xref: /linux/fs/ocfs2/suballoc.c (revision 93d90ad708b8da6efc0e487b66111aa9db7f70c7)
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 		mutex_unlock(&inode->i_mutex);
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 	if (ac->ac_find_loc_priv) {
153 		kfree(ac->ac_find_loc_priv);
154 		ac->ac_find_loc_priv = NULL;
155 	}
156 }
157 
158 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
159 {
160 	ocfs2_free_ac_resource(ac);
161 	kfree(ac);
162 }
163 
164 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
165 {
166 	return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
167 }
168 
169 #define do_error(fmt, ...)						\
170 	do{								\
171 		if (resize)					\
172 			mlog(ML_ERROR, fmt "\n", ##__VA_ARGS__);	\
173 		else							\
174 			ocfs2_error(sb, fmt, ##__VA_ARGS__);		\
175 	} while (0)
176 
177 static int ocfs2_validate_gd_self(struct super_block *sb,
178 				  struct buffer_head *bh,
179 				  int resize)
180 {
181 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
182 
183 	if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
184 		do_error("Group descriptor #%llu has bad signature %.*s",
185 			 (unsigned long long)bh->b_blocknr, 7,
186 			 gd->bg_signature);
187 		return -EINVAL;
188 	}
189 
190 	if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
191 		do_error("Group descriptor #%llu has an invalid bg_blkno "
192 			 "of %llu",
193 			 (unsigned long long)bh->b_blocknr,
194 			 (unsigned long long)le64_to_cpu(gd->bg_blkno));
195 		return -EINVAL;
196 	}
197 
198 	if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
199 		do_error("Group descriptor #%llu has an invalid "
200 			 "fs_generation of #%u",
201 			 (unsigned long long)bh->b_blocknr,
202 			 le32_to_cpu(gd->bg_generation));
203 		return -EINVAL;
204 	}
205 
206 	if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
207 		do_error("Group descriptor #%llu has bit count %u but "
208 			 "claims that %u are free",
209 			 (unsigned long long)bh->b_blocknr,
210 			 le16_to_cpu(gd->bg_bits),
211 			 le16_to_cpu(gd->bg_free_bits_count));
212 		return -EINVAL;
213 	}
214 
215 	if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
216 		do_error("Group descriptor #%llu has bit count %u but "
217 			 "max bitmap bits of %u",
218 			 (unsigned long long)bh->b_blocknr,
219 			 le16_to_cpu(gd->bg_bits),
220 			 8 * le16_to_cpu(gd->bg_size));
221 		return -EINVAL;
222 	}
223 
224 	return 0;
225 }
226 
227 static int ocfs2_validate_gd_parent(struct super_block *sb,
228 				    struct ocfs2_dinode *di,
229 				    struct buffer_head *bh,
230 				    int resize)
231 {
232 	unsigned int max_bits;
233 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
234 
235 	if (di->i_blkno != gd->bg_parent_dinode) {
236 		do_error("Group descriptor #%llu has bad parent "
237 			 "pointer (%llu, expected %llu)",
238 			 (unsigned long long)bh->b_blocknr,
239 			 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
240 			 (unsigned long long)le64_to_cpu(di->i_blkno));
241 		return -EINVAL;
242 	}
243 
244 	max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
245 	if (le16_to_cpu(gd->bg_bits) > max_bits) {
246 		do_error("Group descriptor #%llu has bit count of %u",
247 			 (unsigned long long)bh->b_blocknr,
248 			 le16_to_cpu(gd->bg_bits));
249 		return -EINVAL;
250 	}
251 
252 	/* In resize, we may meet the case bg_chain == cl_next_free_rec. */
253 	if ((le16_to_cpu(gd->bg_chain) >
254 	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
255 	    ((le16_to_cpu(gd->bg_chain) ==
256 	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
257 		do_error("Group descriptor #%llu has bad chain %u",
258 			 (unsigned long long)bh->b_blocknr,
259 			 le16_to_cpu(gd->bg_chain));
260 		return -EINVAL;
261 	}
262 
263 	return 0;
264 }
265 
266 #undef do_error
267 
268 /*
269  * This version only prints errors.  It does not fail the filesystem, and
270  * exists only for resize.
271  */
272 int ocfs2_check_group_descriptor(struct super_block *sb,
273 				 struct ocfs2_dinode *di,
274 				 struct buffer_head *bh)
275 {
276 	int rc;
277 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
278 
279 	BUG_ON(!buffer_uptodate(bh));
280 
281 	/*
282 	 * If the ecc fails, we return the error but otherwise
283 	 * leave the filesystem running.  We know any error is
284 	 * local to this block.
285 	 */
286 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
287 	if (rc) {
288 		mlog(ML_ERROR,
289 		     "Checksum failed for group descriptor %llu\n",
290 		     (unsigned long long)bh->b_blocknr);
291 	} else
292 		rc = ocfs2_validate_gd_self(sb, bh, 1);
293 	if (!rc)
294 		rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
295 
296 	return rc;
297 }
298 
299 static int ocfs2_validate_group_descriptor(struct super_block *sb,
300 					   struct buffer_head *bh)
301 {
302 	int rc;
303 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
304 
305 	trace_ocfs2_validate_group_descriptor(
306 					(unsigned long long)bh->b_blocknr);
307 
308 	BUG_ON(!buffer_uptodate(bh));
309 
310 	/*
311 	 * If the ecc fails, we return the error but otherwise
312 	 * leave the filesystem running.  We know any error is
313 	 * local to this block.
314 	 */
315 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
316 	if (rc)
317 		return rc;
318 
319 	/*
320 	 * Errors after here are fatal.
321 	 */
322 
323 	return ocfs2_validate_gd_self(sb, bh, 0);
324 }
325 
326 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
327 				u64 gd_blkno, struct buffer_head **bh)
328 {
329 	int rc;
330 	struct buffer_head *tmp = *bh;
331 
332 	rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
333 			      ocfs2_validate_group_descriptor);
334 	if (rc)
335 		goto out;
336 
337 	rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
338 	if (rc) {
339 		brelse(tmp);
340 		goto out;
341 	}
342 
343 	/* If ocfs2_read_block() got us a new bh, pass it up. */
344 	if (!*bh)
345 		*bh = tmp;
346 
347 out:
348 	return rc;
349 }
350 
351 static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
352 					  struct ocfs2_group_desc *bg,
353 					  struct ocfs2_chain_list *cl,
354 					  u64 p_blkno, unsigned int clusters)
355 {
356 	struct ocfs2_extent_list *el = &bg->bg_list;
357 	struct ocfs2_extent_rec *rec;
358 
359 	BUG_ON(!ocfs2_supports_discontig_bg(osb));
360 	if (!el->l_next_free_rec)
361 		el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
362 	rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
363 	rec->e_blkno = cpu_to_le64(p_blkno);
364 	rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
365 				  le16_to_cpu(cl->cl_bpc));
366 	rec->e_leaf_clusters = cpu_to_le16(clusters);
367 	le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
368 	le16_add_cpu(&bg->bg_free_bits_count,
369 		     clusters * le16_to_cpu(cl->cl_bpc));
370 	le16_add_cpu(&el->l_next_free_rec, 1);
371 }
372 
373 static int ocfs2_block_group_fill(handle_t *handle,
374 				  struct inode *alloc_inode,
375 				  struct buffer_head *bg_bh,
376 				  u64 group_blkno,
377 				  unsigned int group_clusters,
378 				  u16 my_chain,
379 				  struct ocfs2_chain_list *cl)
380 {
381 	int status = 0;
382 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
383 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
384 	struct super_block * sb = alloc_inode->i_sb;
385 
386 	if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
387 		ocfs2_error(alloc_inode->i_sb, "group block (%llu) != "
388 			    "b_blocknr (%llu)",
389 			    (unsigned long long)group_blkno,
390 			    (unsigned long long) bg_bh->b_blocknr);
391 		status = -EIO;
392 		goto bail;
393 	}
394 
395 	status = ocfs2_journal_access_gd(handle,
396 					 INODE_CACHE(alloc_inode),
397 					 bg_bh,
398 					 OCFS2_JOURNAL_ACCESS_CREATE);
399 	if (status < 0) {
400 		mlog_errno(status);
401 		goto bail;
402 	}
403 
404 	memset(bg, 0, sb->s_blocksize);
405 	strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
406 	bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
407 	bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
408 						osb->s_feature_incompat));
409 	bg->bg_chain = cpu_to_le16(my_chain);
410 	bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
411 	bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
412 	bg->bg_blkno = cpu_to_le64(group_blkno);
413 	if (group_clusters == le16_to_cpu(cl->cl_cpg))
414 		bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
415 	else
416 		ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
417 					      group_clusters);
418 
419 	/* set the 1st bit in the bitmap to account for the descriptor block */
420 	ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
421 	bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
422 
423 	ocfs2_journal_dirty(handle, bg_bh);
424 
425 	/* There is no need to zero out or otherwise initialize the
426 	 * other blocks in a group - All valid FS metadata in a block
427 	 * group stores the superblock fs_generation value at
428 	 * allocation time. */
429 
430 bail:
431 	if (status)
432 		mlog_errno(status);
433 	return status;
434 }
435 
436 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
437 {
438 	u16 curr, best;
439 
440 	best = curr = 0;
441 	while (curr < le16_to_cpu(cl->cl_count)) {
442 		if (le32_to_cpu(cl->cl_recs[best].c_total) >
443 		    le32_to_cpu(cl->cl_recs[curr].c_total))
444 			best = curr;
445 		curr++;
446 	}
447 	return best;
448 }
449 
450 static struct buffer_head *
451 ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
452 			       struct inode *alloc_inode,
453 			       struct ocfs2_alloc_context *ac,
454 			       struct ocfs2_chain_list *cl)
455 {
456 	int status;
457 	u32 bit_off, num_bits;
458 	u64 bg_blkno;
459 	struct buffer_head *bg_bh;
460 	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
461 
462 	status = ocfs2_claim_clusters(handle, ac,
463 				      le16_to_cpu(cl->cl_cpg), &bit_off,
464 				      &num_bits);
465 	if (status < 0) {
466 		if (status != -ENOSPC)
467 			mlog_errno(status);
468 		goto bail;
469 	}
470 
471 	/* setup the group */
472 	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
473 	trace_ocfs2_block_group_alloc_contig(
474 	     (unsigned long long)bg_blkno, alloc_rec);
475 
476 	bg_bh = sb_getblk(osb->sb, bg_blkno);
477 	if (!bg_bh) {
478 		status = -ENOMEM;
479 		mlog_errno(status);
480 		goto bail;
481 	}
482 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
483 
484 	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
485 					bg_blkno, num_bits, alloc_rec, cl);
486 	if (status < 0) {
487 		brelse(bg_bh);
488 		mlog_errno(status);
489 	}
490 
491 bail:
492 	return status ? ERR_PTR(status) : bg_bh;
493 }
494 
495 static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
496 					handle_t *handle,
497 					struct ocfs2_alloc_context *ac,
498 					unsigned int min_bits,
499 					u32 *bit_off, u32 *num_bits)
500 {
501 	int status = 0;
502 
503 	while (min_bits) {
504 		status = ocfs2_claim_clusters(handle, ac, min_bits,
505 					      bit_off, num_bits);
506 		if (status != -ENOSPC)
507 			break;
508 
509 		min_bits >>= 1;
510 	}
511 
512 	return status;
513 }
514 
515 static int ocfs2_block_group_grow_discontig(handle_t *handle,
516 					    struct inode *alloc_inode,
517 					    struct buffer_head *bg_bh,
518 					    struct ocfs2_alloc_context *ac,
519 					    struct ocfs2_chain_list *cl,
520 					    unsigned int min_bits)
521 {
522 	int status;
523 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
524 	struct ocfs2_group_desc *bg =
525 		(struct ocfs2_group_desc *)bg_bh->b_data;
526 	unsigned int needed = le16_to_cpu(cl->cl_cpg) -
527 			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
528 	u32 p_cpos, clusters;
529 	u64 p_blkno;
530 	struct ocfs2_extent_list *el = &bg->bg_list;
531 
532 	status = ocfs2_journal_access_gd(handle,
533 					 INODE_CACHE(alloc_inode),
534 					 bg_bh,
535 					 OCFS2_JOURNAL_ACCESS_CREATE);
536 	if (status < 0) {
537 		mlog_errno(status);
538 		goto bail;
539 	}
540 
541 	while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
542 				le16_to_cpu(el->l_count))) {
543 		if (min_bits > needed)
544 			min_bits = needed;
545 		status = ocfs2_block_group_claim_bits(osb, handle, ac,
546 						      min_bits, &p_cpos,
547 						      &clusters);
548 		if (status < 0) {
549 			if (status != -ENOSPC)
550 				mlog_errno(status);
551 			goto bail;
552 		}
553 		p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
554 		ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
555 					      clusters);
556 
557 		min_bits = clusters;
558 		needed = le16_to_cpu(cl->cl_cpg) -
559 			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
560 	}
561 
562 	if (needed > 0) {
563 		/*
564 		 * We have used up all the extent rec but can't fill up
565 		 * the cpg. So bail out.
