• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* -*- mode: c; c-basic-offset: 8; -*-
3  * vim: noexpandtab sw=8 ts=8 sts=0:
4  *
5  * suballoc.c
6  *
7  * metadata alloc and free
8  * Inspired by ext3 block groups.
9  *
10  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
11  */
12 
13 #include <linux/fs.h>
14 #include <linux/types.h>
15 #include <linux/slab.h>
16 #include <linux/highmem.h>
17 
18 #include <cluster/masklog.h>
19 
20 #include "ocfs2.h"
21 
22 #include "alloc.h"
23 #include "blockcheck.h"
24 #include "dlmglue.h"
25 #include "inode.h"
26 #include "journal.h"
27 #include "localalloc.h"
28 #include "suballoc.h"
29 #include "super.h"
30 #include "sysfile.h"
31 #include "uptodate.h"
32 #include "ocfs2_trace.h"
33 
34 #include "buffer_head_io.h"
35 
36 #define NOT_ALLOC_NEW_GROUP		0
37 #define ALLOC_NEW_GROUP			0x1
38 #define ALLOC_GROUPS_FROM_GLOBAL	0x2
39 
40 #define OCFS2_MAX_TO_STEAL		1024
41 
42 struct ocfs2_suballoc_result {
43 	u64		sr_bg_blkno;	/* The bg we allocated from.  Set
44 					   to 0 when a block group is
45 					   contiguous. */
46 	u64		sr_bg_stable_blkno; /*
47 					     * Doesn't change, always
48 					     * set to target block
49 					     * group descriptor
50 					     * block.
51 					     */
52 	u64		sr_blkno;	/* The first allocated block */
53 	unsigned int	sr_bit_offset;	/* The bit in the bg */
54 	unsigned int	sr_bits;	/* How many bits we claimed */
55 };
56 
ocfs2_group_from_res(struct ocfs2_suballoc_result * res)57 static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
58 {
59 	if (res->sr_blkno == 0)
60 		return 0;
61 
62 	if (res->sr_bg_blkno)
63 		return res->sr_bg_blkno;
64 
65 	return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset);
66 }
67 
68 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
69 static int ocfs2_block_group_fill(handle_t *handle,
70 				  struct inode *alloc_inode,
71 				  struct buffer_head *bg_bh,
72 				  u64 group_blkno,
73 				  unsigned int group_clusters,
74 				  u16 my_chain,
75 				  struct ocfs2_chain_list *cl);
76 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
77 				   struct inode *alloc_inode,
78 				   struct buffer_head *bh,
79 				   u64 max_block,
80 				   u64 *last_alloc_group,
81 				   int flags);
82 
83 static int ocfs2_cluster_group_search(struct inode *inode,
84 				      struct buffer_head *group_bh,
85 				      u32 bits_wanted, u32 min_bits,
86 				      u64 max_block,
87 				      struct ocfs2_suballoc_result *res);
88 static int ocfs2_block_group_search(struct inode *inode,
89 				    struct buffer_head *group_bh,
90 				    u32 bits_wanted, u32 min_bits,
91 				    u64 max_block,
92 				    struct ocfs2_suballoc_result *res);
93 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
94 				     handle_t *handle,
95 				     u32 bits_wanted,
96 				     u32 min_bits,
97 				     struct ocfs2_suballoc_result *res);
98 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
99 					 int nr);
100 static int ocfs2_relink_block_group(handle_t *handle,
101 				    struct inode *alloc_inode,
102 				    struct buffer_head *fe_bh,
103 				    struct buffer_head *bg_bh,
104 				    struct buffer_head *prev_bg_bh,
105 				    u16 chain);
106 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
107 						     u32 wanted);
108 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
109 						   u64 bg_blkno,
110 						   u16 bg_bit_off);
111 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
112 						u64 data_blkno,
113 						u64 *bg_blkno,
114 						u16 *bg_bit_off);
115 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
116 					     u32 bits_wanted, u64 max_block,
117 					     int flags,
118 					     struct ocfs2_alloc_context **ac);
119 
ocfs2_free_ac_resource(struct ocfs2_alloc_context * ac)120 void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
121 {
122 	struct inode *inode = ac->ac_inode;
123 
124 	if (inode) {
125 		if (ac->ac_which != OCFS2_AC_USE_LOCAL)
126 			ocfs2_inode_unlock(inode, 1);
127 
128 		inode_unlock(inode);
129 
130 		iput(inode);
131 		ac->ac_inode = NULL;
132 	}
133 	brelse(ac->ac_bh);
134 	ac->ac_bh = NULL;
135 	ac->ac_resv = NULL;
136 	kfree(ac->ac_find_loc_priv);
137 	ac->ac_find_loc_priv = NULL;
138 }
139 
ocfs2_free_alloc_context(struct ocfs2_alloc_context * ac)140 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
141 {
142 	ocfs2_free_ac_resource(ac);
143 	kfree(ac);
144 }
145 
ocfs2_bits_per_group(struct ocfs2_chain_list * cl)146 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
147 {
148 	return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
149 }
150 
151 #define do_error(fmt, ...)						\
152 do {									\
153 	if (resize)							\
154 		mlog(ML_ERROR, fmt, ##__VA_ARGS__);			\
155 	else								\
156 		return ocfs2_error(sb, fmt, ##__VA_ARGS__);		\
157 } while (0)
158 
ocfs2_validate_gd_self(struct super_block * sb,struct buffer_head * bh,int resize)159 static int ocfs2_validate_gd_self(struct super_block *sb,
160 				  struct buffer_head *bh,
161 				  int resize)
162 {
163 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
164 
165 	if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
166 		do_error("Group descriptor #%llu has bad signature %.*s\n",
167 			 (unsigned long long)bh->b_blocknr, 7,
168 			 gd->bg_signature);
169 	}
170 
171 	if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
172 		do_error("Group descriptor #%llu has an invalid bg_blkno of %llu\n",
173 			 (unsigned long long)bh->b_blocknr,
174 			 (unsigned long long)le64_to_cpu(gd->bg_blkno));
175 	}
176 
177 	if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
178 		do_error("Group descriptor #%llu has an invalid fs_generation of #%u\n",
179 			 (unsigned long long)bh->b_blocknr,
180 			 le32_to_cpu(gd->bg_generation));
181 	}
182 
183 	if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
184 		do_error("Group descriptor #%llu has bit count %u but claims that %u are free\n",
185 			 (unsigned long long)bh->b_blocknr,
186 			 le16_to_cpu(gd->bg_bits),
187 			 le16_to_cpu(gd->bg_free_bits_count));
188 	}
189 
190 	if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
191 		do_error("Group descriptor #%llu has bit count %u but max bitmap bits of %u\n",
192 			 (unsigned long long)bh->b_blocknr,
193 			 le16_to_cpu(gd->bg_bits),
194 			 8 * le16_to_cpu(gd->bg_size));
195 	}
196 
197 	return 0;
198 }
199 
ocfs2_validate_gd_parent(struct super_block * sb,struct ocfs2_dinode * di,struct buffer_head * bh,int resize)200 static int ocfs2_validate_gd_parent(struct super_block *sb,
201 				    struct ocfs2_dinode *di,
202 				    struct buffer_head *bh,
203 				    int resize)
204 {
205 	unsigned int max_bits;
206 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
207 
208 	if (di->i_blkno != gd->bg_parent_dinode) {
209 		do_error("Group descriptor #%llu has bad parent pointer (%llu, expected %llu)\n",
210 			 (unsigned long long)bh->b_blocknr,
211 			 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
212 			 (unsigned long long)le64_to_cpu(di->i_blkno));
213 	}
214 
215 	max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
216 	if (le16_to_cpu(gd->bg_bits) > max_bits) {
217 		do_error("Group descriptor #%llu has bit count of %u\n",
218 			 (unsigned long long)bh->b_blocknr,
219 			 le16_to_cpu(gd->bg_bits));
220 	}
221 
222 	/* In resize, we may meet the case bg_chain == cl_next_free_rec. */
223 	if ((le16_to_cpu(gd->bg_chain) >
224 	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
225 	    ((le16_to_cpu(gd->bg_chain) ==
226 	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
227 		do_error("Group descriptor #%llu has bad chain %u\n",
228 			 (unsigned long long)bh->b_blocknr,
229 			 le16_to_cpu(gd->bg_chain));
230 	}
231 
232 	return 0;
233 }
234 
235 #undef do_error
236 
237 /*
238  * This version only prints errors.  It does not fail the filesystem, and
239  * exists only for resize.
240  */
ocfs2_check_group_descriptor(struct super_block * sb,struct ocfs2_dinode * di,struct buffer_head * bh)241 int ocfs2_check_group_descriptor(struct super_block *sb,
242 				 struct ocfs2_dinode *di,
243 				 struct buffer_head *bh)
244 {
245 	int rc;
246 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
247 
248 	BUG_ON(!buffer_uptodate(bh));
249 
250 	/*
251 	 * If the ecc fails, we return the error but otherwise
252 	 * leave the filesystem running.  We know any error is
253 	 * local to this block.
254 	 */
255 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
256 	if (rc) {
257 		mlog(ML_ERROR,
258 		     "Checksum failed for group descriptor %llu\n",
259 		     (unsigned long long)bh->b_blocknr);
260 	} else
261 		rc = ocfs2_validate_gd_self(sb, bh, 1);
262 	if (!rc)
263 		rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
264 
265 	return rc;
266 }
267 
ocfs2_validate_group_descriptor(struct super_block * sb,struct buffer_head * bh)268 static int ocfs2_validate_group_descriptor(struct super_block *sb,
269 					   struct buffer_head *bh)
270 {
271 	int rc;
272 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
273 
274 	trace_ocfs2_validate_group_descriptor(
275 					(unsigned long long)bh->b_blocknr);
276 
277 	BUG_ON(!buffer_uptodate(bh));
278 
279 	/*
280 	 * If the ecc fails, we return the error but otherwise
281 	 * leave the filesystem running.  We know any error is
282 	 * local to this block.
283 	 */
284 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
285 	if (rc)
286 		return rc;
287 
288 	/*
289 	 * Errors after here are fatal.
