Merge tag 'libnvdimm-for-4.4' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm...
[firefly-linux-kernel-4.4.55.git] / fs / ocfs2 / suballoc.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * suballoc.c
5  *
6  * metadata alloc and free
7  * Inspired by ext3 block groups.
8  *
9  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
10  *
11  * This program is free software; you can redistribute it and/or
12  * modify it under the terms of the GNU General Public
13  * License as published by the Free Software Foundation; either
14  * version 2 of the License, or (at your option) any later version.
15  *
16  * This program is distributed in the hope that it will be useful,
17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
19  * General Public License for more details.
20  *
21  * You should have received a copy of the GNU General Public
22  * License along with this program; if not, write to the
23  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
24  * Boston, MA 021110-1307, USA.
25  */
26
27 #include <linux/fs.h>
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
31
32 #include <cluster/masklog.h>
33
34 #include "ocfs2.h"
35
36 #include "alloc.h"
37 #include "blockcheck.h"
38 #include "dlmglue.h"
39 #include "inode.h"
40 #include "journal.h"
41 #include "localalloc.h"
42 #include "suballoc.h"
43 #include "super.h"
44 #include "sysfile.h"
45 #include "uptodate.h"
46 #include "ocfs2_trace.h"
47
48 #include "buffer_head_io.h"
49
50 #define NOT_ALLOC_NEW_GROUP             0
51 #define ALLOC_NEW_GROUP                 0x1
52 #define ALLOC_GROUPS_FROM_GLOBAL        0x2
53
54 #define OCFS2_MAX_TO_STEAL              1024
55
56 struct ocfs2_suballoc_result {
57         u64             sr_bg_blkno;    /* The bg we allocated from.  Set
58                                            to 0 when a block group is
59                                            contiguous. */
60         u64             sr_bg_stable_blkno; /*
61                                              * Doesn't change, always
62                                              * set to target block
63                                              * group descriptor
64                                              * block.
65                                              */
66         u64             sr_blkno;       /* The first allocated block */
67         unsigned int    sr_bit_offset;  /* The bit in the bg */
68         unsigned int    sr_bits;        /* How many bits we claimed */
69 };
70
71 static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
72 {
73         if (res->sr_blkno == 0)
74                 return 0;
75
76         if (res->sr_bg_blkno)
77                 return res->sr_bg_blkno;
78
79         return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset);
80 }
81
82 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg);
83 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe);
84 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
85 static int ocfs2_block_group_fill(handle_t *handle,
86                                   struct inode *alloc_inode,
87                                   struct buffer_head *bg_bh,
88                                   u64 group_blkno,
89                                   unsigned int group_clusters,
90                                   u16 my_chain,
91                                   struct ocfs2_chain_list *cl);
92 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
93                                    struct inode *alloc_inode,
94                                    struct buffer_head *bh,
95                                    u64 max_block,
96                                    u64 *last_alloc_group,
97                                    int flags);
98
99 static int ocfs2_cluster_group_search(struct inode *inode,
100                                       struct buffer_head *group_bh,
101                                       u32 bits_wanted, u32 min_bits,
102                                       u64 max_block,
103                                       struct ocfs2_suballoc_result *res);
104 static int ocfs2_block_group_search(struct inode *inode,
105                                     struct buffer_head *group_bh,
106                                     u32 bits_wanted, u32 min_bits,
107                                     u64 max_block,
108                                     struct ocfs2_suballoc_result *res);
109 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
110                                      handle_t *handle,
111                                      u32 bits_wanted,
112                                      u32 min_bits,
113                                      struct ocfs2_suballoc_result *res);
114 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
115                                          int nr);
116 static int ocfs2_relink_block_group(handle_t *handle,
117                                     struct inode *alloc_inode,
118                                     struct buffer_head *fe_bh,
119                                     struct buffer_head *bg_bh,
120                                     struct buffer_head *prev_bg_bh,
121                                     u16 chain);
122 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
123                                                      u32 wanted);
124 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
125                                                    u64 bg_blkno,
126                                                    u16 bg_bit_off);
127 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
128                                                 u64 data_blkno,
129                                                 u64 *bg_blkno,
130                                                 u16 *bg_bit_off);
131 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
132                                              u32 bits_wanted, u64 max_block,
133                                              int flags,
134                                              struct ocfs2_alloc_context **ac);
135
136 void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
137 {
138         struct inode *inode = ac->ac_inode;
139
140         if (inode) {
141                 if (ac->ac_which != OCFS2_AC_USE_LOCAL)
142                         ocfs2_inode_unlock(inode, 1);
143
144                 mutex_unlock(&inode->i_mutex);
145
146                 iput(inode);
147                 ac->ac_inode = NULL;
148         }
149         brelse(ac->ac_bh);
150         ac->ac_bh = NULL;
151         ac->ac_resv = NULL;
152         kfree(ac->ac_find_loc_priv);
153         ac->ac_find_loc_priv = NULL;
154 }
155
156 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
157 {
158         ocfs2_free_ac_resource(ac);
159         kfree(ac);
160 }
161
162 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
163 {
164         return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
165 }
166
167 #define do_error(fmt, ...)                                              \
168 do {                                                                    \
169         if (resize)                                                     \
170                 mlog(ML_ERROR, fmt, ##__VA_ARGS__);                     \
171         else                                                            \
172                 return ocfs2_error(sb, fmt, ##__VA_ARGS__);             \
173 } while (0)
174
175 static int ocfs2_validate_gd_self(struct super_block *sb,
176                                   struct buffer_head *bh,
177                                   int resize)
178 {
179         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
180
181         if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
182                 do_error("Group descriptor #%llu has bad signature %.*s\n",
183                          (unsigned long long)bh->b_blocknr, 7,
184                          gd->bg_signature);
185         }
186
187         if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
188                 do_error("Group descriptor #%llu has an invalid bg_blkno of %llu\n",
189                          (unsigned long long)bh->b_blocknr,
190                          (unsigned long long)le64_to_cpu(gd->bg_blkno));
191         }
192
193         if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
194                 do_error("Group descriptor #%llu has an invalid fs_generation of #%u\n",
195                          (unsigned long long)bh->b_blocknr,
196                          le32_to_cpu(gd->bg_generation));
197         }
198
199         if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
200                 do_error("Group descriptor #%llu has bit count %u but claims that %u are free\n",
201                          (unsigned long long)bh->b_blocknr,
202                          le16_to_cpu(gd->bg_bits),
203                          le16_to_cpu(gd->bg_free_bits_count));
204         }
205
206         if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
207                 do_error("Group descriptor #%llu has bit count %u but max bitmap bits of %u\n",
208                          (unsigned long long)bh->b_blocknr,
209                          le16_to_cpu(gd->bg_bits),
210                          8 * le16_to_cpu(gd->bg_size));
211         }
212
213         return 0;
214 }
215
216 static int ocfs2_validate_gd_parent(struct super_block *sb,
217                                     struct ocfs2_dinode *di,
218                                     struct buffer_head *bh,
219                                     int resize)
220 {
221         unsigned int max_bits;
222         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
223
224         if (di->i_blkno != gd->bg_parent_dinode) {
225                 do_error("Group descriptor #%llu has bad parent pointer (%llu, expected %llu)\n",
226                          (unsigned long long)bh->b_blocknr,
227                          (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
228                          (unsigned long long)le64_to_cpu(di->i_blkno));
229         }
230
231         max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
232         if (le16_to_cpu(gd->bg_bits) > max_bits) {
233                 do_error("Group descriptor #%llu has bit count of %u\n",
234                          (unsigned long long)bh->b_blocknr,
235                          le16_to_cpu(gd->bg_bits));
236         }
237
238         /* In resize, we may meet the case bg_chain == cl_next_free_rec. */
239         if ((le16_to_cpu(gd->bg_chain) >
240              le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
241             ((le16_to_cpu(gd->bg_chain) ==
242              le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
243                 do_error("Group descriptor #%llu has bad chain %u\n",
244                          (unsigned long long)bh->b_blocknr,
245                          le16_to_cpu(gd->bg_chain));
246         }
247
248         return 0;
249 }
250
251 #undef do_error
252
253 /*
254  * This version only prints errors.  It does not fail the filesystem, and
255  * exists only for resize.
256  */
257 int ocfs2_check_group_descriptor(struct super_block *sb,
258                                  struct ocfs2_dinode *di,
259                                  struct buffer_head *bh)
260 {
261         int rc;
262         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
263
264         BUG_ON(!buffer_uptodate(bh));
265
266         /*
267          * If the ecc fails, we return the error but otherwise
268          * leave the filesystem running.  We know any error is
269          * local to this block.
270          */
271         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
272         if (rc) {
273                 mlog(ML_ERROR,
274                      "Checksum failed for group descriptor %llu\n",
275                      (unsigned long long)bh->b_blocknr);
276         } else
277                 rc = ocfs2_validate_gd_self(sb, bh, 1);
278         if (!rc)
279                 rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
280
281         return rc;
282 }
283
284 static int ocfs2_validate_group_descriptor(struct super_block *sb,
285                                            struct buffer_head *bh)
286 {
287         int rc;
288         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
289
290         trace_ocfs2_validate_group_descriptor(
291                                         (unsigned long long)bh->b_blocknr);
292
293         BUG_ON(!buffer_uptodate(bh));
294
295         /*
296          * If the ecc fails, we return the error but otherwise
297          * leave the filesystem running.  We know any error is
298          * local to this block.
299          */
300         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
301         if (rc)
302                 return rc;
303
304         /*
305          * Errors after here are fatal.