566 		 */
567 		status = -ENOSPC;
568 		goto bail;
569 	}
570 
571 	ocfs2_journal_dirty(handle, bg_bh);
572 
573 bail:
574 	return status;
575 }
576 
577 static void ocfs2_bg_alloc_cleanup(handle_t *handle,
578 				   struct ocfs2_alloc_context *cluster_ac,
579 				   struct inode *alloc_inode,
580 				   struct buffer_head *bg_bh)
581 {
582 	int i, ret;
583 	struct ocfs2_group_desc *bg;
584 	struct ocfs2_extent_list *el;
585 	struct ocfs2_extent_rec *rec;
586 
587 	if (!bg_bh)
588 		return;
589 
590 	bg = (struct ocfs2_group_desc *)bg_bh->b_data;
591 	el = &bg->bg_list;
592 	for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
593 		rec = &el->l_recs[i];
594 		ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
595 					  cluster_ac->ac_bh,
596 					  le64_to_cpu(rec->e_blkno),
597 					  le16_to_cpu(rec->e_leaf_clusters));
598 		if (ret)
599 			mlog_errno(ret);
600 		/* Try all the clusters to free */
601 	}
602 
603 	ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
604 	brelse(bg_bh);
605 }
606 
607 static struct buffer_head *
608 ocfs2_block_group_alloc_discontig(handle_t *handle,
609 				  struct inode *alloc_inode,
610 				  struct ocfs2_alloc_context *ac,
611 				  struct ocfs2_chain_list *cl)
612 {
613 	int status;
614 	u32 bit_off, num_bits;
615 	u64 bg_blkno;
616 	unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
617 	struct buffer_head *bg_bh = NULL;
618 	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
619 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
620 
621 	if (!ocfs2_supports_discontig_bg(osb)) {
622 		status = -ENOSPC;
623 		goto bail;
624 	}
625 
626 	status = ocfs2_extend_trans(handle,
627 				    ocfs2_calc_bg_discontig_credits(osb->sb));
628 	if (status) {
629 		mlog_errno(status);
630 		goto bail;
631 	}
632 
633 	/*
634 	 * We're going to be grabbing from multiple cluster groups.
635 	 * We don't have enough credits to relink them all, and the
636 	 * cluster groups will be staying in cache for the duration of
637 	 * this operation.
638 	 */
639 	ac->ac_disable_chain_relink = 1;
640 
641 	/* Claim the first region */
642 	status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
643 					      &bit_off, &num_bits);
644 	if (status < 0) {
645 		if (status != -ENOSPC)
646 			mlog_errno(status);
647 		goto bail;
648 	}
649 	min_bits = num_bits;
650 
651 	/* setup the group */
652 	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
653 	trace_ocfs2_block_group_alloc_discontig(
654 				(unsigned long long)bg_blkno, alloc_rec);
655 
656 	bg_bh = sb_getblk(osb->sb, bg_blkno);
657 	if (!bg_bh) {
658 		status = -ENOMEM;
659 		mlog_errno(status);
660 		goto bail;
661 	}
662 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
663 
664 	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
665 					bg_blkno, num_bits, alloc_rec, cl);
666 	if (status < 0) {
667 		mlog_errno(status);
668 		goto bail;
669 	}
670 
671 	status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
672 						  bg_bh, ac, cl, min_bits);
673 	if (status)
674 		mlog_errno(status);
675 
676 bail:
677 	if (status)
678 		ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
679 	return status ? ERR_PTR(status) : bg_bh;
680 }
681 
682 /*
683  * We expect the block group allocator to already be locked.
684  */
685 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
686 				   struct inode *alloc_inode,
687 				   struct buffer_head *bh,
688 				   u64 max_block,
689 				   u64 *last_alloc_group,
690 				   int flags)
691 {
692 	int status, credits;
693 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
694 	struct ocfs2_chain_list *cl;
695 	struct ocfs2_alloc_context *ac = NULL;
696 	handle_t *handle = NULL;
697 	u16 alloc_rec;
698 	struct buffer_head *bg_bh = NULL;
699 	struct ocfs2_group_desc *bg;
700 
701 	BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
702 
703 	cl = &fe->id2.i_chain;
704 	status = ocfs2_reserve_clusters_with_limit(osb,
705 						   le16_to_cpu(cl->cl_cpg),
706 						   max_block, flags, &ac);
707 	if (status < 0) {
708 		if (status != -ENOSPC)
709 			mlog_errno(status);
710 		goto bail;
711 	}
712 
713 	credits = ocfs2_calc_group_alloc_credits(osb->sb,
714 						 le16_to_cpu(cl->cl_cpg));
715 	handle = ocfs2_start_trans(osb, credits);
716 	if (IS_ERR(handle)) {
717 		status = PTR_ERR(handle);
718 		handle = NULL;
719 		mlog_errno(status);
720 		goto bail;
721 	}
722 
723 	if (last_alloc_group && *last_alloc_group != 0) {
724 		trace_ocfs2_block_group_alloc(
725 				(unsigned long long)*last_alloc_group);
726 		ac->ac_last_group = *last_alloc_group;
727 	}
728 
729 	bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
730 					       ac, cl);
731 	if (IS_ERR(bg_bh) && (PTR_ERR(bg_bh) == -ENOSPC))
732 		bg_bh = ocfs2_block_group_alloc_discontig(handle,
733 							  alloc_inode,
734 							  ac, cl);
735 	if (IS_ERR(bg_bh)) {
736 		status = PTR_ERR(bg_bh);
737 		bg_bh = NULL;
738 		if (status != -ENOSPC)
739 			mlog_errno(status);
740 		goto bail;
741 	}
742 	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
743 
744 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
745 					 bh, OCFS2_JOURNAL_ACCESS_WRITE);
746 	if (status < 0) {
747 		mlog_errno(status);
748 		goto bail;
749 	}
750 
751 	alloc_rec = le16_to_cpu(bg->bg_chain);
752 	le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
753 		     le16_to_cpu(bg->bg_free_bits_count));
754 	le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
755 		     le16_to_cpu(bg->bg_bits));
756 	cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
757 	if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
758 		le16_add_cpu(&cl->cl_next_free_rec, 1);
759 
760 	le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
761 					le16_to_cpu(bg->bg_free_bits_count));
762 	le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
763 	le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
764 
765 	ocfs2_journal_dirty(handle, bh);
766 
767 	spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
768 	OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
769 	fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
770 					     le32_to_cpu(fe->i_clusters)));
771 	spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
772 	i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
773 	alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
774 	ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
775 
776 	status = 0;
777 
778 	/* save the new last alloc group so that the caller can cache it. */
779 	if (last_alloc_group)
780 		*last_alloc_group = ac->ac_last_group;
781 
782 bail:
783 	if (handle)
784 		ocfs2_commit_trans(osb, handle);
785 
786 	if (ac)
787 		ocfs2_free_alloc_context(ac);
788 
789 	brelse(bg_bh);
790 
791 	if (status)
792 		mlog_errno(status);
793 	return status;
794 }
795 
796 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
797 				       struct ocfs2_alloc_context *ac,
798 				       int type,
799 				       u32 slot,
800 				       u64 *last_alloc_group,
801 				       int flags)
802 {
803 	int status;
804 	u32 bits_wanted = ac->ac_bits_wanted;
805 	struct inode *alloc_inode;
806 	struct buffer_head *bh = NULL;
807 	struct ocfs2_dinode *fe;
808 	u32 free_bits;
809 
810 	alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
811 	if (!alloc_inode) {
812 		mlog_errno(-EINVAL);
813 		return -EINVAL;
814 	}
815 
816 	mutex_lock(&alloc_inode->i_mutex);
817 
818 	status = ocfs2_inode_lock(alloc_inode, &bh, 1);
819 	if (status < 0) {
820 		mutex_unlock(&alloc_inode->i_mutex);
821 		iput(alloc_inode);
822 
823 		mlog_errno(status);
824 		return status;
825 	}
826 
827 	ac->ac_inode = alloc_inode;
828 	ac->ac_alloc_slot = slot;
829 
830 	fe = (struct ocfs2_dinode *) bh->b_data;
831 
832 	/* The bh was validated by the inode read inside
833 	 * ocfs2_inode_lock().  Any corruption is a code bug. */
834 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
835 
836 	if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
837 		ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
838 			    (unsigned long long)le64_to_cpu(fe->i_blkno));
839 		status = -EIO;
840 		goto bail;
841 	}
842 
843 	free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
844 		le32_to_cpu(fe->id1.bitmap1.i_used);
845 
846 	if (bits_wanted > free_bits) {
847 		/* cluster bitmap never grows */
848 		if (ocfs2_is_cluster_bitmap(alloc_inode)) {
849 			trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
850 								free_bits);
851 			status = -ENOSPC;
852 			goto bail;
853 		}
854 
855 		if (!(flags & ALLOC_NEW_GROUP)) {
856 			trace_ocfs2_reserve_suballoc_bits_no_new_group(
857 						slot, bits_wanted, free_bits);
858 			status = -ENOSPC;
859 			goto bail;
860 		}
861 
862 		status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
863 						 ac->ac_max_block,
864 						 last_alloc_group, flags);
865 		if (status < 0) {
866 			if (status != -ENOSPC)
867 				mlog_errno(status);
868 			goto bail;
869 		}
870 		atomic_inc(&osb->alloc_stats.bg_extends);
871 
872 		/* You should never ask for this much metadata */
873 		BUG_ON(bits_wanted >
874 		       (le32_to_cpu(fe->id1.bitmap1.i_total)
875 			- le32_to_cpu(fe->id1.bitmap1.i_used)));
876 	}
877 
878 	get_bh(bh);
879 	ac->ac_bh = bh;
880 bail:
881 	brelse(bh);
882 
883 	if (status)
884 		mlog_errno(status);
885 	return status;
886 }
887 
888 static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
889 {
890 	spin_lock(&osb->osb_lock);
891 	osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
892 	spin_unlock(&osb->osb_lock);
893 	atomic_set(&osb->s_num_inodes_stolen, 0);
894 }
895 
896 static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
897 {
898 	spin_lock(&osb->osb_lock);
899 	osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
900 	spin_unlock(&osb->osb_lock);