290 	 */
291 
292 	return ocfs2_validate_gd_self(sb, bh, 0);
293 }
294 
ocfs2_read_group_descriptor(struct inode * inode,struct ocfs2_dinode * di,u64 gd_blkno,struct buffer_head ** bh)295 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
296 				u64 gd_blkno, struct buffer_head **bh)
297 {
298 	int rc;
299 	struct buffer_head *tmp = *bh;
300 
301 	rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
302 			      ocfs2_validate_group_descriptor);
303 	if (rc)
304 		goto out;
305 
306 	rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
307 	if (rc) {
308 		brelse(tmp);
309 		goto out;
310 	}
311 
312 	/* If ocfs2_read_block() got us a new bh, pass it up. */
313 	if (!*bh)
314 		*bh = tmp;
315 
316 out:
317 	return rc;
318 }
319 
ocfs2_bg_discontig_add_extent(struct ocfs2_super * osb,struct ocfs2_group_desc * bg,struct ocfs2_chain_list * cl,u64 p_blkno,unsigned int clusters)320 static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
321 					  struct ocfs2_group_desc *bg,
322 					  struct ocfs2_chain_list *cl,
323 					  u64 p_blkno, unsigned int clusters)
324 {
325 	struct ocfs2_extent_list *el = &bg->bg_list;
326 	struct ocfs2_extent_rec *rec;
327 
328 	BUG_ON(!ocfs2_supports_discontig_bg(osb));
329 	if (!el->l_next_free_rec)
330 		el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
331 	rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
332 	rec->e_blkno = cpu_to_le64(p_blkno);
333 	rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
334 				  le16_to_cpu(cl->cl_bpc));
335 	rec->e_leaf_clusters = cpu_to_le16(clusters);
336 	le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
337 	le16_add_cpu(&bg->bg_free_bits_count,
338 		     clusters * le16_to_cpu(cl->cl_bpc));
339 	le16_add_cpu(&el->l_next_free_rec, 1);
340 }
341 
ocfs2_block_group_fill(handle_t * handle,struct inode * alloc_inode,struct buffer_head * bg_bh,u64 group_blkno,unsigned int group_clusters,u16 my_chain,struct ocfs2_chain_list * cl)342 static int ocfs2_block_group_fill(handle_t *handle,
343 				  struct inode *alloc_inode,
344 				  struct buffer_head *bg_bh,
345 				  u64 group_blkno,
346 				  unsigned int group_clusters,
347 				  u16 my_chain,
348 				  struct ocfs2_chain_list *cl)
349 {
350 	int status = 0;
351 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
352 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
353 	struct super_block * sb = alloc_inode->i_sb;
354 
355 	if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
356 		status = ocfs2_error(alloc_inode->i_sb,
357 				     "group block (%llu) != b_blocknr (%llu)\n",
358 				     (unsigned long long)group_blkno,
359 				     (unsigned long long) bg_bh->b_blocknr);
360 		goto bail;
361 	}
362 
363 	status = ocfs2_journal_access_gd(handle,
364 					 INODE_CACHE(alloc_inode),
365 					 bg_bh,
366 					 OCFS2_JOURNAL_ACCESS_CREATE);
367 	if (status < 0) {
368 		mlog_errno(status);
369 		goto bail;
370 	}
371 
372 	memset(bg, 0, sb->s_blocksize);
373 	strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
374 	bg->bg_generation = cpu_to_le32(osb->fs_generation);
375 	bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
376 						osb->s_feature_incompat));
377 	bg->bg_chain = cpu_to_le16(my_chain);
378 	bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
379 	bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
380 	bg->bg_blkno = cpu_to_le64(group_blkno);
381 	if (group_clusters == le16_to_cpu(cl->cl_cpg))
382 		bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
383 	else
384 		ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
385 					      group_clusters);
386 
387 	/* set the 1st bit in the bitmap to account for the descriptor block */
388 	ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
389 	bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
390 
391 	ocfs2_journal_dirty(handle, bg_bh);
392 
393 	/* There is no need to zero out or otherwise initialize the
394 	 * other blocks in a group - All valid FS metadata in a block
395 	 * group stores the superblock fs_generation value at
396 	 * allocation time. */
397 
398 bail:
399 	if (status)
400 		mlog_errno(status);
401 	return status;
402 }
403 
ocfs2_find_smallest_chain(struct ocfs2_chain_list * cl)404 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
405 {
406 	u16 curr, best;
407 
408 	best = curr = 0;
409 	while (curr < le16_to_cpu(cl->cl_count)) {
410 		if (le32_to_cpu(cl->cl_recs[best].c_total) >
411 		    le32_to_cpu(cl->cl_recs[curr].c_total))
412 			best = curr;
413 		curr++;
414 	}
415 	return best;
416 }
417 
418 static struct buffer_head *
ocfs2_block_group_alloc_contig(struct ocfs2_super * osb,handle_t * handle,struct inode * alloc_inode,struct ocfs2_alloc_context * ac,struct ocfs2_chain_list * cl)419 ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
420 			       struct inode *alloc_inode,
421 			       struct ocfs2_alloc_context *ac,
422 			       struct ocfs2_chain_list *cl)
423 {
424 	int status;
425 	u32 bit_off, num_bits;
426 	u64 bg_blkno;
427 	struct buffer_head *bg_bh;
428 	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
429 
430 	status = ocfs2_claim_clusters(handle, ac,
431 				      le16_to_cpu(cl->cl_cpg), &bit_off,
432 				      &num_bits);
433 	if (status < 0) {
434 		if (status != -ENOSPC)
435 			mlog_errno(status);
436 		goto bail;
437 	}
438 
439 	/* setup the group */
440 	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
441 	trace_ocfs2_block_group_alloc_contig(
442 	     (unsigned long long)bg_blkno, alloc_rec);
443 
444 	bg_bh = sb_getblk(osb->sb, bg_blkno);
445 	if (!bg_bh) {
446 		status = -ENOMEM;
447 		mlog_errno(status);
448 		goto bail;
449 	}
450 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
451 
452 	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
453 					bg_blkno, num_bits, alloc_rec, cl);
454 	if (status < 0) {
455 		brelse(bg_bh);
456 		mlog_errno(status);
457 	}
458 
459 bail:
460 	return status ? ERR_PTR(status) : bg_bh;
461 }
462 
ocfs2_block_group_claim_bits(struct ocfs2_super * osb,handle_t * handle,struct ocfs2_alloc_context * ac,unsigned int min_bits,u32 * bit_off,u32 * num_bits)463 static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
464 					handle_t *handle,
465 					struct ocfs2_alloc_context *ac,
466 					unsigned int min_bits,
467 					u32 *bit_off, u32 *num_bits)
468 {
469 	int status = 0;
470 
471 	while (min_bits) {
472 		status = ocfs2_claim_clusters(handle, ac, min_bits,
473 					      bit_off, num_bits);
474 		if (status != -ENOSPC)
475 			break;
476 
477 		min_bits >>= 1;
478 	}
479 
480 	return status;
481 }
482 
ocfs2_block_group_grow_discontig(handle_t * handle,struct inode * alloc_inode,struct buffer_head * bg_bh,struct ocfs2_alloc_context * ac,struct ocfs2_chain_list * cl,unsigned int min_bits)483 static int ocfs2_block_group_grow_discontig(handle_t *handle,
484 					    struct inode *alloc_inode,
485 					    struct buffer_head *bg_bh,
486 					    struct ocfs2_alloc_context *ac,
487 					    struct ocfs2_chain_list *cl,
488 					    unsigned int min_bits)
489 {
490 	int status;
491 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
492 	struct ocfs2_group_desc *bg =
493 		(struct ocfs2_group_desc *)bg_bh->b_data;
494 	unsigned int needed = le16_to_cpu(cl->cl_cpg) -
495 			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
496 	u32 p_cpos, clusters;
497 	u64 p_blkno;
498 	struct ocfs2_extent_list *el = &bg->bg_list;
499 
500 	status = ocfs2_journal_access_gd(handle,
501 					 INODE_CACHE(alloc_inode),
502 					 bg_bh,
503 					 OCFS2_JOURNAL_ACCESS_CREATE);
504 	if (status < 0) {
505 		mlog_errno(status);
506 		goto bail;
507 	}
508 
509 	while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
510 				le16_to_cpu(el->l_count))) {
511 		if (min_bits > needed)
512 			min_bits = needed;
513 		status = ocfs2_block_group_claim_bits(osb, handle, ac,
514 						      min_bits, &p_cpos,
515 						      &clusters);
516 		if (status < 0) {
517 			if (status != -ENOSPC)
518 				mlog_errno(status);
519 			goto bail;
520 		}
521 		p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
522 		ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
523 					      clusters);
524 
525 		min_bits = clusters;
526 		needed = le16_to_cpu(cl->cl_cpg) -
527 			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
528 	}
529 
530 	if (needed > 0) {
531 		/*
532 		 * We have used up all the extent rec but can't fill up
533 		 * the cpg. So bail out.
534 		 */
535 		status = -ENOSPC;
536 		goto bail;
537 	}
538 
539 	ocfs2_journal_dirty(handle, bg_bh);
540 
541 bail:
542 	return status;
543 }
544 
ocfs2_bg_alloc_cleanup(handle_t * handle,struct ocfs2_alloc_context * cluster_ac,struct inode * alloc_inode,struct buffer_head * bg_bh)545 static void ocfs2_bg_alloc_cleanup(handle_t *handle,
546 				   struct ocfs2_alloc_context *cluster_ac,
547 				   struct inode *alloc_inode,
548 				   struct buffer_head *bg_bh)
549 {
550 	int i, ret;
551 	struct ocfs2_group_desc *bg;
552 	struct ocfs2_extent_list *el;
553 	struct ocfs2_extent_rec *rec;
554 
555 	if (!bg_bh)
556 		return;
557 
558 	bg = (struct ocfs2_group_desc *)bg_bh->b_data;
559 	el = &bg->bg_list;
560 	for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
561 		rec = &el->l_recs[i];
562 		ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
563 					  cluster_ac->ac_bh,
564 					  le64_to_cpu(rec->e_blkno),
565 					  le16_to_cpu(rec->e_leaf_clusters));
566 		if (ret)
567 			mlog_errno(ret);
568 		/* Try all the clusters to free */
569 	}
570 
571 	ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
572 	brelse(bg_bh);
573 }
574 
575 static struct buffer_head *
ocfs2_block_group_alloc_discontig(handle_t * handle,struct inode * alloc_inode,struct ocfs2_alloc_context * ac,struct ocfs2_chain_list * cl)576 ocfs2_block_group_alloc_discontig(handle_t *handle,
577 				  struct inode *alloc_inode,
578 				  struct ocfs2_alloc_context *ac,
579 				  struct ocfs2_chain_list *cl)
580 {
581 	int status;
582 	u32 bit_off, num_bits;
583 	u64 bg_blkno;
584 	unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
585 	struct buffer_head *bg_bh = NULL;
586 	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
587 	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
588 
589 	if (!ocfs2_supports_discontig_bg(osb)) {
590 		status = -ENOSPC;
591 		goto bail;
592 	}
593 
594 	status = ocfs2_extend_trans(handle,
595 				    ocfs2_calc_bg_discontig_credits(osb->sb));
596 	if (status) {
597 		mlog_errno(status);
598 		goto bail;
599 	}
600 
601 	/*
602 	 * We're going to be grabbing from multiple cluster groups.
603 	 * We don't have enough credits to relink them all, and the
604 	 * cluster groups will be staying in cache for the duration of
605 	 * this operation.
606 	 */
607 	ac->ac_disable_chain_relink = 1;
608 
609 	/* Claim the first region */
610 	status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
611 					      &bit_off, &num_bits);
612 	if (status < 0) {
613 		if (status != -ENOSPC)
614 			mlog_errno(status);
615 		goto bail;
616 	}
617 	min_bits = num_bits;
618 
619 	/* setup the group */
620 	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
621 	trace_ocfs2_block_group_alloc_discontig(
622 				(unsigned long long)bg_blkno, alloc_rec);
623 
624 	bg_bh = sb_getblk(osb->sb, bg_blkno);
625 	if (!bg_bh) {
626 		status = -ENOMEM;
627 		mlog_errno(status);
628 		goto bail;
629 	}
630 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
631 
632 	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
633 					bg_blkno, num_bits, alloc_rec, cl);
634 	if (status < 0) {
635 		mlog_errno(status);
636 		goto bail;
637 	}
638 
639 	status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
640 						  bg_bh, ac, cl, min_bits);
641 	if (status)
642 		mlog_errno(status);
643 
644 bail:
645 	if (status)
646 		ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
647 	return status ? ERR_PTR(status) : bg_bh;
648 }
649 
650 /*
651  * We expect the block group allocator to already be locked.
