projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ocfs2: Remove ENTRY from masklog.
[deliverable/linux.git] / fs / ocfs2 / suballoc.c
CommitLineData
ccd979bd
MF		 1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * suballoc.c
5 *
6 * metadata alloc and free
7 * Inspired by ext3 block groups.
8 *
9 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public
13 * License as published by the Free Software Foundation; either
14 * version 2 of the License, or (at your option) any later version.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
20 *
21 * You should have received a copy of the GNU General Public
22 * License along with this program; if not, write to the
23 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
24 * Boston, MA 021110-1307, USA.
25 */
26
27#include <linux/fs.h>
28#include <linux/types.h>
29#include <linux/slab.h>
30#include <linux/highmem.h>
31
32#define MLOG_MASK_PREFIX ML_DISK_ALLOC
33#include <cluster/masklog.h>
34
35#include "ocfs2.h"
36
37#include "alloc.h"
d6b32bbb 38#include "blockcheck.h"
ccd979bd
MF		 39#include "dlmglue.h"
40#include "inode.h"
41#include "journal.h"
42#include "localalloc.h"
43#include "suballoc.h"
44#include "super.h"
45#include "sysfile.h"
46#include "uptodate.h"
47
48#include "buffer_head_io.h"
49
ffda89a3 50#define NOT_ALLOC_NEW_GROUP 0
60ca81e8
TM		 51#define ALLOC_NEW_GROUP 0x1
52#define ALLOC_GROUPS_FROM_GLOBAL 0x2
ffda89a3 53
b89c5428 54#define OCFS2_MAX_TO_STEAL 1024
4d0ddb2c 55
7d1fe093 56struct ocfs2_suballoc_result {
2b6cb576
JB		 57 u64 sr_bg_blkno; /* The bg we allocated from. Set
58 to 0 when a block group is
59 contiguous. */
e49e2767
MF		 60 u64 sr_bg_stable_blkno; /*
61 * Doesn't change, always
62 * set to target block
63 * group descriptor
64 * block.
65 */
ba206635 66 u64 sr_blkno; /* The first allocated block */
7d1fe093
JB		 67 unsigned int sr_bit_offset; /* The bit in the bg */
68 unsigned int sr_bits; /* How many bits we claimed */
69};
70
b2b6ebf5
MF		 71static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
72{
73 if (res->sr_blkno == 0)
74 return 0;
75
76 if (res->sr_bg_blkno)
77 return res->sr_bg_blkno;
78
79 return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset);
80}
81
ccd979bd
MF		 82static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg);
83static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe);
84static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
1fabe148 85static int ocfs2_block_group_fill(handle_t *handle,
ccd979bd
MF		 86 struct inode *alloc_inode,
87 struct buffer_head *bg_bh,
88 u64 group_blkno,
798db35f 89 unsigned int group_clusters,
ccd979bd
MF		 90 u16 my_chain,
91 struct ocfs2_chain_list *cl);
92static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
93 struct inode *alloc_inode,
1187c968 94 struct buffer_head *bh,
60ca81e8 95 u64 max_block,
feb473a6 96 u64 *last_alloc_group,
60ca81e8 97 int flags);
ccd979bd 98
ccd979bd
MF		 99static int ocfs2_cluster_group_search(struct inode *inode,
100 struct buffer_head *group_bh,
101 u32 bits_wanted, u32 min_bits,
1187c968 102 u64 max_block,
7d1fe093 103 struct ocfs2_suballoc_result *res);
ccd979bd
MF		 104static int ocfs2_block_group_search(struct inode *inode,
105 struct buffer_head *group_bh,
106 u32 bits_wanted, u32 min_bits,
1187c968 107 u64 max_block,
7d1fe093 108 struct ocfs2_suballoc_result *res);
aa8f8e93 109static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1fabe148 110 handle_t *handle,
ccd979bd
MF		 111 u32 bits_wanted,
112 u32 min_bits,
7d1fe093 113 struct ocfs2_suballoc_result *res);
ccd979bd
MF		 114static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
115 int nr);
1fabe148 116static inline int ocfs2_block_group_set_bits(handle_t *handle,
ccd979bd
MF		 117 struct inode *alloc_inode,
118 struct ocfs2_group_desc *bg,
119 struct buffer_head *group_bh,
120 unsigned int bit_off,
121 unsigned int num_bits);
1fabe148 122static int ocfs2_relink_block_group(handle_t *handle,
ccd979bd
MF		 123 struct inode *alloc_inode,
124 struct buffer_head *fe_bh,
125 struct buffer_head *bg_bh,
126 struct buffer_head *prev_bg_bh,
127 u16 chain);
128static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
129 u32 wanted);
ccd979bd
MF		 130static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
131 u64 bg_blkno,
132 u16 bg_bit_off);
ccd979bd
MF		 133static inline void ocfs2_block_to_cluster_group(struct inode *inode,
134 u64 data_blkno,
135 u64 *bg_blkno,
136 u16 *bg_bit_off);
1187c968
JB		 137static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
138 u32 bits_wanted, u64 max_block,
60ca81e8 139 int flags,
1187c968 140 struct ocfs2_alloc_context **ac);
ccd979bd 141
9c7af40b 142void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
ccd979bd 143{
da5cbf2f
MF		 144 struct inode *inode = ac->ac_inode;
145
146 if (inode) {
147 if (ac->ac_which != OCFS2_AC_USE_LOCAL)
e63aecb6 148 ocfs2_inode_unlock(inode, 1);
da5cbf2f
MF		 149
150 mutex_unlock(&inode->i_mutex);
151
152 iput(inode);
4d0ddb2c 153 ac->ac_inode = NULL;
da5cbf2f 154 }
a81cb88b
MF		 155 brelse(ac->ac_bh);
156 ac->ac_bh = NULL;
e3b4a97d 157 ac->ac_resv = NULL;
e49e2767
MF		 158 if (ac->ac_find_loc_priv) {
159 kfree(ac->ac_find_loc_priv);
160 ac->ac_find_loc_priv = NULL;
161 }
4d0ddb2c
TM		 162}
163
164void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
165{
166 ocfs2_free_ac_resource(ac);
ccd979bd
MF		 167 kfree(ac);
168}
169
170static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
171{
172 return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
173}
174
57e3e797
JB		 175#define do_error(fmt, ...) \
176 do{ \
78c37eb0 177 if (resize) \
57e3e797
JB		 178 mlog(ML_ERROR, fmt "\n", ##__VA_ARGS__); \
179 else \
180 ocfs2_error(sb, fmt, ##__VA_ARGS__); \
181 } while (0)
182
970e4936
JB		 183static int ocfs2_validate_gd_self(struct super_block *sb,
184 struct buffer_head *bh,
78c37eb0 185 int resize)
970e4936
JB		 186{
187 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
188
7bf72ede 189 if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
68f64d47
JB		 190 do_error("Group descriptor #%llu has bad signature %.*s",
191 (unsigned long long)bh->b_blocknr, 7,
57e3e797
JB		 192 gd->bg_signature);
193 return -EINVAL;
7bf72ede
MF		 194 }
195
68f64d47
JB		 196 if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
197 do_error("Group descriptor #%llu has an invalid bg_blkno "
198 "of %llu",
199 (unsigned long long)bh->b_blocknr,
200 (unsigned long long)le64_to_cpu(gd->bg_blkno));
201 return -EINVAL;
202 }
203
204 if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
205 do_error("Group descriptor #%llu has an invalid "
206 "fs_generation of #%u",
207 (unsigned long long)bh->b_blocknr,
208 le32_to_cpu(gd->bg_generation));
209 return -EINVAL;
210 }
211
970e4936
JB		 212 if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
213 do_error("Group descriptor #%llu has bit count %u but "
214 "claims that %u are free",
215 (unsigned long long)bh->b_blocknr,
216 le16_to_cpu(gd->bg_bits),
217 le16_to_cpu(gd->bg_free_bits_count));
218 return -EINVAL;
219 }
220
221 if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
222 do_error("Group descriptor #%llu has bit count %u but "
223 "max bitmap bits of %u",
224 (unsigned long long)bh->b_blocknr,
225 le16_to_cpu(gd->bg_bits),
226 8 * le16_to_cpu(gd->bg_size));
227 return -EINVAL;
228 }
229
230 return 0;
231}
232
233static int ocfs2_validate_gd_parent(struct super_block *sb,
234 struct ocfs2_dinode *di,
235 struct buffer_head *bh,
78c37eb0 236 int resize)
970e4936
JB		 237{
238 unsigned int max_bits;
239 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
240
7bf72ede 241 if (di->i_blkno != gd->bg_parent_dinode) {
68f64d47 242 do_error("Group descriptor #%llu has bad parent "
57e3e797 243 "pointer (%llu, expected %llu)",
68f64d47 244 (unsigned long long)bh->b_blocknr,
57e3e797
JB		 245 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
246 (unsigned long long)le64_to_cpu(di->i_blkno));
247 return -EINVAL;
7bf72ede
MF		 248 }
249
250 max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
251 if (le16_to_cpu(gd->bg_bits) > max_bits) {
68f64d47
JB		 252 do_error("Group descriptor #%llu has bit count of %u",
253 (unsigned long long)bh->b_blocknr,
57e3e797
JB		 254 le16_to_cpu(gd->bg_bits));
255 return -EINVAL;
7bf72ede
MF		 256 }
257
78c37eb0
TM		 258 /* In resize, we may meet the case bg_chain == cl_next_free_rec. */
259 if ((le16_to_cpu(gd->bg_chain) >
260 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
261 ((le16_to_cpu(gd->bg_chain) ==
262 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
68f64d47
JB		 263 do_error("Group descriptor #%llu has bad chain %u",
264 (unsigned long long)bh->b_blocknr,
57e3e797
JB		 265 le16_to_cpu(gd->bg_chain));
266 return -EINVAL;
7bf72ede
MF		 267 }
268
970e4936
JB		 269 return 0;
270}
7bf72ede 271
57e3e797 272#undef do_error
7bf72ede 273
970e4936
JB		 274/*
275 * This version only prints errors. It does not fail the filesystem, and
276 * exists only for resize.
277 */
278int ocfs2_check_group_descriptor(struct super_block *sb,
279 struct ocfs2_dinode *di,
280 struct buffer_head *bh)
281{
282 int rc;
d6b32bbb
JB		 283 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
284
285 BUG_ON(!buffer_uptodate(bh));
970e4936 286
d6b32bbb
JB		 287 /*
288 * If the ecc fails, we return the error but otherwise
289 * leave the filesystem running. We know any error is
290 * local to this block.
291 */
292 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
13723d00
JB		 293 if (rc) {
294 mlog(ML_ERROR,
295 "Checksum failed for group descriptor %llu\n",
296 (unsigned long long)bh->b_blocknr);
297 } else
d6b32bbb 298 rc = ocfs2_validate_gd_self(sb, bh, 1);
970e4936
JB		 299 if (!rc)
300 rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
301
302 return rc;
303}
304
305static int ocfs2_validate_group_descriptor(struct super_block *sb,
306 struct buffer_head *bh)
307{
d6b32bbb
JB		 308 int rc;
309 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
310
970e4936
JB		 311 mlog(0, "Validating group descriptor %llu\n",
312 (unsigned long long)bh->b_blocknr);
313
d6b32bbb
JB		 314 BUG_ON(!buffer_uptodate(bh));
315
316 /*
317 * If the ecc fails, we return the error but otherwise
318 * leave the filesystem running. We know any error is
319 * local to this block.
320 */
321 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
322 if (rc)
323 return rc;
324
325 /*
326 * Errors after here are fatal.