306          */
307
308         return ocfs2_validate_gd_self(sb, bh, 0);
309 }
310
311 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
312                                 u64 gd_blkno, struct buffer_head **bh)
313 {
314         int rc;
315         struct buffer_head *tmp = *bh;
316
317         rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
318                               ocfs2_validate_group_descriptor);
319         if (rc)
320                 goto out;
321
322         rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
323         if (rc) {
324                 brelse(tmp);
325                 goto out;
326         }
327
328         /* If ocfs2_read_block() got us a new bh, pass it up. */
329         if (!*bh)
330                 *bh = tmp;
331
332 out:
333         return rc;
334 }
335
336 static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
337                                           struct ocfs2_group_desc *bg,
338                                           struct ocfs2_chain_list *cl,
339                                           u64 p_blkno, unsigned int clusters)
340 {
341         struct ocfs2_extent_list *el = &bg->bg_list;
342         struct ocfs2_extent_rec *rec;
343
344         BUG_ON(!ocfs2_supports_discontig_bg(osb));
345         if (!el->l_next_free_rec)
346                 el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
347         rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
348         rec->e_blkno = cpu_to_le64(p_blkno);
349         rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
350                                   le16_to_cpu(cl->cl_bpc));
351         rec->e_leaf_clusters = cpu_to_le16(clusters);
352         le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
353         le16_add_cpu(&bg->bg_free_bits_count,
354                      clusters * le16_to_cpu(cl->cl_bpc));
355         le16_add_cpu(&el->l_next_free_rec, 1);
356 }
357
358 static int ocfs2_block_group_fill(handle_t *handle,
359                                   struct inode *alloc_inode,
360                                   struct buffer_head *bg_bh,
361                                   u64 group_blkno,
362                                   unsigned int group_clusters,
363                                   u16 my_chain,
364                                   struct ocfs2_chain_list *cl)
365 {
366         int status = 0;
367         struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
368         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
369         struct super_block * sb = alloc_inode->i_sb;
370
371         if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
372                 status = ocfs2_error(alloc_inode->i_sb,
373                                      "group block (%llu) != b_blocknr (%llu)\n",
374                                      (unsigned long long)group_blkno,
375                                      (unsigned long long) bg_bh->b_blocknr);
376                 goto bail;
377         }
378
379         status = ocfs2_journal_access_gd(handle,
380                                          INODE_CACHE(alloc_inode),
381                                          bg_bh,
382                                          OCFS2_JOURNAL_ACCESS_CREATE);
383         if (status < 0) {
384                 mlog_errno(status);
385                 goto bail;
386         }
387
388         memset(bg, 0, sb->s_blocksize);
389         strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
390         bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
391         bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
392                                                 osb->s_feature_incompat));
393         bg->bg_chain = cpu_to_le16(my_chain);
394         bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
395         bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
396         bg->bg_blkno = cpu_to_le64(group_blkno);
397         if (group_clusters == le16_to_cpu(cl->cl_cpg))
398                 bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
399         else
400                 ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
401                                               group_clusters);
402
403         /* set the 1st bit in the bitmap to account for the descriptor block */
404         ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
405         bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
406
407         ocfs2_journal_dirty(handle, bg_bh);
408
409         /* There is no need to zero out or otherwise initialize the
410          * other blocks in a group - All valid FS metadata in a block
411          * group stores the superblock fs_generation value at
412          * allocation time. */
413
414 bail:
415         if (status)
416                 mlog_errno(status);
417         return status;
418 }
419
420 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
421 {
422         u16 curr, best;
423
424         best = curr = 0;
425         while (curr < le16_to_cpu(cl->cl_count)) {
426                 if (le32_to_cpu(cl->cl_recs[best].c_total) >
427                     le32_to_cpu(cl->cl_recs[curr].c_total))
428                         best = curr;
429                 curr++;
430         }
431         return best;
432 }
433
434 static struct buffer_head *
435 ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
436                                struct inode *alloc_inode,
437                                struct ocfs2_alloc_context *ac,
438                                struct ocfs2_chain_list *cl)
439 {
440         int status;
441         u32 bit_off, num_bits;
442         u64 bg_blkno;
443         struct buffer_head *bg_bh;
444         unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
445
446         status = ocfs2_claim_clusters(handle, ac,
447                                       le16_to_cpu(cl->cl_cpg), &bit_off,
448                                       &num_bits);
449         if (status < 0) {
450                 if (status != -ENOSPC)
451                         mlog_errno(status);
452                 goto bail;
453         }
454
455         /* setup the group */
456         bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
457         trace_ocfs2_block_group_alloc_contig(
458              (unsigned long long)bg_blkno, alloc_rec);
459
460         bg_bh = sb_getblk(osb->sb, bg_blkno);
461         if (!bg_bh) {
462                 status = -ENOMEM;
463                 mlog_errno(status);
464                 goto bail;
465         }
466         ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
467
468         status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
469                                         bg_blkno, num_bits, alloc_rec, cl);
470         if (status < 0) {
471                 brelse(bg_bh);
472                 mlog_errno(status);
473         }
474
475 bail:
476         return status ? ERR_PTR(status) : bg_bh;
477 }
478
479 static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
480                                         handle_t *handle,
481                                         struct ocfs2_alloc_context *ac,
482                                         unsigned int min_bits,
483                                         u32 *bit_off, u32 *num_bits)
484 {
485         int status = 0;
486
487         while (min_bits) {
488                 status = ocfs2_claim_clusters(handle, ac, min_bits,
489                                               bit_off, num_bits);
490                 if (status != -ENOSPC)
491                         break;
492
493                 min_bits >>= 1;
494         }
495
496         return status;
497 }
498
499 static int ocfs2_block_group_grow_discontig(handle_t *handle,
500                                             struct inode *alloc_inode,
501                                             struct buffer_head *bg_bh,
502                                             struct ocfs2_alloc_context *ac,
503                                             struct ocfs2_chain_list *cl,
504                                             unsigned int min_bits)
505 {
506         int status;
507         struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
508         struct ocfs2_group_desc *bg =
509                 (struct ocfs2_group_desc *)bg_bh->b_data;
510         unsigned int needed = le16_to_cpu(cl->cl_cpg) -
511                          le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
512         u32 p_cpos, clusters;
513         u64 p_blkno;
514         struct ocfs2_extent_list *el = &bg->bg_list;
515
516         status = ocfs2_journal_access_gd(handle,
517                                          INODE_CACHE(alloc_inode),
518                                          bg_bh,
519                                          OCFS2_JOURNAL_ACCESS_CREATE);
520         if (status < 0) {
521                 mlog_errno(status);
522                 goto bail;
523         }
524
525         while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
526                                 le16_to_cpu(el->l_count))) {
527                 if (min_bits > needed)
528                         min_bits = needed;
529                 status = ocfs2_block_group_claim_bits(osb, handle, ac,
530                                                       min_bits, &p_cpos,
531                                                       &clusters);
532                 if (status < 0) {
533                         if (status != -ENOSPC)
534                                 mlog_errno(status);
535                         goto bail;
536                 }
537                 p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
538                 ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
539                                               clusters);
540
541                 min_bits = clusters;
542                 needed = le16_to_cpu(cl->cl_cpg) -
543                          le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
544         }
545
546         if (needed > 0) {
547                 /*
548                  * We have used up all the extent rec but can't fill up
549                  * the cpg. So bail out.
550                  */
551                 status = -ENOSPC;
552                 goto bail;
553         }
554
555         ocfs2_journal_dirty(handle, bg_bh);
556
557 bail:
558         return status;
559 }
560
561 static void ocfs2_bg_alloc_cleanup(handle_t *handle,
562                                    struct ocfs2_alloc_context *cluster_ac,
563                                    struct inode *alloc_inode,
564                                    struct buffer_head *bg_bh)
565 {
566         int i, ret;
567         struct ocfs2_group_desc *bg;
568         struct ocfs2_extent_list *el;
569         struct ocfs2_extent_rec *rec;
570
571         if (!bg_bh)
572                 return;
573
574         bg = (struct ocfs2_group_desc *)bg_bh->b_data;
575         el = &bg->bg_list;
576         for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
577                 rec = &el->l_recs[i];
578                 ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
579                                           cluster_ac->ac_bh,
580                                           le64_to_cpu(rec->e_blkno),
581                                           le16_to_cpu(rec->e_leaf_clusters));
582                 if (ret)
583                         mlog_errno(ret);
584                 /* Try all the clusters to free */
585         }
586
587         ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
588         brelse(bg_bh);
589 }
590
591 static struct buffer_head *
592 ocfs2_block_group_alloc_discontig(handle_t *handle,
593                                   struct inode *alloc_inode,
594                                   struct ocfs2_alloc_context *ac,
595                                   struct ocfs2_chain_list *cl)
596 {
597         int status;
598         u32 bit_off, num_bits;
599         u64 bg_blkno;
600         unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
601         struct buffer_head *bg_bh = NULL;
602         unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
603         struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
604
605         if (!ocfs2_supports_discontig_bg(osb)) {
606                 status = -ENOSPC;
607                 goto bail;
608         }
609
610         status = ocfs2_extend_trans(handle,
611                                     ocfs2_calc_bg_discontig_credits(osb->sb));
612         if (status) {
613                 mlog_errno(status);
614                 goto bail;
615         }
616
617         /*
618          * We're going to be grabbing from multiple cluster groups.
619          * We don't have enough credits to relink them all, and the
620          * cluster groups will be staying in cache for the duration of
621          * this operation.
622          */
623         ac->ac_disable_chain_relink = 1;
624
625         /* Claim the first region */
626         status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
627                                               &bit_off, &num_bits);
628         if (status < 0) {
629                 if (status != -ENOSPC)
630                         mlog_errno(status);
631                 goto bail;
632         }
633         min_bits = num_bits;
634
635         /* setup the group */
636         bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
637         trace_ocfs2_block_group_alloc_discontig(
638                                 (unsigned long long)bg_blkno, alloc_rec);
639
640         bg_bh = sb_getblk(osb->sb, bg_blkno);
641         if (!bg_bh) {
642                 status = -ENOMEM;
643                 mlog_errno(status);
644                 goto bail;
645         }
646         ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
647
648         status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
649                                         bg_blkno, num_bits, alloc_rec, cl);
650         if (status < 0) {
651                 mlog_errno(status);
652                 goto bail;
653         }
654
655         status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
656                                                   bg_bh, ac, cl, min_bits);
657         if (status)
658                 mlog_errno(status);
659
660 bail:
661         if (status)
662                 ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
663         return status ? ERR_PTR(status) : bg_bh;
664 }
665
666 /*
667  * We expect the block group allocator to already be locked.