652  */
ocfs2_block_group_alloc(struct ocfs2_super * osb,struct inode * alloc_inode,struct buffer_head * bh,u64 max_block,u64 * last_alloc_group,int flags)653 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
654 				   struct inode *alloc_inode,
655 				   struct buffer_head *bh,
656 				   u64 max_block,
657 				   u64 *last_alloc_group,
658 				   int flags)
659 {
660 	int status, credits;
661 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
662 	struct ocfs2_chain_list *cl;
663 	struct ocfs2_alloc_context *ac = NULL;
664 	handle_t *handle = NULL;
665 	u16 alloc_rec;
666 	struct buffer_head *bg_bh = NULL;
667 	struct ocfs2_group_desc *bg;
668 
669 	BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
670 
671 	cl = &fe->id2.i_chain;
672 	status = ocfs2_reserve_clusters_with_limit(osb,
673 						   le16_to_cpu(cl->cl_cpg),
674 						   max_block, flags, &ac);
675 	if (status < 0) {
676 		if (status != -ENOSPC)
677 			mlog_errno(status);
678 		goto bail;
679 	}
680 
681 	credits = ocfs2_calc_group_alloc_credits(osb->sb,
682 						 le16_to_cpu(cl->cl_cpg));
683 	handle = ocfs2_start_trans(osb, credits);
684 	if (IS_ERR(handle)) {
685 		status = PTR_ERR(handle);
686 		handle = NULL;
687 		mlog_errno(status);
688 		goto bail;
689 	}
690 
691 	if (last_alloc_group && *last_alloc_group != 0) {
692 		trace_ocfs2_block_group_alloc(
693 				(unsigned long long)*last_alloc_group);
694 		ac->ac_last_group = *last_alloc_group;
695 	}
696 
697 	bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
698 					       ac, cl);
699 	if (PTR_ERR(bg_bh) == -ENOSPC)
700 		bg_bh = ocfs2_block_group_alloc_discontig(handle,
701 							  alloc_inode,
702 							  ac, cl);
703 	if (IS_ERR(bg_bh)) {
704 		status = PTR_ERR(bg_bh);
705 		bg_bh = NULL;
706 		if (status != -ENOSPC)
707 			mlog_errno(status);
708 		goto bail;
709 	}
710 	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
711 
712 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
713 					 bh, OCFS2_JOURNAL_ACCESS_WRITE);
714 	if (status < 0) {
715 		mlog_errno(status);
716 		goto bail;
717 	}
718 
719 	alloc_rec = le16_to_cpu(bg->bg_chain);
720 	le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
721 		     le16_to_cpu(bg->bg_free_bits_count));
722 	le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
723 		     le16_to_cpu(bg->bg_bits));
724 	cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
725 	if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
726 		le16_add_cpu(&cl->cl_next_free_rec, 1);
727 
728 	le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
729 					le16_to_cpu(bg->bg_free_bits_count));
730 	le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
731 	le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
732 
733 	ocfs2_journal_dirty(handle, bh);
734 
735 	spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
736 	OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
737 	fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
738 					     le32_to_cpu(fe->i_clusters)));
739 	spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
740 	i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
741 	alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
742 	ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
743 
744 	status = 0;
745 
746 	/* save the new last alloc group so that the caller can cache it. */
747 	if (last_alloc_group)
748 		*last_alloc_group = ac->ac_last_group;
749 
750 bail:
751 	if (handle)
752 		ocfs2_commit_trans(osb, handle);
753 
754 	if (ac)
755 		ocfs2_free_alloc_context(ac);
756 
757 	brelse(bg_bh);
758 
759 	if (status)
760 		mlog_errno(status);
761 	return status;
762 }
763 
ocfs2_reserve_suballoc_bits(struct ocfs2_super * osb,struct ocfs2_alloc_context * ac,int type,u32 slot,u64 * last_alloc_group,int flags)764 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
765 				       struct ocfs2_alloc_context *ac,
766 				       int type,
767 				       u32 slot,
768 				       u64 *last_alloc_group,
769 				       int flags)
770 {
771 	int status;
772 	u32 bits_wanted = ac->ac_bits_wanted;
773 	struct inode *alloc_inode;
774 	struct buffer_head *bh = NULL;
775 	struct ocfs2_dinode *fe;
776 	u32 free_bits;
777 
778 	alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
779 	if (!alloc_inode) {
780 		mlog_errno(-EINVAL);
781 		return -EINVAL;
782 	}
783 
784 	inode_lock(alloc_inode);
785 
786 	status = ocfs2_inode_lock(alloc_inode, &bh, 1);
787 	if (status < 0) {
788 		inode_unlock(alloc_inode);
789 		iput(alloc_inode);
790 
791 		mlog_errno(status);
792 		return status;
793 	}
794 
795 	ac->ac_inode = alloc_inode;
796 	ac->ac_alloc_slot = slot;
797 
798 	fe = (struct ocfs2_dinode *) bh->b_data;
799 
800 	/* The bh was validated by the inode read inside
801 	 * ocfs2_inode_lock().  Any corruption is a code bug. */
802 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
803 
804 	if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
805 		status = ocfs2_error(alloc_inode->i_sb,
806 				     "Invalid chain allocator %llu\n",
807 				     (unsigned long long)le64_to_cpu(fe->i_blkno));
808 		goto bail;
809 	}
810 
811 	free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
812 		le32_to_cpu(fe->id1.bitmap1.i_used);
813 
814 	if (bits_wanted > free_bits) {
815 		/* cluster bitmap never grows */
816 		if (ocfs2_is_cluster_bitmap(alloc_inode)) {
817 			trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
818 								free_bits);
819 			status = -ENOSPC;
820 			goto bail;
821 		}
822 
823 		if (!(flags & ALLOC_NEW_GROUP)) {
824 			trace_ocfs2_reserve_suballoc_bits_no_new_group(
825 						slot, bits_wanted, free_bits);
826 			status = -ENOSPC;
827 			goto bail;
828 		}
829 
830 		status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
831 						 ac->ac_max_block,
832 						 last_alloc_group, flags);
833 		if (status < 0) {
834 			if (status != -ENOSPC)
835 				mlog_errno(status);
836 			goto bail;
837 		}
838 		atomic_inc(&osb->alloc_stats.bg_extends);
839 
840 		/* You should never ask for this much metadata */
841 		BUG_ON(bits_wanted >
842 		       (le32_to_cpu(fe->id1.bitmap1.i_total)
843 			- le32_to_cpu(fe->id1.bitmap1.i_used)));
844 	}
845 
846 	get_bh(bh);
847 	ac->ac_bh = bh;
848 bail:
849 	brelse(bh);
850 
851 	if (status)
852 		mlog_errno(status);
853 	return status;
854 }
855 
ocfs2_init_inode_steal_slot(struct ocfs2_super * osb)856 static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
857 {
858 	spin_lock(&osb->osb_lock);
859 	osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
860 	spin_unlock(&osb->osb_lock);
861 	atomic_set(&osb->s_num_inodes_stolen, 0);
862 }
863 
ocfs2_init_meta_steal_slot(struct ocfs2_super * osb)864 static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
865 {
866 	spin_lock(&osb->osb_lock);
867 	osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
868 	spin_unlock(&osb->osb_lock);
869 	atomic_set(&osb->s_num_meta_stolen, 0);
870 }
871 
ocfs2_init_steal_slots(struct ocfs2_super * osb)872 void ocfs2_init_steal_slots(struct ocfs2_super *osb)
873 {
874 	ocfs2_init_inode_steal_slot(osb);
875 	ocfs2_init_meta_steal_slot(osb);
876 }
877 
__ocfs2_set_steal_slot(struct ocfs2_super * osb,int slot,int type)878 static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
879 {
880 	spin_lock(&osb->osb_lock);
881 	if (type == INODE_ALLOC_SYSTEM_INODE)
882 		osb->s_inode_steal_slot = (u16)slot;
883 	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
884 		osb->s_meta_steal_slot = (u16)slot;
885 	spin_unlock(&osb->osb_lock);
886 }
887 
__ocfs2_get_steal_slot(struct ocfs2_super * osb,int type)888 static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
889 {
890 	int slot = OCFS2_INVALID_SLOT;
891 
892 	spin_lock(&osb->osb_lock);
893 	if (type == INODE_ALLOC_SYSTEM_INODE)
894 		slot = osb->s_inode_steal_slot;
895 	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
896 		slot = osb->s_meta_steal_slot;
897 	spin_unlock(&osb->osb_lock);
898 
899 	return slot;
900 }
901 
ocfs2_get_inode_steal_slot(struct ocfs2_super * osb)902 static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
903 {
904 	return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
905 }
906 
ocfs2_get_meta_steal_slot(struct ocfs2_super * osb)907 static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
908 {
909 	return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
910 }
911 
ocfs2_steal_resource(struct ocfs2_super * osb,struct ocfs2_alloc_context * ac,int type)912 static int ocfs2_steal_resource(struct ocfs2_super *osb,
913 				struct ocfs2_alloc_context *ac,
914 				int type)
915 {
916 	int i, status = -ENOSPC;
917 	int slot = __ocfs2_get_steal_slot(osb, type);
918 
919 	/* Start to steal resource from the first slot after ours. */
920 	if (slot == OCFS2_INVALID_SLOT)
921 		slot = osb->slot_num + 1;
922 
923 	for (i = 0; i < osb->max_slots; i++, slot++) {
924 		if (slot == osb->max_slots)
925 			slot = 0;
926 
927 		if (slot == osb->slot_num)
928 			continue;
929 
930 		status = ocfs2_reserve_suballoc_bits(osb, ac,
931 						     type,
932 						     (u32)slot, NULL,
933 						     NOT_ALLOC_NEW_GROUP);
934 		if (status >= 0) {
935 			__ocfs2_set_steal_slot(osb, slot, type);
936 			break;
937 		}
938 
939 		ocfs2_free_ac_resource(ac);
940 	}
941 
942 	return status;
943 }
944 
ocfs2_steal_inode(struct ocfs2_super * osb,struct ocfs2_alloc_context * ac)945 static int ocfs2_steal_inode(struct ocfs2_super *osb,
946 			     struct ocfs2_alloc_context *ac)
947 {
948 	return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
949 }
950 
ocfs2_steal_meta(struct ocfs2_super * osb,struct ocfs2_alloc_context * ac)951 static int ocfs2_steal_meta(struct ocfs2_super *osb,
952 			    struct ocfs2_alloc_context *ac)
953 {
954 	return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
955 }
956 
ocfs2_reserve_new_metadata_blocks(struct ocfs2_super * osb,int blocks,struct ocfs2_alloc_context ** ac)957 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
958 				      int blocks,
959 				      struct ocfs2_alloc_context **ac)
960 {
961 	int status;
962 	int slot = ocfs2_get_meta_steal_slot(osb);
963 
964 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
965 	if (!(*ac)) {
966 		status = -ENOMEM;
967 		mlog_errno(status);
968 		goto bail;
969 	}
970 
971 	(*ac)->ac_bits_wanted = blocks;
972 	(*ac)->ac_which = OCFS2_AC_USE_META;
973 	(*ac)->ac_group_search = ocfs2_block_group_search;
974 
975 	if (slot != OCFS2_INVALID_SLOT &&
976 		atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
977 		goto extent_steal;
978 
979 	atomic_set(&osb->s_num_meta_stolen, 0);
980 	status = ocfs2_reserve_suballoc_bits(osb, (*ac),
981 					     EXTENT_ALLOC_SYSTEM_INODE,
982 					     (u32)osb->slot_num, NULL,
983 					     ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
984 
985 
986 	if (status >= 0) {
987 		status = 0;
988 		if (slot != OCFS2_INVALID_SLOT)
989 			ocfs2_init_meta_steal_slot(osb);
990 		goto bail;
991 	} else if (status < 0 && status != -ENOSPC) {
992 		mlog_errno(status);
993 		goto bail;
994 	}
995 
996 	ocfs2_free_ac_resource(*ac);
997 
998 extent_steal:
999 	status = ocfs2_steal_meta(osb, *ac);
1000 	atomic_inc(&osb->s_num_meta_stolen);
1001 	if (status < 0) {
1002 		if (status != -ENOSPC)
1003 			mlog_errno(status);
1004 		goto bail;
1005 	}
1006 
1007 	status = 0;
1008 bail:
1009 	if ((status < 0) && *ac) {
1010 		ocfs2_free_alloc_context(*ac);
1011 		*ac = NULL;
1012 	}
1013 
1014 	if (status)
1015 		mlog_errno(status);
1016 	return status;
1017 }
1018 
ocfs2_reserve_new_metadata(struct ocfs2_super * osb,struct ocfs2_extent_list * root_el,struct ocfs2_alloc_context ** ac)1019 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1020 			       struct ocfs2_extent_list *root_el,
1021 			       struct ocfs2_alloc_context **ac)
1022 {
1023 	return ocfs2_reserve_new_metadata_blocks(osb,
1024 					ocfs2_extend_meta_needed(root_el),
1025 					ac);
1026 }
1027 
ocfs2_reserve_new_inode(struct ocfs2_super * osb,struct ocfs2_alloc_context ** ac)1028 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
1029 			    struct ocfs2_alloc_context **ac)
1030 {
1031 	int status;
1032 	int slot = ocfs2_get_inode_steal_slot(osb);
1033 	u64 alloc_group;
1034 
1035 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1036 	if (!(*ac)) {
1037 		status = -ENOMEM;
1038 		mlog_errno(status);
1039 		goto bail;
1040 	}
1041 
1042 	(*ac)->ac_bits_wanted = 1;
1043 	(*ac)->ac_which = OCFS2_AC_USE_INODE;
1044 
1045 	(*ac)->ac_group_search = ocfs2_block_group_search;
1046 
1047 	/*
1048 	 * stat(2) can't handle i_ino > 32bits, so we tell the
1049 	 * lower levels not to allocate us a block group past that
1050 	 * limit.  The 'inode64' mount option avoids this behavior.
1051 	 */
1052 	if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1053 		(*ac)->ac_max_block = (u32)~0U;
1054 
1055 	/*
1056 	 * slot is set when we successfully steal inode from other nodes.
1057 	 * It is reset in 3 places:
1058 	 * 1. when we flush the truncate log
1059 	 * 2. when we complete local alloc recovery.
1060 	 * 3. when we successfully allocate from our own slot.
1061 	 * After it is set, we will go on stealing inodes until we find the
1062 	 * need to check our slots to see whether there is some space for us.
1063 	 */
1064 	if (slot != OCFS2_INVALID_SLOT &&
1065 	    atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
1066 		goto inode_steal;
1067 
1068 	atomic_set(&osb->s_num_inodes_stolen, 0);
1069 	alloc_group = osb->osb_inode_alloc_group;
1070 	status = ocfs2_reserve_suballoc_bits(osb, *ac,
1071 					     INODE_ALLOC_SYSTEM_INODE,
1072 					     (u32)osb->slot_num,
1073 					     &alloc_group,
1074 					     ALLOC_NEW_GROUP |
1075 					     ALLOC_GROUPS_FROM_GLOBAL);
1076 	if (status >= 0) {
1077 		status = 0;
1078 
1079 		spin_lock(&osb->osb_lock);
1080 		osb->osb_inode_alloc_group = alloc_group;
1081 		spin_unlock(&osb->osb_lock);
1082 		trace_ocfs2_reserve_new_inode_new_group(
1083 			(unsigned long long)alloc_group);
1084 
1085 		/*
1086 		 * Some inodes must be freed by us, so try to allocate
1087 		 * from our own next time.
