  * sync-data inodes."
1265  *
1266  * Note: OCFS2 already does this differently for metadata vs data
1267  * allocations, as those bitmaps are separate and undo access is never
1268  * called on a metadata group descriptor.
1269  */
1270 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1271 					 int nr)
1272 {
1273 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1274 	int ret;
1275 
1276 	if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1277 		return 0;
1278 
1279 	if (!buffer_jbd(bg_bh))
1280 		return 1;
1281 
1282 	jbd_lock_bh_state(bg_bh);
1283 	bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
1284 	if (bg)
1285 		ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1286 	else
1287 		ret = 1;
1288 	jbd_unlock_bh_state(bg_bh);
1289 
1290 	return ret;
1291 }
1292 
1293 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1294 					     struct buffer_head *bg_bh,
1295 					     unsigned int bits_wanted,
1296 					     unsigned int total_bits,
1297 					     struct ocfs2_suballoc_result *res)
1298 {
1299 	void *bitmap;
1300 	u16 best_offset, best_size;
1301 	int offset, start, found, status = 0;
1302 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1303 
1304 	/* Callers got this descriptor from
1305 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1306 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1307 
1308 	found = start = best_offset = best_size = 0;
1309 	bitmap = bg->bg_bitmap;
1310 
1311 	while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1312 		if (offset == total_bits)
1313 			break;
1314 
1315 		if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1316 			/* We found a zero, but we can't use it as it
1317 			 * hasn't been put to disk yet! */
1318 			found = 0;
1319 			start = offset + 1;
1320 		} else if (offset == start) {
1321 			/* we found a zero */
1322 			found++;
1323 			/* move start to the next bit to test */
1324 			start++;
1325 		} else {
1326 			/* got a zero after some ones */
1327 			found = 1;
1328 			start = offset + 1;
1329 		}
1330 		if (found > best_size) {
1331 			best_size = found;
1332 			best_offset = start - found;
1333 		}
1334 		/* we got everything we needed */
1335 		if (found == bits_wanted) {
1336 			/* mlog(0, "Found it all!\n"); */
1337 			break;
1338 		}
1339 	}
1340 
1341 	if (best_size) {
1342 		res->sr_bit_offset = best_offset;
1343 		res->sr_bits = best_size;
1344 	} else {
1345 		status = -ENOSPC;
1346 		/* No error log here -- see the comment above
1347 		 * ocfs2_test_bg_bit_allocatable */
1348 	}
1349 
1350 	return status;
1351 }
1352 
1353 int ocfs2_block_group_set_bits(handle_t *handle,
1354 					     struct inode *alloc_inode,
1355 					     struct ocfs2_group_desc *bg,
1356 					     struct buffer_head *group_bh,
1357 					     unsigned int bit_off,
1358 					     unsigned int num_bits)
1359 {
1360 	int status;
1361 	void *bitmap = bg->bg_bitmap;
1362 	int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1363 
1364 	/* All callers get the descriptor via
1365 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1366 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1367 	BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1368 
1369 	trace_ocfs2_block_group_set_bits(bit_off, num_bits);
1370 
1371 	if (ocfs2_is_cluster_bitmap(alloc_inode))
1372 		journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1373 
1374 	status = ocfs2_journal_access_gd(handle,
1375 					 INODE_CACHE(alloc_inode),
1376 					 group_bh,
1377 					 journal_type);
1378 	if (status < 0) {
1379 		mlog_errno(status);
1380 		goto bail;
1381 	}
1382 
1383 	le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1384 	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1385 		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
1386 				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
1387 				   le16_to_cpu(bg->bg_bits),
1388 				   le16_to_cpu(bg->bg_free_bits_count),
1389 				   num_bits);
1390 	}
1391 	while(num_bits--)
1392 		ocfs2_set_bit(bit_off++, bitmap);
1393 
1394 	ocfs2_journal_dirty(handle, group_bh);
1395 
1396 bail:
1397 	return status;
1398 }
1399 
1400 /* find the one with the most empty bits */
1401 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1402 {
1403 	u16 curr, best;
1404 
1405 	BUG_ON(!cl->cl_next_free_rec);
1406 
1407 	best = curr = 0;
1408 	while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1409 		if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1410 		    le32_to_cpu(cl->cl_recs[best].c_free))
1411 			best = curr;
1412 		curr++;
1413 	}
1414 
1415 	BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1416 	return best;
1417 }
1418 
1419 static int ocfs2_relink_block_group(handle_t *handle,
1420 				    struct inode *alloc_inode,
1421 				    struct buffer_head *fe_bh,
1422 				    struct buffer_head *bg_bh,
1423 				    struct buffer_head *prev_bg_bh,
1424 				    u16 chain)
1425 {
1426 	int status;
1427 	/* there is a really tiny chance the journal calls could fail,
1428 	 * but we wouldn't want inconsistent blocks in *any* case. */
1429 	u64 bg_ptr, prev_bg_ptr;
1430 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1431 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1432 	struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1433 
1434 	/* The caller got these descriptors from
1435 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1436 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1437 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1438 
1439 	trace_ocfs2_relink_block_group(
1440 		(unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1441 		(unsigned long long)le64_to_cpu(bg->bg_blkno),
1442 		(unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1443 
1444 	bg_ptr = le64_to_cpu(bg->bg_next_group);
1445 	prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1446 
1447 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1448 					 prev_bg_bh,
1449 					 OCFS2_JOURNAL_ACCESS_WRITE);
1450 	if (status < 0)
1451 		goto out;
1452 
1453 	prev_bg->bg_next_group = bg->bg_next_group;
1454 	ocfs2_journal_dirty(handle, prev_bg_bh);
1455 
1456 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1457 					 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1458 	if (status < 0)
1459 		goto out_rollback_prev_bg;
1460 
1461 	bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1462 	ocfs2_journal_dirty(handle, bg_bh);
1463 
1464 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1465 					 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1466 	if (status < 0)
1467 		goto out_rollback_bg;
1468 
1469 	fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1470 	ocfs2_journal_dirty(handle, fe_bh);
