xref: /openbmc/linux/fs/ocfs2/suballoc.c (revision 8e9356c6)
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 
775 	status = 0;
776 
777 	/* save the new last alloc group so that the caller can cache it. */
778 	if (last_alloc_group)
779 		*last_alloc_group = ac->ac_last_group;
780 
781 bail:
782 	if (handle)
783 		ocfs2_commit_trans(osb, handle);
784 
785 	if (ac)
786 		ocfs2_free_alloc_context(ac);
787 
788 	brelse(bg_bh);
789 
790 	if (status)
791 		mlog_errno(status);
792 	return status;
793 }
794 
795 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
796 				       struct ocfs2_alloc_context *ac,
797 				       int type,
798 				       u32 slot,
799 				       u64 *last_alloc_group,
800 				       int flags)
801 {
802 	int status;
803 	u32 bits_wanted = ac->ac_bits_wanted;
804 	struct inode *alloc_inode;
805 	struct buffer_head *bh = NULL;
806 	struct ocfs2_dinode *fe;
807 	u32 free_bits;
808 
809 	alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
810 	if (!alloc_inode) {
811 		mlog_errno(-EINVAL);
812 		return -EINVAL;
813 	}
814 
815 	mutex_lock(&alloc_inode->i_mutex);
816 
817 	status = ocfs2_inode_lock(alloc_inode, &bh, 1);
818 	if (status < 0) {
819 		mutex_unlock(&alloc_inode->i_mutex);
820 		iput(alloc_inode);
821 
822 		mlog_errno(status);
823 		return status;
824 	}
825 
826 	ac->ac_inode = alloc_inode;
827 	ac->ac_alloc_slot = slot;
828 
829 	fe = (struct ocfs2_dinode *) bh->b_data;
830 
831 	/* The bh was validated by the inode read inside
832 	 * ocfs2_inode_lock().  Any corruption is a code bug. */
833 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
834 
835 	if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
836 		ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
837 			    (unsigned long long)le64_to_cpu(fe->i_blkno));
838 		status = -EIO;
839 		goto bail;
840 	}
841 
842 	free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
843 		le32_to_cpu(fe->id1.bitmap1.i_used);
844 
845 	if (bits_wanted > free_bits) {
846 		/* cluster bitmap never grows */
847 		if (ocfs2_is_cluster_bitmap(alloc_inode)) {
848 			trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
849 								free_bits);
850 			status = -ENOSPC;
851 			goto bail;
852 		}
853 
854 		if (!(flags & ALLOC_NEW_GROUP)) {
855 			trace_ocfs2_reserve_suballoc_bits_no_new_group(
856 						slot, bits_wanted, free_bits);
857 			status = -ENOSPC;
858 			goto bail;
859 		}
860 
861 		status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
862 						 ac->ac_max_block,
863 						 last_alloc_group, flags);
864 		if (status < 0) {
865 			if (status != -ENOSPC)
866 				mlog_errno(status);
867 			goto bail;
868 		}
869 		atomic_inc(&osb->alloc_stats.bg_extends);
870 
871 		/* You should never ask for this much metadata */
872 		BUG_ON(bits_wanted >
873 		       (le32_to_cpu(fe->id1.bitmap1.i_total)
874 			- le32_to_cpu(fe->id1.bitmap1.i_used)));
875 	}
876 
877 	get_bh(bh);
878 	ac->ac_bh = bh;
879 bail:
880 	brelse(bh);
881 
882 	if (status)
883 		mlog_errno(status);
884 	return status;
885 }
886 
887 static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
888 {
889 	spin_lock(&osb->osb_lock);
890 	osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
891 	spin_unlock(&osb->osb_lock);
892 	atomic_set(&osb->s_num_inodes_stolen, 0);
893 }
894 
895 static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
896 {
897 	spin_lock(&osb->osb_lock);
898 	osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
899 	spin_unlock(&osb->osb_lock);
900 	atomic_set(&osb->s_num_meta_stolen, 0);
901 }
902 
903 void ocfs2_init_steal_slots(struct ocfs2_super *osb)
904 {
905 	ocfs2_init_inode_steal_slot(osb);
906 	ocfs2_init_meta_steal_slot(osb);
907 }
908 
909 static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
910 {
911 	spin_lock(&osb->osb_lock);
912 	if (type == INODE_ALLOC_SYSTEM_INODE)
913 		osb->s_inode_steal_slot = slot;
914 	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
915 		osb->s_meta_steal_slot = slot;
916 	spin_unlock(&osb->osb_lock);
917 }
918 
919 static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
920 {
921 	int slot = OCFS2_INVALID_SLOT;
922 
923 	spin_lock(&osb->osb_lock);
924 	if (type == INODE_ALLOC_SYSTEM_INODE)
925 		slot = osb->s_inode_steal_slot;
926 	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
927 		slot = osb->s_meta_steal_slot;
928 	spin_unlock(&osb->osb_lock);
929 
930 	return slot;
931 }
932 
933 static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
934 {
935 	return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
936 }
937 
938 static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
939 {
940 	return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
941 }
942 
943 static int ocfs2_steal_resource(struct ocfs2_super *osb,
944 				struct ocfs2_alloc_context *ac,
945 				int type)
946 {
947 	int i, status = -ENOSPC;
948 	int slot = __ocfs2_get_steal_slot(osb, type);
949 
950 	/* Start to steal resource from the first slot after ours. */
951 	if (slot == OCFS2_INVALID_SLOT)
952 		slot = osb->slot_num + 1;
953 
954 	for (i = 0; i < osb->max_slots; i++, slot++) {
955 		if (slot == osb->max_slots)
956 			slot = 0;
957 
958 		if (slot == osb->slot_num)
959 			continue;
960 
961 		status = ocfs2_reserve_suballoc_bits(osb, ac,
962 						     type,
963 						     (u32)slot, NULL,
964 						     NOT_ALLOC_NEW_GROUP);
965 		if (status >= 0) {
966 			__ocfs2_set_steal_slot(osb, slot, type);
967 			break;
968 		}
969 
970 		ocfs2_free_ac_resource(ac);
971 	}
972 
973 	return status;
974 }
975 
976 static int ocfs2_steal_inode(struct ocfs2_super *osb,
977 			     struct ocfs2_alloc_context *ac)
978 {
979 	return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
980 }
981 
982 static int ocfs2_steal_meta(struct ocfs2_super *osb,
983 			    struct ocfs2_alloc_context *ac)
984 {
985 	return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
986 }
987 
988 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
989 				      int blocks,
990 				      struct ocfs2_alloc_context **ac)
991 {
992 	int status;
993 	int slot = ocfs2_get_meta_steal_slot(osb);
994 
995 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
996 	if (!(*ac)) {
997 		status = -ENOMEM;
998 		mlog_errno(status);
999 		goto bail;
1000 	}
1001 
1002 	(*ac)->ac_bits_wanted = blocks;
1003 	(*ac)->ac_which = OCFS2_AC_USE_META;
1004 	(*ac)->ac_group_search = ocfs2_block_group_search;
1005 
1006 	if (slot != OCFS2_INVALID_SLOT &&
1007 		atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
1008 		goto extent_steal;
1009 
1010 	atomic_set(&osb->s_num_meta_stolen, 0);
1011 	status = ocfs2_reserve_suballoc_bits(osb, (*ac),
1012 					     EXTENT_ALLOC_SYSTEM_INODE,
1013 					     (u32)osb->slot_num, NULL,
1014 					     ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
1015 
1016 
1017 	if (status >= 0) {
1018 		status = 0;
1019 		if (slot != OCFS2_INVALID_SLOT)
1020 			ocfs2_init_meta_steal_slot(osb);
1021 		goto bail;
1022 	} else if (status < 0 && status != -ENOSPC) {
1023 		mlog_errno(status);
1024 		goto bail;
1025 	}
1026 
1027 	ocfs2_free_ac_resource(*ac);
1028 
1029 extent_steal:
1030 	status = ocfs2_steal_meta(osb, *ac);
1031 	atomic_inc(&osb->s_num_meta_stolen);
1032 	if (status < 0) {
1033 		if (status != -ENOSPC)
1034 			mlog_errno(status);
1035 		goto bail;
1036 	}
1037 
1038 	status = 0;
1039 bail:
1040 	if ((status < 0) && *ac) {
1041 		ocfs2_free_alloc_context(*ac);
1042 		*ac = NULL;
1043 	}
1044 
1045 	if (status)
1046 		mlog_errno(status);
1047 	return status;
1048 }
1049 
1050 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1051 			       struct ocfs2_extent_list *root_el,
1052 			       struct ocfs2_alloc_context **ac)
1053 {
1054 	return ocfs2_reserve_new_metadata_blocks(osb,
1055 					ocfs2_extend_meta_needed(root_el),
1056 					ac);
1057 }
1058 
1059 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
1060 			    struct ocfs2_alloc_context **ac)
1061 {
1062 	int status;
1063 	int slot = ocfs2_get_inode_steal_slot(osb);
1064 	u64 alloc_group;
1065 
1066 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1067 	if (!(*ac)) {
1068 		status = -ENOMEM;
1069 		mlog_errno(status);
1070 		goto bail;
1071 	}
1072 
1073 	(*ac)->ac_bits_wanted = 1;
1074 	(*ac)->ac_which = OCFS2_AC_USE_INODE;
1075 
1076 	(*ac)->ac_group_search = ocfs2_block_group_search;
1077 
1078 	/*
1079 	 * stat(2) can't handle i_ino > 32bits, so we tell the
1080 	 * lower levels not to allocate us a block group past that
1081 	 * limit.  The 'inode64' mount option avoids this behavior.