668  */
669 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
670                                    struct inode *alloc_inode,
671                                    struct buffer_head *bh,
672                                    u64 max_block,
673                                    u64 *last_alloc_group,
674                                    int flags)
675 {
676         int status, credits;
677         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
678         struct ocfs2_chain_list *cl;
679         struct ocfs2_alloc_context *ac = NULL;
680         handle_t *handle = NULL;
681         u16 alloc_rec;
682         struct buffer_head *bg_bh = NULL;
683         struct ocfs2_group_desc *bg;
684
685         BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
686
687         cl = &fe->id2.i_chain;
688         status = ocfs2_reserve_clusters_with_limit(osb,
689                                                    le16_to_cpu(cl->cl_cpg),
690                                                    max_block, flags, &ac);
691         if (status < 0) {
692                 if (status != -ENOSPC)
693                         mlog_errno(status);
694                 goto bail;
695         }
696
697         credits = ocfs2_calc_group_alloc_credits(osb->sb,
698                                                  le16_to_cpu(cl->cl_cpg));
699         handle = ocfs2_start_trans(osb, credits);
700         if (IS_ERR(handle)) {
701                 status = PTR_ERR(handle);
702                 handle = NULL;
703                 mlog_errno(status);
704                 goto bail;
705         }
706
707         if (last_alloc_group && *last_alloc_group != 0) {
708                 trace_ocfs2_block_group_alloc(
709                                 (unsigned long long)*last_alloc_group);
710                 ac->ac_last_group = *last_alloc_group;
711         }
712
713         bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
714                                                ac, cl);
715         if (IS_ERR(bg_bh) && (PTR_ERR(bg_bh) == -ENOSPC))
716                 bg_bh = ocfs2_block_group_alloc_discontig(handle,
717                                                           alloc_inode,
718                                                           ac, cl);
719         if (IS_ERR(bg_bh)) {
720                 status = PTR_ERR(bg_bh);
721                 bg_bh = NULL;
722                 if (status != -ENOSPC)
723                         mlog_errno(status);
724                 goto bail;
725         }
726         bg = (struct ocfs2_group_desc *) bg_bh->b_data;
727
728         status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
729                                          bh, OCFS2_JOURNAL_ACCESS_WRITE);
730         if (status < 0) {
731                 mlog_errno(status);
732                 goto bail;
733         }
734
735         alloc_rec = le16_to_cpu(bg->bg_chain);
736         le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
737                      le16_to_cpu(bg->bg_free_bits_count));
738         le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
739                      le16_to_cpu(bg->bg_bits));
740         cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
741         if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
742                 le16_add_cpu(&cl->cl_next_free_rec, 1);
743
744         le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
745                                         le16_to_cpu(bg->bg_free_bits_count));
746         le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
747         le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
748
749         ocfs2_journal_dirty(handle, bh);
750
751         spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
752         OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
753         fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
754                                              le32_to_cpu(fe->i_clusters)));
755         spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
756         i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
757         alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
758         ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
759
760         status = 0;
761
762         /* save the new last alloc group so that the caller can cache it. */
763         if (last_alloc_group)
764                 *last_alloc_group = ac->ac_last_group;
765
766 bail:
767         if (handle)
768                 ocfs2_commit_trans(osb, handle);
769
770         if (ac)
771                 ocfs2_free_alloc_context(ac);
772
773         brelse(bg_bh);
774
775         if (status)
776                 mlog_errno(status);
777         return status;
778 }
779
780 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
781                                        struct ocfs2_alloc_context *ac,
782                                        int type,
783                                        u32 slot,
784                                        u64 *last_alloc_group,
785                                        int flags)
786 {
787         int status;
788         u32 bits_wanted = ac->ac_bits_wanted;
789         struct inode *alloc_inode;
790         struct buffer_head *bh = NULL;
791         struct ocfs2_dinode *fe;
792         u32 free_bits;
793
794         alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
795         if (!alloc_inode) {
796                 mlog_errno(-EINVAL);
797                 return -EINVAL;
798         }
799
800         mutex_lock(&alloc_inode->i_mutex);
801
802         status = ocfs2_inode_lock(alloc_inode, &bh, 1);
803         if (status < 0) {
804                 mutex_unlock(&alloc_inode->i_mutex);
805                 iput(alloc_inode);
806
807                 mlog_errno(status);
808                 return status;
809         }
810
811         ac->ac_inode = alloc_inode;
812         ac->ac_alloc_slot = slot;
813
814         fe = (struct ocfs2_dinode *) bh->b_data;
815
816         /* The bh was validated by the inode read inside
817          * ocfs2_inode_lock().  Any corruption is a code bug. */
818         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
819
820         if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
821                 status = ocfs2_error(alloc_inode->i_sb,
822                                      "Invalid chain allocator %llu\n",
823                                      (unsigned long long)le64_to_cpu(fe->i_blkno));
824                 goto bail;
825         }
826
827         free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
828                 le32_to_cpu(fe->id1.bitmap1.i_used);
829
830         if (bits_wanted > free_bits) {
831                 /* cluster bitmap never grows */
832                 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
833                         trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
834                                                                 free_bits);
835                         status = -ENOSPC;
836                         goto bail;
837                 }
838
839                 if (!(flags & ALLOC_NEW_GROUP)) {
840                         trace_ocfs2_reserve_suballoc_bits_no_new_group(
841                                                 slot, bits_wanted, free_bits);
842                         status = -ENOSPC;
843                         goto bail;
844                 }
845
846                 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
847                                                  ac->ac_max_block,
848                                                  last_alloc_group, flags);
849                 if (status < 0) {
850                         if (status != -ENOSPC)
851                                 mlog_errno(status);
852                         goto bail;
853                 }
854                 atomic_inc(&osb->alloc_stats.bg_extends);
855
856                 /* You should never ask for this much metadata */
857                 BUG_ON(bits_wanted >
858                        (le32_to_cpu(fe->id1.bitmap1.i_total)
859                         - le32_to_cpu(fe->id1.bitmap1.i_used)));
860         }
861
862         get_bh(bh);
863         ac->ac_bh = bh;
864 bail:
865         brelse(bh);
866
867         if (status)
868                 mlog_errno(status);
869         return status;
870 }
871
872 static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
873 {
874         spin_lock(&osb->osb_lock);
875         osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
876         spin_unlock(&osb->osb_lock);
877         atomic_set(&osb->s_num_inodes_stolen, 0);
878 }
879
880 static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
881 {
882         spin_lock(&osb->osb_lock);
883         osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
884         spin_unlock(&osb->osb_lock);
885         atomic_set(&osb->s_num_meta_stolen, 0);
886 }
887
888 void ocfs2_init_steal_slots(struct ocfs2_super *osb)
889 {
890         ocfs2_init_inode_steal_slot(osb);
891         ocfs2_init_meta_steal_slot(osb);
892 }
893
894 static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
895 {
896         spin_lock(&osb->osb_lock);
897         if (type == INODE_ALLOC_SYSTEM_INODE)
898                 osb->s_inode_steal_slot = slot;
899         else if (type == EXTENT_ALLOC_SYSTEM_INODE)
900                 osb->s_meta_steal_slot = slot;
901         spin_unlock(&osb->osb_lock);
902 }
903
904 static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
905 {
906         int slot = OCFS2_INVALID_SLOT;
907
908         spin_lock(&osb->osb_lock);
909         if (type == INODE_ALLOC_SYSTEM_INODE)
910                 slot = osb->s_inode_steal_slot;
911         else if (type == EXTENT_ALLOC_SYSTEM_INODE)
912                 slot = osb->s_meta_steal_slot;
913         spin_unlock(&osb->osb_lock);
914
915         return slot;
916 }
917
918 static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
919 {
920         return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
921 }
922
923 static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
924 {
925         return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
926 }
927
928 static int ocfs2_steal_resource(struct ocfs2_super *osb,
929                                 struct ocfs2_alloc_context *ac,
930                                 int type)
931 {
932         int i, status = -ENOSPC;
933         int slot = __ocfs2_get_steal_slot(osb, type);
934
935         /* Start to steal resource from the first slot after ours. */
936         if (slot == OCFS2_INVALID_SLOT)
937                 slot = osb->slot_num + 1;
938
939         for (i = 0; i < osb->max_slots; i++, slot++) {
940                 if (slot == osb->max_slots)
941                         slot = 0;
942
943                 if (slot == osb->slot_num)
944                         continue;
945
946                 status = ocfs2_reserve_suballoc_bits(osb, ac,
947                                                      type,
948                                                      (u32)slot, NULL,
949                                                      NOT_ALLOC_NEW_GROUP);
950                 if (status >= 0) {
951                         __ocfs2_set_steal_slot(osb, slot, type);
952                         break;
953                 }
954
955                 ocfs2_free_ac_resource(ac);
956         }
957
958         return status;
959 }
960
961 static int ocfs2_steal_inode(struct ocfs2_super *osb,
962                              struct ocfs2_alloc_context *ac)
963 {
964         return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
965 }
966
967 static int ocfs2_steal_meta(struct ocfs2_super *osb,
968                             struct ocfs2_alloc_context *ac)
969 {
970         return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
971 }
972
973 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
974                                       int blocks,
975                                       struct ocfs2_alloc_context **ac)
976 {
977         int status;
978         int slot = ocfs2_get_meta_steal_slot(osb);
979
980         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
981         if (!(*ac)) {
982                 status = -ENOMEM;
983                 mlog_errno(status);
984                 goto bail;
985         }
986
987         (*ac)->ac_bits_wanted = blocks;
988         (*ac)->ac_which = OCFS2_AC_USE_META;
989         (*ac)->ac_group_search = ocfs2_block_group_search;
990
991         if (slot != OCFS2_INVALID_SLOT &&
992                 atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
993                 goto extent_steal;
994
995         atomic_set(&osb->s_num_meta_stolen, 0);
996         status = ocfs2_reserve_suballoc_bits(osb, (*ac),
997                                              EXTENT_ALLOC_SYSTEM_INODE,
998                                              (u32)osb->slot_num, NULL,
999                                              ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
1000
1001
1002         if (status >= 0) {
1003                 status = 0;
1004                 if (slot != OCFS2_INVALID_SLOT)
1005                         ocfs2_init_meta_steal_slot(osb);
1006                 goto bail;
1007         } else if (status < 0 && status != -ENOSPC) {
1008                 mlog_errno(status);
1009                 goto bail;
1010         }
1011
1012         ocfs2_free_ac_resource(*ac);
1013
1014 extent_steal:
1015         status = ocfs2_steal_meta(osb, *ac);
1016         atomic_inc(&osb->s_num_meta_stolen);
1017         if (status < 0) {
1018                 if (status != -ENOSPC)
1019                         mlog_errno(status);
1020                 goto bail;
1021         }
1022
1023         status = 0;
1024 bail:
1025         if ((status < 0) && *ac) {
1026                 ocfs2_free_alloc_context(*ac);
1027                 *ac = NULL;
1028         }
1029
1030         if (status)
1031                 mlog_errno(status);
1032         return status;
1033 }
1034
1035 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
1036                                struct ocfs2_extent_list *root_el,
1037                                struct ocfs2_alloc_context **ac)
1038 {
1039         return ocfs2_reserve_new_metadata_blocks(osb,
1040                                         ocfs2_extend_meta_needed(root_el),
1041                                         ac);
1042 }
1043
1044 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
1045                             struct ocfs2_alloc_context **ac)
1046 {
1047         int status;
1048         int slot = ocfs2_get_inode_steal_slot(osb);
1049         u64 alloc_group;
1050
1051         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1052         if (!(*ac)) {
1053                 status = -ENOMEM;
1054                 mlog_errno(status);
1055                 goto bail;
1056         }
1057
1058         (*ac)->ac_bits_wanted = 1;
1059         (*ac)->ac_which = OCFS2_AC_USE_INODE;
1060
1061         (*ac)->ac_group_search = ocfs2_block_group_search;
1062
1063         /*
1064          * stat(2) can't handle i_ino > 32bits, so we tell the
1065          * lower levels not to allocate us a block group past that
1066          * limit.  The 'inode64' mount option avoids this behavior.
1067          */
1068         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
1069                 (*ac)->ac_max_block = (u32)~0U;
1070
1071         /*
1072          * slot is set when we successfully steal inode from other nodes.
1073          * It is reset in 3 places:
1074          * 1. when we flush the truncate log
1075          * 2. when we complete local alloc recovery.
1076          * 3. when we successfully allocate from our own slot.
1077          * After it is set, we will go on stealing inodes until we find the
1078          * need to check our slots to see whether there is some space for us.
1079          */
1080         if (slot != OCFS2_INVALID_SLOT &&
1081             atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
1082                 goto inode_steal;
1083
1084         atomic_set(&osb->s_num_inodes_stolen, 0);
1085         alloc_group = osb->osb_inode_alloc_group;
1086         status = ocfs2_reserve_suballoc_bits(osb, *ac,
1087                                              INODE_ALLOC_SYSTEM_INODE,
1088                                              (u32)osb->slot_num,
1089                                              &alloc_group,
1090                                              ALLOC_NEW_GROUP |
1091                                              ALLOC_GROUPS_FROM_GLOBAL);
1092         if (status >= 0) {
1093                 status = 0;
1094
1095                 spin_lock(&osb->osb_lock);
1096                 osb->osb_inode_alloc_group = alloc_group;
1097                 spin_unlock(&osb->osb_lock);
1098                 trace_ocfs2_reserve_new_inode_new_group(
1099                         (unsigned long long)alloc_group);
1100
1101                 /*
1102                  * Some inodes must be freed by us, so try to allocate
1103                  * from our own next time.