1471 
1472 out:
1473 	if (status < 0)
1474 		mlog_errno(status);
1475 	return status;
1476 
1477 out_rollback_bg:
1478 	bg->bg_next_group = cpu_to_le64(bg_ptr);
1479 out_rollback_prev_bg:
1480 	prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1481 	goto out;
1482 }
1483 
1484 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1485 						     u32 wanted)
1486 {
1487 	return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1488 }
1489 
1490 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1491  * value on error. */
1492 static int ocfs2_cluster_group_search(struct inode *inode,
1493 				      struct buffer_head *group_bh,
1494 				      u32 bits_wanted, u32 min_bits,
1495 				      u64 max_block,
1496 				      struct ocfs2_suballoc_result *res)
1497 {
1498 	int search = -ENOSPC;
1499 	int ret;
1500 	u64 blkoff;
1501 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1502 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1503 	unsigned int max_bits, gd_cluster_off;
1504 
1505 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1506 
1507 	if (gd->bg_free_bits_count) {
1508 		max_bits = le16_to_cpu(gd->bg_bits);
1509 
1510 		/* Tail groups in cluster bitmaps which aren't cpg
1511 		 * aligned are prone to partial extension by a failed
1512 		 * fs resize. If the file system resize never got to
1513 		 * update the dinode cluster count, then we don't want
1514 		 * to trust any clusters past it, regardless of what
1515 		 * the group descriptor says. */
1516 		gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1517 							  le64_to_cpu(gd->bg_blkno));
1518 		if ((gd_cluster_off + max_bits) >
1519 		    OCFS2_I(inode)->ip_clusters) {
1520 			max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1521 			trace_ocfs2_cluster_group_search_wrong_max_bits(
1522 				(unsigned long long)le64_to_cpu(gd->bg_blkno),
1523 				le16_to_cpu(gd->bg_bits),
1524 				OCFS2_I(inode)->ip_clusters, max_bits);
1525 		}
1526 
1527 		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1528 							group_bh, bits_wanted,
1529 							max_bits, res);
1530 		if (ret)
1531 			return ret;
1532 
1533 		if (max_block) {
1534 			blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1535 							  gd_cluster_off +
1536 							  res->sr_bit_offset +
1537 							  res->sr_bits);
1538 			trace_ocfs2_cluster_group_search_max_block(
1539 				(unsigned long long)blkoff,
1540 				(unsigned long long)max_block);
1541 			if (blkoff > max_block)
1542 				return -ENOSPC;
1543 		}
1544 
1545 		/* ocfs2_block_group_find_clear_bits() might
1546 		 * return success, but we still want to return
1547 		 * -ENOSPC unless it found the minimum number
1548 		 * of bits. */
1549 		if (min_bits <= res->sr_bits)
1550 			search = 0; /* success */
1551 		else if (res->sr_bits) {
1552 			/*
1553 			 * Don't show bits which we'll be returning
1554 			 * for allocation to the local alloc bitmap.
1555 			 */
1556 			ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
1557 		}
1558 	}
1559 
1560 	return search;
1561 }
1562 
1563 static int ocfs2_block_group_search(struct inode *inode,
1564 				    struct buffer_head *group_bh,
1565 				    u32 bits_wanted, u32 min_bits,
1566 				    u64 max_block,
1567 				    struct ocfs2_suballoc_result *res)
1568 {
1569 	int ret = -ENOSPC;
1570 	u64 blkoff;
1571 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1572 
1573 	BUG_ON(min_bits != 1);
1574 	BUG_ON(ocfs2_is_cluster_bitmap(inode));
1575 
1576 	if (bg->bg_free_bits_count) {
1577 		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1578 							group_bh, bits_wanted,
1579 							le16_to_cpu(bg->bg_bits),
1580 							res);
1581 		if (!ret && max_block) {
1582 			blkoff = le64_to_cpu(bg->bg_blkno) +
1583 				res->sr_bit_offset + res->sr_bits;
1584 			trace_ocfs2_block_group_search_max_block(
1585 				(unsigned long long)blkoff,
1586 				(unsigned long long)max_block);
1587 			if (blkoff > max_block)
1588 				ret = -ENOSPC;
1589 		}
1590 	}
1591 
1592 	return ret;
1593 }
1594 
1595 int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1596 				       handle_t *handle,
1597 				       struct buffer_head *di_bh,
1598 				       u32 num_bits,
1599 				       u16 chain)
1600 {
1601 	int ret;
1602 	u32 tmp_used;
1603 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1604 	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1605 
1606 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1607 				      OCFS2_JOURNAL_ACCESS_WRITE);
1608 	if (ret < 0) {
1609 		mlog_errno(ret);
1610 		goto out;
1611 	}
1612 
1613 	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1614 	di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1615 	le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1616 	ocfs2_journal_dirty(handle, di_bh);
1617 
1618 out:
1619 	return ret;
1620 }
1621 
1622 void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
1623 				       struct buffer_head *di_bh,
1624 				       u32 num_bits,
1625 				       u16 chain)
1626 {
1627 	u32 tmp_used;
1628 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1629 	struct ocfs2_chain_list *cl;
1630 
1631 	cl = (struct ocfs2_chain_list *)&di->id2.i_chain;
1632 	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1633 	di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits);
1634 	le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits);
1635 }
1636 
1637 static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1638 					 struct ocfs2_extent_rec *rec,
1639 					 struct ocfs2_chain_list *cl)
1640 {
1641 	unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1642 	unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1643 	unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1644 
1645 	if (res->sr_bit_offset < bitoff)
1646 		return 0;
1647 	if (res->sr_bit_offset >= (bitoff + bitcount))
1648 		return 0;
1649 	res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1650 		(res->sr_bit_offset - bitoff);
1651 	if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1652 		res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1653 	return 1;
1654 }
1655 
1656 static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1657 					  struct ocfs2_group_desc *bg,
1658 					  struct ocfs2_suballoc_result *res)
1659 {
1660 	int i;
1661 	u64 bg_blkno = res->sr_bg_blkno;  /* Save off */
1662 	struct ocfs2_extent_rec *rec;
1663 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1664 	struct ocfs2_chain_list *cl = &di->id2.i_chain;
1665 
1666 	if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1667 		res->sr_blkno = 0;
1668 		return;
1669 	}
1670 
1671 	res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1672 	res->sr_bg_blkno = 0;  /* Clear it for contig block groups */