1082 	 */
1083 	if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1084 		(*ac)->ac_max_block = (u32)~0U;
1085 
1086 	/*
1087 	 * slot is set when we successfully steal inode from other nodes.
1088 	 * It is reset in 3 places:
1089 	 * 1. when we flush the truncate log
1090 	 * 2. when we complete local alloc recovery.
1091 	 * 3. when we successfully allocate from our own slot.
1092 	 * After it is set, we will go on stealing inodes until we find the
1093 	 * need to check our slots to see whether there is some space for us.
1094 	 */
1095 	if (slot != OCFS2_INVALID_SLOT &&
1096 	    atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
1097 		goto inode_steal;
1098 
1099 	atomic_set(&osb->s_num_inodes_stolen, 0);
1100 	alloc_group = osb->osb_inode_alloc_group;
1101 	status = ocfs2_reserve_suballoc_bits(osb, *ac,
1102 					     INODE_ALLOC_SYSTEM_INODE,
1103 					     (u32)osb->slot_num,
1104 					     &alloc_group,
1105 					     ALLOC_NEW_GROUP |
1106 					     ALLOC_GROUPS_FROM_GLOBAL);
1107 	if (status >= 0) {
1108 		status = 0;
1109 
1110 		spin_lock(&osb->osb_lock);
1111 		osb->osb_inode_alloc_group = alloc_group;
1112 		spin_unlock(&osb->osb_lock);
1113 		trace_ocfs2_reserve_new_inode_new_group(
1114 			(unsigned long long)alloc_group);
1115 
1116 		/*
1117 		 * Some inodes must be freed by us, so try to allocate
1118 		 * from our own next time.
1119 		 */
1120 		if (slot != OCFS2_INVALID_SLOT)
1121 			ocfs2_init_inode_steal_slot(osb);
1122 		goto bail;
1123 	} else if (status < 0 && status != -ENOSPC) {
1124 		mlog_errno(status);
1125 		goto bail;
1126 	}
1127 
1128 	ocfs2_free_ac_resource(*ac);
1129 
1130 inode_steal:
1131 	status = ocfs2_steal_inode(osb, *ac);
1132 	atomic_inc(&osb->s_num_inodes_stolen);
1133 	if (status < 0) {
1134 		if (status != -ENOSPC)
1135 			mlog_errno(status);
1136 		goto bail;
1137 	}
1138 
1139 	status = 0;
1140 bail:
1141 	if ((status < 0) && *ac) {
1142 		ocfs2_free_alloc_context(*ac);
1143 		*ac = NULL;
1144 	}
1145 
1146 	if (status)
1147 		mlog_errno(status);
1148 	return status;
1149 }
1150 
1151 /* local alloc code has to do the same thing, so rather than do this
1152  * twice.. */
1153 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1154 				      struct ocfs2_alloc_context *ac)
1155 {
1156 	int status;
1157 
1158 	ac->ac_which = OCFS2_AC_USE_MAIN;
1159 	ac->ac_group_search = ocfs2_cluster_group_search;
1160 
1161 	status = ocfs2_reserve_suballoc_bits(osb, ac,
1162 					     GLOBAL_BITMAP_SYSTEM_INODE,
1163 					     OCFS2_INVALID_SLOT, NULL,
1164 					     ALLOC_NEW_GROUP);
1165 	if (status < 0 && status != -ENOSPC) {
1166 		mlog_errno(status);
1167 		goto bail;
1168 	}
1169 
1170 bail:
1171 	return status;
1172 }
1173 
1174 /* Callers don't need to care which bitmap (local alloc or main) to
1175  * use so we figure it out for them, but unfortunately this clutters
1176  * things a bit. */
1177 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1178 					     u32 bits_wanted, u64 max_block,
1179 					     int flags,
1180 					     struct ocfs2_alloc_context **ac)
1181 {
1182 	int status;
1183 
1184 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1185 	if (!(*ac)) {
1186 		status = -ENOMEM;
1187 		mlog_errno(status);
1188 		goto bail;
1189 	}
1190 
1191 	(*ac)->ac_bits_wanted = bits_wanted;
1192 	(*ac)->ac_max_block = max_block;
1193 
1194 	status = -ENOSPC;
1195 	if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1196 	    ocfs2_alloc_should_use_local(osb, bits_wanted)) {
1197 		status = ocfs2_reserve_local_alloc_bits(osb,
1198 							bits_wanted,
1199 							*ac);
1200 		if ((status < 0) && (status != -ENOSPC)) {
1201 			mlog_errno(status);
1202 			goto bail;
1203 		}
1204 	}
1205 
1206 	if (status == -ENOSPC) {
1207 		status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1208 		if (status < 0) {
1209 			if (status != -ENOSPC)
1210 				mlog_errno(status);
1211 			goto bail;
1212 		}
1213 	}
1214 
1215 	status = 0;
1216 bail:
1217 	if ((status < 0) && *ac) {
1218 		ocfs2_free_alloc_context(*ac);
1219 		*ac = NULL;
1220 	}
1221 
1222 	if (status)
1223 		mlog_errno(status);
1224 	return status;
1225 }
1226 
1227 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1228 			   u32 bits_wanted,
1229 			   struct ocfs2_alloc_context **ac)
1230 {
1231 	return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1232 						 ALLOC_NEW_GROUP, ac);
1233 }
1234 
1235 /*
1236  * More or less lifted from ext3. I'll leave their description below:
1237  *
1238  * "For ext3 allocations, we must not reuse any blocks which are
1239  * allocated in the bitmap buffer's "last committed data" copy.  This
1240  * prevents deletes from freeing up the page for reuse until we have
1241  * committed the delete transaction.
1242  *
1243  * If we didn't do this, then deleting something and reallocating it as
1244  * data would allow the old block to be overwritten before the
1245  * transaction committed (because we force data to disk before commit).
1246  * This would lead to corruption if we crashed between overwriting the
1247  * data and committing the delete.
1248  *
1249  * @@@ We may want to make this allocation behaviour conditional on
1250  * data-writes at some point, and disable it for metadata allocations or
1251  * sync-data inodes."
1252  *
1253  * Note: OCFS2 already does this differently for metadata vs data
1254  * allocations, as those bitmaps are separate and undo access is never
1255  * called on a metadata group descriptor.
