drm/radeon: add new RS880 pci id
[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 #define MLOG_MASK_PREFIX ML_DISK_ALLOC
33 #include <cluster/masklog.h>
34
35 #include "ocfs2.h"
36
37 #include "alloc.h"
38 #include "blockcheck.h"
39 #include "dlmglue.h"
40 #include "inode.h"
41 #include "journal.h"
42 #include "localalloc.h"
43 #include "suballoc.h"
44 #include "super.h"
45 #include "sysfile.h"
46 #include "uptodate.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_INODES_TO_STEAL       1024
55
56 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg);
57 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe);
58 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
59 static int ocfs2_block_group_fill(handle_t *handle,
60                                   struct inode *alloc_inode,
61                                   struct buffer_head *bg_bh,
62                                   u64 group_blkno,
63                                   u16 my_chain,
64                                   struct ocfs2_chain_list *cl);
65 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
66                                    struct inode *alloc_inode,
67                                    struct buffer_head *bh,
68                                    u64 max_block,
69                                    u64 *last_alloc_group,
70                                    int flags);
71
72 static int ocfs2_cluster_group_search(struct inode *inode,
73                                       struct buffer_head *group_bh,
74                                       u32 bits_wanted, u32 min_bits,
75                                       u64 max_block,
76                                       u16 *bit_off, u16 *bits_found);
77 static int ocfs2_block_group_search(struct inode *inode,
78                                     struct buffer_head *group_bh,
79                                     u32 bits_wanted, u32 min_bits,
80                                     u64 max_block,
81                                     u16 *bit_off, u16 *bits_found);
82 static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
83                                      struct ocfs2_alloc_context *ac,
84                                      handle_t *handle,
85                                      u32 bits_wanted,
86                                      u32 min_bits,
87                                      u16 *bit_off,
88                                      unsigned int *num_bits,
89                                      u64 *bg_blkno);
90 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
91                                          int nr);
92 static inline int ocfs2_block_group_set_bits(handle_t *handle,
93                                              struct inode *alloc_inode,
94                                              struct ocfs2_group_desc *bg,
95                                              struct buffer_head *group_bh,
96                                              unsigned int bit_off,
97                                              unsigned int num_bits);
98 static inline int ocfs2_block_group_clear_bits(handle_t *handle,
99                                                struct inode *alloc_inode,
100                                                struct ocfs2_group_desc *bg,
101                                                struct buffer_head *group_bh,
102                                                unsigned int bit_off,
103                                                unsigned int num_bits);
104
105 static int ocfs2_relink_block_group(handle_t *handle,
106                                     struct inode *alloc_inode,
107                                     struct buffer_head *fe_bh,
108                                     struct buffer_head *bg_bh,
109                                     struct buffer_head *prev_bg_bh,
110                                     u16 chain);
111 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
112                                                      u32 wanted);
113 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
114                                                    u64 bg_blkno,
115                                                    u16 bg_bit_off);
116 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
117                                                 u64 data_blkno,
118                                                 u64 *bg_blkno,
119                                                 u16 *bg_bit_off);
120 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
121                                              u32 bits_wanted, u64 max_block,
122                                              int flags,
123                                              struct ocfs2_alloc_context **ac);
124
125 void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
126 {
127         struct inode *inode = ac->ac_inode;
128
129         if (inode) {
130                 if (ac->ac_which != OCFS2_AC_USE_LOCAL)
131                         ocfs2_inode_unlock(inode, 1);
132
133                 mutex_unlock(&inode->i_mutex);
134
135                 iput(inode);
136                 ac->ac_inode = NULL;
137         }
138         brelse(ac->ac_bh);
139         ac->ac_bh = NULL;
140 }
141
142 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
143 {
144         ocfs2_free_ac_resource(ac);
145         kfree(ac);
146 }
147
148 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
149 {
150         return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
151 }
152
153 #define do_error(fmt, ...)                                              \
154         do{                                                             \
155                 if (clean_error)                                        \
156                         mlog(ML_ERROR, fmt "\n", ##__VA_ARGS__);        \
