]> nv-tegra.nvidia Code Review - linux-2.6.git/blob - fs/ocfs2/suballoc.c
Merge branch 'rc-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/mmarek/kbuild
[linux-2.6.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 = -EIO;
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_allow_chain_relink = 0;
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 = -EIO;
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 fe_ptr, 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         fe_ptr = le64_to_cpu(fe->id2.i_chain.cl_recs[chain].c_blkno);
1441         bg_ptr = le64_to_cpu(bg->bg_next_group);
1442         prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1443
1444         status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1445                                          prev_bg_bh,
1446                                          OCFS2_JOURNAL_ACCESS_WRITE);
1447         if (status < 0) {
1448                 mlog_errno(status);
1449                 goto out_rollback;
1450         }
1451
1452         prev_bg->bg_next_group = bg->bg_next_group;
1453         ocfs2_journal_dirty(handle, prev_bg_bh);
1454
1455         status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1456                                          bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1457         if (status < 0) {
1458                 mlog_errno(status);
1459                 goto out_rollback;
1460         }
1461
1462         bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1463         ocfs2_journal_dirty(handle, bg_bh);
1464
1465         status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1466                                          fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1467         if (status < 0) {
1468                 mlog_errno(status);
1469                 goto out_rollback;
1470         }
1471
1472         fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1473         ocfs2_journal_dirty(handle, fe_bh);
1474
1475 out_rollback:
1476         if (status < 0) {
1477                 fe->id2.i_chain.cl_recs[chain].c_blkno = cpu_to_le64(fe_ptr);
1478                 bg->bg_next_group = cpu_to_le64(bg_ptr);
1479                 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1480         }
1481
1482         if (status)
1483                 mlog_errno(status);
1484         return status;
1485 }
1486
1487 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1488                                                      u32 wanted)
1489 {
1490         return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1491 }
1492
1493 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1494  * value on error. */
1495 static int ocfs2_cluster_group_search(struct inode *inode,
1496                                       struct buffer_head *group_bh,
1497                                       u32 bits_wanted, u32 min_bits,
1498                                       u64 max_block,
1499                                       struct ocfs2_suballoc_result *res)
1500 {
1501         int search = -ENOSPC;
1502         int ret;
1503         u64 blkoff;
1504         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1505         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1506         unsigned int max_bits, gd_cluster_off;
1507
1508         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1509
1510         if (gd->bg_free_bits_count) {
1511                 max_bits = le16_to_cpu(gd->bg_bits);
1512
1513                 /* Tail groups in cluster bitmaps which aren't cpg
1514                  * aligned are prone to partial extension by a failed
1515                  * fs resize. If the file system resize never got to
1516                  * update the dinode cluster count, then we don't want
1517                  * to trust any clusters past it, regardless of what
1518                  * the group descriptor says. */
1519                 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1520                                                           le64_to_cpu(gd->bg_blkno));
1521                 if ((gd_cluster_off + max_bits) >
1522                     OCFS2_I(inode)->ip_clusters) {
1523                         max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1524                         trace_ocfs2_cluster_group_search_wrong_max_bits(
1525                                 (unsigned long long)le64_to_cpu(gd->bg_blkno),
1526                                 le16_to_cpu(gd->bg_bits),
1527                                 OCFS2_I(inode)->ip_clusters, max_bits);
1528                 }
1529
1530                 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1531                                                         group_bh, bits_wanted,
1532                                                         max_bits, res);
1533                 if (ret)
1534                         return ret;
1535
1536                 if (max_block) {
1537                         blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1538                                                           gd_cluster_off +
1539                                                           res->sr_bit_offset +
1540                                                           res->sr_bits);
1541                         trace_ocfs2_cluster_group_search_max_block(
1542                                 (unsigned long long)blkoff,
1543                                 (unsigned long long)max_block);
1544                         if (blkoff > max_block)
1545                                 return -ENOSPC;
1546                 }
1547
1548                 /* ocfs2_block_group_find_clear_bits() might
1549                  * return success, but we still want to return
1550                  * -ENOSPC unless it found the minimum number
1551                  * of bits. */
1552                 if (min_bits <= res->sr_bits)
1553                         search = 0; /* success */
1554                 else if (res->sr_bits) {
1555                         /*
1556                          * Don't show bits which we'll be returning
1557                          * for allocation to the local alloc bitmap.