1256  */
1257 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1258 					 int nr)
1259 {
1260 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1261 	int ret;
1262 
1263 	if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1264 		return 0;
1265 
1266 	if (!buffer_jbd(bg_bh))
1267 		return 1;
1268 
1269 	jbd_lock_bh_state(bg_bh);
1270 	bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
1271 	if (bg)
1272 		ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1273 	else
1274 		ret = 1;
1275 	jbd_unlock_bh_state(bg_bh);
1276 
1277 	return ret;
1278 }
1279 
1280 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1281 					     struct buffer_head *bg_bh,
1282 					     unsigned int bits_wanted,
1283 					     unsigned int total_bits,
1284 					     struct ocfs2_suballoc_result *res)
1285 {
1286 	void *bitmap;
1287 	u16 best_offset, best_size;
1288 	int offset, start, found, status = 0;
1289 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1290 
1291 	/* Callers got this descriptor from
1292 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1293 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1294 
1295 	found = start = best_offset = best_size = 0;
1296 	bitmap = bg->bg_bitmap;
1297 
1298 	while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1299 		if (offset == total_bits)
1300 			break;
1301 
1302 		if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1303 			/* We found a zero, but we can't use it as it
1304 			 * hasn't been put to disk yet! */
1305 			found = 0;
1306 			start = offset + 1;
1307 		} else if (offset == start) {
1308 			/* we found a zero */
1309 			found++;
1310 			/* move start to the next bit to test */
1311 			start++;
1312 		} else {
1313 			/* got a zero after some ones */
1314 			found = 1;
1315 			start = offset + 1;
1316 		}
1317 		if (found > best_size) {
1318 			best_size = found;
1319 			best_offset = start - found;
1320 		}
1321 		/* we got everything we needed */
1322 		if (found == bits_wanted) {
1323 			/* mlog(0, "Found it all!\n"); */
1324 			break;
1325 		}
1326 	}
1327 
1328 	if (best_size) {
1329 		res->sr_bit_offset = best_offset;
1330 		res->sr_bits = best_size;
1331 	} else {
1332 		status = -ENOSPC;
1333 		/* No error log here -- see the comment above
1334 		 * ocfs2_test_bg_bit_allocatable */
1335 	}
1336 
1337 	return status;
1338 }
1339 
1340 int ocfs2_block_group_set_bits(handle_t *handle,
1341 					     struct inode *alloc_inode,
1342 					     struct ocfs2_group_desc *bg,
1343 					     struct buffer_head *group_bh,
1344 					     unsigned int bit_off,
1345 					     unsigned int num_bits)
1346 {
1347 	int status;
1348 	void *bitmap = bg->bg_bitmap;
1349 	int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1350 
1351 	/* All callers get the descriptor via
1352 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1353 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1354 	BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1355 
1356 	trace_ocfs2_block_group_set_bits(bit_off, num_bits);
1357 
1358 	if (ocfs2_is_cluster_bitmap(alloc_inode))
1359 		journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1360 
1361 	status = ocfs2_journal_access_gd(handle,
1362 					 INODE_CACHE(alloc_inode),
1363 					 group_bh,
1364 					 journal_type);
1365 	if (status < 0) {
1366 		mlog_errno(status);
1367 		goto bail;
1368 	}
1369 
1370 	le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1371 	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1372 		ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
1373 			    " count %u but claims %u are freed. num_bits %d",
1374 			    (unsigned long long)le64_to_cpu(bg->bg_blkno),
1375 			    le16_to_cpu(bg->bg_bits),
1376 			    le16_to_cpu(bg->bg_free_bits_count), num_bits);
1377 		return -EROFS;
1378 	}
1379 	while(num_bits--)
1380 		ocfs2_set_bit(bit_off++, bitmap);
1381 
1382 	ocfs2_journal_dirty(handle, group_bh);
1383 
1384 bail:
1385 	return status;
1386 }
1387 
1388 /* find the one with the most empty bits */
1389 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1390 {
1391 	u16 curr, best;
1392 
1393 	BUG_ON(!cl->cl_next_free_rec);
1394 
1395 	best = curr = 0;
1396 	while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1397 		if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1398 		    le32_to_cpu(cl->cl_recs[best].c_free))
1399 			best = curr;
1400 		curr++;
1401 	}
1402 
1403 	BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1404 	return best;
1405 }
1406 
1407 static int ocfs2_relink_block_group(handle_t *handle,
1408 				    struct inode *alloc_inode,
1409 				    struct buffer_head *fe_bh,
1410 				    struct buffer_head *bg_bh,
1411 				    struct buffer_head *prev_bg_bh,
1412 				    u16 chain)
1413 {
1414 	int status;
1415 	/* there is a really tiny chance the journal calls could fail,
1416 	 * but we wouldn't want inconsistent blocks in *any* case. */
1417 	u64 bg_ptr, prev_bg_ptr;
1418 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1419 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1420 	struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1421 
1422 	/* The caller got these descriptors from
1423 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1424 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1425 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1426 
1427 	trace_ocfs2_relink_block_group(
1428 		(unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1429 		(unsigned long long)le64_to_cpu(bg->bg_blkno),
1430 		(unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1431 
1432 	bg_ptr = le64_to_cpu(bg->bg_next_group);
1433 	prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1434 
1435 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1436 					 prev_bg_bh,
1437 					 OCFS2_JOURNAL_ACCESS_WRITE);
1438 	if (status < 0)
1439 		goto out;
1440 
1441 	prev_bg->bg_next_group = bg->bg_next_group;
1442 	ocfs2_journal_dirty(handle, prev_bg_bh);
1443 
1444 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1445 					 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1446 	if (status < 0)
1447 		goto out_rollback_prev_bg;
1448 
1449 	bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1450 	ocfs2_journal_dirty(handle, bg_bh);
1451 
1452 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1453 					 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1454 	if (status < 0)
1455 		goto out_rollback_bg;
1456 
1457 	fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1458 	ocfs2_journal_dirty(handle, fe_bh);
1459 
1460 out:
1461 	if (status < 0)
1462 		mlog_errno(status);
1463 	return status;
1464 
1465 out_rollback_bg:
1466 	bg->bg_next_group = cpu_to_le64(bg_ptr);
1467 out_rollback_prev_bg:
1468 	prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1469 	goto out;
1470 }
1471 
1472 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1473 						     u32 wanted)
1474 {
1475 	return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1476 }
1477 
1478 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1479  * value on error. */
1480 static int ocfs2_cluster_group_search(struct inode *inode,
1481 				      struct buffer_head *group_bh,
1482 				      u32 bits_wanted, u32 min_bits,
1483 				      u64 max_block,
1484 				      struct ocfs2_suballoc_result *res)
1485 {
1486 	int search = -ENOSPC;
1487 	int ret;
1488 	u64 blkoff;
1489 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1490 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1491 	unsigned int max_bits, gd_cluster_off;
1492 
1493 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1494 
1495 	if (gd->bg_free_bits_count) {
1496 		max_bits = le16_to_cpu(gd->bg_bits);
1497 
1498 		/* Tail groups in cluster bitmaps which aren't cpg
1499 		 * aligned are prone to partial extension by a failed
1500 		 * fs resize. If the file system resize never got to
1501 		 * update the dinode cluster count, then we don't want
1502 		 * to trust any clusters past it, regardless of what
1503 		 * the group descriptor says. */
1504 		gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1505 							  le64_to_cpu(gd->bg_blkno));
1506 		if ((gd_cluster_off + max_bits) >
1507 		    OCFS2_I(inode)->ip_clusters) {
1508 			max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1509 			trace_ocfs2_cluster_group_search_wrong_max_bits(
1510 				(unsigned long long)le64_to_cpu(gd->bg_blkno),
1511 				le16_to_cpu(gd->bg_bits),
1512 				OCFS2_I(inode)->ip_clusters, max_bits);
1513 		}
1514 
1515 		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1516 							group_bh, bits_wanted,
1517 							max_bits, res);
1518 		if (ret)
1519 			return ret;
1520 
1521 		if (max_block) {
1522 			blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1523 							  gd_cluster_off +
1524 							  res->sr_bit_offset +
1525 							  res->sr_bits);
1526 			trace_ocfs2_cluster_group_search_max_block(
1527 				(unsigned long long)blkoff,
1528 				(unsigned long long)max_block);
1529 			if (blkoff > max_block)
1530 				return -ENOSPC;
1531 		}
1532 
1533 		/* ocfs2_block_group_find_clear_bits() might
1534 		 * return success, but we still want to return
1535 		 * -ENOSPC unless it found the minimum number
1536 		 * of bits. */
1537 		if (min_bits <= res->sr_bits)
1538 			search = 0; /* success */
1539 		else if (res->sr_bits) {
1540 			/*
1541 			 * Don't show bits which we'll be returning
1542 			 * for allocation to the local alloc bitmap.