157                 else                                                    \
158                         ocfs2_error(sb, fmt, ##__VA_ARGS__);            \
159         } while (0)
160
161 static int ocfs2_validate_gd_self(struct super_block *sb,
162                                   struct buffer_head *bh,
163                                   int clean_error)
164 {
165         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
166
167         if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
168                 do_error("Group descriptor #%llu has bad signature %.*s",
169                          (unsigned long long)bh->b_blocknr, 7,
170                          gd->bg_signature);
171                 return -EINVAL;
172         }
173
174         if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
175                 do_error("Group descriptor #%llu has an invalid bg_blkno "
176                          "of %llu",
177                          (unsigned long long)bh->b_blocknr,
178                          (unsigned long long)le64_to_cpu(gd->bg_blkno));
179                 return -EINVAL;
180         }
181
182         if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
183                 do_error("Group descriptor #%llu has an invalid "
184                          "fs_generation of #%u",
185                          (unsigned long long)bh->b_blocknr,
186                          le32_to_cpu(gd->bg_generation));
187                 return -EINVAL;
188         }
189
190         if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
191                 do_error("Group descriptor #%llu has bit count %u but "
192                          "claims that %u are free",
193                          (unsigned long long)bh->b_blocknr,
194                          le16_to_cpu(gd->bg_bits),
195                          le16_to_cpu(gd->bg_free_bits_count));
196                 return -EINVAL;
197         }
198
199         if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
200                 do_error("Group descriptor #%llu has bit count %u but "
201                          "max bitmap bits of %u",
202                          (unsigned long long)bh->b_blocknr,
203                          le16_to_cpu(gd->bg_bits),
204                          8 * le16_to_cpu(gd->bg_size));
205                 return -EINVAL;
206         }
207
208         return 0;
209 }
210
211 static int ocfs2_validate_gd_parent(struct super_block *sb,
212                                     struct ocfs2_dinode *di,
213                                     struct buffer_head *bh,
214                                     int clean_error)
215 {
216         unsigned int max_bits;
217         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
218
219         if (di->i_blkno != gd->bg_parent_dinode) {
220                 do_error("Group descriptor #%llu has bad parent "
221                          "pointer (%llu, expected %llu)",
222                          (unsigned long long)bh->b_blocknr,
223                          (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
224                          (unsigned long long)le64_to_cpu(di->i_blkno));
225                 return -EINVAL;
226         }
227
228         max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
229         if (le16_to_cpu(gd->bg_bits) > max_bits) {
230                 do_error("Group descriptor #%llu has bit count of %u",
231                          (unsigned long long)bh->b_blocknr,
232                          le16_to_cpu(gd->bg_bits));
233                 return -EINVAL;
234         }
235
236         if (le16_to_cpu(gd->bg_chain) >=
237             le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) {
238                 do_error("Group descriptor #%llu has bad chain %u",
239                          (unsigned long long)bh->b_blocknr,
240                          le16_to_cpu(gd->bg_chain));
241                 return -EINVAL;
242         }
243
244         return 0;
245 }
246
247 #undef do_error
248
249 /*
250  * This version only prints errors.  It does not fail the filesystem, and
251  * exists only for resize.
252  */
253 int ocfs2_check_group_descriptor(struct super_block *sb,
254                                  struct ocfs2_dinode *di,
255                                  struct buffer_head *bh)
256 {
257         int rc;
258         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
259
260         BUG_ON(!buffer_uptodate(bh));
261
262         /*
263          * If the ecc fails, we return the error but otherwise
264          * leave the filesystem running.  We know any error is
265          * local to this block.
266          */
267         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
268         if (rc) {
269                 mlog(ML_ERROR,
270                      "Checksum failed for group descriptor %llu\n",
271                      (unsigned long long)bh->b_blocknr);
272         } else
273                 rc = ocfs2_validate_gd_self(sb, bh, 1);
274         if (!rc)
275                 rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
276
277         return rc;
278 }
279
280 static int ocfs2_validate_group_descriptor(struct super_block *sb,
281                                            struct buffer_head *bh)
282 {
283         int rc;
284         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
285
286         mlog(0, "Validating group descriptor %llu\n",
287              (unsigned long long)bh->b_blocknr);
288
289         BUG_ON(!buffer_uptodate(bh));
290
291         /*
292          * If the ecc fails, we return the error but otherwise
293          * leave the filesystem running.  We know any error is
294          * local to this block.
295          */
296         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
297         if (rc)
298                 return rc;
299
300         /*
301          * Errors after here are fatal.