1104                  */
1105                 if (slot != OCFS2_INVALID_SLOT)
1106                         ocfs2_init_inode_steal_slot(osb);
1107                 goto bail;
1108         } else if (status < 0 && status != -ENOSPC) {
1109                 mlog_errno(status);
1110                 goto bail;
1111         }
1112
1113         ocfs2_free_ac_resource(*ac);
1114
1115 inode_steal:
1116         status = ocfs2_steal_inode(osb, *ac);
1117         atomic_inc(&osb->s_num_inodes_stolen);
1118         if (status < 0) {
1119                 if (status != -ENOSPC)
1120                         mlog_errno(status);
1121                 goto bail;
1122         }
1123
1124         status = 0;
1125 bail:
1126         if ((status < 0) && *ac) {
1127                 ocfs2_free_alloc_context(*ac);
1128                 *ac = NULL;
1129         }
1130
1131         if (status)
1132                 mlog_errno(status);
1133         return status;
1134 }
1135
1136 /* local alloc code has to do the same thing, so rather than do this
1137  * twice.. */
1138 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
1139                                       struct ocfs2_alloc_context *ac)
1140 {
1141         int status;
1142
1143         ac->ac_which = OCFS2_AC_USE_MAIN;
1144         ac->ac_group_search = ocfs2_cluster_group_search;
1145
1146         status = ocfs2_reserve_suballoc_bits(osb, ac,
1147                                              GLOBAL_BITMAP_SYSTEM_INODE,
1148                                              OCFS2_INVALID_SLOT, NULL,
1149                                              ALLOC_NEW_GROUP);
1150         if (status < 0 && status != -ENOSPC) {
1151                 mlog_errno(status);
1152                 goto bail;
1153         }
1154
1155 bail:
1156         return status;
1157 }
1158
1159 /* Callers don't need to care which bitmap (local alloc or main) to
1160  * use so we figure it out for them, but unfortunately this clutters
1161  * things a bit. */
1162 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
1163                                              u32 bits_wanted, u64 max_block,
1164                                              int flags,
1165                                              struct ocfs2_alloc_context **ac)
1166 {
1167         int status;
1168
1169         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
1170         if (!(*ac)) {
1171                 status = -ENOMEM;
1172                 mlog_errno(status);
1173                 goto bail;
1174         }
1175
1176         (*ac)->ac_bits_wanted = bits_wanted;
1177         (*ac)->ac_max_block = max_block;
1178
1179         status = -ENOSPC;
1180         if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
1181             ocfs2_alloc_should_use_local(osb, bits_wanted)) {
1182                 status = ocfs2_reserve_local_alloc_bits(osb,
1183                                                         bits_wanted,
1184                                                         *ac);
1185                 if ((status < 0) && (status != -ENOSPC)) {
1186                         mlog_errno(status);
1187                         goto bail;
1188                 }
1189         }
1190
1191         if (status == -ENOSPC) {
1192                 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
1193                 if (status < 0) {
1194                         if (status != -ENOSPC)
1195                                 mlog_errno(status);
1196                         goto bail;
1197                 }
1198         }
1199
1200         status = 0;
1201 bail:
1202         if ((status < 0) && *ac) {
1203                 ocfs2_free_alloc_context(*ac);
1204                 *ac = NULL;
1205         }
1206
1207         if (status)
1208                 mlog_errno(status);
1209         return status;
1210 }
1211
1212 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
1213                            u32 bits_wanted,
1214                            struct ocfs2_alloc_context **ac)
1215 {
1216         return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
1217                                                  ALLOC_NEW_GROUP, ac);
1218 }
1219
1220 /*
1221  * More or less lifted from ext3. I'll leave their description below:
1222  *
1223  * "For ext3 allocations, we must not reuse any blocks which are
1224  * allocated in the bitmap buffer's "last committed data" copy.  This
1225  * prevents deletes from freeing up the page for reuse until we have
1226  * committed the delete transaction.
1227  *
1228  * If we didn't do this, then deleting something and reallocating it as
1229  * data would allow the old block to be overwritten before the
1230  * transaction committed (because we force data to disk before commit).
1231  * This would lead to corruption if we crashed between overwriting the
1232  * data and committing the delete.
1233  *
1234  * @@@ We may want to make this allocation behaviour conditional on
1235  * data-writes at some point, and disable it for metadata allocations or
1236  * sync-data inodes."
1237  *
1238  * Note: OCFS2 already does this differently for metadata vs data
1239  * allocations, as those bitmaps are separate and undo access is never
1240  * called on a metadata group descriptor.
1241  */
1242 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
1243                                          int nr)
1244 {
1245         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1246         int ret;
1247
1248         if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
1249                 return 0;
1250
1251         if (!buffer_jbd(bg_bh))
1252                 return 1;
1253
1254         jbd_lock_bh_state(bg_bh);
1255         bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
1256         if (bg)
1257                 ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
1258         else
1259                 ret = 1;
1260         jbd_unlock_bh_state(bg_bh);
1261
1262         return ret;
1263 }
1264
1265 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
1266                                              struct buffer_head *bg_bh,
1267                                              unsigned int bits_wanted,
1268                                              unsigned int total_bits,
1269                                              struct ocfs2_suballoc_result *res)
1270 {
1271         void *bitmap;
1272         u16 best_offset, best_size;
1273         int offset, start, found, status = 0;
1274         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1275
1276         /* Callers got this descriptor from
1277          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1278         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1279
1280         found = start = best_offset = best_size = 0;
1281         bitmap = bg->bg_bitmap;
1282
1283         while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
1284                 if (offset == total_bits)
1285                         break;
1286
1287                 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
1288                         /* We found a zero, but we can't use it as it
1289                          * hasn't been put to disk yet! */
1290                         found = 0;
1291                         start = offset + 1;
1292                 } else if (offset == start) {
1293                         /* we found a zero */
1294                         found++;
1295                         /* move start to the next bit to test */
1296                         start++;
1297                 } else {
1298                         /* got a zero after some ones */
1299                         found = 1;
1300                         start = offset + 1;
1301                 }
1302                 if (found > best_size) {
1303                         best_size = found;
1304                         best_offset = start - found;
1305                 }
1306                 /* we got everything we needed */
1307                 if (found == bits_wanted) {
1308                         /* mlog(0, "Found it all!\n"); */
1309                         break;
1310                 }
1311         }
1312
1313         if (best_size) {
1314                 res->sr_bit_offset = best_offset;
1315                 res->sr_bits = best_size;
1316         } else {
1317                 status = -ENOSPC;
1318                 /* No error log here -- see the comment above
1319                  * ocfs2_test_bg_bit_allocatable */
1320         }
1321
1322         return status;
1323 }
1324
1325 int ocfs2_block_group_set_bits(handle_t *handle,
1326                                              struct inode *alloc_inode,
1327                                              struct ocfs2_group_desc *bg,
1328                                              struct buffer_head *group_bh,
1329                                              unsigned int bit_off,
1330                                              unsigned int num_bits)
1331 {
1332         int status;
1333         void *bitmap = bg->bg_bitmap;
1334         int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1335
1336         /* All callers get the descriptor via
1337          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1338         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1339         BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1340
1341         trace_ocfs2_block_group_set_bits(bit_off, num_bits);
1342
1343         if (ocfs2_is_cluster_bitmap(alloc_inode))
1344                 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1345
1346         status = ocfs2_journal_access_gd(handle,
1347                                          INODE_CACHE(alloc_inode),
1348                                          group_bh,
1349                                          journal_type);
1350         if (status < 0) {
1351                 mlog_errno(status);
1352                 goto bail;
1353         }
1354
1355         le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1356         if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
1357                 return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
1358                                    (unsigned long long)le64_to_cpu(bg->bg_blkno),
1359                                    le16_to_cpu(bg->bg_bits),
1360                                    le16_to_cpu(bg->bg_free_bits_count),
1361                                    num_bits);
1362         }
1363         while(num_bits--)
1364                 ocfs2_set_bit(bit_off++, bitmap);
1365
1366         ocfs2_journal_dirty(handle, group_bh);
1367
1368 bail:
1369         return status;
1370 }
1371
1372 /* find the one with the most empty bits */
1373 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1374 {
1375         u16 curr, best;
1376
1377         BUG_ON(!cl->cl_next_free_rec);
1378
1379         best = curr = 0;
1380         while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1381                 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1382                     le32_to_cpu(cl->cl_recs[best].c_free))
1383                         best = curr;
1384                 curr++;
1385         }
1386
1387         BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1388         return best;
1389 }
1390
1391 static int ocfs2_relink_block_group(handle_t *handle,
1392                                     struct inode *alloc_inode,
1393                                     struct buffer_head *fe_bh,
1394                                     struct buffer_head *bg_bh,
1395                                     struct buffer_head *prev_bg_bh,
1396                                     u16 chain)
1397 {
1398         int status;
1399         /* there is a really tiny chance the journal calls could fail,
1400          * but we wouldn't want inconsistent blocks in *any* case. */
1401         u64 bg_ptr, prev_bg_ptr;
1402         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1403         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1404         struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1405
1406         /* The caller got these descriptors from
1407          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1408         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1409         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1410
1411         trace_ocfs2_relink_block_group(
1412                 (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1413                 (unsigned long long)le64_to_cpu(bg->bg_blkno),
1414                 (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1415
1416         bg_ptr = le64_to_cpu(bg->bg_next_group);
1417         prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1418
1419         status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1420                                          prev_bg_bh,
1421                                          OCFS2_JOURNAL_ACCESS_WRITE);
1422         if (status < 0)
1423                 goto out;
1424
1425         prev_bg->bg_next_group = bg->bg_next_group;
1426         ocfs2_journal_dirty(handle, prev_bg_bh);
1427
1428         status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1429                                          bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1430         if (status < 0)
1431                 goto out_rollback_prev_bg;
1432
1433         bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1434         ocfs2_journal_dirty(handle, bg_bh);
1435
1436         status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1437                                          fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1438         if (status < 0)
1439                 goto out_rollback_bg;
1440
1441         fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1442         ocfs2_journal_dirty(handle, fe_bh);
1443
1444 out:
1445         if (status < 0)
1446                 mlog_errno(status);
1447         return status;
1448
1449 out_rollback_bg:
1450         bg->bg_next_group = cpu_to_le64(bg_ptr);
1451 out_rollback_prev_bg:
1452         prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1453         goto out;
1454 }
1455
1456 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1457                                                      u32 wanted)
1458 {
1459         return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1460 }
1461
1462 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1463  * value on error. */
1464 static int ocfs2_cluster_group_search(struct inode *inode,
1465                                       struct buffer_head *group_bh,
1466                                       u32 bits_wanted, u32 min_bits,
1467                                       u64 max_block,
1468                                       struct ocfs2_suballoc_result *res)
1469 {
1470         int search = -ENOSPC;
1471         int ret;
1472         u64 blkoff;
1473         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1474         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1475         unsigned int max_bits, gd_cluster_off;
1476
1477         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1478
1479         if (gd->bg_free_bits_count) {
1480                 max_bits = le16_to_cpu(gd->bg_bits);
1481
1482                 /* Tail groups in cluster bitmaps which aren't cpg
1483                  * aligned are prone to partial extension by a failed
1484                  * fs resize. If the file system resize never got to
1485                  * update the dinode cluster count, then we don't want
1486                  * to trust any clusters past it, regardless of what
1487                  * the group descriptor says. */
1488                 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1489                                                           le64_to_cpu(gd->bg_blkno));
1490                 if ((gd_cluster_off + max_bits) >
1491                     OCFS2_I(inode)->ip_clusters) {
1492                         max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1493                         trace_ocfs2_cluster_group_search_wrong_max_bits(
1494                                 (unsigned long long)le64_to_cpu(gd->bg_blkno),
1495                                 le16_to_cpu(gd->bg_bits),
1496                                 OCFS2_I(inode)->ip_clusters, max_bits);
1497                 }
1498
1499                 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1500                                                         group_bh, bits_wanted,
1501                                                         max_bits, res);
1502                 if (ret)
1503                         return ret;
1504
1505                 if (max_block) {
1506                         blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1507                                                           gd_cluster_off +
1508                                                           res->sr_bit_offset +
1509                                                           res->sr_bits);
1510                         trace_ocfs2_cluster_group_search_max_block(
1511                                 (unsigned long long)blkoff,
1512                                 (unsigned long long)max_block);
1513                         if (blkoff > max_block)
1514                                 return -ENOSPC;
1515                 }
1516
1517                 /* ocfs2_block_group_find_clear_bits() might
1518                  * return success, but we still want to return
1519                  * -ENOSPC unless it found the minimum number
1520                  * of bits. */
1521                 if (min_bits <= res->sr_bits)
1522                         search = 0; /* success */
1523                 else if (res->sr_bits) {
1524                         /*
1525                          * Don't show bits which we'll be returning
1526                          * for allocation to the local alloc bitmap.