1543 			 */
1544 			ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
1545 		}
1546 	}
1547 
1548 	return search;
1549 }
1550 
1551 static int ocfs2_block_group_search(struct inode *inode,
1552 				    struct buffer_head *group_bh,
1553 				    u32 bits_wanted, u32 min_bits,
1554 				    u64 max_block,
1555 				    struct ocfs2_suballoc_result *res)
1556 {
1557 	int ret = -ENOSPC;
1558 	u64 blkoff;
1559 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1560 
1561 	BUG_ON(min_bits != 1);
1562 	BUG_ON(ocfs2_is_cluster_bitmap(inode));
1563 
1564 	if (bg->bg_free_bits_count) {
1565 		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1566 							group_bh, bits_wanted,
1567 							le16_to_cpu(bg->bg_bits),
1568 							res);
1569 		if (!ret && max_block) {
1570 			blkoff = le64_to_cpu(bg->bg_blkno) +
1571 				res->sr_bit_offset + res->sr_bits;
1572 			trace_ocfs2_block_group_search_max_block(
1573 				(unsigned long long)blkoff,
1574 				(unsigned long long)max_block);
1575 			if (blkoff > max_block)
1576 				ret = -ENOSPC;
1577 		}
1578 	}
1579 
1580 	return ret;
1581 }
1582 
1583 int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1584 				       handle_t *handle,
1585 				       struct buffer_head *di_bh,
1586 				       u32 num_bits,
1587 				       u16 chain)
1588 {
1589 	int ret;
1590 	u32 tmp_used;
1591 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1592 	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1593 
1594 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1595 				      OCFS2_JOURNAL_ACCESS_WRITE);
1596 	if (ret < 0) {
1597 		mlog_errno(ret);
1598 		goto out;
1599 	}
1600 
1601 	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1602 	di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1603 	le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1604 	ocfs2_journal_dirty(handle, di_bh);
1605 
1606 out:
1607 	return ret;
1608 }
1609 
1610 static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1611 					 struct ocfs2_extent_rec *rec,
1612 					 struct ocfs2_chain_list *cl)
1613 {
1614 	unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1615 	unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1616 	unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1617 
1618 	if (res->sr_bit_offset < bitoff)
1619 		return 0;
1620 	if (res->sr_bit_offset >= (bitoff + bitcount))
1621 		return 0;
1622 	res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1623 		(res->sr_bit_offset - bitoff);
1624 	if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1625 		res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1626 	return 1;
1627 }
1628 
1629 static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1630 					  struct ocfs2_group_desc *bg,
1631 					  struct ocfs2_suballoc_result *res)
1632 {
1633 	int i;
1634 	u64 bg_blkno = res->sr_bg_blkno;  /* Save off */
1635 	struct ocfs2_extent_rec *rec;
1636 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1637 	struct ocfs2_chain_list *cl = &di->id2.i_chain;
1638 
1639 	if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1640 		res->sr_blkno = 0;
1641 		return;
1642 	}
1643 
1644 	res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1645 	res->sr_bg_blkno = 0;  /* Clear it for contig block groups */
1646 	if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1647 	    !bg->bg_list.l_next_free_rec)
1648 		return;
1649 
1650 	for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1651 		rec = &bg->bg_list.l_recs[i];
1652 		if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1653 			res->sr_bg_blkno = bg_blkno;  /* Restore */
1654 			break;
1655 		}
1656 	}
1657 }
1658 
1659 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1660 				  handle_t *handle,
1661 				  u32 bits_wanted,
1662 				  u32 min_bits,
1663 				  struct ocfs2_suballoc_result *res,
1664 				  u16 *bits_left)
1665 {
1666 	int ret;
1667 	struct buffer_head *group_bh = NULL;
1668 	struct ocfs2_group_desc *gd;
1669 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1670 	struct inode *alloc_inode = ac->ac_inode;
1671 
1672 	ret = ocfs2_read_group_descriptor(alloc_inode, di,
1673 					  res->sr_bg_blkno, &group_bh);
1674 	if (ret < 0) {
1675 		mlog_errno(ret);
1676 		return ret;
1677 	}
1678 
1679 	gd = (struct ocfs2_group_desc *) group_bh->b_data;
1680 	ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1681 				  ac->ac_max_block, res);
1682 	if (ret < 0) {
1683 		if (ret != -ENOSPC)
1684 			mlog_errno(ret);
1685 		goto out;
1686 	}
1687 
1688 	if (!ret)
1689 		ocfs2_bg_discontig_fix_result(ac, gd, res);
1690 
1691 	/*
1692 	 * sr_bg_blkno might have been changed by
1693 	 * ocfs2_bg_discontig_fix_result
1694 	 */
1695 	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1696 
1697 	if (ac->ac_find_loc_only)
1698 		goto out_loc_only;
1699 
1700 	ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1701 					       res->sr_bits,
1702 					       le16_to_cpu(gd->bg_chain));
1703 	if (ret < 0) {
1704 		mlog_errno(ret);
1705 		goto out;
1706 	}
1707 
1708 	ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1709 					 res->sr_bit_offset, res->sr_bits);
1710 	if (ret < 0)
1711 		mlog_errno(ret);
1712 
1713 out_loc_only:
1714 	*bits_left = le16_to_cpu(gd->bg_free_bits_count);
1715 
1716 out:
1717 	brelse(group_bh);
1718 
1719 	return ret;
1720 }
1721 
1722 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1723 			      handle_t *handle,
1724 			      u32 bits_wanted,
1725 			      u32 min_bits,
1726 			      struct ocfs2_suballoc_result *res,
1727 			      u16 *bits_left)
1728 {
1729 	int status;
1730 	u16 chain;
1731 	u64 next_group;
1732 	struct inode *alloc_inode = ac->ac_inode;
1733 	struct buffer_head *group_bh = NULL;
1734 	struct buffer_head *prev_group_bh = NULL;
1735 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1736 	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1737 	struct ocfs2_group_desc *bg;
1738 
1739 	chain = ac->ac_chain;
1740 	trace_ocfs2_search_chain_begin(
1741 		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1742 		bits_wanted, chain);
1743 
1744 	status = ocfs2_read_group_descriptor(alloc_inode, fe,
1745 					     le64_to_cpu(cl->cl_recs[chain].c_blkno),
1746 					     &group_bh);
1747 	if (status < 0) {
1748 		mlog_errno(status);
1749 		goto bail;
1750 	}
1751 	bg = (struct ocfs2_group_desc *) group_bh->b_data;
1752 
1753 	status = -ENOSPC;
1754 	/* for now, the chain search is a bit simplistic. We just use
1755 	 * the 1st group with any empty bits. */
1756 	while ((status = ac->ac_group_search(alloc_inode, group_bh,
1757 					     bits_wanted, min_bits,
1758 					     ac->ac_max_block,
1759 					     res)) == -ENOSPC) {
1760 		if (!bg->bg_next_group)
1761 			break;
1762 
1763 		brelse(prev_group_bh);
1764 		prev_group_bh = NULL;
1765 
1766 		next_group = le64_to_cpu(bg->bg_next_group);
1767 		prev_group_bh = group_bh;
1768 		group_bh = NULL;
1769 		status = ocfs2_read_group_descriptor(alloc_inode, fe,
1770 						     next_group, &group_bh);
1771 		if (status < 0) {
1772 			mlog_errno(status);
1773 			goto bail;
1774 		}
1775 		bg = (struct ocfs2_group_desc *) group_bh->b_data;
1776 	}
1777 	if (status < 0) {
1778 		if (status != -ENOSPC)
1779 			mlog_errno(status);
1780 		goto bail;
1781 	}
1782 
1783 	trace_ocfs2_search_chain_succ(
1784 		(unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1785 
1786 	res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1787 
1788 	BUG_ON(res->sr_bits == 0);
1789 	if (!status)
1790 		ocfs2_bg_discontig_fix_result(ac, bg, res);
1791 
1792 	/*
1793 	 * sr_bg_blkno might have been changed by
1794 	 * ocfs2_bg_discontig_fix_result
1795 	 */
1796 	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1797 
1798 	/*
1799 	 * Keep track of previous block descriptor read. When
1800 	 * we find a target, if we have read more than X
1801 	 * number of descriptors, and the target is reasonably
1802 	 * empty, relink him to top of his chain.
1803 	 *
1804 	 * We've read 0 extra blocks and only send one more to
1805 	 * the transaction, yet the next guy to search has a
1806 	 * much easier time.
1807 	 *
1808 	 * Do this *after* figuring out how many bits we're taking out
1809 	 * of our target group.