302          */
303
304         return ocfs2_validate_gd_self(sb, bh, 0);
305 }
306
307 int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
308                                 u64 gd_blkno, struct buffer_head **bh)
309 {
310         int rc;
311         struct buffer_head *tmp = *bh;
312
313         rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
314                               ocfs2_validate_group_descriptor);
315         if (rc)
316                 goto out;
317
318         rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
319         if (rc) {
320                 brelse(tmp);
321                 goto out;
322         }
323
324         /* If ocfs2_read_block() got us a new bh, pass it up. */
325         if (!*bh)
326                 *bh = tmp;
327
328 out:
329         return rc;
330 }
331
332 static int ocfs2_block_group_fill(handle_t *handle,
333                                   struct inode *alloc_inode,
334                                   struct buffer_head *bg_bh,
335                                   u64 group_blkno,
336                                   u16 my_chain,
337                                   struct ocfs2_chain_list *cl)
338 {
339         int status = 0;
340         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
341         struct super_block * sb = alloc_inode->i_sb;
342
343         mlog_entry_void();
344
345         if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
346                 ocfs2_error(alloc_inode->i_sb, "group block (%llu) != "
347                             "b_blocknr (%llu)",
348                             (unsigned long long)group_blkno,
349                             (unsigned long long) bg_bh->b_blocknr);
350                 status = -EIO;
351                 goto bail;
352         }
353
354         status = ocfs2_journal_access_gd(handle,
355                                          INODE_CACHE(alloc_inode),
356                                          bg_bh,
357                                          OCFS2_JOURNAL_ACCESS_CREATE);
358         if (status < 0) {
359                 mlog_errno(status);
360                 goto bail;
361         }
362
363         memset(bg, 0, sb->s_blocksize);
364         strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
365         bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
366         bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb));
367         bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
368         bg->bg_chain = cpu_to_le16(my_chain);
369         bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
370         bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
371         bg->bg_blkno = cpu_to_le64(group_blkno);
372         /* set the 1st bit in the bitmap to account for the descriptor block */
373         ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
374         bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
375
376         status = ocfs2_journal_dirty(handle, bg_bh);
377         if (status < 0)
378                 mlog_errno(status);
379
380         /* There is no need to zero out or otherwise initialize the
381          * other blocks in a group - All valid FS metadata in a block
382          * group stores the superblock fs_generation value at
383          * allocation time. */
384
385 bail:
386         mlog_exit(status);
387         return status;
388 }
389
390 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
391 {
392         u16 curr, best;
393
394         best = curr = 0;
395         while (curr < le16_to_cpu(cl->cl_count)) {
396                 if (le32_to_cpu(cl->cl_recs[best].c_total) >
397                     le32_to_cpu(cl->cl_recs[curr].c_total))
398                         best = curr;
399                 curr++;
400         }
401         return best;
402 }
403
404 /*
405  * We expect the block group allocator to already be locked.
406  */
407 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
408                                    struct inode *alloc_inode,
409                                    struct buffer_head *bh,
410                                    u64 max_block,
411                                    u64 *last_alloc_group,
412                                    int flags)
413 {
414         int status, credits;
415         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
416         struct ocfs2_chain_list *cl;
417         struct ocfs2_alloc_context *ac = NULL;
418         handle_t *handle = NULL;
419         u32 bit_off, num_bits;
420         u16 alloc_rec;
421         u64 bg_blkno;
422         struct buffer_head *bg_bh = NULL;
423         struct ocfs2_group_desc *bg;
424
425         BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
426
427         mlog_entry_void();
428
429         cl = &fe->id2.i_chain;
430         status = ocfs2_reserve_clusters_with_limit(osb,
431                                                    le16_to_cpu(cl->cl_cpg),
432                                                    max_block, flags, &ac);
433         if (status < 0) {
434                 if (status != -ENOSPC)
435                         mlog_errno(status);
436                 goto bail;
437         }
438
439         credits = ocfs2_calc_group_alloc_credits(osb->sb,
440                                                  le16_to_cpu(cl->cl_cpg));
441         handle = ocfs2_start_trans(osb, credits);
442         if (IS_ERR(handle)) {
443                 status = PTR_ERR(handle);
444                 handle = NULL;
445                 mlog_errno(status);
446                 goto bail;
447         }
448
449         if (last_alloc_group && *last_alloc_group != 0) {
450                 mlog(0, "use old allocation group %llu for block group alloc\n",
451                      (unsigned long long)*last_alloc_group);
452                 ac->ac_last_group = *last_alloc_group;
453         }
454         status = ocfs2_claim_clusters(osb,
455                                       handle,
456                                       ac,
457                                       le16_to_cpu(cl->cl_cpg),
458                                       &bit_off,
459                                       &num_bits);
460         if (status < 0) {
461                 if (status != -ENOSPC)