901 	atomic_set(&osb->s_num_meta_stolen, 0);
902 }
903 
904 void ocfs2_init_steal_slots(struct ocfs2_super *osb)
905 {
906 	ocfs2_init_inode_steal_slot(osb);
907 	ocfs2_init_meta_steal_slot(osb);
908 }
909 
910 static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
911 {
912 	spin_lock(&osb->osb_lock);
913 	if (type == INODE_ALLOC_SYSTEM_INODE)
914 		osb->s_inode_steal_slot = slot;
915 	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
916 		osb->s_meta_steal_slot = slot;
917 	spin_unlock(&osb->osb_lock);
918 }
919 
920 static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
921 {
922 	int slot = OCFS2_INVALID_SLOT;
923 
924 	spin_lock(&osb->osb_lock);
925 	if (type == INODE_ALLOC_SYSTEM_INODE)
926 		slot = osb->s_inode_steal_slot;
927 	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
928 		slot = osb->s_meta_steal_slot;
929 	spin_unlock(&osb->osb_lock);
930 
931 	return slot;
932 }
933 
934 static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
935 {
936 	return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
937 }
938 
939 static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
940 {
941 	return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
942 }
943 
944 static int ocfs2_steal_resource(struct ocfs2_super *osb,
945 				struct ocfs2_alloc_context *ac,
946 				int type)
947 {
948 	int i, status = -ENOSPC;
949 	int slot = __ocfs2_get_steal_slot(osb, type);
950 
951 	/* Start to steal resource from the first slot after ours. */
952 	if (slot == OCFS2_INVALID_SLOT)
953 		slot = osb->slot_num + 1;
954 
955 	for (i = 0; i < osb->max_slots; i++, slot++) {
956 		if (slot == osb->max_slots)
957 			slot = 0;
958 
959 		if (slot == osb->slot_num)
960 			continue;
961 
962 		status = ocfs2_reserve_suballoc_bits(osb, ac,
963 						     type,
964 						     (u32)slot, NULL,
965 						     NOT_ALLOC_NEW_GROUP);
966 		if (status >= 0) {
967 			__ocfs2_set_steal_slot(osb, slot, type);
968 			break;
969 		}
970 
971 		ocfs2_free_ac_resource(ac);
972 	}
973 
974 	return status;
975 }
976 
977 static int ocfs2_steal_inode(struct ocfs2_super *osb,
978 			     struct ocfs2_alloc_context *ac)
979 {
980 	return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
981 }
982 
983 static int ocfs2_steal_meta(struct ocfs2_super *osb,
984 			    struct ocfs2_alloc_context *ac)
985 {
986 	return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
987 }
988 
989 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
990 				      int blocks,
991 				      struct ocfs2_alloc_context **ac)
992 {
993 	int status;
994 	int slot = ocfs2_get_meta_steal_slot(osb);
995 
996 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
997 	if (!(*ac)) {
998 		status = -ENOMEM;
999 		mlog_errno(status);
1000 		goto bail;
1001 	}
1002 
1003 	(*ac)->ac_bits_wanted = blocks;
1004 	(*ac)->ac_which = OCFS2_AC_USE_META;
1005 	(*ac)->ac_group_search = ocfs2_block_group_search;
1006 
1007 	if (slot != OCFS2_INVALID_SLOT &&
1008 		atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
1009 		goto extent_steal;
1010 
1011 	atomic_set(&osb->s_num_meta_stolen, 0);
1012 	status = ocfs2_reserve_suballoc_bits(osb, (*ac),
1013 					     EXTENT_ALLOC_SYSTEM_INODE,
1014 					     (u32)osb->slot_num, NULL,
1015 					     ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
1016 
1017 
1018 	if (status >= 0) {
1019 		status = 0;
1020 		if (slot != OCFS2_INVALID_SLOT)
1021 			ocfs2_init_meta_steal_slot(osb);
1022 		goto bail;
1023 	} else if (status < 0 && status != -ENOSPC) {
1024 		mlog_errno(status);
1025 		goto bail;
1026 	}
1027 
1028 	ocfs2_free_ac_resource(*ac);
1029 
1030 extent_steal:
1031 	status = ocfs2_steal_meta(osb, *ac);
1032 	atomic_inc(&osb->s_num_meta_stolen);
1033 	if (status < 0) {
1034 		if (status != -ENOSPC)
1035 			mlog_errno(status);
1036 		goto bail;
1037 	}
1038 
1039 	status = 0;
1040 bail:
1041 	if ((status < 0) && *ac) {
1042 		ocfs2_free_alloc_context(*ac);
1043 		*ac = NULL;
1044 	}
1045 
1046 	if (status)
1047 		mlog_errno(status);
1048 	return status;
1049 }
1050 
1051 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1052 			       struct ocfs2_extent_list *root_el,
1053 			       struct ocfs2_alloc_context **ac)
1054 {
1055 	return ocfs2_reserve_new_metadata_blocks(osb,
1056 					ocfs2_extend_meta_needed(root_el),
1057 					ac);
1058 }
1059 
1060 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
1061 			    struct ocfs2_alloc_context **ac)
1062 {
1063 	int status;
1064 	int slot = ocfs2_get_inode_steal_slot(osb);
1065 	u64 alloc_group;
1066 
1067 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1068 	if (!(*ac)) {
1069 		status = -ENOMEM;
1070 		mlog_errno(status);
1071 		goto bail;
1072 	}
1073 
1074 	(*ac)->ac_bits_wanted = 1;
1075 	(*ac)->ac_which = OCFS2_AC_USE_INODE;
1076 
1077 	(*ac)->ac_group_search = ocfs2_block_group_search;
1078 
1079 	/*
1080 	 * stat(2) can't handle i_ino > 32bits, so we tell the
1081 	 * lower levels not to allocate us a block group past that
1082 	 * limit.  The 'inode64' mount option avoids this behavior.
1083 	 */
1084 	if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1085 		(*ac)->ac_max_block = (u32)~0U;
1086 
1087 	/*
1088 	 * slot is set when we successfully steal inode from other nodes.
1089 	 * It is reset in 3 places:
1090 	 * 1. when we flush the truncate log
1091 	 * 2. when we complete local alloc recovery.
1092 	 * 3. when we successfully allocate from our own slot.
1093 	 * After it is set, we will go on stealing inodes until we find the
1094 	 * need to check our slots to see whether there is some space for us.
1095 	 */
1096 	if (slot != OCFS2_INVALID_SLOT &&
1097 	    atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
1098 		goto inode_steal;
1099 
1100 	atomic_set(&osb->s_num_inodes_stolen, 0);
1101 	alloc_group = osb->osb_inode_alloc_group;
1102 	status = ocfs2_reserve_suballoc_bits(osb, *ac,
1103 					     INODE_ALLOC_SYSTEM_INODE,
1104 					     (u32)osb->slot_num,
1105 					     &alloc_group,
1106 					     ALLOC_NEW_GROUP |
1107 					     ALLOC_GROUPS_FROM_GLOBAL);
1108 	if (status >= 0) {
1109 		status = 0;
1110 
1111 		spin_lock(&osb->osb_lock);
1112 		osb->osb_inode_alloc_group = alloc_group;
1113 		spin_unlock(&osb->osb_lock);
1114 		trace_ocfs2_reserve_new_inode_new_group(
1115 			(unsigned long long)alloc_group);
1116 
1117 		/*
1118 		 * Some inodes must be freed by us, so try to allocate
1119 		 * from our own next time.
1120 		 */
1121 		if (slot != OCFS2_INVALID_SLOT)
1122 			ocfs2_init_inode_steal_slot(osb);
1123 		goto bail;
1124 	} else if (status < 0 && status != -ENOSPC) {
1125 		mlog_errno(status);
1126 		goto bail;
1127 	}
1128 
1129 	ocfs2_free_ac_resource(*ac);
1130 
1131 inode_steal:
1132 	status = ocfs2_steal_inode(osb, *ac);
1133 	atomic_inc(&osb->s_num_inodes_stolen);
1134 	if (status < 0) {
1135 		if (status != -ENOSPC)
1136 			mlog_errno(status);
1137 		goto bail;
1138 	}
1139 
1140 	status = 0;
1141 bail:
1142 	if ((status < 0) && *ac) {
1143 		ocfs2_free_alloc_context(*ac);
1144 		*ac = NULL;
1145 	}
1146 
1147 	if (status)
1148 		mlog_errno(status);
1149 	return status;
1150 }
1151 
1152 /* local alloc code has to do the same thing, so rather than do this
1153  * twice.. */
1154 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1155 				      struct ocfs2_alloc_context *ac)
1156 {
1157 	int status;
1158 
1159 	ac->ac_which = OCFS2_AC_USE_MAIN;
1160 	ac->ac_group_search = ocfs2_cluster_group_search;
1161 
1162 	status = ocfs2_reserve_suballoc_bits(osb, ac,
1163 					     GLOBAL_BITMAP_SYSTEM_INODE,
1164 					     OCFS2_INVALID_SLOT, NULL,
1165 					     ALLOC_NEW_GROUP);
1166 	if (status < 0 && status != -ENOSPC) {
1167 		mlog_errno(status);
1168 		goto bail;
1169 	}
1170 
1171 bail:
1172 	return status;
1173 }
1174 
1175 /* Callers don't need to care which bitmap (local alloc or main) to
1176  * use so we figure it out for them, but unfortunately this clutters
1177  * things a bit. */
1178 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1179 					     u32 bits_wanted, u64 max_block,
1180 					     int flags,
1181 					     struct ocfs2_alloc_context **ac)
1182 {
1183 	int status;
1184 
1185 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1186 	if (!(*ac)) {
1187 		status = -ENOMEM;
1188 		mlog_errno(status);
1189 		goto bail;
1190 	}
1191 
1192 	(*ac)->ac_bits_wanted = bits_wanted;
1193 	(*ac)->ac_max_block = max_block;
1194 
1195 	status = -ENOSPC;
1196 	if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1197 	    ocfs2_alloc_should_use_local(osb, bits_wanted)) {
1198 		status = ocfs2_reserve_local_alloc_bits(osb,
1199 							bits_wanted,
1200 							*ac);
1201 		if ((status < 0) && (status != -ENOSPC)) {
1202 			mlog_errno(status);
1203 			goto bail;
1204 		}
1205 	}
1206 
1207 	if (status == -ENOSPC) {
1208 		status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1209 		if (status < 0) {
1210 			if (status != -ENOSPC)
1211 				mlog_errno(status);
1212 			goto bail;
1213 		}
1214 	}
1215 
1216 	status = 0;
1217 bail:
1218 	if ((status < 0) && *ac) {
1219 		ocfs2_free_alloc_context(*ac);
1220 		*ac = NULL;
1221 	}
1222 
1223 	if (status)
1224 		mlog_errno(status);
1225 	return status;
1226 }
1227 
1228 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1229 			   u32 bits_wanted,
1230 			   struct ocfs2_alloc_context **ac)
1231 {
1232 	return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1233 						 ALLOC_NEW_GROUP, ac);
1234 }
1235 
1236 /*
1237  * More or less lifted from ext3. I'll leave their description below:
1238  *
1239  * "For ext3 allocations, we must not reuse any blocks which are
1240  * allocated in the bitmap buffer's "last committed data" copy.  This
1241  * prevents deletes from freeing up the page for reuse until we have
1242  * committed the delete transaction.