1810 	 */
1811 	if (!ac->ac_disable_chain_relink &&
1812 	    (prev_group_bh) &&
1813 	    (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1814 		status = ocfs2_relink_block_group(handle, alloc_inode,
1815 						  ac->ac_bh, group_bh,
1816 						  prev_group_bh, chain);
1817 		if (status < 0) {
1818 			mlog_errno(status);
1819 			goto bail;
1820 		}
1821 	}
1822 
1823 	if (ac->ac_find_loc_only)
1824 		goto out_loc_only;
1825 
1826 	status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1827 						  ac->ac_bh, res->sr_bits,
1828 						  chain);
1829 	if (status) {
1830 		mlog_errno(status);
1831 		goto bail;
1832 	}
1833 
1834 	status = ocfs2_block_group_set_bits(handle,
1835 					    alloc_inode,
1836 					    bg,
1837 					    group_bh,
1838 					    res->sr_bit_offset,
1839 					    res->sr_bits);
1840 	if (status < 0) {
1841 		mlog_errno(status);
1842 		goto bail;
1843 	}
1844 
1845 	trace_ocfs2_search_chain_end(
1846 			(unsigned long long)le64_to_cpu(fe->i_blkno),
1847 			res->sr_bits);
1848 
1849 out_loc_only:
1850 	*bits_left = le16_to_cpu(bg->bg_free_bits_count);
1851 bail:
1852 	brelse(group_bh);
1853 	brelse(prev_group_bh);
1854 
1855 	if (status)
1856 		mlog_errno(status);
1857 	return status;
1858 }
1859 
1860 /* will give out up to bits_wanted contiguous bits. */
1861 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1862 				     handle_t *handle,
1863 				     u32 bits_wanted,
1864 				     u32 min_bits,
1865 				     struct ocfs2_suballoc_result *res)
1866 {
1867 	int status;
1868 	u16 victim, i;
1869 	u16 bits_left = 0;
1870 	u64 hint = ac->ac_last_group;
1871 	struct ocfs2_chain_list *cl;
1872 	struct ocfs2_dinode *fe;
1873 
1874 	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1875 	BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1876 	BUG_ON(!ac->ac_bh);
1877 
1878 	fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1879 
1880 	/* The bh was validated by the inode read during
1881 	 * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1882 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1883 
1884 	if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1885 	    le32_to_cpu(fe->id1.bitmap1.i_total)) {
1886 		ocfs2_error(ac->ac_inode->i_sb,
1887 			    "Chain allocator dinode %llu has %u used "
1888 			    "bits but only %u total.",
1889 			    (unsigned long long)le64_to_cpu(fe->i_blkno),
1890 			    le32_to_cpu(fe->id1.bitmap1.i_used),
1891 			    le32_to_cpu(fe->id1.bitmap1.i_total));
1892 		status = -EIO;
1893 		goto bail;
1894 	}
1895 
1896 	res->sr_bg_blkno = hint;
1897 	if (res->sr_bg_blkno) {
1898 		/* Attempt to short-circuit the usual search mechanism
1899 		 * by jumping straight to the most recently used
1900 		 * allocation group. This helps us maintain some
1901 		 * contiguousness across allocations. */
1902 		status = ocfs2_search_one_group(ac, handle, bits_wanted,
1903 						min_bits, res, &bits_left);
1904 		if (!status)
1905 			goto set_hint;
1906 		if (status < 0 && status != -ENOSPC) {
1907 			mlog_errno(status);
1908 			goto bail;
1909 		}
1910 	}
1911 
1912 	cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1913 
1914 	victim = ocfs2_find_victim_chain(cl);
1915 	ac->ac_chain = victim;
1916 
1917 	status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1918 				    res, &bits_left);
1919 	if (!status) {
1920 		hint = ocfs2_group_from_res(res);
1921 		goto set_hint;
1922 	}
1923 	if (status < 0 && status != -ENOSPC) {
1924 		mlog_errno(status);
1925 		goto bail;
1926 	}
1927 
1928 	trace_ocfs2_claim_suballoc_bits(victim);
1929 
1930 	/* If we didn't pick a good victim, then just default to
1931 	 * searching each chain in order. Don't allow chain relinking
1932 	 * because we only calculate enough journal credits for one
1933 	 * relink per alloc. */
1934 	ac->ac_disable_chain_relink = 1;
1935 	for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1936 		if (i == victim)
1937 			continue;
1938 		if (!cl->cl_recs[i].c_free)
1939 			continue;
1940 
1941 		ac->ac_chain = i;
1942 		status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1943 					    res, &bits_left);
1944 		if (!status) {
1945 			hint = ocfs2_group_from_res(res);
1946 			break;
1947 		}
1948 		if (status < 0 && status != -ENOSPC) {
1949 			mlog_errno(status);
1950 			goto bail;
1951 		}
1952 	}
1953 
1954 set_hint:
1955 	if (status != -ENOSPC) {
1956 		/* If the next search of this group is not likely to
1957 		 * yield a suitable extent, then we reset the last
1958 		 * group hint so as to not waste a disk read */
1959 		if (bits_left < min_bits)
1960 			ac->ac_last_group = 0;
1961 		else
1962 			ac->ac_last_group = hint;
1963 	}
1964 
1965 bail:
1966 	if (status)
1967 		mlog_errno(status);
1968 	return status;
1969 }
1970 
1971 int ocfs2_claim_metadata(handle_t *handle,
1972 			 struct ocfs2_alloc_context *ac,
1973 			 u32 bits_wanted,
1974 			 u64 *suballoc_loc,
1975 			 u16 *suballoc_bit_start,
1976 			 unsigned int *num_bits,
1977 			 u64 *blkno_start)
1978 {
1979 	int status;
1980 	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
1981 
1982 	BUG_ON(!ac);
1983 	BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1984 	BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
1985 
1986 	status = ocfs2_claim_suballoc_bits(ac,
1987 					   handle,
1988 					   bits_wanted,
1989 					   1,
1990 					   &res);
1991 	if (status < 0) {
1992 		mlog_errno(status);
1993 		goto bail;
1994 	}
1995 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
1996 
1997 	*suballoc_loc = res.sr_bg_blkno;
1998 	*suballoc_bit_start = res.sr_bit_offset;
1999 	*blkno_start = res.sr_blkno;
2000 	ac->ac_bits_given += res.sr_bits;
2001 	*num_bits = res.sr_bits;
2002 	status = 0;
2003 bail:
2004 	if (status)
2005 		mlog_errno(status);
2006 	return status;
2007 }
2008 
2009 static void ocfs2_init_inode_ac_group(struct inode *dir,
2010 				      struct buffer_head *parent_di_bh,
2011 				      struct ocfs2_alloc_context *ac)
2012 {
2013 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2014 	/*
2015 	 * Try to allocate inodes from some specific group.
2016 	 *
2017 	 * If the parent dir has recorded the last group used in allocation,
2018 	 * cool, use it. Otherwise if we try to allocate new inode from the
2019 	 * same slot the parent dir belongs to, use the same chunk.
2020 	 *
2021 	 * We are very careful here to avoid the mistake of setting
2022 	 * ac_last_group to a group descriptor from a different (unlocked) slot.
2023 	 */
2024 	if (OCFS2_I(dir)->ip_last_used_group &&
2025 	    OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2026 		ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2027 	else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2028 		if (di->i_suballoc_loc)
2029 			ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2030 		else
2031 			ac->ac_last_group = ocfs2_which_suballoc_group(
2032 					le64_to_cpu(di->i_blkno),
2033 					le16_to_cpu(di->i_suballoc_bit));
2034 	}
2035 }
2036 
2037 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2038 					     struct ocfs2_alloc_context *ac)
2039 {
2040 	OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2041 	OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2042 }
2043 
2044 int ocfs2_find_new_inode_loc(struct inode *dir,
2045 			     struct buffer_head *parent_fe_bh,
2046 			     struct ocfs2_alloc_context *ac,
2047 			     u64 *fe_blkno)
2048 {
2049 	int ret;
2050 	handle_t *handle = NULL;
2051 	struct ocfs2_suballoc_result *res;
2052 
2053 	BUG_ON(!ac);
2054 	BUG_ON(ac->ac_bits_given != 0);
2055 	BUG_ON(ac->ac_bits_wanted != 1);
2056 	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2057 
2058 	res = kzalloc(sizeof(*res), GFP_NOFS);
2059 	if (res == NULL) {
2060 		ret = -ENOMEM;
2061 		mlog_errno(ret);
2062 		goto out;
2063 	}
2064 
2065 	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2066 
2067 	/*
2068 	 * The handle started here is for chain relink. Alternatively,
2069 	 * we could just disable relink for these calls.
2070 	 */
2071 	handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2072 	if (IS_ERR(handle)) {
2073 		ret = PTR_ERR(handle);
2074 		handle = NULL;
2075 		mlog_errno(ret);
2076 		goto out;
2077 	}
2078 
2079 	/*
2080 	 * This will instruct ocfs2_claim_suballoc_bits and
2081 	 * ocfs2_search_one_group to search but save actual allocation
2082 	 * for later.
