]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/ocfs2/suballoc.c
ocfs2: flock: drop cross-node lock when failed locally
[mirror_ubuntu-artful-kernel.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
ccd979bd
MF
32#include <cluster/masklog.h>
33
34#include "ocfs2.h"
35
36#include "alloc.h"
d6b32bbb 37#include "blockcheck.h"
ccd979bd
MF
38#include "dlmglue.h"
39#include "inode.h"
40#include "journal.h"
41#include "localalloc.h"
42#include "suballoc.h"
43#include "super.h"
44#include "sysfile.h"
45#include "uptodate.h"
2f73e135 46#include "ocfs2_trace.h"
ccd979bd
MF
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 int ocfs2_relink_block_group(handle_t *handle,
ccd979bd
MF
117 struct inode *alloc_inode,
118 struct buffer_head *fe_bh,
119 struct buffer_head *bg_bh,
120 struct buffer_head *prev_bg_bh,
121 u16 chain);
122static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
123 u32 wanted);
ccd979bd
MF
124static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
125 u64 bg_blkno,
126 u16 bg_bit_off);
ccd979bd
MF
127static inline void ocfs2_block_to_cluster_group(struct inode *inode,
128 u64 data_blkno,
129 u64 *bg_blkno,
130 u16 *bg_bit_off);
1187c968
JB
131static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
132 u32 bits_wanted, u64 max_block,
60ca81e8 133 int flags,
1187c968 134 struct ocfs2_alloc_context **ac);
ccd979bd 135
9c7af40b 136void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
ccd979bd 137{
da5cbf2f
MF
138 struct inode *inode = ac->ac_inode;
139
140 if (inode) {
141 if (ac->ac_which != OCFS2_AC_USE_LOCAL)
e63aecb6 142 ocfs2_inode_unlock(inode, 1);
da5cbf2f
MF
143
144 mutex_unlock(&inode->i_mutex);
145
146 iput(inode);
4d0ddb2c 147 ac->ac_inode = NULL;
da5cbf2f 148 }
a81cb88b
MF
149 brelse(ac->ac_bh);
150 ac->ac_bh = NULL;
e3b4a97d 151 ac->ac_resv = NULL;
e49e2767
MF
152 if (ac->ac_find_loc_priv) {
153 kfree(ac->ac_find_loc_priv);
154 ac->ac_find_loc_priv = NULL;
155 }
4d0ddb2c
TM
156}
157
158void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
159{
160 ocfs2_free_ac_resource(ac);
ccd979bd
MF
161 kfree(ac);
162}
163
164static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
165{
166 return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
167}
168
57e3e797
JB
169#define do_error(fmt, ...) \
170 do{ \
78c37eb0 171 if (resize) \
57e3e797
JB
172 mlog(ML_ERROR, fmt "\n", ##__VA_ARGS__); \
173 else \
174 ocfs2_error(sb, fmt, ##__VA_ARGS__); \
175 } while (0)
176
970e4936
JB
177static int ocfs2_validate_gd_self(struct super_block *sb,
178 struct buffer_head *bh,
78c37eb0 179 int resize)
970e4936
JB
180{
181 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
182
7bf72ede 183 if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
68f64d47
JB
184 do_error("Group descriptor #%llu has bad signature %.*s",
185 (unsigned long long)bh->b_blocknr, 7,
57e3e797
JB
186 gd->bg_signature);
187 return -EINVAL;
7bf72ede
MF
188 }
189
68f64d47
JB
190 if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
191 do_error("Group descriptor #%llu has an invalid bg_blkno "
192 "of %llu",
193 (unsigned long long)bh->b_blocknr,
194 (unsigned long long)le64_to_cpu(gd->bg_blkno));
195 return -EINVAL;
196 }
197
198 if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
199 do_error("Group descriptor #%llu has an invalid "
200 "fs_generation of #%u",
201 (unsigned long long)bh->b_blocknr,
202 le32_to_cpu(gd->bg_generation));
203 return -EINVAL;
204 }
205
970e4936
JB
206 if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
207 do_error("Group descriptor #%llu has bit count %u but "
208 "claims that %u are free",
209 (unsigned long long)bh->b_blocknr,
210 le16_to_cpu(gd->bg_bits),
211 le16_to_cpu(gd->bg_free_bits_count));
212 return -EINVAL;
213 }
214
215 if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
216 do_error("Group descriptor #%llu has bit count %u but "
217 "max bitmap bits of %u",
218 (unsigned long long)bh->b_blocknr,
219 le16_to_cpu(gd->bg_bits),
220 8 * le16_to_cpu(gd->bg_size));
221 return -EINVAL;
222 }
223
224 return 0;
225}
226
227static int ocfs2_validate_gd_parent(struct super_block *sb,
228 struct ocfs2_dinode *di,
229 struct buffer_head *bh,
78c37eb0 230 int resize)
970e4936
JB
231{
232 unsigned int max_bits;
233 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
234
7bf72ede 235 if (di->i_blkno != gd->bg_parent_dinode) {
68f64d47 236 do_error("Group descriptor #%llu has bad parent "
57e3e797 237 "pointer (%llu, expected %llu)",
68f64d47 238 (unsigned long long)bh->b_blocknr,
57e3e797
JB
239 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
240 (unsigned long long)le64_to_cpu(di->i_blkno));
241 return -EINVAL;
7bf72ede
MF
242 }
243
244 max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
245 if (le16_to_cpu(gd->bg_bits) > max_bits) {
68f64d47
JB
246 do_error("Group descriptor #%llu has bit count of %u",
247 (unsigned long long)bh->b_blocknr,
57e3e797
JB
248 le16_to_cpu(gd->bg_bits));
249 return -EINVAL;
7bf72ede
MF
250 }
251
78c37eb0
TM
252 /* In resize, we may meet the case bg_chain == cl_next_free_rec. */
253 if ((le16_to_cpu(gd->bg_chain) >
254 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
255 ((le16_to_cpu(gd->bg_chain) ==
256 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
68f64d47
JB
257 do_error("Group descriptor #%llu has bad chain %u",
258 (unsigned long long)bh->b_blocknr,
57e3e797
JB
259 le16_to_cpu(gd->bg_chain));
260 return -EINVAL;
7bf72ede
MF
261 }
262
970e4936
JB
263 return 0;
264}
7bf72ede 265
57e3e797 266#undef do_error
7bf72ede 267
970e4936
JB
268/*
269 * This version only prints errors. It does not fail the filesystem, and
270 * exists only for resize.
271 */
272int ocfs2_check_group_descriptor(struct super_block *sb,
273 struct ocfs2_dinode *di,
274 struct buffer_head *bh)
275{
276 int rc;
d6b32bbb
JB
277 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
278
279 BUG_ON(!buffer_uptodate(bh));
970e4936 280
d6b32bbb
JB
281 /*
282 * If the ecc fails, we return the error but otherwise
283 * leave the filesystem running. We know any error is
284 * local to this block.
285 */
286 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
13723d00
JB
287 if (rc) {
288 mlog(ML_ERROR,
289 "Checksum failed for group descriptor %llu\n",
290 (unsigned long long)bh->b_blocknr);
291 } else
d6b32bbb 292 rc = ocfs2_validate_gd_self(sb, bh, 1);
970e4936
JB
293 if (!rc)
294 rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
295
296 return rc;
297}
298
299static int ocfs2_validate_group_descriptor(struct super_block *sb,
300 struct buffer_head *bh)
301{
d6b32bbb
JB
302 int rc;
303 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
304
2f73e135
TM
305 trace_ocfs2_validate_group_descriptor(
306 (unsigned long long)bh->b_blocknr);
970e4936 307
d6b32bbb
JB
308 BUG_ON(!buffer_uptodate(bh));
309
310 /*
311 * If the ecc fails, we return the error but otherwise
312 * leave the filesystem running. We know any error is
313 * local to this block.
314 */
315 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
316 if (rc)
317 return rc;
318
319 /*
320 * Errors after here are fatal.
321 */
322
970e4936 323 return ocfs2_validate_gd_self(sb, bh, 0);
7bf72ede
MF
324}
325
68f64d47
JB
326int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
327 u64 gd_blkno, struct buffer_head **bh)
328{
329 int rc;
330 struct buffer_head *tmp = *bh;
331
8cb471e8 332 rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
970e4936 333 ocfs2_validate_group_descriptor);
68f64d47
JB
334 if (rc)
335 goto out;
336
970e4936 337 rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
68f64d47
JB
338 if (rc) {
339 brelse(tmp);
340 goto out;
341 }
342
343 /* If ocfs2_read_block() got us a new bh, pass it up. */
344 if (!*bh)
345 *bh = tmp;
346
347out:
348 return rc;
349}
350
798db35f
JB
351static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
352 struct ocfs2_group_desc *bg,
353 struct ocfs2_chain_list *cl,
47dea423 354 u64 p_blkno, unsigned int clusters)
798db35f
JB
355{
356 struct ocfs2_extent_list *el = &bg->bg_list;
357 struct ocfs2_extent_rec *rec;
358
4711954e 359 BUG_ON(!ocfs2_supports_discontig_bg(osb));
798db35f
JB
360 if (!el->l_next_free_rec)
361 el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
362 rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
4711954e 363 rec->e_blkno = cpu_to_le64(p_blkno);
798db35f
JB
364 rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
365 le16_to_cpu(cl->cl_bpc));
47dea423 366 rec->e_leaf_clusters = cpu_to_le16(clusters);
798db35f 367 le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
4711954e
TM
368 le16_add_cpu(&bg->bg_free_bits_count,
369 clusters * le16_to_cpu(cl->cl_bpc));
798db35f
JB
370 le16_add_cpu(&el->l_next_free_rec, 1);
371}
372
1fabe148 373static int ocfs2_block_group_fill(handle_t *handle,
ccd979bd
MF
374 struct inode *alloc_inode,
375 struct buffer_head *bg_bh,
376 u64 group_blkno,
798db35f 377 unsigned int group_clusters,
ccd979bd
MF
378 u16 my_chain,
379 struct ocfs2_chain_list *cl)
