481017e1dac5ac64100850f4b8b59f84f2a7c9f6
[platform/kernel/linux-rpi.git] / fs / ocfs2 / suballoc.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * suballoc.c
4  *
5  * metadata alloc and free
6  * Inspired by ext3 block groups.
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  */
10
11 #include <linux/fs.h>
12 #include <linux/types.h>
13 #include <linux/slab.h>
14 #include <linux/highmem.h>
15
16 #include <cluster/masklog.h>
17
18 #include "ocfs2.h"
19
20 #include "alloc.h"
21 #include "blockcheck.h"
22 #include "dlmglue.h"
23 #include "inode.h"
24 #include "journal.h"
25 #include "localalloc.h"
26 #include "suballoc.h"
27 #include "super.h"
28 #include "sysfile.h"
29 #include "uptodate.h"
30 #include "ocfs2_trace.h"
31
32 #include "buffer_head_io.h"
33
34 #define NOT_ALLOC_NEW_GROUP             0
35 #define ALLOC_NEW_GROUP                 0x1
36 #define ALLOC_GROUPS_FROM_GLOBAL        0x2
37
38 #define OCFS2_MAX_TO_STEAL              1024
39
40 struct ocfs2_suballoc_result {
41         u64             sr_bg_blkno;    /* The bg we allocated from.  Set
42                                            to 0 when a block group is
43                                            contiguous. */
44         u64             sr_bg_stable_blkno; /*
45                                              * Doesn't change, always
46                                              * set to target block
47                                              * group descriptor
48                                              * block.
49                                              */
50         u64             sr_blkno;       /* The first allocated block */
51         unsigned int    sr_bit_offset;  /* The bit in the bg */
52         unsigned int    sr_bits;        /* How many bits we claimed */
53 };
54
55 static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
56 {
57         if (res->sr_blkno == 0)
58                 return 0;
59
60         if (res->sr_bg_blkno)
61                 return res->sr_bg_blkno;
62
63         return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset);
64 }
65
66 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
67 static int ocfs2_block_group_fill(handle_t *handle,
68                                   struct inode *alloc_inode,
69                                   struct buffer_head *bg_bh,
70                                   u64 group_blkno,
71                                   unsigned int group_clusters,
72                                   u16 my_chain,
73                                   struct ocfs2_chain_list *cl);
74 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
75                                    struct inode *alloc_inode,
76                                    struct buffer_head *bh,
77                                    u64 max_block,
78                                    u64 *last_alloc_group,
79                                    int flags);
80
81 static int ocfs2_cluster_group_search(struct inode *inode,
82                                       struct buffer_head *group_bh,
83                                       u32 bits_wanted, u32 min_bits,
84                                       u64 max_block,
85                                       struct ocfs2_suballoc_result *res);
86 static int ocfs2_block_group_search(struct inode *inode,
87                                     struct buffer_head *group_bh,
88                                     u32 bits_wanted, u32 min_bits,
89                                     u64 max_block,
90                                     struct ocfs2_suballoc_result *res);
91 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
92                                      handle_t *handle,
93                                      u32 bits_wanted,
94                                      u32 min_bits,
95                                      struct ocfs2_suballoc_result *res);
96 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
97                                          int nr);
98 static int ocfs2_relink_block_group(handle_t *handle,
99                                     struct inode *alloc_inode,
100                                     struct buffer_head *fe_bh,
101                                     struct buffer_head *bg_bh,
102                                     struct buffer_head *prev_bg_bh,
103                                     u16 chain);
104 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
105                                                      u32 wanted);
106 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
107                                                    u64 bg_blkno,
108                                                    u16 bg_bit_off);
109 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
110                                                 u64 data_blkno,
111                                                 u64 *bg_blkno,
112                                                 u16 *bg_bit_off);
113 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
114                                              u32 bits_wanted, u64 max_block,
115                                              int flags,
116                                              struct ocfs2_alloc_context **ac);
117
118 void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
119 {
120         struct inode *inode = ac->ac_inode;
121
122         if (inode) {
123                 if (ac->ac_which != OCFS2_AC_USE_LOCAL)
124                         ocfs2_inode_unlock(inode, 1);
125
126                 inode_unlock(inode);
127
128                 iput(inode);
129                 ac->ac_inode = NULL;
130         }
131         brelse(ac->ac_bh);
132         ac->ac_bh = NULL;
133         ac->ac_resv = NULL;
134         kfree(ac->ac_find_loc_priv);
135         ac->ac_find_loc_priv = NULL;
136 }
137
138 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
139 {
140         ocfs2_free_ac_resource(ac);
141         kfree(ac);
142 }
143
144 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
145 {
146         return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
147 }
148
149 #define do_error(fmt, ...)                                              \
150 do {                                                                    \
151         if (resize)                                                     \
152                 mlog(ML_ERROR, fmt, ##__VA_ARGS__);                     \
153         else                                                            \
154                 return ocfs2_error(sb, fmt, ##__VA_ARGS__);             \
155 } while (0)
156
157 static int ocfs2_validate_gd_self(struct super_block *sb,
158                                   struct buffer_head *bh,
159                                   int resize)
160 {
161         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
162
163         if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
164                 do_error("Group descriptor #%llu has bad signature %.*s\n",
165                          (unsigned long long)bh->b_blocknr, 7,
166                          gd->bg_signature);
167         }
168
169         if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
170                 do_error("Group descriptor #%llu has an invalid bg_blkno of %llu\n",
171                          (unsigned long long)bh->b_blocknr,
172                          (unsigned long long)le64_to_cpu(gd->bg_blkno));
173         }
174
175         if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
176                 do_error("Group descriptor #%llu has an invalid fs_generation of #%u\n",
177                          (unsigned long long)bh->b_blocknr,
178                          le32_to_cpu(gd->bg_generation));
179         }
180
181         if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
182                 do_error("Group descriptor #%llu has bit count %u but claims that %u are free\n",
183                          (unsigned long long)bh->b_blocknr,
184                          le16_to_cpu(gd->bg_bits),
185                          le16_to_cpu(gd->bg_free_bits_count));
186         }
187
188         if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
189                 do_error("Group descriptor #%llu has bit count %u but max bitmap bits of %u\n",
190                          (unsigned long long)bh->b_blocknr,
191                          le16_to_cpu(gd->bg_bits),
192                          8 * le16_to_cpu(gd->bg_size));
193         }
194
195         return 0;
196 }
197
198 static int ocfs2_validate_gd_parent(struct super_block *sb,
199                                     struct ocfs2_dinode *di,
200                                     struct buffer_head *bh,
201                                     int resize)
202 {
203         unsigned int max_bits;
204         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
205
206         if (di->i_blkno != gd->bg_parent_dinode) {
207                 do_error("Group descriptor #%llu has bad parent pointer (%llu, expected %llu)\n",
208                          (unsigned long long)bh->b_blocknr,
209                          (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
210                          (unsigned long long)le64_to_cpu(di->i_blkno));
211         }
212
213         max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
214         if (le16_to_cpu(gd->bg_bits) > max_bits) {
215                 do_error("Group descriptor #%llu has bit count of %u\n",
216                          (unsigned long long)bh->b_blocknr,
217                          le16_to_cpu(gd->bg_bits));
218         }
219
220         /* In resize, we may meet the case bg_chain == cl_next_free_rec. */
221         if ((le16_to_cpu(gd->bg_chain) >
222              le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
223             ((le16_to_cpu(gd->bg_chain) ==
224              le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
225                 do_error("Group descriptor #%llu has bad chain %u\n",
226                          (unsigned long long)bh->b_blocknr,
227                          le16_to_cpu(gd->bg_chain));
228         }
229
230         return 0;
231 }
232
233 #undef do_error
234
235 /*
236  * This version only prints errors.  It does not fail the filesystem, and
237  * exists only for resize.
238  */
239 int ocfs2_check_group_descriptor(struct super_block *sb,
240                                  struct ocfs2_dinode *di,
241                                  struct buffer_head *bh)
242 {
243         int rc;
244         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
245
246         BUG_ON(!buffer_uptodate(bh));
247
248         /*
249          * If the ecc fails, we return the error but otherwise
250          * leave the filesystem running.  We know any error is
251          * local to this block.
252          */
253         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
254         if (rc) {
255                 mlog(ML_ERROR,
256                      "Checksum failed for group descriptor %llu\n",
257                      (unsigned long long)bh->b_blocknr);
258         } else
259                 rc = ocfs2_validate_gd_self(sb, bh, 1);
260         if (!rc)
261                 rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
262
263         return rc;
264 }
265
266 static int ocfs2_validate_group_descriptor(struct super_block *sb,
267                                            struct buffer_head *bh)
268 {
269         int rc;
270         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
271
272         trace_ocfs2_validate_group_descriptor(
273                                         (unsigned long long)bh->b_blocknr);
274
275         BUG_ON(!buffer_uptodate(bh));
276
277         /*
278          * If the ecc fails, we return the error but otherwise
279          * leave the filesystem running.  We know any error is
280          * local to this block.
281          */
282         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
283         if (rc)
284                 return rc;
285
286         /*
287          * Errors after here are fatal.