1673 	if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1674 	    !bg->bg_list.l_next_free_rec)
1675 		return;
1676 
1677 	for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1678 		rec = &bg->bg_list.l_recs[i];
1679 		if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1680 			res->sr_bg_blkno = bg_blkno;  /* Restore */
1681 			break;
1682 		}
1683 	}
1684 }
1685 
1686 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1687 				  handle_t *handle,
1688 				  u32 bits_wanted,
1689 				  u32 min_bits,
1690 				  struct ocfs2_suballoc_result *res,
1691 				  u16 *bits_left)
1692 {
1693 	int ret;
1694 	struct buffer_head *group_bh = NULL;
1695 	struct ocfs2_group_desc *gd;
1696 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1697 	struct inode *alloc_inode = ac->ac_inode;
1698 
1699 	ret = ocfs2_read_group_descriptor(alloc_inode, di,
1700 					  res->sr_bg_blkno, &group_bh);
1701 	if (ret < 0) {
1702 		mlog_errno(ret);
1703 		return ret;
1704 	}
1705 
1706 	gd = (struct ocfs2_group_desc *) group_bh->b_data;
1707 	ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1708 				  ac->ac_max_block, res);
1709 	if (ret < 0) {
1710 		if (ret != -ENOSPC)
1711 			mlog_errno(ret);
1712 		goto out;
1713 	}
1714 
1715 	if (!ret)
1716 		ocfs2_bg_discontig_fix_result(ac, gd, res);
1717 
1718 	/*
1719 	 * sr_bg_blkno might have been changed by
1720 	 * ocfs2_bg_discontig_fix_result
1721 	 */
1722 	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1723 
1724 	if (ac->ac_find_loc_only)
1725 		goto out_loc_only;
1726 
1727 	ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1728 					       res->sr_bits,
1729 					       le16_to_cpu(gd->bg_chain));
1730 	if (ret < 0) {
1731 		mlog_errno(ret);
1732 		goto out;
1733 	}
1734 
1735 	ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1736 					 res->sr_bit_offset, res->sr_bits);
1737 	if (ret < 0) {
1738 		ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh,
1739 					       res->sr_bits,
1740 					       le16_to_cpu(gd->bg_chain));
1741 		mlog_errno(ret);
1742 	}
1743 
1744 out_loc_only:
1745 	*bits_left = le16_to_cpu(gd->bg_free_bits_count);
1746 
1747 out:
1748 	brelse(group_bh);
1749 
1750 	return ret;
1751 }
1752 
1753 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1754 			      handle_t *handle,
1755 			      u32 bits_wanted,
1756 			      u32 min_bits,
1757 			      struct ocfs2_suballoc_result *res,
1758 			      u16 *bits_left)
1759 {
1760 	int status;
1761 	u16 chain;
1762 	u64 next_group;
1763 	struct inode *alloc_inode = ac->ac_inode;
1764 	struct buffer_head *group_bh = NULL;
1765 	struct buffer_head *prev_group_bh = NULL;
1766 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1767 	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1768 	struct ocfs2_group_desc *bg;
1769 
1770 	chain = ac->ac_chain;
1771 	trace_ocfs2_search_chain_begin(
1772 		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1773 		bits_wanted, chain);
1774 
1775 	status = ocfs2_read_group_descriptor(alloc_inode, fe,
1776 					     le64_to_cpu(cl->cl_recs[chain].c_blkno),
1777 					     &group_bh);
1778 	if (status < 0) {
1779 		mlog_errno(status);
1780 		goto bail;
1781 	}
1782 	bg = (struct ocfs2_group_desc *) group_bh->b_data;
1783 
1784 	status = -ENOSPC;
1785 	/* for now, the chain search is a bit simplistic. We just use
1786 	 * the 1st group with any empty bits. */
1787 	while ((status = ac->ac_group_search(alloc_inode, group_bh,
1788 					     bits_wanted, min_bits,
1789 					     ac->ac_max_block,
1790 					     res)) == -ENOSPC) {
1791 		if (!bg->bg_next_group)
1792 			break;
1793 
1794 		brelse(prev_group_bh);
1795 		prev_group_bh = NULL;
1796 
1797 		next_group = le64_to_cpu(bg->bg_next_group);
1798 		prev_group_bh = group_bh;
1799 		group_bh = NULL;
1800 		status = ocfs2_read_group_descriptor(alloc_inode, fe,
1801 						     next_group, &group_bh);
1802 		if (status < 0) {
1803 			mlog_errno(status);
1804 			goto bail;
1805 		}
1806 		bg = (struct ocfs2_group_desc *) group_bh->b_data;
1807 	}
1808 	if (status < 0) {
1809 		if (status != -ENOSPC)
1810 			mlog_errno(status);
1811 		goto bail;
1812 	}
1813 
1814 	trace_ocfs2_search_chain_succ(
1815 		(unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1816 
1817 	res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1818 
1819 	BUG_ON(res->sr_bits == 0);
1820 	if (!status)
1821 		ocfs2_bg_discontig_fix_result(ac, bg, res);
1822 
1823 	/*
1824 	 * sr_bg_blkno might have been changed by
1825 	 * ocfs2_bg_discontig_fix_result
1826 	 */
1827 	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1828 
1829 	/*
1830 	 * Keep track of previous block descriptor read. When
1831 	 * we find a target, if we have read more than X
1832 	 * number of descriptors, and the target is reasonably
1833 	 * empty, relink him to top of his chain.
1834 	 *
1835 	 * We've read 0 extra blocks and only send one more to
1836 	 * the transaction, yet the next guy to search has a
1837 	 * much easier time.
1838 	 *
1839 	 * Do this *after* figuring out how many bits we're taking out
1840 	 * of our target group.
1841 	 */
1842 	if (!ac->ac_disable_chain_relink &&
1843 	    (prev_group_bh) &&
1844 	    (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1845 		status = ocfs2_relink_block_group(handle, alloc_inode,
1846 						  ac->ac_bh, group_bh,
1847 						  prev_group_bh, chain);
1848 		if (status < 0) {
1849 			mlog_errno(status);
1850 			goto bail;
1851 		}
1852 	}
1853 
1854 	if (ac->ac_find_loc_only)
1855 		goto out_loc_only;
1856 
1857 	status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1858 						  ac->ac_bh, res->sr_bits,
1859 						  chain);
1860 	if (status) {
1861 		mlog_errno(status);
1862 		goto bail;
1863 	}
1864 
1865 	status = ocfs2_block_group_set_bits(handle,
1866 					    alloc_inode,
1867 					    bg,
1868 					    group_bh,
1869 					    res->sr_bit_offset,
1870 					    res->sr_bits);
1871 	if (status < 0) {
1872 		ocfs2_rollback_alloc_dinode_counts(alloc_inode,
1873 					ac->ac_bh, res->sr_bits, chain);
1874 		mlog_errno(status);
1875 		goto bail;
1876 	}
1877 
1878 	trace_ocfs2_search_chain_end(
1879 			(unsigned long long)le64_to_cpu(fe->i_blkno),
1880 			res->sr_bits);
1881 
1882 out_loc_only:
1883 	*bits_left = le16_to_cpu(bg->bg_free_bits_count);
1884 bail:
1885 	brelse(group_bh);
1886 	brelse(prev_group_bh);
1887 
1888 	if (status)
1889 		mlog_errno(status);
1890 	return status;
1891 }
1892 
1893 /* will give out up to bits_wanted contiguous bits. */
1894 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1895 				     handle_t *handle,
1896 				     u32 bits_wanted,