1527                          */
1528                         ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
1529                 }
1530         }
1531
1532         return search;
1533 }
1534
1535 static int ocfs2_block_group_search(struct inode *inode,
1536                                     struct buffer_head *group_bh,
1537                                     u32 bits_wanted, u32 min_bits,
1538                                     u64 max_block,
1539                                     struct ocfs2_suballoc_result *res)
1540 {
1541         int ret = -ENOSPC;
1542         u64 blkoff;
1543         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1544
1545         BUG_ON(min_bits != 1);
1546         BUG_ON(ocfs2_is_cluster_bitmap(inode));
1547
1548         if (bg->bg_free_bits_count) {
1549                 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1550                                                         group_bh, bits_wanted,
1551                                                         le16_to_cpu(bg->bg_bits),
1552                                                         res);
1553                 if (!ret && max_block) {
1554                         blkoff = le64_to_cpu(bg->bg_blkno) +
1555                                 res->sr_bit_offset + res->sr_bits;
1556                         trace_ocfs2_block_group_search_max_block(
1557                                 (unsigned long long)blkoff,
1558                                 (unsigned long long)max_block);
1559                         if (blkoff > max_block)
1560                                 ret = -ENOSPC;
1561                 }
1562         }
1563
1564         return ret;
1565 }
1566
1567 int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1568                                        handle_t *handle,
1569                                        struct buffer_head *di_bh,
1570                                        u32 num_bits,
1571                                        u16 chain)
1572 {
1573         int ret;
1574         u32 tmp_used;
1575         struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1576         struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1577
1578         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1579                                       OCFS2_JOURNAL_ACCESS_WRITE);
1580         if (ret < 0) {
1581                 mlog_errno(ret);
1582                 goto out;
1583         }
1584
1585         tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1586         di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1587         le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1588         ocfs2_journal_dirty(handle, di_bh);
1589
1590 out:
1591         return ret;
1592 }
1593
1594 void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
1595                                        struct buffer_head *di_bh,
1596                                        u32 num_bits,
1597                                        u16 chain)
1598 {
1599         u32 tmp_used;
1600         struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1601         struct ocfs2_chain_list *cl;
1602
1603         cl = (struct ocfs2_chain_list *)&di->id2.i_chain;
1604         tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1605         di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits);
1606         le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits);
1607 }
1608
1609 static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1610                                          struct ocfs2_extent_rec *rec,
1611                                          struct ocfs2_chain_list *cl)
1612 {
1613         unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1614         unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1615         unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1616
1617         if (res->sr_bit_offset < bitoff)
1618                 return 0;
1619         if (res->sr_bit_offset >= (bitoff + bitcount))
1620                 return 0;
1621         res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1622                 (res->sr_bit_offset - bitoff);
1623         if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1624                 res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1625         return 1;
1626 }
1627
1628 static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1629                                           struct ocfs2_group_desc *bg,
1630                                           struct ocfs2_suballoc_result *res)
1631 {
1632         int i;
1633         u64 bg_blkno = res->sr_bg_blkno;  /* Save off */
1634         struct ocfs2_extent_rec *rec;
1635         struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1636         struct ocfs2_chain_list *cl = &di->id2.i_chain;
1637
1638         if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1639                 res->sr_blkno = 0;
1640                 return;
1641         }
1642
1643         res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1644         res->sr_bg_blkno = 0;  /* Clear it for contig block groups */
1645         if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1646             !bg->bg_list.l_next_free_rec)
1647                 return;
1648
1649         for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1650                 rec = &bg->bg_list.l_recs[i];
1651                 if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1652                         res->sr_bg_blkno = bg_blkno;  /* Restore */
1653                         break;
1654                 }
1655         }
1656 }
1657
1658 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1659                                   handle_t *handle,
1660                                   u32 bits_wanted,
1661                                   u32 min_bits,
1662                                   struct ocfs2_suballoc_result *res,
1663                                   u16 *bits_left)
1664 {
1665         int ret;
1666         struct buffer_head *group_bh = NULL;
1667         struct ocfs2_group_desc *gd;
1668         struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1669         struct inode *alloc_inode = ac->ac_inode;
1670
1671         ret = ocfs2_read_group_descriptor(alloc_inode, di,
1672                                           res->sr_bg_blkno, &group_bh);
1673         if (ret < 0) {
1674                 mlog_errno(ret);
1675                 return ret;
1676         }
1677
1678         gd = (struct ocfs2_group_desc *) group_bh->b_data;
1679         ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1680                                   ac->ac_max_block, res);
1681         if (ret < 0) {
1682                 if (ret != -ENOSPC)
1683                         mlog_errno(ret);
1684                 goto out;
1685         }
1686
1687         if (!ret)
1688                 ocfs2_bg_discontig_fix_result(ac, gd, res);
1689
1690         /*
1691          * sr_bg_blkno might have been changed by
1692          * ocfs2_bg_discontig_fix_result
1693          */
1694         res->sr_bg_stable_blkno = group_bh->b_blocknr;
1695
1696         if (ac->ac_find_loc_only)
1697                 goto out_loc_only;
1698
1699         ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1700                                                res->sr_bits,
1701                                                le16_to_cpu(gd->bg_chain));
1702         if (ret < 0) {
1703                 mlog_errno(ret);
1704                 goto out;
1705         }
1706
1707         ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1708                                          res->sr_bit_offset, res->sr_bits);
1709         if (ret < 0) {
1710                 ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh,
1711                                                res->sr_bits,
1712                                                le16_to_cpu(gd->bg_chain));
1713                 mlog_errno(ret);
1714         }
1715
1716 out_loc_only:
1717         *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1718
1719 out:
1720         brelse(group_bh);
1721
1722         return ret;
1723 }
1724
1725 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1726                               handle_t *handle,
1727                               u32 bits_wanted,
1728                               u32 min_bits,
1729                               struct ocfs2_suballoc_result *res,
1730                               u16 *bits_left)
1731 {
1732         int status;
1733         u16 chain;
1734         u64 next_group;
1735         struct inode *alloc_inode = ac->ac_inode;
1736         struct buffer_head *group_bh = NULL;
1737         struct buffer_head *prev_group_bh = NULL;
1738         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1739         struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1740         struct ocfs2_group_desc *bg;
1741
1742         chain = ac->ac_chain;
1743         trace_ocfs2_search_chain_begin(
1744                 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1745                 bits_wanted, chain);
1746
1747         status = ocfs2_read_group_descriptor(alloc_inode, fe,
1748                                              le64_to_cpu(cl->cl_recs[chain].c_blkno),
1749                                              &group_bh);
1750         if (status < 0) {
1751                 mlog_errno(status);
1752                 goto bail;
1753         }
1754         bg = (struct ocfs2_group_desc *) group_bh->b_data;
1755
1756         status = -ENOSPC;
1757         /* for now, the chain search is a bit simplistic. We just use
1758          * the 1st group with any empty bits. */
1759         while ((status = ac->ac_group_search(alloc_inode, group_bh,
1760                                              bits_wanted, min_bits,
1761                                              ac->ac_max_block,
1762                                              res)) == -ENOSPC) {
1763                 if (!bg->bg_next_group)
1764                         break;
1765
1766                 brelse(prev_group_bh);
1767                 prev_group_bh = NULL;
1768
1769                 next_group = le64_to_cpu(bg->bg_next_group);
1770                 prev_group_bh = group_bh;
1771                 group_bh = NULL;
1772                 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1773                                                      next_group, &group_bh);
1774                 if (status < 0) {
1775                         mlog_errno(status);
1776                         goto bail;
1777                 }
1778                 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1779         }
1780         if (status < 0) {
1781                 if (status != -ENOSPC)
1782                         mlog_errno(status);
1783                 goto bail;
1784         }
1785
1786         trace_ocfs2_search_chain_succ(
1787                 (unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1788
1789         res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1790
1791         BUG_ON(res->sr_bits == 0);
1792         if (!status)
1793                 ocfs2_bg_discontig_fix_result(ac, bg, res);
1794
1795         /*
1796          * sr_bg_blkno might have been changed by
1797          * ocfs2_bg_discontig_fix_result
1798          */
1799         res->sr_bg_stable_blkno = group_bh->b_blocknr;
1800
1801         /*
1802          * Keep track of previous block descriptor read. When
1803          * we find a target, if we have read more than X
1804          * number of descriptors, and the target is reasonably
1805          * empty, relink him to top of his chain.
1806          *
1807          * We've read 0 extra blocks and only send one more to
1808          * the transaction, yet the next guy to search has a
1809          * much easier time.
1810          *
1811          * Do this *after* figuring out how many bits we're taking out
1812          * of our target group.