2083 	 */
2084 	ac->ac_find_loc_only = 1;
2085 
2086 	ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2087 	if (ret < 0) {
2088 		mlog_errno(ret);
2089 		goto out;
2090 	}
2091 
2092 	ac->ac_find_loc_priv = res;
2093 	*fe_blkno = res->sr_blkno;
2094 
2095 out:
2096 	if (handle)
2097 		ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2098 
2099 	if (ret)
2100 		kfree(res);
2101 
2102 	return ret;
2103 }
2104 
2105 int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2106 				 struct inode *dir,
2107 				 struct ocfs2_alloc_context *ac,
2108 				 u64 *suballoc_loc,
2109 				 u16 *suballoc_bit,
2110 				 u64 di_blkno)
2111 {
2112 	int ret;
2113 	u16 chain;
2114 	struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2115 	struct buffer_head *bg_bh = NULL;
2116 	struct ocfs2_group_desc *bg;
2117 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2118 
2119 	/*
2120 	 * Since di_blkno is being passed back in, we check for any
2121 	 * inconsistencies which may have happened between
2122 	 * calls. These are code bugs as di_blkno is not expected to
2123 	 * change once returned from ocfs2_find_new_inode_loc()
2124 	 */
2125 	BUG_ON(res->sr_blkno != di_blkno);
2126 
2127 	ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2128 					  res->sr_bg_stable_blkno, &bg_bh);
2129 	if (ret) {
2130 		mlog_errno(ret);
2131 		goto out;
2132 	}
2133 
2134 	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2135 	chain = le16_to_cpu(bg->bg_chain);
2136 
2137 	ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2138 					       ac->ac_bh, res->sr_bits,
2139 					       chain);
2140 	if (ret) {
2141 		mlog_errno(ret);
2142 		goto out;
2143 	}
2144 
2145 	ret = ocfs2_block_group_set_bits(handle,
2146 					 ac->ac_inode,
2147 					 bg,
2148 					 bg_bh,
2149 					 res->sr_bit_offset,
2150 					 res->sr_bits);
2151 	if (ret < 0) {
2152 		mlog_errno(ret);
2153 		goto out;
2154 	}
2155 
2156 	trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2157 					   res->sr_bits);
2158 
2159 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2160 
2161 	BUG_ON(res->sr_bits != 1);
2162 
2163 	*suballoc_loc = res->sr_bg_blkno;
2164 	*suballoc_bit = res->sr_bit_offset;
2165 	ac->ac_bits_given++;
2166 	ocfs2_save_inode_ac_group(dir, ac);
2167 
2168 out:
2169 	brelse(bg_bh);
2170 
2171 	return ret;
2172 }
2173 
2174 int ocfs2_claim_new_inode(handle_t *handle,
2175 			  struct inode *dir,
2176 			  struct buffer_head *parent_fe_bh,
2177 			  struct ocfs2_alloc_context *ac,
2178 			  u64 *suballoc_loc,
2179 			  u16 *suballoc_bit,
2180 			  u64 *fe_blkno)
2181 {
2182 	int status;
2183 	struct ocfs2_suballoc_result res;
2184 
2185 	BUG_ON(!ac);
2186 	BUG_ON(ac->ac_bits_given != 0);
2187 	BUG_ON(ac->ac_bits_wanted != 1);
2188 	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2189 
2190 	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2191 
2192 	status = ocfs2_claim_suballoc_bits(ac,
2193 					   handle,
2194 					   1,
2195 					   1,
2196 					   &res);
2197 	if (status < 0) {
2198 		mlog_errno(status);
2199 		goto bail;
2200 	}
2201 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2202 
2203 	BUG_ON(res.sr_bits != 1);
2204 
2205 	*suballoc_loc = res.sr_bg_blkno;
2206 	*suballoc_bit = res.sr_bit_offset;
2207 	*fe_blkno = res.sr_blkno;
2208 	ac->ac_bits_given++;
2209 	ocfs2_save_inode_ac_group(dir, ac);
2210 	status = 0;
2211 bail:
2212 	if (status)
2213 		mlog_errno(status);
2214 	return status;
2215 }
2216 
2217 /* translate a group desc. blkno and it's bitmap offset into
2218  * disk cluster offset. */
2219 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2220 						   u64 bg_blkno,
2221 						   u16 bg_bit_off)
2222 {
2223 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2224 	u32 cluster = 0;
2225 
2226 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2227 
2228 	if (bg_blkno != osb->first_cluster_group_blkno)
2229 		cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2230 	cluster += (u32) bg_bit_off;
2231 	return cluster;
2232 }
2233 
2234 /* given a cluster offset, calculate which block group it belongs to
2235  * and return that block offset. */
2236 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2237 {
2238 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2239 	u32 group_no;
2240 
2241 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2242 
2243 	group_no = cluster / osb->bitmap_cpg;
2244 	if (!group_no)
2245 		return osb->first_cluster_group_blkno;
2246 	return ocfs2_clusters_to_blocks(inode->i_sb,
2247 					group_no * osb->bitmap_cpg);
2248 }
2249 
2250 /* given the block number of a cluster start, calculate which cluster
2251  * group and descriptor bitmap offset that corresponds to. */
2252 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2253 						u64 data_blkno,
2254 						u64 *bg_blkno,
2255 						u16 *bg_bit_off)
2256 {
2257 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2258 	u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2259 
2260 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2261 
2262 	*bg_blkno = ocfs2_which_cluster_group(inode,
2263 					      data_cluster);
2264 
2265 	if (*bg_blkno == osb->first_cluster_group_blkno)
2266 		*bg_bit_off = (u16) data_cluster;
2267 	else
2268 		*bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2269 							     data_blkno - *bg_blkno);
2270 }
2271 
2272 /*
2273  * min_bits - minimum contiguous chunk from this total allocation we
2274  * can handle. set to what we asked for originally for a full
2275  * contig. allocation, set to '1' to indicate we can deal with extents
2276  * of any size.
2277  */
2278 int __ocfs2_claim_clusters(handle_t *handle,
2279 			   struct ocfs2_alloc_context *ac,
2280 			   u32 min_clusters,
2281 			   u32 max_clusters,
2282 			   u32 *cluster_start,
2283 			   u32 *num_clusters)
2284 {
2285 	int status;
2286 	unsigned int bits_wanted = max_clusters;
2287 	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2288 	struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2289 
2290 	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2291 
2292 	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2293 	       && ac->ac_which != OCFS2_AC_USE_MAIN);
2294 
2295 	if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2296 		WARN_ON(min_clusters > 1);
2297 
2298 		status = ocfs2_claim_local_alloc_bits(osb,
2299 						      handle,
2300 						      ac,
2301 						      bits_wanted,
2302 						      cluster_start,
2303 						      num_clusters);
2304 		if (!status)
2305 			atomic_inc(&osb->alloc_stats.local_data);
2306 	} else {
2307 		if (min_clusters > (osb->bitmap_cpg - 1)) {
2308 			/* The only paths asking for contiguousness
2309 			 * should know about this already. */
2310 			mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2311 			     "group bitmap size %u!\n", min_clusters,
2312 			     osb->bitmap_cpg);
2313 			status = -ENOSPC;
2314 			goto bail;
2315 		}
2316 		/* clamp the current request down to a realistic size. */
2317 		if (bits_wanted > (osb->bitmap_cpg - 1))
2318 			bits_wanted = osb->bitmap_cpg - 1;
2319 
2320 		status = ocfs2_claim_suballoc_bits(ac,
2321 						   handle,
2322 						   bits_wanted,
2323 						   min_clusters,
2324 						   &res);
2325 		if (!status) {
2326 			BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2327 			*cluster_start =
2328 				ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2329 								 res.sr_bg_blkno,
2330 								 res.sr_bit_offset);
2331 			atomic_inc(&osb->alloc_stats.bitmap_data);
2332 			*num_clusters = res.sr_bits;
2333 		}
2334 	}
2335 	if (status < 0) {
2336 		if (status != -ENOSPC)
2337 			mlog_errno(status);
2338 		goto bail;
2339 	}
2340 
2341 	ac->ac_bits_given += *num_clusters;
2342 
2343 bail:
2344 	if (status)
2345 		mlog_errno(status);
2346 	return status;
2347 }
2348 
2349 int ocfs2_claim_clusters(handle_t *handle,
2350 			 struct ocfs2_alloc_context *ac,
2351 			 u32 min_clusters,
2352 			 u32 *cluster_start,
2353 			 u32 *num_clusters)
2354 {
2355 	unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2356 
2357 	return __ocfs2_claim_clusters(handle, ac, min_clusters,
2358 				      bits_wanted, cluster_start, num_clusters);
2359 }
2360 
2361 static int ocfs2_block_group_clear_bits(handle_t *handle,
2362 					struct inode *alloc_inode,
2363 					struct ocfs2_group_desc *bg,
2364 					struct buffer_head *group_bh,
2365 					unsigned int bit_off,
2366 					unsigned int num_bits,
2367 					void (*undo_fn)(unsigned int bit,
2368 							unsigned long *bmap))
2369 {
2370 	int status;
2371 	unsigned int tmp;
2372 	struct ocfs2_group_desc *undo_bg = NULL;
2373 
2374 	/* The caller got this descriptor from
2375 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
2376 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2377 
2378 	trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2379 
2380 	BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2381 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2382 					 group_bh,
2383 					 undo_fn ?