1897 				     u32 min_bits,
1898 				     struct ocfs2_suballoc_result *res)
1899 {
1900 	int status;
1901 	u16 victim, i;
1902 	u16 bits_left = 0;
1903 	u64 hint = ac->ac_last_group;
1904 	struct ocfs2_chain_list *cl;
1905 	struct ocfs2_dinode *fe;
1906 
1907 	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1908 	BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1909 	BUG_ON(!ac->ac_bh);
1910 
1911 	fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1912 
1913 	/* The bh was validated by the inode read during
1914 	 * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1915 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1916 
1917 	if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1918 	    le32_to_cpu(fe->id1.bitmap1.i_total)) {
1919 		status = ocfs2_error(ac->ac_inode->i_sb,
1920 				     "Chain allocator dinode %llu has %u used bits but only %u total\n",
1921 				     (unsigned long long)le64_to_cpu(fe->i_blkno),
1922 				     le32_to_cpu(fe->id1.bitmap1.i_used),
1923 				     le32_to_cpu(fe->id1.bitmap1.i_total));
1924 		goto bail;
1925 	}
1926 
1927 	res->sr_bg_blkno = hint;
1928 	if (res->sr_bg_blkno) {
1929 		/* Attempt to short-circuit the usual search mechanism
1930 		 * by jumping straight to the most recently used
1931 		 * allocation group. This helps us maintain some
1932 		 * contiguousness across allocations. */
1933 		status = ocfs2_search_one_group(ac, handle, bits_wanted,
1934 						min_bits, res, &bits_left);
1935 		if (!status)
1936 			goto set_hint;
1937 		if (status < 0 && status != -ENOSPC) {
1938 			mlog_errno(status);
1939 			goto bail;
1940 		}
1941 	}
1942 
1943 	cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1944 
1945 	victim = ocfs2_find_victim_chain(cl);
1946 	ac->ac_chain = victim;
1947 
1948 	status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1949 				    res, &bits_left);
1950 	if (!status) {
1951 		if (ocfs2_is_cluster_bitmap(ac->ac_inode))
1952 			hint = res->sr_bg_blkno;
1953 		else
1954 			hint = ocfs2_group_from_res(res);
1955 		goto set_hint;
1956 	}
1957 	if (status < 0 && status != -ENOSPC) {
1958 		mlog_errno(status);
1959 		goto bail;
1960 	}
1961 
1962 	trace_ocfs2_claim_suballoc_bits(victim);
1963 
1964 	/* If we didn't pick a good victim, then just default to
1965 	 * searching each chain in order. Don't allow chain relinking
1966 	 * because we only calculate enough journal credits for one
1967 	 * relink per alloc. */
1968 	ac->ac_disable_chain_relink = 1;
1969 	for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1970 		if (i == victim)
1971 			continue;
1972 		if (!cl->cl_recs[i].c_free)
1973 			continue;
1974 
1975 		ac->ac_chain = i;
1976 		status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1977 					    res, &bits_left);
1978 		if (!status) {
1979 			hint = ocfs2_group_from_res(res);
1980 			break;
1981 		}
1982 		if (status < 0 && status != -ENOSPC) {
1983 			mlog_errno(status);
1984 			goto bail;
1985 		}
1986 	}
1987 
1988 set_hint:
1989 	if (status != -ENOSPC) {
1990 		/* If the next search of this group is not likely to
1991 		 * yield a suitable extent, then we reset the last
1992 		 * group hint so as to not waste a disk read */
1993 		if (bits_left < min_bits)
1994 			ac->ac_last_group = 0;
1995 		else
1996 			ac->ac_last_group = hint;
1997 	}
1998 
1999 bail:
2000 	if (status)
2001 		mlog_errno(status);
2002 	return status;
2003 }
2004 
2005 int ocfs2_claim_metadata(handle_t *handle,
2006 			 struct ocfs2_alloc_context *ac,
2007 			 u32 bits_wanted,
2008 			 u64 *suballoc_loc,
2009 			 u16 *suballoc_bit_start,
2010 			 unsigned int *num_bits,
2011 			 u64 *blkno_start)
2012 {
2013 	int status;
2014 	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2015 
2016 	BUG_ON(!ac);
2017 	BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
2018 	BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
2019 
2020 	status = ocfs2_claim_suballoc_bits(ac,
2021 					   handle,
2022 					   bits_wanted,
2023 					   1,
2024 					   &res);
2025 	if (status < 0) {
2026 		mlog_errno(status);
2027 		goto bail;
2028 	}
2029 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2030 
2031 	*suballoc_loc = res.sr_bg_blkno;
2032 	*suballoc_bit_start = res.sr_bit_offset;
2033 	*blkno_start = res.sr_blkno;
2034 	ac->ac_bits_given += res.sr_bits;
2035 	*num_bits = res.sr_bits;
2036 	status = 0;
2037 bail:
2038 	if (status)
2039 		mlog_errno(status);
2040 	return status;
2041 }
2042 
2043 static void ocfs2_init_inode_ac_group(struct inode *dir,
2044 				      struct buffer_head *parent_di_bh,
2045 				      struct ocfs2_alloc_context *ac)
2046 {
2047 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2048 	/*
2049 	 * Try to allocate inodes from some specific group.
2050 	 *
2051 	 * If the parent dir has recorded the last group used in allocation,
2052 	 * cool, use it. Otherwise if we try to allocate new inode from the
2053 	 * same slot the parent dir belongs to, use the same chunk.
2054 	 *
2055 	 * We are very careful here to avoid the mistake of setting
2056 	 * ac_last_group to a group descriptor from a different (unlocked) slot.
2057 	 */
2058 	if (OCFS2_I(dir)->ip_last_used_group &&
2059 	    OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2060 		ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2061 	else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2062 		if (di->i_suballoc_loc)
2063 			ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2064 		else
2065 			ac->ac_last_group = ocfs2_which_suballoc_group(
2066 					le64_to_cpu(di->i_blkno),
2067 					le16_to_cpu(di->i_suballoc_bit));
2068 	}
2069 }
2070 
2071 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2072 					     struct ocfs2_alloc_context *ac)
2073 {
2074 	OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2075 	OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2076 }
2077 
2078 int ocfs2_find_new_inode_loc(struct inode *dir,
2079 			     struct buffer_head *parent_fe_bh,
2080 			     struct ocfs2_alloc_context *ac,
2081 			     u64 *fe_blkno)
2082 {
2083 	int ret;
2084 	handle_t *handle = NULL;
2085 	struct ocfs2_suballoc_result *res;
2086 
2087 	BUG_ON(!ac);
2088 	BUG_ON(ac->ac_bits_given != 0);
2089 	BUG_ON(ac->ac_bits_wanted != 1);
2090 	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2091 
2092 	res = kzalloc(sizeof(*res), GFP_NOFS);
2093 	if (res == NULL) {
2094 		ret = -ENOMEM;
2095 		mlog_errno(ret);
2096 		goto out;
2097 	}
2098 
2099 	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2100 
2101 	/*
2102 	 * The handle started here is for chain relink. Alternatively,
2103 	 * we could just disable relink for these calls.