462                         mlog_errno(status);
463                 goto bail;
464         }
465
466         alloc_rec = ocfs2_find_smallest_chain(cl);
467
468         /* setup the group */
469         bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
470         mlog(0, "new descriptor, record %u, at block %llu\n",
471              alloc_rec, (unsigned long long)bg_blkno);
472
473         bg_bh = sb_getblk(osb->sb, bg_blkno);
474         if (!bg_bh) {
475                 status = -EIO;
476                 mlog_errno(status);
477                 goto bail;
478         }
479         ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
480
481         status = ocfs2_block_group_fill(handle,
482                                         alloc_inode,
483                                         bg_bh,
484                                         bg_blkno,
485                                         alloc_rec,
486                                         cl);
487         if (status < 0) {
488                 mlog_errno(status);
489                 goto bail;
490         }
491
492         bg = (struct ocfs2_group_desc *) bg_bh->b_data;
493
494         status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
495                                          bh, OCFS2_JOURNAL_ACCESS_WRITE);
496         if (status < 0) {
497                 mlog_errno(status);
498                 goto bail;
499         }
500
501         le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
502                      le16_to_cpu(bg->bg_free_bits_count));
503         le32_add_cpu(&cl->cl_recs[alloc_rec].c_total, le16_to_cpu(bg->bg_bits));
504         cl->cl_recs[alloc_rec].c_blkno  = cpu_to_le64(bg_blkno);
505         if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
506                 le16_add_cpu(&cl->cl_next_free_rec, 1);
507
508         le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
509                                         le16_to_cpu(bg->bg_free_bits_count));
510         le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
511         le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
512
513         status = ocfs2_journal_dirty(handle, bh);
514         if (status < 0) {
515                 mlog_errno(status);
516                 goto bail;
517         }
518
519         spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
520         OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
521         fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
522                                              le32_to_cpu(fe->i_clusters)));
523         spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
524         i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
525         alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
526
527         status = 0;
528
529         /* save the new last alloc group so that the caller can cache it. */
530         if (last_alloc_group)
531                 *last_alloc_group = ac->ac_last_group;
532
533 bail:
534         if (handle)
535                 ocfs2_commit_trans(osb, handle);
536
537         if (ac)
538                 ocfs2_free_alloc_context(ac);
539
540         brelse(bg_bh);
541
542         mlog_exit(status);
543         return status;
544 }
545
546 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
547                                        struct ocfs2_alloc_context *ac,
548                                        int type,
549                                        u32 slot,
550                                        u64 *last_alloc_group,
551                                        int flags)
552 {
553         int status;
554         u32 bits_wanted = ac->ac_bits_wanted;
555         struct inode *alloc_inode;
556         struct buffer_head *bh = NULL;
557         struct ocfs2_dinode *fe;
558         u32 free_bits;
559
560         mlog_entry_void();
561
562         alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
563         if (!alloc_inode) {
564                 mlog_errno(-EINVAL);
565                 return -EINVAL;
566         }
567
568         mutex_lock(&alloc_inode->i_mutex);
569
570         status = ocfs2_inode_lock(alloc_inode, &bh, 1);
571         if (status < 0) {
572                 mutex_unlock(&alloc_inode->i_mutex);
573                 iput(alloc_inode);
574
575                 mlog_errno(status);
576                 return status;
577         }
578
579         ac->ac_inode = alloc_inode;
580         ac->ac_alloc_slot = slot;
581
582         fe = (struct ocfs2_dinode *) bh->b_data;
583
584         /* The bh was validated by the inode read inside
585          * ocfs2_inode_lock().  Any corruption is a code bug. */
586         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
587
588         if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
589                 ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
590                             (unsigned long long)le64_to_cpu(fe->i_blkno));
591                 status = -EIO;
592                 goto bail;
593         }
594
595         free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
596                 le32_to_cpu(fe->id1.bitmap1.i_used);
597
598         if (bits_wanted > free_bits) {
599                 /* cluster bitmap never grows */
600                 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
601                         mlog(0, "Disk Full: wanted=%u, free_bits=%u\n",
602                              bits_wanted, free_bits);
603                         status = -ENOSPC;
604                         goto bail;
605                 }
606
607                 if (!(flags & ALLOC_NEW_GROUP)) {
608                         mlog(0, "Alloc File %u Full: wanted=%u, free_bits=%u, "
609                              "and we don't alloc a new group for it.\n",
610                              slot, bits_wanted, free_bits);
611                         status = -ENOSPC;
612                         goto bail;
613                 }
614
615                 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
616                                                  ac->ac_max_block,
617                                                  last_alloc_group, flags);
618                 if (status < 0) {