2384 					 OCFS2_JOURNAL_ACCESS_UNDO :
2385 					 OCFS2_JOURNAL_ACCESS_WRITE);
2386 	if (status < 0) {
2387 		mlog_errno(status);
2388 		goto bail;
2389 	}
2390 
2391 	if (undo_fn) {
2392 		jbd_lock_bh_state(group_bh);
2393 		undo_bg = (struct ocfs2_group_desc *)
2394 					bh2jh(group_bh)->b_committed_data;
2395 		BUG_ON(!undo_bg);
2396 	}
2397 
2398 	tmp = num_bits;
2399 	while(tmp--) {
2400 		ocfs2_clear_bit((bit_off + tmp),
2401 				(unsigned long *) bg->bg_bitmap);
2402 		if (undo_fn)
2403 			undo_fn(bit_off + tmp,
2404 				(unsigned long *) undo_bg->bg_bitmap);
2405 	}
2406 	le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2407 	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2408 		ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
2409 			    " count %u but claims %u are freed. num_bits %d",
2410 			    (unsigned long long)le64_to_cpu(bg->bg_blkno),
2411 			    le16_to_cpu(bg->bg_bits),
2412 			    le16_to_cpu(bg->bg_free_bits_count), num_bits);
2413 		return -EROFS;
2414 	}
2415 
2416 	if (undo_fn)
2417 		jbd_unlock_bh_state(group_bh);
2418 
2419 	ocfs2_journal_dirty(handle, group_bh);
2420 bail:
2421 	return status;
2422 }
2423 
2424 /*
2425  * expects the suballoc inode to already be locked.
2426  */
2427 static int _ocfs2_free_suballoc_bits(handle_t *handle,
2428 				     struct inode *alloc_inode,
2429 				     struct buffer_head *alloc_bh,
2430 				     unsigned int start_bit,
2431 				     u64 bg_blkno,
2432 				     unsigned int count,
2433 				     void (*undo_fn)(unsigned int bit,
2434 						     unsigned long *bitmap))
2435 {
2436 	int status = 0;
2437 	u32 tmp_used;
2438 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2439 	struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2440 	struct buffer_head *group_bh = NULL;
2441 	struct ocfs2_group_desc *group;
2442 
2443 	/* The alloc_bh comes from ocfs2_free_dinode() or
2444 	 * ocfs2_free_clusters().  The callers have all locked the
2445 	 * allocator and gotten alloc_bh from the lock call.  This
2446 	 * validates the dinode buffer.  Any corruption that has happened
2447 	 * is a code bug. */
2448 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2449 	BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2450 
2451 	trace_ocfs2_free_suballoc_bits(
2452 		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2453 		(unsigned long long)bg_blkno,
2454 		start_bit, count);
2455 
2456 	status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2457 					     &group_bh);
2458 	if (status < 0) {
2459 		mlog_errno(status);
2460 		goto bail;
2461 	}
2462 	group = (struct ocfs2_group_desc *) group_bh->b_data;
2463 
2464 	BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2465 
2466 	status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2467 					      group, group_bh,
2468 					      start_bit, count, undo_fn);
2469 	if (status < 0) {
2470 		mlog_errno(status);
2471 		goto bail;
2472 	}
2473 
2474 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2475 					 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2476 	if (status < 0) {
2477 		mlog_errno(status);
2478 		goto bail;
2479 	}
2480 
2481 	le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2482 		     count);
2483 	tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2484 	fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2485 	ocfs2_journal_dirty(handle, alloc_bh);
2486 
2487 bail:
2488 	brelse(group_bh);
2489 
2490 	if (status)
2491 		mlog_errno(status);
2492 	return status;
2493 }
2494 
2495 int ocfs2_free_suballoc_bits(handle_t *handle,
2496 			     struct inode *alloc_inode,
2497 			     struct buffer_head *alloc_bh,
2498 			     unsigned int start_bit,
2499 			     u64 bg_blkno,
2500 			     unsigned int count)
2501 {
2502 	return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2503 					 start_bit, bg_blkno, count, NULL);
2504 }
2505 
2506 int ocfs2_free_dinode(handle_t *handle,
2507 		      struct inode *inode_alloc_inode,
2508 		      struct buffer_head *inode_alloc_bh,
2509 		      struct ocfs2_dinode *di)
2510 {
2511 	u64 blk = le64_to_cpu(di->i_blkno);
2512 	u16 bit = le16_to_cpu(di->i_suballoc_bit);
2513 	u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2514 
2515 	if (di->i_suballoc_loc)
2516 		bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2517 	return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2518 					inode_alloc_bh, bit, bg_blkno, 1);
2519 }
2520 
2521 static int _ocfs2_free_clusters(handle_t *handle,
2522 				struct inode *bitmap_inode,
2523 				struct buffer_head *bitmap_bh,
2524 				u64 start_blk,
2525 				unsigned int num_clusters,
2526 				void (*undo_fn)(unsigned int bit,
2527 						unsigned long *bitmap))
2528 {
2529 	int status;
2530 	u16 bg_start_bit;
2531 	u64 bg_blkno;
2532 	struct ocfs2_dinode *fe;
2533 
2534 	/* You can't ever have a contiguous set of clusters
2535 	 * bigger than a block group bitmap so we never have to worry
2536 	 * about looping on them.
2537 	 * This is expensive. We can safely remove once this stuff has
2538 	 * gotten tested really well. */
2539 	BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
2540 
2541 	fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
2542 
2543 	ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2544 				     &bg_start_bit);
2545 
2546 	trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2547 			(unsigned long long)start_blk,
2548 			bg_start_bit, num_clusters);
2549 
2550 	status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2551 					   bg_start_bit, bg_blkno,
2552 					   num_clusters, undo_fn);
2553 	if (status < 0) {
2554 		mlog_errno(status);
2555 		goto out;
2556 	}
2557 
2558 	ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2559 					 num_clusters);
2560 
2561 out:
2562 	if (status)
2563 		mlog_errno(status);
2564 	return status;
2565 }
2566 
2567 int ocfs2_free_clusters(handle_t *handle,
2568 			struct inode *bitmap_inode,
2569 			struct buffer_head *bitmap_bh,
2570 			u64 start_blk,
2571 			unsigned int num_clusters)
2572 {
2573 	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2574 				    start_blk, num_clusters,
2575 				    _ocfs2_set_bit);
2576 }
2577 
2578 /*
2579  * Give never-used clusters back to the global bitmap.  We don't need
2580  * to protect these bits in the undo buffer.
2581  */
2582 int ocfs2_release_clusters(handle_t *handle,
2583 			   struct inode *bitmap_inode,
2584 			   struct buffer_head *bitmap_bh,
2585 			   u64 start_blk,
2586 			   unsigned int num_clusters)
2587 {
2588 	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2589 				    start_blk, num_clusters,
2590 				    _ocfs2_clear_bit);
2591 }
2592 
2593 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
2594 {
2595 	printk("Block Group:\n");
2596 	printk("bg_signature:       %s\n", bg->bg_signature);
2597 	printk("bg_size:            %u\n", bg->bg_size);
2598 	printk("bg_bits:            %u\n", bg->bg_bits);
2599 	printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
2600 	printk("bg_chain:           %u\n", bg->bg_chain);
2601 	printk("bg_generation:      %u\n", le32_to_cpu(bg->bg_generation));
2602 	printk("bg_next_group:      %llu\n",
2603 	       (unsigned long long)bg->bg_next_group);
2604 	printk("bg_parent_dinode:   %llu\n",
2605 	       (unsigned long long)bg->bg_parent_dinode);
2606 	printk("bg_blkno:           %llu\n",
2607 	       (unsigned long long)bg->bg_blkno);
2608 }
2609 
2610 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2611 {
2612 	int i;
2613 
2614 	printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
2615 	printk("i_signature:                  %s\n", fe->i_signature);
2616 	printk("i_size:                       %llu\n",
2617 	       (unsigned long long)fe->i_size);
2618 	printk("i_clusters:                   %u\n", fe->i_clusters);
2619 	printk("i_generation:                 %u\n",
2620 	       le32_to_cpu(fe->i_generation));
2621 	printk("id1.bitmap1.i_used:           %u\n",
2622 	       le32_to_cpu(fe->id1.bitmap1.i_used));
2623 	printk("id1.bitmap1.i_total:          %u\n",
2624 	       le32_to_cpu(fe->id1.bitmap1.i_total));
2625 	printk("id2.i_chain.cl_cpg:           %u\n", fe->id2.i_chain.cl_cpg);
2626 	printk("id2.i_chain.cl_bpc:           %u\n", fe->id2.i_chain.cl_bpc);
2627 	printk("id2.i_chain.cl_count:         %u\n", fe->id2.i_chain.cl_count);
2628 	printk("id2.i_chain.cl_next_free_rec: %u\n",
2629 	       fe->id2.i_chain.cl_next_free_rec);
2630 	for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2631 		printk("fe->id2.i_chain.cl_recs[%d].c_free:  %u\n", i,
2632 		       fe->id2.i_chain.cl_recs[i].c_free);
2633 		printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2634 		       fe->id2.i_chain.cl_recs[i].c_total);
2635 		printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2636 		       (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
2637 	}
2638 }
2639 
2640 /*
2641  * For a given allocation, determine which allocators will need to be
2642  * accessed, and lock them, reserving the appropriate number of bits.