327 */
328
970e4936 329 return ocfs2_validate_gd_self(sb, bh, 0);
7bf72ede
MF		 330}
331
68f64d47
JB		 332int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
333 u64 gd_blkno, struct buffer_head **bh)
334{
335 int rc;
336 struct buffer_head *tmp = *bh;
337
8cb471e8 338 rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
970e4936 339 ocfs2_validate_group_descriptor);
68f64d47
JB		 340 if (rc)
341 goto out;
342
970e4936 343 rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
68f64d47
JB		 344 if (rc) {
345 brelse(tmp);
346 goto out;
347 }
348
349 /* If ocfs2_read_block() got us a new bh, pass it up. */
350 if (!*bh)
351 *bh = tmp;
352
353out:
354 return rc;
355}
356
798db35f
JB		 357static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
358 struct ocfs2_group_desc *bg,
359 struct ocfs2_chain_list *cl,
47dea423 360 u64 p_blkno, unsigned int clusters)
798db35f
JB		 361{
362 struct ocfs2_extent_list *el = &bg->bg_list;
363 struct ocfs2_extent_rec *rec;
364
4711954e 365 BUG_ON(!ocfs2_supports_discontig_bg(osb));
798db35f
JB		 366 if (!el->l_next_free_rec)
367 el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
368 rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
4711954e 369 rec->e_blkno = cpu_to_le64(p_blkno);
798db35f
JB		 370 rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
371 le16_to_cpu(cl->cl_bpc));
47dea423 372 rec->e_leaf_clusters = cpu_to_le16(clusters);
798db35f 373 le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
4711954e
TM		 374 le16_add_cpu(&bg->bg_free_bits_count,
375 clusters * le16_to_cpu(cl->cl_bpc));
798db35f
JB		 376 le16_add_cpu(&el->l_next_free_rec, 1);
377}
378
1fabe148 379static int ocfs2_block_group_fill(handle_t *handle,
ccd979bd
MF		 380 struct inode *alloc_inode,
381 struct buffer_head *bg_bh,
382 u64 group_blkno,
798db35f 383 unsigned int group_clusters,
ccd979bd
MF		 384 u16 my_chain,
385 struct ocfs2_chain_list *cl)
386{
387 int status = 0;
798db35f 388 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
ccd979bd
MF		 389 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
390 struct super_block * sb = alloc_inode->i_sb;
391
ccd979bd 392 if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
b0697053
MF		 393 ocfs2_error(alloc_inode->i_sb, "group block (%llu) != "
394 "b_blocknr (%llu)",
395 (unsigned long long)group_blkno,
ccd979bd
MF		 396 (unsigned long long) bg_bh->b_blocknr);
397 status = -EIO;
398 goto bail;
399 }
400
13723d00 401 status = ocfs2_journal_access_gd(handle,
0cf2f763 402 INODE_CACHE(alloc_inode),
13723d00
JB		 403 bg_bh,
404 OCFS2_JOURNAL_ACCESS_CREATE);
ccd979bd
MF		 405 if (status < 0) {
406 mlog_errno(status);
407 goto bail;
408 }
409
410 memset(bg, 0, sb->s_blocksize);
411 strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
412 bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
8571882c
TM		 413 bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
414 osb->s_feature_incompat));
ccd979bd
MF		 415 bg->bg_chain = cpu_to_le16(my_chain);
416 bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
417 bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
418 bg->bg_blkno = cpu_to_le64(group_blkno);
798db35f
JB		 419 if (group_clusters == le16_to_cpu(cl->cl_cpg))
420 bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
421 else
4711954e 422 ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
798db35f
JB		 423 group_clusters);
424
ccd979bd
MF		 425 /* set the 1st bit in the bitmap to account for the descriptor block */
426 ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
427 bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
428
ec20cec7 429 ocfs2_journal_dirty(handle, bg_bh);
ccd979bd
MF		 430
431 /* There is no need to zero out or otherwise initialize the
432 * other blocks in a group - All valid FS metadata in a block
433 * group stores the superblock fs_generation value at
434 * allocation time. */
435
436bail:
437 mlog_exit(status);
438 return status;
439}
440
441static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
442{
443 u16 curr, best;
444
445 best = curr = 0;
446 while (curr < le16_to_cpu(cl->cl_count)) {
447 if (le32_to_cpu(cl->cl_recs[best].c_total) >
448 le32_to_cpu(cl->cl_recs[curr].c_total))
449 best = curr;
450 curr++;
451 }
452 return best;
453}
454
798db35f
JB		 455static struct buffer_head *
456ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
457 struct inode *alloc_inode,
458 struct ocfs2_alloc_context *ac,
459 struct ocfs2_chain_list *cl)
460{
461 int status;
462 u32 bit_off, num_bits;
463 u64 bg_blkno;
464 struct buffer_head *bg_bh;
465 unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
466
1ed9b777 467 status = ocfs2_claim_clusters(handle, ac,
798db35f
JB		 468 le16_to_cpu(cl->cl_cpg), &bit_off,
469 &num_bits);
470 if (status < 0) {
471 if (status != -ENOSPC)
472 mlog_errno(status);
473 goto bail;
474 }
475
476 /* setup the group */
477 bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
478 mlog(0, "new descriptor, record %u, at block %llu\n",
479 alloc_rec, (unsigned long long)bg_blkno);
480
481 bg_bh = sb_getblk(osb->sb, bg_blkno);
482 if (!bg_bh) {
483 status = -EIO;
484 mlog_errno(status);
485 goto bail;
486 }
487 ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
488
489 status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
490 bg_blkno, num_bits, alloc_rec, cl);
491 if (status < 0) {
492 brelse(bg_bh);
493 mlog_errno(status);
494 }
495
496bail:
497 return status ? ERR_PTR(status) : bg_bh;
498}
499
500static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
501 handle_t *handle,
502 struct ocfs2_alloc_context *ac,
503 unsigned int min_bits,
504 u32 *bit_off, u32 *num_bits)
505{
18d3a98f 506 int status = 0;
798db35f
JB		 507
508 while (min_bits) {
1ed9b777 509 status = ocfs2_claim_clusters(handle, ac, min_bits,
798db35f
JB		 510 bit_off, num_bits);
511 if (status != -ENOSPC)
512 break;
513
514 min_bits >>= 1;
515 }
516
517 return status;
518}
519
520static int ocfs2_block_group_grow_discontig(handle_t *handle,
521 struct inode *alloc_inode,
522 struct buffer_head *bg_bh,
523 struct ocfs2_alloc_context *ac,
524 struct ocfs2_chain_list *cl,
525 unsigned int min_bits)
526{
527 int status;
528 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
529 struct ocfs2_group_desc *bg =
530 (struct ocfs2_group_desc *)bg_bh->b_data;
4711954e
TM		 531 unsigned int needed = le16_to_cpu(cl->cl_cpg) -
532 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
798db35f
JB		 533 u32 p_cpos, clusters;
534 u64 p_blkno;
535 struct ocfs2_extent_list *el = &bg->bg_list;
536
537 status = ocfs2_journal_access_gd(handle,
538 INODE_CACHE(alloc_inode),
539 bg_bh,
540 OCFS2_JOURNAL_ACCESS_CREATE);
541 if (status < 0) {
542 mlog_errno(status);
543 goto bail;
544 }
545
546 while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
547 le16_to_cpu(el->l_count))) {
798db35f
JB		 548 if (min_bits > needed)
549 min_bits = needed;
550 status = ocfs2_block_group_claim_bits(osb, handle, ac,
551 min_bits, &p_cpos,
552 &clusters);
553 if (status < 0) {
554 if (status != -ENOSPC)
555 mlog_errno(status);
556 goto bail;
557 }
558 p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
559 ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
560 clusters);
561
562 min_bits = clusters;
4711954e
TM		 563 needed = le16_to_cpu(cl->cl_cpg) -
564 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
798db35f
JB		 565 }
566
567 if (needed > 0) {
4711954e
TM		 568 /*
569 * We have used up all the extent rec but can't fill up
570 * the cpg. So bail out.
571 */
572 status = -ENOSPC;
573 goto bail;
798db35f
JB		 574 }
575
576 ocfs2_journal_dirty(handle, bg_bh);
577
578bail:
579 return status;
580}
581
8b06bc59
JB		 582static void ocfs2_bg_alloc_cleanup(handle_t *handle,
583 struct ocfs2_alloc_context *cluster_ac,
584 struct inode *alloc_inode,
585 struct buffer_head *bg_bh)
798db35f 586{
8b06bc59 587 int i, ret;
798db35f
JB		 588 struct ocfs2_group_desc *bg;
589 struct ocfs2_extent_list *el;
590 struct ocfs2_extent_rec *rec;
591
592 if (!bg_bh)
593 return;
594
595 bg = (struct ocfs2_group_desc *)bg_bh->b_data;
596 el = &bg->bg_list;
597 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
598 rec = &el->l_recs[i];
8b06bc59
JB		 599 ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
600 cluster_ac->ac_bh,
601 le64_to_cpu(rec->e_blkno),
602 le32_to_cpu(rec->e_leaf_clusters));
603 if (ret)
604 mlog_errno(ret);
605 /* Try all the clusters to free */
798db35f
JB		 606 }
607
608 ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
609 brelse(bg_bh);
610}
611
612static struct buffer_head *
613ocfs2_block_group_alloc_discontig(handle_t *handle,
614 struct inode *alloc_inode,
615 struct ocfs2_alloc_context *ac,
8b06bc59 616 struct ocfs2_chain_list *cl)
798db35f
JB		 617{
618 int status;
619 u32 bit_off, num_bits;
620 u64 bg_blkno;
621 unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
622 struct buffer_head *bg_bh = NULL;
623 unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
624 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
625
4711954e 626 if (!ocfs2_supports_discontig_bg(osb)) {
798db35f
JB		 627 status = -ENOSPC;
628 goto bail;
629 }
630
8b06bc59
JB		 631 status = ocfs2_extend_trans(handle,
632 ocfs2_calc_bg_discontig_credits(osb->sb));
633 if (status) {
634 mlog_errno(status);
635 goto bail;
636 }
637
95ec0adf
JB		 638 /*
639 * We're going to be grabbing from multiple cluster groups.
640 * We don't have enough credits to relink them all, and the
641 * cluster groups will be staying in cache for the duration of
642 * this operation.
643 */
644 ac->ac_allow_chain_relink = 0;
645
798db35f
JB		 646 /* Claim the first region */
647 status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
648 &bit_off, &num_bits);
649 if (status < 0) {
650 if (status != -ENOSPC)
651 mlog_errno(status);
652 goto bail;
653 }
654 min_bits = num_bits;
655
656 /* setup the group */
657 bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
658 mlog(0, "new descriptor, record %u, at block %llu\n",
659 alloc_rec, (unsigned long long)bg_blkno);
660
661 bg_bh = sb_getblk(osb->sb, bg_blkno);
662 if (!bg_bh) {
663 status = -EIO;
664 mlog_errno(status);
665 goto bail;
666 }
667 ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
668
669 status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
670 bg_blkno, num_bits, alloc_rec, cl);
671 if (status < 0) {
672 mlog_errno(status);
673 goto bail;
674 }
675
676 status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
677 bg_bh, ac, cl, min_bits);
678 if (status)
679 mlog_errno(status);
680
681bail:
682 if (status)
8b06bc59 683 ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
798db35f
JB		 684 return status ? ERR_PTR(status) : bg_bh;
685}
686
ccd979bd
MF		 687/*
688 * We expect the block group allocator to already be locked.