1813          */
1814         if (!ac->ac_disable_chain_relink &&
1815             (prev_group_bh) &&
1816             (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1817                 status = ocfs2_relink_block_group(handle, alloc_inode,
1818                                                   ac->ac_bh, group_bh,
1819                                                   prev_group_bh, chain);
1820                 if (status < 0) {
1821                         mlog_errno(status);
1822                         goto bail;
1823                 }
1824         }
1825
1826         if (ac->ac_find_loc_only)
1827                 goto out_loc_only;
1828
1829         status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1830                                                   ac->ac_bh, res->sr_bits,
1831                                                   chain);
1832         if (status) {
1833                 mlog_errno(status);
1834                 goto bail;
1835         }
1836
1837         status = ocfs2_block_group_set_bits(handle,
1838                                             alloc_inode,
1839                                             bg,
1840                                             group_bh,
1841                                             res->sr_bit_offset,
1842                                             res->sr_bits);
1843         if (status < 0) {
1844                 ocfs2_rollback_alloc_dinode_counts(alloc_inode,
1845                                         ac->ac_bh, res->sr_bits, chain);
1846                 mlog_errno(status);
1847                 goto bail;
1848         }
1849
1850         trace_ocfs2_search_chain_end(
1851                         (unsigned long long)le64_to_cpu(fe->i_blkno),
1852                         res->sr_bits);
1853
1854 out_loc_only:
1855         *bits_left = le16_to_cpu(bg->bg_free_bits_count);
1856 bail:
1857         brelse(group_bh);
1858         brelse(prev_group_bh);
1859
1860         if (status)
1861                 mlog_errno(status);
1862         return status;
1863 }
1864
1865 /* will give out up to bits_wanted contiguous bits. */
1866 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1867                                      handle_t *handle,
1868                                      u32 bits_wanted,
1869                                      u32 min_bits,
1870                                      struct ocfs2_suballoc_result *res)
1871 {
1872         int status;
1873         u16 victim, i;
1874         u16 bits_left = 0;
1875         u64 hint = ac->ac_last_group;
1876         struct ocfs2_chain_list *cl;
1877         struct ocfs2_dinode *fe;
1878
1879         BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1880         BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1881         BUG_ON(!ac->ac_bh);
1882
1883         fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1884
1885         /* The bh was validated by the inode read during
1886          * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1887         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1888
1889         if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1890             le32_to_cpu(fe->id1.bitmap1.i_total)) {
1891                 status = ocfs2_error(ac->ac_inode->i_sb,
1892                                      "Chain allocator dinode %llu has %u used bits but only %u total\n",
1893                                      (unsigned long long)le64_to_cpu(fe->i_blkno),
1894                                      le32_to_cpu(fe->id1.bitmap1.i_used),
1895                                      le32_to_cpu(fe->id1.bitmap1.i_total));
1896                 goto bail;
1897         }
1898
1899         res->sr_bg_blkno = hint;
1900         if (res->sr_bg_blkno) {
1901                 /* Attempt to short-circuit the usual search mechanism
1902                  * by jumping straight to the most recently used
1903                  * allocation group. This helps us maintain some
1904                  * contiguousness across allocations. */
1905                 status = ocfs2_search_one_group(ac, handle, bits_wanted,
1906                                                 min_bits, res, &bits_left);
1907                 if (!status)
1908                         goto set_hint;
1909                 if (status < 0 && status != -ENOSPC) {
1910                         mlog_errno(status);
1911                         goto bail;
1912                 }
1913         }
1914
1915         cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1916
1917         victim = ocfs2_find_victim_chain(cl);
1918         ac->ac_chain = victim;
1919
1920         status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1921                                     res, &bits_left);
1922         if (!status) {
1923                 if (ocfs2_is_cluster_bitmap(ac->ac_inode))
1924                         hint = res->sr_bg_blkno;
1925                 else
1926                         hint = ocfs2_group_from_res(res);
1927                 goto set_hint;
1928         }
1929         if (status < 0 && status != -ENOSPC) {
1930                 mlog_errno(status);
1931                 goto bail;
1932         }
1933
1934         trace_ocfs2_claim_suballoc_bits(victim);
1935
1936         /* If we didn't pick a good victim, then just default to
1937          * searching each chain in order. Don't allow chain relinking
1938          * because we only calculate enough journal credits for one
1939          * relink per alloc. */
1940         ac->ac_disable_chain_relink = 1;
1941         for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1942                 if (i == victim)
1943                         continue;
1944                 if (!cl->cl_recs[i].c_free)
1945                         continue;
1946
1947                 ac->ac_chain = i;
1948                 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1949                                             res, &bits_left);
1950                 if (!status) {
1951                         hint = ocfs2_group_from_res(res);
1952                         break;
1953                 }
1954                 if (status < 0 && status != -ENOSPC) {
1955                         mlog_errno(status);
1956                         goto bail;
1957                 }
1958         }
1959
1960 set_hint:
1961         if (status != -ENOSPC) {
1962                 /* If the next search of this group is not likely to
1963                  * yield a suitable extent, then we reset the last
1964                  * group hint so as to not waste a disk read */
1965                 if (bits_left < min_bits)
1966                         ac->ac_last_group = 0;
1967                 else
1968                         ac->ac_last_group = hint;
1969         }
1970
1971 bail:
1972         if (status)
1973                 mlog_errno(status);
1974         return status;
1975 }
1976
1977 int ocfs2_claim_metadata(handle_t *handle,
1978                          struct ocfs2_alloc_context *ac,
1979                          u32 bits_wanted,
1980                          u64 *suballoc_loc,
1981                          u16 *suballoc_bit_start,
1982                          unsigned int *num_bits,
1983                          u64 *blkno_start)
1984 {
1985         int status;
1986         struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
1987
1988         BUG_ON(!ac);
1989         BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1990         BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
1991
1992         status = ocfs2_claim_suballoc_bits(ac,
1993                                            handle,
1994                                            bits_wanted,
1995                                            1,
1996                                            &res);
1997         if (status < 0) {
1998                 mlog_errno(status);
1999                 goto bail;
2000         }
2001         atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2002
2003         *suballoc_loc = res.sr_bg_blkno;
2004         *suballoc_bit_start = res.sr_bit_offset;
2005         *blkno_start = res.sr_blkno;
2006         ac->ac_bits_given += res.sr_bits;
2007         *num_bits = res.sr_bits;
2008         status = 0;
2009 bail:
2010         if (status)
2011                 mlog_errno(status);
2012         return status;
2013 }
2014
2015 static void ocfs2_init_inode_ac_group(struct inode *dir,
2016                                       struct buffer_head *parent_di_bh,
2017                                       struct ocfs2_alloc_context *ac)
2018 {
2019         struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2020         /*
2021          * Try to allocate inodes from some specific group.
2022          *
2023          * If the parent dir has recorded the last group used in allocation,
2024          * cool, use it. Otherwise if we try to allocate new inode from the
2025          * same slot the parent dir belongs to, use the same chunk.
2026          *
2027          * We are very careful here to avoid the mistake of setting
2028          * ac_last_group to a group descriptor from a different (unlocked) slot.
2029          */
2030         if (OCFS2_I(dir)->ip_last_used_group &&
2031             OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2032                 ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2033         else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2034                 if (di->i_suballoc_loc)
2035                         ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2036                 else
2037                         ac->ac_last_group = ocfs2_which_suballoc_group(
2038                                         le64_to_cpu(di->i_blkno),
2039                                         le16_to_cpu(di->i_suballoc_bit));
2040         }
2041 }
2042
2043 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2044                                              struct ocfs2_alloc_context *ac)
2045 {
2046         OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2047         OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2048 }
2049
2050 int ocfs2_find_new_inode_loc(struct inode *dir,
2051                              struct buffer_head *parent_fe_bh,
2052                              struct ocfs2_alloc_context *ac,
2053                              u64 *fe_blkno)
2054 {
2055         int ret;
2056         handle_t *handle = NULL;
2057         struct ocfs2_suballoc_result *res;
2058
2059         BUG_ON(!ac);
2060         BUG_ON(ac->ac_bits_given != 0);
2061         BUG_ON(ac->ac_bits_wanted != 1);
2062         BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2063
2064         res = kzalloc(sizeof(*res), GFP_NOFS);
2065         if (res == NULL) {
2066                 ret = -ENOMEM;
2067                 mlog_errno(ret);
2068                 goto out;
2069         }
2070
2071         ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2072
2073         /*
2074          * The handle started here is for chain relink. Alternatively,
2075          * we could just disable relink for these calls.
2076          */
2077         handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2078         if (IS_ERR(handle)) {
2079                 ret = PTR_ERR(handle);
2080                 handle = NULL;
2081                 mlog_errno(ret);
2082                 goto out;
2083         }
2084
2085         /*
2086          * This will instruct ocfs2_claim_suballoc_bits and
2087          * ocfs2_search_one_group to search but save actual allocation
2088          * for later.
2089          */
2090         ac->ac_find_loc_only = 1;
2091
2092         ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2093         if (ret < 0) {
2094                 mlog_errno(ret);
2095                 goto out;
2096         }
2097
2098         ac->ac_find_loc_priv = res;
2099         *fe_blkno = res->sr_blkno;
2100         ocfs2_update_inode_fsync_trans(handle, dir, 0);
2101 out:
2102         if (handle)
2103                 ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2104
2105         if (ret)
2106                 kfree(res);
2107
2108         return ret;
2109 }
2110
2111 int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2112                                  struct inode *dir,
2113                                  struct ocfs2_alloc_context *ac,
2114                                  u64 *suballoc_loc,
2115                                  u16 *suballoc_bit,
2116                                  u64 di_blkno)
2117 {
2118         int ret;
2119         u16 chain;
2120         struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2121         struct buffer_head *bg_bh = NULL;
2122         struct ocfs2_group_desc *bg;
2123         struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2124
2125         /*
2126          * Since di_blkno is being passed back in, we check for any
2127          * inconsistencies which may have happened between
2128          * calls. These are code bugs as di_blkno is not expected to
2129          * change once returned from ocfs2_find_new_inode_loc()
2130          */
2131         BUG_ON(res->sr_blkno != di_blkno);
2132
2133         ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2134                                           res->sr_bg_stable_blkno, &bg_bh);
2135         if (ret) {
2136                 mlog_errno(ret);
2137                 goto out;
2138         }
2139
2140         bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2141         chain = le16_to_cpu(bg->bg_chain);
2142
2143         ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2144                                                ac->ac_bh, res->sr_bits,
2145                                                chain);
2146         if (ret) {
2147                 mlog_errno(ret);
2148                 goto out;
2149         }
2150
2151         ret = ocfs2_block_group_set_bits(handle,
2152                                          ac->ac_inode,
2153                                          bg,
2154                                          bg_bh,
2155                                          res->sr_bit_offset,
2156                                          res->sr_bits);
2157         if (ret < 0) {
2158                 ocfs2_rollback_alloc_dinode_counts(ac->ac_inode,
2159                                                ac->ac_bh, res->sr_bits, chain);
2160                 mlog_errno(ret);
2161                 goto out;
2162         }
2163
2164         trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2165                                            res->sr_bits);
2166
2167         atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2168
2169         BUG_ON(res->sr_bits != 1);
2170
2171         *suballoc_loc = res->sr_bg_blkno;
2172         *suballoc_bit = res->sr_bit_offset;
2173         ac->ac_bits_given++;
2174         ocfs2_save_inode_ac_group(dir, ac);
2175
2176 out:
2177         brelse(bg_bh);
2178
2179         return ret;
2180 }
2181
2182 int ocfs2_claim_new_inode(handle_t *handle,
2183                           struct inode *dir,
2184                           struct buffer_head *parent_fe_bh,
2185                           struct ocfs2_alloc_context *ac,
2186                           u64 *suballoc_loc,
2187                           u16 *suballoc_bit,
2188                           u64 *fe_blkno)
2189 {
2190         int status;
2191         struct ocfs2_suballoc_result res;
2192
2193         BUG_ON(!ac);
2194         BUG_ON(ac->ac_bits_given != 0);
2195         BUG_ON(ac->ac_bits_wanted != 1);
2196         BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2197
2198         ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2199
2200         status = ocfs2_claim_suballoc_bits(ac,
2201                                            handle,
2202                                            1,
2203                                            1,
2204                                            &res);
2205         if (status < 0) {
2206                 mlog_errno(status);
2207                 goto bail;
2208         }
2209         atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2210
2211         BUG_ON(res.sr_bits != 1);
2212
2213         *suballoc_loc = res.sr_bg_blkno;
2214         *suballoc_bit = res.sr_bit_offset;
2215         *fe_blkno = res.sr_blkno;
2216         ac->ac_bits_given++;
2217         ocfs2_save_inode_ac_group(dir, ac);
2218         status = 0;
2219 bail:
2220         if (status)
2221                 mlog_errno(status);
2222         return status;
2223 }
2224
2225 /* translate a group desc. blkno and it's bitmap offset into
2226  * disk cluster offset. */
2227 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2228                                                    u64 bg_blkno,
2229                                                    u16 bg_bit_off)
2230 {
2231         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2232         u32 cluster = 0;
2233
2234         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2235
2236         if (bg_blkno != osb->first_cluster_group_blkno)
2237                 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2238         cluster += (u32) bg_bit_off;
2239         return cluster;
2240 }
2241
2242 /* given a cluster offset, calculate which block group it belongs to
2243  * and return that block offset. */
2244 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2245 {
2246         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2247         u32 group_no;
2248
2249         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2250
2251         group_no = cluster / osb->bitmap_cpg;
2252         if (!group_no)
2253                 return osb->first_cluster_group_blkno;
2254         return ocfs2_clusters_to_blocks(inode->i_sb,
2255                                         group_no * osb->bitmap_cpg);
2256 }
2257
2258 /* given the block number of a cluster start, calculate which cluster
2259  * group and descriptor bitmap offset that corresponds to. */
2260 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2261                                                 u64 data_blkno,
2262                                                 u64 *bg_blkno,
2263                                                 u16 *bg_bit_off)
2264 {
2265         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2266         u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2267
2268         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2269
2270         *bg_blkno = ocfs2_which_cluster_group(inode,
2271                                               data_cluster);
2272
2273         if (*bg_blkno == osb->first_cluster_group_blkno)
2274                 *bg_bit_off = (u16) data_cluster;
2275         else
2276                 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2277                                                              data_blkno - *bg_blkno);
2278 }
2279
2280 /*
2281  * min_bits - minimum contiguous chunk from this total allocation we
2282  * can handle. set to what we asked for originally for a full
2283  * contig. allocation, set to '1' to indicate we can deal with extents
2284  * of any size.