619                         if (status != -ENOSPC)
620                                 mlog_errno(status);
621                         goto bail;
622                 }
623                 atomic_inc(&osb->alloc_stats.bg_extends);
624
625                 /* You should never ask for this much metadata */
626                 BUG_ON(bits_wanted >
627                        (le32_to_cpu(fe->id1.bitmap1.i_total)
628                         - le32_to_cpu(fe->id1.bitmap1.i_used)));
629         }
630
631         get_bh(bh);
632         ac->ac_bh = bh;
633 bail:
634         brelse(bh);
635
636         mlog_exit(status);
637         return status;
638 }
639
640 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
641                                       int blocks,
642                                       struct ocfs2_alloc_context **ac)
643 {
644         int status;
645         u32 slot;
646
647         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
648         if (!(*ac)) {
649                 status = -ENOMEM;
650                 mlog_errno(status);
651                 goto bail;
652         }
653
654         (*ac)->ac_bits_wanted = blocks;
655         (*ac)->ac_which = OCFS2_AC_USE_META;
656         slot = osb->slot_num;
657         (*ac)->ac_group_search = ocfs2_block_group_search;
658
659         status = ocfs2_reserve_suballoc_bits(osb, (*ac),
660                                              EXTENT_ALLOC_SYSTEM_INODE,
661                                              slot, NULL, ALLOC_NEW_GROUP);
662         if (status < 0) {
663                 if (status != -ENOSPC)
664                         mlog_errno(status);
665                 goto bail;
666         }
667
668         status = 0;
669 bail:
670         if ((status < 0) && *ac) {
671                 ocfs2_free_alloc_context(*ac);
672                 *ac = NULL;
673         }
674
675         mlog_exit(status);
676         return status;
677 }
678
679 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
680                                struct ocfs2_extent_list *root_el,
681                                struct ocfs2_alloc_context **ac)
682 {
683         return ocfs2_reserve_new_metadata_blocks(osb,
684                                         ocfs2_extend_meta_needed(root_el),
685                                         ac);
686 }
687
688 static int ocfs2_steal_inode_from_other_nodes(struct ocfs2_super *osb,
689                                               struct ocfs2_alloc_context *ac)
690 {
691         int i, status = -ENOSPC;
692         s16 slot = ocfs2_get_inode_steal_slot(osb);
693
694         /* Start to steal inodes from the first slot after ours. */
695         if (slot == OCFS2_INVALID_SLOT)
696                 slot = osb->slot_num + 1;
697
698         for (i = 0; i < osb->max_slots; i++, slot++) {
699                 if (slot == osb->max_slots)
700                         slot = 0;
701
702                 if (slot == osb->slot_num)
703                         continue;
704
705                 status = ocfs2_reserve_suballoc_bits(osb, ac,
706                                                      INODE_ALLOC_SYSTEM_INODE,
707                                                      slot, NULL,
708                                                      NOT_ALLOC_NEW_GROUP);
709                 if (status >= 0) {
710                         ocfs2_set_inode_steal_slot(osb, slot);
711                         break;
712                 }
713
714                 ocfs2_free_ac_resource(ac);
715         }
716
717         return status;
718 }
719
720 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
721                             struct ocfs2_alloc_context **ac)
722 {
723         int status;
724         s16 slot = ocfs2_get_inode_steal_slot(osb);
725         u64 alloc_group;
726
727         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
728         if (!(*ac)) {
729                 status = -ENOMEM;
730                 mlog_errno(status);
731                 goto bail;
732         }
733
734         (*ac)->ac_bits_wanted = 1;
735         (*ac)->ac_which = OCFS2_AC_USE_INODE;
736
737         (*ac)->ac_group_search = ocfs2_block_group_search;
738
739         /*
740          * stat(2) can't handle i_ino > 32bits, so we tell the
741          * lower levels not to allocate us a block group past that
742          * limit.  The 'inode64' mount option avoids this behavior.
743          */
744         if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
745                 (*ac)->ac_max_block = (u32)~0U;
746
747         /*
748          * slot is set when we successfully steal inode from other nodes.
749          * It is reset in 3 places:
750          * 1. when we flush the truncate log
751          * 2. when we complete local alloc recovery.
752          * 3. when we successfully allocate from our own slot.
753          * After it is set, we will go on stealing inodes until we find the
754          * need to check our slots to see whether there is some space for us.
755          */
756         if (slot != OCFS2_INVALID_SLOT &&
757             atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_INODES_TO_STEAL)
758                 goto inode_steal;
759
760         atomic_set(&osb->s_num_inodes_stolen, 0);
761         alloc_group = osb->osb_inode_alloc_group;
762         status = ocfs2_reserve_suballoc_bits(osb, *ac,
763                                              INODE_ALLOC_SYSTEM_INODE,
764                                              osb->slot_num,
765                                              &alloc_group,
766                                              ALLOC_NEW_GROUP |
767                                              ALLOC_GROUPS_FROM_GLOBAL);
768         if (status >= 0) {
769                 status = 0;
770
771                 spin_lock(&osb->osb_lock);
772                 osb->osb_inode_alloc_group = alloc_group;
773                 spin_unlock(&osb->osb_lock);
774                 mlog(0, "after reservation, new allocation group is "
775                      "%llu\n", (unsigned long long)alloc_group);
776
777                 /*
778                  * Some inodes must be freed by us, so try to allocate
779                  * from our own next time.