380{
381 int status = 0;
798db35f 382 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
ccd979bd
MF
383 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
384 struct super_block * sb = alloc_inode->i_sb;
385
ccd979bd 386 if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
b0697053
MF
387 ocfs2_error(alloc_inode->i_sb, "group block (%llu) != "
388 "b_blocknr (%llu)",
389 (unsigned long long)group_blkno,
ccd979bd
MF
390 (unsigned long long) bg_bh->b_blocknr);
391 status = -EIO;
392 goto bail;
393 }
394
13723d00 395 status = ocfs2_journal_access_gd(handle,
0cf2f763 396 INODE_CACHE(alloc_inode),
13723d00
JB
397 bg_bh,
398 OCFS2_JOURNAL_ACCESS_CREATE);
ccd979bd
MF
399 if (status < 0) {
400 mlog_errno(status);
401 goto bail;
402 }
403
404 memset(bg, 0, sb->s_blocksize);
405 strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
406 bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
8571882c
TM
407 bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
408 osb->s_feature_incompat));
ccd979bd
MF
409 bg->bg_chain = cpu_to_le16(my_chain);
410 bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
411 bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
412 bg->bg_blkno = cpu_to_le64(group_blkno);
798db35f
JB
413 if (group_clusters == le16_to_cpu(cl->cl_cpg))
414 bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
415 else
4711954e 416 ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
798db35f
JB
417 group_clusters);
418
ccd979bd
MF
419 /* set the 1st bit in the bitmap to account for the descriptor block */
420 ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
421 bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
422
ec20cec7 423 ocfs2_journal_dirty(handle, bg_bh);
ccd979bd
MF
424
425 /* There is no need to zero out or otherwise initialize the
426 * other blocks in a group - All valid FS metadata in a block
427 * group stores the superblock fs_generation value at
428 * allocation time. */
429
430bail:
c1e8d35e
TM
431 if (status)
432 mlog_errno(status);
ccd979bd
MF
433 return status;
434}
435
436static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
437{
438 u16 curr, best;
439
440 best = curr = 0;
441 while (curr < le16_to_cpu(cl->cl_count)) {
442 if (le32_to_cpu(cl->cl_recs[best].c_total) >
443 le32_to_cpu(cl->cl_recs[curr].c_total))
444 best = curr;
445 curr++;
446 }
447 return best;
448}
449
798db35f
JB
450static struct buffer_head *
451ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
452 struct inode *alloc_inode,
453 struct ocfs2_alloc_context *ac,
454 struct ocfs2_chain_list *cl)
455{
456 int status;
457 u32 bit_off, num_bits;
458 u64 bg_blkno;
459 struct buffer_head *bg_bh;
460 unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
461
1ed9b777 462 status = ocfs2_claim_clusters(handle, ac,
798db35f
JB
463 le16_to_cpu(cl->cl_cpg), &bit_off,
464 &num_bits);
465 if (status < 0) {
466 if (status != -ENOSPC)
467 mlog_errno(status);
468 goto bail;
469 }
470
471 /* setup the group */
472 bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
2f73e135
TM
473 trace_ocfs2_block_group_alloc_contig(
474 (unsigned long long)bg_blkno, alloc_rec);
798db35f
JB
475
476 bg_bh = sb_getblk(osb->sb, bg_blkno);
477 if (!bg_bh) {
7391a294 478 status = -ENOMEM;
798db35f
JB
479 mlog_errno(status);
480 goto bail;
481 }
482 ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
483
484 status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
485 bg_blkno, num_bits, alloc_rec, cl);
486 if (status < 0) {
487 brelse(bg_bh);
488 mlog_errno(status);
489 }
490
491bail:
492 return status ? ERR_PTR(status) : bg_bh;
493}
494
495static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
496 handle_t *handle,
497 struct ocfs2_alloc_context *ac,
498 unsigned int min_bits,
499 u32 *bit_off, u32 *num_bits)
500{
18d3a98f 501 int status = 0;
798db35f
JB
502
503 while (min_bits) {
1ed9b777 504 status = ocfs2_claim_clusters(handle, ac, min_bits,
798db35f
JB
505 bit_off, num_bits);
506 if (status != -ENOSPC)
507 break;
508
509 min_bits >>= 1;
510 }
511
512 return status;
513}
514
515static int ocfs2_block_group_grow_discontig(handle_t *handle,
516 struct inode *alloc_inode,
517 struct buffer_head *bg_bh,
518 struct ocfs2_alloc_context *ac,
519 struct ocfs2_chain_list *cl,
520 unsigned int min_bits)
521{
522 int status;
523 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
524 struct ocfs2_group_desc *bg =
525 (struct ocfs2_group_desc *)bg_bh->b_data;
4711954e
TM
526 unsigned int needed = le16_to_cpu(cl->cl_cpg) -
527 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
798db35f
JB
528 u32 p_cpos, clusters;
529 u64 p_blkno;
530 struct ocfs2_extent_list *el = &bg->bg_list;
531
532 status = ocfs2_journal_access_gd(handle,
533 INODE_CACHE(alloc_inode),
534 bg_bh,
535 OCFS2_JOURNAL_ACCESS_CREATE);
536 if (status < 0) {
537 mlog_errno(status);
538 goto bail;
539 }
540
541 while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
542 le16_to_cpu(el->l_count))) {
798db35f
JB
543 if (min_bits > needed)
544 min_bits = needed;
545 status = ocfs2_block_group_claim_bits(osb, handle, ac,
546 min_bits, &p_cpos,
547 &clusters);
548 if (status < 0) {
549 if (status != -ENOSPC)
550 mlog_errno(status);
551 goto bail;
552 }
553 p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
554 ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
555 clusters);
556
557 min_bits = clusters;
4711954e
TM
558 needed = le16_to_cpu(cl->cl_cpg) -
559 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
798db35f
JB
560 }
561
562 if (needed > 0) {
4711954e
TM
563 /*
564 * We have used up all the extent rec but can't fill up
565 * the cpg. So bail out.
566 */
567 status = -ENOSPC;
568 goto bail;
798db35f
JB
569 }
570
571 ocfs2_journal_dirty(handle, bg_bh);
572
573bail:
574 return status;
575}
576
8b06bc59
JB
577static void ocfs2_bg_alloc_cleanup(handle_t *handle,
578 struct ocfs2_alloc_context *cluster_ac,
579 struct inode *alloc_inode,
580 struct buffer_head *bg_bh)
798db35f 581{
8b06bc59 582 int i, ret;
798db35f
JB
583 struct ocfs2_group_desc *bg;
584 struct ocfs2_extent_list *el;
585 struct ocfs2_extent_rec *rec;
586
587 if (!bg_bh)
588 return;
589
590 bg = (struct ocfs2_group_desc *)bg_bh->b_data;
591 el = &bg->bg_list;
592 for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
593 rec = &el->l_recs[i];
8b06bc59
JB
594 ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
595 cluster_ac->ac_bh,
596 le64_to_cpu(rec->e_blkno),
72094e43 597 le16_to_cpu(rec->e_leaf_clusters));
8b06bc59
JB
598 if (ret)
599 mlog_errno(ret);
600 /* Try all the clusters to free */
798db35f
JB
601 }
602
603 ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
604 brelse(bg_bh);
605}
606
607static struct buffer_head *
608ocfs2_block_group_alloc_discontig(handle_t *handle,
609 struct inode *alloc_inode,
610 struct ocfs2_alloc_context *ac,
8b06bc59 611 struct ocfs2_chain_list *cl)
798db35f
JB
612{
613 int status;
614 u32 bit_off, num_bits;
615 u64 bg_blkno;
616 unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
617 struct buffer_head *bg_bh = NULL;
618 unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
619 struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
620
4711954e 621 if (!ocfs2_supports_discontig_bg(osb)) {
798db35f
JB
622 status = -ENOSPC;
623 goto bail;
624 }
625
8b06bc59
JB
626 status = ocfs2_extend_trans(handle,
627 ocfs2_calc_bg_discontig_credits(osb->sb));
628 if (status) {
629 mlog_errno(status);
630 goto bail;
631 }
632
95ec0adf
JB
633 /*
634 * We're going to be grabbing from multiple cluster groups.
635 * We don't have enough credits to relink them all, and the
636 * cluster groups will be staying in cache for the duration of
637 * this operation.
638 */
309a85b6 639 ac->ac_disable_chain_relink = 1;
95ec0adf 640
798db35f
JB
641 /* Claim the first region */
642 status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
643 &bit_off, &num_bits);
644 if (status < 0) {
645 if (status != -ENOSPC)
646 mlog_errno(status);
647 goto bail;
648 }
649 min_bits = num_bits;
650
651 /* setup the group */
652 bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
2f73e135
TM
653 trace_ocfs2_block_group_alloc_discontig(
654 (unsigned long long)bg_blkno, alloc_rec);
798db35f
JB
655
656 bg_bh = sb_getblk(osb->sb, bg_blkno);
657 if (!bg_bh) {
7391a294 658 status = -ENOMEM;
798db35f
JB
659 mlog_errno(status);
660 goto bail;
661 }
662 ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
663
664 status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
665 bg_blkno, num_bits, alloc_rec, cl);
666 if (status < 0) {
667 mlog_errno(status);
668 goto bail;
669 }
670
671 status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
672 bg_bh, ac, cl, min_bits);
673 if (status)
674 mlog_errno(status);
675
676bail:
677 if (status)
8b06bc59 678 ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
798db35f
JB
679 return status ? ERR_PTR(status) : bg_bh;
680}
681
ccd979bd
MF
682/*
683 * We expect the block group allocator to already be locked.