689 */
690static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
691 struct inode *alloc_inode,
1187c968 692 struct buffer_head *bh,
60ca81e8 693 u64 max_block,
feb473a6 694 u64 *last_alloc_group,
60ca81e8 695 int flags)
ccd979bd
MF		 696{
697 int status, credits;
698 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
699 struct ocfs2_chain_list *cl;
700 struct ocfs2_alloc_context *ac = NULL;
1fabe148 701 handle_t *handle = NULL;
4711954e 702 u16 alloc_rec;
ccd979bd
MF		 703 struct buffer_head *bg_bh = NULL;
704 struct ocfs2_group_desc *bg;
705
706 BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
707
ccd979bd 708 cl = &fe->id2.i_chain;
1187c968
JB		 709 status = ocfs2_reserve_clusters_with_limit(osb,
710 le16_to_cpu(cl->cl_cpg),
60ca81e8 711 max_block, flags, &ac);
ccd979bd
MF		 712 if (status < 0) {
713 if (status != -ENOSPC)
714 mlog_errno(status);
715 goto bail;
716 }
717
718 credits = ocfs2_calc_group_alloc_credits(osb->sb,
719 le16_to_cpu(cl->cl_cpg));
65eff9cc 720 handle = ocfs2_start_trans(osb, credits);
ccd979bd
MF		 721 if (IS_ERR(handle)) {
722 status = PTR_ERR(handle);
723 handle = NULL;
724 mlog_errno(status);
725 goto bail;
726 }
727
feb473a6
TM		 728 if (last_alloc_group && *last_alloc_group != 0) {
729 mlog(0, "use old allocation group %llu for block group alloc\n",
730 (unsigned long long)*last_alloc_group);
731 ac->ac_last_group = *last_alloc_group;
732 }
798db35f
JB		 733
734 bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
735 ac, cl);
736 if (IS_ERR(bg_bh) && (PTR_ERR(bg_bh) == -ENOSPC))
737 bg_bh = ocfs2_block_group_alloc_discontig(handle,
738 alloc_inode,
8b06bc59 739 ac, cl);
798db35f
JB		 740 if (IS_ERR(bg_bh)) {
741 status = PTR_ERR(bg_bh);
742 bg_bh = NULL;
ccd979bd
MF		 743 if (status != -ENOSPC)
744 mlog_errno(status);
745 goto bail;
746 }
ccd979bd
MF		 747 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
748
0cf2f763 749 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
13723d00 750 bh, OCFS2_JOURNAL_ACCESS_WRITE);
ccd979bd
MF		 751 if (status < 0) {
752 mlog_errno(status);
753 goto bail;
754 }
755
4711954e
TM		 756 alloc_rec = le16_to_cpu(bg->bg_chain);
757 le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
ccd979bd 758 le16_to_cpu(bg->bg_free_bits_count));
4711954e 759 le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
798db35f 760 le16_to_cpu(bg->bg_bits));
0a463b74 761 cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
ccd979bd
MF		 762 if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
763 le16_add_cpu(&cl->cl_next_free_rec, 1);
764
765 le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
766 le16_to_cpu(bg->bg_free_bits_count));
767 le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
768 le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
769
ec20cec7 770 ocfs2_journal_dirty(handle, bh);
ccd979bd
MF		 771
772 spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
773 OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
774 fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
775 le32_to_cpu(fe->i_clusters)));
776 spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
777 i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
8110b073 778 alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
ccd979bd
MF		 779
780 status = 0;
feb473a6
TM		 781
782 /* save the new last alloc group so that the caller can cache it. */
783 if (last_alloc_group)
784 *last_alloc_group = ac->ac_last_group;
785
ccd979bd
MF		 786bail:
787 if (handle)
02dc1af4 788 ocfs2_commit_trans(osb, handle);
ccd979bd
MF		 789
790 if (ac)
791 ocfs2_free_alloc_context(ac);
792
a81cb88b 793 brelse(bg_bh);
ccd979bd
MF		 794
795 mlog_exit(status);
796 return status;
797}
798
799static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
da5cbf2f
MF		 800 struct ocfs2_alloc_context *ac,
801 int type,
ffda89a3 802 u32 slot,
feb473a6 803 u64 *last_alloc_group,
60ca81e8 804 int flags)
ccd979bd
MF		 805{
806 int status;
807 u32 bits_wanted = ac->ac_bits_wanted;
da5cbf2f 808 struct inode *alloc_inode;
ccd979bd 809 struct buffer_head *bh = NULL;
ccd979bd
MF		 810 struct ocfs2_dinode *fe;
811 u32 free_bits;
812
da5cbf2f
MF		 813 alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
814 if (!alloc_inode) {
815 mlog_errno(-EINVAL);
816 return -EINVAL;
817 }
ccd979bd 818
da5cbf2f
MF		 819 mutex_lock(&alloc_inode->i_mutex);
820
e63aecb6 821 status = ocfs2_inode_lock(alloc_inode, &bh, 1);
ccd979bd 822 if (status < 0) {
da5cbf2f
MF		 823 mutex_unlock(&alloc_inode->i_mutex);
824 iput(alloc_inode);
825
ccd979bd 826 mlog_errno(status);
da5cbf2f 827 return status;
ccd979bd
MF		 828 }
829
da5cbf2f 830 ac->ac_inode = alloc_inode;
a4a48911 831 ac->ac_alloc_slot = slot;
da5cbf2f 832
ccd979bd 833 fe = (struct ocfs2_dinode *) bh->b_data;
10995aa2
JB		 834
835 /* The bh was validated by the inode read inside
836 * ocfs2_inode_lock(). Any corruption is a code bug. */
837 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
838
ccd979bd 839 if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
b0697053
MF		 840 ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
841 (unsigned long long)le64_to_cpu(fe->i_blkno));
ccd979bd
MF		 842 status = -EIO;
843 goto bail;
844 }
845
846 free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
847 le32_to_cpu(fe->id1.bitmap1.i_used);
848
849 if (bits_wanted > free_bits) {
850 /* cluster bitmap never grows */
851 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
852 mlog(0, "Disk Full: wanted=%u, free_bits=%u\n",
853 bits_wanted, free_bits);
854 status = -ENOSPC;
855 goto bail;
856 }
857
60ca81e8 858 if (!(flags & ALLOC_NEW_GROUP)) {
ffda89a3
TM		 859 mlog(0, "Alloc File %u Full: wanted=%u, free_bits=%u, "
860 "and we don't alloc a new group for it.\n",
861 slot, bits_wanted, free_bits);
862 status = -ENOSPC;
863 goto bail;
864 }
865
1187c968 866 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
feb473a6
TM		 867 ac->ac_max_block,
868 last_alloc_group, flags);
ccd979bd
MF		 869 if (status < 0) {
870 if (status != -ENOSPC)
871 mlog_errno(status);
872 goto bail;
873 }
874 atomic_inc(&osb->alloc_stats.bg_extends);
875
876 /* You should never ask for this much metadata */
877 BUG_ON(bits_wanted >
878 (le32_to_cpu(fe->id1.bitmap1.i_total)
879 - le32_to_cpu(fe->id1.bitmap1.i_used)));
880 }
881
882 get_bh(bh);
883 ac->ac_bh = bh;
884bail:
a81cb88b 885 brelse(bh);
ccd979bd
MF		 886
887 mlog_exit(status);
888 return status;
889}
890
b89c5428
TY		 891static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
892{
893 spin_lock(&osb->osb_lock);
894 osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
895 spin_unlock(&osb->osb_lock);
896 atomic_set(&osb->s_num_inodes_stolen, 0);
897}
898
899static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
900{
901 spin_lock(&osb->osb_lock);
902 osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
903 spin_unlock(&osb->osb_lock);
904 atomic_set(&osb->s_num_meta_stolen, 0);
905}
906
907void ocfs2_init_steal_slots(struct ocfs2_super *osb)
908{
909 ocfs2_init_inode_steal_slot(osb);
910 ocfs2_init_meta_steal_slot(osb);
911}
912
913static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
914{
915 spin_lock(&osb->osb_lock);
916 if (type == INODE_ALLOC_SYSTEM_INODE)
917 osb->s_inode_steal_slot = slot;
918 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
919 osb->s_meta_steal_slot = slot;
920 spin_unlock(&osb->osb_lock);
921}
922
923static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
924{
925 int slot = OCFS2_INVALID_SLOT;
926
927 spin_lock(&osb->osb_lock);
928 if (type == INODE_ALLOC_SYSTEM_INODE)
929 slot = osb->s_inode_steal_slot;
930 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
931 slot = osb->s_meta_steal_slot;
932 spin_unlock(&osb->osb_lock);
933
934 return slot;
935}
936
937static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
938{
939 return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
940}
941
942static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
943{
944 return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
945}
946
947static int ocfs2_steal_resource(struct ocfs2_super *osb,
948 struct ocfs2_alloc_context *ac,
949 int type)
950{
951 int i, status = -ENOSPC;
952 int slot = __ocfs2_get_steal_slot(osb, type);
953
954 /* Start to steal resource from the first slot after ours. */
955 if (slot == OCFS2_INVALID_SLOT)
956 slot = osb->slot_num + 1;
957
958 for (i = 0; i < osb->max_slots; i++, slot++) {
959 if (slot == osb->max_slots)
960 slot = 0;
961
962 if (slot == osb->slot_num)
963 continue;
964
965 status = ocfs2_reserve_suballoc_bits(osb, ac,
966 type,
967 (u32)slot, NULL,
968 NOT_ALLOC_NEW_GROUP);
969 if (status >= 0) {
970 __ocfs2_set_steal_slot(osb, slot, type);
971 break;
972 }
973
974 ocfs2_free_ac_resource(ac);
975 }
976
977 return status;
978}
979
980static int ocfs2_steal_inode(struct ocfs2_super *osb,
981 struct ocfs2_alloc_context *ac)
982{
983 return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
984}
985
986static int ocfs2_steal_meta(struct ocfs2_super *osb,
987 struct ocfs2_alloc_context *ac)
988{
989 return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
990}
991
cf1d6c76
TY		 992int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
993 int blocks,
994 struct ocfs2_alloc_context **ac)
ccd979bd
MF		 995{
996 int status;
b89c5428 997 int slot = ocfs2_get_meta_steal_slot(osb);
ccd979bd 998
cd861280 999 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
ccd979bd
MF		 1000 if (!(*ac)) {
1001 status = -ENOMEM;
1002 mlog_errno(status);
1003 goto bail;
1004 }
1005
cf1d6c76 1006 (*ac)->ac_bits_wanted = blocks;
ccd979bd 1007 (*ac)->ac_which = OCFS2_AC_USE_META;
ccd979bd
MF		 1008 (*ac)->ac_group_search = ocfs2_block_group_search;
1009
b89c5428
TY		 1010 if (slot != OCFS2_INVALID_SLOT &&
1011 atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
1012 goto extent_steal;
1013
1014 atomic_set(&osb->s_num_meta_stolen, 0);
da5cbf2f 1015 status = ocfs2_reserve_suballoc_bits(osb, (*ac),
ffda89a3 1016 EXTENT_ALLOC_SYSTEM_INODE,
b89c5428 1017 (u32)osb->slot_num, NULL,
33d5d380 1018 ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
b89c5428
TY		 1019
1020
1021 if (status >= 0) {
1022 status = 0;
1023 if (slot != OCFS2_INVALID_SLOT)
1024 ocfs2_init_meta_steal_slot(osb);
1025 goto bail;
1026 } else if (status < 0 && status != -ENOSPC) {
1027 mlog_errno(status);
1028 goto bail;
1029 }
1030
1031 ocfs2_free_ac_resource(*ac);
1032
1033extent_steal:
1034 status = ocfs2_steal_meta(osb, *ac);
1035 atomic_inc(&osb->s_num_meta_stolen);
ccd979bd
MF		 1036 if (status < 0) {
1037 if (status != -ENOSPC)
1038 mlog_errno(status);
1039 goto bail;
1040 }
1041
1042 status = 0;
1043bail:
1044 if ((status < 0) && *ac) {
1045 ocfs2_free_alloc_context(*ac);
1046 *ac = NULL;
1047 }
1048
ccd979bd
MF		 1049 mlog_exit(status);
1050 return status;
1051}
1052
cf1d6c76
TY		 1053int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1054 struct ocfs2_extent_list *root_el,
1055 struct ocfs2_alloc_context **ac)
1056{
1057 return ocfs2_reserve_new_metadata_blocks(osb,
1058 ocfs2_extend_meta_needed(root_el),
1059 ac);
1060}
1061
ccd979bd 1062int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
ccd979bd
MF		 1063 struct ocfs2_alloc_context **ac)
1064{
1065 int status;
b89c5428 1066 int slot = ocfs2_get_inode_steal_slot(osb);
feb473a6 1067 u64 alloc_group;
ccd979bd 1068
cd861280 1069 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
ccd979bd
MF		 1070 if (!(*ac)) {
1071 status = -ENOMEM;
1072 mlog_errno(status);
1073 goto bail;
1074 }
1075
1076 (*ac)->ac_bits_wanted = 1;
ccd979bd
MF		 1077 (*ac)->ac_which = OCFS2_AC_USE_INODE;
1078
ccd979bd
MF		 1079 (*ac)->ac_group_search = ocfs2_block_group_search;
1080
1187c968
JB		 1081 /*
1082 * stat(2) can't handle i_ino > 32bits, so we tell the
1083 * lower levels not to allocate us a block group past that
12462f1d 1084 * limit. The 'inode64' mount option avoids this behavior.
1187c968 1085 */
12462f1d
JB		 1086 if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1087 (*ac)->ac_max_block = (u32)~0U;
1187c968 1088
4d0ddb2c
TM		 1089 /*
1090 * slot is set when we successfully steal inode from other nodes.
1091 * It is reset in 3 places:
1092 * 1. when we flush the truncate log
1093 * 2. when we complete local alloc recovery.
1094 * 3. when we successfully allocate from our own slot.
1095 * After it is set, we will go on stealing inodes until we find the
1096 * need to check our slots to see whether there is some space for us.
1097 */
1098 if (slot != OCFS2_INVALID_SLOT &&
b89c5428 1099 atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
4d0ddb2c
TM		 1100 goto inode_steal;
1101
1102 atomic_set(&osb->s_num_inodes_stolen, 0);
feb473a6 1103 alloc_group = osb->osb_inode_alloc_group;
da5cbf2f
MF		 1104 status = ocfs2_reserve_suballoc_bits(osb, *ac,
1105 INODE_ALLOC_SYSTEM_INODE,
b89c5428 1106 (u32)osb->slot_num,
feb473a6 1107 &alloc_group,
60ca81e8
TM		 1108 ALLOC_NEW_GROUP |
1109 ALLOC_GROUPS_FROM_GLOBAL);
4d0ddb2c
TM		 1110 if (status >= 0) {
1111 status = 0;
1112
feb473a6
TM		 1113 spin_lock(&osb->osb_lock);
1114 osb->osb_inode_alloc_group = alloc_group;
1115 spin_unlock(&osb->osb_lock);
1116 mlog(0, "after reservation, new allocation group is "
1117 "%llu\n", (unsigned long long)alloc_group);
1118
4d0ddb2c
TM		 1119 /*
1120 * Some inodes must be freed by us, so try to allocate
1121 * from our own next time.