288          */
289
290         return ocfs2_validate_gd_self(sb, bh, 0);
291 }
292
293 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
294                                 u64 gd_blkno, struct buffer_head **bh)
295 {
296         int rc;
297         struct buffer_head *tmp = *bh;
298
299         rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
300                               ocfs2_validate_group_descriptor);
301         if (rc)
302                 goto out;
303
304         rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
305         if (rc) {
306                 brelse(tmp);
307                 goto out;
308         }
309
310         /* If ocfs2_read_block() got us a new bh, pass it up. */
311         if (!*bh)
312                 *bh = tmp;
313
314 out:
315         return rc;
316 }
317
318 static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
319                                           struct ocfs2_group_desc *bg,
320                                           struct ocfs2_chain_list *cl,
321                                           u64 p_blkno, unsigned int clusters)
322 {
323         struct ocfs2_extent_list *el = &bg->bg_list;
324         struct ocfs2_extent_rec *rec;
325
326         BUG_ON(!ocfs2_supports_discontig_bg(osb));
327         if (!el->l_next_free_rec)
328                 el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
329         rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
330         rec->e_blkno = cpu_to_le64(p_blkno);
331         rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
332                                   le16_to_cpu(cl->cl_bpc));
333         rec->e_leaf_clusters = cpu_to_le16(clusters);
334         le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
335         le16_add_cpu(&bg->bg_free_bits_count,
336                      clusters * le16_to_cpu(cl->cl_bpc));
337         le16_add_cpu(&el->l_next_free_rec, 1);
338 }
339
340 static int ocfs2_block_group_fill(handle_t *handle,
341                                   struct inode *alloc_inode,
342                                   struct buffer_head *bg_bh,
343                                   u64 group_blkno,
344                                   unsigned int group_clusters,
345                                   u16 my_chain,
346                                   struct ocfs2_chain_list *cl)
347 {
348         int status = 0;
349         struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
350         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
351         struct super_block * sb = alloc_inode->i_sb;
352
353         if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
354                 status = ocfs2_error(alloc_inode->i_sb,
355                                      "group block (%llu) != b_blocknr (%llu)\n",
356                                      (unsigned long long)group_blkno,
357                                      (unsigned long long) bg_bh->b_blocknr);
358                 goto bail;
359         }
360
361         status = ocfs2_journal_access_gd(handle,
362                                          INODE_CACHE(alloc_inode),
363                                          bg_bh,
364                                          OCFS2_JOURNAL_ACCESS_CREATE);
365         if (status < 0) {
366                 mlog_errno(status);
367                 goto bail;
368         }
369
370         memset(bg, 0, sb->s_blocksize);
371         strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
372         bg->bg_generation = cpu_to_le32(osb->fs_generation);
373         bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
374                                                 osb->s_feature_incompat));
375         bg->bg_chain = cpu_to_le16(my_chain);
376         bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
377         bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
378         bg->bg_blkno = cpu_to_le64(group_blkno);
379         if (group_clusters == le16_to_cpu(cl->cl_cpg))
380                 bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
381         else
382                 ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
383                                               group_clusters);
384
385         /* set the 1st bit in the bitmap to account for the descriptor block */
386         ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
387         bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
388
389         ocfs2_journal_dirty(handle, bg_bh);
390
391         /* There is no need to zero out or otherwise initialize the
392          * other blocks in a group - All valid FS metadata in a block
393          * group stores the superblock fs_generation value at
394          * allocation time. */
395
396 bail:
397         if (status)
398                 mlog_errno(status);
399         return status;
400 }
401
402 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
403 {
404         u16 curr, best;
405
406         best = curr = 0;
407         while (curr < le16_to_cpu(cl->cl_count)) {
408                 if (le32_to_cpu(cl->cl_recs[best].c_total) >
409                     le32_to_cpu(cl->cl_recs[curr].c_total))
410                         best = curr;
411                 curr++;
412         }
413         return best;
414 }
415
416 static struct buffer_head *
417 ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
418                                struct inode *alloc_inode,
419                                struct ocfs2_alloc_context *ac,
420                                struct ocfs2_chain_list *cl)
421 {
422         int status;
423         u32 bit_off, num_bits;
424         u64 bg_blkno;
425         struct buffer_head *bg_bh;
426         unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
427
428         status = ocfs2_claim_clusters(handle, ac,
429                                       le16_to_cpu(cl->cl_cpg), &bit_off,
430                                       &num_bits);
431         if (status < 0) {
432                 if (status != -ENOSPC)
433                         mlog_errno(status);
434                 goto bail;
435         }
436
437         /* setup the group */
438         bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
439         trace_ocfs2_block_group_alloc_contig(
440              (unsigned long long)bg_blkno, alloc_rec);
441
442         bg_bh = sb_getblk(osb->sb, bg_blkno);
443         if (!bg_bh) {
444                 status = -ENOMEM;
445                 mlog_errno(status);
446                 goto bail;
447         }
448         ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
449
450         status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
451                                         bg_blkno, num_bits, alloc_rec, cl);
452         if (status < 0) {
453                 brelse(bg_bh);
454                 mlog_errno(status);
455         }
456
457 bail:
458         return status ? ERR_PTR(status) : bg_bh;
459 }
460
461 static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
462                                         handle_t *handle,
463                                         struct ocfs2_alloc_context *ac,
464                                         unsigned int min_bits,
465                                         u32 *bit_off, u32 *num_bits)
466 {
467         int status = 0;
468
469         while (min_bits) {
470                 status = ocfs2_claim_clusters(handle, ac, min_bits,
471                                               bit_off, num_bits);
472                 if (status != -ENOSPC)
473                         break;
474
475                 min_bits >>= 1;
476         }
477
478         return status;
479 }
480
481 static int ocfs2_block_group_grow_discontig(handle_t *handle,
482                                             struct inode *alloc_inode,
483                                             struct buffer_head *bg_bh,
484                                             struct ocfs2_alloc_context *ac,
485                                             struct ocfs2_chain_list *cl,
486                                             unsigned int min_bits)
487 {
488         int status;
489         struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
490         struct ocfs2_group_desc *bg =
491                 (struct ocfs2_group_desc *)bg_bh->b_data;
492         unsigned int needed = le16_to_cpu(cl->cl_cpg) -
493                          le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
494         u32 p_cpos, clusters;
495         u64 p_blkno;
496         struct ocfs2_extent_list *el = &bg->bg_list;
497
498         status = ocfs2_journal_access_gd(handle,
499                                          INODE_CACHE(alloc_inode),
500                                          bg_bh,
501                                          OCFS2_JOURNAL_ACCESS_CREATE);
502         if (status < 0) {
503                 mlog_errno(status);
504                 goto bail;
505         }
506
507         while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
508                                 le16_to_cpu(el->l_count))) {
509                 if (min_bits > needed)
510                         min_bits = needed;
511                 status = ocfs2_block_group_claim_bits(osb, handle, ac,
512                                                       min_bits, &p_cpos,
513                                                       &clusters);
514                 if (status < 0) {
515                         if (status != -ENOSPC)
516                                 mlog_errno(status);
517                         goto bail;
518                 }
519                 p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
520                 ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
521                                               clusters);
522
523                 min_bits = clusters;
524                 needed = le16_to_cpu(cl->cl_cpg) -
525                          le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
526         }
527
528         if (needed > 0) {
529                 /*
530                  * We have used up all the extent rec but can't fill up
531                  * the cpg. So bail out.
532                  */
533                 status = -ENOSPC;
534                 goto bail;
535         }
536
537         ocfs2_journal_dirty(handle, bg_bh);
538
539 bail:
540         return status;
541 }
542
543 static void ocfs2_bg_alloc_cleanup(handle_t *handle,
544                                    struct ocfs2_alloc_context *cluster_ac,
545                                    struct inode *alloc_inode,
546                                    struct buffer_head *bg_bh)
547 {
548         int i, ret;
549         struct ocfs2_group_desc *bg;
550         struct ocfs2_extent_list *el;
551         struct ocfs2_extent_rec *rec;
552
553         if (!bg_bh)
554                 return;
555
556         bg = (struct ocfs2_group_desc *)bg_bh->b_data;
557         el = &bg->bg_list;
558         for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
559                 rec = &el->l_recs[i];
560                 ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
561                                           cluster_ac->ac_bh,
562                                           le64_to_cpu(rec->e_blkno),
563                                           le16_to_cpu(rec->e_leaf_clusters));
564                 if (ret)
565                         mlog_errno(ret);
566                 /* Try all the clusters to free */
567         }
568
569         ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
570         brelse(bg_bh);
571 }
572
573 static struct buffer_head *
574 ocfs2_block_group_alloc_discontig(handle_t *handle,
575                                   struct inode *alloc_inode,
576                                   struct ocfs2_alloc_context *ac,
577                                   struct ocfs2_chain_list *cl)
578 {
579         int status;
580         u32 bit_off, num_bits;
581         u64 bg_blkno;
582         unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
583         struct buffer_head *bg_bh = NULL;
584         unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
585         struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
586
587         if (!ocfs2_supports_discontig_bg(osb)) {
588                 status = -ENOSPC;
589                 goto bail;
590         }
591
592         status = ocfs2_extend_trans(handle,
593                                     ocfs2_calc_bg_discontig_credits(osb->sb));
594         if (status) {
595                 mlog_errno(status);
596                 goto bail;
597         }
598
599         /*
600          * We're going to be grabbing from multiple cluster groups.
601          * We don't have enough credits to relink them all, and the
602          * cluster groups will be staying in cache for the duration of
603          * this operation.
604          */
605         ac->ac_disable_chain_relink = 1;
606
607         /* Claim the first region */
608         status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
609                                               &bit_off, &num_bits);
610         if (status < 0) {
611                 if (status != -ENOSPC)
612                         mlog_errno(status);
613                 goto bail;
614         }
615         min_bits = num_bits;
616
617         /* setup the group */
618         bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
619         trace_ocfs2_block_group_alloc_discontig(
620                                 (unsigned long long)bg_blkno, alloc_rec);
621
622         bg_bh = sb_getblk(osb->sb, bg_blkno);
623         if (!bg_bh) {
624                 status = -ENOMEM;
625                 mlog_errno(status);
626                 goto bail;
627         }
628         ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
629
630         status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
631                                         bg_blkno, num_bits, alloc_rec, cl);
632         if (status < 0) {
633                 mlog_errno(status);
634                 goto bail;
635         }
636
637         status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
638                                                   bg_bh, ac, cl, min_bits);
639         if (status)
640                 mlog_errno(status);
641
642 bail:
643         if (status)
644                 ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
645         return status ? ERR_PTR(status) : bg_bh;
646 }
647
648 /*
649  * We expect the block group allocator to already be locked.