684 */
685static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
686 struct inode *alloc_inode,
1187c968 687 struct buffer_head *bh,
60ca81e8 688 u64 max_block,
feb473a6 689 u64 *last_alloc_group,
60ca81e8 690 int flags)
ccd979bd
MF
691{
692 int status, credits;
693 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
694 struct ocfs2_chain_list *cl;
695 struct ocfs2_alloc_context *ac = NULL;
1fabe148 696 handle_t *handle = NULL;
4711954e 697 u16 alloc_rec;
ccd979bd
MF
698 struct buffer_head *bg_bh = NULL;
699 struct ocfs2_group_desc *bg;
700
701 BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
702
ccd979bd 703 cl = &fe->id2.i_chain;
1187c968
JB
704 status = ocfs2_reserve_clusters_with_limit(osb,
705 le16_to_cpu(cl->cl_cpg),
60ca81e8 706 max_block, flags, &ac);
ccd979bd
MF
707 if (status < 0) {
708 if (status != -ENOSPC)
709 mlog_errno(status);
710 goto bail;
711 }
712
713 credits = ocfs2_calc_group_alloc_credits(osb->sb,
714 le16_to_cpu(cl->cl_cpg));
65eff9cc 715 handle = ocfs2_start_trans(osb, credits);
ccd979bd
MF
716 if (IS_ERR(handle)) {
717 status = PTR_ERR(handle);
718 handle = NULL;
719 mlog_errno(status);
720 goto bail;
721 }
722
feb473a6 723 if (last_alloc_group && *last_alloc_group != 0) {
2f73e135
TM
724 trace_ocfs2_block_group_alloc(
725 (unsigned long long)*last_alloc_group);
feb473a6
TM
726 ac->ac_last_group = *last_alloc_group;
727 }
798db35f
JB
728
729 bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
730 ac, cl);
731 if (IS_ERR(bg_bh) && (PTR_ERR(bg_bh) == -ENOSPC))
732 bg_bh = ocfs2_block_group_alloc_discontig(handle,
733 alloc_inode,
8b06bc59 734 ac, cl);
798db35f
JB
735 if (IS_ERR(bg_bh)) {
736 status = PTR_ERR(bg_bh);
737 bg_bh = NULL;
ccd979bd
MF
738 if (status != -ENOSPC)
739 mlog_errno(status);
740 goto bail;
741 }
ccd979bd
MF
742 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
743
0cf2f763 744 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
13723d00 745 bh, OCFS2_JOURNAL_ACCESS_WRITE);
ccd979bd
MF
746 if (status < 0) {
747 mlog_errno(status);
748 goto bail;
749 }
750
4711954e
TM
751 alloc_rec = le16_to_cpu(bg->bg_chain);
752 le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
ccd979bd 753 le16_to_cpu(bg->bg_free_bits_count));
4711954e 754 le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
798db35f 755 le16_to_cpu(bg->bg_bits));
0a463b74 756 cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
ccd979bd
MF
757 if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
758 le16_add_cpu(&cl->cl_next_free_rec, 1);
759
760 le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
761 le16_to_cpu(bg->bg_free_bits_count));
762 le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
763 le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
764
ec20cec7 765 ocfs2_journal_dirty(handle, bh);
ccd979bd
MF
766
767 spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
768 OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
769 fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
770 le32_to_cpu(fe->i_clusters)));
771 spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
772 i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
8110b073 773 alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
6fdb702d 774 ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
ccd979bd
MF
775
776 status = 0;
feb473a6
TM
777
778 /* save the new last alloc group so that the caller can cache it. */
779 if (last_alloc_group)
780 *last_alloc_group = ac->ac_last_group;
781
ccd979bd
MF
782bail:
783 if (handle)
02dc1af4 784 ocfs2_commit_trans(osb, handle);
ccd979bd
MF
785
786 if (ac)
787 ocfs2_free_alloc_context(ac);
788
a81cb88b 789 brelse(bg_bh);
ccd979bd 790
c1e8d35e
TM
791 if (status)
792 mlog_errno(status);
ccd979bd
MF
793 return status;
794}
795
796static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
da5cbf2f
MF
797 struct ocfs2_alloc_context *ac,
798 int type,
ffda89a3 799 u32 slot,
feb473a6 800 u64 *last_alloc_group,
60ca81e8 801 int flags)
ccd979bd
MF
802{
803 int status;
804 u32 bits_wanted = ac->ac_bits_wanted;
da5cbf2f 805 struct inode *alloc_inode;
ccd979bd 806 struct buffer_head *bh = NULL;
ccd979bd
MF
807 struct ocfs2_dinode *fe;
808 u32 free_bits;
809
da5cbf2f
MF
810 alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
811 if (!alloc_inode) {
812 mlog_errno(-EINVAL);
813 return -EINVAL;
814 }
ccd979bd 815
da5cbf2f
MF
816 mutex_lock(&alloc_inode->i_mutex);
817
e63aecb6 818 status = ocfs2_inode_lock(alloc_inode, &bh, 1);
ccd979bd 819 if (status < 0) {
da5cbf2f
MF
820 mutex_unlock(&alloc_inode->i_mutex);
821 iput(alloc_inode);
822
ccd979bd 823 mlog_errno(status);
da5cbf2f 824 return status;
ccd979bd
MF
825 }
826
da5cbf2f 827 ac->ac_inode = alloc_inode;
a4a48911 828 ac->ac_alloc_slot = slot;
da5cbf2f 829
ccd979bd 830 fe = (struct ocfs2_dinode *) bh->b_data;
10995aa2
JB
831
832 /* The bh was validated by the inode read inside
833 * ocfs2_inode_lock(). Any corruption is a code bug. */
834 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
835
ccd979bd 836 if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
b0697053
MF
837 ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
838 (unsigned long long)le64_to_cpu(fe->i_blkno));
ccd979bd
MF
839 status = -EIO;
840 goto bail;
841 }
842
843 free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
844 le32_to_cpu(fe->id1.bitmap1.i_used);
845
846 if (bits_wanted > free_bits) {
847 /* cluster bitmap never grows */
848 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
2f73e135
TM
849 trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
850 free_bits);
ccd979bd
MF
851 status = -ENOSPC;
852 goto bail;
853 }
854
60ca81e8 855 if (!(flags & ALLOC_NEW_GROUP)) {
2f73e135
TM
856 trace_ocfs2_reserve_suballoc_bits_no_new_group(
857 slot, bits_wanted, free_bits);
ffda89a3
TM
858 status = -ENOSPC;
859 goto bail;
860 }
861
1187c968 862 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
feb473a6
TM
863 ac->ac_max_block,
864 last_alloc_group, flags);
ccd979bd
MF
865 if (status < 0) {
866 if (status != -ENOSPC)
867 mlog_errno(status);
868 goto bail;
869 }
870 atomic_inc(&osb->alloc_stats.bg_extends);
871
872 /* You should never ask for this much metadata */
873 BUG_ON(bits_wanted >
874 (le32_to_cpu(fe->id1.bitmap1.i_total)
875 - le32_to_cpu(fe->id1.bitmap1.i_used)));
876 }
877
878 get_bh(bh);
879 ac->ac_bh = bh;
880bail:
a81cb88b 881 brelse(bh);
ccd979bd 882
c1e8d35e
TM
883 if (status)
884 mlog_errno(status);
ccd979bd
MF
885 return status;
886}
887
b89c5428
TY
888static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
889{
890 spin_lock(&osb->osb_lock);
891 osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
892 spin_unlock(&osb->osb_lock);
893 atomic_set(&osb->s_num_inodes_stolen, 0);
894}
895
896static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
897{
898 spin_lock(&osb->osb_lock);
899 osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
900 spin_unlock(&osb->osb_lock);
901 atomic_set(&osb->s_num_meta_stolen, 0);
902}
903
904void ocfs2_init_steal_slots(struct ocfs2_super *osb)
905{
906 ocfs2_init_inode_steal_slot(osb);
907 ocfs2_init_meta_steal_slot(osb);
908}
909
910static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
911{
912 spin_lock(&osb->osb_lock);
913 if (type == INODE_ALLOC_SYSTEM_INODE)
914 osb->s_inode_steal_slot = slot;
915 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
916 osb->s_meta_steal_slot = slot;
917 spin_unlock(&osb->osb_lock);
918}
919
920static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
921{
922 int slot = OCFS2_INVALID_SLOT;
923
924 spin_lock(&osb->osb_lock);
925 if (type == INODE_ALLOC_SYSTEM_INODE)
926 slot = osb->s_inode_steal_slot;
927 else if (type == EXTENT_ALLOC_SYSTEM_INODE)
928 slot = osb->s_meta_steal_slot;
929 spin_unlock(&osb->osb_lock);
930
931 return slot;
932}
933
934static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
935{
936 return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
937}
938
939static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
940{
941 return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
942}
943
944static int ocfs2_steal_resource(struct ocfs2_super *osb,
945 struct ocfs2_alloc_context *ac,
946 int type)
947{
948 int i, status = -ENOSPC;
949 int slot = __ocfs2_get_steal_slot(osb, type);
950
951 /* Start to steal resource from the first slot after ours. */
952 if (slot == OCFS2_INVALID_SLOT)
953 slot = osb->slot_num + 1;
954
955 for (i = 0; i < osb->max_slots; i++, slot++) {
956 if (slot == osb->max_slots)
957 slot = 0;
958
959 if (slot == osb->slot_num)
960 continue;
961
962 status = ocfs2_reserve_suballoc_bits(osb, ac,
963 type,
964 (u32)slot, NULL,
965 NOT_ALLOC_NEW_GROUP);
966 if (status >= 0) {
967 __ocfs2_set_steal_slot(osb, slot, type);
968 break;
969 }
970
971 ocfs2_free_ac_resource(ac);
972 }
973
974 return status;
975}
976
977static int ocfs2_steal_inode(struct ocfs2_super *osb,
978 struct ocfs2_alloc_context *ac)
979{
980 return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
981}
982
983static int ocfs2_steal_meta(struct ocfs2_super *osb,
984 struct ocfs2_alloc_context *ac)
985{
986 return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
987}
988
cf1d6c76
TY
989int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
990 int blocks,
991 struct ocfs2_alloc_context **ac)
ccd979bd
MF
992{
993 int status;
b89c5428 994 int slot = ocfs2_get_meta_steal_slot(osb);
ccd979bd 995
cd861280 996 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
ccd979bd
MF
997 if (!(*ac)) {
998 status = -ENOMEM;
999 mlog_errno(status);
1000 goto bail;
1001 }
1002
cf1d6c76 1003 (*ac)->ac_bits_wanted = blocks;
ccd979bd 1004 (*ac)->ac_which = OCFS2_AC_USE_META;
ccd979bd
MF
1005 (*ac)->ac_group_search = ocfs2_block_group_search;
1006
b89c5428
TY
1007 if (slot != OCFS2_INVALID_SLOT &&
1008 atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
1009 goto extent_steal;
1010
1011 atomic_set(&osb->s_num_meta_stolen, 0);
da5cbf2f 1012 status = ocfs2_reserve_suballoc_bits(osb, (*ac),
ffda89a3 1013 EXTENT_ALLOC_SYSTEM_INODE,
b89c5428 1014 (u32)osb->slot_num, NULL,
33d5d380 1015 ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
b89c5428
TY
1016
1017
1018 if (status >= 0) {
1019 status = 0;
1020 if (slot != OCFS2_INVALID_SLOT)
1021 ocfs2_init_meta_steal_slot(osb);
1022 goto bail;
1023 } else if (status < 0 && status != -ENOSPC) {
1024 mlog_errno(status);
1025 goto bail;
1026 }
1027
1028 ocfs2_free_ac_resource(*ac);
1029
1030extent_steal:
1031 status = ocfs2_steal_meta(osb, *ac);
1032 atomic_inc(&osb->s_num_meta_stolen);
ccd979bd
MF
1033 if (status < 0) {
1034 if (status != -ENOSPC)
1035 mlog_errno(status);
1036 goto bail;
1037 }
1038
1039 status = 0;
1040bail:
1041 if ((status < 0) && *ac) {
1042 ocfs2_free_alloc_context(*ac);
1043 *ac = NULL;
1044 }
1045
c1e8d35e
TM
1046 if (status)
1047 mlog_errno(status);
ccd979bd
MF
1048 return status;
1049}
1050
cf1d6c76
TY
1051int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1052 struct ocfs2_extent_list *root_el,
1053 struct ocfs2_alloc_context **ac)
1054{
1055 return ocfs2_reserve_new_metadata_blocks(osb,
1056 ocfs2_extend_meta_needed(root_el),
1057 ac);
1058}
1059
ccd979bd 1060int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
ccd979bd
MF
1061 struct ocfs2_alloc_context **ac)
1062{
1063 int status;
b89c5428 1064 int slot = ocfs2_get_inode_steal_slot(osb);
feb473a6 1065 u64 alloc_group;
ccd979bd 1066
cd861280 1067 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
ccd979bd
MF
1068 if (!(*ac)) {
1069 status = -ENOMEM;
1070 mlog_errno(status);
1071 goto bail;
1072 }
1073
1074 (*ac)->ac_bits_wanted = 1;
ccd979bd
MF
1075 (*ac)->ac_which = OCFS2_AC_USE_INODE;
1076
ccd979bd
MF
1077 (*ac)->ac_group_search = ocfs2_block_group_search;
1078
1187c968
JB
1079 /*
1080 * stat(2) can't handle i_ino > 32bits, so we tell the
1081 * lower levels not to allocate us a block group past that
12462f1d 1082 * limit. The 'inode64' mount option avoids this behavior.
1187c968 1083 */
12462f1d
JB
1084 if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1085 (*ac)->ac_max_block = (u32)~0U;
1187c968 1086
4d0ddb2c
TM
1087 /*
1088 * slot is set when we successfully steal inode from other nodes.
1089 * It is reset in 3 places:
1090 * 1. when we flush the truncate log
1091 * 2. when we complete local alloc recovery.
1092 * 3. when we successfully allocate from our own slot.
1093 * After it is set, we will go on stealing inodes until we find the
1094 * need to check our slots to see whether there is some space for us.
1095 */
1096 if (slot != OCFS2_INVALID_SLOT &&
b89c5428 1097 atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
4d0ddb2c
TM
1098 goto inode_steal;
1099
1100 atomic_set(&osb->s_num_inodes_stolen, 0);
feb473a6 1101 alloc_group = osb->osb_inode_alloc_group;
da5cbf2f
MF
1102 status = ocfs2_reserve_suballoc_bits(osb, *ac,
1103 INODE_ALLOC_SYSTEM_INODE,
b89c5428 1104 (u32)osb->slot_num,
feb473a6 1105 &alloc_group,
60ca81e8
TM
1106 ALLOC_NEW_GROUP |
1107 ALLOC_GROUPS_FROM_GLOBAL);
4d0ddb2c
TM
1108 if (status >= 0) {
1109 status = 0;
1110
feb473a6
TM
1111 spin_lock(&osb->osb_lock);
1112 osb->osb_inode_alloc_group = alloc_group;
1113 spin_unlock(&osb->osb_lock);
2f73e135
TM
1114 trace_ocfs2_reserve_new_inode_new_group(
1115 (unsigned long long)alloc_group);
feb473a6 1116
4d0ddb2c
TM
1117 /*
1118 * Some inodes must be freed by us, so try to allocate
1119 * from our own next time.