1122 */
1123 if (slot != OCFS2_INVALID_SLOT)
1124 ocfs2_init_inode_steal_slot(osb);
1125 goto bail;
1126 } else if (status < 0 && status != -ENOSPC) {
1127 mlog_errno(status);
1128 goto bail;
1129 }
1130
1131 ocfs2_free_ac_resource(*ac);
1132
1133inode_steal:
b89c5428 1134 status = ocfs2_steal_inode(osb, *ac);
4d0ddb2c 1135 atomic_inc(&osb->s_num_inodes_stolen);
ccd979bd
MF		 1136 if (status < 0) {
1137 if (status != -ENOSPC)
1138 mlog_errno(status);
1139 goto bail;
1140 }
1141
1142 status = 0;
1143bail:
1144 if ((status < 0) && *ac) {
1145 ocfs2_free_alloc_context(*ac);
1146 *ac = NULL;
1147 }
1148
ccd979bd
MF		 1149 mlog_exit(status);
1150 return status;
1151}
1152
1153/* local alloc code has to do the same thing, so rather than do this
1154 * twice.. */
1155int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1156 struct ocfs2_alloc_context *ac)
1157{
1158 int status;
1159
ccd979bd
MF		 1160 ac->ac_which = OCFS2_AC_USE_MAIN;
1161 ac->ac_group_search = ocfs2_cluster_group_search;
1162
da5cbf2f
MF		 1163 status = ocfs2_reserve_suballoc_bits(osb, ac,
1164 GLOBAL_BITMAP_SYSTEM_INODE,
feb473a6 1165 OCFS2_INVALID_SLOT, NULL,
ffda89a3 1166 ALLOC_NEW_GROUP);
da5cbf2f 1167 if (status < 0 && status != -ENOSPC) {
ccd979bd 1168 mlog_errno(status);
da5cbf2f
MF		 1169 goto bail;
1170 }
1171
ccd979bd
MF		 1172bail:
1173 return status;
1174}
1175
1176/* Callers don't need to care which bitmap (local alloc or main) to
1177 * use so we figure it out for them, but unfortunately this clutters
1178 * things a bit. */
1187c968
JB		 1179static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1180 u32 bits_wanted, u64 max_block,
60ca81e8 1181 int flags,
1187c968 1182 struct ocfs2_alloc_context **ac)
ccd979bd
MF		 1183{
1184 int status;
1185
cd861280 1186 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
ccd979bd
MF		 1187 if (!(*ac)) {
1188 status = -ENOMEM;
1189 mlog_errno(status);
1190 goto bail;
1191 }
1192
1193 (*ac)->ac_bits_wanted = bits_wanted;
1187c968 1194 (*ac)->ac_max_block = max_block;
ccd979bd
MF		 1195
1196 status = -ENOSPC;
60ca81e8
TM		 1197 if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1198 ocfs2_alloc_should_use_local(osb, bits_wanted)) {
ccd979bd 1199 status = ocfs2_reserve_local_alloc_bits(osb,
ccd979bd
MF		 1200 bits_wanted,
1201 *ac);
a57c8fd2 1202 if ((status < 0) && (status != -ENOSPC)) {
ccd979bd
MF		 1203 mlog_errno(status);
1204 goto bail;
ccd979bd
MF		 1205 }
1206 }
1207
1208 if (status == -ENOSPC) {
1209 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1210 if (status < 0) {
1211 if (status != -ENOSPC)
1212 mlog_errno(status);
1213 goto bail;
1214 }
1215 }
1216
1217 status = 0;
1218bail:
1219 if ((status < 0) && *ac) {
1220 ocfs2_free_alloc_context(*ac);
1221 *ac = NULL;
1222 }
1223
1224 mlog_exit(status);
1225 return status;
1226}
1227
1187c968
JB		 1228int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1229 u32 bits_wanted,
1230 struct ocfs2_alloc_context **ac)
1231{
60ca81e8
TM		 1232 return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1233 ALLOC_NEW_GROUP, ac);
1187c968
JB		 1234}
1235
ccd979bd
MF		 1236/*
1237 * More or less lifted from ext3. I'll leave their description below:
1238 *
1239 * "For ext3 allocations, we must not reuse any blocks which are
1240 * allocated in the bitmap buffer's "last committed data" copy. This
1241 * prevents deletes from freeing up the page for reuse until we have
1242 * committed the delete transaction.
1243 *
1244 * If we didn't do this, then deleting something and reallocating it as
1245 * data would allow the old block to be overwritten before the
1246 * transaction committed (because we force data to disk before commit).
1247 * This would lead to corruption if we crashed between overwriting the
1248 * data and committing the delete.
1249 *
1250 * @@@ We may want to make this allocation behaviour conditional on
1251 * data-writes at some point, and disable it for metadata allocations or
1252 * sync-data inodes."
1253 *
1254 * Note: OCFS2 already does this differently for metadata vs data
c78bad11 1255 * allocations, as those bitmaps are separate and undo access is never
ccd979bd
MF		 1256 * called on a metadata group descriptor.
1257 */
1258static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1259 int nr)
1260{
1261 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
94e41ecf 1262 int ret;
ccd979bd
MF		 1263
1264 if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1265 return 0;
94e41ecf
SM		 1266
1267 if (!buffer_jbd(bg_bh))
ccd979bd
MF		 1268 return 1;
1269
94e41ecf 1270 jbd_lock_bh_state(bg_bh);
ccd979bd 1271 bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
94e41ecf
SM		 1272 if (bg)
1273 ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1274 else
1275 ret = 1;
1276 jbd_unlock_bh_state(bg_bh);
1277
1278 return ret;
ccd979bd
MF		 1279}
1280
1281static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1282 struct buffer_head *bg_bh,
1283 unsigned int bits_wanted,
7bf72ede 1284 unsigned int total_bits,
7d1fe093 1285 struct ocfs2_suballoc_result *res)
ccd979bd
MF		 1286{
1287 void *bitmap;
1288 u16 best_offset, best_size;
1289 int offset, start, found, status = 0;
1290 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1291
42035306
JB		 1292 /* Callers got this descriptor from
1293 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1294 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
ccd979bd
MF		 1295
1296 found = start = best_offset = best_size = 0;
1297 bitmap = bg->bg_bitmap;
1298
7bf72ede
MF		 1299 while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1300 if (offset == total_bits)
ccd979bd
MF		 1301 break;
1302
1303 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1304 /* We found a zero, but we can't use it as it
1305 * hasn't been put to disk yet! */
1306 found = 0;
1307 start = offset + 1;
1308 } else if (offset == start) {
1309 /* we found a zero */
1310 found++;
1311 /* move start to the next bit to test */
1312 start++;
1313 } else {
1314 /* got a zero after some ones */
1315 found = 1;
1316 start = offset + 1;
1317 }
1318 if (found > best_size) {
1319 best_size = found;
1320 best_offset = start - found;
1321 }
1322 /* we got everything we needed */
1323 if (found == bits_wanted) {
1324 /* mlog(0, "Found it all!\n"); */
1325 break;
1326 }
1327 }
1328
7d1fe093
JB		 1329 if (best_size) {
1330 res->sr_bit_offset = best_offset;
1331 res->sr_bits = best_size;
ccd979bd
MF		 1332 } else {
1333 status = -ENOSPC;
1334 /* No error log here -- see the comment above
1335 * ocfs2_test_bg_bit_allocatable */
1336 }
1337
1338 return status;
1339}
1340
1fabe148 1341static inline int ocfs2_block_group_set_bits(handle_t *handle,
ccd979bd
MF		 1342 struct inode *alloc_inode,
1343 struct ocfs2_group_desc *bg,
1344 struct buffer_head *group_bh,
1345 unsigned int bit_off,
1346 unsigned int num_bits)
1347{
1348 int status;
1349 void *bitmap = bg->bg_bitmap;
1350 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1351
42035306
JB		 1352 /* All callers get the descriptor via
1353 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1354 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
ccd979bd
MF		 1355 BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1356
1357 mlog(0, "block_group_set_bits: off = %u, num = %u\n", bit_off,
1358 num_bits);
1359
1360 if (ocfs2_is_cluster_bitmap(alloc_inode))
1361 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1362
13723d00 1363 status = ocfs2_journal_access_gd(handle,
0cf2f763 1364 INODE_CACHE(alloc_inode),
13723d00
JB		 1365 group_bh,
1366 journal_type);
ccd979bd
MF		 1367 if (status < 0) {
1368 mlog_errno(status);
1369 goto bail;
1370 }
1371
1372 le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
9b5cd10e
SE		 1373 if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1374 ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
1375 " count %u but claims %u are freed. num_bits %d",
1376 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1377 le16_to_cpu(bg->bg_bits),
1378 le16_to_cpu(bg->bg_free_bits_count), num_bits);
1379 return -EROFS;
1380 }
ccd979bd
MF		 1381 while(num_bits--)
1382 ocfs2_set_bit(bit_off++, bitmap);
1383
ec20cec7 1384 ocfs2_journal_dirty(handle, group_bh);
ccd979bd
MF		 1385
1386bail:
1387 mlog_exit(status);
1388 return status;
1389}
1390
1391/* find the one with the most empty bits */
1392static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1393{
1394 u16 curr, best;
1395
1396 BUG_ON(!cl->cl_next_free_rec);
1397
1398 best = curr = 0;
1399 while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1400 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1401 le32_to_cpu(cl->cl_recs[best].c_free))
1402 best = curr;
1403 curr++;
1404 }
1405
1406 BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1407 return best;
1408}
1409
1fabe148 1410static int ocfs2_relink_block_group(handle_t *handle,
ccd979bd
MF		 1411 struct inode *alloc_inode,
1412 struct buffer_head *fe_bh,
1413 struct buffer_head *bg_bh,
1414 struct buffer_head *prev_bg_bh,
1415 u16 chain)
1416{
1417 int status;
1418 /* there is a really tiny chance the journal calls could fail,
1419 * but we wouldn't want inconsistent blocks in *any* case. */
1420 u64 fe_ptr, bg_ptr, prev_bg_ptr;
1421 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1422 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1423 struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1424
42035306
JB		 1425 /* The caller got these descriptors from
1426 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1427 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1428 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
ccd979bd 1429
b0697053 1430 mlog(0, "Suballoc %llu, chain %u, move group %llu to top, prev = %llu\n",
1ca1a111
MF		 1431 (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1432 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1433 (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
ccd979bd
MF		 1434
1435 fe_ptr = le64_to_cpu(fe->id2.i_chain.cl_recs[chain].c_blkno);
1436 bg_ptr = le64_to_cpu(bg->bg_next_group);
1437 prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1438
0cf2f763
JB		 1439 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1440 prev_bg_bh,
13723d00 1441 OCFS2_JOURNAL_ACCESS_WRITE);
ccd979bd
MF		 1442 if (status < 0) {
1443 mlog_errno(status);
1444 goto out_rollback;
1445 }
1446
1447 prev_bg->bg_next_group = bg->bg_next_group;
ec20cec7 1448 ocfs2_journal_dirty(handle, prev_bg_bh);
ccd979bd 1449
0cf2f763
JB		 1450 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1451 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
ccd979bd
MF		 1452 if (status < 0) {
1453 mlog_errno(status);
1454 goto out_rollback;
1455 }
1456
1457 bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
ec20cec7 1458 ocfs2_journal_dirty(handle, bg_bh);
ccd979bd 1459
0cf2f763
JB		 1460 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1461 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
ccd979bd
MF		 1462 if (status < 0) {
1463 mlog_errno(status);
1464 goto out_rollback;
1465 }
1466
1467 fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
ec20cec7 1468 ocfs2_journal_dirty(handle, fe_bh);
ccd979bd 1469
ccd979bd
MF		 1470out_rollback:
1471 if (status < 0) {
1472 fe->id2.i_chain.cl_recs[chain].c_blkno = cpu_to_le64(fe_ptr);
1473 bg->bg_next_group = cpu_to_le64(bg_ptr);
1474 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1475 }
42035306 1476
ccd979bd
MF		 1477 mlog_exit(status);
1478 return status;
1479}
1480
1481static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1482 u32 wanted)
1483{
1484 return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1485}
1486
1487/* return 0 on success, -ENOSPC to keep searching and any other < 0
1488 * value on error. */
1489static int ocfs2_cluster_group_search(struct inode *inode,
1490 struct buffer_head *group_bh,
1491 u32 bits_wanted, u32 min_bits,
1187c968 1492 u64 max_block,
7d1fe093 1493 struct ocfs2_suballoc_result *res)
ccd979bd
MF		 1494{
1495 int search = -ENOSPC;
1496 int ret;
1187c968 1497 u64 blkoff;
7bf72ede 1498 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
9c7af40b 1499 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7bf72ede 1500 unsigned int max_bits, gd_cluster_off;
ccd979bd
MF		 1501
1502 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1503
7bf72ede
MF		 1504 if (gd->bg_free_bits_count) {
1505 max_bits = le16_to_cpu(gd->bg_bits);
1506
1507 /* Tail groups in cluster bitmaps which aren't cpg
1508 * aligned are prone to partial extention by a failed
1509 * fs resize. If the file system resize never got to
1510 * update the dinode cluster count, then we don't want
1511 * to trust any clusters past it, regardless of what
1512 * the group descriptor says. */
1513 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1514 le64_to_cpu(gd->bg_blkno));
1515 if ((gd_cluster_off + max_bits) >
1516 OCFS2_I(inode)->ip_clusters) {
1517 max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1518 mlog(0, "Desc %llu, bg_bits %u, clusters %u, use %u\n",
1519 (unsigned long long)le64_to_cpu(gd->bg_blkno),
1520 le16_to_cpu(gd->bg_bits),
1521 OCFS2_I(inode)->ip_clusters, max_bits);
1522 }
1523
ccd979bd
MF		 1524 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1525 group_bh, bits_wanted,
7d1fe093 1526 max_bits, res);
ccd979bd
MF		 1527 if (ret)
1528 return ret;
1529
1187c968
JB		 1530 if (max_block) {
1531 blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1532 gd_cluster_off +
7d1fe093
JB		 1533 res->sr_bit_offset +
1534 res->sr_bits);
1187c968
JB		 1535 mlog(0, "Checking %llu against %llu\n",
1536 (unsigned long long)blkoff,
1537 (unsigned long long)max_block);
1538 if (blkoff > max_block)
1539 return -ENOSPC;
1540 }
1541
ccd979bd
MF		 1542 /* ocfs2_block_group_find_clear_bits() might
1543 * return success, but we still want to return
1544 * -ENOSPC unless it found the minimum number
1545 * of bits. */
7d1fe093 1546 if (min_bits <= res->sr_bits)
ccd979bd 1547 search = 0; /* success */
7d1fe093 1548 else if (res->sr_bits) {
9c7af40b
MF		 1549 /*
1550 * Don't show bits which we'll be returning
1551 * for allocation to the local alloc bitmap.