1088 		 */
1089 		if (slot != OCFS2_INVALID_SLOT)
1090 			ocfs2_init_inode_steal_slot(osb);
1091 		goto bail;
1092 	} else if (status < 0 && status != -ENOSPC) {
1093 		mlog_errno(status);
1094 		goto bail;
1095 	}
1096 
1097 	ocfs2_free_ac_resource(*ac);
1098 
1099 inode_steal:
1100 	status = ocfs2_steal_inode(osb, *ac);
1101 	atomic_inc(&osb->s_num_inodes_stolen);
1102 	if (status < 0) {
1103 		if (status != -ENOSPC)
1104 			mlog_errno(status);
1105 		goto bail;
1106 	}
1107 
1108 	status = 0;
1109 bail:
1110 	if ((status < 0) && *ac) {
1111 		ocfs2_free_alloc_context(*ac);
1112 		*ac = NULL;
1113 	}
1114 
1115 	if (status)
1116 		mlog_errno(status);
1117 	return status;
1118 }
1119 
1120 /* local alloc code has to do the same thing, so rather than do this
1121  * twice.. */
ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super * osb,struct ocfs2_alloc_context * ac)1122 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1123 				      struct ocfs2_alloc_context *ac)
1124 {
1125 	int status;
1126 
1127 	ac->ac_which = OCFS2_AC_USE_MAIN;
1128 	ac->ac_group_search = ocfs2_cluster_group_search;
1129 
1130 	status = ocfs2_reserve_suballoc_bits(osb, ac,
1131 					     GLOBAL_BITMAP_SYSTEM_INODE,
1132 					     OCFS2_INVALID_SLOT, NULL,
1133 					     ALLOC_NEW_GROUP);
1134 	if (status < 0 && status != -ENOSPC)
1135 		mlog_errno(status);
1136 
1137 	return status;
1138 }
1139 
1140 /* Callers don't need to care which bitmap (local alloc or main) to
1141  * use so we figure it out for them, but unfortunately this clutters
1142  * things a bit. */
ocfs2_reserve_clusters_with_limit(struct ocfs2_super * osb,u32 bits_wanted,u64 max_block,int flags,struct ocfs2_alloc_context ** ac)1143 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1144 					     u32 bits_wanted, u64 max_block,
1145 					     int flags,
1146 					     struct ocfs2_alloc_context **ac)
1147 {
1148 	int status, ret = 0;
1149 	int retried = 0;
1150 
1151 	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1152 	if (!(*ac)) {
1153 		status = -ENOMEM;
1154 		mlog_errno(status);
1155 		goto bail;
1156 	}
1157 
1158 	(*ac)->ac_bits_wanted = bits_wanted;
1159 	(*ac)->ac_max_block = max_block;
1160 
1161 	status = -ENOSPC;
1162 	if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1163 	    ocfs2_alloc_should_use_local(osb, bits_wanted)) {
1164 		status = ocfs2_reserve_local_alloc_bits(osb,
1165 							bits_wanted,
1166 							*ac);
1167 		if ((status < 0) && (status != -ENOSPC)) {
1168 			mlog_errno(status);
1169 			goto bail;
1170 		}
1171 	}
1172 
1173 	if (status == -ENOSPC) {
1174 retry:
1175 		status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1176 		/* Retry if there is sufficient space cached in truncate log */
1177 		if (status == -ENOSPC && !retried) {
1178 			retried = 1;
1179 			ocfs2_inode_unlock((*ac)->ac_inode, 1);
1180 			inode_unlock((*ac)->ac_inode);
1181 
1182 			ret = ocfs2_try_to_free_truncate_log(osb, bits_wanted);
1183 			if (ret == 1) {
1184 				iput((*ac)->ac_inode);
1185 				(*ac)->ac_inode = NULL;
1186 				goto retry;
1187 			}
1188 
1189 			if (ret < 0)
1190 				mlog_errno(ret);
1191 
1192 			inode_lock((*ac)->ac_inode);
1193 			ret = ocfs2_inode_lock((*ac)->ac_inode, NULL, 1);
1194 			if (ret < 0) {
1195 				mlog_errno(ret);
1196 				inode_unlock((*ac)->ac_inode);
1197 				iput((*ac)->ac_inode);
1198 				(*ac)->ac_inode = NULL;
1199 				goto bail;
1200 			}
1201 		}
1202 		if (status < 0) {
1203 			if (status != -ENOSPC)
1204 				mlog_errno(status);
1205 			goto bail;
1206 		}
1207 	}
1208 
1209 	status = 0;
1210 bail:
1211 	if ((status < 0) && *ac) {
1212 		ocfs2_free_alloc_context(*ac);
1213 		*ac = NULL;
1214 	}
1215 
1216 	if (status)
1217 		mlog_errno(status);
1218 	return status;
1219 }
1220 
ocfs2_reserve_clusters(struct ocfs2_super * osb,u32 bits_wanted,struct ocfs2_alloc_context ** ac)1221 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1222 			   u32 bits_wanted,
1223 			   struct ocfs2_alloc_context **ac)
1224 {
1225 	return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1226 						 ALLOC_NEW_GROUP, ac);
1227 }
1228 
1229 /*
1230  * More or less lifted from ext3. I'll leave their description below:
1231  *
1232  * "For ext3 allocations, we must not reuse any blocks which are
1233  * allocated in the bitmap buffer's "last committed data" copy.  This
1234  * prevents deletes from freeing up the page for reuse until we have
1235  * committed the delete transaction.
1236  *
1237  * If we didn't do this, then deleting something and reallocating it as
1238  * data would allow the old block to be overwritten before the
1239  * transaction committed (because we force data to disk before commit).
1240  * This would lead to corruption if we crashed between overwriting the
1241  * data and committing the delete.
1242  *
1243  * @@@ We may want to make this allocation behaviour conditional on
1244  * data-writes at some point, and disable it for metadata allocations or
1245  * sync-data inodes."
1246  *
1247  * Note: OCFS2 already does this differently for metadata vs data
1248  * allocations, as those bitmaps are separate and undo access is never
1249  * called on a metadata group descriptor.
1250  */
ocfs2_test_bg_bit_allocatable(struct buffer_head * bg_bh,int nr)1251 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1252 					 int nr)
1253 {
1254 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1255 	struct journal_head *jh;
1256 	int ret;
1257 
1258 	if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1259 		return 0;
1260 
1261 	jh = jbd2_journal_grab_journal_head(bg_bh);
1262 	if (!jh)
1263 		return 1;
1264 
1265 	spin_lock(&jh->b_state_lock);
1266 	bg = (struct ocfs2_group_desc *) jh->b_committed_data;
1267 	if (bg)
1268 		ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1269 	else
1270 		ret = 1;
1271 	spin_unlock(&jh->b_state_lock);
1272 	jbd2_journal_put_journal_head(jh);
1273 
1274 	return ret;
1275 }
1276 
ocfs2_block_group_find_clear_bits(struct ocfs2_super * osb,struct buffer_head * bg_bh,unsigned int bits_wanted,unsigned int total_bits,struct ocfs2_suballoc_result * res)1277 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1278 					     struct buffer_head *bg_bh,
1279 					     unsigned int bits_wanted,
1280 					     unsigned int total_bits,
1281 					     struct ocfs2_suballoc_result *res)
1282 {
1283 	void *bitmap;
1284 	u16 best_offset, best_size;
1285 	int offset, start, found, status = 0;
1286 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1287 
1288 	/* Callers got this descriptor from
1289 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1290 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1291 
1292 	found = start = best_offset = best_size = 0;
1293 	bitmap = bg->bg_bitmap;
1294 
1295 	while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1296 		if (offset == total_bits)
1297 			break;
1298 
1299 		if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1300 			/* We found a zero, but we can't use it as it
1301 			 * hasn't been put to disk yet! */
1302 			found = 0;
1303 			start = offset + 1;
1304 		} else if (offset == start) {
1305 			/* we found a zero */
1306 			found++;
1307 			/* move start to the next bit to test */
1308 			start++;
1309 		} else {
1310 			/* got a zero after some ones */
1311 			found = 1;
1312 			start = offset + 1;
1313 		}
1314 		if (found > best_size) {
1315 			best_size = found;
1316 			best_offset = start - found;
1317 		}
1318 		/* we got everything we needed */
1319 		if (found == bits_wanted) {
1320 			/* mlog(0, "Found it all!\n"); */
1321 			break;
1322 		}
1323 	}
1324 
1325 	if (best_size) {
1326 		res->sr_bit_offset = best_offset;
1327 		res->sr_bits = best_size;
1328 	} else {
1329 		status = -ENOSPC;
1330 		/* No error log here -- see the comment above
1331 		 * ocfs2_test_bg_bit_allocatable */
1332 	}
1333 
1334 	return status;
1335 }
1336 
ocfs2_block_group_set_bits(handle_t * handle,struct inode * alloc_inode,struct ocfs2_group_desc * bg,struct buffer_head * group_bh,unsigned int bit_off,unsigned int num_bits)1337 int ocfs2_block_group_set_bits(handle_t *handle,
1338 					     struct inode *alloc_inode,
1339 					     struct ocfs2_group_desc *bg,
1340 					     struct buffer_head *group_bh,
1341 					     unsigned int bit_off,
1342 					     unsigned int num_bits)
1343 {
1344 	int status;
1345 	void *bitmap = bg->bg_bitmap;
1346 	int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1347 
1348 	/* All callers get the descriptor via
1349 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1350 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1351 	BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1352 
1353 	trace_ocfs2_block_group_set_bits(bit_off, num_bits);
1354 
1355 	if (ocfs2_is_cluster_bitmap(alloc_inode))
1356 		journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1357 
1358 	status = ocfs2_journal_access_gd(handle,
1359 					 INODE_CACHE(alloc_inode),
1360 					 group_bh,
1361 					 journal_type);
1362 	if (status < 0) {
1363 		mlog_errno(status);
1364 		goto bail;
1365 	}
1366 
1367 	le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1368 	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1369 		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
1370 				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
1371 				   le16_to_cpu(bg->bg_bits),
1372 				   le16_to_cpu(bg->bg_free_bits_count),
1373 				   num_bits);
1374 	}
1375 	while(num_bits--)
1376 		ocfs2_set_bit(bit_off++, bitmap);
1377 
1378 	ocfs2_journal_dirty(handle, group_bh);
1379 
1380 bail:
1381 	return status;
1382 }
1383 
1384 /* find the one with the most empty bits */
ocfs2_find_victim_chain(struct ocfs2_chain_list * cl)1385 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1386 {
1387 	u16 curr, best;
1388 
1389 	BUG_ON(!cl->cl_next_free_rec);
1390 
1391 	best = curr = 0;
1392 	while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1393 		if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1394 		    le32_to_cpu(cl->cl_recs[best].c_free))
1395 			best = curr;
1396 		curr++;
1397 	}
1398 
1399 	BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1400 	return best;
1401 }
1402 
ocfs2_relink_block_group(handle_t * handle,struct inode * alloc_inode,struct buffer_head * fe_bh,struct buffer_head * bg_bh,struct buffer_head * prev_bg_bh,u16 chain)1403 static int ocfs2_relink_block_group(handle_t *handle,
1404 				    struct inode *alloc_inode,
1405 				    struct buffer_head *fe_bh,
1406 				    struct buffer_head *bg_bh,
1407 				    struct buffer_head *prev_bg_bh,
1408 				    u16 chain)
1409 {
1410 	int status;
1411 	/* there is a really tiny chance the journal calls could fail,
1412 	 * but we wouldn't want inconsistent blocks in *any* case. */
1413 	u64 bg_ptr, prev_bg_ptr;
1414 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1415 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1416 	struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1417 
1418 	/* The caller got these descriptors from
1419 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1420 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1421 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1422 
1423 	trace_ocfs2_relink_block_group(
1424 		(unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1425 		(unsigned long long)le64_to_cpu(bg->bg_blkno),
1426 		(unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1427 
1428 	bg_ptr = le64_to_cpu(bg->bg_next_group);
1429 	prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1430 
1431 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1432 					 prev_bg_bh,
1433 					 OCFS2_JOURNAL_ACCESS_WRITE);
1434 	if (status < 0)
1435 		goto out;
1436 
1437 	prev_bg->bg_next_group = bg->bg_next_group;
1438 	ocfs2_journal_dirty(handle, prev_bg_bh);
1439 
1440 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1441 					 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1442 	if (status < 0)
1443 		goto out_rollback_prev_bg;
1444 
1445 	bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1446 	ocfs2_journal_dirty(handle, bg_bh);
1447 
1448 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1449 					 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1450 	if (status < 0)
1451 		goto out_rollback_bg;
1452 
1453 	fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1454 	ocfs2_journal_dirty(handle, fe_bh);
1455 
1456 out:
1457 	if (status < 0)
1458 		mlog_errno(status);
1459 	return status;
1460 
1461 out_rollback_bg:
1462 	bg->bg_next_group = cpu_to_le64(bg_ptr);
1463 out_rollback_prev_bg:
1464 	prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1465 	goto out;
1466 }
1467 
ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc * bg,u32 wanted)1468 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1469 						     u32 wanted)
1470 {
1471 	return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1472 }
1473 
1474 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1475  * value on error. */
ocfs2_cluster_group_search(struct inode * inode,struct buffer_head * group_bh,u32 bits_wanted,u32 min_bits,u64 max_block,struct ocfs2_suballoc_result * res)1476 static int ocfs2_cluster_group_search(struct inode *inode,
1477 				      struct buffer_head *group_bh,
1478 				      u32 bits_wanted, u32 min_bits,
1479 				      u64 max_block,
1480 				      struct ocfs2_suballoc_result *res)
1481 {
1482 	int search = -ENOSPC;
1483 	int ret;
1484 	u64 blkoff;
1485 	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1486 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1487 	unsigned int max_bits, gd_cluster_off;
1488 
1489 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1490 
1491 	if (gd->bg_free_bits_count) {
1492 		max_bits = le16_to_cpu(gd->bg_bits);
1493 
1494 		/* Tail groups in cluster bitmaps which aren't cpg
1495 		 * aligned are prone to partial extension by a failed
1496 		 * fs resize. If the file system resize never got to
1497 		 * update the dinode cluster count, then we don't want
1498 		 * to trust any clusters past it, regardless of what
1499 		 * the group descriptor says. */
1500 		gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1501 							  le64_to_cpu(gd->bg_blkno));
1502 		if ((gd_cluster_off + max_bits) >
1503 		    OCFS2_I(inode)->ip_clusters) {
1504 			max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1505 			trace_ocfs2_cluster_group_search_wrong_max_bits(
1506 				(unsigned long long)le64_to_cpu(gd->bg_blkno),
1507 				le16_to_cpu(gd->bg_bits),
1508 				OCFS2_I(inode)->ip_clusters, max_bits);
1509 		}
1510 
1511 		ret = ocfs2_block_group_find_clear_bits(osb,
1512 							group_bh, bits_wanted,
1513 							max_bits, res);
1514 		if (ret)
1515 			return ret;
1516 
1517 		if (max_block) {
1518 			blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1519 							  gd_cluster_off +
1520 							  res->sr_bit_offset +
1521 							  res->sr_bits);
1522 			trace_ocfs2_cluster_group_search_max_block(
1523 				(unsigned long long)blkoff,
1524 				(unsigned long long)max_block);
1525 			if (blkoff > max_block)
1526 				return -ENOSPC;
1527 		}
1528 
1529 		/* ocfs2_block_group_find_clear_bits() might
1530 		 * return success, but we still want to return
1531 		 * -ENOSPC unless it found the minimum number
1532 		 * of bits. */
1533 		if (min_bits <= res->sr_bits)
1534 			search = 0; /* success */
1535 		else if (res->sr_bits) {
1536 			/*
1537 			 * Don't show bits which we'll be returning
1538 			 * for allocation to the local alloc bitmap.