650  */
651 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
652                                    struct inode *alloc_inode,
653                                    struct buffer_head *bh,
654                                    u64 max_block,
655                                    u64 *last_alloc_group,
656                                    int flags)
657 {
658         int status, credits;
659         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
660         struct ocfs2_chain_list *cl;
661         struct ocfs2_alloc_context *ac = NULL;
662         handle_t *handle = NULL;
663         u16 alloc_rec;
664         struct buffer_head *bg_bh = NULL;
665         struct ocfs2_group_desc *bg;
666
667         BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
668
669         cl = &fe->id2.i_chain;
670         status = ocfs2_reserve_clusters_with_limit(osb,
671                                                    le16_to_cpu(cl->cl_cpg),
672                                                    max_block, flags, &ac);
673         if (status < 0) {
674                 if (status != -ENOSPC)
675                         mlog_errno(status);
676                 goto bail;
677         }
678
679         credits = ocfs2_calc_group_alloc_credits(osb->sb,
680                                                  le16_to_cpu(cl->cl_cpg));
681         handle = ocfs2_start_trans(osb, credits);
682         if (IS_ERR(handle)) {
683                 status = PTR_ERR(handle);
684                 handle = NULL;
685                 mlog_errno(status);
686                 goto bail;
687         }
688
689         if (last_alloc_group && *last_alloc_group != 0) {
690                 trace_ocfs2_block_group_alloc(
691                                 (unsigned long long)*last_alloc_group);
692                 ac->ac_last_group = *last_alloc_group;
693         }
694
695         bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
696                                                ac, cl);
697         if (PTR_ERR(bg_bh) == -ENOSPC)
698                 bg_bh = ocfs2_block_group_alloc_discontig(handle,
699                                                           alloc_inode,
700                                                           ac, cl);
701         if (IS_ERR(bg_bh)) {
702                 status = PTR_ERR(bg_bh);
703                 bg_bh = NULL;
704                 if (status != -ENOSPC)
705                         mlog_errno(status);
706                 goto bail;
707         }
708         bg = (struct ocfs2_group_desc *) bg_bh->b_data;
709
710         status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
711                                          bh, OCFS2_JOURNAL_ACCESS_WRITE);
712         if (status < 0) {
713                 mlog_errno(status);
714                 goto bail;
715         }
716
717         alloc_rec = le16_to_cpu(bg->bg_chain);
718         le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
719                      le16_to_cpu(bg->bg_free_bits_count));
720         le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
721                      le16_to_cpu(bg->bg_bits));
722         cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
723         if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
724                 le16_add_cpu(&cl->cl_next_free_rec, 1);
725
726         le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
727                                         le16_to_cpu(bg->bg_free_bits_count));
728         le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
729         le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
730
731         ocfs2_journal_dirty(handle, bh);
732
733         spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
734         OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
735         fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
736                                              le32_to_cpu(fe->i_clusters)));
737         spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
738         i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
739         alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
740         ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
741
742         status = 0;
743
744         /* save the new last alloc group so that the caller can cache it. */
745         if (last_alloc_group)
746                 *last_alloc_group = ac->ac_last_group;
747
748 bail:
749         if (handle)
750                 ocfs2_commit_trans(osb, handle);
751
752         if (ac)
753                 ocfs2_free_alloc_context(ac);
754
755         brelse(bg_bh);
756
757         if (status)
758                 mlog_errno(status);
759         return status;
760 }
761
762 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
763                                        struct ocfs2_alloc_context *ac,
764                                        int type,
765                                        u32 slot,
766                                        u64 *last_alloc_group,
767                                        int flags)
768 {
769         int status;
770         u32 bits_wanted = ac->ac_bits_wanted;
771         struct inode *alloc_inode;
772         struct buffer_head *bh = NULL;
773         struct ocfs2_dinode *fe;
774         u32 free_bits;
775
776         alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
777         if (!alloc_inode) {
778                 mlog_errno(-EINVAL);
779                 return -EINVAL;
780         }
781
782         inode_lock(alloc_inode);
783
784         status = ocfs2_inode_lock(alloc_inode, &bh, 1);
785         if (status < 0) {
786                 inode_unlock(alloc_inode);
787                 iput(alloc_inode);
788
789                 mlog_errno(status);
790                 return status;
791         }
792
793         ac->ac_inode = alloc_inode;
794         ac->ac_alloc_slot = slot;
795
796         fe = (struct ocfs2_dinode *) bh->b_data;
797
798         /* The bh was validated by the inode read inside
799          * ocfs2_inode_lock().  Any corruption is a code bug. */
800         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
801
802         if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
803                 status = ocfs2_error(alloc_inode->i_sb,
804                                      "Invalid chain allocator %llu\n",
805                                      (unsigned long long)le64_to_cpu(fe->i_blkno));
806                 goto bail;
807         }
808
809         free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
810                 le32_to_cpu(fe->id1.bitmap1.i_used);
811
812         if (bits_wanted > free_bits) {
813                 /* cluster bitmap never grows */
814                 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
815                         trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
816                                                                 free_bits);
817                         status = -ENOSPC;
818                         goto bail;
819                 }
820
821                 if (!(flags & ALLOC_NEW_GROUP)) {
822                         trace_ocfs2_reserve_suballoc_bits_no_new_group(
823                                                 slot, bits_wanted, free_bits);
824                         status = -ENOSPC;
825                         goto bail;
826                 }
827
828                 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
829                                                  ac->ac_max_block,
830                                                  last_alloc_group, flags);
831                 if (status < 0) {
832                         if (status != -ENOSPC)
833                                 mlog_errno(status);
834                         goto bail;
835                 }
836                 atomic_inc(&osb->alloc_stats.bg_extends);
837
838                 /* You should never ask for this much metadata */
839                 BUG_ON(bits_wanted >
840                        (le32_to_cpu(fe->id1.bitmap1.i_total)
841                         - le32_to_cpu(fe->id1.bitmap1.i_used)));
842         }
843
844         get_bh(bh);
845         ac->ac_bh = bh;
846 bail:
847         brelse(bh);
848
849         if (status)
850                 mlog_errno(status);
851         return status;
852 }
853
854 static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
855 {
856         spin_lock(&osb->osb_lock);
857         osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
858         spin_unlock(&osb->osb_lock);
859         atomic_set(&osb->s_num_inodes_stolen, 0);
860 }
861
862 static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
863 {
864         spin_lock(&osb->osb_lock);
865         osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
866         spin_unlock(&osb->osb_lock);
867         atomic_set(&osb->s_num_meta_stolen, 0);
868 }
869
870 void ocfs2_init_steal_slots(struct ocfs2_super *osb)
871 {
872         ocfs2_init_inode_steal_slot(osb);
873         ocfs2_init_meta_steal_slot(osb);
874 }
875
876 static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
877 {
878         spin_lock(&osb->osb_lock);
879         if (type == INODE_ALLOC_SYSTEM_INODE)
880                 osb->s_inode_steal_slot = (u16)slot;
881         else if (type == EXTENT_ALLOC_SYSTEM_INODE)
882                 osb->s_meta_steal_slot = (u16)slot;
883         spin_unlock(&osb->osb_lock);
884 }
885
886 static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
887 {
888         int slot = OCFS2_INVALID_SLOT;
889
890         spin_lock(&osb->osb_lock);
891         if (type == INODE_ALLOC_SYSTEM_INODE)
892                 slot = osb->s_inode_steal_slot;
893         else if (type == EXTENT_ALLOC_SYSTEM_INODE)
894                 slot = osb->s_meta_steal_slot;
895         spin_unlock(&osb->osb_lock);
896
897         return slot;
898 }
899
900 static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
901 {
902         return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
903 }
904
905 static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
906 {
907         return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
908 }
909
910 static int ocfs2_steal_resource(struct ocfs2_super *osb,
911                                 struct ocfs2_alloc_context *ac,
912                                 int type)
913 {
914         int i, status = -ENOSPC;
915         int slot = __ocfs2_get_steal_slot(osb, type);
916
917         /* Start to steal resource from the first slot after ours. */
918         if (slot == OCFS2_INVALID_SLOT)
919                 slot = osb->slot_num + 1;
920
921         for (i = 0; i < osb->max_slots; i++, slot++) {
922                 if (slot == osb->max_slots)
923                         slot = 0;
924
925                 if (slot == osb->slot_num)
926                         continue;
927
928                 status = ocfs2_reserve_suballoc_bits(osb, ac,
929                                                      type,
930                                                      (u32)slot, NULL,
931                                                      NOT_ALLOC_NEW_GROUP);
932                 if (status >= 0) {
933                         __ocfs2_set_steal_slot(osb, slot, type);
934                         break;
935                 }
936
937                 ocfs2_free_ac_resource(ac);
938         }
939
940         return status;
941 }
942
943 static int ocfs2_steal_inode(struct ocfs2_super *osb,
944                              struct ocfs2_alloc_context *ac)
945 {
946         return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
947 }
948
949 static int ocfs2_steal_meta(struct ocfs2_super *osb,
950                             struct ocfs2_alloc_context *ac)
951 {
952         return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
953 }
954
955 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
956                                       int blocks,
957                                       struct ocfs2_alloc_context **ac)
958 {
959         int status;
960         int slot = ocfs2_get_meta_steal_slot(osb);
961
962         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
963         if (!(*ac)) {
964                 status = -ENOMEM;
965                 mlog_errno(status);
966                 goto bail;
967         }
968
969         (*ac)->ac_bits_wanted = blocks;
970         (*ac)->ac_which = OCFS2_AC_USE_META;
971         (*ac)->ac_group_search = ocfs2_block_group_search;
972
973         if (slot != OCFS2_INVALID_SLOT &&
974                 atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
975                 goto extent_steal;
976
977         atomic_set(&osb->s_num_meta_stolen, 0);
978         status = ocfs2_reserve_suballoc_bits(osb, (*ac),
979                                              EXTENT_ALLOC_SYSTEM_INODE,
980                                              (u32)osb->slot_num, NULL,
981                                              ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
982
983
984         if (status >= 0) {
985                 status = 0;
986                 if (slot != OCFS2_INVALID_SLOT)
987                         ocfs2_init_meta_steal_slot(osb);
988                 goto bail;
989         } else if (status < 0 && status != -ENOSPC) {
990                 mlog_errno(status);
991                 goto bail;
992         }
993
994         ocfs2_free_ac_resource(*ac);
995
996 extent_steal:
997         status = ocfs2_steal_meta(osb, *ac);
998         atomic_inc(&osb->s_num_meta_stolen);
999         if (status < 0) {
1000                 if (status != -ENOSPC)
1001                         mlog_errno(status);
1002                 goto bail;
1003         }
1004
1005         status = 0;
1006 bail:
1007         if ((status < 0) && *ac) {
1008                 ocfs2_free_alloc_context(*ac);
1009                 *ac = NULL;
1010         }
1011
1012         if (status)
1013                 mlog_errno(status);
1014         return status;
1015 }
1016
1017 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1018                                struct ocfs2_extent_list *root_el,
1019                                struct ocfs2_alloc_context **ac)
1020 {
1021         return ocfs2_reserve_new_metadata_blocks(osb,
1022                                         ocfs2_extend_meta_needed(root_el),
1023                                         ac);
1024 }
1025
1026 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
1027                             struct ocfs2_alloc_context **ac)
1028 {
1029         int status;
1030         int slot = ocfs2_get_inode_steal_slot(osb);
1031         u64 alloc_group;
1032
1033         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1034         if (!(*ac)) {
1035                 status = -ENOMEM;
1036                 mlog_errno(status);
1037                 goto bail;
1038         }
1039
1040         (*ac)->ac_bits_wanted = 1;
1041         (*ac)->ac_which = OCFS2_AC_USE_INODE;
1042
1043         (*ac)->ac_group_search = ocfs2_block_group_search;
1044
1045         /*
1046          * stat(2) can't handle i_ino > 32bits, so we tell the
1047          * lower levels not to allocate us a block group past that
1048          * limit.  The 'inode64' mount option avoids this behavior.
1049          */
1050         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1051                 (*ac)->ac_max_block = (u32)~0U;
1052
1053         /*
1054          * slot is set when we successfully steal inode from other nodes.
1055          * It is reset in 3 places:
1056          * 1. when we flush the truncate log
1057          * 2. when we complete local alloc recovery.
1058          * 3. when we successfully allocate from our own slot.
1059          * After it is set, we will go on stealing inodes until we find the
1060          * need to check our slots to see whether there is some space for us.
1061          */
1062         if (slot != OCFS2_INVALID_SLOT &&
1063             atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
1064                 goto inode_steal;
1065
1066         atomic_set(&osb->s_num_inodes_stolen, 0);
1067         alloc_group = osb->osb_inode_alloc_group;
1068         status = ocfs2_reserve_suballoc_bits(osb, *ac,
1069                                              INODE_ALLOC_SYSTEM_INODE,
1070                                              (u32)osb->slot_num,
1071                                              &alloc_group,
1072                                              ALLOC_NEW_GROUP |
1073                                              ALLOC_GROUPS_FROM_GLOBAL);
1074         if (status >= 0) {
1075                 status = 0;
1076
1077                 spin_lock(&osb->osb_lock);
1078                 osb->osb_inode_alloc_group = alloc_group;
1079                 spin_unlock(&osb->osb_lock);
1080                 trace_ocfs2_reserve_new_inode_new_group(
1081                         (unsigned long long)alloc_group);
1082
1083                 /*
1084                  * Some inodes must be freed by us, so try to allocate
1085                  * from our own next time.