1558                          */
1559                         ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
1560                 }
1561         }
1562
1563         return search;
1564 }
1565
1566 static int ocfs2_block_group_search(struct inode *inode,
1567                                     struct buffer_head *group_bh,
1568                                     u32 bits_wanted, u32 min_bits,
1569                                     u64 max_block,
1570                                     struct ocfs2_suballoc_result *res)
1571 {
1572         int ret = -ENOSPC;
1573         u64 blkoff;
1574         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1575
1576         BUG_ON(min_bits != 1);
1577         BUG_ON(ocfs2_is_cluster_bitmap(inode));
1578
1579         if (bg->bg_free_bits_count) {
1580                 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1581                                                         group_bh, bits_wanted,
1582                                                         le16_to_cpu(bg->bg_bits),
1583                                                         res);
1584                 if (!ret && max_block) {
1585                         blkoff = le64_to_cpu(bg->bg_blkno) +
1586                                 res->sr_bit_offset + res->sr_bits;
1587                         trace_ocfs2_block_group_search_max_block(
1588                                 (unsigned long long)blkoff,
1589                                 (unsigned long long)max_block);
1590                         if (blkoff > max_block)
1591                                 ret = -ENOSPC;
1592                 }
1593         }
1594
1595         return ret;
1596 }
1597
1598 static int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1599                                        handle_t *handle,
1600                                        struct buffer_head *di_bh,
1601                                        u32 num_bits,
1602                                        u16 chain)
1603 {
1604         int ret;
1605         u32 tmp_used;
1606         struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1607         struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1608
1609         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1610                                       OCFS2_JOURNAL_ACCESS_WRITE);
1611         if (ret < 0) {
1612                 mlog_errno(ret);
1613                 goto out;
1614         }
1615
1616         tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1617         di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1618         le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1619         ocfs2_journal_dirty(handle, di_bh);
1620
1621 out:
1622         return ret;
1623 }
1624
1625 static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
1626                                          struct ocfs2_extent_rec *rec,
1627                                          struct ocfs2_chain_list *cl)
1628 {
1629         unsigned int bpc = le16_to_cpu(cl->cl_bpc);
1630         unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
1631         unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
1632
1633         if (res->sr_bit_offset < bitoff)
1634                 return 0;
1635         if (res->sr_bit_offset >= (bitoff + bitcount))
1636                 return 0;
1637         res->sr_blkno = le64_to_cpu(rec->e_blkno) +
1638                 (res->sr_bit_offset - bitoff);
1639         if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
1640                 res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
1641         return 1;
1642 }
1643
1644 static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
1645                                           struct ocfs2_group_desc *bg,
1646                                           struct ocfs2_suballoc_result *res)
1647 {
1648         int i;
1649         u64 bg_blkno = res->sr_bg_blkno;  /* Save off */
1650         struct ocfs2_extent_rec *rec;
1651         struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1652         struct ocfs2_chain_list *cl = &di->id2.i_chain;
1653
1654         if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
1655                 res->sr_blkno = 0;
1656                 return;
1657         }
1658
1659         res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
1660         res->sr_bg_blkno = 0;  /* Clear it for contig block groups */
1661         if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
1662             !bg->bg_list.l_next_free_rec)
1663                 return;
1664
1665         for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
1666                 rec = &bg->bg_list.l_recs[i];
1667                 if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
1668                         res->sr_bg_blkno = bg_blkno;  /* Restore */
1669                         break;
1670                 }
1671         }
1672 }
1673
1674 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1675                                   handle_t *handle,
1676                                   u32 bits_wanted,
1677                                   u32 min_bits,
1678                                   struct ocfs2_suballoc_result *res,
1679                                   u16 *bits_left)
1680 {
1681         int ret;
1682         struct buffer_head *group_bh = NULL;
1683         struct ocfs2_group_desc *gd;
1684         struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1685         struct inode *alloc_inode = ac->ac_inode;
1686
1687         ret = ocfs2_read_group_descriptor(alloc_inode, di,
1688                                           res->sr_bg_blkno, &group_bh);
1689         if (ret < 0) {
1690                 mlog_errno(ret);
1691                 return ret;
1692         }
1693
1694         gd = (struct ocfs2_group_desc *) group_bh->b_data;
1695         ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1696                                   ac->ac_max_block, res);
1697         if (ret < 0) {
1698                 if (ret != -ENOSPC)
1699                         mlog_errno(ret);
1700                 goto out;
1701         }
1702
1703         if (!ret)
1704                 ocfs2_bg_discontig_fix_result(ac, gd, res);
1705
1706         /*
1707          * sr_bg_blkno might have been changed by
1708          * ocfs2_bg_discontig_fix_result
1709          */
1710         res->sr_bg_stable_blkno = group_bh->b_blocknr;
1711
1712         if (ac->ac_find_loc_only)
1713                 goto out_loc_only;
1714
1715         ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1716                                                res->sr_bits,
1717                                                le16_to_cpu(gd->bg_chain));
1718         if (ret < 0) {
1719                 mlog_errno(ret);
1720                 goto out;
1721         }
1722
1723         ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1724                                          res->sr_bit_offset, res->sr_bits);
1725         if (ret < 0)
1726                 mlog_errno(ret);
1727
1728 out_loc_only:
1729         *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1730
1731 out:
1732         brelse(group_bh);
1733
1734         return ret;
1735 }
1736
1737 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1738                               handle_t *handle,
1739                               u32 bits_wanted,
1740                               u32 min_bits,
1741                               struct ocfs2_suballoc_result *res,
1742                               u16 *bits_left)
1743 {
1744         int status;
1745         u16 chain;
1746         u64 next_group;
1747         struct inode *alloc_inode = ac->ac_inode;
1748         struct buffer_head *group_bh = NULL;
1749         struct buffer_head *prev_group_bh = NULL;
1750         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1751         struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1752         struct ocfs2_group_desc *bg;
1753
1754         chain = ac->ac_chain;
1755         trace_ocfs2_search_chain_begin(
1756                 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
1757                 bits_wanted, chain);
1758
1759         status = ocfs2_read_group_descriptor(alloc_inode, fe,
1760                                              le64_to_cpu(cl->cl_recs[chain].c_blkno),
1761                                              &group_bh);
1762         if (status < 0) {
1763                 mlog_errno(status);
1764                 goto bail;
1765         }
1766         bg = (struct ocfs2_group_desc *) group_bh->b_data;
1767
1768         status = -ENOSPC;
1769         /* for now, the chain search is a bit simplistic. We just use
1770          * the 1st group with any empty bits. */
1771         while ((status = ac->ac_group_search(alloc_inode, group_bh,
1772                                              bits_wanted, min_bits,
1773                                              ac->ac_max_block,
1774                                              res)) == -ENOSPC) {
1775                 if (!bg->bg_next_group)
1776                         break;
1777
1778                 brelse(prev_group_bh);
1779                 prev_group_bh = NULL;
1780
1781                 next_group = le64_to_cpu(bg->bg_next_group);
1782                 prev_group_bh = group_bh;
1783                 group_bh = NULL;
1784                 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1785                                                      next_group, &group_bh);
1786                 if (status < 0) {
1787                         mlog_errno(status);
1788                         goto bail;
1789                 }
1790                 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1791         }
1792         if (status < 0) {
1793                 if (status != -ENOSPC)
1794                         mlog_errno(status);
1795                 goto bail;
1796         }
1797
1798         trace_ocfs2_search_chain_succ(
1799                 (unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
1800
1801         res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
1802
1803         BUG_ON(res->sr_bits == 0);
1804         if (!status)
1805                 ocfs2_bg_discontig_fix_result(ac, bg, res);
1806
1807         /*
1808          * sr_bg_blkno might have been changed by
1809          * ocfs2_bg_discontig_fix_result
1810          */
1811         res->sr_bg_stable_blkno = group_bh->b_blocknr;
1812
1813         /*
1814          * Keep track of previous block descriptor read. When
1815          * we find a target, if we have read more than X
1816          * number of descriptors, and the target is reasonably
1817          * empty, relink him to top of his chain.