1243  *
1244  * If we didn't do this, then deleting something and reallocating it as
1245  * data would allow the old block to be overwritten before the
1246  * transaction committed (because we force data to disk before commit).
1247  * This would lead to corruption if we crashed between overwriting the
1248  * data and committing the delete.
1249  *
1250  * @@@ We may want to make this allocation behaviour conditional on
1251  * data-writes at some point, and disable it for metadata allocations or
1252  * sync-data inodes."
1253  *
1254  * Note: OCFS2 already does this differently for metadata vs data
1255  * allocations, as those bitmaps are separate and undo access is never
1256  * called on a metadata group descriptor.
1257  */
1258 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1259 					 int nr)
1260 {
1261 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1262 	int ret;
1263 
1264 	if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1265 		return 0;
1266 
1267 	if (!buffer_jbd(bg_bh))
1268 		return 1;
1269 
1270 	jbd_lock_bh_state(bg_bh);
1271 	bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
1272 	if (bg)
1273 		ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1274 	else
1275 		ret = 1;
1276 	jbd_unlock_bh_state(bg_bh);
1277 
1278 	return ret;
1279 }
1280 
1281 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1282 					     struct buffer_head *bg_bh,
1283 					     unsigned int bits_wanted,
1284 					     unsigned int total_bits,
1285 					     struct ocfs2_suballoc_result *res)
1286 {
1287 	void *bitmap;
1288 	u16 best_offset, best_size;
1289 	int offset, start, found, status = 0;
1290 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1291 
1292 	/* Callers got this descriptor from
1293 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1294 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1295 
1296 	found = start = best_offset = best_size = 0;
1297 	bitmap = bg->bg_bitmap;
1298 
1299 	while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1300 		if (offset == total_bits)
1301 			break;
1302 
1303 		if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1304 			/* We found a zero, but we can't use it as it
1305 			 * hasn't been put to disk yet! */
1306 			found = 0;
1307 			start = offset + 1;
1308 		} else if (offset == start) {
1309 			/* we found a zero */
1310 			found++;
1311 			/* move start to the next bit to test */
1312 			start++;
1313 		} else {
1314 			/* got a zero after some ones */
1315 			found = 1;
1316 			start = offset + 1;
1317 		}
1318 		if (found > best_size) {
1319 			best_size = found;
1320 			best_offset = start - found;
1321 		}
1322 		/* we got everything we needed */
1323 		if (found == bits_wanted) {
1324 			/* mlog(0, "Found it all!\n"); */
1325 			break;
1326 		}
1327 	}
1328 
1329 	if (best_size) {
1330 		res->sr_bit_offset = best_offset;
1331 		res->sr_bits = best_size;
1332 	} else {
1333 		status = -ENOSPC;
1334 		/* No error log here -- see the comment above
1335 		 * ocfs2_test_bg_bit_allocatable */
1336 	}
1337 
1338 	return status;
1339 }
1340 
1341 int ocfs2_block_group_set_bits(handle_t *handle,
1342 					     struct inode *alloc_inode,
1343 					     struct ocfs2_group_desc *bg,
1344 					     struct buffer_head *group_bh,
1345 					     unsigned int bit_off,
1346 					     unsigned int num_bits)
1347 {
1348 	int status;
1349 	void *bitmap = bg->bg_bitmap;
1350 	int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1351 
1352 	/* All callers get the descriptor via
1353 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1354 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1355 	BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1356 
1357 	trace_ocfs2_block_group_set_bits(bit_off, num_bits);
1358 
1359 	if (ocfs2_is_cluster_bitmap(alloc_inode))
1360 		journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1361 
1362 	status = ocfs2_journal_access_gd(handle,
1363 					 INODE_CACHE(alloc_inode),
1364 					 group_bh,
1365 					 journal_type);
1366 	if (status < 0) {
1367 		mlog_errno(status);
1368 		goto bail;
1369 	}
1370 
1371 	le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1372 	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1373 		ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
1374 			    " count %u but claims %u are freed. num_bits %d",
1375 			    (unsigned long long)le64_to_cpu(bg->bg_blkno),
1376 			    le16_to_cpu(bg->bg_bits),
1377 			    le16_to_cpu(bg->bg_free_bits_count), num_bits);
1378 		return -EROFS;
1379 	}
1380 	while(num_bits--)
1381 		ocfs2_set_bit(bit_off++, bitmap);
1382 
1383 	ocfs2_journal_dirty(handle, group_bh);
1384 
1385 bail:
1386 	return status;
1387 }
1388 
1389 /* find the one with the most empty bits */
1390 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1391 {
1392 	u16 curr, best;
1393 
1394 	BUG_ON(!cl->cl_next_free_rec);
1395 
1396 	best = curr = 0;
1397 	while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1398 		if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1399 		    le32_to_cpu(cl->cl_recs[best].c_free))
1400 			best = curr;
1401 		curr++;
1402 	}
1403 
1404 	BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1405 	return best;
1406 }
1407 
1408 static int ocfs2_relink_block_group(handle_t *handle,
1409 				    struct inode *alloc_inode,
1410 				    struct buffer_head *fe_bh,
1411 				    struct buffer_head *bg_bh,
1412 				    struct buffer_head *prev_bg_bh,
1413 				    u16 chain)
1414 {
1415 	int status;
1416 	/* there is a really tiny chance the journal calls could fail,
1417 	 * but we wouldn't want inconsistent blocks in *any* case. */
1418 	u64 bg_ptr, prev_bg_ptr;
1419 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1420 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1421 	struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1422 
1423 	/* The caller got these descriptors from
1424 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1425 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1426 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1427 
1428 	trace_ocfs2_relink_block_group(
1429 		(unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1430 		(unsigned long long)le64_to_cpu(bg->bg_blkno),
1431 		(unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1432 
1433 	bg_ptr = le64_to_cpu(bg->bg_next_group);
1434 	prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1435 
1436 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1437 					 prev_bg_bh,
1438 					 OCFS2_JOURNAL_ACCESS_WRITE);
1439 	if (status < 0)
1440 		goto out;
1441 
1442 	prev_bg->bg_next_group = bg->bg_next_group;
1443 	ocfs2_journal_dirty(handle, prev_bg_bh);
1444 
1445 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1446 					 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1447 	if (status < 0)
1448 		goto out_rollback_prev_bg;
1449 
1450 	bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1451 	ocfs2_journal_dirty(handle, bg_bh);
1452 
1453 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1454 					 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1455 	if (status < 0)
1456 		goto out_rollback_bg;
1457 
1458 	fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1459 	ocfs2_journal_dirty(handle, fe_bh);
1460 
1461 out:
1462 	if (status < 0)
1463 		mlog_errno(status);
1464 	return status;
1465 
1466 out_rollback_bg:
1467 	bg->bg_next_group = cpu_to_le64(bg_ptr);
1468 out_rollback_prev_bg:
1469 	prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1470 	goto out;
1471 }
1472 
1473 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1474 						     u32 wanted)
1475 {
1476 	return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1477 }
1478 
1479 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1480  * value on error. */
1481 static int ocfs2_cluster_group_search(struct inode *inode,
1482 				      struct buffer_head *group_bh,
1483 				      u32 bits_wanted, u32 min_bits,
1484 				      u64 max_block,
1485 				      struct ocfs2_suballoc_result *res)
1486 {
1487 	int search = -ENOSPC;
1488 	int ret;
1489 	u64 blkoff;
1490 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1491 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1492 	unsigned int max_bits, gd_cluster_off;
1493 
1494 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1495 
1496 	if (gd->bg_free_bits_count) {
1497 		max_bits = le16_to_cpu(gd->bg_bits);
1498 
1499 		/* Tail groups in cluster bitmaps which aren't cpg
1500 		 * aligned are prone to partial extension by a failed
1501 		 * fs resize. If the file system resize never got to
1502 		 * update the dinode cluster count, then we don't want
1503 		 * to trust any clusters past it, regardless of what
1504 		 * the group descriptor says. */
1505 		gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1506 							  le64_to_cpu(gd->bg_blkno));
1507 		if ((gd_cluster_off + max_bits) >
1508 		    OCFS2_I(inode)->ip_clusters) {
1509 			max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1510 			trace_ocfs2_cluster_group_search_wrong_max_bits(
1511 				(unsigned long long)le64_to_cpu(gd->bg_blkno),
1512 				le16_to_cpu(gd->bg_bits),
1513 				OCFS2_I(inode)->ip_clusters, max_bits);
1514 		}
1515 
1516 		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1517 							group_bh, bits_wanted,
1518 							max_bits, res);
1519 		if (ret)
1520 			return ret;
1521 
1522 		if (max_block) {
1523 			blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1524 							  gd_cluster_off +
1525 							  res->sr_bit_offset +
1526 							  res->sr_bits);
1527 			trace_ocfs2_cluster_group_search_max_block(
1528 				(unsigned long long)blkoff,
1529 				(unsigned long long)max_block);
1530 			if (blkoff > max_block)
1531 				return -ENOSPC;
1532 		}
1533 
1534 		/* ocfs2_block_group_find_clear_bits() might
1535 		 * return success, but we still want to return
1536 		 * -ENOSPC unless it found the minimum number
1537 		 * of bits. */
1538 		if (min_bits <= res->sr_bits)
1539 			search = 0; /* success */
1540 		else if (res->sr_bits) {
1541 			/*
1542 			 * Don't show bits which we'll be returning
1543 			 * for allocation to the local alloc bitmap.