1086                  */
1087                 if (slot != OCFS2_INVALID_SLOT)
1088                         ocfs2_init_inode_steal_slot(osb);
1089                 goto bail;
1090         } else if (status < 0 && status != -ENOSPC) {
1091                 mlog_errno(status);
1092                 goto bail;
1093         }
1094
1095         ocfs2_free_ac_resource(*ac);
1096
1097 inode_steal:
1098         status = ocfs2_steal_inode(osb, *ac);
1099         atomic_inc(&osb->s_num_inodes_stolen);
1100         if (status < 0) {
1101                 if (status != -ENOSPC)
1102                         mlog_errno(status);
1103                 goto bail;
1104         }
1105
1106         status = 0;
1107 bail:
1108         if ((status < 0) && *ac) {
1109                 ocfs2_free_alloc_context(*ac);
1110                 *ac = NULL;
1111         }
1112
1113         if (status)
1114                 mlog_errno(status);
1115         return status;
1116 }
1117
1118 /* local alloc code has to do the same thing, so rather than do this
1119  * twice.. */
1120 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1121                                       struct ocfs2_alloc_context *ac)
1122 {
1123         int status;
1124
1125         ac->ac_which = OCFS2_AC_USE_MAIN;
1126         ac->ac_group_search = ocfs2_cluster_group_search;
1127
1128         status = ocfs2_reserve_suballoc_bits(osb, ac,
1129                                              GLOBAL_BITMAP_SYSTEM_INODE,
1130                                              OCFS2_INVALID_SLOT, NULL,
1131                                              ALLOC_NEW_GROUP);
1132         if (status < 0 && status != -ENOSPC)
1133                 mlog_errno(status);
1134
1135         return status;
1136 }
1137
1138 /* Callers don't need to care which bitmap (local alloc or main) to
1139  * use so we figure it out for them, but unfortunately this clutters
1140  * things a bit. */
1141 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1142                                              u32 bits_wanted, u64 max_block,
1143                                              int flags,
1144                                              struct ocfs2_alloc_context **ac)
1145 {
1146         int status, ret = 0;
1147         int retried = 0;
1148
1149         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1150         if (!(*ac)) {
1151                 status = -ENOMEM;
1152                 mlog_errno(status);
1153                 goto bail;
1154         }
1155
1156         (*ac)->ac_bits_wanted = bits_wanted;
1157         (*ac)->ac_max_block = max_block;
1158
1159         status = -ENOSPC;
1160         if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1161             ocfs2_alloc_should_use_local(osb, bits_wanted)) {
1162                 status = ocfs2_reserve_local_alloc_bits(osb,
1163                                                         bits_wanted,
1164                                                         *ac);
1165                 if ((status < 0) && (status != -ENOSPC)) {
1166                         mlog_errno(status);
1167                         goto bail;
1168                 }
1169         }
1170
1171         if (status == -ENOSPC) {
1172 retry:
1173                 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1174                 /* Retry if there is sufficient space cached in truncate log */
1175                 if (status == -ENOSPC && !retried) {
1176                         retried = 1;
1177                         ocfs2_inode_unlock((*ac)->ac_inode, 1);
1178                         inode_unlock((*ac)->ac_inode);
1179
1180                         ret = ocfs2_try_to_free_truncate_log(osb, bits_wanted);
1181                         if (ret == 1) {
1182                                 iput((*ac)->ac_inode);
1183                                 (*ac)->ac_inode = NULL;
1184                                 goto retry;
1185                         }
1186
1187                         if (ret < 0)
1188                                 mlog_errno(ret);
1189
1190                         inode_lock((*ac)->ac_inode);
1191                         ret = ocfs2_inode_lock((*ac)->ac_inode, NULL, 1);
1192                         if (ret < 0) {
1193                                 mlog_errno(ret);
1194                                 inode_unlock((*ac)->ac_inode);
1195                                 iput((*ac)->ac_inode);
1196                                 (*ac)->ac_inode = NULL;
1197                                 goto bail;
1198                         }
1199                 }
1200                 if (status < 0) {
1201                         if (status != -ENOSPC)
1202                                 mlog_errno(status);
1203                         goto bail;
1204                 }
1205         }
1206
1207         status = 0;
1208 bail:
1209         if ((status < 0) && *ac) {
1210                 ocfs2_free_alloc_context(*ac);
1211                 *ac = NULL;
1212         }
1213
1214         if (status)
1215                 mlog_errno(status);
1216         return status;
1217 }
1218
1219 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1220                            u32 bits_wanted,
1221                            struct ocfs2_alloc_context **ac)
1222 {
1223         return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1224                                                  ALLOC_NEW_GROUP, ac);
1225 }
1226
1227 /*
1228  * More or less lifted from ext3. I'll leave their description below:
1229  *
1230  * "For ext3 allocations, we must not reuse any blocks which are
1231  * allocated in the bitmap buffer's "last committed data" copy.  This
1232  * prevents deletes from freeing up the page for reuse until we have
1233  * committed the delete transaction.
1234  *
1235  * If we didn't do this, then deleting something and reallocating it as
1236  * data would allow the old block to be overwritten before the
1237  * transaction committed (because we force data to disk before commit).
1238  * This would lead to corruption if we crashed between overwriting the
1239  * data and committing the delete.
1240  *
1241  * @@@ We may want to make this allocation behaviour conditional on
1242  * data-writes at some point, and disable it for metadata allocations or
1243  * sync-data inodes."
1244  *
1245  * Note: OCFS2 already does this differently for metadata vs data
1246  * allocations, as those bitmaps are separate and undo access is never
1247  * called on a metadata group descriptor.
1248  */
1249 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1250                                          int nr)
1251 {
1252         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1253         struct journal_head *jh;
1254         int ret = 1;
1255
1256         if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1257                 return 0;
1258
1259         if (!buffer_jbd(bg_bh))
1260                 return 1;
1261
1262         jbd_lock_bh_journal_head(bg_bh);
1263         if (buffer_jbd(bg_bh)) {
1264                 jh = bh2jh(bg_bh);
1265                 spin_lock(&jh->b_state_lock);
1266                 bg = (struct ocfs2_group_desc *) jh->b_committed_data;
1267                 if (bg)
1268                         ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1269                 else
1270                         ret = 1;
1271                 spin_unlock(&jh->b_state_lock);
1272         }
1273         jbd_unlock_bh_journal_head(bg_bh);
1274
1275         return ret;
1276 }
1277
1278 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1279                                              struct buffer_head *bg_bh,
1280                                              unsigned int bits_wanted,
1281                                              unsigned int total_bits,
1282                                              struct ocfs2_suballoc_result *res)
1283 {
1284         void *bitmap;
1285         u16 best_offset, best_size;
1286         int offset, start, found, status = 0;
1287         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1288
1289         /* Callers got this descriptor from
1290          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1291         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1292
1293         found = start = best_offset = best_size = 0;
1294         bitmap = bg->bg_bitmap;
1295
1296         while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1297                 if (offset == total_bits)
1298                         break;
1299
1300                 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1301                         /* We found a zero, but we can't use it as it
1302                          * hasn't been put to disk yet! */
1303                         found = 0;
1304                         start = offset + 1;
1305                 } else if (offset == start) {
1306                         /* we found a zero */
1307                         found++;
1308                         /* move start to the next bit to test */
1309                         start++;
1310                 } else {
1311                         /* got a zero after some ones */
1312                         found = 1;
1313                         start = offset + 1;
1314                 }
1315                 if (found > best_size) {
1316                         best_size = found;
1317                         best_offset = start - found;
1318                 }
1319                 /* we got everything we needed */
1320                 if (found == bits_wanted) {
1321                         /* mlog(0, "Found it all!\n"); */
1322                         break;
1323                 }
1324         }
1325
1326         if (best_size) {
1327                 res->sr_bit_offset = best_offset;
1328                 res->sr_bits = best_size;
1329         } else {
1330                 status = -ENOSPC;
1331                 /* No error log here -- see the comment above
1332                  * ocfs2_test_bg_bit_allocatable */
1333         }
1334
1335         return status;
1336 }
1337
1338 int ocfs2_block_group_set_bits(handle_t *handle,
1339                                              struct inode *alloc_inode,
1340                                              struct ocfs2_group_desc *bg,
1341                                              struct buffer_head *group_bh,
1342                                              unsigned int bit_off,
1343                                              unsigned int num_bits)
1344 {
1345         int status;
1346         void *bitmap = bg->bg_bitmap;
1347         int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1348
1349         /* All callers get the descriptor via
1350          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1351         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1352         BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1353
1354         trace_ocfs2_block_group_set_bits(bit_off, num_bits);
1355
1356         if (ocfs2_is_cluster_bitmap(alloc_inode))
1357                 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1358
1359         status = ocfs2_journal_access_gd(handle,
1360                                          INODE_CACHE(alloc_inode),
1361                                          group_bh,
1362                                          journal_type);
1363         if (status < 0) {
1364                 mlog_errno(status);
1365                 goto bail;
1366         }
1367
1368         le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1369         if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1370                 return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
1371                                    (unsigned long long)le64_to_cpu(bg->bg_blkno),
1372                                    le16_to_cpu(bg->bg_bits),
1373                                    le16_to_cpu(bg->bg_free_bits_count),
1374                                    num_bits);
1375         }
1376         while(num_bits--)
1377                 ocfs2_set_bit(bit_off++, bitmap);
1378
1379         ocfs2_journal_dirty(handle, group_bh);
1380
1381 bail:
1382         return status;
1383 }
1384
1385 /* find the one with the most empty bits */
1386 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1387 {
1388         u16 curr, best;
1389
1390         BUG_ON(!cl->cl_next_free_rec);
1391
1392         best = curr = 0;
1393         while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1394                 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1395                     le32_to_cpu(cl->cl_recs[best].c_free))
1396                         best = curr;
1397                 curr++;
1398         }
1399
1400         BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1401         return best;
1402 }
1403
1404 static int ocfs2_relink_block_group(handle_t *handle,
1405                                     struct inode *alloc_inode,
1406                                     struct buffer_head *fe_bh,
1407                                     struct buffer_head *bg_bh,
1408                                     struct buffer_head *prev_bg_bh,
1409                                     u16 chain)
1410 {
1411         int status;
1412         /* there is a really tiny chance the journal calls could fail,
1413          * but we wouldn't want inconsistent blocks in *any* case. */
1414         u64 bg_ptr, prev_bg_ptr;
1415         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1416         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1417         struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1418
1419         /* The caller got these descriptors from
1420          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1421         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1422         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1423
1424         trace_ocfs2_relink_block_group(
1425                 (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1426                 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1427                 (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1428
1429         bg_ptr = le64_to_cpu(bg->bg_next_group);
1430         prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1431
1432         status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1433                                          prev_bg_bh,
1434                                          OCFS2_JOURNAL_ACCESS_WRITE);
1435         if (status < 0)
1436                 goto out;
1437
1438         prev_bg->bg_next_group = bg->bg_next_group;
1439         ocfs2_journal_dirty(handle, prev_bg_bh);
1440
1441         status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1442                                          bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1443         if (status < 0)
1444                 goto out_rollback_prev_bg;
1445
1446         bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1447         ocfs2_journal_dirty(handle, bg_bh);
1448
1449         status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1450                                          fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1451         if (status < 0)
1452                 goto out_rollback_bg;
1453
1454         fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1455         ocfs2_journal_dirty(handle, fe_bh);
1456
1457 out:
1458         if (status < 0)
1459                 mlog_errno(status);
1460         return status;
1461
1462 out_rollback_bg:
1463         bg->bg_next_group = cpu_to_le64(bg_ptr);
1464 out_rollback_prev_bg:
1465         prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1466         goto out;
1467 }
1468
1469 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1470                                                      u32 wanted)
1471 {
1472         return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1473 }
1474
1475 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1476  * value on error. */
1477 static int ocfs2_cluster_group_search(struct inode *inode,
1478                                       struct buffer_head *group_bh,
1479                                       u32 bits_wanted, u32 min_bits,
1480                                       u64 max_block,
1481                                       struct ocfs2_suballoc_result *res)
1482 {
1483         int search = -ENOSPC;
1484         int ret;
1485         u64 blkoff;
1486         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1487         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1488         unsigned int max_bits, gd_cluster_off;
1489
1490         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1491
1492         if (gd->bg_free_bits_count) {
1493                 max_bits = le16_to_cpu(gd->bg_bits);
1494
1495                 /* Tail groups in cluster bitmaps which aren't cpg
1496                  * aligned are prone to partial extension by a failed
1497                  * fs resize. If the file system resize never got to
1498                  * update the dinode cluster count, then we don't want
1499                  * to trust any clusters past it, regardless of what
1500                  * the group descriptor says. */
1501                 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1502                                                           le64_to_cpu(gd->bg_blkno));
1503                 if ((gd_cluster_off + max_bits) >
1504                     OCFS2_I(inode)->ip_clusters) {
1505                         max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1506                         trace_ocfs2_cluster_group_search_wrong_max_bits(
1507                                 (unsigned long long)le64_to_cpu(gd->bg_blkno),
1508                                 le16_to_cpu(gd->bg_bits),
1509                                 OCFS2_I(inode)->ip_clusters, max_bits);
1510                 }
1511
1512                 ret = ocfs2_block_group_find_clear_bits(osb,
1513                                                         group_bh, bits_wanted,
1514                                                         max_bits, res);
1515                 if (ret)
1516                         return ret;
1517
1518                 if (max_block) {
1519                         blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1520                                                           gd_cluster_off +
1521                                                           res->sr_bit_offset +
1522                                                           res->sr_bits);
1523                         trace_ocfs2_cluster_group_search_max_block(
1524                                 (unsigned long long)blkoff,
1525                                 (unsigned long long)max_block);
1526                         if (blkoff > max_block)
1527                                 return -ENOSPC;
1528                 }
1529
1530                 /* ocfs2_block_group_find_clear_bits() might
1531                  * return success, but we still want to return
1532                  * -ENOSPC unless it found the minimum number
1533                  * of bits. */
1534                 if (min_bits <= res->sr_bits)
1535                         search = 0; /* success */
1536                 else if (res->sr_bits) {
1537                         /*
1538                          * Don't show bits which we'll be returning
1539                          * for allocation to the local alloc bitmap.