2104 	 */
2105 	handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2106 	if (IS_ERR(handle)) {
2107 		ret = PTR_ERR(handle);
2108 		handle = NULL;
2109 		mlog_errno(ret);
2110 		goto out;
2111 	}
2112 
2113 	/*
2114 	 * This will instruct ocfs2_claim_suballoc_bits and
2115 	 * ocfs2_search_one_group to search but save actual allocation
2116 	 * for later.
2117 	 */
2118 	ac->ac_find_loc_only = 1;
2119 
2120 	ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2121 	if (ret < 0) {
2122 		mlog_errno(ret);
2123 		goto out;
2124 	}
2125 
2126 	ac->ac_find_loc_priv = res;
2127 	*fe_blkno = res->sr_blkno;
2128 	ocfs2_update_inode_fsync_trans(handle, dir, 0);
2129 out:
2130 	if (handle)
2131 		ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2132 
2133 	if (ret)
2134 		kfree(res);
2135 
2136 	return ret;
2137 }
2138 
2139 int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2140 				 struct inode *dir,
2141 				 struct ocfs2_alloc_context *ac,
2142 				 u64 *suballoc_loc,
2143 				 u16 *suballoc_bit,
2144 				 u64 di_blkno)
2145 {
2146 	int ret;
2147 	u16 chain;
2148 	struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2149 	struct buffer_head *bg_bh = NULL;
2150 	struct ocfs2_group_desc *bg;
2151 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2152 
2153 	/*
2154 	 * Since di_blkno is being passed back in, we check for any
2155 	 * inconsistencies which may have happened between
2156 	 * calls. These are code bugs as di_blkno is not expected to
2157 	 * change once returned from ocfs2_find_new_inode_loc()
2158 	 */
2159 	BUG_ON(res->sr_blkno != di_blkno);
2160 
2161 	ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2162 					  res->sr_bg_stable_blkno, &bg_bh);
2163 	if (ret) {
2164 		mlog_errno(ret);
2165 		goto out;
2166 	}
2167 
2168 	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2169 	chain = le16_to_cpu(bg->bg_chain);
2170 
2171 	ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2172 					       ac->ac_bh, res->sr_bits,
2173 					       chain);
2174 	if (ret) {
2175 		mlog_errno(ret);
2176 		goto out;
2177 	}
2178 
2179 	ret = ocfs2_block_group_set_bits(handle,
2180 					 ac->ac_inode,
2181 					 bg,
2182 					 bg_bh,
2183 					 res->sr_bit_offset,
2184 					 res->sr_bits);
2185 	if (ret < 0) {
2186 		ocfs2_rollback_alloc_dinode_counts(ac->ac_inode,
2187 					       ac->ac_bh, res->sr_bits, chain);
2188 		mlog_errno(ret);
2189 		goto out;
2190 	}
2191 
2192 	trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2193 					   res->sr_bits);
2194 
2195 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2196 
2197 	BUG_ON(res->sr_bits != 1);
2198 
2199 	*suballoc_loc = res->sr_bg_blkno;
2200 	*suballoc_bit = res->sr_bit_offset;
2201 	ac->ac_bits_given++;
2202 	ocfs2_save_inode_ac_group(dir, ac);
2203 
2204 out:
2205 	brelse(bg_bh);
2206 
2207 	return ret;
2208 }
2209 
2210 int ocfs2_claim_new_inode(handle_t *handle,
2211 			  struct inode *dir,
2212 			  struct buffer_head *parent_fe_bh,
2213 			  struct ocfs2_alloc_context *ac,
2214 			  u64 *suballoc_loc,
2215 			  u16 *suballoc_bit,
2216 			  u64 *fe_blkno)
2217 {
2218 	int status;
2219 	struct ocfs2_suballoc_result res;
2220 
2221 	BUG_ON(!ac);
2222 	BUG_ON(ac->ac_bits_given != 0);
2223 	BUG_ON(ac->ac_bits_wanted != 1);
2224 	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2225 
2226 	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2227 
2228 	status = ocfs2_claim_suballoc_bits(ac,
2229 					   handle,
2230 					   1,
2231 					   1,
2232 					   &res);
2233 	if (status < 0) {
2234 		mlog_errno(status);
2235 		goto bail;
2236 	}
2237 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2238 
2239 	BUG_ON(res.sr_bits != 1);
2240 
2241 	*suballoc_loc = res.sr_bg_blkno;
2242 	*suballoc_bit = res.sr_bit_offset;
2243 	*fe_blkno = res.sr_blkno;
2244 	ac->ac_bits_given++;
2245 	ocfs2_save_inode_ac_group(dir, ac);
2246 	status = 0;
2247 bail:
2248 	if (status)
2249 		mlog_errno(status);
2250 	return status;
2251 }
2252 
2253 /* translate a group desc. blkno and it's bitmap offset into
2254  * disk cluster offset. */
2255 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2256 						   u64 bg_blkno,
2257 						   u16 bg_bit_off)
2258 {
2259 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2260 	u32 cluster = 0;
2261 
2262 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2263 
2264 	if (bg_blkno != osb->first_cluster_group_blkno)
2265 		cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2266 	cluster += (u32) bg_bit_off;
2267 	return cluster;
2268 }
2269 
2270 /* given a cluster offset, calculate which block group it belongs to
2271  * and return that block offset. */
2272 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2273 {
2274 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2275 	u32 group_no;
2276 
2277 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2278 
2279 	group_no = cluster / osb->bitmap_cpg;
2280 	if (!group_no)
2281 		return osb->first_cluster_group_blkno;
2282 	return ocfs2_clusters_to_blocks(inode->i_sb,
2283 					group_no * osb->bitmap_cpg);
2284 }
2285 
2286 /* given the block number of a cluster start, calculate which cluster
2287  * group and descriptor bitmap offset that corresponds to. */
2288 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2289 						u64 data_blkno,
2290 						u64 *bg_blkno,
2291 						u16 *bg_bit_off)
2292 {
2293 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2294 	u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2295 
2296 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2297 
2298 	*bg_blkno = ocfs2_which_cluster_group(inode,
2299 					      data_cluster);
2300 
2301 	if (*bg_blkno == osb->first_cluster_group_blkno)
2302 		*bg_bit_off = (u16) data_cluster;
2303 	else
2304 		*bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2305 							     data_blkno - *bg_blkno);
2306 }
2307 
2308 /*
2309  * min_bits - minimum contiguous chunk from this total allocation we
2310  * can handle. set to what we asked for originally for a full
2311  * contig. allocation, set to '1' to indicate we can deal with extents
2312  * of any size.