2285  */
2286 int __ocfs2_claim_clusters(handle_t *handle,
2287                            struct ocfs2_alloc_context *ac,
2288                            u32 min_clusters,
2289                            u32 max_clusters,
2290                            u32 *cluster_start,
2291                            u32 *num_clusters)
2292 {
2293         int status;
2294         unsigned int bits_wanted = max_clusters;
2295         struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2296         struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2297
2298         BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2299
2300         BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2301                && ac->ac_which != OCFS2_AC_USE_MAIN);
2302
2303         if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2304                 WARN_ON(min_clusters > 1);
2305
2306                 status = ocfs2_claim_local_alloc_bits(osb,
2307                                                       handle,
2308                                                       ac,
2309                                                       bits_wanted,
2310                                                       cluster_start,
2311                                                       num_clusters);
2312                 if (!status)
2313                         atomic_inc(&osb->alloc_stats.local_data);
2314         } else {
2315                 if (min_clusters > (osb->bitmap_cpg - 1)) {
2316                         /* The only paths asking for contiguousness
2317                          * should know about this already. */
2318                         mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2319                              "group bitmap size %u!\n", min_clusters,
2320                              osb->bitmap_cpg);
2321                         status = -ENOSPC;
2322                         goto bail;
2323                 }
2324                 /* clamp the current request down to a realistic size. */
2325                 if (bits_wanted > (osb->bitmap_cpg - 1))
2326                         bits_wanted = osb->bitmap_cpg - 1;
2327
2328                 status = ocfs2_claim_suballoc_bits(ac,
2329                                                    handle,
2330                                                    bits_wanted,
2331                                                    min_clusters,
2332                                                    &res);
2333                 if (!status) {
2334                         BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2335                         *cluster_start =
2336                                 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2337                                                                  res.sr_bg_blkno,
2338                                                                  res.sr_bit_offset);
2339                         atomic_inc(&osb->alloc_stats.bitmap_data);
2340                         *num_clusters = res.sr_bits;
2341                 }
2342         }
2343         if (status < 0) {
2344                 if (status != -ENOSPC)
2345                         mlog_errno(status);
2346                 goto bail;
2347         }
2348
2349         ac->ac_bits_given += *num_clusters;
2350
2351 bail:
2352         if (status)
2353                 mlog_errno(status);
2354         return status;
2355 }
2356
2357 int ocfs2_claim_clusters(handle_t *handle,
2358                          struct ocfs2_alloc_context *ac,
2359                          u32 min_clusters,
2360                          u32 *cluster_start,
2361                          u32 *num_clusters)
2362 {
2363         unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2364
2365         return __ocfs2_claim_clusters(handle, ac, min_clusters,
2366                                       bits_wanted, cluster_start, num_clusters);
2367 }
2368
2369 static int ocfs2_block_group_clear_bits(handle_t *handle,
2370                                         struct inode *alloc_inode,
2371                                         struct ocfs2_group_desc *bg,
2372                                         struct buffer_head *group_bh,
2373                                         unsigned int bit_off,
2374                                         unsigned int num_bits,
2375                                         void (*undo_fn)(unsigned int bit,
2376                                                         unsigned long *bmap))
2377 {
2378         int status;
2379         unsigned int tmp;
2380         struct ocfs2_group_desc *undo_bg = NULL;
2381
2382         /* The caller got this descriptor from
2383          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
2384         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2385
2386         trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2387
2388         BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2389         status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2390                                          group_bh,
2391                                          undo_fn ?
2392                                          OCFS2_JOURNAL_ACCESS_UNDO :
2393                                          OCFS2_JOURNAL_ACCESS_WRITE);
2394         if (status < 0) {
2395                 mlog_errno(status);
2396                 goto bail;
2397         }
2398
2399         if (undo_fn) {
2400                 jbd_lock_bh_state(group_bh);
2401                 undo_bg = (struct ocfs2_group_desc *)
2402                                         bh2jh(group_bh)->b_committed_data;
2403                 BUG_ON(!undo_bg);
2404         }
2405
2406         tmp = num_bits;
2407         while(tmp--) {
2408                 ocfs2_clear_bit((bit_off + tmp),
2409                                 (unsigned long *) bg->bg_bitmap);
2410                 if (undo_fn)
2411                         undo_fn(bit_off + tmp,
2412                                 (unsigned long *) undo_bg->bg_bitmap);
2413         }
2414         le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2415         if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2416                 return ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit count %u but claims %u are freed. num_bits %d\n",
2417                                    (unsigned long long)le64_to_cpu(bg->bg_blkno),
2418                                    le16_to_cpu(bg->bg_bits),
2419                                    le16_to_cpu(bg->bg_free_bits_count),
2420                                    num_bits);
2421         }
2422
2423         if (undo_fn)
2424                 jbd_unlock_bh_state(group_bh);
2425
2426         ocfs2_journal_dirty(handle, group_bh);
2427 bail:
2428         return status;
2429 }
2430
2431 /*
2432  * expects the suballoc inode to already be locked.
2433  */
2434 static int _ocfs2_free_suballoc_bits(handle_t *handle,
2435                                      struct inode *alloc_inode,
2436                                      struct buffer_head *alloc_bh,
2437                                      unsigned int start_bit,
2438                                      u64 bg_blkno,
2439                                      unsigned int count,
2440                                      void (*undo_fn)(unsigned int bit,
2441                                                      unsigned long *bitmap))
2442 {
2443         int status = 0;
2444         u32 tmp_used;
2445         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2446         struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2447         struct buffer_head *group_bh = NULL;
2448         struct ocfs2_group_desc *group;
2449
2450         /* The alloc_bh comes from ocfs2_free_dinode() or
2451          * ocfs2_free_clusters().  The callers have all locked the
2452          * allocator and gotten alloc_bh from the lock call.  This
2453          * validates the dinode buffer.  Any corruption that has happened
2454          * is a code bug. */
2455         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2456         BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2457
2458         trace_ocfs2_free_suballoc_bits(
2459                 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2460                 (unsigned long long)bg_blkno,
2461                 start_bit, count);
2462
2463         status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2464                                              &group_bh);
2465         if (status < 0) {
2466                 mlog_errno(status);
2467                 goto bail;
2468         }
2469         group = (struct ocfs2_group_desc *) group_bh->b_data;
2470
2471         BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2472
2473         status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2474                                               group, group_bh,
2475                                               start_bit, count, undo_fn);
2476         if (status < 0) {
2477                 mlog_errno(status);
2478                 goto bail;
2479         }
2480
2481         status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2482                                          alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2483         if (status < 0) {
2484                 mlog_errno(status);
2485                 ocfs2_block_group_set_bits(handle, alloc_inode, group, group_bh,
2486                                 start_bit, count);
2487                 goto bail;
2488         }
2489
2490         le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2491                      count);
2492         tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2493         fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2494         ocfs2_journal_dirty(handle, alloc_bh);
2495
2496 bail:
2497         brelse(group_bh);
2498
2499         if (status)
2500                 mlog_errno(status);
2501         return status;
2502 }
2503
2504 int ocfs2_free_suballoc_bits(handle_t *handle,
2505                              struct inode *alloc_inode,
2506                              struct buffer_head *alloc_bh,
2507                              unsigned int start_bit,
2508                              u64 bg_blkno,
2509                              unsigned int count)
2510 {
2511         return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2512                                          start_bit, bg_blkno, count, NULL);
2513 }
2514
2515 int ocfs2_free_dinode(handle_t *handle,
2516                       struct inode *inode_alloc_inode,
2517                       struct buffer_head *inode_alloc_bh,
2518                       struct ocfs2_dinode *di)
2519 {
2520         u64 blk = le64_to_cpu(di->i_blkno);
2521         u16 bit = le16_to_cpu(di->i_suballoc_bit);
2522         u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2523
2524         if (di->i_suballoc_loc)
2525                 bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2526         return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2527                                         inode_alloc_bh, bit, bg_blkno, 1);
2528 }
2529
2530 static int _ocfs2_free_clusters(handle_t *handle,
2531                                 struct inode *bitmap_inode,
2532                                 struct buffer_head *bitmap_bh,
2533                                 u64 start_blk,
2534                                 unsigned int num_clusters,
2535                                 void (*undo_fn)(unsigned int bit,
2536                                                 unsigned long *bitmap))
2537 {
2538         int status;
2539         u16 bg_start_bit;
2540         u64 bg_blkno;
2541         struct ocfs2_dinode *fe;
2542
2543         /* You can't ever have a contiguous set of clusters
2544          * bigger than a block group bitmap so we never have to worry
2545          * about looping on them.