1540                          */
1541                         ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
1542                 }
1543         }
1544
1545         return search;
1546 }
1547
1548 static int ocfs2_block_group_search(struct inode *inode,
1549                                     struct buffer_head *group_bh,
1550                                     u32 bits_wanted, u32 min_bits,
1551                                     u64 max_block,
1552                                     struct ocfs2_suballoc_result *res)
1553 {
1554         int ret = -ENOSPC;
1555         u64 blkoff;
1556         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1557
1558         BUG_ON(min_bits != 1);
1559         BUG_ON(ocfs2_is_cluster_bitmap(inode));
1560
1561         if (bg->bg_free_bits_count) {
1562                 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1563                                                         group_bh, bits_wanted,
1564                                                         le16_to_cpu(bg->bg_bits),
1565                                                         res);
1566                 if (!ret && max_block) {
1567                         blkoff = le64_to_cpu(bg->bg_blkno) +
1568                                 res->sr_bit_offset + res->sr_bits;
1569                         trace_ocfs2_block_group_search_max_block(
1570                                 (unsigned long long)blkoff,
1571                                 (unsigned long long)max_block);
1572                         if (blkoff > max_block)
1573                                 ret = -ENOSPC;
1574                 }
1575         }
1576
1577         return ret;
1578 }
1579
1580 int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1581                                        handle_t *handle,
1582                                        struct buffer_head *di_bh,
1583                                        u32 num_bits,
1584                                        u16 chain)
1585 {
1586         int ret;
1587         u32 tmp_used;
1588         struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1589         struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1590
1591         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1592                                       OCFS2_JOURNAL_ACCESS_WRITE);
1593         if (ret < 0) {
1594                 mlog_errno(ret);
1595                 goto out;
1596         }
1597
1598         tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1599         di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1600         le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1601         ocfs2_journal_dirty(handle, di_bh);
1602
1603 out:
1604         return ret;
1605 }
1606
1607 void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
1608                                        struct buffer_head *di_bh,
1609                                        u32 num_bits,
1610                                        u16 chain)
1611 {
1612         u32 tmp_used;
1613         struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1614         struct ocfs2_chain_list *cl;
1615
1616         cl = (struct ocfs2_chain_list *)&di->id2.i_chain;
1617         tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1618         di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits);
1619         le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits);
1620 }
1621
1622 static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1623                                          struct ocfs2_extent_rec *rec,
1624                                          struct ocfs2_chain_list *cl)
1625 {
1626         unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1627         unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1628         unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1629
1630         if (res->sr_bit_offset < bitoff)
1631                 return 0;
1632         if (res->sr_bit_offset >= (bitoff + bitcount))
1633                 return 0;
1634         res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1635                 (res->sr_bit_offset - bitoff);
1636         if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1637                 res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1638         return 1;
1639 }
1640
1641 static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1642                                           struct ocfs2_group_desc *bg,
1643                                           struct ocfs2_suballoc_result *res)
1644 {
1645         int i;
1646         u64 bg_blkno = res->sr_bg_blkno;  /* Save off */
1647         struct ocfs2_extent_rec *rec;
1648         struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1649         struct ocfs2_chain_list *cl = &di->id2.i_chain;
1650
1651         if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1652                 res->sr_blkno = 0;
1653                 return;
1654         }
1655
1656         res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1657         res->sr_bg_blkno = 0;  /* Clear it for contig block groups */
1658         if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1659             !bg->bg_list.l_next_free_rec)
1660                 return;
1661
1662         for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1663                 rec = &bg->bg_list.l_recs[i];
1664                 if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1665                         res->sr_bg_blkno = bg_blkno;  /* Restore */
1666                         break;
1667                 }
1668         }
1669 }
1670
1671 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1672                                   handle_t *handle,
1673                                   u32 bits_wanted,
1674                                   u32 min_bits,
1675                                   struct ocfs2_suballoc_result *res,
1676                                   u16 *bits_left)
1677 {
1678         int ret;
1679         struct buffer_head *group_bh = NULL;
1680         struct ocfs2_group_desc *gd;
1681         struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1682         struct inode *alloc_inode = ac->ac_inode;
1683
1684         ret = ocfs2_read_group_descriptor(alloc_inode, di,
1685                                           res->sr_bg_blkno, &group_bh);
1686         if (ret < 0) {
1687                 mlog_errno(ret);
1688                 return ret;
1689         }
1690
1691         gd = (struct ocfs2_group_desc *) group_bh->b_data;
1692         ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1693                                   ac->ac_max_block, res);
1694         if (ret < 0) {
1695                 if (ret != -ENOSPC)
1696                         mlog_errno(ret);
1697                 goto out;
1698         }
1699
1700         if (!ret)
1701                 ocfs2_bg_discontig_fix_result(ac, gd, res);
1702
1703         /*
1704          * sr_bg_blkno might have been changed by
1705          * ocfs2_bg_discontig_fix_result
1706          */
1707         res->sr_bg_stable_blkno = group_bh->b_blocknr;
1708
1709         if (ac->ac_find_loc_only)
1710                 goto out_loc_only;
1711
1712         ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1713                                                res->sr_bits,
1714                                                le16_to_cpu(gd->bg_chain));
1715         if (ret < 0) {
1716                 mlog_errno(ret);
1717                 goto out;
1718         }
1719
1720         ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1721                                          res->sr_bit_offset, res->sr_bits);
1722         if (ret < 0) {
1723                 ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh,
1724                                                res->sr_bits,
1725                                                le16_to_cpu(gd->bg_chain));
1726                 mlog_errno(ret);
1727         }
1728
1729 out_loc_only:
1730         *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1731
1732 out:
1733         brelse(group_bh);
1734
1735         return ret;
1736 }
1737
1738 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1739                               handle_t *handle,
1740                               u32 bits_wanted,
1741                               u32 min_bits,
1742                               struct ocfs2_suballoc_result *res,
1743                               u16 *bits_left)
1744 {
1745         int status;
1746         u16 chain;
1747         u64 next_group;
1748         struct inode *alloc_inode = ac->ac_inode;
1749         struct buffer_head *group_bh = NULL;
1750         struct buffer_head *prev_group_bh = NULL;
1751         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1752         struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1753         struct ocfs2_group_desc *bg;
1754
1755         chain = ac->ac_chain;
1756         trace_ocfs2_search_chain_begin(
1757                 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1758                 bits_wanted, chain);
1759
1760         status = ocfs2_read_group_descriptor(alloc_inode, fe,
1761                                              le64_to_cpu(cl->cl_recs[chain].c_blkno),
1762                                              &group_bh);
1763         if (status < 0) {
1764                 mlog_errno(status);
1765                 goto bail;
1766         }
1767         bg = (struct ocfs2_group_desc *) group_bh->b_data;
1768
1769         status = -ENOSPC;
1770         /* for now, the chain search is a bit simplistic. We just use
1771          * the 1st group with any empty bits. */
1772         while ((status = ac->ac_group_search(alloc_inode, group_bh,
1773                                              bits_wanted, min_bits,
1774                                              ac->ac_max_block,
1775                                              res)) == -ENOSPC) {
1776                 if (!bg->bg_next_group)
1777                         break;
1778
1779                 brelse(prev_group_bh);
1780                 prev_group_bh = NULL;
1781
1782                 next_group = le64_to_cpu(bg->bg_next_group);
1783                 prev_group_bh = group_bh;
1784                 group_bh = NULL;
1785                 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1786                                                      next_group, &group_bh);
1787                 if (status < 0) {
1788                         mlog_errno(status);
1789                         goto bail;
1790                 }
1791                 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1792         }
1793         if (status < 0) {
1794                 if (status != -ENOSPC)
1795                         mlog_errno(status);
1796                 goto bail;
1797         }
1798
1799         trace_ocfs2_search_chain_succ(
1800                 (unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1801
1802         res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1803
1804         BUG_ON(res->sr_bits == 0);
1805         if (!status)
1806                 ocfs2_bg_discontig_fix_result(ac, bg, res);
1807
1808         /*
1809          * sr_bg_blkno might have been changed by
1810          * ocfs2_bg_discontig_fix_result
1811          */
1812         res->sr_bg_stable_blkno = group_bh->b_blocknr;
1813
1814         /*
1815          * Keep track of previous block descriptor read. When
1816          * we find a target, if we have read more than X
1817          * number of descriptors, and the target is reasonably
1818          * empty, relink him to top of his chain.