1539 			 */
1540 			ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
1541 		}
1542 	}
1543 
1544 	return search;
1545 }
1546 
ocfs2_block_group_search(struct inode * inode,struct buffer_head * group_bh,u32 bits_wanted,u32 min_bits,u64 max_block,struct ocfs2_suballoc_result * res)1547 static int ocfs2_block_group_search(struct inode *inode,
1548 				    struct buffer_head *group_bh,
1549 				    u32 bits_wanted, u32 min_bits,
1550 				    u64 max_block,
1551 				    struct ocfs2_suballoc_result *res)
1552 {
1553 	int ret = -ENOSPC;
1554 	u64 blkoff;
1555 	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1556 
1557 	BUG_ON(min_bits != 1);
1558 	BUG_ON(ocfs2_is_cluster_bitmap(inode));
1559 
1560 	if (bg->bg_free_bits_count) {
1561 		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1562 							group_bh, bits_wanted,
1563 							le16_to_cpu(bg->bg_bits),
1564 							res);
1565 		if (!ret && max_block) {
1566 			blkoff = le64_to_cpu(bg->bg_blkno) +
1567 				res->sr_bit_offset + res->sr_bits;
1568 			trace_ocfs2_block_group_search_max_block(
1569 				(unsigned long long)blkoff,
1570 				(unsigned long long)max_block);
1571 			if (blkoff > max_block)
1572 				ret = -ENOSPC;
1573 		}
1574 	}
1575 
1576 	return ret;
1577 }
1578 
ocfs2_alloc_dinode_update_counts(struct inode * inode,handle_t * handle,struct buffer_head * di_bh,u32 num_bits,u16 chain)1579 int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1580 				       handle_t *handle,
1581 				       struct buffer_head *di_bh,
1582 				       u32 num_bits,
1583 				       u16 chain)
1584 {
1585 	int ret;
1586 	u32 tmp_used;
1587 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1588 	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1589 
1590 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1591 				      OCFS2_JOURNAL_ACCESS_WRITE);
1592 	if (ret < 0) {
1593 		mlog_errno(ret);
1594 		goto out;
1595 	}
1596 
1597 	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1598 	di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1599 	le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1600 	ocfs2_journal_dirty(handle, di_bh);
1601 
1602 out:
1603 	return ret;
1604 }
1605 
ocfs2_rollback_alloc_dinode_counts(struct inode * inode,struct buffer_head * di_bh,u32 num_bits,u16 chain)1606 void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
1607 				       struct buffer_head *di_bh,
1608 				       u32 num_bits,
1609 				       u16 chain)
1610 {
1611 	u32 tmp_used;
1612 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1613 	struct ocfs2_chain_list *cl;
1614 
1615 	cl = (struct ocfs2_chain_list *)&di->id2.i_chain;
1616 	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1617 	di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits);
1618 	le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits);
1619 }
1620 
ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result * res,struct ocfs2_extent_rec * rec,struct ocfs2_chain_list * cl)1621 static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1622 					 struct ocfs2_extent_rec *rec,
1623 					 struct ocfs2_chain_list *cl)
1624 {
1625 	unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1626 	unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1627 	unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1628 
1629 	if (res->sr_bit_offset < bitoff)
1630 		return 0;
1631 	if (res->sr_bit_offset >= (bitoff + bitcount))
1632 		return 0;
1633 	res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1634 		(res->sr_bit_offset - bitoff);
1635 	if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1636 		res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1637 	return 1;
1638 }
1639 
ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context * ac,struct ocfs2_group_desc * bg,struct ocfs2_suballoc_result * res)1640 static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1641 					  struct ocfs2_group_desc *bg,
1642 					  struct ocfs2_suballoc_result *res)
1643 {
1644 	int i;
1645 	u64 bg_blkno = res->sr_bg_blkno;  /* Save off */
1646 	struct ocfs2_extent_rec *rec;
1647 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1648 	struct ocfs2_chain_list *cl = &di->id2.i_chain;
1649 
1650 	if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1651 		res->sr_blkno = 0;
1652 		return;
1653 	}
1654 
1655 	res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1656 	res->sr_bg_blkno = 0;  /* Clear it for contig block groups */
1657 	if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1658 	    !bg->bg_list.l_next_free_rec)
1659 		return;
1660 
1661 	for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1662 		rec = &bg->bg_list.l_recs[i];
1663 		if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1664 			res->sr_bg_blkno = bg_blkno;  /* Restore */
1665 			break;
1666 		}
1667 	}
1668 }
1669 
ocfs2_search_one_group(struct ocfs2_alloc_context * ac,handle_t * handle,u32 bits_wanted,u32 min_bits,struct ocfs2_suballoc_result * res,u16 * bits_left)1670 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1671 				  handle_t *handle,
1672 				  u32 bits_wanted,
1673 				  u32 min_bits,
1674 				  struct ocfs2_suballoc_result *res,
1675 				  u16 *bits_left)
1676 {
1677 	int ret;
1678 	struct buffer_head *group_bh = NULL;
1679 	struct ocfs2_group_desc *gd;
1680 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1681 	struct inode *alloc_inode = ac->ac_inode;
1682 
1683 	ret = ocfs2_read_group_descriptor(alloc_inode, di,
1684 					  res->sr_bg_blkno, &group_bh);
1685 	if (ret < 0) {
1686 		mlog_errno(ret);
1687 		return ret;
1688 	}
1689 
1690 	gd = (struct ocfs2_group_desc *) group_bh->b_data;
1691 	ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1692 				  ac->ac_max_block, res);
1693 	if (ret < 0) {
1694 		if (ret != -ENOSPC)
1695 			mlog_errno(ret);
1696 		goto out;
1697 	}
1698 
1699 	if (!ret)
1700 		ocfs2_bg_discontig_fix_result(ac, gd, res);
1701 
1702 	/*
1703 	 * sr_bg_blkno might have been changed by
1704 	 * ocfs2_bg_discontig_fix_result
1705 	 */
1706 	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1707 
1708 	if (ac->ac_find_loc_only)
1709 		goto out_loc_only;
1710 
1711 	ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1712 					       res->sr_bits,
1713 					       le16_to_cpu(gd->bg_chain));
1714 	if (ret < 0) {
1715 		mlog_errno(ret);
1716 		goto out;
1717 	}
1718 
1719 	ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1720 					 res->sr_bit_offset, res->sr_bits);
1721 	if (ret < 0) {
1722 		ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh,
1723 					       res->sr_bits,
1724 					       le16_to_cpu(gd->bg_chain));
1725 		mlog_errno(ret);
1726 	}
1727 
1728 out_loc_only:
1729 	*bits_left = le16_to_cpu(gd->bg_free_bits_count);
1730 
1731 out:
1732 	brelse(group_bh);
1733 
1734 	return ret;
1735 }
1736 
ocfs2_search_chain(struct ocfs2_alloc_context * ac,handle_t * handle,u32 bits_wanted,u32 min_bits,struct ocfs2_suballoc_result * res,u16 * bits_left)1737 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1738 			      handle_t *handle,
1739 			      u32 bits_wanted,
1740 			      u32 min_bits,
1741 			      struct ocfs2_suballoc_result *res,
1742 			      u16 *bits_left)
1743 {
1744 	int status;
1745 	u16 chain;
1746 	u64 next_group;
1747 	struct inode *alloc_inode = ac->ac_inode;
1748 	struct buffer_head *group_bh = NULL;
1749 	struct buffer_head *prev_group_bh = NULL;
1750 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1751 	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1752 	struct ocfs2_group_desc *bg;
1753 
1754 	chain = ac->ac_chain;
1755 	trace_ocfs2_search_chain_begin(
1756 		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1757 		bits_wanted, chain);
1758 
1759 	status = ocfs2_read_group_descriptor(alloc_inode, fe,
1760 					     le64_to_cpu(cl->cl_recs[chain].c_blkno),
1761 					     &group_bh);
1762 	if (status < 0) {
1763 		mlog_errno(status);
1764 		goto bail;
1765 	}
1766 	bg = (struct ocfs2_group_desc *) group_bh->b_data;
1767 
1768 	status = -ENOSPC;
1769 	/* for now, the chain search is a bit simplistic. We just use
1770 	 * the 1st group with any empty bits. */
1771 	while ((status = ac->ac_group_search(alloc_inode, group_bh,
1772 					     bits_wanted, min_bits,
1773 					     ac->ac_max_block,
1774 					     res)) == -ENOSPC) {
1775 		if (!bg->bg_next_group)
1776 			break;
1777 
1778 		brelse(prev_group_bh);
1779 		prev_group_bh = NULL;
1780 
1781 		next_group = le64_to_cpu(bg->bg_next_group);
1782 		prev_group_bh = group_bh;
1783 		group_bh = NULL;
1784 		status = ocfs2_read_group_descriptor(alloc_inode, fe,
1785 						     next_group, &group_bh);
1786 		if (status < 0) {
1787 			mlog_errno(status);
1788 			goto bail;
1789 		}
1790 		bg = (struct ocfs2_group_desc *) group_bh->b_data;
1791 	}
1792 	if (status < 0) {
1793 		if (status != -ENOSPC)
1794 			mlog_errno(status);
1795 		goto bail;
1796 	}
1797 
1798 	trace_ocfs2_search_chain_succ(
1799 		(unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1800 
1801 	res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1802 
1803 	BUG_ON(res->sr_bits == 0);
1804 	if (!status)
1805 		ocfs2_bg_discontig_fix_result(ac, bg, res);
1806 
1807 	/*
1808 	 * sr_bg_blkno might have been changed by
1809 	 * ocfs2_bg_discontig_fix_result
1810 	 */
1811 	res->sr_bg_stable_blkno = group_bh->b_blocknr;
1812 
1813 	/*
1814 	 * Keep track of previous block descriptor read. When
1815 	 * we find a target, if we have read more than X
1816 	 * number of descriptors, and the target is reasonably
1817 	 * empty, relink him to top of his chain.