1818          *
1819          * We've read 0 extra blocks and only send one more to
1820          * the transaction, yet the next guy to search has a
1821          * much easier time.
1822          *
1823          * Do this *after* figuring out how many bits we're taking out
1824          * of our target group.
1825          */
1826         if (ac->ac_allow_chain_relink &&
1827             (prev_group_bh) &&
1828             (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
1829                 status = ocfs2_relink_block_group(handle, alloc_inode,
1830                                                   ac->ac_bh, group_bh,
1831                                                   prev_group_bh, chain);
1832                 if (status < 0) {
1833                         mlog_errno(status);
1834                         goto bail;
1835                 }
1836         }
1837
1838         if (ac->ac_find_loc_only)
1839                 goto out_loc_only;
1840
1841         status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
1842                                                   ac->ac_bh, res->sr_bits,
1843                                                   chain);
1844         if (status) {
1845                 mlog_errno(status);
1846                 goto bail;
1847         }
1848
1849         status = ocfs2_block_group_set_bits(handle,
1850                                             alloc_inode,
1851                                             bg,
1852                                             group_bh,
1853                                             res->sr_bit_offset,
1854                                             res->sr_bits);
1855         if (status < 0) {
1856                 mlog_errno(status);
1857                 goto bail;
1858         }
1859
1860         trace_ocfs2_search_chain_end(
1861                         (unsigned long long)le64_to_cpu(fe->i_blkno),
1862                         res->sr_bits);
1863
1864 out_loc_only:
1865         *bits_left = le16_to_cpu(bg->bg_free_bits_count);
1866 bail:
1867         brelse(group_bh);
1868         brelse(prev_group_bh);
1869
1870         if (status)
1871                 mlog_errno(status);
1872         return status;
1873 }
1874
1875 /* will give out up to bits_wanted contiguous bits. */
1876 static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
1877                                      handle_t *handle,
1878                                      u32 bits_wanted,
1879                                      u32 min_bits,
1880                                      struct ocfs2_suballoc_result *res)
1881 {
1882         int status;
1883         u16 victim, i;
1884         u16 bits_left = 0;
1885         u64 hint = ac->ac_last_group;
1886         struct ocfs2_chain_list *cl;
1887         struct ocfs2_dinode *fe;
1888
1889         BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1890         BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1891         BUG_ON(!ac->ac_bh);
1892
1893         fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1894
1895         /* The bh was validated by the inode read during
1896          * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1897         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1898
1899         if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1900             le32_to_cpu(fe->id1.bitmap1.i_total)) {
1901                 ocfs2_error(ac->ac_inode->i_sb,
1902                             "Chain allocator dinode %llu has %u used "
1903                             "bits but only %u total.",
1904                             (unsigned long long)le64_to_cpu(fe->i_blkno),
1905                             le32_to_cpu(fe->id1.bitmap1.i_used),
1906                             le32_to_cpu(fe->id1.bitmap1.i_total));
1907                 status = -EIO;
1908                 goto bail;
1909         }
1910
1911         res->sr_bg_blkno = hint;
1912         if (res->sr_bg_blkno) {
1913                 /* Attempt to short-circuit the usual search mechanism
1914                  * by jumping straight to the most recently used
1915                  * allocation group. This helps us maintain some
1916                  * contiguousness across allocations. */
1917                 status = ocfs2_search_one_group(ac, handle, bits_wanted,
1918                                                 min_bits, res, &bits_left);
1919                 if (!status)
1920                         goto set_hint;
1921                 if (status < 0 && status != -ENOSPC) {
1922                         mlog_errno(status);
1923                         goto bail;
1924                 }
1925         }
1926
1927         cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1928
1929         victim = ocfs2_find_victim_chain(cl);
1930         ac->ac_chain = victim;
1931         ac->ac_allow_chain_relink = 1;
1932
1933         status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1934                                     res, &bits_left);
1935         if (!status) {
1936                 hint = ocfs2_group_from_res(res);
1937                 goto set_hint;
1938         }
1939         if (status < 0 && status != -ENOSPC) {
1940                 mlog_errno(status);
1941                 goto bail;
1942         }
1943
1944         trace_ocfs2_claim_suballoc_bits(victim);
1945
1946         /* If we didn't pick a good victim, then just default to
1947          * searching each chain in order. Don't allow chain relinking
1948          * because we only calculate enough journal credits for one
1949          * relink per alloc. */
1950         ac->ac_allow_chain_relink = 0;
1951         for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1952                 if (i == victim)
1953                         continue;
1954                 if (!cl->cl_recs[i].c_free)
1955                         continue;
1956
1957                 ac->ac_chain = i;
1958                 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1959                                             res, &bits_left);
1960                 if (!status) {
1961                         hint = ocfs2_group_from_res(res);
1962                         break;
1963                 }
1964                 if (status < 0 && status != -ENOSPC) {
1965                         mlog_errno(status);
1966                         goto bail;
1967                 }
1968         }
1969
1970 set_hint:
1971         if (status != -ENOSPC) {
1972                 /* If the next search of this group is not likely to
1973                  * yield a suitable extent, then we reset the last
1974                  * group hint so as to not waste a disk read */
1975                 if (bits_left < min_bits)
1976                         ac->ac_last_group = 0;
1977                 else
1978                         ac->ac_last_group = hint;
1979         }
1980
1981 bail:
1982         if (status)
1983                 mlog_errno(status);
1984         return status;
1985 }
1986
1987 int ocfs2_claim_metadata(handle_t *handle,
1988                          struct ocfs2_alloc_context *ac,
1989                          u32 bits_wanted,
1990                          u64 *suballoc_loc,
1991                          u16 *suballoc_bit_start,
1992                          unsigned int *num_bits,
1993                          u64 *blkno_start)
1994 {
1995         int status;
1996         struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
1997
1998         BUG_ON(!ac);
1999         BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
2000         BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
2001
2002         status = ocfs2_claim_suballoc_bits(ac,
2003                                            handle,
2004                                            bits_wanted,
2005                                            1,
2006                                            &res);
2007         if (status < 0) {
2008                 mlog_errno(status);
2009                 goto bail;
2010         }
2011         atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2012
2013         *suballoc_loc = res.sr_bg_blkno;
2014         *suballoc_bit_start = res.sr_bit_offset;
2015         *blkno_start = res.sr_blkno;
2016         ac->ac_bits_given += res.sr_bits;
2017         *num_bits = res.sr_bits;
2018         status = 0;
2019 bail:
2020         if (status)
2021                 mlog_errno(status);
2022         return status;
2023 }
2024
2025 static void ocfs2_init_inode_ac_group(struct inode *dir,
2026                                       struct buffer_head *parent_di_bh,
2027                                       struct ocfs2_alloc_context *ac)
2028 {
2029         struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
2030         /*
2031          * Try to allocate inodes from some specific group.