1819          *
1820          * We've read 0 extra blocks and only send one more to
1821          * the transaction, yet the next guy to search has a
1822          * much easier time.
1823          *
1824          * Do this *after* figuring out how many bits we're taking out
1825          * of our target group.
1826          */
1827         if (!ac->ac_disable_chain_relink &&
1828             (prev_group_bh) &&
1829             (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1830                 status = ocfs2_relink_block_group(handle, alloc_inode,
1831                                                   ac->ac_bh, group_bh,
1832                                                   prev_group_bh, chain);
1833                 if (status < 0) {
1834                         mlog_errno(status);
1835                         goto bail;
1836                 }
1837         }
1838
1839         if (ac->ac_find_loc_only)
1840                 goto out_loc_only;
1841
1842         status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1843                                                   ac->ac_bh, res->sr_bits,
1844                                                   chain);
1845         if (status) {
1846                 mlog_errno(status);
1847                 goto bail;
1848         }
1849
1850         status = ocfs2_block_group_set_bits(handle,
1851                                             alloc_inode,
1852                                             bg,
1853                                             group_bh,
1854                                             res->sr_bit_offset,
1855                                             res->sr_bits);
1856         if (status < 0) {
1857                 ocfs2_rollback_alloc_dinode_counts(alloc_inode,
1858                                         ac->ac_bh, res->sr_bits, chain);
1859                 mlog_errno(status);
1860                 goto bail;
1861         }
1862
1863         trace_ocfs2_search_chain_end(
1864                         (unsigned long long)le64_to_cpu(fe->i_blkno),
1865                         res->sr_bits);
1866
1867 out_loc_only:
1868         *bits_left = le16_to_cpu(bg->bg_free_bits_count);
1869 bail:
1870         brelse(group_bh);
1871         brelse(prev_group_bh);
1872
1873         if (status)
1874                 mlog_errno(status);
1875         return status;
1876 }
1877
1878 /* will give out up to bits_wanted contiguous bits. */
1879 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1880                                      handle_t *handle,
1881                                      u32 bits_wanted,
1882                                      u32 min_bits,
1883                                      struct ocfs2_suballoc_result *res)
1884 {
1885         int status;
1886         u16 victim, i;
1887         u16 bits_left = 0;
1888         u64 hint = ac->ac_last_group;
1889         struct ocfs2_chain_list *cl;
1890         struct ocfs2_dinode *fe;
1891
1892         BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1893         BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1894         BUG_ON(!ac->ac_bh);
1895
1896         fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1897
1898         /* The bh was validated by the inode read during
1899          * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1900         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1901
1902         if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1903             le32_to_cpu(fe->id1.bitmap1.i_total)) {
1904                 status = ocfs2_error(ac->ac_inode->i_sb,
1905                                      "Chain allocator dinode %llu has %u used bits but only %u total\n",
1906                                      (unsigned long long)le64_to_cpu(fe->i_blkno),
1907                                      le32_to_cpu(fe->id1.bitmap1.i_used),
1908                                      le32_to_cpu(fe->id1.bitmap1.i_total));
1909                 goto bail;
1910         }
1911
1912         res->sr_bg_blkno = hint;
1913         if (res->sr_bg_blkno) {
1914                 /* Attempt to short-circuit the usual search mechanism
1915                  * by jumping straight to the most recently used
1916                  * allocation group. This helps us maintain some
1917                  * contiguousness across allocations. */
1918                 status = ocfs2_search_one_group(ac, handle, bits_wanted,
1919                                                 min_bits, res, &bits_left);
1920                 if (!status)
1921                         goto set_hint;
1922                 if (status < 0 && status != -ENOSPC) {
1923                         mlog_errno(status);
1924                         goto bail;
1925                 }
1926         }
1927
1928         cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1929
1930         victim = ocfs2_find_victim_chain(cl);
1931         ac->ac_chain = victim;
1932
1933         status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1934                                     res, &bits_left);
1935         if (!status) {
1936                 if (ocfs2_is_cluster_bitmap(ac->ac_inode))
1937                         hint = res->sr_bg_blkno;
1938                 else
1939                         hint = ocfs2_group_from_res(res);
1940                 goto set_hint;
1941         }
1942         if (status < 0 && status != -ENOSPC) {
1943                 mlog_errno(status);
1944                 goto bail;
1945         }
1946
1947         trace_ocfs2_claim_suballoc_bits(victim);
1948
1949         /* If we didn't pick a good victim, then just default to
1950          * searching each chain in order. Don't allow chain relinking
1951          * because we only calculate enough journal credits for one
1952          * relink per alloc. */
1953         ac->ac_disable_chain_relink = 1;
1954         for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1955                 if (i == victim)
1956                         continue;
1957                 if (!cl->cl_recs[i].c_free)
1958                         continue;
1959
1960                 ac->ac_chain = i;
1961                 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1962                                             res, &bits_left);
1963                 if (!status) {
1964                         hint = ocfs2_group_from_res(res);
1965                         break;
1966                 }
1967                 if (status < 0 && status != -ENOSPC) {
1968                         mlog_errno(status);
1969                         goto bail;
1970                 }
1971         }
1972
1973 set_hint:
1974         if (status != -ENOSPC) {
1975                 /* If the next search of this group is not likely to
1976                  * yield a suitable extent, then we reset the last
1977                  * group hint so as to not waste a disk read */
1978                 if (bits_left < min_bits)
1979                         ac->ac_last_group = 0;
1980                 else
1981                         ac->ac_last_group = hint;
1982         }
1983
1984 bail:
1985         if (status)
1986                 mlog_errno(status);
1987         return status;
1988 }
1989
1990 int ocfs2_claim_metadata(handle_t *handle,
1991                          struct ocfs2_alloc_context *ac,
1992                          u32 bits_wanted,
1993                          u64 *suballoc_loc,
1994                          u16 *suballoc_bit_start,
1995                          unsigned int *num_bits,
1996                          u64 *blkno_start)
1997 {
1998         int status;
1999         struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2000
2001         BUG_ON(!ac);
2002         BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
2003         BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
2004
2005         status = ocfs2_claim_suballoc_bits(ac,
2006                                            handle,
2007                                            bits_wanted,
2008                                            1,
2009                                            &res);
2010         if (status < 0) {
2011                 mlog_errno(status);
2012                 goto bail;
2013         }
2014         atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2015
2016         *suballoc_loc = res.sr_bg_blkno;
2017         *suballoc_bit_start = res.sr_bit_offset;
2018         *blkno_start = res.sr_blkno;
2019         ac->ac_bits_given += res.sr_bits;
2020         *num_bits = res.sr_bits;
2021         status = 0;
2022 bail:
2023         if (status)
2024                 mlog_errno(status);
2025         return status;
2026 }
2027
2028 static void ocfs2_init_inode_ac_group(struct inode *dir,
2029                                       struct buffer_head *parent_di_bh,
2030                                       struct ocfs2_alloc_context *ac)
2031 {
2032         struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2033         /*
2034          * Try to allocate inodes from some specific group.
2035          *
2036          * If the parent dir has recorded the last group used in allocation,
2037          * cool, use it. Otherwise if we try to allocate new inode from the
2038          * same slot the parent dir belongs to, use the same chunk.
2039          *
2040          * We are very careful here to avoid the mistake of setting
2041          * ac_last_group to a group descriptor from a different (unlocked) slot.
2042          */
2043         if (OCFS2_I(dir)->ip_last_used_group &&
2044             OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2045                 ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2046         else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2047                 if (di->i_suballoc_loc)
2048                         ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2049                 else
2050                         ac->ac_last_group = ocfs2_which_suballoc_group(
2051                                         le64_to_cpu(di->i_blkno),
2052                                         le16_to_cpu(di->i_suballoc_bit));
2053         }
2054 }
2055
2056 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2057                                              struct ocfs2_alloc_context *ac)
2058 {
2059         OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2060         OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2061 }
2062
2063 int ocfs2_find_new_inode_loc(struct inode *dir,
2064                              struct buffer_head *parent_fe_bh,
2065                              struct ocfs2_alloc_context *ac,
2066                              u64 *fe_blkno)
2067 {
2068         int ret;
2069         handle_t *handle = NULL;
2070         struct ocfs2_suballoc_result *res;
2071
2072         BUG_ON(!ac);
2073         BUG_ON(ac->ac_bits_given != 0);
2074         BUG_ON(ac->ac_bits_wanted != 1);
2075         BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2076
2077         res = kzalloc(sizeof(*res), GFP_NOFS);
2078         if (res == NULL) {
2079                 ret = -ENOMEM;
2080                 mlog_errno(ret);
2081                 goto out;
2082         }
2083
2084         ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2085
2086         /*
2087          * The handle started here is for chain relink. Alternatively,
2088          * we could just disable relink for these calls.
2089          */
2090         handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2091         if (IS_ERR(handle)) {
2092                 ret = PTR_ERR(handle);
2093                 handle = NULL;
2094                 mlog_errno(ret);
2095                 goto out;
2096         }
2097
2098         /*
2099          * This will instruct ocfs2_claim_suballoc_bits and
2100          * ocfs2_search_one_group to search but save actual allocation
2101          * for later.