1120 */
1121 if (slot != OCFS2_INVALID_SLOT)
1122 ocfs2_init_inode_steal_slot(osb);
1123 goto bail;
1124 } else if (status < 0 && status != -ENOSPC) {
1125 mlog_errno(status);
1126 goto bail;
1127 }
1128
1129 ocfs2_free_ac_resource(*ac);
1130
1131inode_steal:
b89c5428 1132 status = ocfs2_steal_inode(osb, *ac);
4d0ddb2c 1133 atomic_inc(&osb->s_num_inodes_stolen);
ccd979bd
MF
1134 if (status < 0) {
1135 if (status != -ENOSPC)
1136 mlog_errno(status);
1137 goto bail;
1138 }
1139
1140 status = 0;
1141bail:
1142 if ((status < 0) && *ac) {
1143 ocfs2_free_alloc_context(*ac);
1144 *ac = NULL;
1145 }
1146
c1e8d35e
TM
1147 if (status)
1148 mlog_errno(status);
ccd979bd
MF
1149 return status;
1150}
1151
1152/* local alloc code has to do the same thing, so rather than do this
1153 * twice.. */
1154int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1155 struct ocfs2_alloc_context *ac)
1156{
1157 int status;
1158
ccd979bd
MF
1159 ac->ac_which = OCFS2_AC_USE_MAIN;
1160 ac->ac_group_search = ocfs2_cluster_group_search;
1161
da5cbf2f
MF
1162 status = ocfs2_reserve_suballoc_bits(osb, ac,
1163 GLOBAL_BITMAP_SYSTEM_INODE,
feb473a6 1164 OCFS2_INVALID_SLOT, NULL,
ffda89a3 1165 ALLOC_NEW_GROUP);
da5cbf2f 1166 if (status < 0 && status != -ENOSPC) {
ccd979bd 1167 mlog_errno(status);
da5cbf2f
MF
1168 goto bail;
1169 }
1170
ccd979bd
MF
1171bail:
1172 return status;
1173}
1174
1175/* Callers don't need to care which bitmap (local alloc or main) to
1176 * use so we figure it out for them, but unfortunately this clutters
1177 * things a bit. */
1187c968
JB
1178static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1179 u32 bits_wanted, u64 max_block,
60ca81e8 1180 int flags,
1187c968 1181 struct ocfs2_alloc_context **ac)
ccd979bd
MF
1182{
1183 int status;
1184
cd861280 1185 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
ccd979bd
MF
1186 if (!(*ac)) {
1187 status = -ENOMEM;
1188 mlog_errno(status);
1189 goto bail;
1190 }
1191
1192 (*ac)->ac_bits_wanted = bits_wanted;
1187c968 1193 (*ac)->ac_max_block = max_block;
ccd979bd
MF
1194
1195 status = -ENOSPC;
60ca81e8
TM
1196 if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1197 ocfs2_alloc_should_use_local(osb, bits_wanted)) {
ccd979bd 1198 status = ocfs2_reserve_local_alloc_bits(osb,
ccd979bd
MF
1199 bits_wanted,
1200 *ac);
a57c8fd2 1201 if ((status < 0) && (status != -ENOSPC)) {
ccd979bd
MF
1202 mlog_errno(status);
1203 goto bail;
ccd979bd
MF
1204 }
1205 }
1206
1207 if (status == -ENOSPC) {
1208 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1209 if (status < 0) {
1210 if (status != -ENOSPC)
1211 mlog_errno(status);
1212 goto bail;
1213 }
1214 }
1215
1216 status = 0;
1217bail:
1218 if ((status < 0) && *ac) {
1219 ocfs2_free_alloc_context(*ac);
1220 *ac = NULL;
1221 }
1222
c1e8d35e
TM
1223 if (status)
1224 mlog_errno(status);
ccd979bd
MF
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
0a2fcd89 1341int 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
2f73e135 1357 trace_ocfs2_block_group_set_bits(bit_off, num_bits);
ccd979bd
MF
1358
1359 if (ocfs2_is_cluster_bitmap(alloc_inode))
1360 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1361
13723d00 1362 status = ocfs2_journal_access_gd(handle,
0cf2f763 1363 INODE_CACHE(alloc_inode),
13723d00
JB
1364 group_bh,
1365 journal_type);
ccd979bd
MF
1366 if (status < 0) {
1367 mlog_errno(status);
1368 goto bail;
1369 }
1370
1371 le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
9b5cd10e
SE
1372 if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1373 ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
1374 " count %u but claims %u are freed. num_bits %d",
1375 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1376 le16_to_cpu(bg->bg_bits),
1377 le16_to_cpu(bg->bg_free_bits_count), num_bits);
1378 return -EROFS;
1379 }
ccd979bd
MF
1380 while(num_bits--)
1381 ocfs2_set_bit(bit_off++, bitmap);
1382
ec20cec7 1383 ocfs2_journal_dirty(handle, group_bh);
ccd979bd
MF
1384
1385bail:
ccd979bd
MF
1386 return status;
1387}
1388
1389/* find the one with the most empty bits */
1390static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1391{
1392 u16 curr, best;
1393
1394 BUG_ON(!cl->cl_next_free_rec);
1395
1396 best = curr = 0;
1397 while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1398 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1399 le32_to_cpu(cl->cl_recs[best].c_free))
1400 best = curr;
1401 curr++;
1402 }
1403
1404 BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1405 return best;
1406}
1407
1fabe148 1408static int ocfs2_relink_block_group(handle_t *handle,
ccd979bd
MF
1409 struct inode *alloc_inode,
1410 struct buffer_head *fe_bh,
1411 struct buffer_head *bg_bh,
1412 struct buffer_head *prev_bg_bh,
1413 u16 chain)
1414{
1415 int status;
1416 /* there is a really tiny chance the journal calls could fail,
1417 * but we wouldn't want inconsistent blocks in *any* case. */
49309841 1418 u64 bg_ptr, prev_bg_ptr;
ccd979bd
MF
1419 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1420 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1421 struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1422
42035306
JB
1423 /* The caller got these descriptors from
1424 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
1425 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1426 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
ccd979bd 1427
2f73e135
TM
1428 trace_ocfs2_relink_block_group(
1429 (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1430 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1431 (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
ccd979bd 1432
ccd979bd
MF
1433 bg_ptr = le64_to_cpu(bg->bg_next_group);
1434 prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1435
0cf2f763
JB
1436 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1437 prev_bg_bh,
13723d00 1438 OCFS2_JOURNAL_ACCESS_WRITE);
25e28921 1439 if (status < 0)
49309841 1440 goto out;
ccd979bd
MF
1441
1442 prev_bg->bg_next_group = bg->bg_next_group;
ec20cec7 1443 ocfs2_journal_dirty(handle, prev_bg_bh);
ccd979bd 1444
0cf2f763
JB
1445 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1446 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
25e28921 1447 if (status < 0)
49309841 1448 goto out_rollback_prev_bg;
ccd979bd
MF
1449
1450 bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
ec20cec7 1451 ocfs2_journal_dirty(handle, bg_bh);
ccd979bd 1452
0cf2f763
JB
1453 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1454 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
25e28921 1455 if (status < 0)
49309841 1456 goto out_rollback_bg;
ccd979bd
MF
1457
1458 fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
ec20cec7 1459 ocfs2_journal_dirty(handle, fe_bh);
ccd979bd 1460
49309841 1461out:
25e28921
AM
1462 if (status < 0)
1463 mlog_errno(status);
49309841 1464 return status;
42035306 1465
49309841
JL
1466out_rollback_bg:
1467 bg->bg_next_group = cpu_to_le64(bg_ptr);
1468out_rollback_prev_bg:
1469 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
25e28921 1470 goto out;
ccd979bd
MF
1471}
1472
1473static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1474 u32 wanted)
1475{
1476 return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1477}
1478
1479/* return 0 on success, -ENOSPC to keep searching and any other < 0
1480 * value on error. */
1481static int ocfs2_cluster_group_search(struct inode *inode,
1482 struct buffer_head *group_bh,
1483 u32 bits_wanted, u32 min_bits,
1187c968 1484 u64 max_block,
7d1fe093 1485 struct ocfs2_suballoc_result *res)
ccd979bd
MF
1486{
1487 int search = -ENOSPC;
1488 int ret;
1187c968 1489 u64 blkoff;
7bf72ede 1490 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
9c7af40b 1491 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7bf72ede 1492 unsigned int max_bits, gd_cluster_off;
ccd979bd
MF
1493
1494 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1495
7bf72ede
MF
1496 if (gd->bg_free_bits_count) {
1497 max_bits = le16_to_cpu(gd->bg_bits);
1498
1499 /* Tail groups in cluster bitmaps which aren't cpg
25985edc 1500 * aligned are prone to partial extension by a failed
7bf72ede
MF
1501 * fs resize. If the file system resize never got to
1502 * update the dinode cluster count, then we don't want
1503 * to trust any clusters past it, regardless of what
1504 * the group descriptor says. */
1505 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1506 le64_to_cpu(gd->bg_blkno));
1507 if ((gd_cluster_off + max_bits) >
1508 OCFS2_I(inode)->ip_clusters) {
1509 max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
2f73e135
TM
1510 trace_ocfs2_cluster_group_search_wrong_max_bits(
1511 (unsigned long long)le64_to_cpu(gd->bg_blkno),
1512 le16_to_cpu(gd->bg_bits),
1513 OCFS2_I(inode)->ip_clusters, max_bits);
7bf72ede
MF
1514 }
1515
ccd979bd
MF
1516 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1517 group_bh, bits_wanted,
7d1fe093 1518 max_bits, res);
ccd979bd
MF
1519 if (ret)
1520 return ret;
1521
1187c968
JB
1522 if (max_block) {
1523 blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1524 gd_cluster_off +
7d1fe093
JB
1525 res->sr_bit_offset +
1526 res->sr_bits);
2f73e135
TM
1527 trace_ocfs2_cluster_group_search_max_block(
1528 (unsigned long long)blkoff,
1529 (unsigned long long)max_block);
1187c968
JB
1530 if (blkoff > max_block)
1531 return -ENOSPC;
1532 }
1533
ccd979bd
MF
1534 /* ocfs2_block_group_find_clear_bits() might
1535 * return success, but we still want to return
1536 * -ENOSPC unless it found the minimum number
1537 * of bits. */
7d1fe093 1538 if (min_bits <= res->sr_bits)
ccd979bd 1539 search = 0; /* success */
7d1fe093 1540 else if (res->sr_bits) {
9c7af40b
MF
1541 /*
1542 * Don't show bits which we'll be returning
1543 * for allocation to the local alloc bitmap.