780                  */
781                 if (slot != OCFS2_INVALID_SLOT)
782                         ocfs2_init_inode_steal_slot(osb);
783                 goto bail;
784         } else if (status < 0 && status != -ENOSPC) {
785                 mlog_errno(status);
786                 goto bail;
787         }
788
789         ocfs2_free_ac_resource(*ac);
790
791 inode_steal:
792         status = ocfs2_steal_inode_from_other_nodes(osb, *ac);
793         atomic_inc(&osb->s_num_inodes_stolen);
794         if (status < 0) {
795                 if (status != -ENOSPC)
796                         mlog_errno(status);
797                 goto bail;
798         }
799
800         status = 0;
801 bail:
802         if ((status < 0) && *ac) {
803                 ocfs2_free_alloc_context(*ac);
804                 *ac = NULL;
805         }
806
807         mlog_exit(status);
808         return status;
809 }
810
811 /* local alloc code has to do the same thing, so rather than do this
812  * twice.. */
813 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
814                                       struct ocfs2_alloc_context *ac)
815 {
816         int status;
817
818         ac->ac_which = OCFS2_AC_USE_MAIN;
819         ac->ac_group_search = ocfs2_cluster_group_search;
820
821         status = ocfs2_reserve_suballoc_bits(osb, ac,
822                                              GLOBAL_BITMAP_SYSTEM_INODE,
823                                              OCFS2_INVALID_SLOT, NULL,
824                                              ALLOC_NEW_GROUP);
825         if (status < 0 && status != -ENOSPC) {
826                 mlog_errno(status);
827                 goto bail;
828         }
829
830 bail:
831         return status;
832 }
833
834 /* Callers don't need to care which bitmap (local alloc or main) to
835  * use so we figure it out for them, but unfortunately this clutters
836  * things a bit. */
837 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
838                                              u32 bits_wanted, u64 max_block,
839                                              int flags,
840                                              struct ocfs2_alloc_context **ac)
841 {
842         int status;
843
844         mlog_entry_void();
845
846         *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
847         if (!(*ac)) {
848                 status = -ENOMEM;
849                 mlog_errno(status);
850                 goto bail;
851         }
852
853         (*ac)->ac_bits_wanted = bits_wanted;
854         (*ac)->ac_max_block = max_block;
855
856         status = -ENOSPC;
857         if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
858             ocfs2_alloc_should_use_local(osb, bits_wanted)) {
859                 status = ocfs2_reserve_local_alloc_bits(osb,
860                                                         bits_wanted,
861                                                         *ac);
862                 if (status == -EFBIG) {
863                         /* The local alloc window is outside ac_max_block.
864                          * use the main bitmap. */
865                         status = -ENOSPC;
866                 } else if ((status < 0) && (status != -ENOSPC)) {
867                         mlog_errno(status);
868                         goto bail;
869                 }
870         }
871
872         if (status == -ENOSPC) {
873                 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
874                 if (status < 0) {
875                         if (status != -ENOSPC)
876                                 mlog_errno(status);
877                         goto bail;
878                 }
879         }
880
881         status = 0;
882 bail:
883         if ((status < 0) && *ac) {
884                 ocfs2_free_alloc_context(*ac);
885                 *ac = NULL;
886         }
887
888         mlog_exit(status);
889         return status;
890 }
891
892 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
893                            u32 bits_wanted,
894                            struct ocfs2_alloc_context **ac)
895 {
896         return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
897                                                  ALLOC_NEW_GROUP, ac);
898 }
899
900 /*
901  * More or less lifted from ext3. I'll leave their description below:
902  *
903  * "For ext3 allocations, we must not reuse any blocks which are
904  * allocated in the bitmap buffer's "last committed data" copy.  This
905  * prevents deletes from freeing up the page for reuse until we have
906  * committed the delete transaction.
907  *
908  * If we didn't do this, then deleting something and reallocating it as
909  * data would allow the old block to be overwritten before the
910  * transaction committed (because we force data to disk before commit).
911  * This would lead to corruption if we crashed between overwriting the
912  * data and committing the delete.
913  *
914  * @@@ We may want to make this allocation behaviour conditional on
915  * data-writes at some point, and disable it for metadata allocations or
916  * sync-data inodes."
917  *
918  * Note: OCFS2 already does this differently for metadata vs data
919  * allocations, as those bitmaps are separate and undo access is never
920  * called on a metadata group descriptor.
921  */
922 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
923                                          int nr)
924 {
925         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
926         int ret;
927
928         if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
929                 return 0;
930
931         if (!buffer_jbd(bg_bh))
932                 return 1;
933
934         jbd_lock_bh_state(bg_bh);
935         bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
936         if (bg)
937                 ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
938         else
939                 ret = 1;
940         jbd_unlock_bh_state(bg_bh);
941
942         return ret;
943 }
944
945 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
946                                              struct buffer_head *bg_bh,
947                                              unsigned int bits_wanted,
948                                              unsigned int total_bits,
949                                              u16 *bit_off,
950                                              u16 *bits_found)
951 {
952         void *bitmap;
953         u16 best_offset, best_size;
954         int offset, start, found, status = 0;
955         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
956
957         /* Callers got this descriptor from
958          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
959         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
960
961         found = start = best_offset = best_size = 0;
962         bitmap = bg->bg_bitmap;
963
964         while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
965                 if (offset == total_bits)
966                         break;
967
968                 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
969                         /* We found a zero, but we can't use it as it
970                          * hasn't been put to disk yet! */