2102          */
2103         ac->ac_find_loc_only = 1;
2104
2105         ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2106         if (ret < 0) {
2107                 mlog_errno(ret);
2108                 goto out;
2109         }
2110
2111         ac->ac_find_loc_priv = res;
2112         *fe_blkno = res->sr_blkno;
2113         ocfs2_update_inode_fsync_trans(handle, dir, 0);
2114 out:
2115         if (handle)
2116                 ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2117
2118         if (ret)
2119                 kfree(res);
2120
2121         return ret;
2122 }
2123
2124 int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2125                                  struct inode *dir,
2126                                  struct ocfs2_alloc_context *ac,
2127                                  u64 *suballoc_loc,
2128                                  u16 *suballoc_bit,
2129                                  u64 di_blkno)
2130 {
2131         int ret;
2132         u16 chain;
2133         struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2134         struct buffer_head *bg_bh = NULL;
2135         struct ocfs2_group_desc *bg;
2136         struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2137
2138         /*
2139          * Since di_blkno is being passed back in, we check for any
2140          * inconsistencies which may have happened between
2141          * calls. These are code bugs as di_blkno is not expected to
2142          * change once returned from ocfs2_find_new_inode_loc()
2143          */
2144         BUG_ON(res->sr_blkno != di_blkno);
2145
2146         ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2147                                           res->sr_bg_stable_blkno, &bg_bh);
2148         if (ret) {
2149                 mlog_errno(ret);
2150                 goto out;
2151         }
2152
2153         bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2154         chain = le16_to_cpu(bg->bg_chain);
2155
2156         ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2157                                                ac->ac_bh, res->sr_bits,
2158                                                chain);
2159         if (ret) {
2160                 mlog_errno(ret);
2161                 goto out;
2162         }
2163
2164         ret = ocfs2_block_group_set_bits(handle,
2165                                          ac->ac_inode,
2166                                          bg,
2167                                          bg_bh,
2168                                          res->sr_bit_offset,
2169                                          res->sr_bits);
2170         if (ret < 0) {
2171                 ocfs2_rollback_alloc_dinode_counts(ac->ac_inode,
2172                                                ac->ac_bh, res->sr_bits, chain);
2173                 mlog_errno(ret);
2174                 goto out;
2175         }
2176
2177         trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2178                                            res->sr_bits);
2179
2180         atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2181
2182         BUG_ON(res->sr_bits != 1);
2183
2184         *suballoc_loc = res->sr_bg_blkno;
2185         *suballoc_bit = res->sr_bit_offset;
2186         ac->ac_bits_given++;
2187         ocfs2_save_inode_ac_group(dir, ac);
2188
2189 out:
2190         brelse(bg_bh);
2191
2192         return ret;
2193 }
2194
2195 int ocfs2_claim_new_inode(handle_t *handle,
2196                           struct inode *dir,
2197                           struct buffer_head *parent_fe_bh,
2198                           struct ocfs2_alloc_context *ac,
2199                           u64 *suballoc_loc,
2200                           u16 *suballoc_bit,
2201                           u64 *fe_blkno)
2202 {
2203         int status;
2204         struct ocfs2_suballoc_result res;
2205
2206         BUG_ON(!ac);
2207         BUG_ON(ac->ac_bits_given != 0);
2208         BUG_ON(ac->ac_bits_wanted != 1);
2209         BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2210
2211         ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2212
2213         status = ocfs2_claim_suballoc_bits(ac,
2214                                            handle,
2215                                            1,
2216                                            1,
2217                                            &res);
2218         if (status < 0) {
2219                 mlog_errno(status);
2220                 goto bail;
2221         }
2222         atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2223
2224         BUG_ON(res.sr_bits != 1);
2225
2226         *suballoc_loc = res.sr_bg_blkno;
2227         *suballoc_bit = res.sr_bit_offset;
2228         *fe_blkno = res.sr_blkno;
2229         ac->ac_bits_given++;
2230         ocfs2_save_inode_ac_group(dir, ac);
2231         status = 0;
2232 bail:
2233         if (status)
2234                 mlog_errno(status);
2235         return status;
2236 }
2237
2238 /* translate a group desc. blkno and it's bitmap offset into
2239  * disk cluster offset. */
2240 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2241                                                    u64 bg_blkno,
2242                                                    u16 bg_bit_off)
2243 {
2244         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2245         u32 cluster = 0;
2246
2247         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2248
2249         if (bg_blkno != osb->first_cluster_group_blkno)
2250                 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2251         cluster += (u32) bg_bit_off;
2252         return cluster;
2253 }
2254
2255 /* given a cluster offset, calculate which block group it belongs to
2256  * and return that block offset. */
2257 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2258 {
2259         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2260         u32 group_no;
2261
2262         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2263
2264         group_no = cluster / osb->bitmap_cpg;
2265         if (!group_no)
2266                 return osb->first_cluster_group_blkno;
2267         return ocfs2_clusters_to_blocks(inode->i_sb,
2268                                         group_no * osb->bitmap_cpg);
2269 }
2270
2271 /* given the block number of a cluster start, calculate which cluster
2272  * group and descriptor bitmap offset that corresponds to. */
2273 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2274                                                 u64 data_blkno,
2275                                                 u64 *bg_blkno,
2276                                                 u16 *bg_bit_off)
2277 {
2278         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2279         u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2280
2281         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2282
2283         *bg_blkno = ocfs2_which_cluster_group(inode,
2284                                               data_cluster);
2285
2286         if (*bg_blkno == osb->first_cluster_group_blkno)
2287                 *bg_bit_off = (u16) data_cluster;
2288         else
2289                 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2290                                                              data_blkno - *bg_blkno);
2291 }
2292
2293 /*
2294  * min_bits - minimum contiguous chunk from this total allocation we
2295  * can handle. set to what we asked for originally for a full
2296  * contig. allocation, set to '1' to indicate we can deal with extents
2297  * of any size.
2298  */
2299 int __ocfs2_claim_clusters(handle_t *handle,
2300                            struct ocfs2_alloc_context *ac,
2301                            u32 min_clusters,
2302                            u32 max_clusters,
2303                            u32 *cluster_start,
2304                            u32 *num_clusters)
2305 {
2306         int status;
2307         unsigned int bits_wanted = max_clusters;
2308         struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2309         struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2310
2311         BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2312
2313         BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2314                && ac->ac_which != OCFS2_AC_USE_MAIN);
2315
2316         if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2317                 WARN_ON(min_clusters > 1);
2318
2319                 status = ocfs2_claim_local_alloc_bits(osb,
2320                                                       handle,
2321                                                       ac,
2322                                                       bits_wanted,
2323                                                       cluster_start,
2324                                                       num_clusters);
2325                 if (!status)
2326                         atomic_inc(&osb->alloc_stats.local_data);
2327         } else {
2328                 if (min_clusters > (osb->bitmap_cpg - 1)) {
2329                         /* The only paths asking for contiguousness
2330                          * should know about this already. */
2331                         mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2332                              "group bitmap size %u!\n", min_clusters,
2333                              osb->bitmap_cpg);
2334                         status = -ENOSPC;
2335                         goto bail;
2336                 }
2337                 /* clamp the current request down to a realistic size. */
2338                 if (bits_wanted > (osb->bitmap_cpg - 1))
2339                         bits_wanted = osb->bitmap_cpg - 1;
2340
2341                 status = ocfs2_claim_suballoc_bits(ac,
2342                                                    handle,
2343                                                    bits_wanted,
2344                                                    min_clusters,
2345                                                    &res);
2346                 if (!status) {
2347                         BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2348                         *cluster_start =
2349                                 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2350                                                                  res.sr_bg_blkno,
2351                                                                  res.sr_bit_offset);
2352                         atomic_inc(&osb->alloc_stats.bitmap_data);
2353                         *num_clusters = res.sr_bits;
2354                 }
2355         }
2356         if (status < 0) {
2357                 if (status != -ENOSPC)
2358                         mlog_errno(status);
2359                 goto bail;
2360         }
2361
2362         ac->ac_bits_given += *num_clusters;
2363
2364 bail:
2365         if (status)
2366                 mlog_errno(status);
2367         return status;
2368 }
2369
2370 int ocfs2_claim_clusters(handle_t *handle,
2371                          struct ocfs2_alloc_context *ac,
2372                          u32 min_clusters,
2373                          u32 *cluster_start,
2374                          u32 *num_clusters)
2375 {
2376         unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2377
2378         return __ocfs2_claim_clusters(handle, ac, min_clusters,
2379                                       bits_wanted, cluster_start, num_clusters);
2380 }
2381
2382 static int ocfs2_block_group_clear_bits(handle_t *handle,
2383                                         struct inode *alloc_inode,
2384                                         struct ocfs2_group_desc *bg,
2385                                         struct buffer_head *group_bh,
2386                                         unsigned int bit_off,
2387                                         unsigned int num_bits,
2388                                         void (*undo_fn)(unsigned int bit,
2389                                                         unsigned long *bmap))
2390 {
2391         int status;
2392         unsigned int tmp;
2393         struct ocfs2_group_desc *undo_bg = NULL;
2394         struct journal_head *jh;
2395
2396         /* The caller got this descriptor from
2397          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
2398         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2399
2400         trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2401
2402         BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2403         status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2404                                          group_bh,
2405                                          undo_fn ?
2406                                          OCFS2_JOURNAL_ACCESS_UNDO :
2407                                          OCFS2_JOURNAL_ACCESS_WRITE);
2408         if (status < 0) {
2409                 mlog_errno(status);
2410                 goto bail;
2411         }
2412
2413         jh = bh2jh(group_bh);
2414         if (undo_fn) {
2415                 spin_lock(&jh->b_state_lock);
2416                 undo_bg = (struct ocfs2_group_desc *) jh->b_committed_data;
2417                 BUG_ON(!undo_bg);
2418         }
2419
2420         tmp = num_bits;
2421         while(tmp--) {
2422                 ocfs2_clear_bit((bit_off + tmp),
2423                                 (unsigned long *) bg->bg_bitmap);
2424                 if (undo_fn)
2425                         undo_fn(bit_off + tmp,
2426                                 (unsigned long *) undo_bg->bg_bitmap);
2427         }
2428         le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2429         if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2430                 if (undo_fn)
2431                         spin_unlock(&jh->b_state_lock);
2432                 return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
2433                                    (unsigned long long)le64_to_cpu(bg->bg_blkno),
2434                                    le16_to_cpu(bg->bg_bits),
2435                                    le16_to_cpu(bg->bg_free_bits_count),
2436                                    num_bits);
2437         }
2438
2439         if (undo_fn)
2440                 spin_unlock(&jh->b_state_lock);
2441
2442         ocfs2_journal_dirty(handle, group_bh);
2443 bail:
2444         return status;
2445 }
2446
2447 /*
2448  * expects the suballoc inode to already be locked.