2313  */
2314 int __ocfs2_claim_clusters(handle_t *handle,
2315 			   struct ocfs2_alloc_context *ac,
2316 			   u32 min_clusters,
2317 			   u32 max_clusters,
2318 			   u32 *cluster_start,
2319 			   u32 *num_clusters)
2320 {
2321 	int status;
2322 	unsigned int bits_wanted = max_clusters;
2323 	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2324 	struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2325 
2326 	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2327 
2328 	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2329 	       && ac->ac_which != OCFS2_AC_USE_MAIN);
2330 
2331 	if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2332 		WARN_ON(min_clusters > 1);
2333 
2334 		status = ocfs2_claim_local_alloc_bits(osb,
2335 						      handle,
2336 						      ac,
2337 						      bits_wanted,
2338 						      cluster_start,
2339 						      num_clusters);
2340 		if (!status)
2341 			atomic_inc(&osb->alloc_stats.local_data);
2342 	} else {
2343 		if (min_clusters > (osb->bitmap_cpg - 1)) {
2344 			/* The only paths asking for contiguousness
2345 			 * should know about this already. */
2346 			mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2347 			     "group bitmap size %u!\n", min_clusters,
2348 			     osb->bitmap_cpg);
2349 			status = -ENOSPC;
2350 			goto bail;
2351 		}
2352 		/* clamp the current request down to a realistic size. */
2353 		if (bits_wanted > (osb->bitmap_cpg - 1))
2354 			bits_wanted = osb->bitmap_cpg - 1;
2355 
2356 		status = ocfs2_claim_suballoc_bits(ac,
2357 						   handle,
2358 						   bits_wanted,
2359 						   min_clusters,
2360 						   &res);
2361 		if (!status) {
2362 			BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2363 			*cluster_start =
2364 				ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2365 								 res.sr_bg_blkno,
2366 								 res.sr_bit_offset);
2367 			atomic_inc(&osb->alloc_stats.bitmap_data);
2368 			*num_clusters = res.sr_bits;
2369 		}
2370 	}
2371 	if (status < 0) {
2372 		if (status != -ENOSPC)
2373 			mlog_errno(status);
2374 		goto bail;
2375 	}
2376 
2377 	ac->ac_bits_given += *num_clusters;
2378 
2379 bail:
2380 	if (status)
2381 		mlog_errno(status);
2382 	return status;
2383 }
2384 
2385 int ocfs2_claim_clusters(handle_t *handle,
2386 			 struct ocfs2_alloc_context *ac,
2387 			 u32 min_clusters,
2388 			 u32 *cluster_start,
2389 			 u32 *num_clusters)
2390 {
2391 	unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2392 
2393 	return __ocfs2_claim_clusters(handle, ac, min_clusters,
2394 				      bits_wanted, cluster_start, num_clusters);
2395 }
2396 
2397 static int ocfs2_block_group_clear_bits(handle_t *handle,
2398 					struct inode *alloc_inode,
2399 					struct ocfs2_group_desc *bg,
2400 					struct buffer_head *group_bh,
2401 					unsigned int bit_off,
2402 					unsigned int num_bits,
2403 					void (*undo_fn)(unsigned int bit,
2404 							unsigned long *bmap))
2405 {
2406 	int status;
2407 	unsigned int tmp;
2408 	struct ocfs2_group_desc *undo_bg = NULL;
2409 
2410 	/* The caller got this descriptor from
2411 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
2412 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2413 
2414 	trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2415 
2416 	BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2417 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2418 					 group_bh,
2419 					 undo_fn ?
2420 					 OCFS2_JOURNAL_ACCESS_UNDO :
2421 					 OCFS2_JOURNAL_ACCESS_WRITE);
2422 	if (status < 0) {
2423 		mlog_errno(status);
2424 		goto bail;
2425 	}
2426 
2427 	if (undo_fn) {
2428 		jbd_lock_bh_state(group_bh);
2429 		undo_bg = (struct ocfs2_group_desc *)
2430 					bh2jh(group_bh)->b_committed_data;
2431 		BUG_ON(!undo_bg);
2432 	}
2433 
2434 	tmp = num_bits;
2435 	while(tmp--) {
2436 		ocfs2_clear_bit((bit_off + tmp),
2437 				(unsigned long *) bg->bg_bitmap);
2438 		if (undo_fn)
2439 			undo_fn(bit_off + tmp,
2440 				(unsigned long *) undo_bg->bg_bitmap);
2441 	}
2442 	le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2443 	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2444 		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
2445 				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
2446 				   le16_to_cpu(bg->bg_bits),
2447 				   le16_to_cpu(bg->bg_free_bits_count),
2448 				   num_bits);
2449 	}
2450 
2451 	if (undo_fn)
2452 		jbd_unlock_bh_state(group_bh);
2453 
2454 	ocfs2_journal_dirty(handle, group_bh);
2455 bail:
2456 	return status;
2457 }
2458 
2459 /*
2460  * expects the suballoc inode to already be locked.
2461  */
2462 static int _ocfs2_free_suballoc_bits(handle_t *handle,
2463 				     struct inode *alloc_inode,
2464 				     struct buffer_head *alloc_bh,
2465 				     unsigned int start_bit,
2466 				     u64 bg_blkno,
2467 				     unsigned int count,
2468 				     void (*undo_fn)(unsigned int bit,
2469 						     unsigned long *bitmap))
2470 {
2471 	int status = 0;
2472 	u32 tmp_used;
2473 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2474 	struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2475 	struct buffer_head *group_bh = NULL;
2476 	struct ocfs2_group_desc *group;
2477 
2478 	/* The alloc_bh comes from ocfs2_free_dinode() or
2479 	 * ocfs2_free_clusters().  The callers have all locked the
2480 	 * allocator and gotten alloc_bh from the lock call.  This
2481 	 * validates the dinode buffer.  Any corruption that has happened
2482 	 * is a code bug. */
2483 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2484 	BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2485 
2486 	trace_ocfs2_free_suballoc_bits(
2487 		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2488 		(unsigned long long)bg_blkno,
2489 		start_bit, count);
2490 
2491 	status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2492 					     &group_bh);
2493 	if (status < 0) {
2494 		mlog_errno(status);
2495 		goto bail;
2496 	}
2497 	group = (struct ocfs2_group_desc *) group_bh->b_data;
2498 
2499 	BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2500 
2501 	status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2502 					      group, group_bh,
2503 					      start_bit, count, undo_fn);
2504 	if (status < 0) {
2505 		mlog_errno(status);
2506 		goto bail;
2507 	}
2508 
2509 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2510 					 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2511 	if (status < 0) {
2512 		mlog_errno(status);
2513 		ocfs2_block_group_set_bits(handle, alloc_inode, group, group_bh,
2514 				start_bit, count);
2515 		goto bail;
2516 	}
2517 
2518 	le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2519 		     count);
2520 	tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2521 	fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2522 	ocfs2_journal_dirty(handle, alloc_bh);
2523 
2524 bail:
2525 	brelse(group_bh);
2526 
2527 	if (status)
2528 		mlog_errno(status);
2529 	return status;
2530 }
2531 
2532 int ocfs2_free_suballoc_bits(handle_t *handle,
2533 			     struct inode *alloc_inode,
2534 			     struct buffer_head *alloc_bh,
2535 			     unsigned int start_bit,
2536 			     u64 bg_blkno,
2537 			     unsigned int count)
2538 {
2539 	return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2540 					 start_bit, bg_blkno, count, NULL);
2541 }
2542 
2543 int ocfs2_free_dinode(handle_t *handle,
2544 		      struct inode *inode_alloc_inode,
2545 		      struct buffer_head *inode_alloc_bh,
2546 		      struct ocfs2_dinode *di)
2547 {
2548 	u64 blk = le64_to_cpu(di->i_blkno);
2549 	u16 bit = le16_to_cpu(di->i_suballoc_bit);
2550 	u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2551 
2552 	if (di->i_suballoc_loc)
2553 		bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2554 	return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2555 					inode_alloc_bh, bit, bg_blkno, 1);
2556 }
2557 
2558 static int _ocfs2_free_clusters(handle_t *handle,
2559 				struct inode *bitmap_inode,
2560 				struct buffer_head *bitmap_bh,
2561 				u64 start_blk,
2562 				unsigned int num_clusters,
2563 				void (*undo_fn)(unsigned int bit,
2564 						unsigned long *bitmap))
2565 {
2566 	int status;
2567 	u16 bg_start_bit;
2568 	u64 bg_blkno;
2569 	struct ocfs2_dinode *fe;
2570 
2571 	/* You can't ever have a contiguous set of clusters
2572 	 * bigger than a block group bitmap so we never have to worry
2573 	 * about looping on them.