2032          *
2033          * If the parent dir has recorded the last group used in allocation,
2034          * cool, use it. Otherwise if we try to allocate new inode from the
2035          * same slot the parent dir belongs to, use the same chunk.
2036          *
2037          * We are very careful here to avoid the mistake of setting
2038          * ac_last_group to a group descriptor from a different (unlocked) slot.
2039          */
2040         if (OCFS2_I(dir)->ip_last_used_group &&
2041             OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
2042                 ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
2043         else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
2044                 if (di->i_suballoc_loc)
2045                         ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
2046                 else
2047                         ac->ac_last_group = ocfs2_which_suballoc_group(
2048                                         le64_to_cpu(di->i_blkno),
2049                                         le16_to_cpu(di->i_suballoc_bit));
2050         }
2051 }
2052
2053 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
2054                                              struct ocfs2_alloc_context *ac)
2055 {
2056         OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
2057         OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
2058 }
2059
2060 int ocfs2_find_new_inode_loc(struct inode *dir,
2061                              struct buffer_head *parent_fe_bh,
2062                              struct ocfs2_alloc_context *ac,
2063                              u64 *fe_blkno)
2064 {
2065         int ret;
2066         handle_t *handle = NULL;
2067         struct ocfs2_suballoc_result *res;
2068
2069         BUG_ON(!ac);
2070         BUG_ON(ac->ac_bits_given != 0);
2071         BUG_ON(ac->ac_bits_wanted != 1);
2072         BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2073
2074         res = kzalloc(sizeof(*res), GFP_NOFS);
2075         if (res == NULL) {
2076                 ret = -ENOMEM;
2077                 mlog_errno(ret);
2078                 goto out;
2079         }
2080
2081         ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2082
2083         /*
2084          * The handle started here is for chain relink. Alternatively,
2085          * we could just disable relink for these calls.
2086          */
2087         handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
2088         if (IS_ERR(handle)) {
2089                 ret = PTR_ERR(handle);
2090                 handle = NULL;
2091                 mlog_errno(ret);
2092                 goto out;
2093         }
2094
2095         /*
2096          * This will instruct ocfs2_claim_suballoc_bits and
2097          * ocfs2_search_one_group to search but save actual allocation
2098          * for later.
2099          */
2100         ac->ac_find_loc_only = 1;
2101
2102         ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
2103         if (ret < 0) {
2104                 mlog_errno(ret);
2105                 goto out;
2106         }
2107
2108         ac->ac_find_loc_priv = res;
2109         *fe_blkno = res->sr_blkno;
2110
2111 out:
2112         if (handle)
2113                 ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
2114
2115         if (ret)
2116                 kfree(res);
2117
2118         return ret;
2119 }
2120
2121 int ocfs2_claim_new_inode_at_loc(handle_t *handle,
2122                                  struct inode *dir,
2123                                  struct ocfs2_alloc_context *ac,
2124                                  u64 *suballoc_loc,
2125                                  u16 *suballoc_bit,
2126                                  u64 di_blkno)
2127 {
2128         int ret;
2129         u16 chain;
2130         struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
2131         struct buffer_head *bg_bh = NULL;
2132         struct ocfs2_group_desc *bg;
2133         struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
2134
2135         /*
2136          * Since di_blkno is being passed back in, we check for any
2137          * inconsistencies which may have happened between
2138          * calls. These are code bugs as di_blkno is not expected to
2139          * change once returned from ocfs2_find_new_inode_loc()
2140          */
2141         BUG_ON(res->sr_blkno != di_blkno);
2142
2143         ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
2144                                           res->sr_bg_stable_blkno, &bg_bh);
2145         if (ret) {
2146                 mlog_errno(ret);
2147                 goto out;
2148         }
2149
2150         bg = (struct ocfs2_group_desc *) bg_bh->b_data;
2151         chain = le16_to_cpu(bg->bg_chain);
2152
2153         ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
2154                                                ac->ac_bh, res->sr_bits,
2155                                                chain);
2156         if (ret) {
2157                 mlog_errno(ret);
2158                 goto out;
2159         }
2160
2161         ret = ocfs2_block_group_set_bits(handle,
2162                                          ac->ac_inode,
2163                                          bg,
2164                                          bg_bh,
2165                                          res->sr_bit_offset,
2166                                          res->sr_bits);
2167         if (ret < 0) {
2168                 mlog_errno(ret);
2169                 goto out;
2170         }
2171
2172         trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
2173                                            res->sr_bits);
2174
2175         atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2176
2177         BUG_ON(res->sr_bits != 1);
2178
2179         *suballoc_loc = res->sr_bg_blkno;
2180         *suballoc_bit = res->sr_bit_offset;
2181         ac->ac_bits_given++;
2182         ocfs2_save_inode_ac_group(dir, ac);
2183
2184 out:
2185         brelse(bg_bh);
2186
2187         return ret;
2188 }
2189
2190 int ocfs2_claim_new_inode(handle_t *handle,
2191                           struct inode *dir,
2192                           struct buffer_head *parent_fe_bh,
2193                           struct ocfs2_alloc_context *ac,
2194                           u64 *suballoc_loc,
2195                           u16 *suballoc_bit,
2196                           u64 *fe_blkno)
2197 {
2198         int status;
2199         struct ocfs2_suballoc_result res;
2200
2201         BUG_ON(!ac);
2202         BUG_ON(ac->ac_bits_given != 0);
2203         BUG_ON(ac->ac_bits_wanted != 1);
2204         BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
2205
2206         ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
2207
2208         status = ocfs2_claim_suballoc_bits(ac,
2209                                            handle,
2210                                            1,
2211                                            1,
2212                                            &res);
2213         if (status < 0) {
2214                 mlog_errno(status);
2215                 goto bail;
2216         }
2217         atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
2218
2219         BUG_ON(res.sr_bits != 1);
2220
2221         *suballoc_loc = res.sr_bg_blkno;
2222         *suballoc_bit = res.sr_bit_offset;
2223         *fe_blkno = res.sr_blkno;
2224         ac->ac_bits_given++;
2225         ocfs2_save_inode_ac_group(dir, ac);
2226         status = 0;
2227 bail:
2228         if (status)
2229                 mlog_errno(status);
2230         return status;
2231 }
2232
2233 /* translate a group desc. blkno and it's bitmap offset into