2449  */
2450 static int _ocfs2_free_suballoc_bits(handle_t *handle,
2451                                      struct inode *alloc_inode,
2452                                      struct buffer_head *alloc_bh,
2453                                      unsigned int start_bit,
2454                                      u64 bg_blkno,
2455                                      unsigned int count,
2456                                      void (*undo_fn)(unsigned int bit,
2457                                                      unsigned long *bitmap))
2458 {
2459         int status = 0;
2460         u32 tmp_used;
2461         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2462         struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2463         struct buffer_head *group_bh = NULL;
2464         struct ocfs2_group_desc *group;
2465
2466         /* The alloc_bh comes from ocfs2_free_dinode() or
2467          * ocfs2_free_clusters().  The callers have all locked the
2468          * allocator and gotten alloc_bh from the lock call.  This
2469          * validates the dinode buffer.  Any corruption that has happened
2470          * is a code bug. */
2471         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2472         BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2473
2474         trace_ocfs2_free_suballoc_bits(
2475                 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2476                 (unsigned long long)bg_blkno,
2477                 start_bit, count);
2478
2479         status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2480                                              &group_bh);
2481         if (status < 0) {
2482                 mlog_errno(status);
2483                 goto bail;
2484         }
2485         group = (struct ocfs2_group_desc *) group_bh->b_data;
2486
2487         BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2488
2489         status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2490                                               group, group_bh,
2491                                               start_bit, count, undo_fn);
2492         if (status < 0) {
2493                 mlog_errno(status);
2494                 goto bail;
2495         }
2496
2497         status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2498                                          alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2499         if (status < 0) {
2500                 mlog_errno(status);
2501                 ocfs2_block_group_set_bits(handle, alloc_inode, group, group_bh,
2502                                 start_bit, count);
2503                 goto bail;
2504         }
2505
2506         le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2507                      count);
2508         tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2509         fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2510         ocfs2_journal_dirty(handle, alloc_bh);
2511
2512 bail:
2513         brelse(group_bh);
2514         return status;
2515 }
2516
2517 int ocfs2_free_suballoc_bits(handle_t *handle,
2518                              struct inode *alloc_inode,
2519                              struct buffer_head *alloc_bh,
2520                              unsigned int start_bit,
2521                              u64 bg_blkno,
2522                              unsigned int count)
2523 {
2524         return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2525                                          start_bit, bg_blkno, count, NULL);
2526 }
2527
2528 int ocfs2_free_dinode(handle_t *handle,
2529                       struct inode *inode_alloc_inode,
2530                       struct buffer_head *inode_alloc_bh,
2531                       struct ocfs2_dinode *di)
2532 {
2533         u64 blk = le64_to_cpu(di->i_blkno);
2534         u16 bit = le16_to_cpu(di->i_suballoc_bit);
2535         u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2536
2537         if (di->i_suballoc_loc)
2538                 bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2539         return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2540                                         inode_alloc_bh, bit, bg_blkno, 1);
2541 }
2542
2543 static int _ocfs2_free_clusters(handle_t *handle,
2544                                 struct inode *bitmap_inode,
2545                                 struct buffer_head *bitmap_bh,
2546                                 u64 start_blk,
2547                                 unsigned int num_clusters,
2548                                 void (*undo_fn)(unsigned int bit,
2549                                                 unsigned long *bitmap))
2550 {
2551         int status;
2552         u16 bg_start_bit;
2553         u64 bg_blkno;
2554
2555         /* You can't ever have a contiguous set of clusters
2556          * bigger than a block group bitmap so we never have to worry
2557          * about looping on them.
2558          * This is expensive. We can safely remove once this stuff has
2559          * gotten tested really well. */
2560         BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb,
2561                                 ocfs2_blocks_to_clusters(bitmap_inode->i_sb,
2562                                                          start_blk)));
2563
2564
2565         ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2566                                      &bg_start_bit);
2567
2568         trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2569                         (unsigned long long)start_blk,
2570                         bg_start_bit, num_clusters);
2571
2572         status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2573                                            bg_start_bit, bg_blkno,
2574                                            num_clusters, undo_fn);
2575         if (status < 0) {
2576                 mlog_errno(status);
2577                 goto out;
2578         }
2579
2580         ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2581                                          num_clusters);
2582
2583 out:
2584         return status;
2585 }
2586
2587 int ocfs2_free_clusters(handle_t *handle,
2588                         struct inode *bitmap_inode,
2589                         struct buffer_head *bitmap_bh,
2590                         u64 start_blk,
2591                         unsigned int num_clusters)
2592 {
2593         return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2594                                     start_blk, num_clusters,
2595                                     _ocfs2_set_bit);
2596 }
2597
2598 /*
2599  * Give never-used clusters back to the global bitmap.  We don't need
2600  * to protect these bits in the undo buffer.
2601  */
2602 int ocfs2_release_clusters(handle_t *handle,
2603                            struct inode *bitmap_inode,
2604                            struct buffer_head *bitmap_bh,
2605                            u64 start_blk,
2606                            unsigned int num_clusters)
2607 {
2608         return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2609                                     start_blk, num_clusters,
2610                                     _ocfs2_clear_bit);
2611 }
2612
2613 /*
2614  * For a given allocation, determine which allocators will need to be
2615  * accessed, and lock them, reserving the appropriate number of bits.
2616  *
2617  * Sparse file systems call this from ocfs2_write_begin_nolock()
2618  * and ocfs2_allocate_unwritten_extents().
2619  *
2620  * File systems which don't support holes call this from
2621  * ocfs2_extend_allocation().
2622  */
2623 int ocfs2_lock_allocators(struct inode *inode,
2624                           struct ocfs2_extent_tree *et,
2625                           u32 clusters_to_add, u32 extents_to_split,
2626                           struct ocfs2_alloc_context **data_ac,
2627                           struct ocfs2_alloc_context **meta_ac)
2628 {
2629         int ret = 0, num_free_extents;
2630         unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2631         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2632
2633         *meta_ac = NULL;
2634         if (data_ac)
2635                 *data_ac = NULL;
2636
2637         BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2638
2639         num_free_extents = ocfs2_num_free_extents(et);
2640         if (num_free_extents < 0) {
2641                 ret = num_free_extents;
2642                 mlog_errno(ret);
2643                 goto out;
2644         }
2645
2646         /*
2647          * Sparse allocation file systems need to be more conservative
2648          * with reserving room for expansion - the actual allocation
2649          * happens while we've got a journal handle open so re-taking
2650          * a cluster lock (because we ran out of room for another
2651          * extent) will violate ordering rules.
2652          *
2653          * Most of the time we'll only be seeing this 1 cluster at a time
2654          * anyway.
2655          *
2656          * Always lock for any unwritten extents - we might want to
2657          * add blocks during a split.
2658          */
2659         if (!num_free_extents ||
2660             (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2661                 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2662                 if (ret < 0) {
2663                         if (ret != -ENOSPC)
2664                                 mlog_errno(ret);
2665                         goto out;
2666                 }
2667         }
2668
2669         if (clusters_to_add == 0)
2670                 goto out;
2671
2672         ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2673         if (ret < 0) {
2674                 if (ret != -ENOSPC)
2675                         mlog_errno(ret);
2676                 goto out;
2677         }
2678
2679 out:
2680         if (ret) {
2681                 if (*meta_ac) {
2682                         ocfs2_free_alloc_context(*meta_ac);
2683                         *meta_ac = NULL;
2684                 }
2685
2686                 /*
2687                  * We cannot have an error and a non null *data_ac.
2688                  */
2689         }
2690
2691         return ret;
2692 }
2693
2694 /*
2695  * Read the inode specified by blkno to get suballoc_slot and
2696  * suballoc_bit.
2697  */
2698 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2699                                        u16 *suballoc_slot, u64 *group_blkno,
2700                                        u16 *suballoc_bit)
2701 {
2702         int status;
2703         struct buffer_head *inode_bh = NULL;
2704         struct ocfs2_dinode *inode_fe;
2705
2706         trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2707
2708         /* dirty read disk */
2709         status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2710         if (status < 0) {
2711                 mlog(ML_ERROR, "read block %llu failed %d\n",
2712                      (unsigned long long)blkno, status);
2713                 goto bail;
2714         }
2715
2716         inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2717         if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2718                 mlog(ML_ERROR, "invalid inode %llu requested\n",
2719                      (unsigned long long)blkno);
2720                 status = -EINVAL;
2721                 goto bail;
2722         }
2723
2724         if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2725             (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2726                 mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2727                      (unsigned long long)blkno,
2728                      (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2729                 status = -EINVAL;
2730                 goto bail;
2731         }
2732
2733         if (suballoc_slot)
2734                 *suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2735         if (suballoc_bit)
2736                 *suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2737         if (group_blkno)
2738                 *group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2739
2740 bail:
2741         brelse(inode_bh);
2742
2743         if (status)
2744                 mlog_errno(status);
2745         return status;
2746 }
2747
2748 /*
2749  * test whether bit is SET in allocator bitmap or not.  on success, 0
2750  * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
2751  * is returned and *res is meaningless.  Call this after you have
2752  * cluster locked against suballoc, or you may get a result based on
2753  * non-up2date contents
2754  */
2755 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2756                                    struct inode *suballoc,
2757                                    struct buffer_head *alloc_bh,
2758                                    u64 group_blkno, u64 blkno,
2759                                    u16 bit, int *res)
2760 {
2761         struct ocfs2_dinode *alloc_di;
2762         struct ocfs2_group_desc *group;
2763         struct buffer_head *group_bh = NULL;
2764         u64 bg_blkno;
2765         int status;
2766
2767         trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2768                                       (unsigned int)bit);
2769
2770         alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2771         if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2772                 mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2773                      (unsigned int)bit,
2774                      ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2775                 status = -EINVAL;
2776                 goto bail;
2777         }
2778
2779         bg_blkno = group_blkno ? group_blkno :
2780                    ocfs2_which_suballoc_group(blkno, bit);
2781         status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2782                                              &group_bh);
2783         if (status < 0) {
2784                 mlog(ML_ERROR, "read group %llu failed %d\n",
2785                      (unsigned long long)bg_blkno, status);
2786                 goto bail;
2787         }
2788
2789         group = (struct ocfs2_group_desc *) group_bh->b_data;
2790         *res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2791
2792 bail:
2793         brelse(group_bh);
2794
2795         if (status)
2796                 mlog_errno(status);
2797         return status;
2798 }
2799
2800 /*
2801  * Test if the bit representing this inode (blkno) is set in the
2802  * suballocator.
2803  *
2804  * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2805  *
2806  * In the event of failure, a negative value is returned and *res is
2807  * meaningless.
2808  *
2809  * Callers must make sure to hold nfs_sync_lock to prevent
2810  * ocfs2_delete_inode() on another node from accessing the same
2811  * suballocator concurrently.
2812  */
2813 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2814 {
2815         int status;
2816         u64 group_blkno = 0;
2817         u16 suballoc_bit = 0, suballoc_slot = 0;
2818         struct inode *inode_alloc_inode;
2819         struct buffer_head *alloc_bh = NULL;
2820
2821         trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2822
2823         status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2824                                              &group_blkno, &suballoc_bit);
2825         if (status < 0) {
2826                 mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2827                 goto bail;
2828         }
2829
2830         if (suballoc_slot == (u16)OCFS2_INVALID_SLOT)
2831                 inode_alloc_inode = ocfs2_get_system_file_inode(osb,
2832                         GLOBAL_INODE_ALLOC_SYSTEM_INODE, suballoc_slot);
2833         else
2834                 inode_alloc_inode = ocfs2_get_system_file_inode(osb,
2835                         INODE_ALLOC_SYSTEM_INODE, suballoc_slot);
2836         if (!inode_alloc_inode) {
2837                 /* the error code could be inaccurate, but we are not able to
2838                  * get the correct one. */
2839                 status = -EINVAL;
2840                 mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2841                      (u32)suballoc_slot);
2842                 goto bail;
2843         }
2844
2845         inode_lock(inode_alloc_inode);
2846         status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2847         if (status < 0) {
2848                 inode_unlock(inode_alloc_inode);
2849                 iput(inode_alloc_inode);
2850                 mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2851                      (u32)suballoc_slot, status);
2852                 goto bail;
2853         }
2854
2855         status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2856                                          group_blkno, blkno, suballoc_bit, res);
2857         if (status < 0)
2858                 mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2859
2860         ocfs2_inode_unlock(inode_alloc_inode, 0);
2861         inode_unlock(inode_alloc_inode);
2862
2863         iput(inode_alloc_inode);
2864         brelse(alloc_bh);
2865 bail:
2866         if (status)
2867                 mlog_errno(status);
2868         return status;
2869 }