1818 	 *
1819 	 * We've read 0 extra blocks and only send one more to
1820 	 * the transaction, yet the next guy to search has a
1821 	 * much easier time.
1822 	 *
1823 	 * Do this *after* figuring out how many bits we're taking out
1824 	 * of our target group.
1825 	 */
1826 	if (!ac->ac_disable_chain_relink &&
1827 	    (prev_group_bh) &&
1828 	    (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1829 		status = ocfs2_relink_block_group(handle, alloc_inode,
1830 						  ac->ac_bh, group_bh,
1831 						  prev_group_bh, chain);
1832 		if (status < 0) {
1833 			mlog_errno(status);
1834 			goto bail;
1835 		}
1836 	}
1837 
1838 	if (ac->ac_find_loc_only)
1839 		goto out_loc_only;
1840 
1841 	status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1842 						  ac->ac_bh, res->sr_bits,
1843 						  chain);
1844 	if (status) {
1845 		mlog_errno(status);
1846 		goto bail;
1847 	}
1848 
1849 	status = ocfs2_block_group_set_bits(handle,
1850 					    alloc_inode,
1851 					    bg,
1852 					    group_bh,
1853 					    res->sr_bit_offset,
1854 					    res->sr_bits);
1855 	if (status < 0) {
1856 		ocfs2_rollback_alloc_dinode_counts(alloc_inode,
1857 					ac->ac_bh, res->sr_bits, chain);
1858 		mlog_errno(status);
1859 		goto bail;
1860 	}
1861 
1862 	trace_ocfs2_search_chain_end(
1863 			(unsigned long long)le64_to_cpu(fe->i_blkno),
1864 			res->sr_bits);
1865 
1866 out_loc_only:
1867 	*bits_left = le16_to_cpu(bg->bg_free_bits_count);
1868 bail:
1869 	brelse(group_bh);
1870 	brelse(prev_group_bh);
1871 
1872 	if (status)
1873 		mlog_errno(status);
1874 	return status;
1875 }
1876 
1877 /* will give out up to bits_wanted contiguous bits. */
ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context * ac,handle_t * handle,u32 bits_wanted,u32 min_bits,struct ocfs2_suballoc_result * res)1878 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1879 				     handle_t *handle,
1880 				     u32 bits_wanted,
1881 				     u32 min_bits,
1882 				     struct ocfs2_suballoc_result *res)
1883 {
1884 	int status;
1885 	u16 victim, i;
1886 	u16 bits_left = 0;
1887 	u64 hint = ac->ac_last_group;
1888 	struct ocfs2_chain_list *cl;
1889 	struct ocfs2_dinode *fe;
1890 
1891 	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1892 	BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1893 	BUG_ON(!ac->ac_bh);
1894 
1895 	fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1896 
1897 	/* The bh was validated by the inode read during
1898 	 * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1899 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1900 
1901 	if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1902 	    le32_to_cpu(fe->id1.bitmap1.i_total)) {
1903 		status = ocfs2_error(ac->ac_inode->i_sb,
1904 				     "Chain allocator dinode %llu has %u used bits but only %u total\n",
1905 				     (unsigned long long)le64_to_cpu(fe->i_blkno),
1906 				     le32_to_cpu(fe->id1.bitmap1.i_used),
1907 				     le32_to_cpu(fe->id1.bitmap1.i_total));
1908 		goto bail;
1909 	}
1910 
1911 	res->sr_bg_blkno = hint;
1912 	if (res->sr_bg_blkno) {
1913 		/* Attempt to short-circuit the usual search mechanism
1914 		 * by jumping straight to the most recently used
1915 		 * allocation group. This helps us maintain some
1916 		 * contiguousness across allocations. */
1917 		status = ocfs2_search_one_group(ac, handle, bits_wanted,
1918 						min_bits, res, &bits_left);
1919 		if (!status)
1920 			goto set_hint;
1921 		if (status < 0 && status != -ENOSPC) {
1922 			mlog_errno(status);
1923 			goto bail;
1924 		}
1925 	}
1926 
1927 	cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1928 
1929 	victim = ocfs2_find_victim_chain(cl);
1930 	ac->ac_chain = victim;
1931 
1932 	status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1933 				    res, &bits_left);
1934 	if (!status) {
1935 		if (ocfs2_is_cluster_bitmap(ac->ac_inode))
1936 			hint = res->sr_bg_blkno;
1937 		else
1938 			hint = ocfs2_group_from_res(res);
1939 		goto set_hint;
1940 	}
1941 	if (status < 0 && status != -ENOSPC) {
1942 		mlog_errno(status);
1943 		goto bail;
1944 	}
1945 
1946 	trace_ocfs2_claim_suballoc_bits(victim);
1947 
1948 	/* If we didn't pick a good victim, then just default to
1949 	 * searching each chain in order. Don't allow chain relinking
1950 	 * because we only calculate enough journal credits for one
1951 	 * relink per alloc. */
1952 	ac->ac_disable_chain_relink = 1;
1953 	for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1954 		if (i == victim)
1955 			continue;
1956 		if (!cl->cl_recs[i].c_free)
1957 			continue;
1958 
1959 		ac->ac_chain = i;
1960 		status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1961 					    res, &bits_left);
1962 		if (!status) {
1963 			hint = ocfs2_group_from_res(res);
1964 			break;
1965 		}
1966 		if (status < 0 && status != -ENOSPC) {
1967 			mlog_errno(status);
1968 			goto bail;
1969 		}
1970 	}
1971 
1972 set_hint:
1973 	if (status != -ENOSPC) {
1974 		/* If the next search of this group is not likely to
1975 		 * yield a suitable extent, then we reset the last
1976 		 * group hint so as to not waste a disk read */
1977 		if (bits_left < min_bits)
1978 			ac->ac_last_group = 0;
1979 		else
1980 			ac->ac_last_group = hint;
1981 	}
1982 
1983 bail:
1984 	if (status)
1985 		mlog_errno(status);
1986 	return status;
1987 }
1988 
ocfs2_claim_metadata(handle_t * handle,struct ocfs2_alloc_context * ac,u32 bits_wanted,u64 * suballoc_loc,u16 * suballoc_bit_start,unsigned int * num_bits,u64 * blkno_start)1989 int ocfs2_claim_metadata(handle_t *handle,
1990 			 struct ocfs2_alloc_context *ac,
1991 			 u32 bits_wanted,
1992 			 u64 *suballoc_loc,
1993 			 u16 *suballoc_bit_start,
1994 			 unsigned int *num_bits,
1995 			 u64 *blkno_start)
1996 {
1997 	int status;
1998 	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
1999 
2000 	BUG_ON(!ac);
2001 	BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
2002 	BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
2003 
2004 	status = ocfs2_claim_suballoc_bits(ac,
2005 					   handle,
2006 					   bits_wanted,
2007 					   1,
2008 					   &res);
2009 	if (status < 0) {
2010 		mlog_errno(status);
2011 		goto bail;
2012 	}
2013 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2014 
2015 	*suballoc_loc = res.sr_bg_blkno;
2016 	*suballoc_bit_start = res.sr_bit_offset;
2017 	*blkno_start = res.sr_blkno;
2018 	ac->ac_bits_given += res.sr_bits;
2019 	*num_bits = res.sr_bits;
2020 	status = 0;
2021 bail:
2022 	if (status)
2023 		mlog_errno(status);
2024 	return status;
2025 }
2026 
ocfs2_init_inode_ac_group(struct inode * dir,struct buffer_head * parent_di_bh,struct ocfs2_alloc_context * ac)2027 static void ocfs2_init_inode_ac_group(struct inode *dir,
2028 				      struct buffer_head *parent_di_bh,
2029 				      struct ocfs2_alloc_context *ac)
2030 {
2031 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2032 	/*
2033 	 * Try to allocate inodes from some specific group.
2034 	 *
2035 	 * If the parent dir has recorded the last group used in allocation,
2036 	 * cool, use it. Otherwise if we try to allocate new inode from the
2037 	 * same slot the parent dir belongs to, use the same chunk.
2038 	 *
2039 	 * We are very careful here to avoid the mistake of setting
2040 	 * ac_last_group to a group descriptor from a different (unlocked) slot.
2041 	 */
2042 	if (OCFS2_I(dir)->ip_last_used_group &&
2043 	    OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2044 		ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2045 	else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2046 		if (di->i_suballoc_loc)
2047 			ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2048 		else
2049 			ac->ac_last_group = ocfs2_which_suballoc_group(
2050 					le64_to_cpu(di->i_blkno),
2051 					le16_to_cpu(di->i_suballoc_bit));
2052 	}
2053 }
2054 
ocfs2_save_inode_ac_group(struct inode * dir,struct ocfs2_alloc_context * ac)2055 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2056 					     struct ocfs2_alloc_context *ac)
2057 {
2058 	OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2059 	OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2060 }
2061 
ocfs2_find_new_inode_loc(struct inode * dir,struct buffer_head * parent_fe_bh,struct ocfs2_alloc_context * ac,u64 * fe_blkno)2062 int ocfs2_find_new_inode_loc(struct inode *dir,
2063 			     struct buffer_head *parent_fe_bh,
2064 			     struct ocfs2_alloc_context *ac,
2065 			     u64 *fe_blkno)
2066 {
2067 	int ret;
2068 	handle_t *handle = NULL;
2069 	struct ocfs2_suballoc_result *res;
2070 
2071 	BUG_ON(!ac);
2072 	BUG_ON(ac->ac_bits_given != 0);
2073 	BUG_ON(ac->ac_bits_wanted != 1);
2074 	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2075 
2076 	res = kzalloc(sizeof(*res), GFP_NOFS);
2077 	if (res == NULL) {
2078 		ret = -ENOMEM;
2079 		mlog_errno(ret);
2080 		goto out;
2081 	}
2082 
2083 	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2084 
2085 	/*
2086 	 * The handle started here is for chain relink. Alternatively,
2087 	 * we could just disable relink for these calls.
2088 	 */
2089 	handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2090 	if (IS_ERR(handle)) {
2091 		ret = PTR_ERR(handle);
2092 		handle = NULL;
2093 		mlog_errno(ret);
2094 		goto out;
2095 	}
2096 
2097 	/*
2098 	 * This will instruct ocfs2_claim_suballoc_bits and
2099 	 * ocfs2_search_one_group to search but save actual allocation
2100 	 * for later.