971                         found = 0;
972                         start = offset + 1;
973                 } else if (offset == start) {
974                         /* we found a zero */
975                         found++;
976                         /* move start to the next bit to test */
977                         start++;
978                 } else {
979                         /* got a zero after some ones */
980                         found = 1;
981                         start = offset + 1;
982                 }
983                 if (found > best_size) {
984                         best_size = found;
985                         best_offset = start - found;
986                 }
987                 /* we got everything we needed */
988                 if (found == bits_wanted) {
989                         /* mlog(0, "Found it all!\n"); */
990                         break;
991                 }
992         }
993
994         /* XXX: I think the first clause is equivalent to the second
995          *      - jlbec */
996         if (found == bits_wanted) {
997                 *bit_off = start - found;
998                 *bits_found = found;
999         } else if (best_size) {
1000                 *bit_off = best_offset;
1001                 *bits_found = best_size;
1002         } else {
1003                 status = -ENOSPC;
1004                 /* No error log here -- see the comment above
1005                  * ocfs2_test_bg_bit_allocatable */
1006         }
1007
1008         return status;
1009 }
1010
1011 static inline int ocfs2_block_group_set_bits(handle_t *handle,
1012                                              struct inode *alloc_inode,
1013                                              struct ocfs2_group_desc *bg,
1014                                              struct buffer_head *group_bh,
1015                                              unsigned int bit_off,
1016                                              unsigned int num_bits)
1017 {
1018         int status;
1019         void *bitmap = bg->bg_bitmap;
1020         int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1021
1022         mlog_entry_void();
1023
1024         /* All callers get the descriptor via
1025          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1026         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1027         BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
1028
1029         mlog(0, "block_group_set_bits: off = %u, num = %u\n", bit_off,
1030              num_bits);
1031
1032         if (ocfs2_is_cluster_bitmap(alloc_inode))
1033                 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1034
1035         status = ocfs2_journal_access_gd(handle,
1036                                          INODE_CACHE(alloc_inode),
1037                                          group_bh,
1038                                          journal_type);
1039         if (status < 0) {
1040                 mlog_errno(status);
1041                 goto bail;
1042         }
1043
1044         le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
1045
1046         while(num_bits--)
1047                 ocfs2_set_bit(bit_off++, bitmap);
1048
1049         status = ocfs2_journal_dirty(handle,
1050                                      group_bh);
1051         if (status < 0) {
1052                 mlog_errno(status);
1053                 goto bail;
1054         }
1055
1056 bail:
1057         mlog_exit(status);
1058         return status;
1059 }
1060
1061 /* find the one with the most empty bits */
1062 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
1063 {
1064         u16 curr, best;
1065
1066         BUG_ON(!cl->cl_next_free_rec);
1067
1068         best = curr = 0;
1069         while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
1070                 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
1071                     le32_to_cpu(cl->cl_recs[best].c_free))
1072                         best = curr;
1073                 curr++;
1074         }
1075
1076         BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
1077         return best;
1078 }
1079
1080 static int ocfs2_relink_block_group(handle_t *handle,
1081                                     struct inode *alloc_inode,
1082                                     struct buffer_head *fe_bh,
1083                                     struct buffer_head *bg_bh,
1084                                     struct buffer_head *prev_bg_bh,
1085                                     u16 chain)
1086 {
1087         int status;
1088         /* there is a really tiny chance the journal calls could fail,
1089          * but we wouldn't want inconsistent blocks in *any* case. */
1090         u64 fe_ptr, bg_ptr, prev_bg_ptr;
1091         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1092         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
1093         struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
1094
1095         /* The caller got these descriptors from
1096          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1097         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1098         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
1099
1100         mlog(0, "Suballoc %llu, chain %u, move group %llu to top, prev = %llu\n",
1101              (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
1102              (unsigned long long)le64_to_cpu(bg->bg_blkno),
1103              (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
1104
1105         fe_ptr = le64_to_cpu(fe->id2.i_chain.cl_recs[chain].c_blkno);
1106         bg_ptr = le64_to_cpu(bg->bg_next_group);
1107         prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
1108
1109         status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1110                                          prev_bg_bh,
1111                                          OCFS2_JOURNAL_ACCESS_WRITE);
1112         if (status < 0) {
1113                 mlog_errno(status);
1114                 goto out_rollback;
1115         }
1116
1117         prev_bg->bg_next_group = bg->bg_next_group;
1118
1119         status = ocfs2_journal_dirty(handle, prev_bg_bh);
1120         if (status < 0) {
1121                 mlog_errno(status);
1122                 goto out_rollback;
1123         }
1124
1125         status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1126                                          bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1127         if (status < 0) {
1128                 mlog_errno(status);
1129                 goto out_rollback;
1130         }
1131
1132         bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
1133
1134         status = ocfs2_journal_dirty(handle, bg_bh);
1135         if (status < 0) {
1136                 mlog_errno(status);
1137                 goto out_rollback;
1138         }
1139
1140         status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1141                                          fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1142         if (status < 0) {
1143                 mlog_errno(status);
1144                 goto out_rollback;
1145         }
1146
1147         fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
1148