1552 */
7d1fe093 1553 ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
ccd979bd
MF		 1554 }
1555 }
1556
1557 return search;
1558}
1559
1560static int ocfs2_block_group_search(struct inode *inode,
1561 struct buffer_head *group_bh,
1562 u32 bits_wanted, u32 min_bits,
1187c968 1563 u64 max_block,
7d1fe093 1564 struct ocfs2_suballoc_result *res)
ccd979bd
MF		 1565{
1566 int ret = -ENOSPC;
1187c968 1567 u64 blkoff;
ccd979bd
MF		 1568 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1569
1570 BUG_ON(min_bits != 1);
1571 BUG_ON(ocfs2_is_cluster_bitmap(inode));
1572
1187c968 1573 if (bg->bg_free_bits_count) {
ccd979bd
MF		 1574 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1575 group_bh, bits_wanted,
7bf72ede 1576 le16_to_cpu(bg->bg_bits),
7d1fe093 1577 res);
1187c968 1578 if (!ret && max_block) {
7d1fe093
JB		 1579 blkoff = le64_to_cpu(bg->bg_blkno) +
1580 res->sr_bit_offset + res->sr_bits;
1187c968
JB		 1581 mlog(0, "Checking %llu against %llu\n",
1582 (unsigned long long)blkoff,
1583 (unsigned long long)max_block);
1584 if (blkoff > max_block)
1585 ret = -ENOSPC;
1586 }
1587 }
ccd979bd
MF		 1588
1589 return ret;
1590}
1591
883d4cae 1592static int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1fabe148 1593 handle_t *handle,
883d4cae
MF		 1594 struct buffer_head *di_bh,
1595 u32 num_bits,
1596 u16 chain)
1597{
1598 int ret;
1599 u32 tmp_used;
1600 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1601 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1602
0cf2f763 1603 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
13723d00 1604 OCFS2_JOURNAL_ACCESS_WRITE);
883d4cae
MF		 1605 if (ret < 0) {
1606 mlog_errno(ret);
1607 goto out;
1608 }
1609
1610 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1611 di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1612 le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
ec20cec7 1613 ocfs2_journal_dirty(handle, di_bh);
883d4cae
MF		 1614
1615out:
1616 return ret;
1617}
1618
ba206635
JB		 1619static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1620 struct ocfs2_extent_rec *rec,
1621 struct ocfs2_chain_list *cl)
13e434cf
JB		 1622{
1623 unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1624 unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1625 unsigned int bitcount = le32_to_cpu(rec->e_leaf_clusters) * bpc;
1626
1627 if (res->sr_bit_offset < bitoff)
1628 return 0;
1629 if (res->sr_bit_offset >= (bitoff + bitcount))
1630 return 0;
ba206635
JB		 1631 res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1632 (res->sr_bit_offset - bitoff);
13e434cf
JB		 1633 if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1634 res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1635 return 1;
1636}
1637
ba206635
JB		 1638static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1639 struct ocfs2_group_desc *bg,
1640 struct ocfs2_suballoc_result *res)
13e434cf
JB		 1641{
1642 int i;
2b6cb576 1643 u64 bg_blkno = res->sr_bg_blkno; /* Save off */
13e434cf
JB		 1644 struct ocfs2_extent_rec *rec;
1645 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1646 struct ocfs2_chain_list *cl = &di->id2.i_chain;
1647
ba206635
JB		 1648 if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1649 res->sr_blkno = 0;
13e434cf 1650 return;
ba206635 1651 }
13e434cf 1652
ba206635 1653 res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
2b6cb576 1654 res->sr_bg_blkno = 0; /* Clear it for contig block groups */
4711954e 1655 if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
ba206635 1656 !bg->bg_list.l_next_free_rec)
13e434cf
JB		 1657 return;
1658
1659 for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1660 rec = &bg->bg_list.l_recs[i];
2b6cb576
JB		 1661 if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1662 res->sr_bg_blkno = bg_blkno; /* Restore */
13e434cf 1663 break;
2b6cb576 1664 }
13e434cf
JB		 1665 }
1666}
1667
883d4cae 1668static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1fabe148 1669 handle_t *handle,
883d4cae
MF		 1670 u32 bits_wanted,
1671 u32 min_bits,
7d1fe093 1672 struct ocfs2_suballoc_result *res,
883d4cae
MF		 1673 u16 *bits_left)
1674{
1675 int ret;
883d4cae
MF		 1676 struct buffer_head *group_bh = NULL;
1677 struct ocfs2_group_desc *gd;
68f64d47 1678 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
883d4cae 1679 struct inode *alloc_inode = ac->ac_inode;
883d4cae 1680
7d1fe093
JB		 1681 ret = ocfs2_read_group_descriptor(alloc_inode, di,
1682 res->sr_bg_blkno, &group_bh);
883d4cae
MF		 1683 if (ret < 0) {
1684 mlog_errno(ret);
1685 return ret;
1686 }
1687
1688 gd = (struct ocfs2_group_desc *) group_bh->b_data;
883d4cae 1689 ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
7d1fe093 1690 ac->ac_max_block, res);
883d4cae
MF		 1691 if (ret < 0) {
1692 if (ret != -ENOSPC)
1693 mlog_errno(ret);
1694 goto out;
1695 }
1696
13e434cf 1697 if (!ret)
ba206635 1698 ocfs2_bg_discontig_fix_result(ac, gd, res);
13e434cf 1699
e49e2767
MF		 1700 /*
1701 * sr_bg_blkno might have been changed by
1702 * ocfs2_bg_discontig_fix_result
1703 */
1704 res->sr_bg_stable_blkno = group_bh->b_blocknr;
1705
1706 if (ac->ac_find_loc_only)
1707 goto out_loc_only;
1708
883d4cae 1709 ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
7d1fe093 1710 res->sr_bits,
883d4cae
MF		 1711 le16_to_cpu(gd->bg_chain));
1712 if (ret < 0) {
1713 mlog_errno(ret);
1714 goto out;
1715 }
1716
1717 ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
7d1fe093 1718 res->sr_bit_offset, res->sr_bits);
883d4cae
MF		 1719 if (ret < 0)
1720 mlog_errno(ret);
1721
e49e2767 1722out_loc_only:
883d4cae
MF		 1723 *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1724
1725out:
1726 brelse(group_bh);
1727
1728 return ret;
1729}
1730
ccd979bd 1731static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1fabe148 1732 handle_t *handle,
ccd979bd
MF		 1733 u32 bits_wanted,
1734 u32 min_bits,
7d1fe093 1735 struct ocfs2_suballoc_result *res,
883d4cae 1736 u16 *bits_left)
ccd979bd
MF		 1737{
1738 int status;
7d1fe093 1739 u16 chain;
ccd979bd 1740 u64 next_group;
ccd979bd
MF		 1741 struct inode *alloc_inode = ac->ac_inode;
1742 struct buffer_head *group_bh = NULL;
1743 struct buffer_head *prev_group_bh = NULL;
1744 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1745 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1746 struct ocfs2_group_desc *bg;
1747
1748 chain = ac->ac_chain;
b0697053
MF		 1749 mlog(0, "trying to alloc %u bits from chain %u, inode %llu\n",
1750 bits_wanted, chain,
1751 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno);
ccd979bd 1752
68f64d47
JB		 1753 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1754 le64_to_cpu(cl->cl_recs[chain].c_blkno),
1755 &group_bh);
ccd979bd
MF		 1756 if (status < 0) {
1757 mlog_errno(status);
1758 goto bail;
1759 }
1760 bg = (struct ocfs2_group_desc *) group_bh->b_data;
ccd979bd
MF		 1761
1762 status = -ENOSPC;
1763 /* for now, the chain search is a bit simplistic. We just use
1764 * the 1st group with any empty bits. */
1187c968
JB		 1765 while ((status = ac->ac_group_search(alloc_inode, group_bh,
1766 bits_wanted, min_bits,
7d1fe093
JB		 1767 ac->ac_max_block,
1768 res)) == -ENOSPC) {
ccd979bd
MF		 1769 if (!bg->bg_next_group)
1770 break;
a81cb88b
MF		 1771
1772 brelse(prev_group_bh);
1773 prev_group_bh = NULL;
1774
ccd979bd
MF		 1775 next_group = le64_to_cpu(bg->bg_next_group);
1776 prev_group_bh = group_bh;
1777 group_bh = NULL;
68f64d47
JB		 1778 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1779 next_group, &group_bh);
ccd979bd
MF		 1780 if (status < 0) {
1781 mlog_errno(status);
1782 goto bail;
1783 }
1784 bg = (struct ocfs2_group_desc *) group_bh->b_data;
ccd979bd
MF		 1785 }
1786 if (status < 0) {
1787 if (status != -ENOSPC)
1788 mlog_errno(status);
1789 goto bail;
1790 }
1791
b0697053 1792 mlog(0, "alloc succeeds: we give %u bits from block group %llu\n",
7d1fe093 1793 res->sr_bits, (unsigned long long)le64_to_cpu(bg->bg_blkno));
ccd979bd 1794
7d1fe093 1795 res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
ccd979bd 1796
7d1fe093 1797 BUG_ON(res->sr_bits == 0);
13e434cf 1798 if (!status)
ba206635 1799 ocfs2_bg_discontig_fix_result(ac, bg, res);
13e434cf 1800
e49e2767
MF		 1801 /*
1802 * sr_bg_blkno might have been changed by
1803 * ocfs2_bg_discontig_fix_result
1804 */
1805 res->sr_bg_stable_blkno = group_bh->b_blocknr;
ccd979bd
MF		 1806
1807 /*
1808 * Keep track of previous block descriptor read. When
1809 * we find a target, if we have read more than X
1810 * number of descriptors, and the target is reasonably
1811 * empty, relink him to top of his chain.
1812 *
1813 * We've read 0 extra blocks and only send one more to
1814 * the transaction, yet the next guy to search has a
1815 * much easier time.
1816 *
1817 * Do this *after* figuring out how many bits we're taking out
1818 * of our target group.