2234  * disk cluster offset. */
2235 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
2236                                                    u64 bg_blkno,
2237                                                    u16 bg_bit_off)
2238 {
2239         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2240         u32 cluster = 0;
2241
2242         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2243
2244         if (bg_blkno != osb->first_cluster_group_blkno)
2245                 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
2246         cluster += (u32) bg_bit_off;
2247         return cluster;
2248 }
2249
2250 /* given a cluster offset, calculate which block group it belongs to
2251  * and return that block offset. */
2252 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
2253 {
2254         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2255         u32 group_no;
2256
2257         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2258
2259         group_no = cluster / osb->bitmap_cpg;
2260         if (!group_no)
2261                 return osb->first_cluster_group_blkno;
2262         return ocfs2_clusters_to_blocks(inode->i_sb,
2263                                         group_no * osb->bitmap_cpg);
2264 }
2265
2266 /* given the block number of a cluster start, calculate which cluster
2267  * group and descriptor bitmap offset that corresponds to. */
2268 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
2269                                                 u64 data_blkno,
2270                                                 u64 *bg_blkno,
2271                                                 u16 *bg_bit_off)
2272 {
2273         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2274         u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
2275
2276         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
2277
2278         *bg_blkno = ocfs2_which_cluster_group(inode,
2279                                               data_cluster);
2280
2281         if (*bg_blkno == osb->first_cluster_group_blkno)
2282                 *bg_bit_off = (u16) data_cluster;
2283         else
2284                 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
2285                                                              data_blkno - *bg_blkno);
2286 }
2287
2288 /*
2289  * min_bits - minimum contiguous chunk from this total allocation we
2290  * can handle. set to what we asked for originally for a full
2291  * contig. allocation, set to '1' to indicate we can deal with extents
2292  * of any size.
2293  */
2294 int __ocfs2_claim_clusters(handle_t *handle,
2295                            struct ocfs2_alloc_context *ac,
2296                            u32 min_clusters,
2297                            u32 max_clusters,
2298                            u32 *cluster_start,
2299                            u32 *num_clusters)
2300 {
2301         int status;
2302         unsigned int bits_wanted = max_clusters;
2303         struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
2304         struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
2305
2306         BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
2307
2308         BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
2309                && ac->ac_which != OCFS2_AC_USE_MAIN);
2310
2311         if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
2312                 WARN_ON(min_clusters > 1);
2313
2314                 status = ocfs2_claim_local_alloc_bits(osb,
2315                                                       handle,
2316                                                       ac,
2317                                                       bits_wanted,
2318                                                       cluster_start,
2319                                                       num_clusters);
2320                 if (!status)
2321                         atomic_inc(&osb->alloc_stats.local_data);
2322         } else {
2323                 if (min_clusters > (osb->bitmap_cpg - 1)) {
2324                         /* The only paths asking for contiguousness
2325                          * should know about this already. */
2326                         mlog(ML_ERROR, "minimum allocation requested %u exceeds "
2327                              "group bitmap size %u!\n", min_clusters,
2328                              osb->bitmap_cpg);
2329                         status = -ENOSPC;
2330                         goto bail;
2331                 }
2332                 /* clamp the current request down to a realistic size. */
2333                 if (bits_wanted > (osb->bitmap_cpg - 1))
2334                         bits_wanted = osb->bitmap_cpg - 1;
2335
2336                 status = ocfs2_claim_suballoc_bits(ac,
2337                                                    handle,
2338                                                    bits_wanted,
2339                                                    min_clusters,
2340                                                    &res);
2341                 if (!status) {
2342                         BUG_ON(res.sr_blkno); /* cluster alloc can't set */
2343                         *cluster_start =
2344                                 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
2345                                                                  res.sr_bg_blkno,
2346                                                                  res.sr_bit_offset);
2347                         atomic_inc(&osb->alloc_stats.bitmap_data);
2348                         *num_clusters = res.sr_bits;
2349                 }
2350         }
2351         if (status < 0) {
2352                 if (status != -ENOSPC)
2353                         mlog_errno(status);
2354                 goto bail;
2355         }
2356
2357         ac->ac_bits_given += *num_clusters;
2358
2359 bail:
2360         if (status)
2361                 mlog_errno(status);
2362         return status;
2363 }
2364
2365 int ocfs2_claim_clusters(handle_t *handle,
2366                          struct ocfs2_alloc_context *ac,
2367                          u32 min_clusters,
2368                          u32 *cluster_start,
2369                          u32 *num_clusters)
2370 {
2371         unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
2372
2373         return __ocfs2_claim_clusters(handle, ac, min_clusters,
2374                                       bits_wanted, cluster_start, num_clusters);
2375 }
2376
2377 static int ocfs2_block_group_clear_bits(handle_t *handle,
2378                                         struct inode *alloc_inode,
2379                                         struct ocfs2_group_desc *bg,
2380                                         struct buffer_head *group_bh,
2381                                         unsigned int bit_off,
2382                                         unsigned int num_bits,
2383                                         void (*undo_fn)(unsigned int bit,
2384                                                         unsigned long *bmap))
2385 {
2386         int status;
2387         unsigned int tmp;
2388         struct ocfs2_group_desc *undo_bg = NULL;
2389
2390         /* The caller got this descriptor from
2391          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
2392         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
2393
2394         trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
2395
2396         BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
2397         status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
2398                                          group_bh,
2399                                          undo_fn ?