2101 	 */
2102 	ac->ac_find_loc_only = 1;
2103 
2104 	ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2105 	if (ret < 0) {
2106 		mlog_errno(ret);
2107 		goto out;
2108 	}
2109 
2110 	ac->ac_find_loc_priv = res;
2111 	*fe_blkno = res->sr_blkno;
2112 	ocfs2_update_inode_fsync_trans(handle, dir, 0);
2113 out:
2114 	if (handle)
2115 		ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2116 
2117 	if (ret)
2118 		kfree(res);
2119 
2120 	return ret;
2121 }
2122 
ocfs2_claim_new_inode_at_loc(handle_t * handle,struct inode * dir,struct ocfs2_alloc_context * ac,u64 * suballoc_loc,u16 * suballoc_bit,u64 di_blkno)2123 int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2124 				 struct inode *dir,
2125 				 struct ocfs2_alloc_context *ac,
2126 				 u64 *suballoc_loc,
2127 				 u16 *suballoc_bit,
2128 				 u64 di_blkno)
2129 {
2130 	int ret;
2131 	u16 chain;
2132 	struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2133 	struct buffer_head *bg_bh = NULL;
2134 	struct ocfs2_group_desc *bg;
2135 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2136 
2137 	/*
2138 	 * Since di_blkno is being passed back in, we check for any
2139 	 * inconsistencies which may have happened between
2140 	 * calls. These are code bugs as di_blkno is not expected to
2141 	 * change once returned from ocfs2_find_new_inode_loc()
2142 	 */
2143 	BUG_ON(res->sr_blkno != di_blkno);
2144 
2145 	ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2146 					  res->sr_bg_stable_blkno, &bg_bh);
2147 	if (ret) {
2148 		mlog_errno(ret);
2149 		goto out;
2150 	}
2151 
2152 	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2153 	chain = le16_to_cpu(bg->bg_chain);
2154 
2155 	ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2156 					       ac->ac_bh, res->sr_bits,
2157 					       chain);
2158 	if (ret) {
2159 		mlog_errno(ret);
2160 		goto out;
2161 	}
2162 
2163 	ret = ocfs2_block_group_set_bits(handle,
2164 					 ac->ac_inode,
2165 					 bg,
2166 					 bg_bh,
2167 					 res->sr_bit_offset,
2168 					 res->sr_bits);
2169 	if (ret < 0) {
2170 		ocfs2_rollback_alloc_dinode_counts(ac->ac_inode,
2171 					       ac->ac_bh, res->sr_bits, chain);
2172 		mlog_errno(ret);
2173 		goto out;
2174 	}
2175 
2176 	trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2177 					   res->sr_bits);
2178 
2179 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2180 
2181 	BUG_ON(res->sr_bits != 1);
2182 
2183 	*suballoc_loc = res->sr_bg_blkno;
2184 	*suballoc_bit = res->sr_bit_offset;
2185 	ac->ac_bits_given++;
2186 	ocfs2_save_inode_ac_group(dir, ac);
2187 
2188 out:
2189 	brelse(bg_bh);
2190 
2191 	return ret;
2192 }
2193 
ocfs2_claim_new_inode(handle_t * handle,struct inode * dir,struct buffer_head * parent_fe_bh,struct ocfs2_alloc_context * ac,u64 * suballoc_loc,u16 * suballoc_bit,u64 * fe_blkno)2194 int ocfs2_claim_new_inode(handle_t *handle,
2195 			  struct inode *dir,
2196 			  struct buffer_head *parent_fe_bh,
2197 			  struct ocfs2_alloc_context *ac,
2198 			  u64 *suballoc_loc,
2199 			  u16 *suballoc_bit,
2200 			  u64 *fe_blkno)
2201 {
2202 	int status;
2203 	struct ocfs2_suballoc_result res;
2204 
2205 	BUG_ON(!ac);
2206 	BUG_ON(ac->ac_bits_given != 0);
2207 	BUG_ON(ac->ac_bits_wanted != 1);
2208 	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2209 
2210 	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2211 
2212 	status = ocfs2_claim_suballoc_bits(ac,
2213 					   handle,
2214 					   1,
2215 					   1,
2216 					   &res);
2217 	if (status < 0) {
2218 		mlog_errno(status);
2219 		goto bail;
2220 	}
2221 	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2222 
2223 	BUG_ON(res.sr_bits != 1);
2224 
2225 	*suballoc_loc = res.sr_bg_blkno;
2226 	*suballoc_bit = res.sr_bit_offset;
2227 	*fe_blkno = res.sr_blkno;
2228 	ac->ac_bits_given++;
2229 	ocfs2_save_inode_ac_group(dir, ac);
2230 	status = 0;
2231 bail:
2232 	if (status)
2233 		mlog_errno(status);
2234 	return status;
2235 }
2236 
2237 /* translate a group desc. blkno and it's bitmap offset into
2238  * disk cluster offset. */
ocfs2_desc_bitmap_to_cluster_off(struct inode * inode,u64 bg_blkno,u16 bg_bit_off)2239 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2240 						   u64 bg_blkno,
2241 						   u16 bg_bit_off)
2242 {
2243 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2244 	u32 cluster = 0;
2245 
2246 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2247 
2248 	if (bg_blkno != osb->first_cluster_group_blkno)
2249 		cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2250 	cluster += (u32) bg_bit_off;
2251 	return cluster;
2252 }
2253 
2254 /* given a cluster offset, calculate which block group it belongs to
2255  * and return that block offset. */
ocfs2_which_cluster_group(struct inode * inode,u32 cluster)2256 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2257 {
2258 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2259 	u32 group_no;
2260 
2261 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2262 
2263 	group_no = cluster / osb->bitmap_cpg;
2264 	if (!group_no)
2265 		return osb->first_cluster_group_blkno;
2266 	return ocfs2_clusters_to_blocks(inode->i_sb,
2267 					group_no * osb->bitmap_cpg);
2268 }
2269 
2270 /* given the block number of a cluster start, calculate which cluster
2271  * group and descriptor bitmap offset that corresponds to. */
ocfs2_block_to_cluster_group(struct inode * inode,u64 data_blkno,u64 * bg_blkno,u16 * bg_bit_off)2272 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2273 						u64 data_blkno,
2274 						u64 *bg_blkno,
2275 						u16 *bg_bit_off)
2276 {
2277 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2278 	u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2279 
2280 	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2281 
2282 	*bg_blkno = ocfs2_which_cluster_group(inode,
2283 					      data_cluster);
2284 
2285 	if (*bg_blkno == osb->first_cluster_group_blkno)
2286 		*bg_bit_off = (u16) data_cluster;
2287 	else
2288 		*bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2289 							     data_blkno - *bg_blkno);
2290 }
2291 
2292 /*
2293  * min_bits - minimum contiguous chunk from this total allocation we
2294  * can handle. set to what we asked for originally for a full
2295  * contig. allocation, set to '1' to indicate we can deal with extents
2296  * of any size.
2297  */
__ocfs2_claim_clusters(handle_t * handle,struct ocfs2_alloc_context * ac,u32 min_clusters,u32 max_clusters,u32 * cluster_start,u32 * num_clusters)2298 int __ocfs2_claim_clusters(handle_t *handle,
2299 			   struct ocfs2_alloc_context *ac,
2300 			   u32 min_clusters,
2301 			   u32 max_clusters,
2302 			   u32 *cluster_start,
2303 			   u32 *num_clusters)
2304 {
2305 	int status;
2306 	unsigned int bits_wanted = max_clusters;
2307 	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2308 	struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2309 
2310 	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2311 
2312 	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2313 	       && ac->ac_which != OCFS2_AC_USE_MAIN);
2314 
2315 	if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2316 		WARN_ON(min_clusters > 1);
2317 
2318 		status = ocfs2_claim_local_alloc_bits(osb,
2319 						      handle,
2320 						      ac,
2321 						      bits_wanted,
2322 						      cluster_start,
2323 						      num_clusters);
2324 		if (!status)
2325 			atomic_inc(&osb->alloc_stats.local_data);
2326 	} else {
2327 		if (min_clusters > (osb->bitmap_cpg - 1)) {
2328 			/* The only paths asking for contiguousness
2329 			 * should know about this already. */
2330 			mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2331 			     "group bitmap size %u!\n", min_clusters,
2332 			     osb->bitmap_cpg);
2333 			status = -ENOSPC;
2334 			goto bail;
2335 		}
2336 		/* clamp the current request down to a realistic size. */
2337 		if (bits_wanted > (osb->bitmap_cpg - 1))
2338 			bits_wanted = osb->bitmap_cpg - 1;
2339 
2340 		status = ocfs2_claim_suballoc_bits(ac,
2341 						   handle,
2342 						   bits_wanted,
2343 						   min_clusters,
2344 						   &res);
2345 		if (!status) {
2346 			BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2347 			*cluster_start =
2348 				ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2349 								 res.sr_bg_blkno,
2350 								 res.sr_bit_offset);
2351 			atomic_inc(&osb->alloc_stats.bitmap_data);
2352 			*num_clusters = res.sr_bits;
2353 		}
2354 	}
2355 	if (status < 0) {
2356 		if (status != -ENOSPC)
2357 			mlog_errno(status);
2358 		goto bail;
2359 	}
2360 
2361 	ac->ac_bits_given += *num_clusters;
2362 
2363 bail:
2364 	if (status)
2365 		mlog_errno(status);
2366 	return status;
2367 }
2368 
ocfs2_claim_clusters(handle_t * handle,struct ocfs2_alloc_context * ac,u32 min_clusters,u32 * cluster_start,u32 * num_clusters)2369 int ocfs2_claim_clusters(handle_t *handle,
2370 			 struct ocfs2_alloc_context *ac,
2371 			 u32 min_clusters,
2372 			 u32 *cluster_start,
2373 			 u32 *num_clusters)
2374 {
2375 	unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2376 
2377 	return __ocfs2_claim_clusters(handle, ac, min_clusters,
2378 				      bits_wanted, cluster_start, num_clusters);
2379 }
2380 
ocfs2_block_group_clear_bits(handle_t * handle,struct inode * alloc_inode,struct ocfs2_group_desc * bg,struct buffer_head * group_bh,unsigned int bit_off,unsigned int num_bits,void (* undo_fn)(unsigned int bit,unsigned long * bmap))2381 static int ocfs2_block_group_clear_bits(handle_t *handle,
2382 					struct inode *alloc_inode,
2383 					struct ocfs2_group_desc *bg,
2384 					struct buffer_head *group_bh,
2385 					unsigned int bit_off,
2386 					unsigned int num_bits,
2387 					void (*undo_fn)(unsigned int bit,
2388 							unsigned long *bmap))
2389 {
2390 	int status;
2391 	unsigned int tmp;
2392 	struct ocfs2_group_desc *undo_bg = NULL;
2393 	struct journal_head *jh;
2394 
2395 	/* The caller got this descriptor from
2396 	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
2397 	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2398 
2399 	trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2400 
2401 	BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2402 	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2403 					 group_bh,
2404 					 undo_fn ?
2405 					 OCFS2_JOURNAL_ACCESS_UNDO :
2406 					 OCFS2_JOURNAL_ACCESS_WRITE);
2407 	if (status < 0) {
2408 		mlog_errno(status);
2409 		goto bail;
2410 	}
2411 
2412 	jh = bh2jh(group_bh);
2413 	if (undo_fn) {
2414 		spin_lock(&jh->b_state_lock);
2415 		undo_bg = (struct ocfs2_group_desc *) jh->b_committed_data;
2416 		BUG_ON(!undo_bg);
2417 	}
2418 
2419 	tmp = num_bits;
2420 	while(tmp--) {
2421 		ocfs2_clear_bit((bit_off + tmp),
2422 				(unsigned long *) bg->bg_bitmap);
2423 		if (undo_fn)
2424 			undo_fn(bit_off + tmp,
2425 				(unsigned long *) undo_bg->bg_bitmap);
2426 	}
2427 	le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2428 	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2429 		if (undo_fn)
2430 			spin_unlock(&jh->b_state_lock);
2431 		return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
2432 				   (unsigned long long)le64_to_cpu(bg->bg_blkno),
2433 				   le16_to_cpu(bg->bg_bits),
2434 				   le16_to_cpu(bg->bg_free_bits_count),
2435 				   num_bits);
2436 	}
2437 
2438 	if (undo_fn)
2439 		spin_unlock(&jh->b_state_lock);
2440 
2441 	ocfs2_journal_dirty(handle, group_bh);
2442 bail:
2443 	return status;
2444 }
2445 
2446 /*
2447  * expects the suballoc inode to already be locked.
2448  */
_ocfs2_free_suballoc_bits(handle_t * handle,struct inode * alloc_inode,struct buffer_head * alloc_bh,unsigned int start_bit,u64 bg_blkno,unsigned int count,void (* undo_fn)(unsigned int bit,unsigned long * bitmap))2449 static int _ocfs2_free_suballoc_bits(handle_t *handle,
2450 				     struct inode *alloc_inode,
2451 				     struct buffer_head *alloc_bh,
2452 				     unsigned int start_bit,
2453 				     u64 bg_blkno,
2454 				     unsigned int count,
2455 				     void (*undo_fn)(unsigned int bit,
2456 						     unsigned long *bitmap))
2457 {
2458 	int status = 0;
2459 	u32 tmp_used;
2460 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2461 	struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2462 	struct buffer_head *group_bh = NULL;
2463 	struct ocfs2_group_desc *group;
2464 
2465 	/* The alloc_bh comes from ocfs2_free_dinode() or
2466 	 * ocfs2_free_clusters().  The callers have all locked the
2467 	 * allocator and gotten alloc_bh from the lock call.  This
2468 	 * validates the dinode buffer.  Any corruption that has happened
2469 	 * is a code bug. */
2470 	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2471 	BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2472 
2473 	trace_ocfs2_free_suballoc_bits(
2474 		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2475 		(unsigned long long)bg_blkno,
2476 		start_bit, count);
2477 
2478 	status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2479 					     &group_bh);
2480 	if (status < 0) {
2481 		mlog_errno(status);
2482 		goto bail;
2483 	}
2484 	group = (struct ocfs2_group_desc *) group_bh->b_data;
2485 
2486 	BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2487 
2488 	status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2489 					      group, group_bh,
2490 					      start_bit, count, undo_fn);
2491 	if (status < 0) {
2492 		mlog_errno(status);
2493 		goto bail;
2494 	}
2495 
2496 	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2497 					 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2498 	if (status < 0) {
2499 		mlog_errno(status);
2500 		ocfs2_block_group_set_bits(handle, alloc_inode, group, group_bh,
2501 				start_bit, count);
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 	return status;
2514 }
2515 
ocfs2_free_suballoc_bits(handle_t * handle,struct inode * alloc_inode,struct buffer_head * alloc_bh,unsigned int start_bit,u64 bg_blkno,unsigned int count)2516 int ocfs2_free_suballoc_bits(handle_t *handle,
2517 			     struct inode *alloc_inode,
2518 			     struct buffer_head *alloc_bh,
2519 			     unsigned int start_bit,
2520 			     u64 bg_blkno,
2521 			     unsigned int count)
2522 {
2523 	return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2524 					 start_bit, bg_blkno, count, NULL);
2525 }
2526 
ocfs2_free_dinode(handle_t * handle,struct inode * inode_alloc_inode,struct buffer_head * inode_alloc_bh,struct ocfs2_dinode * di)2527 int ocfs2_free_dinode(handle_t *handle,
2528 		      struct inode *inode_alloc_inode,
2529 		      struct buffer_head *inode_alloc_bh,
2530 		      struct ocfs2_dinode *di)
2531 {
2532 	u64 blk = le64_to_cpu(di->i_blkno);
2533 	u16 bit = le16_to_cpu(di->i_suballoc_bit);
2534 	u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2535 
2536 	if (di->i_suballoc_loc)
2537 		bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2538 	return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2539 					inode_alloc_bh, bit, bg_blkno, 1);
2540 }
2541 
_ocfs2_free_clusters(handle_t * handle,struct inode * bitmap_inode,struct buffer_head * bitmap_bh,u64 start_blk,unsigned int num_clusters,void (* undo_fn)(unsigned int bit,unsigned long * bitmap))2542 static int _ocfs2_free_clusters(handle_t *handle,
2543 				struct inode *bitmap_inode,
2544 				struct buffer_head *bitmap_bh,
2545 				u64 start_blk,
2546 				unsigned int num_clusters,
2547 				void (*undo_fn)(unsigned int bit,
2548 						unsigned long *bitmap))
2549 {
2550 	int status;
2551 	u16 bg_start_bit;
2552 	u64 bg_blkno;
2553 
2554 	/* You can't ever have a contiguous set of clusters
2555 	 * bigger than a block group bitmap so we never have to worry
2556 	 * about looping on them.