1544 			 */
1545 			ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
1546 		}
1547 	}
1548 
1549 	return search;
1550 }
1551 
1552 static int ocfs2_block_group_search(struct inode *inode,
1553 				    struct buffer_head *group_bh,
1554 				    u32 bits_wanted, u32 min_bits,
1555 				    u64 max_block,
1556 				    struct ocfs2_suballoc_result *res)
1557 {
1558 	int ret = -ENOSPC;
1559 	u64 blkoff;
1560 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1561 
1562 	BUG_ON(min_bits != 1);
1563 	BUG_ON(ocfs2_is_cluster_bitmap(inode));
1564 
1565 	if (bg->bg_free_bits_count) {
1566 		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1567 							group_bh, bits_wanted,
1568 							le16_to_cpu(bg->bg_bits),
1569 							res);
1570 		if (!ret && max_block) {
1571 			blkoff = le64_to_cpu(bg->bg_blkno) +
1572 				res->sr_bit_offset + res->sr_bits;
1573 			trace_ocfs2_block_group_search_max_block(
1574 				(unsigned long long)blkoff,
1575 				(unsigned long long)max_block);
1576 			if (blkoff > max_block)
1577 				ret = -ENOSPC;
1578 		}
1579 	}
1580 
1581 	return ret;
1582 }
1583 
1584 int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1585 				       handle_t *handle,
1586 				       struct buffer_head *di_bh,
1587 				       u32 num_bits,
1588 				       u16 chain)
1589 {
1590 	int ret;
1591 	u32 tmp_used;
1592 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1593 	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1594 
1595 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1596 				      OCFS2_JOURNAL_ACCESS_WRITE);
1597 	if (ret < 0) {
1598 		mlog_errno(ret);
1599 		goto out;
1600 	}
1601 
1602 	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1603 	di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1604 	le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1605 	ocfs2_journal_dirty(handle, di_bh);
1606 
1607 out:
1608 	return ret;
1609 }
1610 
1611 void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
1612 				       struct buffer_head *di_bh,
1613 				       u32 num_bits,
1614 				       u16 chain)
1615 {
1616 	u32 tmp_used;
1617 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1618 	struct ocfs2_chain_list *cl;
1619 
1620 	cl = (struct ocfs2_chain_list *)&di->id2.i_chain;
1621 	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1622 	di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits);
1623 	le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits);
1624 }
1625 
1626 static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1627 					 struct ocfs2_extent_rec *rec,
1628 					 struct ocfs2_chain_list *cl)
1629 {
1630 	unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1631 	unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1632 	unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1633 
1634 	if (res->sr_bit_offset < bitoff)
1635 		return 0;
1636 	if (res->sr_bit_offset >= (bitoff + bitcount))
1637 		return 0;
1638 	res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1639 		(res->sr_bit_offset - bitoff);
1640 	if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1641 		res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1642 	return 1;
1643 }
1644 
1645 static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1646 					  struct ocfs2_group_desc *bg,
1647 					  struct ocfs2_suballoc_result *res)
1648 {
1649 	int i;
1650 	u64 bg_blkno = res->sr_bg_blkno;  /* Save off */
1651 	struct ocfs2_extent_rec *rec;
1652 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1653 	struct ocfs2_chain_list *cl = &di->id2.i_chain;
1654 
1655 	if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1656 		res->sr_blkno = 0;
1657 		return;
1658 	}
1659 
1660 	res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1661 	res->sr_bg_blkno = 0;  /* Clear it for contig block groups */
1662 	if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1663 	    !bg->bg_list.l_next_free_rec)
1664 		return;
1665 
1666 	for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1667 		rec = &bg->bg_list.l_recs[i];
1668 		if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1669 			res->sr_bg_blkno = bg_blkno;  /* Restore */
1670 			break;
1671 		}
1672 	}
1673 }
1674 
1675 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1676 				  handle_t *handle,
1677 				  u32 bits_wanted,
1678 				  u32 min_bits,
1679 				  struct ocfs2_suballoc_result *res,
1680 				  u16 *bits_left)
1681 {
1682 	int ret;
1683 	struct buffer_head *group_bh = NULL;
1684 	struct ocfs2_group_desc *gd;
1685 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1686 	struct inode *alloc_inode = ac->ac_inode;
1687 
1688 	ret = ocfs2_read_group_descriptor(alloc_inode, di,
1689 					  res->sr_bg_blkno, &group_bh);
1690 	if (ret < 0) {
1691 		mlog_errno(ret);
1692 		return ret;
1693 	}
1694 
1695 	gd = (struct ocfs2_group_desc *) group_bh->b_data;
1696 	ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1697 				  ac->ac_max_block, res);
1698 	if (ret < 0) {
1699 		if (ret != -ENOSPC)
1700 			mlog_errno(ret);
1701 		goto out;
1702 	}
1703 
1704 	if (!ret)
1705 		ocfs2_bg_discontig_fix_result(ac, gd, res);
1706 
1707 	/*
1708 	 * sr_bg_blkno might have been changed by
1709 	 * ocfs2_bg_discontig_fix_result
1710 	 */
1711 	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1712 
1713 	if (ac->ac_find_loc_only)
1714 		goto out_loc_only;
1715 
1716 	ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1717 					       res->sr_bits,
1718 					       le16_to_cpu(gd->bg_chain));
1719 	if (ret < 0) {
1720 		mlog_errno(ret);
1721 		goto out;
1722 	}
1723 
1724 	ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1725 					 res->sr_bit_offset, res->sr_bits);
1726 	if (ret < 0) {
1727 		ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh,
1728 					       res->sr_bits,
1729 					       le16_to_cpu(gd->bg_chain));
1730 		mlog_errno(ret);
1731 	}
1732 
1733 out_loc_only:
1734 	*bits_left = le16_to_cpu(gd->bg_free_bits_count);
1735 
1736 out:
1737 	brelse(group_bh);
1738 
1739 	return ret;
1740 }
1741 
1742 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1743 			      handle_t *handle,
1744 			      u32 bits_wanted,
1745 			      u32 min_bits,
1746 			      struct ocfs2_suballoc_result *res,
1747 			      u16 *bits_left)
1748 {
1749 	int status;
1750 	u16 chain;
1751 	u64 next_group;
1752 	struct inode *alloc_inode = ac->ac_inode;
1753 	struct buffer_head *group_bh = NULL;
1754 	struct buffer_head *prev_group_bh = NULL;
1755 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1756 	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1757 	struct ocfs2_group_desc *bg;
1758 
1759 	chain = ac->ac_chain;
1760 	trace_ocfs2_search_chain_begin(
1761 		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1762 		bits_wanted, chain);
1763 
1764 	status = ocfs2_read_group_descriptor(alloc_inode, fe,
1765 					     le64_to_cpu(cl->cl_recs[chain].c_blkno),
1766 					     &group_bh);
1767 	if (status < 0) {
1768 		mlog_errno(status);
1769 		goto bail;
1770 	}
1771 	bg = (struct ocfs2_group_desc *) group_bh->b_data;
1772 
1773 	status = -ENOSPC;
1774 	/* for now, the chain search is a bit simplistic. We just use
1775 	 * the 1st group with any empty bits. */
1776 	while ((status = ac->ac_group_search(alloc_inode, group_bh,
1777 					     bits_wanted, min_bits,
1778 					     ac->ac_max_block,
1779 					     res)) == -ENOSPC) {
1780 		if (!bg->bg_next_group)
1781 			break;
1782 
1783 		brelse(prev_group_bh);
1784 		prev_group_bh = NULL;
1785 
1786 		next_group = le64_to_cpu(bg->bg_next_group);
1787 		prev_group_bh = group_bh;
1788 		group_bh = NULL;
1789 		status = ocfs2_read_group_descriptor(alloc_inode, fe,
1790 						     next_group, &group_bh);
1791 		if (status < 0) {
1792 			mlog_errno(status);
1793 			goto bail;
1794 		}
1795 		bg = (struct ocfs2_group_desc *) group_bh->b_data;
1796 	}
1797 	if (status < 0) {
1798 		if (status != -ENOSPC)
1799 			mlog_errno(status);
1800 		goto bail;
1801 	}
1802 
1803 	trace_ocfs2_search_chain_succ(
1804 		(unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1805 
1806 	res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1807 
1808 	BUG_ON(res->sr_bits == 0);
1809 	if (!status)
1810 		ocfs2_bg_discontig_fix_result(ac, bg, res);
1811 
1812 	/*
1813 	 * sr_bg_blkno might have been changed by
1814 	 * ocfs2_bg_discontig_fix_result
1815 	 */
1816 	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1817 
1818 	/*
1819 	 * Keep track of previous block descriptor read. When
1820 	 * we find a target, if we have read more than X
1821 	 * number of descriptors, and the target is reasonably
1822 	 * empty, relink him to top of his chain.
1823 	 *
1824 	 * We've read 0 extra blocks and only send one more to
1825 	 * the transaction, yet the next guy to search has a
1826 	 * much easier time.
1827 	 *
1828 	 * Do this *after* figuring out how many bits we're taking out
1829 	 * of our target group.