2400                                          OCFS2_JOURNAL_ACCESS_UNDO :
2401                                          OCFS2_JOURNAL_ACCESS_WRITE);
2402         if (status < 0) {
2403                 mlog_errno(status);
2404                 goto bail;
2405         }
2406
2407         if (undo_fn) {
2408                 jbd_lock_bh_state(group_bh);
2409                 undo_bg = (struct ocfs2_group_desc *)
2410                                         bh2jh(group_bh)->b_committed_data;
2411                 BUG_ON(!undo_bg);
2412         }
2413
2414         tmp = num_bits;
2415         while(tmp--) {
2416                 ocfs2_clear_bit((bit_off + tmp),
2417                                 (unsigned long *) bg->bg_bitmap);
2418                 if (undo_fn)
2419                         undo_fn(bit_off + tmp,
2420                                 (unsigned long *) undo_bg->bg_bitmap);
2421         }
2422         le16_add_cpu(&bg->bg_free_bits_count, num_bits);
2423         if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
2424                 ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
2425                             " count %u but claims %u are freed. num_bits %d",
2426                             (unsigned long long)le64_to_cpu(bg->bg_blkno),
2427                             le16_to_cpu(bg->bg_bits),
2428                             le16_to_cpu(bg->bg_free_bits_count), num_bits);
2429                 return -EROFS;
2430         }
2431
2432         if (undo_fn)
2433                 jbd_unlock_bh_state(group_bh);
2434
2435         ocfs2_journal_dirty(handle, group_bh);
2436 bail:
2437         return status;
2438 }
2439
2440 /*
2441  * expects the suballoc inode to already be locked.
2442  */
2443 static int _ocfs2_free_suballoc_bits(handle_t *handle,
2444                                      struct inode *alloc_inode,
2445                                      struct buffer_head *alloc_bh,
2446                                      unsigned int start_bit,
2447                                      u64 bg_blkno,
2448                                      unsigned int count,
2449                                      void (*undo_fn)(unsigned int bit,
2450                                                      unsigned long *bitmap))
2451 {
2452         int status = 0;
2453         u32 tmp_used;
2454         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
2455         struct ocfs2_chain_list *cl = &fe->id2.i_chain;
2456         struct buffer_head *group_bh = NULL;
2457         struct ocfs2_group_desc *group;
2458
2459         /* The alloc_bh comes from ocfs2_free_dinode() or
2460          * ocfs2_free_clusters().  The callers have all locked the
2461          * allocator and gotten alloc_bh from the lock call.  This
2462          * validates the dinode buffer.  Any corruption that has happened
2463          * is a code bug. */
2464         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
2465         BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
2466
2467         trace_ocfs2_free_suballoc_bits(
2468                 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
2469                 (unsigned long long)bg_blkno,
2470                 start_bit, count);
2471
2472         status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
2473                                              &group_bh);
2474         if (status < 0) {
2475                 mlog_errno(status);
2476                 goto bail;
2477         }
2478         group = (struct ocfs2_group_desc *) group_bh->b_data;
2479
2480         BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
2481
2482         status = ocfs2_block_group_clear_bits(handle, alloc_inode,
2483                                               group, group_bh,
2484                                               start_bit, count, undo_fn);
2485         if (status < 0) {
2486                 mlog_errno(status);
2487                 goto bail;
2488         }
2489
2490         status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
2491                                          alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2492         if (status < 0) {
2493                 mlog_errno(status);
2494                 goto bail;
2495         }
2496
2497         le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2498                      count);
2499         tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2500         fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2501         ocfs2_journal_dirty(handle, alloc_bh);
2502
2503 bail:
2504         brelse(group_bh);
2505
2506         if (status)
2507                 mlog_errno(status);
2508         return status;
2509 }
2510
2511 int ocfs2_free_suballoc_bits(handle_t *handle,
2512                              struct inode *alloc_inode,
2513                              struct buffer_head *alloc_bh,
2514                              unsigned int start_bit,
2515                              u64 bg_blkno,
2516                              unsigned int count)
2517 {
2518         return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
2519                                          start_bit, bg_blkno, count, NULL);
2520 }
2521
2522 int ocfs2_free_dinode(handle_t *handle,
2523                       struct inode *inode_alloc_inode,
2524                       struct buffer_head *inode_alloc_bh,
2525                       struct ocfs2_dinode *di)
2526 {
2527         u64 blk = le64_to_cpu(di->i_blkno);
2528         u16 bit = le16_to_cpu(di->i_suballoc_bit);
2529         u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2530
2531         if (di->i_suballoc_loc)
2532                 bg_blkno = le64_to_cpu(di->i_suballoc_loc);
2533         return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2534                                         inode_alloc_bh, bit, bg_blkno, 1);
2535 }
2536
2537 static int _ocfs2_free_clusters(handle_t *handle,
2538                                 struct inode *bitmap_inode,
2539                                 struct buffer_head *bitmap_bh,
2540                                 u64 start_blk,
2541                                 unsigned int num_clusters,
2542                                 void (*undo_fn)(unsigned int bit,
2543                                                 unsigned long *bitmap))
2544 {
2545         int status;
2546         u16 bg_start_bit;
2547         u64 bg_blkno;
2548         struct ocfs2_dinode *fe;
2549
2550         /* You can't ever have a contiguous set of clusters
2551          * bigger than a block group bitmap so we never have to worry
2552          * about looping on them.