1819 */
1820 if (ac->ac_allow_chain_relink &&
1821 (prev_group_bh) &&
7d1fe093 1822 (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
ccd979bd
MF		 1823 status = ocfs2_relink_block_group(handle, alloc_inode,
1824 ac->ac_bh, group_bh,
1825 prev_group_bh, chain);
1826 if (status < 0) {
1827 mlog_errno(status);
1828 goto bail;
1829 }
1830 }
1831
e49e2767
MF		 1832 if (ac->ac_find_loc_only)
1833 goto out_loc_only;
1834
d5134982
MF		 1835 status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1836 ac->ac_bh, res->sr_bits,
1837 chain);
1838 if (status) {
ccd979bd
MF		 1839 mlog_errno(status);
1840 goto bail;
1841 }
1842
ccd979bd
MF		 1843 status = ocfs2_block_group_set_bits(handle,
1844 alloc_inode,
1845 bg,
1846 group_bh,
7d1fe093
JB		 1847 res->sr_bit_offset,
1848 res->sr_bits);
ccd979bd
MF		 1849 if (status < 0) {
1850 mlog_errno(status);
1851 goto bail;
1852 }
1853
7d1fe093 1854 mlog(0, "Allocated %u bits from suballocator %llu\n", res->sr_bits,
1ca1a111 1855 (unsigned long long)le64_to_cpu(fe->i_blkno));
ccd979bd 1856
e49e2767 1857out_loc_only:
883d4cae 1858 *bits_left = le16_to_cpu(bg->bg_free_bits_count);
ccd979bd 1859bail:
a81cb88b
MF		 1860 brelse(group_bh);
1861 brelse(prev_group_bh);
ccd979bd
MF		 1862
1863 mlog_exit(status);
1864 return status;
1865}
1866
1867/* will give out up to bits_wanted contiguous bits. */
aa8f8e93 1868static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1fabe148 1869 handle_t *handle,
ccd979bd
MF		 1870 u32 bits_wanted,
1871 u32 min_bits,
7d1fe093 1872 struct ocfs2_suballoc_result *res)
ccd979bd
MF		 1873{
1874 int status;
1875 u16 victim, i;
883d4cae 1876 u16 bits_left = 0;
b2b6ebf5 1877 u64 hint = ac->ac_last_group;
ccd979bd
MF		 1878 struct ocfs2_chain_list *cl;
1879 struct ocfs2_dinode *fe;
1880
ccd979bd
MF		 1881 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1882 BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1883 BUG_ON(!ac->ac_bh);
1884
1885 fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
10995aa2
JB		 1886
1887 /* The bh was validated by the inode read during
1888 * ocfs2_reserve_suballoc_bits(). Any corruption is a code bug. */
1889 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1890
ccd979bd
MF		 1891 if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1892 le32_to_cpu(fe->id1.bitmap1.i_total)) {
aa8f8e93
JB		 1893 ocfs2_error(ac->ac_inode->i_sb,
1894 "Chain allocator dinode %llu has %u used "
b0697053
MF		 1895 "bits but only %u total.",
1896 (unsigned long long)le64_to_cpu(fe->i_blkno),
ccd979bd
MF		 1897 le32_to_cpu(fe->id1.bitmap1.i_used),
1898 le32_to_cpu(fe->id1.bitmap1.i_total));
1899 status = -EIO;
1900 goto bail;
1901 }
1902
b2b6ebf5 1903 res->sr_bg_blkno = hint;
7d1fe093 1904 if (res->sr_bg_blkno) {
883d4cae
MF		 1905 /* Attempt to short-circuit the usual search mechanism
1906 * by jumping straight to the most recently used
b595076a 1907 * allocation group. This helps us maintain some
883d4cae 1908 * contiguousness across allocations. */
da5cbf2f 1909 status = ocfs2_search_one_group(ac, handle, bits_wanted,
7d1fe093
JB		 1910 min_bits, res, &bits_left);
1911 if (!status)
883d4cae 1912 goto set_hint;
883d4cae
MF		 1913 if (status < 0 && status != -ENOSPC) {
1914 mlog_errno(status);
1915 goto bail;
1916 }
1917 }
1918
ccd979bd
MF		 1919 cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1920
1921 victim = ocfs2_find_victim_chain(cl);
1922 ac->ac_chain = victim;
1923 ac->ac_allow_chain_relink = 1;
1924
7d1fe093
JB		 1925 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1926 res, &bits_left);
b2b6ebf5
MF		 1927 if (!status) {
1928 hint = ocfs2_group_from_res(res);
883d4cae 1929 goto set_hint;
b2b6ebf5 1930 }
ccd979bd
MF		 1931 if (status < 0 && status != -ENOSPC) {
1932 mlog_errno(status);
1933 goto bail;
1934 }
1935
1936 mlog(0, "Search of victim chain %u came up with nothing, "
1937 "trying all chains now.\n", victim);
1938
1939 /* If we didn't pick a good victim, then just default to
1940 * searching each chain in order. Don't allow chain relinking
1941 * because we only calculate enough journal credits for one
1942 * relink per alloc. */
1943 ac->ac_allow_chain_relink = 0;
1944 for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1945 if (i == victim)
1946 continue;
1947 if (!cl->cl_recs[i].c_free)
1948 continue;
1949
1950 ac->ac_chain = i;
da5cbf2f 1951 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
7d1fe093 1952 res, &bits_left);
b2b6ebf5
MF		 1953 if (!status) {
1954 hint = ocfs2_group_from_res(res);
ccd979bd 1955 break;
b2b6ebf5 1956 }
ccd979bd
MF		 1957 if (status < 0 && status != -ENOSPC) {
1958 mlog_errno(status);
1959 goto bail;
1960 }
1961 }
ccd979bd 1962
883d4cae
MF		 1963set_hint:
1964 if (status != -ENOSPC) {
1965 /* If the next search of this group is not likely to
1966 * yield a suitable extent, then we reset the last
1967 * group hint so as to not waste a disk read */
1968 if (bits_left < min_bits)
1969 ac->ac_last_group = 0;
1970 else
b2b6ebf5 1971 ac->ac_last_group = hint;
883d4cae
MF		 1972 }
1973
1974bail:
ccd979bd
MF		 1975 mlog_exit(status);
1976 return status;
1977}
1978
1ed9b777 1979int ocfs2_claim_metadata(handle_t *handle,
ccd979bd
MF		 1980 struct ocfs2_alloc_context *ac,
1981 u32 bits_wanted,
2b6cb576 1982 u64 *suballoc_loc,
ccd979bd
MF		 1983 u16 *suballoc_bit_start,
1984 unsigned int *num_bits,
1985 u64 *blkno_start)
1986{
1987 int status;
ba206635 1988 struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
ccd979bd
MF		 1989
1990 BUG_ON(!ac);
1991 BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1992 BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
ccd979bd 1993
aa8f8e93 1994 status = ocfs2_claim_suballoc_bits(ac,
da5cbf2f 1995 handle,
ccd979bd
MF		 1996 bits_wanted,
1997 1,
7d1fe093 1998 &res);
ccd979bd
MF		 1999 if (status < 0) {
2000 mlog_errno(status);
2001 goto bail;
2002 }
1ed9b777 2003 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
ccd979bd 2004
2b6cb576 2005 *suballoc_loc = res.sr_bg_blkno;
7d1fe093 2006 *suballoc_bit_start = res.sr_bit_offset;
ba206635 2007 *blkno_start = res.sr_blkno;
7d1fe093
JB		 2008 ac->ac_bits_given += res.sr_bits;
2009 *num_bits = res.sr_bits;
ccd979bd
MF		 2010 status = 0;
2011bail:
2012 mlog_exit(status);
2013 return status;
2014}
2015
13821151 2016static void ocfs2_init_inode_ac_group(struct inode *dir,
abf1b3cb 2017 struct buffer_head *parent_di_bh,
13821151
TM		 2018 struct ocfs2_alloc_context *ac)
2019{
abf1b3cb 2020 struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
13821151
TM		 2021 /*
2022 * Try to allocate inodes from some specific group.
2023 *
2024 * If the parent dir has recorded the last group used in allocation,
2025 * cool, use it. Otherwise if we try to allocate new inode from the
2026 * same slot the parent dir belongs to, use the same chunk.
2027 *
2028 * We are very careful here to avoid the mistake of setting
2029 * ac_last_group to a group descriptor from a different (unlocked) slot.
2030 */
2031 if (OCFS2_I(dir)->ip_last_used_group &&
2032 OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2033 ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
abf1b3cb
TM		 2034 else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2035 if (di->i_suballoc_loc)
2036 ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2037 else
2038 ac->ac_last_group = ocfs2_which_suballoc_group(
2039 le64_to_cpu(di->i_blkno),
2040 le16_to_cpu(di->i_suballoc_bit));
2041 }
13821151
TM		 2042}
2043
2044static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2045 struct ocfs2_alloc_context *ac)
2046{
2047 OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2048 OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2049}
2050
e49e2767
MF		 2051int ocfs2_find_new_inode_loc(struct inode *dir,
2052 struct buffer_head *parent_fe_bh,
2053 struct ocfs2_alloc_context *ac,
2054 u64 *fe_blkno)
2055{
2056 int ret;
2057 handle_t *handle = NULL;
2058 struct ocfs2_suballoc_result *res;
2059
2060 BUG_ON(!ac);
2061 BUG_ON(ac->ac_bits_given != 0);
2062 BUG_ON(ac->ac_bits_wanted != 1);
2063 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2064
2065 res = kzalloc(sizeof(*res), GFP_NOFS);
2066 if (res == NULL) {
2067 ret = -ENOMEM;
2068 mlog_errno(ret);
2069 goto out;
2070 }
2071
2072 ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2073
2074 /*
2075 * The handle started here is for chain relink. Alternatively,
2076 * we could just disable relink for these calls.
2077 */
2078 handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2079 if (IS_ERR(handle)) {
2080 ret = PTR_ERR(handle);
2081 handle = NULL;
2082 mlog_errno(ret);
2083 goto out;
2084 }
2085
2086 /*
2087 * This will instruct ocfs2_claim_suballoc_bits and
2088 * ocfs2_search_one_group to search but save actual allocation
2089 * for later.