2643  *
2644  * Sparse file systems call this from ocfs2_write_begin_nolock()
2645  * and ocfs2_allocate_unwritten_extents().
2646  *
2647  * File systems which don't support holes call this from
2648  * ocfs2_extend_allocation().
2649  */
2650 int ocfs2_lock_allocators(struct inode *inode,
2651 			  struct ocfs2_extent_tree *et,
2652 			  u32 clusters_to_add, u32 extents_to_split,
2653 			  struct ocfs2_alloc_context **data_ac,
2654 			  struct ocfs2_alloc_context **meta_ac)
2655 {
2656 	int ret = 0, num_free_extents;
2657 	unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2658 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2659 
2660 	*meta_ac = NULL;
2661 	if (data_ac)
2662 		*data_ac = NULL;
2663 
2664 	BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2665 
2666 	num_free_extents = ocfs2_num_free_extents(osb, et);
2667 	if (num_free_extents < 0) {
2668 		ret = num_free_extents;
2669 		mlog_errno(ret);
2670 		goto out;
2671 	}
2672 
2673 	/*
2674 	 * Sparse allocation file systems need to be more conservative
2675 	 * with reserving room for expansion - the actual allocation
2676 	 * happens while we've got a journal handle open so re-taking
2677 	 * a cluster lock (because we ran out of room for another
2678 	 * extent) will violate ordering rules.
2679 	 *
2680 	 * Most of the time we'll only be seeing this 1 cluster at a time
2681 	 * anyway.
2682 	 *
2683 	 * Always lock for any unwritten extents - we might want to
2684 	 * add blocks during a split.
2685 	 */
2686 	if (!num_free_extents ||
2687 	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2688 		ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2689 		if (ret < 0) {
2690 			if (ret != -ENOSPC)
2691 				mlog_errno(ret);
2692 			goto out;
2693 		}
2694 	}
2695 
2696 	if (clusters_to_add == 0)
2697 		goto out;
2698 
2699 	ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2700 	if (ret < 0) {
2701 		if (ret != -ENOSPC)
2702 			mlog_errno(ret);
2703 		goto out;
2704 	}
2705 
2706 out:
2707 	if (ret) {
2708 		if (*meta_ac) {
2709 			ocfs2_free_alloc_context(*meta_ac);
2710 			*meta_ac = NULL;
2711 		}
2712 
2713 		/*
2714 		 * We cannot have an error and a non null *data_ac.
2715 		 */
2716 	}
2717 
2718 	return ret;
2719 }
2720 
2721 /*
2722  * Read the inode specified by blkno to get suballoc_slot and
2723  * suballoc_bit.
2724  */
2725 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2726 				       u16 *suballoc_slot, u64 *group_blkno,
2727 				       u16 *suballoc_bit)
2728 {
2729 	int status;
2730 	struct buffer_head *inode_bh = NULL;
2731 	struct ocfs2_dinode *inode_fe;
2732 
2733 	trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2734 
2735 	/* dirty read disk */
2736 	status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2737 	if (status < 0) {
2738 		mlog(ML_ERROR, "read block %llu failed %d\n",
2739 		     (unsigned long long)blkno, status);
2740 		goto bail;
2741 	}
2742 
2743 	inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2744 	if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2745 		mlog(ML_ERROR, "invalid inode %llu requested\n",
2746 		     (unsigned long long)blkno);
2747 		status = -EINVAL;
2748 		goto bail;
2749 	}
2750 
2751 	if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2752 	    (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2753 		mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2754 		     (unsigned long long)blkno,
2755 		     (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2756 		status = -EINVAL;
2757 		goto bail;
2758 	}
2759 
2760 	if (suballoc_slot)
2761 		*suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2762 	if (suballoc_bit)
2763 		*suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2764 	if (group_blkno)
2765 		*group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2766 
2767 bail:
2768 	brelse(inode_bh);
2769 
2770 	if (status)
2771 		mlog_errno(status);
2772 	return status;
2773 }
2774 
2775 /*
2776  * test whether bit is SET in allocator bitmap or not.  on success, 0
2777  * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
2778  * is returned and *res is meaningless.  Call this after you have
2779  * cluster locked against suballoc, or you may get a result based on
2780  * non-up2date contents
2781  */
2782 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2783 				   struct inode *suballoc,
2784 				   struct buffer_head *alloc_bh,
2785 				   u64 group_blkno, u64 blkno,
2786 				   u16 bit, int *res)
2787 {
2788 	struct ocfs2_dinode *alloc_di;
2789 	struct ocfs2_group_desc *group;
2790 	struct buffer_head *group_bh = NULL;
2791 	u64 bg_blkno;
2792 	int status;
2793 
2794 	trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2795 				      (unsigned int)bit);
2796 
2797 	alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2798 	if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2799 		mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2800 		     (unsigned int)bit,
2801 		     ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2802 		status = -EINVAL;
2803 		goto bail;
2804 	}
2805 
2806 	bg_blkno = group_blkno ? group_blkno :
2807 		   ocfs2_which_suballoc_group(blkno, bit);
2808 	status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2809 					     &group_bh);
2810 	if (status < 0) {
2811 		mlog(ML_ERROR, "read group %llu failed %d\n",
2812 		     (unsigned long long)bg_blkno, status);
2813 		goto bail;
2814 	}
2815 
2816 	group = (struct ocfs2_group_desc *) group_bh->b_data;
2817 	*res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2818 
2819 bail:
2820 	brelse(group_bh);
2821 
2822 	if (status)
2823 		mlog_errno(status);
2824 	return status;
2825 }
2826 
2827 /*
2828  * Test if the bit representing this inode (blkno) is set in the
2829  * suballocator.
2830  *
2831  * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2832  *
2833  * In the event of failure, a negative value is returned and *res is
2834  * meaningless.
2835  *
2836  * Callers must make sure to hold nfs_sync_lock to prevent
2837  * ocfs2_delete_inode() on another node from accessing the same
2838  * suballocator concurrently.
2839  */
2840 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2841 {
2842 	int status;
2843 	u64 group_blkno = 0;
2844 	u16 suballoc_bit = 0, suballoc_slot = 0;
2845 	struct inode *inode_alloc_inode;
2846 	struct buffer_head *alloc_bh = NULL;
2847 
2848 	trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2849 
2850 	status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2851 					     &group_blkno, &suballoc_bit);
2852 	if (status < 0) {
2853 		mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2854 		goto bail;
2855 	}
2856 
2857 	inode_alloc_inode =
2858 		ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2859 					    suballoc_slot);
2860 	if (!inode_alloc_inode) {
2861 		/* the error code could be inaccurate, but we are not able to
2862 		 * get the correct one. */
2863 		status = -EINVAL;
2864 		mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2865 		     (u32)suballoc_slot);
2866 		goto bail;
2867 	}
2868 
2869 	mutex_lock(&inode_alloc_inode->i_mutex);
2870 	status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2871 	if (status < 0) {
2872 		mutex_unlock(&inode_alloc_inode->i_mutex);
2873 		mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2874 		     (u32)suballoc_slot, status);
2875 		goto bail;
2876 	}
2877 
2878 	status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2879 					 group_blkno, blkno, suballoc_bit, res);
2880 	if (status < 0)
2881 		mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2882 
2883 	ocfs2_inode_unlock(inode_alloc_inode, 0);
2884 	mutex_unlock(&inode_alloc_inode->i_mutex);
2885 
2886 	iput(inode_alloc_inode);
2887 	brelse(alloc_bh);
2888 bail:
2889 	if (status)
2890 		mlog_errno(status);
2891 	return status;
2892 }
2893