2553          * This is expensive. We can safely remove once this stuff has
2554          * gotten tested really well. */
2555         BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
2556
2557         fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
2558
2559         ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2560                                      &bg_start_bit);
2561
2562         trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
2563                         (unsigned long long)start_blk,
2564                         bg_start_bit, num_clusters);
2565
2566         status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2567                                            bg_start_bit, bg_blkno,
2568                                            num_clusters, undo_fn);
2569         if (status < 0) {
2570                 mlog_errno(status);
2571                 goto out;
2572         }
2573
2574         ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2575                                          num_clusters);
2576
2577 out:
2578         if (status)
2579                 mlog_errno(status);
2580         return status;
2581 }
2582
2583 int ocfs2_free_clusters(handle_t *handle,
2584                         struct inode *bitmap_inode,
2585                         struct buffer_head *bitmap_bh,
2586                         u64 start_blk,
2587                         unsigned int num_clusters)
2588 {
2589         return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2590                                     start_blk, num_clusters,
2591                                     _ocfs2_set_bit);
2592 }
2593
2594 /*
2595  * Give never-used clusters back to the global bitmap.  We don't need
2596  * to protect these bits in the undo buffer.
2597  */
2598 int ocfs2_release_clusters(handle_t *handle,
2599                            struct inode *bitmap_inode,
2600                            struct buffer_head *bitmap_bh,
2601                            u64 start_blk,
2602                            unsigned int num_clusters)
2603 {
2604         return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
2605                                     start_blk, num_clusters,
2606                                     _ocfs2_clear_bit);
2607 }
2608
2609 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
2610 {
2611         printk("Block Group:\n");
2612         printk("bg_signature:       %s\n", bg->bg_signature);
2613         printk("bg_size:            %u\n", bg->bg_size);
2614         printk("bg_bits:            %u\n", bg->bg_bits);
2615         printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
2616         printk("bg_chain:           %u\n", bg->bg_chain);
2617         printk("bg_generation:      %u\n", le32_to_cpu(bg->bg_generation));
2618         printk("bg_next_group:      %llu\n",
2619                (unsigned long long)bg->bg_next_group);
2620         printk("bg_parent_dinode:   %llu\n",
2621                (unsigned long long)bg->bg_parent_dinode);
2622         printk("bg_blkno:           %llu\n",
2623                (unsigned long long)bg->bg_blkno);
2624 }
2625
2626 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2627 {
2628         int i;
2629
2630         printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
2631         printk("i_signature:                  %s\n", fe->i_signature);
2632         printk("i_size:                       %llu\n",
2633                (unsigned long long)fe->i_size);
2634         printk("i_clusters:                   %u\n", fe->i_clusters);
2635         printk("i_generation:                 %u\n",
2636                le32_to_cpu(fe->i_generation));
2637         printk("id1.bitmap1.i_used:           %u\n",
2638                le32_to_cpu(fe->id1.bitmap1.i_used));
2639         printk("id1.bitmap1.i_total:          %u\n",
2640                le32_to_cpu(fe->id1.bitmap1.i_total));
2641         printk("id2.i_chain.cl_cpg:           %u\n", fe->id2.i_chain.cl_cpg);
2642         printk("id2.i_chain.cl_bpc:           %u\n", fe->id2.i_chain.cl_bpc);
2643         printk("id2.i_chain.cl_count:         %u\n", fe->id2.i_chain.cl_count);
2644         printk("id2.i_chain.cl_next_free_rec: %u\n",
2645                fe->id2.i_chain.cl_next_free_rec);
2646         for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2647                 printk("fe->id2.i_chain.cl_recs[%d].c_free:  %u\n", i,
2648                        fe->id2.i_chain.cl_recs[i].c_free);
2649                 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2650                        fe->id2.i_chain.cl_recs[i].c_total);
2651                 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2652                        (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
2653         }
2654 }
2655
2656 /*
2657  * For a given allocation, determine which allocators will need to be
2658  * accessed, and lock them, reserving the appropriate number of bits.
2659  *
2660  * Sparse file systems call this from ocfs2_write_begin_nolock()
2661  * and ocfs2_allocate_unwritten_extents().
2662  *
2663  * File systems which don't support holes call this from
2664  * ocfs2_extend_allocation().
2665  */
2666 int ocfs2_lock_allocators(struct inode *inode,
2667                           struct ocfs2_extent_tree *et,
2668                           u32 clusters_to_add, u32 extents_to_split,
2669                           struct ocfs2_alloc_context **data_ac,
2670                           struct ocfs2_alloc_context **meta_ac)
2671 {
2672         int ret = 0, num_free_extents;
2673         unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2674         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2675
2676         *meta_ac = NULL;
2677         if (data_ac)
2678                 *data_ac = NULL;
2679
2680         BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2681
2682         num_free_extents = ocfs2_num_free_extents(osb, et);
2683         if (num_free_extents < 0) {
2684                 ret = num_free_extents;
2685                 mlog_errno(ret);
2686                 goto out;
2687         }
2688
2689         /*
2690          * Sparse allocation file systems need to be more conservative
2691          * with reserving room for expansion - the actual allocation
2692          * happens while we've got a journal handle open so re-taking
2693          * a cluster lock (because we ran out of room for another
2694          * extent) will violate ordering rules.
2695          *
2696          * Most of the time we'll only be seeing this 1 cluster at a time
2697          * anyway.