2546          * This is expensive. We can safely remove once this stuff has
2547          * gotten tested really well. */
2548         BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
2549
2550         fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
2551
2552         ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2553                                      &bg_start_bit);
2554
2555         trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2556                         (unsigned long long)start_blk,
2557                         bg_start_bit, num_clusters);
2558
2559         status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2560                                            bg_start_bit, bg_blkno,
2561                                            num_clusters, undo_fn);
2562         if (status < 0) {
2563                 mlog_errno(status);
2564                 goto out;
2565         }
2566
2567         ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2568                                          num_clusters);
2569
2570 out:
2571         if (status)
2572                 mlog_errno(status);
2573         return status;
2574 }
2575
2576 int ocfs2_free_clusters(handle_t *handle,
2577                         struct inode *bitmap_inode,
2578                         struct buffer_head *bitmap_bh,
2579                         u64 start_blk,
2580                         unsigned int num_clusters)
2581 {
2582         return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2583                                     start_blk, num_clusters,
2584                                     _ocfs2_set_bit);
2585 }
2586
2587 /*
2588  * Give never-used clusters back to the global bitmap.  We don't need
2589  * to protect these bits in the undo buffer.
2590  */
2591 int ocfs2_release_clusters(handle_t *handle,
2592                            struct inode *bitmap_inode,
2593                            struct buffer_head *bitmap_bh,
2594                            u64 start_blk,
2595                            unsigned int num_clusters)
2596 {
2597         return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2598                                     start_blk, num_clusters,
2599                                     _ocfs2_clear_bit);
2600 }
2601
2602 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
2603 {
2604         printk("Block Group:\n");
2605         printk("bg_signature:       %s\n", bg->bg_signature);
2606         printk("bg_size:            %u\n", bg->bg_size);
2607         printk("bg_bits:            %u\n", bg->bg_bits);
2608         printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
2609         printk("bg_chain:           %u\n", bg->bg_chain);
2610         printk("bg_generation:      %u\n", le32_to_cpu(bg->bg_generation));
2611         printk("bg_next_group:      %llu\n",
2612                (unsigned long long)bg->bg_next_group);
2613         printk("bg_parent_dinode:   %llu\n",
2614                (unsigned long long)bg->bg_parent_dinode);
2615         printk("bg_blkno:           %llu\n",
2616                (unsigned long long)bg->bg_blkno);
2617 }
2618
2619 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2620 {
2621         int i;
2622
2623         printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
2624         printk("i_signature:                  %s\n", fe->i_signature);
2625         printk("i_size:                       %llu\n",
2626                (unsigned long long)fe->i_size);
2627         printk("i_clusters:                   %u\n", fe->i_clusters);
2628         printk("i_generation:                 %u\n",
2629                le32_to_cpu(fe->i_generation));
2630         printk("id1.bitmap1.i_used:           %u\n",
2631                le32_to_cpu(fe->id1.bitmap1.i_used));
2632         printk("id1.bitmap1.i_total:          %u\n",
2633                le32_to_cpu(fe->id1.bitmap1.i_total));
2634         printk("id2.i_chain.cl_cpg:           %u\n", fe->id2.i_chain.cl_cpg);
2635         printk("id2.i_chain.cl_bpc:           %u\n", fe->id2.i_chain.cl_bpc);
2636         printk("id2.i_chain.cl_count:         %u\n", fe->id2.i_chain.cl_count);
2637         printk("id2.i_chain.cl_next_free_rec: %u\n",
2638                fe->id2.i_chain.cl_next_free_rec);
2639         for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2640                 printk("fe->id2.i_chain.cl_recs[%d].c_free:  %u\n", i,
2641                        fe->id2.i_chain.cl_recs[i].c_free);
2642                 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2643                        fe->id2.i_chain.cl_recs[i].c_total);
2644                 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2645                        (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
2646         }
2647 }
2648
2649 /*
2650  * For a given allocation, determine which allocators will need to be
2651  * accessed, and lock them, reserving the appropriate number of bits.
2652  *
2653  * Sparse file systems call this from ocfs2_write_begin_nolock()
2654  * and ocfs2_allocate_unwritten_extents().
2655  *
2656  * File systems which don't support holes call this from
2657  * ocfs2_extend_allocation().
2658  */
2659 int ocfs2_lock_allocators(struct inode *inode,
2660                           struct ocfs2_extent_tree *et,
2661                           u32 clusters_to_add, u32 extents_to_split,
2662                           struct ocfs2_alloc_context **data_ac,
2663                           struct ocfs2_alloc_context **meta_ac)
2664 {
2665         int ret = 0, num_free_extents;
2666         unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2667         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2668
2669         *meta_ac = NULL;
2670         if (data_ac)
2671                 *data_ac = NULL;
2672
2673         BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2674
2675         num_free_extents = ocfs2_num_free_extents(osb, et);
2676         if (num_free_extents < 0) {
2677                 ret = num_free_extents;
2678                 mlog_errno(ret);
2679                 goto out;
2680         }
2681
2682         /*
2683          * Sparse allocation file systems need to be more conservative
2684          * with reserving room for expansion - the actual allocation
2685          * happens while we've got a journal handle open so re-taking
2686          * a cluster lock (because we ran out of room for another
2687          * extent) will violate ordering rules.
2688          *
2689          * Most of the time we'll only be seeing this 1 cluster at a time
2690          * anyway.
2691          *
2692          * Always lock for any unwritten extents - we might want to
2693          * add blocks during a split.
2694          */
2695         if (!num_free_extents ||
2696             (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2697                 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2698                 if (ret < 0) {
2699                         if (ret != -ENOSPC)
2700                                 mlog_errno(ret);
2701                         goto out;
2702                 }
2703         }
2704
2705         if (clusters_to_add == 0)
2706                 goto out;
2707
2708         ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2709         if (ret < 0) {
2710                 if (ret != -ENOSPC)
2711                         mlog_errno(ret);
2712                 goto out;
2713         }
2714
2715 out:
2716         if (ret) {
2717                 if (*meta_ac) {
2718                         ocfs2_free_alloc_context(*meta_ac);
2719                         *meta_ac = NULL;
2720                 }
2721
2722                 /*
2723                  * We cannot have an error and a non null *data_ac.
2724                  */
2725         }
2726
2727         return ret;
2728 }
2729
2730 /*
2731  * Read the inode specified by blkno to get suballoc_slot and
2732  * suballoc_bit.
2733  */
2734 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2735                                        u16 *suballoc_slot, u64 *group_blkno,
2736                                        u16 *suballoc_bit)
2737 {
2738         int status;
2739         struct buffer_head *inode_bh = NULL;
2740         struct ocfs2_dinode *inode_fe;
2741
2742         trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2743
2744         /* dirty read disk */
2745         status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2746         if (status < 0) {
2747                 mlog(ML_ERROR, "read block %llu failed %d\n",
2748                      (unsigned long long)blkno, status);
2749                 goto bail;
2750         }
2751
2752         inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2753         if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2754                 mlog(ML_ERROR, "invalid inode %llu requested\n",
2755                      (unsigned long long)blkno);
2756                 status = -EINVAL;
2757                 goto bail;
2758         }
2759
2760         if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2761             (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2762                 mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2763                      (unsigned long long)blkno,
2764                      (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2765                 status = -EINVAL;
2766                 goto bail;
2767         }
2768
2769         if (suballoc_slot)
2770                 *suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2771         if (suballoc_bit)
2772                 *suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2773         if (group_blkno)
2774                 *group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2775
2776 bail:
2777         brelse(inode_bh);
2778
2779         if (status)
2780                 mlog_errno(status);
2781         return status;
2782 }
2783
2784 /*
2785  * test whether bit is SET in allocator bitmap or not.  on success, 0
2786  * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
2787  * is returned and *res is meaningless.  Call this after you have
2788  * cluster locked against suballoc, or you may get a result based on
2789  * non-up2date contents
2790  */
2791 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2792                                    struct inode *suballoc,
2793                                    struct buffer_head *alloc_bh,
2794                                    u64 group_blkno, u64 blkno,
2795                                    u16 bit, int *res)
2796 {
2797         struct ocfs2_dinode *alloc_di;
2798         struct ocfs2_group_desc *group;
2799         struct buffer_head *group_bh = NULL;
2800         u64 bg_blkno;
2801         int status;
2802
2803         trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2804                                       (unsigned int)bit);
2805
2806         alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2807         if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2808                 mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2809                      (unsigned int)bit,
2810                      ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2811                 status = -EINVAL;
2812                 goto bail;
2813         }
2814
2815         bg_blkno = group_blkno ? group_blkno :
2816                    ocfs2_which_suballoc_group(blkno, bit);
2817         status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2818                                              &group_bh);
2819         if (status < 0) {
2820                 mlog(ML_ERROR, "read group %llu failed %d\n",
2821                      (unsigned long long)bg_blkno, status);
2822                 goto bail;
2823         }
2824
2825         group = (struct ocfs2_group_desc *) group_bh->b_data;
2826         *res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2827
2828 bail:
2829         brelse(group_bh);
2830
2831         if (status)
2832                 mlog_errno(status);
2833         return status;
2834 }
2835
2836 /*
2837  * Test if the bit representing this inode (blkno) is set in the
2838  * suballocator.
2839  *
2840  * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2841  *
2842  * In the event of failure, a negative value is returned and *res is
2843  * meaningless.
2844  *
2845  * Callers must make sure to hold nfs_sync_lock to prevent
2846  * ocfs2_delete_inode() on another node from accessing the same
2847  * suballocator concurrently.
2848  */
2849 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2850 {
2851         int status;
2852         u64 group_blkno = 0;
2853         u16 suballoc_bit = 0, suballoc_slot = 0;
2854         struct inode *inode_alloc_inode;
2855         struct buffer_head *alloc_bh = NULL;
2856
2857         trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2858
2859         status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2860                                              &group_blkno, &suballoc_bit);
2861         if (status < 0) {
2862                 mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2863                 goto bail;
2864         }
2865
2866         inode_alloc_inode =
2867                 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2868                                             suballoc_slot);
2869         if (!inode_alloc_inode) {
2870                 /* the error code could be inaccurate, but we are not able to
2871                  * get the correct one. */
2872                 status = -EINVAL;
2873                 mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2874                      (u32)suballoc_slot);
2875                 goto bail;
2876         }
2877
2878         mutex_lock(&inode_alloc_inode->i_mutex);
2879         status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2880         if (status < 0) {
2881                 mutex_unlock(&inode_alloc_inode->i_mutex);
2882                 iput(inode_alloc_inode);
2883                 mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2884                      (u32)suballoc_slot, status);
2885                 goto bail;
2886         }
2887
2888         status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2889                                          group_blkno, blkno, suballoc_bit, res);
2890         if (status < 0)
2891                 mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2892
2893         ocfs2_inode_unlock(inode_alloc_inode, 0);
2894         mutex_unlock(&inode_alloc_inode->i_mutex);
2895
2896         iput(inode_alloc_inode);
2897         brelse(alloc_bh);
2898 bail:
2899         if (status)
2900                 mlog_errno(status);
2901         return status;
2902 }