2574 	 * This is expensive. We can safely remove once this stuff has
2575 	 * gotten tested really well. */
2576 	BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
2577 
2578 	fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
2579 
2580 	ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2581 				     &bg_start_bit);
2582 
2583 	trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2584 			(unsigned long long)start_blk,
2585 			bg_start_bit, num_clusters);
2586 
2587 	status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2588 					   bg_start_bit, bg_blkno,
2589 					   num_clusters, undo_fn);
2590 	if (status < 0) {
2591 		mlog_errno(status);
2592 		goto out;
2593 	}
2594 
2595 	ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2596 					 num_clusters);
2597 
2598 out:
2599 	if (status)
2600 		mlog_errno(status);
2601 	return status;
2602 }
2603 
2604 int ocfs2_free_clusters(handle_t *handle,
2605 			struct inode *bitmap_inode,
2606 			struct buffer_head *bitmap_bh,
2607 			u64 start_blk,
2608 			unsigned int num_clusters)
2609 {
2610 	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2611 				    start_blk, num_clusters,
2612 				    _ocfs2_set_bit);
2613 }
2614 
2615 /*
2616  * Give never-used clusters back to the global bitmap.  We don't need
2617  * to protect these bits in the undo buffer.
2618  */
2619 int ocfs2_release_clusters(handle_t *handle,
2620 			   struct inode *bitmap_inode,
2621 			   struct buffer_head *bitmap_bh,
2622 			   u64 start_blk,
2623 			   unsigned int num_clusters)
2624 {
2625 	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2626 				    start_blk, num_clusters,
2627 				    _ocfs2_clear_bit);
2628 }
2629 
2630 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
2631 {
2632 	printk("Block Group:\n");
2633 	printk("bg_signature:       %s\n", bg->bg_signature);
2634 	printk("bg_size:            %u\n", bg->bg_size);
2635 	printk("bg_bits:            %u\n", bg->bg_bits);
2636 	printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
2637 	printk("bg_chain:           %u\n", bg->bg_chain);
2638 	printk("bg_generation:      %u\n", le32_to_cpu(bg->bg_generation));
2639 	printk("bg_next_group:      %llu\n",
2640 	       (unsigned long long)bg->bg_next_group);
2641 	printk("bg_parent_dinode:   %llu\n",
2642 	       (unsigned long long)bg->bg_parent_dinode);
2643 	printk("bg_blkno:           %llu\n",
2644 	       (unsigned long long)bg->bg_blkno);
2645 }
2646 
2647 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2648 {
2649 	int i;
2650 
2651 	printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
2652 	printk("i_signature:                  %s\n", fe->i_signature);
2653 	printk("i_size:                       %llu\n",
2654 	       (unsigned long long)fe->i_size);
2655 	printk("i_clusters:                   %u\n", fe->i_clusters);
2656 	printk("i_generation:                 %u\n",
2657 	       le32_to_cpu(fe->i_generation));
2658 	printk("id1.bitmap1.i_used:           %u\n",
2659 	       le32_to_cpu(fe->id1.bitmap1.i_used));
2660 	printk("id1.bitmap1.i_total:          %u\n",
2661 	       le32_to_cpu(fe->id1.bitmap1.i_total));
2662 	printk("id2.i_chain.cl_cpg:           %u\n", fe->id2.i_chain.cl_cpg);
2663 	printk("id2.i_chain.cl_bpc:           %u\n", fe->id2.i_chain.cl_bpc);
2664 	printk("id2.i_chain.cl_count:         %u\n", fe->id2.i_chain.cl_count);
2665 	printk("id2.i_chain.cl_next_free_rec: %u\n",
2666 	       fe->id2.i_chain.cl_next_free_rec);
2667 	for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2668 		printk("fe->id2.i_chain.cl_recs[%d].c_free:  %u\n", i,
2669 		       fe->id2.i_chain.cl_recs[i].c_free);
2670 		printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2671 		       fe->id2.i_chain.cl_recs[i].c_total);
2672 		printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2673 		       (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
2674 	}
2675 }
2676 
2677 /*
2678  * For a given allocation, determine which allocators will need to be
2679  * accessed, and lock them, reserving the appropriate number of bits.
2680  *
2681  * Sparse file systems call this from ocfs2_write_begin_nolock()
2682  * and ocfs2_allocate_unwritten_extents().
2683  *
2684  * File systems which don't support holes call this from
2685  * ocfs2_extend_allocation().
2686  */
2687 int ocfs2_lock_allocators(struct inode *inode,
2688 			  struct ocfs2_extent_tree *et,
2689 			  u32 clusters_to_add, u32 extents_to_split,
2690 			  struct ocfs2_alloc_context **data_ac,
2691 			  struct ocfs2_alloc_context **meta_ac)
2692 {
2693 	int ret = 0, num_free_extents;
2694 	unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2695 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2696 
2697 	*meta_ac = NULL;
2698 	if (data_ac)
2699 		*data_ac = NULL;
2700 
2701 	BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2702 
2703 	num_free_extents = ocfs2_num_free_extents(osb, et);
2704 	if (num_free_extents < 0) {
2705 		ret = num_free_extents;
2706 		mlog_errno(ret);
2707 		goto out;
2708 	}
2709 
2710 	/*
2711 	 * Sparse allocation file systems need to be more conservative
2712 	 * with reserving room for expansion - the actual allocation
2713 	 * happens while we've got a journal handle open so re-taking
2714 	 * a cluster lock (because we ran out of room for another
2715 	 * extent) will violate ordering rules.