2698          *
2699          * Always lock for any unwritten extents - we might want to
2700          * add blocks during a split.
2701          */
2702         if (!num_free_extents ||
2703             (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2704                 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2705                 if (ret < 0) {
2706                         if (ret != -ENOSPC)
2707                                 mlog_errno(ret);
2708                         goto out;
2709                 }
2710         }
2711
2712         if (clusters_to_add == 0)
2713                 goto out;
2714
2715         ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2716         if (ret < 0) {
2717                 if (ret != -ENOSPC)
2718                         mlog_errno(ret);
2719                 goto out;
2720         }
2721
2722 out:
2723         if (ret) {
2724                 if (*meta_ac) {
2725                         ocfs2_free_alloc_context(*meta_ac);
2726                         *meta_ac = NULL;
2727                 }
2728
2729                 /*
2730                  * We cannot have an error and a non null *data_ac.
2731                  */
2732         }
2733
2734         return ret;
2735 }
2736
2737 /*
2738  * Read the inode specified by blkno to get suballoc_slot and
2739  * suballoc_bit.
2740  */
2741 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2742                                        u16 *suballoc_slot, u64 *group_blkno,
2743                                        u16 *suballoc_bit)
2744 {
2745         int status;
2746         struct buffer_head *inode_bh = NULL;
2747         struct ocfs2_dinode *inode_fe;
2748
2749         trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
2750
2751         /* dirty read disk */
2752         status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2753         if (status < 0) {
2754                 mlog(ML_ERROR, "read block %llu failed %d\n",
2755                      (unsigned long long)blkno, status);
2756                 goto bail;
2757         }
2758
2759         inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2760         if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2761                 mlog(ML_ERROR, "invalid inode %llu requested\n",
2762                      (unsigned long long)blkno);
2763                 status = -EINVAL;
2764                 goto bail;
2765         }
2766
2767         if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2768             (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2769                 mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2770                      (unsigned long long)blkno,
2771                      (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2772                 status = -EINVAL;
2773                 goto bail;
2774         }
2775
2776         if (suballoc_slot)
2777                 *suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2778         if (suballoc_bit)
2779                 *suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2780         if (group_blkno)
2781                 *group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
2782
2783 bail:
2784         brelse(inode_bh);
2785
2786         if (status)
2787                 mlog_errno(status);
2788         return status;
2789 }
2790
2791 /*
2792  * test whether bit is SET in allocator bitmap or not.  on success, 0
2793  * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
2794  * is returned and *res is meaningless.  Call this after you have
2795  * cluster locked against suballoc, or you may get a result based on
2796  * non-up2date contents
2797  */
2798 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2799                                    struct inode *suballoc,
2800                                    struct buffer_head *alloc_bh,
2801                                    u64 group_blkno, u64 blkno,
2802                                    u16 bit, int *res)
2803 {
2804         struct ocfs2_dinode *alloc_di;
2805         struct ocfs2_group_desc *group;
2806         struct buffer_head *group_bh = NULL;
2807         u64 bg_blkno;
2808         int status;
2809
2810         trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
2811                                       (unsigned int)bit);
2812
2813         alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
2814         if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
2815                 mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2816                      (unsigned int)bit,
2817                      ocfs2_bits_per_group(&alloc_di->id2.i_chain));
2818                 status = -EINVAL;
2819                 goto bail;
2820         }
2821
2822         bg_blkno = group_blkno ? group_blkno :
2823                    ocfs2_which_suballoc_group(blkno, bit);
2824         status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
2825                                              &group_bh);
2826         if (status < 0) {
2827                 mlog(ML_ERROR, "read group %llu failed %d\n",
2828                      (unsigned long long)bg_blkno, status);
2829                 goto bail;
2830         }
2831
2832         group = (struct ocfs2_group_desc *) group_bh->b_data;
2833         *res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2834
2835 bail:
2836         brelse(group_bh);
2837
2838         if (status)
2839                 mlog_errno(status);
2840         return status;
2841 }
2842
2843 /*
2844  * Test if the bit representing this inode (blkno) is set in the
2845  * suballocator.
2846  *
2847  * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2848  *
2849  * In the event of failure, a negative value is returned and *res is
2850  * meaningless.
2851  *
2852  * Callers must make sure to hold nfs_sync_lock to prevent
2853  * ocfs2_delete_inode() on another node from accessing the same
2854  * suballocator concurrently.
2855  */
2856 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2857 {
2858         int status;
2859         u64 group_blkno = 0;
2860         u16 suballoc_bit = 0, suballoc_slot = 0;
2861         struct inode *inode_alloc_inode;
2862         struct buffer_head *alloc_bh = NULL;
2863
2864         trace_ocfs2_test_inode_bit((unsigned long long)blkno);
2865
2866         status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2867                                              &group_blkno, &suballoc_bit);
2868         if (status < 0) {
2869                 mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2870                 goto bail;
2871         }
2872
2873         inode_alloc_inode =
2874                 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2875                                             suballoc_slot);
2876         if (!inode_alloc_inode) {
2877                 /* the error code could be inaccurate, but we are not able to
2878                  * get the correct one. */
2879                 status = -EINVAL;
2880                 mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2881                      (u32)suballoc_slot);
2882                 goto bail;
2883         }
2884
2885         mutex_lock(&inode_alloc_inode->i_mutex);
2886         status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2887         if (status < 0) {
2888                 mutex_unlock(&inode_alloc_inode->i_mutex);
2889                 mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2890                      (u32)suballoc_slot, status);
2891                 goto bail;
2892         }
2893
2894         status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2895                                          group_blkno, blkno, suballoc_bit, res);
2896         if (status < 0)
2897                 mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2898
2899         ocfs2_inode_unlock(inode_alloc_inode, 0);
2900         mutex_unlock(&inode_alloc_inode->i_mutex);
2901
2902         iput(inode_alloc_inode);
2903         brelse(alloc_bh);
2904 bail:
2905         if (status)
2906                 mlog_errno(status);
2907         return status;
2908 }