1830 	 */
1831 	if (!ac->ac_disable_chain_relink &&
1832 	    (prev_group_bh) &&
1833 	    (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1834 		status = ocfs2_relink_block_group(handle, alloc_inode,
1835 						  ac->ac_bh, group_bh,
1836 						  prev_group_bh, chain);
1837 		if (status < 0) {
1838 			mlog_errno(status);
1839 			goto bail;
1840 		}
1841 	}
1842 
1843 	if (ac->ac_find_loc_only)
1844 		goto out_loc_only;
1845 
1846 	status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1847 						  ac->ac_bh, res->sr_bits,
1848 						  chain);
1849 	if (status) {
1850 		mlog_errno(status);
1851 		goto bail;
1852 	}
1853 
1854 	status = ocfs2_block_group_set_bits(handle,
1855 					    alloc_inode,
1856 					    bg,
1857 					    group_bh,
1858 					    res->sr_bit_offset,
1859 					    res->sr_bits);
1860 	if (status < 0) {
1861 		ocfs2_rollback_alloc_dinode_counts(alloc_inode,
1862 					ac->ac_bh, res->sr_bits, chain);
1863 		mlog_errno(status);
1864 		goto bail;
1865 	}
1866 
1867 	trace_ocfs2_search_chain_end(
1868 			(unsigned long long)le64_to_cpu(fe->i_blkno),
1869 			res->sr_bits);
1870 
1871 out_loc_only:
1872 	*bits_left = le16_to_cpu(bg->bg_free_bits_count);
1873 bail:
1874 	brelse(group_bh);
1875 	brelse(prev_group_bh);
1876 
1877 	if (status)
1878 		mlog_errno(status);
1879 	return status;
1880 }
1881 
1882 /* will give out up to bits_wanted contiguous bits. */
1883 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1884 				     handle_t *handle,
1885 				     u32 bits_wanted,
1886 				     u32 min_bits,
1887 				     struct ocfs2_suballoc_result *res)
1888 {
1889 	int status;
1890 	u16 victim, i;
1891 	u16 bits_left = 0;
1892 	u64 hint = ac->ac_last_group;
1893 	struct ocfs2_chain_list *cl;
1894 	struct ocfs2_dinode *fe;
1895 
1896 	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1897 	BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1898 	BUG_ON(!ac->ac_bh);
1899 
1900 	fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1901 
1902 	/* The bh was validated by the inode read during
1903 	 * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1904 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1905 
1906 	if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1907 	    le32_to_cpu(fe->id1.bitmap1.i_total)) {
1908 		ocfs2_error(ac->ac_inode->i_sb,
1909 			    "Chain allocator dinode %llu has %u used "
1910 			    "bits but only %u total.",
1911 			    (unsigned long long)le64_to_cpu(fe->i_blkno),
1912 			    le32_to_cpu(fe->id1.bitmap1.i_used),
1913 			    le32_to_cpu(fe->id1.bitmap1.i_total));
1914 		status = -EIO;
1915 		goto bail;
1916 	}
1917 
1918 	res->sr_bg_blkno = hint;
1919 	if (res->sr_bg_blkno) {
1920 		/* Attempt to short-circuit the usual search mechanism
1921 		 * by jumping straight to the most recently used
1922 		 * allocation group. This helps us maintain some
1923 		 * contiguousness across allocations. */
1924 		status = ocfs2_search_one_group(ac, handle, bits_wanted,
1925 						min_bits, res, &bits_left);
1926 		if (!status)
1927 			goto set_hint;
1928 		if (status < 0 && status != -ENOSPC) {
1929 			mlog_errno(status);
1930 			goto bail;
1931 		}
1932 	}
1933 
1934 	cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1935 
1936 	victim = ocfs2_find_victim_chain(cl);
1937 	ac->ac_chain = victim;
1938 
1939 	status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1940 				    res, &bits_left);
1941 	if (!status) {
1942 		hint = ocfs2_group_from_res(res);
1943 		goto set_hint;
1944 	}
1945 	if (status < 0 && status != -ENOSPC) {
1946 		mlog_errno(status);
1947 		goto bail;
1948 	}
1949 
1950 	trace_ocfs2_claim_suballoc_bits(victim);
1951 
1952 	/* If we didn't pick a good victim, then just default to
1953 	 * searching each chain in order. Don't allow chain relinking
1954 	 * because we only calculate enough journal credits for one
1955 	 * relink per alloc. */
1956 	ac->ac_disable_chain_relink = 1;
1957 	for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1958 		if (i == victim)
1959 			continue;
1960 		if (!cl->cl_recs[i].c_free)
1961 			continue;
1962 
1963 		ac->ac_chain = i;
1964 		status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1965 					    res, &bits_left);
1966 		if (!status) {
1967 			hint = ocfs2_group_from_res(res);
1968 			break;
1969 		}
1970 		if (status < 0 && status != -ENOSPC) {
1971 			mlog_errno(status);
1972 			goto bail;
1973 		}
1974 	}
1975 
1976 set_hint:
1977 	if (status != -ENOSPC) {
1978 		/* If the next search of this group is not likely to
1979 		 * yield a suitable extent, then we reset the last
1980 		 * group hint so as to not waste a disk read */
1981 		if (bits_left < min_bits)
1982 			ac->ac_last_group = 0;
1983 		else
1984 			ac->ac_last_group = hint;
1985 	}
1986 
1987 bail:
1988 	if (status)
1989 		mlog_errno(status);
1990 	return status;
1991 }
1992 
1993 int ocfs2_claim_metadata(handle_t *handle,
1994 			 struct ocfs2_alloc_context *ac,
1995 			 u32 bits_wanted,
1996 			 u64 *suballoc_loc,
1997 			 u16 *suballoc_bit_start,
1998 			 unsigned int *num_bits,
1999 			 u64 *blkno_start)
2000 {
2001 	int status;
2002 	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2003 
2004 	BUG_ON(!ac);
2005 	BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
2006 	BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
2007 
2008 	status = ocfs2_claim_suballoc_bits(ac,
2009 					   handle,
2010 					   bits_wanted,
2011 					   1,
2012 					   &res);
2013 	if (status < 0) {
2014 		mlog_errno(status);
2015 		goto bail;
2016 	}
2017 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2018 
2019 	*suballoc_loc = res.sr_bg_blkno;
2020 	*suballoc_bit_start = res.sr_bit_offset;
2021 	*blkno_start = res.sr_blkno;
2022 	ac->ac_bits_given += res.sr_bits;
2023 	*num_bits = res.sr_bits;
2024 	status = 0;
2025 bail:
2026 	if (status)
2027 		mlog_errno(status);
2028 	return status;
2029 }
2030 
2031 static void ocfs2_init_inode_ac_group(struct inode *dir,
2032 				      struct buffer_head *parent_di_bh,
2033 				      struct ocfs2_alloc_context *ac)
2034 {
2035 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2036 	/*
2037 	 * Try to allocate inodes from some specific group.
2038 	 *
2039 	 * If the parent dir has recorded the last group used in allocation,
2040 	 * cool, use it. Otherwise if we try to allocate new inode from the
2041 	 * same slot the parent dir belongs to, use the same chunk.
2042 	 *
2043 	 * We are very careful here to avoid the mistake of setting
2044 	 * ac_last_group to a group descriptor from a different (unlocked) slot.
2045 	 */
2046 	if (OCFS2_I(dir)->ip_last_used_group &&
2047 	    OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2048 		ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2049 	else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2050 		if (di->i_suballoc_loc)
2051 			ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2052 		else
2053 			ac->ac_last_group = ocfs2_which_suballoc_group(
2054 					le64_to_cpu(di->i_blkno),
2055 					le16_to_cpu(di->i_suballoc_bit));
2056 	}
2057 }
2058 
2059 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2060 					     struct ocfs2_alloc_context *ac)
2061 {
2062 	OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2063 	OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2064 }
2065 
2066 int ocfs2_find_new_inode_loc(struct inode *dir,
2067 			     struct buffer_head *parent_fe_bh,
2068 			     struct ocfs2_alloc_context *ac,
2069 			     u64 *fe_blkno)
2070 {
2071 	int ret;
2072 	handle_t *handle = NULL;
2073 	struct ocfs2_suballoc_result *res;
2074 
2075 	BUG_ON(!ac);
2076 	BUG_ON(ac->ac_bits_given != 0);
2077 	BUG_ON(ac->ac_bits_wanted != 1);
2078 	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2079 
2080 	res = kzalloc(sizeof(*res), GFP_NOFS);
2081 	if (res == NULL) {
2082 		ret = -ENOMEM;
2083 		mlog_errno(ret);
2084 		goto out;
2085 	}
2086 
2087 	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2088 
2089 	/*
2090 	 * The handle started here is for chain relink. Alternatively,
2091 	 * we could just disable relink for these calls.
2092 	 */
2093 	handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2094 	if (IS_ERR(handle)) {
2095 		ret = PTR_ERR(handle);
2096 		handle = NULL;
2097 		mlog_errno(ret);
2098 		goto out;
2099 	}
2100 
2101 	/*
2102 	 * This will instruct ocfs2_claim_suballoc_bits and
2103 	 * ocfs2_search_one_group to search but save actual allocation
2104 	 * for later.
2105 	 */
2106 	ac->ac_find_loc_only = 1;
2107 
2108 	ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2109 	if (ret < 0) {
2110 		mlog_errno(ret);
2111 		goto out;
2112 	}
2113 
2114 	ac->ac_find_loc_priv = res;
2115 	*fe_blkno = res->sr_blkno;
2116 	ocfs2_update_inode_fsync_trans(handle, dir, 0);
2117 out:
2118 	if (handle)
2119 		ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2120 
2121 	if (ret)
2122 		kfree(res);
2123 
2124 	return ret;
2125 }
2126 
2127 int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2128 				 struct inode *dir,
2129 				 struct ocfs2_alloc_context *ac,
2130 				 u64 *suballoc_loc,
2131 				 u16 *suballoc_bit,
2132 				 u64 di_blkno)
2133 {
2134 	int ret;
2135 	u16 chain;
2136 	struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2137 	struct buffer_head *bg_bh = NULL;
2138 	struct ocfs2_group_desc *bg;
2139 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2140 
2141 	/*
2142 	 * Since di_blkno is being passed back in, we check for any
2143 	 * inconsistencies which may have happened between
2144 	 * calls. These are code bugs as di_blkno is not expected to
2145 	 * change once returned from ocfs2_find_new_inode_loc()
2146 	 */
2147 	BUG_ON(res->sr_blkno != di_blkno);
2148 
2149 	ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2150 					  res->sr_bg_stable_blkno, &bg_bh);
2151 	if (ret) {
2152 		mlog_errno(ret);
2153 		goto out;
2154 	}
2155 
2156 	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2157 	chain = le16_to_cpu(bg->bg_chain);
2158 
2159 	ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2160 					       ac->ac_bh, res->sr_bits,
2161 					       chain);
2162 	if (ret) {
2163 		mlog_errno(ret);
2164 		goto out;
2165 	}
2166 
2167 	ret = ocfs2_block_group_set_bits(handle,
2168 					 ac->ac_inode,
2169 					 bg,
2170 					 bg_bh,
2171 					 res->sr_bit_offset,
2172 					 res->sr_bits);
2173 	if (ret < 0) {
2174 		ocfs2_rollback_alloc_dinode_counts(ac->ac_inode,
2175 					       ac->ac_bh, res->sr_bits, chain);
2176 		mlog_errno(ret);
2177 		goto out;
2178 	}
2179 
2180 	trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2181 					   res->sr_bits);
2182 
2183 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2184 
2185 	BUG_ON(res->sr_bits != 1);
2186 
2187 	*suballoc_loc = res->sr_bg_blkno;
2188 	*suballoc_bit = res->sr_bit_offset;
2189 	ac->ac_bits_given++;
2190 	ocfs2_save_inode_ac_group(dir, ac);
2191 
2192 out:
2193 	brelse(bg_bh);
2194 
2195 	return ret;
2196 }
2197 
2198 int ocfs2_claim_new_inode(handle_t *handle,
2199 			  struct inode *dir,
2200 			  struct buffer_head *parent_fe_bh,
2201 			  struct ocfs2_alloc_context *ac,
2202 			  u64 *suballoc_loc,
2203 			  u16 *suballoc_bit,
2204 			  u64 *fe_blkno)
2205 {
2206 	int status;
2207 	struct ocfs2_suballoc_result res;
2208 
2209 	BUG_ON(!ac);
2210 	BUG_ON(ac->ac_bits_given != 0);
2211 	BUG_ON(ac->ac_bits_wanted != 1);
2212 	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2213 
2214 	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2215 
2216 	status = ocfs2_claim_suballoc_bits(ac,
2217 					   handle,
2218 					   1,
2219 					   1,
2220 					   &res);
2221 	if (status < 0) {
2222 		mlog_errno(status);
2223 		goto bail;
2224 	}
2225 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2226 
2227 	BUG_ON(res.sr_bits != 1);
2228 
2229 	*suballoc_loc = res.sr_bg_blkno;
2230 	*suballoc_bit = res.sr_bit_offset;
2231 	*fe_blkno = res.sr_blkno;
2232 	ac->ac_bits_given++;
2233 	ocfs2_save_inode_ac_group(dir, ac);
2234 	status = 0;
2235 bail:
2236 	if (status)
2237 		mlog_errno(status);
2238 	return status;
2239 }
2240 
2241 /* translate a group desc. blkno and it's bitmap offset into
2242  * disk cluster offset. */
2243 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2244 						   u64 bg_blkno,
2245 						   u16 bg_bit_off)
2246 {
2247 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2248 	u32 cluster = 0;
2249 
2250 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2251 
2252 	if (bg_blkno != osb->first_cluster_group_blkno)
2253 		cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2254 	cluster += (u32) bg_bit_off;
2255 	return cluster;
2256 }
2257 
2258 /* given a cluster offset, calculate which block group it belongs to
2259  * and return that block offset. */
2260 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2261 {
2262 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2263 	u32 group_no;
2264 
2265 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2266 
2267 	group_no = cluster / osb->bitmap_cpg;
2268 	if (!group_no)
2269 		return osb->first_cluster_group_blkno;
2270 	return ocfs2_clusters_to_blocks(inode->i_sb,
2271 					group_no * osb->bitmap_cpg);
2272 }
2273 
2274 /* given the block number of a cluster start, calculate which cluster
2275  * group and descriptor bitmap offset that corresponds to. */
2276 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2277 						u64 data_blkno,
2278 						u64 *bg_blkno,
2279 						u16 *bg_bit_off)
2280 {
2281 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2282 	u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2283 
2284 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2285 
2286 	*bg_blkno = ocfs2_which_cluster_group(inode,
2287 					      data_cluster);
2288 
2289 	if (*bg_blkno == osb->first_cluster_group_blkno)
2290 		*bg_bit_off = (u16) data_cluster;
2291 	else
2292 		*bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2293 							     data_blkno - *bg_blkno);
2294 }
2295 
2296 /*
2297  * min_bits - minimum contiguous chunk from this total allocation we
2298  * can handle. set to what we asked for originally for a full
2299  * contig. allocation, set to '1' to indicate we can deal with extents
2300  * of any size.