2090 */
2091 ac->ac_find_loc_only = 1;
2092
2093 ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2094 if (ret < 0) {
2095 mlog_errno(ret);
2096 goto out;
2097 }
2098
2099 ac->ac_find_loc_priv = res;
2100 *fe_blkno = res->sr_blkno;
2101
2102out:
2103 if (handle)
2104 ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2105
2106 if (ret)
2107 kfree(res);
2108
2109 return ret;
2110}
2111
2112int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2113 struct inode *dir,
2114 struct ocfs2_alloc_context *ac,
2115 u64 *suballoc_loc,
2116 u16 *suballoc_bit,
2117 u64 di_blkno)
2118{
2119 int ret;
2120 u16 chain;
2121 struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2122 struct buffer_head *bg_bh = NULL;
2123 struct ocfs2_group_desc *bg;
2124 struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2125
2126 /*
2127 * Since di_blkno is being passed back in, we check for any
2128 * inconsistencies which may have happened between
2129 * calls. These are code bugs as di_blkno is not expected to
2130 * change once returned from ocfs2_find_new_inode_loc()
2131 */
2132 BUG_ON(res->sr_blkno != di_blkno);
2133
2134 ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2135 res->sr_bg_stable_blkno, &bg_bh);
2136 if (ret) {
2137 mlog_errno(ret);
2138 goto out;
2139 }
2140
2141 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2142 chain = le16_to_cpu(bg->bg_chain);
2143
2144 ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2145 ac->ac_bh, res->sr_bits,
2146 chain);
2147 if (ret) {
2148 mlog_errno(ret);
2149 goto out;
2150 }
2151
2152 ret = ocfs2_block_group_set_bits(handle,
2153 ac->ac_inode,
2154 bg,
2155 bg_bh,
2156 res->sr_bit_offset,
2157 res->sr_bits);
2158 if (ret < 0) {
2159 mlog_errno(ret);
2160 goto out;
2161 }
2162
2163 mlog(0, "Allocated %u bits from suballocator %llu\n", res->sr_bits,
2164 (unsigned long long)di_blkno);
2165
2166 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2167
2168 BUG_ON(res->sr_bits != 1);
2169
2170 *suballoc_loc = res->sr_bg_blkno;
2171 *suballoc_bit = res->sr_bit_offset;
2172 ac->ac_bits_given++;
2173 ocfs2_save_inode_ac_group(dir, ac);
2174
2175out:
2176 brelse(bg_bh);
2177
2178 return ret;
2179}
2180
1ed9b777 2181int ocfs2_claim_new_inode(handle_t *handle,
13821151
TM		 2182 struct inode *dir,
2183 struct buffer_head *parent_fe_bh,
ccd979bd 2184 struct ocfs2_alloc_context *ac,
2b6cb576 2185 u64 *suballoc_loc,
ccd979bd
MF		 2186 u16 *suballoc_bit,
2187 u64 *fe_blkno)
2188{
2189 int status;
2b6cb576 2190 struct ocfs2_suballoc_result res;
ccd979bd 2191
ccd979bd
MF		 2192 BUG_ON(!ac);
2193 BUG_ON(ac->ac_bits_given != 0);
2194 BUG_ON(ac->ac_bits_wanted != 1);
2195 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
ccd979bd 2196
13821151
TM		 2197 ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2198
aa8f8e93 2199 status = ocfs2_claim_suballoc_bits(ac,
da5cbf2f 2200 handle,
ccd979bd
MF		 2201 1,
2202 1,
7d1fe093 2203 &res);
ccd979bd
MF		 2204 if (status < 0) {
2205 mlog_errno(status);
2206 goto bail;
2207 }
1ed9b777 2208 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
ccd979bd 2209
7d1fe093 2210 BUG_ON(res.sr_bits != 1);
ccd979bd 2211
2b6cb576 2212 *suballoc_loc = res.sr_bg_blkno;
7d1fe093 2213 *suballoc_bit = res.sr_bit_offset;
ba206635 2214 *fe_blkno = res.sr_blkno;
ccd979bd 2215 ac->ac_bits_given++;
13821151 2216 ocfs2_save_inode_ac_group(dir, ac);
ccd979bd
MF		 2217 status = 0;
2218bail:
2219 mlog_exit(status);
2220 return status;
2221}
2222
2223/* translate a group desc. blkno and it's bitmap offset into
2224 * disk cluster offset. */
2225static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2226 u64 bg_blkno,
2227 u16 bg_bit_off)
2228{
2229 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2230 u32 cluster = 0;
2231
2232 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2233
2234 if (bg_blkno != osb->first_cluster_group_blkno)
2235 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2236 cluster += (u32) bg_bit_off;
2237 return cluster;
2238}
2239
2240/* given a cluster offset, calculate which block group it belongs to
2241 * and return that block offset. */
d659072f 2242u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
ccd979bd
MF		 2243{
2244 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2245 u32 group_no;
2246
2247 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2248
2249 group_no = cluster / osb->bitmap_cpg;
2250 if (!group_no)
2251 return osb->first_cluster_group_blkno;
2252 return ocfs2_clusters_to_blocks(inode->i_sb,
2253 group_no * osb->bitmap_cpg);
2254}
2255
2256/* given the block number of a cluster start, calculate which cluster
2257 * group and descriptor bitmap offset that corresponds to. */
2258static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2259 u64 data_blkno,
2260 u64 *bg_blkno,
2261 u16 *bg_bit_off)
2262{
2263 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2264 u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2265
2266 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2267
2268 *bg_blkno = ocfs2_which_cluster_group(inode,
2269 data_cluster);
2270
2271 if (*bg_blkno == osb->first_cluster_group_blkno)
2272 *bg_bit_off = (u16) data_cluster;
2273 else
2274 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2275 data_blkno - *bg_blkno);
2276}
2277
2278/*
2279 * min_bits - minimum contiguous chunk from this total allocation we
2280 * can handle. set to what we asked for originally for a full
2281 * contig. allocation, set to '1' to indicate we can deal with extents
2282 * of any size.
2283 */
1ed9b777 2284int __ocfs2_claim_clusters(handle_t *handle,
415cb800
MF		 2285 struct ocfs2_alloc_context *ac,
2286 u32 min_clusters,
2287 u32 max_clusters,
2288 u32 *cluster_start,
2289 u32 *num_clusters)
ccd979bd
MF		 2290{
2291 int status;
415cb800 2292 unsigned int bits_wanted = max_clusters;
ba206635 2293 struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
1ed9b777 2294 struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
ccd979bd 2295
ccd979bd
MF		 2296 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2297
2298 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2299 && ac->ac_which != OCFS2_AC_USE_MAIN);
ccd979bd
MF		 2300
2301 if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
33d5d380
MF		 2302 WARN_ON(min_clusters > 1);
2303
ccd979bd
MF		 2304 status = ocfs2_claim_local_alloc_bits(osb,
2305 handle,
2306 ac,
2307 bits_wanted,
2308 cluster_start,
2309 num_clusters);
2310 if (!status)
2311 atomic_inc(&osb->alloc_stats.local_data);
2312 } else {
2313 if (min_clusters > (osb->bitmap_cpg - 1)) {
2314 /* The only paths asking for contiguousness
2315 * should know about this already. */
2fbe8d1e
SM		 2316 mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2317 "group bitmap size %u!\n", min_clusters,
2318 osb->bitmap_cpg);
ccd979bd
MF		 2319 status = -ENOSPC;
2320 goto bail;
2321 }
2322 /* clamp the current request down to a realistic size. */
2323 if (bits_wanted > (osb->bitmap_cpg - 1))
2324 bits_wanted = osb->bitmap_cpg - 1;
2325
aa8f8e93 2326 status = ocfs2_claim_suballoc_bits(ac,
da5cbf2f 2327 handle,
ccd979bd
MF		 2328 bits_wanted,
2329 min_clusters,
7d1fe093 2330 &res);
ccd979bd 2331 if (!status) {
ba206635 2332 BUG_ON(res.sr_blkno); /* cluster alloc can't set */
ccd979bd
MF		 2333 *cluster_start =
2334 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
7d1fe093
JB		 2335 res.sr_bg_blkno,
2336 res.sr_bit_offset);
ccd979bd 2337 atomic_inc(&osb->alloc_stats.bitmap_data);
4711954e 2338 *num_clusters = res.sr_bits;
ccd979bd
MF		 2339 }
2340 }
2341 if (status < 0) {
2342 if (status != -ENOSPC)
2343 mlog_errno(status);
2344 goto bail;
2345 }
2346
4711954e 2347 ac->ac_bits_given += *num_clusters;
ccd979bd
MF		 2348
2349bail:
2350 mlog_exit(status);
2351 return status;
2352}
2353
1ed9b777 2354int ocfs2_claim_clusters(handle_t *handle,
415cb800
MF		 2355 struct ocfs2_alloc_context *ac,
2356 u32 min_clusters,
2357 u32 *cluster_start,
2358 u32 *num_clusters)
2359{
2360 unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2361
1ed9b777 2362 return __ocfs2_claim_clusters(handle, ac, min_clusters,
415cb800
MF		 2363 bits_wanted, cluster_start, num_clusters);
2364}
2365
b4414eea
MF		 2366static int ocfs2_block_group_clear_bits(handle_t *handle,
2367 struct inode *alloc_inode,
2368 struct ocfs2_group_desc *bg,
2369 struct buffer_head *group_bh,
2370 unsigned int bit_off,
2371 unsigned int num_bits,
2372 void (*undo_fn)(unsigned int bit,
2373 unsigned long *bmap))
ccd979bd
MF		 2374{
2375 int status;
2376 unsigned int tmp;
ccd979bd
MF		 2377 struct ocfs2_group_desc *undo_bg = NULL;
2378
42035306
JB		 2379 /* The caller got this descriptor from
2380 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
2381 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
ccd979bd
MF		 2382
2383 mlog(0, "off = %u, num = %u\n", bit_off, num_bits);
2384
b4414eea 2385 BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
0cf2f763 2386 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
b4414eea
MF		 2387 group_bh,
2388 undo_fn ?
2389 OCFS2_JOURNAL_ACCESS_UNDO :
2390 OCFS2_JOURNAL_ACCESS_WRITE);
ccd979bd
MF		 2391 if (status < 0) {
2392 mlog_errno(status);
2393 goto bail;
2394 }
2395
b4414eea 2396 if (undo_fn) {
94e41ecf
SM		 2397 jbd_lock_bh_state(group_bh);
2398 undo_bg = (struct ocfs2_group_desc *)
2399 bh2jh(group_bh)->b_committed_data;
2400 BUG_ON(!undo_bg);
2401 }
ccd979bd
MF		 2402
2403 tmp = num_bits;
2404 while(tmp--) {
2405 ocfs2_clear_bit((bit_off + tmp),
2406 (unsigned long *) bg->bg_bitmap);
b4414eea
MF		 2407 if (undo_fn)
2408 undo_fn(bit_off + tmp,
2409 (unsigned long *) undo_bg->bg_bitmap);
ccd979bd
MF		 2410 }
2411 le16_add_cpu(&bg->bg_free_bits_count, num_bits);
9b5cd10e
SE		 2412 if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2413 ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
2414 " count %u but claims %u are freed. num_bits %d",
2415 (unsigned long long)le64_to_cpu(bg->bg_blkno),
2416 le16_to_cpu(bg->bg_bits),
2417 le16_to_cpu(bg->bg_free_bits_count), num_bits);
2418 return -EROFS;
2419 }
ccd979bd 2420
b4414eea 2421 if (undo_fn)
94e41ecf
SM		 2422 jbd_unlock_bh_state(group_bh);
2423
ec20cec7 2424 ocfs2_journal_dirty(handle, group_bh);
ccd979bd
MF		 2425bail:
2426 return status;
2427}
2428
2429/*
2430 * expects the suballoc inode to already be locked.
2431 */
b4414eea
MF		 2432static int _ocfs2_free_suballoc_bits(handle_t *handle,
2433 struct inode *alloc_inode,
2434 struct buffer_head *alloc_bh,
2435 unsigned int start_bit,
2436 u64 bg_blkno,
2437 unsigned int count,
2438 void (*undo_fn)(unsigned int bit,
2439 unsigned long *bitmap))
ccd979bd
MF		 2440{
2441 int status = 0;
2442 u32 tmp_used;
ccd979bd
MF		 2443 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2444 struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2445 struct buffer_head *group_bh = NULL;
2446 struct ocfs2_group_desc *group;
2447
10995aa2
JB		 2448 /* The alloc_bh comes from ocfs2_free_dinode() or
2449 * ocfs2_free_clusters(). The callers have all locked the
2450 * allocator and gotten alloc_bh from the lock call. This
2451 * validates the dinode buffer. Any corruption that has happended
2452 * is a code bug. */
2453 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
ccd979bd
MF		 2454 BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2455
b0697053
MF		 2456 mlog(0, "%llu: freeing %u bits from group %llu, starting at %u\n",
2457 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno, count,
2458 (unsigned long long)bg_blkno, start_bit);
ccd979bd 2459
68f64d47
JB		 2460 status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2461 &group_bh);
ccd979bd
MF		 2462 if (status < 0) {
2463 mlog_errno(status);
2464 goto bail;
2465 }
ccd979bd 2466 group = (struct ocfs2_group_desc *) group_bh->b_data;
68f64d47 2467
ccd979bd
MF		 2468 BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2469
2470 status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2471 group, group_bh,
b4414eea 2472 start_bit, count, undo_fn);
ccd979bd
MF		 2473 if (status < 0) {
2474 mlog_errno(status);
2475 goto bail;
2476 }
2477
0cf2f763
JB		 2478 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2479 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
ccd979bd
MF		 2480 if (status < 0) {
2481 mlog_errno(status);
2482 goto bail;
2483 }
2484
2485 le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2486 count);
2487 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2488 fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
ec20cec7 2489 ocfs2_journal_dirty(handle, alloc_bh);
ccd979bd
MF		 2490
2491bail:
a81cb88b 2492 brelse(group_bh);
ccd979bd
MF		 2493
2494 mlog_exit(status);
2495 return status;
2496}
2497
b4414eea
MF		 2498int ocfs2_free_suballoc_bits(handle_t *handle,
2499 struct inode *alloc_inode,
2500 struct buffer_head *alloc_bh,
2501 unsigned int start_bit,
2502 u64 bg_blkno,
2503 unsigned int count)
2504{
2505 return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2506 start_bit, bg_blkno, count, NULL);
2507}
2508
1fabe148 2509int ocfs2_free_dinode(handle_t *handle,
ccd979bd
MF		 2510 struct inode *inode_alloc_inode,
2511 struct buffer_head *inode_alloc_bh,
2512 struct ocfs2_dinode *di)
2513{
2514 u64 blk = le64_to_cpu(di->i_blkno);
2515 u16 bit = le16_to_cpu(di->i_suballoc_bit);
2516 u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2517
74380c47
TM		 2518 if (di->i_suballoc_loc)
2519 bg_blkno = le64_to_cpu(di->i_suballoc_loc);
ccd979bd
MF		 2520 return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2521 inode_alloc_bh, bit, bg_blkno, 1);
2522}
2523
b4414eea
MF		 2524static int _ocfs2_free_clusters(handle_t *handle,
2525 struct inode *bitmap_inode,
2526 struct buffer_head *bitmap_bh,
2527 u64 start_blk,
2528 unsigned int num_clusters,
2529 void (*undo_fn)(unsigned int bit,
2530 unsigned long *bitmap))
ccd979bd
MF		 2531{
2532 int status;
2533 u16 bg_start_bit;
2534 u64 bg_blkno;
2535 struct ocfs2_dinode *fe;
2536
2537 /* You can't ever have a contiguous set of clusters
2538 * bigger than a block group bitmap so we never have to worry
ef6b689b
TM		 2539 * about looping on them.