1544 */
7d1fe093 1545 ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
ccd979bd
MF
1546 }
1547 }
1548
1549 return search;
1550}
1551
1552static int ocfs2_block_group_search(struct inode *inode,
1553 struct buffer_head *group_bh,
1554 u32 bits_wanted, u32 min_bits,
1187c968 1555 u64 max_block,
7d1fe093 1556 struct ocfs2_suballoc_result *res)
ccd979bd
MF
1557{
1558 int ret = -ENOSPC;
1187c968 1559 u64 blkoff;
ccd979bd
MF
1560 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1561
1562 BUG_ON(min_bits != 1);
1563 BUG_ON(ocfs2_is_cluster_bitmap(inode));
1564
1187c968 1565 if (bg->bg_free_bits_count) {
ccd979bd
MF
1566 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1567 group_bh, bits_wanted,
7bf72ede 1568 le16_to_cpu(bg->bg_bits),
7d1fe093 1569 res);
1187c968 1570 if (!ret && max_block) {
7d1fe093
JB
1571 blkoff = le64_to_cpu(bg->bg_blkno) +
1572 res->sr_bit_offset + res->sr_bits;
2f73e135
TM
1573 trace_ocfs2_block_group_search_max_block(
1574 (unsigned long long)blkoff,
1575 (unsigned long long)max_block);
1187c968
JB
1576 if (blkoff > max_block)
1577 ret = -ENOSPC;
1578 }
1579 }
ccd979bd
MF
1580
1581 return ret;
1582}
1583
0a2fcd89 1584int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1fabe148 1585 handle_t *handle,
883d4cae
MF
1586 struct buffer_head *di_bh,
1587 u32 num_bits,
1588 u16 chain)
1589{
1590 int ret;
1591 u32 tmp_used;
1592 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1593 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1594
0cf2f763 1595 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
13723d00 1596 OCFS2_JOURNAL_ACCESS_WRITE);
883d4cae
MF
1597 if (ret < 0) {
1598 mlog_errno(ret);
1599 goto out;
1600 }
1601
1602 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1603 di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1604 le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
ec20cec7 1605 ocfs2_journal_dirty(handle, di_bh);
883d4cae
MF
1606
1607out:
1608 return ret;
1609}
1610
ba206635
JB
1611static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1612 struct ocfs2_extent_rec *rec,
1613 struct ocfs2_chain_list *cl)
13e434cf
JB
1614{
1615 unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1616 unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
72094e43 1617 unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
13e434cf
JB
1618
1619 if (res->sr_bit_offset < bitoff)
1620 return 0;
1621 if (res->sr_bit_offset >= (bitoff + bitcount))
1622 return 0;
ba206635
JB
1623 res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1624 (res->sr_bit_offset - bitoff);
13e434cf
JB
1625 if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1626 res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1627 return 1;
1628}
1629
ba206635
JB
1630static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1631 struct ocfs2_group_desc *bg,
1632 struct ocfs2_suballoc_result *res)
13e434cf
JB
1633{
1634 int i;
2b6cb576 1635 u64 bg_blkno = res->sr_bg_blkno; /* Save off */
13e434cf
JB
1636 struct ocfs2_extent_rec *rec;
1637 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1638 struct ocfs2_chain_list *cl = &di->id2.i_chain;
1639
ba206635
JB
1640 if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1641 res->sr_blkno = 0;
13e434cf 1642 return;
ba206635 1643 }
13e434cf 1644
ba206635 1645 res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
2b6cb576 1646 res->sr_bg_blkno = 0; /* Clear it for contig block groups */
4711954e 1647 if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
ba206635 1648 !bg->bg_list.l_next_free_rec)
13e434cf
JB
1649 return;
1650
1651 for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1652 rec = &bg->bg_list.l_recs[i];
2b6cb576
JB
1653 if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1654 res->sr_bg_blkno = bg_blkno; /* Restore */
13e434cf 1655 break;
2b6cb576 1656 }
13e434cf
JB
1657 }
1658}
1659
883d4cae 1660static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1fabe148 1661 handle_t *handle,
883d4cae
MF
1662 u32 bits_wanted,
1663 u32 min_bits,
7d1fe093 1664 struct ocfs2_suballoc_result *res,
883d4cae
MF
1665 u16 *bits_left)
1666{
1667 int ret;
883d4cae
MF
1668 struct buffer_head *group_bh = NULL;
1669 struct ocfs2_group_desc *gd;
68f64d47 1670 struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
883d4cae 1671 struct inode *alloc_inode = ac->ac_inode;
883d4cae 1672
7d1fe093
JB
1673 ret = ocfs2_read_group_descriptor(alloc_inode, di,
1674 res->sr_bg_blkno, &group_bh);
883d4cae
MF
1675 if (ret < 0) {
1676 mlog_errno(ret);
1677 return ret;
1678 }
1679
1680 gd = (struct ocfs2_group_desc *) group_bh->b_data;
883d4cae 1681 ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
7d1fe093 1682 ac->ac_max_block, res);
883d4cae
MF
1683 if (ret < 0) {
1684 if (ret != -ENOSPC)
1685 mlog_errno(ret);
1686 goto out;
1687 }
1688
13e434cf 1689 if (!ret)
ba206635 1690 ocfs2_bg_discontig_fix_result(ac, gd, res);
13e434cf 1691
e49e2767
MF
1692 /*
1693 * sr_bg_blkno might have been changed by
1694 * ocfs2_bg_discontig_fix_result
1695 */
1696 res->sr_bg_stable_blkno = group_bh->b_blocknr;
1697
1698 if (ac->ac_find_loc_only)
1699 goto out_loc_only;
1700
883d4cae 1701 ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
7d1fe093 1702 res->sr_bits,
883d4cae
MF
1703 le16_to_cpu(gd->bg_chain));
1704 if (ret < 0) {
1705 mlog_errno(ret);
1706 goto out;
1707 }
1708
1709 ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
7d1fe093 1710 res->sr_bit_offset, res->sr_bits);
883d4cae
MF
1711 if (ret < 0)
1712 mlog_errno(ret);
1713
e49e2767 1714out_loc_only:
883d4cae
MF
1715 *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1716
1717out:
1718 brelse(group_bh);
1719
1720 return ret;
1721}
1722
ccd979bd 1723static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1fabe148 1724 handle_t *handle,
ccd979bd
MF
1725 u32 bits_wanted,
1726 u32 min_bits,
7d1fe093 1727 struct ocfs2_suballoc_result *res,
883d4cae 1728 u16 *bits_left)
ccd979bd
MF
1729{
1730 int status;
7d1fe093 1731 u16 chain;
ccd979bd 1732 u64 next_group;
ccd979bd
MF
1733 struct inode *alloc_inode = ac->ac_inode;
1734 struct buffer_head *group_bh = NULL;
1735 struct buffer_head *prev_group_bh = NULL;
1736 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1737 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1738 struct ocfs2_group_desc *bg;
1739
1740 chain = ac->ac_chain;
2f73e135
TM
1741 trace_ocfs2_search_chain_begin(
1742 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1743 bits_wanted, chain);
ccd979bd 1744
68f64d47
JB
1745 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1746 le64_to_cpu(cl->cl_recs[chain].c_blkno),
1747 &group_bh);
ccd979bd
MF
1748 if (status < 0) {
1749 mlog_errno(status);
1750 goto bail;
1751 }
1752 bg = (struct ocfs2_group_desc *) group_bh->b_data;
ccd979bd
MF
1753
1754 status = -ENOSPC;
1755 /* for now, the chain search is a bit simplistic. We just use
1756 * the 1st group with any empty bits. */
1187c968
JB
1757 while ((status = ac->ac_group_search(alloc_inode, group_bh,
1758 bits_wanted, min_bits,
7d1fe093
JB
1759 ac->ac_max_block,
1760 res)) == -ENOSPC) {
ccd979bd
MF
1761 if (!bg->bg_next_group)
1762 break;
a81cb88b
MF
1763
1764 brelse(prev_group_bh);
1765 prev_group_bh = NULL;
1766
ccd979bd
MF
1767 next_group = le64_to_cpu(bg->bg_next_group);
1768 prev_group_bh = group_bh;
1769 group_bh = NULL;
68f64d47
JB
1770 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1771 next_group, &group_bh);
ccd979bd
MF
1772 if (status < 0) {
1773 mlog_errno(status);
1774 goto bail;
1775 }
1776 bg = (struct ocfs2_group_desc *) group_bh->b_data;
ccd979bd
MF
1777 }
1778 if (status < 0) {
1779 if (status != -ENOSPC)
1780 mlog_errno(status);
1781 goto bail;
1782 }
1783
2f73e135
TM
1784 trace_ocfs2_search_chain_succ(
1785 (unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
ccd979bd 1786
7d1fe093 1787 res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
ccd979bd 1788
7d1fe093 1789 BUG_ON(res->sr_bits == 0);
13e434cf 1790 if (!status)
ba206635 1791 ocfs2_bg_discontig_fix_result(ac, bg, res);
13e434cf 1792
e49e2767
MF
1793 /*
1794 * sr_bg_blkno might have been changed by
1795 * ocfs2_bg_discontig_fix_result
1796 */
1797 res->sr_bg_stable_blkno = group_bh->b_blocknr;
ccd979bd
MF
1798
1799 /*
1800 * Keep track of previous block descriptor read. When
1801 * we find a target, if we have read more than X
1802 * number of descriptors, and the target is reasonably
1803 * empty, relink him to top of his chain.
1804 *
1805 * We've read 0 extra blocks and only send one more to
1806 * the transaction, yet the next guy to search has a
1807 * much easier time.
1808 *
1809 * Do this *after* figuring out how many bits we're taking out
1810 * of our target group.