2301  */
2302 int __ocfs2_claim_clusters(handle_t *handle,
2303 			   struct ocfs2_alloc_context *ac,
2304 			   u32 min_clusters,
2305 			   u32 max_clusters,
2306 			   u32 *cluster_start,
2307 			   u32 *num_clusters)
2308 {
2309 	int status;
2310 	unsigned int bits_wanted = max_clusters;
2311 	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2312 	struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2313 
2314 	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2315 
2316 	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2317 	       && ac->ac_which != OCFS2_AC_USE_MAIN);
2318 
2319 	if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2320 		WARN_ON(min_clusters > 1);
2321 
2322 		status = ocfs2_claim_local_alloc_bits(osb,
2323 						      handle,
2324 						      ac,
2325 						      bits_wanted,
2326 						      cluster_start,
2327 						      num_clusters);
2328 		if (!status)
2329 			atomic_inc(&osb->alloc_stats.local_data);
2330 	} else {
2331 		if (min_clusters > (osb->bitmap_cpg - 1)) {
2332 			/* The only paths asking for contiguousness
2333 			 * should know about this already. */
2334 			mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2335 			     "group bitmap size %u!\n", min_clusters,
2336 			     osb->bitmap_cpg);
2337 			status = -ENOSPC;
2338 			goto bail;
2339 		}
2340 		/* clamp the current request down to a realistic size. */
2341 		if (bits_wanted > (osb->bitmap_cpg - 1))
2342 			bits_wanted = osb->bitmap_cpg - 1;
2343 
2344 		status = ocfs2_claim_suballoc_bits(ac,
2345 						   handle,
2346 						   bits_wanted,
2347 						   min_clusters,
2348 						   &res);
2349 		if (!status) {
2350 			BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2351 			*cluster_start =
2352 				ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2353 								 res.sr_bg_blkno,
2354 								 res.sr_bit_offset);
2355 			atomic_inc(&osb->alloc_stats.bitmap_data);
2356 			*num_clusters = res.sr_bits;
2357 		}
2358 	}
2359 	if (status < 0) {
2360 		if (status != -ENOSPC)
2361 			mlog_errno(status);
2362 		goto bail;
2363 	}
2364 
2365 	ac->ac_bits_given += *num_clusters;
2366 
2367 bail:
2368 	if (status)
2369 		mlog_errno(status);
2370 	return status;
2371 }
2372 
2373 int ocfs2_claim_clusters(handle_t *handle,
2374 			 struct ocfs2_alloc_context *ac,
2375 			 u32 min_clusters,
2376 			 u32 *cluster_start,
2377 			 u32 *num_clusters)
2378 {
2379 	unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2380 
2381 	return __ocfs2_claim_clusters(handle, ac, min_clusters,
2382 				      bits_wanted, cluster_start, num_clusters);
2383 }
2384 
2385 static int ocfs2_block_group_clear_bits(handle_t *handle,
2386 					struct inode *alloc_inode,
2387 					struct ocfs2_group_desc *bg,
2388 					struct buffer_head *group_bh,
2389 					unsigned int bit_off,
2390 					unsigned int num_bits,
2391 					void (*undo_fn)(unsigned int bit,
2392 							unsigned long *bmap))
2393 {
2394 	int status;
2395 	unsigned int tmp;
2396 	struct ocfs2_group_desc *undo_bg = NULL;
2397 
2398 	/* The caller got this descriptor from
2399 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
2400 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2401 
2402 	trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2403 
2404 	BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2405 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2406 					 group_bh,
2407 					 undo_fn ?
2408 					 OCFS2_JOURNAL_ACCESS_UNDO :
2409 					 OCFS2_JOURNAL_ACCESS_WRITE);
2410 	if (status < 0) {
2411 		mlog_errno(status);
2412 		goto bail;
2413 	}
2414 
2415 	if (undo_fn) {
2416 		jbd_lock_bh_state(group_bh);
2417 		undo_bg = (struct ocfs2_group_desc *)
2418 					bh2jh(group_bh)->b_committed_data;
2419 		BUG_ON(!undo_bg);
2420 	}
2421 
2422 	tmp = num_bits;
2423 	while(tmp--) {
2424 		ocfs2_clear_bit((bit_off + tmp),
2425 				(unsigned long *) bg->bg_bitmap);
2426 		if (undo_fn)
2427 			undo_fn(bit_off + tmp,
2428 				(unsigned long *) undo_bg->bg_bitmap);
2429 	}
2430 	le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2431 	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2432 		ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
2433 			    " count %u but claims %u are freed. num_bits %d",
2434 			    (unsigned long long)le64_to_cpu(bg->bg_blkno),
2435 			    le16_to_cpu(bg->bg_bits),
2436 			    le16_to_cpu(bg->bg_free_bits_count), num_bits);
2437 		return -EROFS;
2438 	}
2439 
2440 	if (undo_fn)
2441 		jbd_unlock_bh_state(group_bh);
2442 
2443 	ocfs2_journal_dirty(handle, group_bh);
2444 bail:
2445 	return status;
2446 }
2447 
2448 /*
2449  * expects the suballoc inode to already be locked.
2450  */
2451 static int _ocfs2_free_suballoc_bits(handle_t *handle,
2452 				     struct inode *alloc_inode,
2453 				     struct buffer_head *alloc_bh,
2454 				     unsigned int start_bit,
2455 				     u64 bg_blkno,
2456 				     unsigned int count,
2457 				     void (*undo_fn)(unsigned int bit,
2458 						     unsigned long *bitmap))
2459 {
2460 	int status = 0;
2461 	u32 tmp_used;
2462 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2463 	struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2464 	struct buffer_head *group_bh = NULL;
2465 	struct ocfs2_group_desc *group;
2466 
2467 	/* The alloc_bh comes from ocfs2_free_dinode() or
2468 	 * ocfs2_free_clusters().  The callers have all locked the
2469 	 * allocator and gotten alloc_bh from the lock call.  This
2470 	 * validates the dinode buffer.  Any corruption that has happened
2471 	 * is a code bug. */
2472 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2473 	BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2474 
2475 	trace_ocfs2_free_suballoc_bits(
2476 		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2477 		(unsigned long long)bg_blkno,
2478 		start_bit, count);
2479 
2480 	status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2481 					     &group_bh);
2482 	if (status < 0) {
2483 		mlog_errno(status);
2484 		goto bail;
2485 	}
2486 	group = (struct ocfs2_group_desc *) group_bh->b_data;
2487 
2488 	BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2489 
2490 	status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2491 					      group, group_bh,
2492 					      start_bit, count, undo_fn);
2493 	if (status < 0) {
2494 		mlog_errno(status);
2495 		goto bail;
2496 	}
2497 
2498 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2499 					 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2500 	if (status < 0) {
2501 		mlog_errno(status);
2502 		goto bail;
2503 	}
2504 
2505 	le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2506 		     count);
2507 	tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2508 	fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2509 	ocfs2_journal_dirty(handle, alloc_bh);
2510 
2511 bail:
2512 	brelse(group_bh);
2513 
2514 	if (status)
2515 		mlog_errno(status);
2516 	return status;
2517 }
2518 
2519 int ocfs2_free_suballoc_bits(handle_t *handle,
2520 			     struct inode *alloc_inode,
2521 			     struct buffer_head *alloc_bh,
2522 			     unsigned int start_bit,
2523 			     u64 bg_blkno,
2524 			     unsigned int count)
2525 {
2526 	return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2527 					 start_bit, bg_blkno, count, NULL);
2528 }
2529 
2530 int ocfs2_free_dinode(handle_t *handle,
2531 		      struct inode *inode_alloc_inode,
2532 		      struct buffer_head *inode_alloc_bh,
2533 		      struct ocfs2_dinode *di)
2534 {
2535 	u64 blk = le64_to_cpu(di->i_blkno);
2536 	u16 bit = le16_to_cpu(di->i_suballoc_bit);
2537 	u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2538 
2539 	if (di->i_suballoc_loc)
2540 		bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2541 	return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2542 					inode_alloc_bh, bit, bg_blkno, 1);
2543 }
2544 
2545 static int _ocfs2_free_clusters(handle_t *handle,
2546 				struct inode *bitmap_inode,
2547 				struct buffer_head *bitmap_bh,
2548 				u64 start_blk,
2549 				unsigned int num_clusters,
2550 				void (*undo_fn)(unsigned int bit,
2551 						unsigned long *bitmap))
2552 {
2553 	int status;
2554 	u16 bg_start_bit;
2555 	u64 bg_blkno;
2556 	struct ocfs2_dinode *fe;
2557 
2558 	/* You can't ever have a contiguous set of clusters
2559 	 * bigger than a block group bitmap so we never have to worry
2560 	 * about looping on them.