2557 	 * This is expensive. We can safely remove once this stuff has
2558 	 * gotten tested really well. */
2559 	BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb,
2560 				ocfs2_blocks_to_clusters(bitmap_inode->i_sb,
2561 							 start_blk)));
2562 
2563 
2564 	ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2565 				     &bg_start_bit);
2566 
2567 	trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2568 			(unsigned long long)start_blk,
2569 			bg_start_bit, num_clusters);
2570 
2571 	status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2572 					   bg_start_bit, bg_blkno,
2573 					   num_clusters, undo_fn);
2574 	if (status < 0) {
2575 		mlog_errno(status);
2576 		goto out;
2577 	}
2578 
2579 	ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2580 					 num_clusters);
2581 
2582 out:
2583 	return status;
2584 }
2585 
ocfs2_free_clusters(handle_t * handle,struct inode * bitmap_inode,struct buffer_head * bitmap_bh,u64 start_blk,unsigned int num_clusters)2586 int ocfs2_free_clusters(handle_t *handle,
2587 			struct inode *bitmap_inode,
2588 			struct buffer_head *bitmap_bh,
2589 			u64 start_blk,
2590 			unsigned int num_clusters)
2591 {
2592 	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2593 				    start_blk, num_clusters,
2594 				    _ocfs2_set_bit);
2595 }
2596 
2597 /*
2598  * Give never-used clusters back to the global bitmap.  We don't need
2599  * to protect these bits in the undo buffer.
2600  */
ocfs2_release_clusters(handle_t * handle,struct inode * bitmap_inode,struct buffer_head * bitmap_bh,u64 start_blk,unsigned int num_clusters)2601 int ocfs2_release_clusters(handle_t *handle,
2602 			   struct inode *bitmap_inode,
2603 			   struct buffer_head *bitmap_bh,
2604 			   u64 start_blk,
2605 			   unsigned int num_clusters)
2606 {
2607 	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2608 				    start_blk, num_clusters,
2609 				    _ocfs2_clear_bit);
2610 }
2611 
2612 /*
2613  * For a given allocation, determine which allocators will need to be
2614  * accessed, and lock them, reserving the appropriate number of bits.
2615  *
2616  * Sparse file systems call this from ocfs2_write_begin_nolock()
2617  * and ocfs2_allocate_unwritten_extents().
2618  *
2619  * File systems which don't support holes call this from
2620  * ocfs2_extend_allocation().
2621  */
ocfs2_lock_allocators(struct inode * inode,struct ocfs2_extent_tree * et,u32 clusters_to_add,u32 extents_to_split,struct ocfs2_alloc_context ** data_ac,struct ocfs2_alloc_context ** meta_ac)2622 int ocfs2_lock_allocators(struct inode *inode,
2623 			  struct ocfs2_extent_tree *et,
2624 			  u32 clusters_to_add, u32 extents_to_split,
2625 			  struct ocfs2_alloc_context **data_ac,
2626 			  struct ocfs2_alloc_context **meta_ac)
2627 {
2628 	int ret = 0, num_free_extents;
2629 	unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2630 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2631 
2632 	*meta_ac = NULL;
2633 	if (data_ac)
2634 		*data_ac = NULL;
2635 
2636 	BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2637 
2638 	num_free_extents = ocfs2_num_free_extents(et);
2639 	if (num_free_extents < 0) {
2640 		ret = num_free_extents;
2641 		mlog_errno(ret);
2642 		goto out;
2643 	}
2644 
2645 	/*
2646 	 * Sparse allocation file systems need to be more conservative
2647 	 * with reserving room for expansion - the actual allocation
2648 	 * happens while we've got a journal handle open so re-taking
2649 	 * a cluster lock (because we ran out of room for another
2650 	 * extent) will violate ordering rules.
2651 	 *
2652 	 * Most of the time we'll only be seeing this 1 cluster at a time
2653 	 * anyway.
2654 	 *
2655 	 * Always lock for any unwritten extents - we might want to
2656 	 * add blocks during a split.
2657 	 */
2658 	if (!num_free_extents ||
2659 	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2660 		ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2661 		if (ret < 0) {
2662 			if (ret != -ENOSPC)
2663 				mlog_errno(ret);
2664 			goto out;
2665 		}
2666 	}
2667 
2668 	if (clusters_to_add == 0)
2669 		goto out;
2670 
2671 	ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2672 	if (ret < 0) {
2673 		if (ret != -ENOSPC)
2674 			mlog_errno(ret);
2675 		goto out;
2676 	}
2677 
2678 out:
2679 	if (ret) {
2680 		if (*meta_ac) {
2681 			ocfs2_free_alloc_context(*meta_ac);
2682 			*meta_ac = NULL;
2683 		}
2684 
2685 		/*
2686 		 * We cannot have an error and a non null *data_ac.
2687 		 */
2688 	}
2689 
2690 	return ret;
2691 }
2692 
2693 /*
2694  * Read the inode specified by blkno to get suballoc_slot and
2695  * suballoc_bit.
2696  */
ocfs2_get_suballoc_slot_bit(struct ocfs2_super * osb,u64 blkno,u16 * suballoc_slot,u64 * group_blkno,u16 * suballoc_bit)2697 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2698 				       u16 *suballoc_slot, u64 *group_blkno,
2699 				       u16 *suballoc_bit)
2700 {
2701 	int status;
2702 	struct buffer_head *inode_bh = NULL;
2703 	struct ocfs2_dinode *inode_fe;
2704 
2705 	trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2706 
2707 	/* dirty read disk */
2708 	status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2709 	if (status < 0) {
2710 		mlog(ML_ERROR, "read block %llu failed %d\n",
2711 		     (unsigned long long)blkno, status);
2712 		goto bail;
2713 	}
2714 
2715 	inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2716 	if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2717 		mlog(ML_ERROR, "invalid inode %llu requested\n",
2718 		     (unsigned long long)blkno);
2719 		status = -EINVAL;
2720 		goto bail;
2721 	}
2722 
2723 	if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2724 	    (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2725 		mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2726 		     (unsigned long long)blkno,
2727 		     (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2728 		status = -EINVAL;
2729 		goto bail;
2730 	}
2731 
2732 	if (suballoc_slot)
2733 		*suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2734 	if (suballoc_bit)
2735 		*suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2736 	if (group_blkno)
2737 		*group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2738 
2739 bail:
2740 	brelse(inode_bh);
2741 
2742 	if (status)
2743 		mlog_errno(status);
2744 	return status;
2745 }
2746 
2747 /*
2748  * test whether bit is SET in allocator bitmap or not.  on success, 0
2749  * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
2750  * is returned and *res is meaningless.  Call this after you have
2751  * cluster locked against suballoc, or you may get a result based on
2752  * non-up2date contents
2753  */
ocfs2_test_suballoc_bit(struct ocfs2_super * osb,struct inode * suballoc,struct buffer_head * alloc_bh,u64 group_blkno,u64 blkno,u16 bit,int * res)2754 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2755 				   struct inode *suballoc,
2756 				   struct buffer_head *alloc_bh,
2757 				   u64 group_blkno, u64 blkno,
2758 				   u16 bit, int *res)
2759 {
2760 	struct ocfs2_dinode *alloc_di;
2761 	struct ocfs2_group_desc *group;
2762 	struct buffer_head *group_bh = NULL;
2763 	u64 bg_blkno;
2764 	int status;
2765 
2766 	trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2767 				      (unsigned int)bit);
2768 
2769 	alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2770 	if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2771 		mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2772 		     (unsigned int)bit,
2773 		     ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2774 		status = -EINVAL;
2775 		goto bail;
2776 	}
2777 
2778 	bg_blkno = group_blkno ? group_blkno :
2779 		   ocfs2_which_suballoc_group(blkno, bit);
2780 	status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2781 					     &group_bh);
2782 	if (status < 0) {
2783 		mlog(ML_ERROR, "read group %llu failed %d\n",
2784 		     (unsigned long long)bg_blkno, status);
2785 		goto bail;
2786 	}
2787 
2788 	group = (struct ocfs2_group_desc *) group_bh->b_data;
2789 	*res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2790 
2791 bail:
2792 	brelse(group_bh);
2793 
2794 	if (status)
2795 		mlog_errno(status);
2796 	return status;
2797 }
2798 
2799 /*
2800  * Test if the bit representing this inode (blkno) is set in the
2801  * suballocator.
2802  *
2803  * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2804  *
2805  * In the event of failure, a negative value is returned and *res is
2806  * meaningless.
2807  *
2808  * Callers must make sure to hold nfs_sync_lock to prevent
2809  * ocfs2_delete_inode() on another node from accessing the same
2810  * suballocator concurrently.
2811  */
ocfs2_test_inode_bit(struct ocfs2_super * osb,u64 blkno,int * res)2812 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2813 {
2814 	int status;
2815 	u64 group_blkno = 0;
2816 	u16 suballoc_bit = 0, suballoc_slot = 0;
2817 	struct inode *inode_alloc_inode;
2818 	struct buffer_head *alloc_bh = NULL;
2819 
2820 	trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2821 
2822 	status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2823 					     &group_blkno, &suballoc_bit);
2824 	if (status < 0) {
2825 		mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2826 		goto bail;
2827 	}
2828 
2829 	if (suballoc_slot == (u16)OCFS2_INVALID_SLOT)
2830 		inode_alloc_inode = ocfs2_get_system_file_inode(osb,
2831 			GLOBAL_INODE_ALLOC_SYSTEM_INODE, suballoc_slot);
2832 	else
2833 		inode_alloc_inode = ocfs2_get_system_file_inode(osb,
2834 			INODE_ALLOC_SYSTEM_INODE, suballoc_slot);
2835 	if (!inode_alloc_inode) {
2836 		/* the error code could be inaccurate, but we are not able to
2837 		 * get the correct one. */
2838 		status = -EINVAL;
2839 		mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2840 		     (u32)suballoc_slot);
2841 		goto bail;
2842 	}
2843 
2844 	inode_lock(inode_alloc_inode);
2845 	status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2846 	if (status < 0) {
2847 		inode_unlock(inode_alloc_inode);
2848 		iput(inode_alloc_inode);
2849 		mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2850 		     (u32)suballoc_slot, status);
2851 		goto bail;
2852 	}
2853 
2854 	status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2855 					 group_blkno, blkno, suballoc_bit, res);
2856 	if (status < 0)
2857 		mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2858 
2859 	ocfs2_inode_unlock(inode_alloc_inode, 0);
2860 	inode_unlock(inode_alloc_inode);
2861 
2862 	iput(inode_alloc_inode);
2863 	brelse(alloc_bh);
2864 bail:
2865 	if (status)
2866 		mlog_errno(status);
2867 	return status;
2868 }
2869