1811 */
309a85b6 1812 if (!ac->ac_disable_chain_relink &&
ccd979bd 1813 (prev_group_bh) &&
7d1fe093 1814 (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
ccd979bd
MF
1815 status = ocfs2_relink_block_group(handle, alloc_inode,
1816 ac->ac_bh, group_bh,
1817 prev_group_bh, chain);
1818 if (status < 0) {
1819 mlog_errno(status);
1820 goto bail;
1821 }
1822 }
1823
e49e2767
MF
1824 if (ac->ac_find_loc_only)
1825 goto out_loc_only;
1826
d5134982
MF
1827 status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1828 ac->ac_bh, res->sr_bits,
1829 chain);
1830 if (status) {
ccd979bd
MF
1831 mlog_errno(status);
1832 goto bail;
1833 }
1834
ccd979bd
MF
1835 status = ocfs2_block_group_set_bits(handle,
1836 alloc_inode,
1837 bg,
1838 group_bh,
7d1fe093
JB
1839 res->sr_bit_offset,
1840 res->sr_bits);
ccd979bd
MF
1841 if (status < 0) {
1842 mlog_errno(status);
1843 goto bail;
1844 }
1845
2f73e135
TM
1846 trace_ocfs2_search_chain_end(
1847 (unsigned long long)le64_to_cpu(fe->i_blkno),
1848 res->sr_bits);
ccd979bd 1849
e49e2767 1850out_loc_only:
883d4cae 1851 *bits_left = le16_to_cpu(bg->bg_free_bits_count);
ccd979bd 1852bail:
a81cb88b
MF
1853 brelse(group_bh);
1854 brelse(prev_group_bh);
ccd979bd 1855
c1e8d35e
TM
1856 if (status)
1857 mlog_errno(status);
ccd979bd
MF
1858 return status;
1859}
1860
1861/* will give out up to bits_wanted contiguous bits. */
aa8f8e93 1862static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1fabe148 1863 handle_t *handle,
ccd979bd
MF
1864 u32 bits_wanted,
1865 u32 min_bits,
7d1fe093 1866 struct ocfs2_suballoc_result *res)
ccd979bd
MF
1867{
1868 int status;
1869 u16 victim, i;
883d4cae 1870 u16 bits_left = 0;
b2b6ebf5 1871 u64 hint = ac->ac_last_group;
ccd979bd
MF
1872 struct ocfs2_chain_list *cl;
1873 struct ocfs2_dinode *fe;
1874
ccd979bd
MF
1875 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1876 BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1877 BUG_ON(!ac->ac_bh);
1878
1879 fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
10995aa2
JB
1880
1881 /* The bh was validated by the inode read during
1882 * ocfs2_reserve_suballoc_bits(). Any corruption is a code bug. */
1883 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1884
ccd979bd
MF
1885 if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1886 le32_to_cpu(fe->id1.bitmap1.i_total)) {
aa8f8e93
JB
1887 ocfs2_error(ac->ac_inode->i_sb,
1888 "Chain allocator dinode %llu has %u used "
b0697053
MF
1889 "bits but only %u total.",
1890 (unsigned long long)le64_to_cpu(fe->i_blkno),
ccd979bd
MF
1891 le32_to_cpu(fe->id1.bitmap1.i_used),
1892 le32_to_cpu(fe->id1.bitmap1.i_total));
1893 status = -EIO;
1894 goto bail;
1895 }
1896
b2b6ebf5 1897 res->sr_bg_blkno = hint;
7d1fe093 1898 if (res->sr_bg_blkno) {
883d4cae
MF
1899 /* Attempt to short-circuit the usual search mechanism
1900 * by jumping straight to the most recently used
b595076a 1901 * allocation group. This helps us maintain some
883d4cae 1902 * contiguousness across allocations. */
da5cbf2f 1903 status = ocfs2_search_one_group(ac, handle, bits_wanted,
7d1fe093
JB
1904 min_bits, res, &bits_left);
1905 if (!status)
883d4cae 1906 goto set_hint;
883d4cae
MF
1907 if (status < 0 && status != -ENOSPC) {
1908 mlog_errno(status);
1909 goto bail;
1910 }
1911 }
1912
ccd979bd
MF
1913 cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1914
1915 victim = ocfs2_find_victim_chain(cl);
1916 ac->ac_chain = victim;
ccd979bd 1917
7d1fe093
JB
1918 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1919 res, &bits_left);
b2b6ebf5
MF
1920 if (!status) {
1921 hint = ocfs2_group_from_res(res);
883d4cae 1922 goto set_hint;
b2b6ebf5 1923 }
ccd979bd
MF
1924 if (status < 0 && status != -ENOSPC) {
1925 mlog_errno(status);
1926 goto bail;
1927 }
1928
2f73e135 1929 trace_ocfs2_claim_suballoc_bits(victim);
ccd979bd
MF
1930
1931 /* If we didn't pick a good victim, then just default to
1932 * searching each chain in order. Don't allow chain relinking
1933 * because we only calculate enough journal credits for one
1934 * relink per alloc. */
309a85b6 1935 ac->ac_disable_chain_relink = 1;
ccd979bd
MF
1936 for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1937 if (i == victim)
1938 continue;
1939 if (!cl->cl_recs[i].c_free)
1940 continue;
1941
1942 ac->ac_chain = i;
da5cbf2f 1943 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
7d1fe093 1944 res, &bits_left);
b2b6ebf5
MF
1945 if (!status) {
1946 hint = ocfs2_group_from_res(res);
ccd979bd 1947 break;
b2b6ebf5 1948 }
ccd979bd
MF
1949 if (status < 0 && status != -ENOSPC) {
1950 mlog_errno(status);
1951 goto bail;
1952 }
1953 }
ccd979bd 1954
883d4cae
MF
1955set_hint:
1956 if (status != -ENOSPC) {
1957 /* If the next search of this group is not likely to
1958 * yield a suitable extent, then we reset the last
1959 * group hint so as to not waste a disk read */
1960 if (bits_left < min_bits)
1961 ac->ac_last_group = 0;
1962 else
b2b6ebf5 1963 ac->ac_last_group = hint;
883d4cae
MF
1964 }
1965
1966bail:
c1e8d35e
TM
1967 if (status)
1968 mlog_errno(status);
ccd979bd
MF
1969 return status;
1970}
1971
1ed9b777 1972int ocfs2_claim_metadata(handle_t *handle,
ccd979bd
MF
1973 struct ocfs2_alloc_context *ac,
1974 u32 bits_wanted,
2b6cb576 1975 u64 *suballoc_loc,
ccd979bd
MF
1976 u16 *suballoc_bit_start,
1977 unsigned int *num_bits,
1978 u64 *blkno_start)
1979{
1980 int status;
ba206635 1981 struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
ccd979bd
MF
1982
1983 BUG_ON(!ac);
1984 BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1985 BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
ccd979bd 1986
aa8f8e93 1987 status = ocfs2_claim_suballoc_bits(ac,
da5cbf2f 1988 handle,
ccd979bd
MF
1989 bits_wanted,
1990 1,
7d1fe093 1991 &res);
ccd979bd
MF
1992 if (status < 0) {
1993 mlog_errno(status);
1994 goto bail;
1995 }
1ed9b777 1996 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
ccd979bd 1997
2b6cb576 1998 *suballoc_loc = res.sr_bg_blkno;
7d1fe093 1999 *suballoc_bit_start = res.sr_bit_offset;
ba206635 2000 *blkno_start = res.sr_blkno;
7d1fe093
JB
2001 ac->ac_bits_given += res.sr_bits;
2002 *num_bits = res.sr_bits;
ccd979bd
MF
2003 status = 0;
2004bail:
c1e8d35e
TM
2005 if (status)
2006 mlog_errno(status);
ccd979bd
MF
2007 return status;
2008}
2009
13821151 2010static void ocfs2_init_inode_ac_group(struct inode *dir,
abf1b3cb 2011 struct buffer_head *parent_di_bh,
13821151
TM
2012 struct ocfs2_alloc_context *ac)
2013{
abf1b3cb 2014 struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
13821151
TM
2015 /*
2016 * Try to allocate inodes from some specific group.
2017 *
2018 * If the parent dir has recorded the last group used in allocation,
2019 * cool, use it. Otherwise if we try to allocate new inode from the
2020 * same slot the parent dir belongs to, use the same chunk.
2021 *
2022 * We are very careful here to avoid the mistake of setting
2023 * ac_last_group to a group descriptor from a different (unlocked) slot.
2024 */
2025 if (OCFS2_I(dir)->ip_last_used_group &&
2026 OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2027 ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
abf1b3cb
TM
2028 else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2029 if (di->i_suballoc_loc)
2030 ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2031 else
2032 ac->ac_last_group = ocfs2_which_suballoc_group(
2033 le64_to_cpu(di->i_blkno),
2034 le16_to_cpu(di->i_suballoc_bit));
2035 }
13821151
TM
2036}
2037
2038static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2039 struct ocfs2_alloc_context *ac)
2040{
2041 OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2042 OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2043}
2044
e49e2767
MF
2045int ocfs2_find_new_inode_loc(struct inode *dir,
2046 struct buffer_head *parent_fe_bh,
2047 struct ocfs2_alloc_context *ac,
2048 u64 *fe_blkno)
2049{
2050 int ret;
2051 handle_t *handle = NULL;
2052 struct ocfs2_suballoc_result *res;
2053
2054 BUG_ON(!ac);
2055 BUG_ON(ac->ac_bits_given != 0);
2056 BUG_ON(ac->ac_bits_wanted != 1);
2057 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2058
2059 res = kzalloc(sizeof(*res), GFP_NOFS);
2060 if (res == NULL) {
2061 ret = -ENOMEM;
2062 mlog_errno(ret);
2063 goto out;
2064 }
2065
2066 ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2067
2068 /*
2069 * The handle started here is for chain relink. Alternatively,
2070 * we could just disable relink for these calls.
2071 */
2072 handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2073 if (IS_ERR(handle)) {
2074 ret = PTR_ERR(handle);
2075 handle = NULL;
2076 mlog_errno(ret);
2077 goto out;
2078 }
2079
2080 /*
2081 * This will instruct ocfs2_claim_suballoc_bits and
2082 * ocfs2_search_one_group to search but save actual allocation
2083 * for later.
2084 */
2085 ac->ac_find_loc_only = 1;
2086
2087 ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2088 if (ret < 0) {
2089 mlog_errno(ret);
2090 goto out;
2091 }
2092
2093 ac->ac_find_loc_priv = res;
2094 *fe_blkno = res->sr_blkno;
6fdb702d 2095 ocfs2_update_inode_fsync_trans(handle, dir, 0);
e49e2767
MF
2096out:
2097 if (handle)
2098 ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2099
2100 if (ret)
2101 kfree(res);
2102
2103 return ret;
2104}
2105
2106int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2107 struct inode *dir,
2108 struct ocfs2_alloc_context *ac,
2109 u64 *suballoc_loc,
2110 u16 *suballoc_bit,
2111 u64 di_blkno)
2112{
2113 int ret;
2114 u16 chain;
2115 struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2116 struct buffer_head *bg_bh = NULL;
2117 struct ocfs2_group_desc *bg;
2118 struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2119
2120 /*
2121 * Since di_blkno is being passed back in, we check for any
2122 * inconsistencies which may have happened between
2123 * calls. These are code bugs as di_blkno is not expected to
2124 * change once returned from ocfs2_find_new_inode_loc()
2125 */
2126 BUG_ON(res->sr_blkno != di_blkno);
2127
2128 ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2129 res->sr_bg_stable_blkno, &bg_bh);
2130 if (ret) {
2131 mlog_errno(ret);
2132 goto out;
2133 }
2134
2135 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2136 chain = le16_to_cpu(bg->bg_chain);
2137
2138 ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2139 ac->ac_bh, res->sr_bits,
2140 chain);
2141 if (ret) {
2142 mlog_errno(ret);
2143 goto out;
2144 }
2145
2146 ret = ocfs2_block_group_set_bits(handle,
2147 ac->ac_inode,
2148 bg,
2149 bg_bh,
2150 res->sr_bit_offset,
2151 res->sr_bits);
2152 if (ret < 0) {
2153 mlog_errno(ret);
2154 goto out;
2155 }
2156
2f73e135
TM
2157 trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2158 res->sr_bits);
e49e2767
MF
2159
2160 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2161
2162 BUG_ON(res->sr_bits != 1);
2163
2164 *suballoc_loc = res->sr_bg_blkno;
2165 *suballoc_bit = res->sr_bit_offset;
2166 ac->ac_bits_given++;
2167 ocfs2_save_inode_ac_group(dir, ac);
2168
2169out:
2170 brelse(bg_bh);
2171
2172 return ret;
2173}
2174
1ed9b777 2175int ocfs2_claim_new_inode(handle_t *handle,
13821151
TM
2176 struct inode *dir,
2177 struct buffer_head *parent_fe_bh,
ccd979bd 2178 struct ocfs2_alloc_context *ac,
2b6cb576 2179 u64 *suballoc_loc,
ccd979bd
MF
2180 u16 *suballoc_bit,
2181 u64 *fe_blkno)
2182{
2183 int status;
2b6cb576 2184 struct ocfs2_suballoc_result res;
ccd979bd 2185
ccd979bd
MF
2186 BUG_ON(!ac);
2187 BUG_ON(ac->ac_bits_given != 0);
2188 BUG_ON(ac->ac_bits_wanted != 1);
2189 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
ccd979bd 2190
13821151
TM
2191 ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2192
aa8f8e93 2193 status = ocfs2_claim_suballoc_bits(ac,
da5cbf2f 2194 handle,
ccd979bd
MF
2195 1,
2196 1,
7d1fe093 2197 &res);
ccd979bd
MF
2198 if (status < 0) {
2199 mlog_errno(status);
2200 goto bail;
2201 }
1ed9b777 2202 atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
ccd979bd 2203
7d1fe093 2204 BUG_ON(res.sr_bits != 1);
ccd979bd 2205
2b6cb576 2206 *suballoc_loc = res.sr_bg_blkno;
7d1fe093 2207 *suballoc_bit = res.sr_bit_offset;
ba206635 2208 *fe_blkno = res.sr_blkno;
ccd979bd 2209 ac->ac_bits_given++;
13821151 2210 ocfs2_save_inode_ac_group(dir, ac);
ccd979bd
MF
2211 status = 0;
2212bail:
c1e8d35e
TM
2213 if (status)
2214 mlog_errno(status);
ccd979bd
MF
2215 return status;
2216}
2217
2218/* translate a group desc. blkno and it's bitmap offset into
2219 * disk cluster offset. */
2220static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2221 u64 bg_blkno,
2222 u16 bg_bit_off)
2223{
2224 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2225 u32 cluster = 0;
2226
2227 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2228
2229 if (bg_blkno != osb->first_cluster_group_blkno)
2230 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2231 cluster += (u32) bg_bit_off;
2232 return cluster;
2233}
2234
2235/* given a cluster offset, calculate which block group it belongs to
2236 * and return that block offset. */
d659072f 2237u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
ccd979bd
MF
2238{
2239 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2240 u32 group_no;
2241
2242 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2243
2244 group_no = cluster / osb->bitmap_cpg;
2245 if (!group_no)
2246 return osb->first_cluster_group_blkno;
2247 return ocfs2_clusters_to_blocks(inode->i_sb,
2248 group_no * osb->bitmap_cpg);
2249}
2250
2251/* given the block number of a cluster start, calculate which cluster
2252 * group and descriptor bitmap offset that corresponds to. */
2253static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2254 u64 data_blkno,
2255 u64 *bg_blkno,
2256 u16 *bg_bit_off)
2257{
2258 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2259 u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2260
2261 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2262
2263 *bg_blkno = ocfs2_which_cluster_group(inode,
2264 data_cluster);
2265
2266 if (*bg_blkno == osb->first_cluster_group_blkno)
2267 *bg_bit_off = (u16) data_cluster;
2268 else
2269 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2270 data_blkno - *bg_blkno);
2271}
2272
2273/*
2274 * min_bits - minimum contiguous chunk from this total allocation we
2275 * can handle. set to what we asked for originally for a full
2276 * contig. allocation, set to '1' to indicate we can deal with extents
2277 * of any size.
