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