2716 	 *
2717 	 * Most of the time we'll only be seeing this 1 cluster at a time
2718 	 * anyway.
2719 	 *
2720 	 * Always lock for any unwritten extents - we might want to
2721 	 * add blocks during a split.
2722 	 */
2723 	if (!num_free_extents ||
2724 	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2725 		ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2726 		if (ret < 0) {
2727 			if (ret != -ENOSPC)
2728 				mlog_errno(ret);
2729 			goto out;
2730 		}
2731 	}
2732 
2733 	if (clusters_to_add == 0)
2734 		goto out;
2735 
2736 	ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2737 	if (ret < 0) {
2738 		if (ret != -ENOSPC)
2739 			mlog_errno(ret);
2740 		goto out;
2741 	}
2742 
2743 out:
2744 	if (ret) {
2745 		if (*meta_ac) {
2746 			ocfs2_free_alloc_context(*meta_ac);
2747 			*meta_ac = NULL;
2748 		}
2749 
2750 		/*
2751 		 * We cannot have an error and a non null *data_ac.
2752 		 */
2753 	}
2754 
2755 	return ret;
2756 }
2757 
2758 /*
2759  * Read the inode specified by blkno to get suballoc_slot and
2760  * suballoc_bit.
2761  */
2762 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2763 				       u16 *suballoc_slot, u64 *group_blkno,
2764 				       u16 *suballoc_bit)
2765 {
2766 	int status;
2767 	struct buffer_head *inode_bh = NULL;
2768 	struct ocfs2_dinode *inode_fe;
2769 
2770 	trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2771 
2772 	/* dirty read disk */
2773 	status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2774 	if (status < 0) {
2775 		mlog(ML_ERROR, "read block %llu failed %d\n",
2776 		     (unsigned long long)blkno, status);
2777 		goto bail;
2778 	}
2779 
2780 	inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2781 	if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2782 		mlog(ML_ERROR, "invalid inode %llu requested\n",
2783 		     (unsigned long long)blkno);
2784 		status = -EINVAL;
2785 		goto bail;
2786 	}
2787 
2788 	if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2789 	    (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2790 		mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2791 		     (unsigned long long)blkno,
2792 		     (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2793 		status = -EINVAL;
2794 		goto bail;
2795 	}
2796 
2797 	if (suballoc_slot)
2798 		*suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2799 	if (suballoc_bit)
2800 		*suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2801 	if (group_blkno)
2802 		*group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2803 
2804 bail:
2805 	brelse(inode_bh);
2806 
2807 	if (status)
2808 		mlog_errno(status);
2809 	return status;
2810 }
2811 
2812 /*
2813  * test whether bit is SET in allocator bitmap or not.  on success, 0
2814  * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
2815  * is returned and *res is meaningless.  Call this after you have
2816  * cluster locked against suballoc, or you may get a result based on
2817  * non-up2date contents
2818  */
2819 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2820 				   struct inode *suballoc,
2821 				   struct buffer_head *alloc_bh,
2822 				   u64 group_blkno, u64 blkno,
2823 				   u16 bit, int *res)
2824 {
2825 	struct ocfs2_dinode *alloc_di;
2826 	struct ocfs2_group_desc *group;
2827 	struct buffer_head *group_bh = NULL;
2828 	u64 bg_blkno;
2829 	int status;
2830 
2831 	trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2832 				      (unsigned int)bit);
2833 
2834 	alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2835 	if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2836 		mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2837 		     (unsigned int)bit,
2838 		     ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2839 		status = -EINVAL;
2840 		goto bail;
2841 	}
2842 
2843 	bg_blkno = group_blkno ? group_blkno :
2844 		   ocfs2_which_suballoc_group(blkno, bit);
2845 	status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2846 					     &group_bh);
2847 	if (status < 0) {
2848 		mlog(ML_ERROR, "read group %llu failed %d\n",
2849 		     (unsigned long long)bg_blkno, status);
2850 		goto bail;
2851 	}
2852 
2853 	group = (struct ocfs2_group_desc *) group_bh->b_data;
2854 	*res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2855 
2856 bail:
2857 	brelse(group_bh);
2858 
2859 	if (status)
2860 		mlog_errno(status);
2861 	return status;
2862 }
2863 
2864 /*
2865  * Test if the bit representing this inode (blkno) is set in the
2866  * suballocator.
2867  *
2868  * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2869  *
2870  * In the event of failure, a negative value is returned and *res is
2871  * meaningless.
2872  *
2873  * Callers must make sure to hold nfs_sync_lock to prevent
2874  * ocfs2_delete_inode() on another node from accessing the same
2875  * suballocator concurrently.
2876  */
2877 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2878 {
2879 	int status;
2880 	u64 group_blkno = 0;
2881 	u16 suballoc_bit = 0, suballoc_slot = 0;
2882 	struct inode *inode_alloc_inode;
2883 	struct buffer_head *alloc_bh = NULL;
2884 
2885 	trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2886 
2887 	status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2888 					     &group_blkno, &suballoc_bit);
2889 	if (status < 0) {
2890 		mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2891 		goto bail;
2892 	}
2893 
2894 	inode_alloc_inode =
2895 		ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2896 					    suballoc_slot);
2897 	if (!inode_alloc_inode) {
2898 		/* the error code could be inaccurate, but we are not able to
2899 		 * get the correct one. */
2900 		status = -EINVAL;
2901 		mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2902 		     (u32)suballoc_slot);
2903 		goto bail;
2904 	}
2905 
2906 	inode_lock(inode_alloc_inode);
2907 	status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2908 	if (status < 0) {
2909 		inode_unlock(inode_alloc_inode);
2910 		iput(inode_alloc_inode);
2911 		mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2912 		     (u32)suballoc_slot, status);
2913 		goto bail;
2914 	}
2915 
2916 	status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2917 					 group_blkno, blkno, suballoc_bit, res);
2918 	if (status < 0)
2919 		mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2920 
2921 	ocfs2_inode_unlock(inode_alloc_inode, 0);
2922 	inode_unlock(inode_alloc_inode);
2923 
2924 	iput(inode_alloc_inode);
2925 	brelse(alloc_bh);
2926 bail:
2927 	if (status)
2928 		mlog_errno(status);
2929 	return status;
2930 }
2931