2278 */
1ed9b777 2279int __ocfs2_claim_clusters(handle_t *handle,
415cb800
MF
2280 struct ocfs2_alloc_context *ac,
2281 u32 min_clusters,
2282 u32 max_clusters,
2283 u32 *cluster_start,
2284 u32 *num_clusters)
ccd979bd
MF
2285{
2286 int status;
415cb800 2287 unsigned int bits_wanted = max_clusters;
ba206635 2288 struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
1ed9b777 2289 struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
ccd979bd 2290
ccd979bd
MF
2291 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2292
2293 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2294 && ac->ac_which != OCFS2_AC_USE_MAIN);
ccd979bd
MF
2295
2296 if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
33d5d380
MF
2297 WARN_ON(min_clusters > 1);
2298
ccd979bd
MF
2299 status = ocfs2_claim_local_alloc_bits(osb,
2300 handle,
2301 ac,
2302 bits_wanted,
2303 cluster_start,
2304 num_clusters);
2305 if (!status)
2306 atomic_inc(&osb->alloc_stats.local_data);
2307 } else {
2308 if (min_clusters > (osb->bitmap_cpg - 1)) {
2309 /* The only paths asking for contiguousness
2310 * should know about this already. */
2fbe8d1e
SM
2311 mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2312 "group bitmap size %u!\n", min_clusters,
2313 osb->bitmap_cpg);
ccd979bd
MF
2314 status = -ENOSPC;
2315 goto bail;
2316 }
2317 /* clamp the current request down to a realistic size. */
2318 if (bits_wanted > (osb->bitmap_cpg - 1))
2319 bits_wanted = osb->bitmap_cpg - 1;
2320
aa8f8e93 2321 status = ocfs2_claim_suballoc_bits(ac,
da5cbf2f 2322 handle,
ccd979bd
MF
2323 bits_wanted,
2324 min_clusters,
7d1fe093 2325 &res);
ccd979bd 2326 if (!status) {
ba206635 2327 BUG_ON(res.sr_blkno); /* cluster alloc can't set */
ccd979bd
MF
2328 *cluster_start =
2329 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
7d1fe093
JB
2330 res.sr_bg_blkno,
2331 res.sr_bit_offset);
ccd979bd 2332 atomic_inc(&osb->alloc_stats.bitmap_data);
4711954e 2333 *num_clusters = res.sr_bits;
ccd979bd
MF
2334 }
2335 }
2336 if (status < 0) {
2337 if (status != -ENOSPC)
2338 mlog_errno(status);
2339 goto bail;
2340 }
2341
4711954e 2342 ac->ac_bits_given += *num_clusters;
ccd979bd
MF
2343
2344bail:
c1e8d35e
TM
2345 if (status)
2346 mlog_errno(status);
ccd979bd
MF
2347 return status;
2348}
2349
1ed9b777 2350int ocfs2_claim_clusters(handle_t *handle,
415cb800
MF
2351 struct ocfs2_alloc_context *ac,
2352 u32 min_clusters,
2353 u32 *cluster_start,
2354 u32 *num_clusters)
2355{
2356 unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2357
1ed9b777 2358 return __ocfs2_claim_clusters(handle, ac, min_clusters,
415cb800
MF
2359 bits_wanted, cluster_start, num_clusters);
2360}
2361
b4414eea
MF
2362static int ocfs2_block_group_clear_bits(handle_t *handle,
2363 struct inode *alloc_inode,
2364 struct ocfs2_group_desc *bg,
2365 struct buffer_head *group_bh,
2366 unsigned int bit_off,
2367 unsigned int num_bits,
2368 void (*undo_fn)(unsigned int bit,
2369 unsigned long *bmap))
ccd979bd
MF
2370{
2371 int status;
2372 unsigned int tmp;
ccd979bd
MF
2373 struct ocfs2_group_desc *undo_bg = NULL;
2374
42035306
JB
2375 /* The caller got this descriptor from
2376 * ocfs2_read_group_descriptor(). Any corruption is a code bug. */
2377 BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
ccd979bd 2378
2f73e135 2379 trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
ccd979bd 2380
b4414eea 2381 BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
0cf2f763 2382 status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
b4414eea
MF
2383 group_bh,
2384 undo_fn ?
2385 OCFS2_JOURNAL_ACCESS_UNDO :
2386 OCFS2_JOURNAL_ACCESS_WRITE);
ccd979bd
MF
2387 if (status < 0) {
2388 mlog_errno(status);
2389 goto bail;
2390 }
2391
b4414eea 2392 if (undo_fn) {
94e41ecf
SM
2393 jbd_lock_bh_state(group_bh);
2394 undo_bg = (struct ocfs2_group_desc *)
2395 bh2jh(group_bh)->b_committed_data;
2396 BUG_ON(!undo_bg);
2397 }
ccd979bd
MF
2398
2399 tmp = num_bits;
2400 while(tmp--) {
2401 ocfs2_clear_bit((bit_off + tmp),
2402 (unsigned long *) bg->bg_bitmap);
b4414eea
MF
2403 if (undo_fn)
2404 undo_fn(bit_off + tmp,
2405 (unsigned long *) undo_bg->bg_bitmap);
ccd979bd
MF
2406 }
2407 le16_add_cpu(&bg->bg_free_bits_count, num_bits);
9b5cd10e
SE
2408 if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2409 ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
2410 " count %u but claims %u are freed. num_bits %d",
2411 (unsigned long long)le64_to_cpu(bg->bg_blkno),
2412 le16_to_cpu(bg->bg_bits),
2413 le16_to_cpu(bg->bg_free_bits_count), num_bits);
2414 return -EROFS;
2415 }
ccd979bd 2416
b4414eea 2417 if (undo_fn)
94e41ecf
SM
2418 jbd_unlock_bh_state(group_bh);
2419
ec20cec7 2420 ocfs2_journal_dirty(handle, group_bh);
ccd979bd
MF
2421bail:
2422 return status;
2423}
2424
2425/*
2426 * expects the suballoc inode to already be locked.
2427 */
b4414eea
MF
2428static int _ocfs2_free_suballoc_bits(handle_t *handle,
2429 struct inode *alloc_inode,
2430 struct buffer_head *alloc_bh,
2431 unsigned int start_bit,
2432 u64 bg_blkno,
2433 unsigned int count,
2434 void (*undo_fn)(unsigned int bit,
2435 unsigned long *bitmap))
ccd979bd
MF
2436{
2437 int status = 0;
2438 u32 tmp_used;
ccd979bd
MF
2439 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2440 struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2441 struct buffer_head *group_bh = NULL;
2442 struct ocfs2_group_desc *group;
2443
10995aa2
JB
2444 /* The alloc_bh comes from ocfs2_free_dinode() or
2445 * ocfs2_free_clusters(). The callers have all locked the
2446 * allocator and gotten alloc_bh from the lock call. This
25985edc 2447 * validates the dinode buffer. Any corruption that has happened
10995aa2
JB
2448 * is a code bug. */
2449 BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
ccd979bd
MF
2450 BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2451
2f73e135
TM
2452 trace_ocfs2_free_suballoc_bits(
2453 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2454 (unsigned long long)bg_blkno,
2455 start_bit, count);
ccd979bd 2456
68f64d47
JB
2457 status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2458 &group_bh);
ccd979bd
MF
2459 if (status < 0) {
2460 mlog_errno(status);
2461 goto bail;
2462 }
ccd979bd 2463 group = (struct ocfs2_group_desc *) group_bh->b_data;
68f64d47 2464
ccd979bd
MF
2465 BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2466
2467 status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2468 group, group_bh,
b4414eea 2469 start_bit, count, undo_fn);
ccd979bd
MF
2470 if (status < 0) {
2471 mlog_errno(status);
2472 goto bail;
2473 }
2474
0cf2f763
JB
2475 status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2476 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
ccd979bd
MF
2477 if (status < 0) {
2478 mlog_errno(status);
2479 goto bail;
2480 }
2481
2482 le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2483 count);
2484 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2485 fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
ec20cec7 2486 ocfs2_journal_dirty(handle, alloc_bh);
ccd979bd
MF
2487
2488bail:
a81cb88b 2489 brelse(group_bh);
ccd979bd 2490
c1e8d35e
TM
2491 if (status)
2492 mlog_errno(status);
ccd979bd
MF
2493 return status;
2494}
2495
b4414eea
MF
2496int ocfs2_free_suballoc_bits(handle_t *handle,
2497 struct inode *alloc_inode,
2498 struct buffer_head *alloc_bh,
2499 unsigned int start_bit,
2500 u64 bg_blkno,
2501 unsigned int count)
2502{
2503 return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2504 start_bit, bg_blkno, count, NULL);
2505}
2506
1fabe148 2507int ocfs2_free_dinode(handle_t *handle,
ccd979bd
MF
2508 struct inode *inode_alloc_inode,
2509 struct buffer_head *inode_alloc_bh,
2510 struct ocfs2_dinode *di)
2511{
2512 u64 blk = le64_to_cpu(di->i_blkno);
2513 u16 bit = le16_to_cpu(di->i_suballoc_bit);
2514 u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2515
74380c47
TM
2516 if (di->i_suballoc_loc)
2517 bg_blkno = le64_to_cpu(di->i_suballoc_loc);
ccd979bd
MF
2518 return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2519 inode_alloc_bh, bit, bg_blkno, 1);
2520}
2521
b4414eea
MF
2522static int _ocfs2_free_clusters(handle_t *handle,
2523 struct inode *bitmap_inode,
2524 struct buffer_head *bitmap_bh,
2525 u64 start_blk,
2526 unsigned int num_clusters,
2527 void (*undo_fn)(unsigned int bit,
2528 unsigned long *bitmap))
ccd979bd
MF
2529{
2530 int status;
2531 u16 bg_start_bit;
2532 u64 bg_blkno;
2533 struct ocfs2_dinode *fe;
2534
2535 /* You can't ever have a contiguous set of clusters
2536 * bigger than a block group bitmap so we never have to worry
ef6b689b
TM
2537 * about looping on them.