2561 	 * This is expensive. We can safely remove once this stuff has
2562 	 * gotten tested really well. */
2563 	BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
2564 
2565 	fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
2566 
2567 	ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2568 				     &bg_start_bit);
2569 
2570 	trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2571 			(unsigned long long)start_blk,
2572 			bg_start_bit, num_clusters);
2573 
2574 	status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2575 					   bg_start_bit, bg_blkno,
2576 					   num_clusters, undo_fn);
2577 	if (status < 0) {
2578 		mlog_errno(status);
2579 		goto out;
2580 	}
2581 
2582 	ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2583 					 num_clusters);
2584 
2585 out:
2586 	if (status)
2587 		mlog_errno(status);
2588 	return status;
2589 }
2590 
2591 int ocfs2_free_clusters(handle_t *handle,
2592 			struct inode *bitmap_inode,
2593 			struct buffer_head *bitmap_bh,
2594 			u64 start_blk,
2595 			unsigned int num_clusters)
2596 {
2597 	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2598 				    start_blk, num_clusters,
2599 				    _ocfs2_set_bit);
2600 }
2601 
2602 /*
2603  * Give never-used clusters back to the global bitmap.  We don't need
2604  * to protect these bits in the undo buffer.
2605  */
2606 int ocfs2_release_clusters(handle_t *handle,
2607 			   struct inode *bitmap_inode,
2608 			   struct buffer_head *bitmap_bh,
2609 			   u64 start_blk,
2610 			   unsigned int num_clusters)
2611 {
2612 	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2613 				    start_blk, num_clusters,
2614 				    _ocfs2_clear_bit);
2615 }
2616 
2617 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
2618 {
2619 	printk("Block Group:\n");
2620 	printk("bg_signature:       %s\n", bg->bg_signature);
2621 	printk("bg_size:            %u\n", bg->bg_size);
2622 	printk("bg_bits:            %u\n", bg->bg_bits);
2623 	printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
2624 	printk("bg_chain:           %u\n", bg->bg_chain);
2625 	printk("bg_generation:      %u\n", le32_to_cpu(bg->bg_generation));
2626 	printk("bg_next_group:      %llu\n",
2627 	       (unsigned long long)bg->bg_next_group);
2628 	printk("bg_parent_dinode:   %llu\n",
2629 	       (unsigned long long)bg->bg_parent_dinode);
2630 	printk("bg_blkno:           %llu\n",
2631 	       (unsigned long long)bg->bg_blkno);
2632 }
2633 
2634 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2635 {
2636 	int i;
2637 
2638 	printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
2639 	printk("i_signature:                  %s\n", fe->i_signature);
2640 	printk("i_size:                       %llu\n",
2641 	       (unsigned long long)fe->i_size);
2642 	printk("i_clusters:                   %u\n", fe->i_clusters);
2643 	printk("i_generation:                 %u\n",
2644 	       le32_to_cpu(fe->i_generation));
2645 	printk("id1.bitmap1.i_used:           %u\n",
2646 	       le32_to_cpu(fe->id1.bitmap1.i_used));
2647 	printk("id1.bitmap1.i_total:          %u\n",
2648 	       le32_to_cpu(fe->id1.bitmap1.i_total));
2649 	printk("id2.i_chain.cl_cpg:           %u\n", fe->id2.i_chain.cl_cpg);
2650 	printk("id2.i_chain.cl_bpc:           %u\n", fe->id2.i_chain.cl_bpc);
2651 	printk("id2.i_chain.cl_count:         %u\n", fe->id2.i_chain.cl_count);
2652 	printk("id2.i_chain.cl_next_free_rec: %u\n",
2653 	       fe->id2.i_chain.cl_next_free_rec);
2654 	for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2655 		printk("fe->id2.i_chain.cl_recs[%d].c_free:  %u\n", i,
2656 		       fe->id2.i_chain.cl_recs[i].c_free);
2657 		printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2658 		       fe->id2.i_chain.cl_recs[i].c_total);
2659 		printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2660 		       (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
2661 	}
2662 }
2663 
2664 /*
2665  * For a given allocation, determine which allocators will need to be
2666  * accessed, and lock them, reserving the appropriate number of bits.
2667  *
2668  * Sparse file systems call this from ocfs2_write_begin_nolock()
2669  * and ocfs2_allocate_unwritten_extents().
2670  *
2671  * File systems which don't support holes call this from
2672  * ocfs2_extend_allocation().
2673  */
2674 int ocfs2_lock_allocators(struct inode *inode,
2675 			  struct ocfs2_extent_tree *et,
2676 			  u32 clusters_to_add, u32 extents_to_split,
2677 			  struct ocfs2_alloc_context **data_ac,
2678 			  struct ocfs2_alloc_context **meta_ac)
2679 {
2680 	int ret = 0, num_free_extents;
2681 	unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2682 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2683 
2684 	*meta_ac = NULL;
2685 	if (data_ac)
2686 		*data_ac = NULL;
2687 
2688 	BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2689 
2690 	num_free_extents = ocfs2_num_free_extents(osb, et);
2691 	if (num_free_extents < 0) {
2692 		ret = num_free_extents;
2693 		mlog_errno(ret);
2694 		goto out;
2695 	}
2696 
2697 	/*
2698 	 * Sparse allocation file systems need to be more conservative
2699 	 * with reserving room for expansion - the actual allocation
2700 	 * happens while we've got a journal handle open so re-taking
2701 	 * a cluster lock (because we ran out of room for another
2702 	 * extent) will violate ordering rules.
2703 	 *
2704 	 * Most of the time we'll only be seeing this 1 cluster at a time
2705 	 * anyway.
2706 	 *
2707 	 * Always lock for any unwritten extents - we might want to
2708 	 * add blocks during a split.
2709 	 */
2710 	if (!num_free_extents ||
2711 	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2712 		ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2713 		if (ret < 0) {
2714 			if (ret != -ENOSPC)
2715 				mlog_errno(ret);
2716 			goto out;
2717 		}
2718 	}
2719 
2720 	if (clusters_to_add == 0)
2721 		goto out;
2722 
2723 	ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2724 	if (ret < 0) {
2725 		if (ret != -ENOSPC)
2726 			mlog_errno(ret);
2727 		goto out;
2728 	}
2729 
2730 out:
2731 	if (ret) {
2732 		if (*meta_ac) {
2733 			ocfs2_free_alloc_context(*meta_ac);
2734 			*meta_ac = NULL;
2735 		}
2736 
2737 		/*
2738 		 * We cannot have an error and a non null *data_ac.
2739 		 */
2740 	}
2741 
2742 	return ret;
2743 }
2744 
2745 /*
2746  * Read the inode specified by blkno to get suballoc_slot and
2747  * suballoc_bit.
2748  */
2749 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2750 				       u16 *suballoc_slot, u64 *group_blkno,
2751 				       u16 *suballoc_bit)
2752 {
2753 	int status;
2754 	struct buffer_head *inode_bh = NULL;
2755 	struct ocfs2_dinode *inode_fe;
2756 
2757 	trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2758 
2759 	/* dirty read disk */
2760 	status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2761 	if (status < 0) {
2762 		mlog(ML_ERROR, "read block %llu failed %d\n",
2763 		     (unsigned long long)blkno, status);
2764 		goto bail;
2765 	}
2766 
2767 	inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2768 	if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2769 		mlog(ML_ERROR, "invalid inode %llu requested\n",
2770 		     (unsigned long long)blkno);
2771 		status = -EINVAL;
2772 		goto bail;
2773 	}
2774 
2775 	if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2776 	    (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2777 		mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2778 		     (unsigned long long)blkno,
2779 		     (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2780 		status = -EINVAL;
2781 		goto bail;
2782 	}
2783 
2784 	if (suballoc_slot)
2785 		*suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2786 	if (suballoc_bit)
2787 		*suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2788 	if (group_blkno)
2789 		*group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2790 
2791 bail:
2792 	brelse(inode_bh);
2793 
2794 	if (status)
2795 		mlog_errno(status);
2796 	return status;
2797 }
2798 
2799 /*
2800  * test whether bit is SET in allocator bitmap or not.  on success, 0
2801  * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
2802  * is returned and *res is meaningless.  Call this after you have
2803  * cluster locked against suballoc, or you may get a result based on
2804  * non-up2date contents
2805  */
2806 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2807 				   struct inode *suballoc,
2808 				   struct buffer_head *alloc_bh,
2809 				   u64 group_blkno, u64 blkno,
2810 				   u16 bit, int *res)
2811 {
2812 	struct ocfs2_dinode *alloc_di;
2813 	struct ocfs2_group_desc *group;
2814 	struct buffer_head *group_bh = NULL;
2815 	u64 bg_blkno;
2816 	int status;
2817 
2818 	trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2819 				      (unsigned int)bit);
2820 
2821 	alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2822 	if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2823 		mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2824 		     (unsigned int)bit,
2825 		     ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2826 		status = -EINVAL;
2827 		goto bail;
2828 	}
2829 
2830 	bg_blkno = group_blkno ? group_blkno :
2831 		   ocfs2_which_suballoc_group(blkno, bit);
2832 	status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2833 					     &group_bh);
2834 	if (status < 0) {
2835 		mlog(ML_ERROR, "read group %llu failed %d\n",
2836 		     (unsigned long long)bg_blkno, status);
2837 		goto bail;
2838 	}
2839 
2840 	group = (struct ocfs2_group_desc *) group_bh->b_data;
2841 	*res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2842 
2843 bail:
2844 	brelse(group_bh);
2845 
2846 	if (status)
2847 		mlog_errno(status);
2848 	return status;
2849 }
2850 
2851 /*
2852  * Test if the bit representing this inode (blkno) is set in the
2853  * suballocator.
2854  *
2855  * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2856  *
2857  * In the event of failure, a negative value is returned and *res is
2858  * meaningless.
2859  *
2860  * Callers must make sure to hold nfs_sync_lock to prevent
2861  * ocfs2_delete_inode() on another node from accessing the same
2862  * suballocator concurrently.
2863  */
2864 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2865 {
2866 	int status;
2867 	u64 group_blkno = 0;
2868 	u16 suballoc_bit = 0, suballoc_slot = 0;
2869 	struct inode *inode_alloc_inode;
2870 	struct buffer_head *alloc_bh = NULL;
2871 
2872 	trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2873 
2874 	status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2875 					     &group_blkno, &suballoc_bit);
2876 	if (status < 0) {
2877 		mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2878 		goto bail;
2879 	}
2880 
2881 	inode_alloc_inode =
2882 		ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2883 					    suballoc_slot);
2884 	if (!inode_alloc_inode) {
2885 		/* the error code could be inaccurate, but we are not able to
2886 		 * get the correct one. */
2887 		status = -EINVAL;
2888 		mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2889 		     (u32)suballoc_slot);
2890 		goto bail;
2891 	}
2892 
2893 	mutex_lock(&inode_alloc_inode->i_mutex);
2894 	status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2895 	if (status < 0) {
2896 		mutex_unlock(&inode_alloc_inode->i_mutex);
2897 		iput(inode_alloc_inode);
2898 		mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2899 		     (u32)suballoc_slot, status);
2900 		goto bail;
2901 	}
2902 
2903 	status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2904 					 group_blkno, blkno, suballoc_bit, res);
2905 	if (status < 0)
2906 		mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2907 
2908 	ocfs2_inode_unlock(inode_alloc_inode, 0);
2909 	mutex_unlock(&inode_alloc_inode->i_mutex);
2910 
2911 	iput(inode_alloc_inode);
2912 	brelse(alloc_bh);
2913 bail:
2914 	if (status)
2915 		mlog_errno(status);
2916 	return status;
2917 }
2918