2540 * This is expensive. We can safely remove once this stuff has
ccd979bd
MF		 2541 * gotten tested really well. */
2542 BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
2543
2544 fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
2545
2546 ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2547 &bg_start_bit);
2548
b0697053
MF		 2549 mlog(0, "want to free %u clusters starting at block %llu\n",
2550 num_clusters, (unsigned long long)start_blk);
2551 mlog(0, "bg_blkno = %llu, bg_start_bit = %u\n",
2552 (unsigned long long)bg_blkno, bg_start_bit);
ccd979bd 2553
b4414eea
MF		 2554 status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2555 bg_start_bit, bg_blkno,
2556 num_clusters, undo_fn);
9c7af40b 2557 if (status < 0) {
ccd979bd 2558 mlog_errno(status);
9c7af40b
MF		 2559 goto out;
2560 }
2561
2562 ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2563 num_clusters);
ccd979bd 2564
9c7af40b 2565out:
ccd979bd
MF		 2566 mlog_exit(status);
2567 return status;
2568}
2569
b4414eea
MF		 2570int ocfs2_free_clusters(handle_t *handle,
2571 struct inode *bitmap_inode,
2572 struct buffer_head *bitmap_bh,
2573 u64 start_blk,
2574 unsigned int num_clusters)
2575{
2576 return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2577 start_blk, num_clusters,
2578 _ocfs2_set_bit);
2579}
2580
2581/*
2582 * Give never-used clusters back to the global bitmap. We don't need
2583 * to protect these bits in the undo buffer.
2584 */
2585int ocfs2_release_clusters(handle_t *handle,
2586 struct inode *bitmap_inode,
2587 struct buffer_head *bitmap_bh,
2588 u64 start_blk,
2589 unsigned int num_clusters)
2590{
2591 return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2592 start_blk, num_clusters,
2593 _ocfs2_clear_bit);
2594}
2595
ccd979bd
MF		 2596static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
2597{
2598 printk("Block Group:\n");
2599 printk("bg_signature: %s\n", bg->bg_signature);
2600 printk("bg_size: %u\n", bg->bg_size);
2601 printk("bg_bits: %u\n", bg->bg_bits);
2602 printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
2603 printk("bg_chain: %u\n", bg->bg_chain);
2604 printk("bg_generation: %u\n", le32_to_cpu(bg->bg_generation));
b0697053
MF		 2605 printk("bg_next_group: %llu\n",
2606 (unsigned long long)bg->bg_next_group);
2607 printk("bg_parent_dinode: %llu\n",
2608 (unsigned long long)bg->bg_parent_dinode);
2609 printk("bg_blkno: %llu\n",
2610 (unsigned long long)bg->bg_blkno);
ccd979bd
MF		 2611}
2612
2613static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2614{
2615 int i;
2616
b0697053 2617 printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
ccd979bd 2618 printk("i_signature: %s\n", fe->i_signature);
b0697053
MF		 2619 printk("i_size: %llu\n",
2620 (unsigned long long)fe->i_size);
ccd979bd
MF		 2621 printk("i_clusters: %u\n", fe->i_clusters);
2622 printk("i_generation: %u\n",
2623 le32_to_cpu(fe->i_generation));
2624 printk("id1.bitmap1.i_used: %u\n",
2625 le32_to_cpu(fe->id1.bitmap1.i_used));
2626 printk("id1.bitmap1.i_total: %u\n",
2627 le32_to_cpu(fe->id1.bitmap1.i_total));
2628 printk("id2.i_chain.cl_cpg: %u\n", fe->id2.i_chain.cl_cpg);
2629 printk("id2.i_chain.cl_bpc: %u\n", fe->id2.i_chain.cl_bpc);
2630 printk("id2.i_chain.cl_count: %u\n", fe->id2.i_chain.cl_count);
2631 printk("id2.i_chain.cl_next_free_rec: %u\n",
2632 fe->id2.i_chain.cl_next_free_rec);
2633 for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2634 printk("fe->id2.i_chain.cl_recs[%d].c_free: %u\n", i,
2635 fe->id2.i_chain.cl_recs[i].c_free);
2636 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2637 fe->id2.i_chain.cl_recs[i].c_total);
b0697053
MF		 2638 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2639 (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
ccd979bd
MF		 2640 }
2641}
e7d4cb6b
TM		 2642
2643/*
2644 * For a given allocation, determine which allocators will need to be
2645 * accessed, and lock them, reserving the appropriate number of bits.
2646 *
2647 * Sparse file systems call this from ocfs2_write_begin_nolock()
2648 * and ocfs2_allocate_unwritten_extents().
2649 *
2650 * File systems which don't support holes call this from
2651 * ocfs2_extend_allocation().
2652 */
f99b9b7c
JB		 2653int ocfs2_lock_allocators(struct inode *inode,
2654 struct ocfs2_extent_tree *et,
e7d4cb6b
TM		 2655 u32 clusters_to_add, u32 extents_to_split,
2656 struct ocfs2_alloc_context **data_ac,
f99b9b7c 2657 struct ocfs2_alloc_context **meta_ac)
e7d4cb6b
TM		 2658{
2659 int ret = 0, num_free_extents;
2660 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2661 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2662
2663 *meta_ac = NULL;
2664 if (data_ac)
2665 *data_ac = NULL;
2666
2667 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2668
3d03a305 2669 num_free_extents = ocfs2_num_free_extents(osb, et);
e7d4cb6b
TM		 2670 if (num_free_extents < 0) {
2671 ret = num_free_extents;
2672 mlog_errno(ret);
2673 goto out;
2674 }
2675
2676 /*
2677 * Sparse allocation file systems need to be more conservative
2678 * with reserving room for expansion - the actual allocation
2679 * happens while we've got a journal handle open so re-taking
2680 * a cluster lock (because we ran out of room for another
2681 * extent) will violate ordering rules.
2682 *
2683 * Most of the time we'll only be seeing this 1 cluster at a time
2684 * anyway.
2685 *
2686 * Always lock for any unwritten extents - we might want to
2687 * add blocks during a split.
2688 */
2689 if (!num_free_extents ||
2690 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
f99b9b7c 2691 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
e7d4cb6b
TM		 2692 if (ret < 0) {
2693 if (ret != -ENOSPC)
2694 mlog_errno(ret);
2695 goto out;
2696 }
2697 }
2698
2699 if (clusters_to_add == 0)
2700 goto out;
2701
2702 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2703 if (ret < 0) {
2704 if (ret != -ENOSPC)
2705 mlog_errno(ret);
2706 goto out;
2707 }
2708
2709out:
2710 if (ret) {
2711 if (*meta_ac) {
2712 ocfs2_free_alloc_context(*meta_ac);
2713 *meta_ac = NULL;
2714 }
2715
2716 /*
2717 * We cannot have an error and a non null *data_ac.
2718 */
2719 }
2720
2721 return ret;
2722}
6ca497a8 2723
2724/*
2725 * Read the inode specified by blkno to get suballoc_slot and
2726 * suballoc_bit.
2727 */
2728static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
889f004a
TM		 2729 u16 *suballoc_slot, u64 *group_blkno,
2730 u16 *suballoc_bit)
6ca497a8 2731{
2732 int status;
2733 struct buffer_head *inode_bh = NULL;
2734 struct ocfs2_dinode *inode_fe;
2735
ef6b689b 2736 mlog(0, "blkno: %llu\n", (unsigned long long)blkno);
6ca497a8 2737
2738 /* dirty read disk */
2739 status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2740 if (status < 0) {
5b09b507
JB		 2741 mlog(ML_ERROR, "read block %llu failed %d\n",
2742 (unsigned long long)blkno, status);
6ca497a8 2743 goto bail;
2744 }
2745
2746 inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2747 if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
5b09b507
JB		 2748 mlog(ML_ERROR, "invalid inode %llu requested\n",
2749 (unsigned long long)blkno);
6ca497a8 2750 status = -EINVAL;
2751 goto bail;
2752 }
2753
0fba8137 2754 if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
6ca497a8 2755 (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2756 mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
5b09b507
JB		 2757 (unsigned long long)blkno,
2758 (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
6ca497a8 2759 status = -EINVAL;
2760 goto bail;
2761 }
2762
2763 if (suballoc_slot)
2764 *suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2765 if (suballoc_bit)
2766 *suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
889f004a
TM		 2767 if (group_blkno)
2768 *group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
6ca497a8 2769
2770bail:
2771 brelse(inode_bh);
2772
2773 mlog_exit(status);
2774 return status;
2775}
2776
2777/*
2778 * test whether bit is SET in allocator bitmap or not. on success, 0
2779 * is returned and *res is 1 for SET; 0 otherwise. when fails, errno
2780 * is returned and *res is meaningless. Call this after you have
2781 * cluster locked against suballoc, or you may get a result based on
2782 * non-up2date contents
2783 */
2784static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2785 struct inode *suballoc,
889f004a
TM		 2786 struct buffer_head *alloc_bh,
2787 u64 group_blkno, u64 blkno,
6ca497a8 2788 u16 bit, int *res)
2789{
abf1b3cb 2790 struct ocfs2_dinode *alloc_di;
6ca497a8 2791 struct ocfs2_group_desc *group;
2792 struct buffer_head *group_bh = NULL;
2793 u64 bg_blkno;
2794 int status;
2795
ef6b689b
TM		 2796 mlog(0, "blkno: %llu bit: %u\n", (unsigned long long)blkno,
2797 (unsigned int)bit);
6ca497a8 2798
abf1b3cb
TM		 2799 alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2800 if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
6ca497a8 2801 mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2802 (unsigned int)bit,
abf1b3cb 2803 ocfs2_bits_per_group(&alloc_di->id2.i_chain));
6ca497a8 2804 status = -EINVAL;
2805 goto bail;
2806 }
2807
889f004a
TM		 2808 bg_blkno = group_blkno ? group_blkno :
2809 ocfs2_which_suballoc_group(blkno, bit);
abf1b3cb 2810 status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
6ca497a8 2811 &group_bh);
2812 if (status < 0) {
5b09b507
JB		 2813 mlog(ML_ERROR, "read group %llu failed %d\n",
2814 (unsigned long long)bg_blkno, status);
6ca497a8 2815 goto bail;
2816 }
2817
2818 group = (struct ocfs2_group_desc *) group_bh->b_data;
2819 *res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2820
2821bail:
2822 brelse(group_bh);
2823
2824 mlog_exit(status);
2825 return status;
2826}
2827
2828/*
2829 * Test if the bit representing this inode (blkno) is set in the
2830 * suballocator.
2831 *
2832 * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2833 *
2834 * In the event of failure, a negative value is returned and *res is
2835 * meaningless.
2836 *
2837 * Callers must make sure to hold nfs_sync_lock to prevent
2838 * ocfs2_delete_inode() on another node from accessing the same
2839 * suballocator concurrently.
2840 */
2841int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2842{
2843 int status;
889f004a 2844 u64 group_blkno = 0;
6ca497a8 2845 u16 suballoc_bit = 0, suballoc_slot = 0;
2846 struct inode *inode_alloc_inode;
2847 struct buffer_head *alloc_bh = NULL;
2848
ef6b689b 2849 mlog(0, "blkno: %llu", (unsigned long long)blkno);
6ca497a8 2850
2851 status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
889f004a 2852 &group_blkno, &suballoc_bit);
6ca497a8 2853 if (status < 0) {
2854 mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2855 goto bail;
2856 }
2857
2858 inode_alloc_inode =
2859 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2860 suballoc_slot);
2861 if (!inode_alloc_inode) {
2862 /* the error code could be inaccurate, but we are not able to
2863 * get the correct one. */
2864 status = -EINVAL;
2865 mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2866 (u32)suballoc_slot);
2867 goto bail;
2868 }
2869
2870 mutex_lock(&inode_alloc_inode->i_mutex);
2871 status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2872 if (status < 0) {
2873 mutex_unlock(&inode_alloc_inode->i_mutex);
2874 mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2875 (u32)suballoc_slot, status);
2876 goto bail;
2877 }
2878
2879 status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
889f004a 2880 group_blkno, blkno, suballoc_bit, res);
6ca497a8 2881 if (status < 0)
2882 mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2883
2884 ocfs2_inode_unlock(inode_alloc_inode, 0);
2885 mutex_unlock(&inode_alloc_inode->i_mutex);
2886
2887 iput(inode_alloc_inode);
2888 brelse(alloc_bh);
2889bail:
2890 mlog_exit(status);
2891 return status;
2892}
Linux kernel - Deliverables
This page took 0.663374 seconds and 5 git commands to generate.