2538 * This is expensive. We can safely remove once this stuff has
ccd979bd
MF
2539 * gotten tested really well. */
2540 BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
2541
2542 fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
2543
2544 ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2545 &bg_start_bit);
2546
2f73e135
TM
2547 trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2548 (unsigned long long)start_blk,
2549 bg_start_bit, num_clusters);
ccd979bd 2550
b4414eea
MF
2551 status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2552 bg_start_bit, bg_blkno,
2553 num_clusters, undo_fn);
9c7af40b 2554 if (status < 0) {
ccd979bd 2555 mlog_errno(status);
9c7af40b
MF
2556 goto out;
2557 }
2558
2559 ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2560 num_clusters);
ccd979bd 2561
9c7af40b 2562out:
c1e8d35e
TM
2563 if (status)
2564 mlog_errno(status);
ccd979bd
MF
2565 return status;
2566}
2567
b4414eea
MF
2568int ocfs2_free_clusters(handle_t *handle,
2569 struct inode *bitmap_inode,
2570 struct buffer_head *bitmap_bh,
2571 u64 start_blk,
2572 unsigned int num_clusters)
2573{
2574 return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2575 start_blk, num_clusters,
2576 _ocfs2_set_bit);
2577}
2578
2579/*
2580 * Give never-used clusters back to the global bitmap. We don't need
2581 * to protect these bits in the undo buffer.
2582 */
2583int ocfs2_release_clusters(handle_t *handle,
2584 struct inode *bitmap_inode,
2585 struct buffer_head *bitmap_bh,
2586 u64 start_blk,
2587 unsigned int num_clusters)
2588{
2589 return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2590 start_blk, num_clusters,
2591 _ocfs2_clear_bit);
2592}
2593
ccd979bd
MF
2594static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
2595{
2596 printk("Block Group:\n");
2597 printk("bg_signature: %s\n", bg->bg_signature);
2598 printk("bg_size: %u\n", bg->bg_size);
2599 printk("bg_bits: %u\n", bg->bg_bits);
2600 printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
2601 printk("bg_chain: %u\n", bg->bg_chain);
2602 printk("bg_generation: %u\n", le32_to_cpu(bg->bg_generation));
b0697053
MF
2603 printk("bg_next_group: %llu\n",
2604 (unsigned long long)bg->bg_next_group);
2605 printk("bg_parent_dinode: %llu\n",
2606 (unsigned long long)bg->bg_parent_dinode);
2607 printk("bg_blkno: %llu\n",
2608 (unsigned long long)bg->bg_blkno);
ccd979bd
MF
2609}
2610
2611static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2612{
2613 int i;
2614
b0697053 2615 printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
ccd979bd 2616 printk("i_signature: %s\n", fe->i_signature);
b0697053
MF
2617 printk("i_size: %llu\n",
2618 (unsigned long long)fe->i_size);
ccd979bd
MF
2619 printk("i_clusters: %u\n", fe->i_clusters);
2620 printk("i_generation: %u\n",
2621 le32_to_cpu(fe->i_generation));
2622 printk("id1.bitmap1.i_used: %u\n",
2623 le32_to_cpu(fe->id1.bitmap1.i_used));
2624 printk("id1.bitmap1.i_total: %u\n",
2625 le32_to_cpu(fe->id1.bitmap1.i_total));
2626 printk("id2.i_chain.cl_cpg: %u\n", fe->id2.i_chain.cl_cpg);
2627 printk("id2.i_chain.cl_bpc: %u\n", fe->id2.i_chain.cl_bpc);
2628 printk("id2.i_chain.cl_count: %u\n", fe->id2.i_chain.cl_count);
2629 printk("id2.i_chain.cl_next_free_rec: %u\n",
2630 fe->id2.i_chain.cl_next_free_rec);
2631 for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2632 printk("fe->id2.i_chain.cl_recs[%d].c_free: %u\n", i,
2633 fe->id2.i_chain.cl_recs[i].c_free);
2634 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2635 fe->id2.i_chain.cl_recs[i].c_total);
b0697053
MF
2636 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2637 (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
ccd979bd
MF
2638 }
2639}
e7d4cb6b
TM
2640
2641/*
2642 * For a given allocation, determine which allocators will need to be
2643 * accessed, and lock them, reserving the appropriate number of bits.
2644 *
2645 * Sparse file systems call this from ocfs2_write_begin_nolock()
2646 * and ocfs2_allocate_unwritten_extents().
2647 *
2648 * File systems which don't support holes call this from
2649 * ocfs2_extend_allocation().
2650 */
f99b9b7c
JB
2651int ocfs2_lock_allocators(struct inode *inode,
2652 struct ocfs2_extent_tree *et,
e7d4cb6b
TM
2653 u32 clusters_to_add, u32 extents_to_split,
2654 struct ocfs2_alloc_context **data_ac,
f99b9b7c 2655 struct ocfs2_alloc_context **meta_ac)
e7d4cb6b
TM
2656{
2657 int ret = 0, num_free_extents;
2658 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2659 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2660
2661 *meta_ac = NULL;
2662 if (data_ac)
2663 *data_ac = NULL;
2664
2665 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2666
3d03a305 2667 num_free_extents = ocfs2_num_free_extents(osb, et);
e7d4cb6b
TM
2668 if (num_free_extents < 0) {
2669 ret = num_free_extents;
2670 mlog_errno(ret);
2671 goto out;
2672 }
2673
2674 /*
2675 * Sparse allocation file systems need to be more conservative
2676 * with reserving room for expansion - the actual allocation
2677 * happens while we've got a journal handle open so re-taking
2678 * a cluster lock (because we ran out of room for another
2679 * extent) will violate ordering rules.
2680 *
2681 * Most of the time we'll only be seeing this 1 cluster at a time
2682 * anyway.
2683 *
2684 * Always lock for any unwritten extents - we might want to
2685 * add blocks during a split.
2686 */
2687 if (!num_free_extents ||
2688 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
f99b9b7c 2689 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
e7d4cb6b
TM
2690 if (ret < 0) {
2691 if (ret != -ENOSPC)
2692 mlog_errno(ret);
2693 goto out;
2694 }
2695 }
2696
2697 if (clusters_to_add == 0)
2698 goto out;
2699
2700 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2701 if (ret < 0) {
2702 if (ret != -ENOSPC)
2703 mlog_errno(ret);
2704 goto out;
2705 }
2706
2707out:
2708 if (ret) {
2709 if (*meta_ac) {
2710 ocfs2_free_alloc_context(*meta_ac);
2711 *meta_ac = NULL;
2712 }
2713
2714 /*
2715 * We cannot have an error and a non null *data_ac.
2716 */
2717 }
2718
2719 return ret;
2720}
6ca497a8 2721
2722/*
2723 * Read the inode specified by blkno to get suballoc_slot and
2724 * suballoc_bit.
2725 */
2726static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
889f004a
TM
2727 u16 *suballoc_slot, u64 *group_blkno,
2728 u16 *suballoc_bit)
6ca497a8 2729{
2730 int status;
2731 struct buffer_head *inode_bh = NULL;
2732 struct ocfs2_dinode *inode_fe;
2733
2f73e135 2734 trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
6ca497a8 2735
2736 /* dirty read disk */
2737 status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2738 if (status < 0) {
5b09b507
JB
2739 mlog(ML_ERROR, "read block %llu failed %d\n",
2740 (unsigned long long)blkno, status);
6ca497a8 2741 goto bail;
2742 }
2743
2744 inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2745 if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
5b09b507
JB
2746 mlog(ML_ERROR, "invalid inode %llu requested\n",
2747 (unsigned long long)blkno);
6ca497a8 2748 status = -EINVAL;
2749 goto bail;
2750 }
2751
0fba8137 2752 if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
6ca497a8 2753 (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2754 mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
5b09b507
JB
2755 (unsigned long long)blkno,
2756 (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
6ca497a8 2757 status = -EINVAL;
2758 goto bail;
2759 }
2760
2761 if (suballoc_slot)
2762 *suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2763 if (suballoc_bit)
2764 *suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
889f004a
TM
2765 if (group_blkno)
2766 *group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
6ca497a8 2767
2768bail:
2769 brelse(inode_bh);
2770
c1e8d35e
TM
2771 if (status)
2772 mlog_errno(status);
6ca497a8 2773 return status;
2774}
2775
2776/*
2777 * test whether bit is SET in allocator bitmap or not. on success, 0
2778 * is returned and *res is 1 for SET; 0 otherwise. when fails, errno
2779 * is returned and *res is meaningless. Call this after you have
2780 * cluster locked against suballoc, or you may get a result based on
2781 * non-up2date contents
2782 */
2783static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2784 struct inode *suballoc,
889f004a
TM
2785 struct buffer_head *alloc_bh,
2786 u64 group_blkno, u64 blkno,
6ca497a8 2787 u16 bit, int *res)
2788{
abf1b3cb 2789 struct ocfs2_dinode *alloc_di;
6ca497a8 2790 struct ocfs2_group_desc *group;
2791 struct buffer_head *group_bh = NULL;
2792 u64 bg_blkno;
2793 int status;
2794
2f73e135
TM
2795 trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2796 (unsigned int)bit);
6ca497a8 2797
abf1b3cb
TM
2798 alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2799 if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
6ca497a8 2800 mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2801 (unsigned int)bit,
abf1b3cb 2802 ocfs2_bits_per_group(&alloc_di->id2.i_chain));
6ca497a8 2803 status = -EINVAL;
2804 goto bail;
2805 }
2806
889f004a
TM
2807 bg_blkno = group_blkno ? group_blkno :
2808 ocfs2_which_suballoc_group(blkno, bit);
abf1b3cb 2809 status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
6ca497a8 2810 &group_bh);
2811 if (status < 0) {
5b09b507
JB
2812 mlog(ML_ERROR, "read group %llu failed %d\n",
2813 (unsigned long long)bg_blkno, status);
6ca497a8 2814 goto bail;
2815 }
2816
2817 group = (struct ocfs2_group_desc *) group_bh->b_data;
2818 *res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2819
2820bail:
2821 brelse(group_bh);
2822
c1e8d35e
TM
2823 if (status)
2824 mlog_errno(status);
6ca497a8 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
2f73e135 2849 trace_ocfs2_test_inode_bit((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:
c1e8d35e
TM
2890 if (status)
2891 mlog_errno(status);
6ca497a8 2892 return status;
2893}