1149         status = ocfs2_journal_dirty(handle, fe_bh);
1150         if (status < 0) {
1151                 mlog_errno(status);
1152                 goto out_rollback;
1153         }
1154
1155         status = 0;
1156 out_rollback:
1157         if (status < 0) {
1158                 fe->id2.i_chain.cl_recs[chain].c_blkno = cpu_to_le64(fe_ptr);
1159                 bg->bg_next_group = cpu_to_le64(bg_ptr);
1160                 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1161         }
1162
1163         mlog_exit(status);
1164         return status;
1165 }
1166
1167 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1168                                                      u32 wanted)
1169 {
1170         return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1171 }
1172
1173 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1174  * value on error. */
1175 static int ocfs2_cluster_group_search(struct inode *inode,
1176                                       struct buffer_head *group_bh,
1177                                       u32 bits_wanted, u32 min_bits,
1178                                       u64 max_block,
1179                                       u16 *bit_off, u16 *bits_found)
1180 {
1181         int search = -ENOSPC;
1182         int ret;
1183         u64 blkoff;
1184         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1185         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1186         u16 tmp_off, tmp_found;
1187         unsigned int max_bits, gd_cluster_off;
1188
1189         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1190
1191         if (gd->bg_free_bits_count) {
1192                 max_bits = le16_to_cpu(gd->bg_bits);
1193
1194                 /* Tail groups in cluster bitmaps which aren't cpg
1195                  * aligned are prone to partial extention by a failed
1196                  * fs resize. If the file system resize never got to
1197                  * update the dinode cluster count, then we don't want
1198                  * to trust any clusters past it, regardless of what
1199                  * the group descriptor says. */
1200                 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1201                                                           le64_to_cpu(gd->bg_blkno));
1202                 if ((gd_cluster_off + max_bits) >
1203                     OCFS2_I(inode)->ip_clusters) {
1204                         max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1205                         mlog(0, "Desc %llu, bg_bits %u, clusters %u, use %u\n",
1206                              (unsigned long long)le64_to_cpu(gd->bg_blkno),
1207                              le16_to_cpu(gd->bg_bits),
1208                              OCFS2_I(inode)->ip_clusters, max_bits);
1209                 }
1210
1211                 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1212                                                         group_bh, bits_wanted,
1213                                                         max_bits,
1214                                                         &tmp_off, &tmp_found);
1215                 if (ret)
1216                         return ret;
1217
1218                 if (max_block) {
1219                         blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1220                                                           gd_cluster_off +
1221                                                           tmp_off + tmp_found);
1222                         mlog(0, "Checking %llu against %llu\n",
1223                              (unsigned long long)blkoff,
1224                              (unsigned long long)max_block);
1225                         if (blkoff > max_block)
1226                                 return -ENOSPC;
1227                 }
1228
1229                 /* ocfs2_block_group_find_clear_bits() might
1230                  * return success, but we still want to return
1231                  * -ENOSPC unless it found the minimum number
1232                  * of bits. */
1233                 if (min_bits <= tmp_found) {
1234                         *bit_off = tmp_off;
1235                         *bits_found = tmp_found;
1236                         search = 0; /* success */
1237                 } else if (tmp_found) {
1238                         /*
1239                          * Don't show bits which we'll be returning
1240                          * for allocation to the local alloc bitmap.
1241                          */
1242                         ocfs2_local_alloc_seen_free_bits(osb, tmp_found);
1243                 }
1244         }
1245
1246         return search;
1247 }
1248
1249 static int ocfs2_block_group_search(struct inode *inode,
1250                                     struct buffer_head *group_bh,
1251                                     u32 bits_wanted, u32 min_bits,
1252                                     u64 max_block,
1253                                     u16 *bit_off, u16 *bits_found)
1254 {
1255         int ret = -ENOSPC;
1256         u64 blkoff;
1257         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1258
1259         BUG_ON(min_bits != 1);
1260         BUG_ON(ocfs2_is_cluster_bitmap(inode));
1261
1262         if (bg->bg_free_bits_count) {
1263                 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1264                                                         group_bh, bits_wanted,
1265                                                         le16_to_cpu(bg->bg_bits),
1266                                                         bit_off, bits_found);
1267                 if (!ret && max_block) {
1268                         blkoff = le64_to_cpu(bg->bg_blkno) + *bit_off +
1269                                 *bits_found;
1270                         mlog(0, "Checking %llu against %llu\n",
1271                              (unsigned long long)blkoff,
1272                              (unsigned long long)max_block);
1273                         if (blkoff > max_block)
1274                                 ret = -ENOSPC;
1275                 }
1276         }
1277
1278         return ret;
1279 }
1280
1281 static int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1282                                        handle_t *handle,
1283                                        struct buffer_head *di_bh,
1284                                        u32 num_bits,
1285                                        u16 chain)
1286 {
1287         int ret;
1288         u32 tmp_used;
1289         struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1290         struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1291
1292         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1293                                       OCFS2_JOURNAL_ACCESS_WRITE);
1294         if (ret < 0) {
1295                 mlog_errno(ret);
1296                 goto out;
1297         }
1298
1299         tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1300         di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1301         le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1302
1303         ret = ocfs2_journal_dirty(handle, di_bh);
1304         if (ret < 0)
1305                 mlog_errno(ret);
1306
1307 out:
1308         return ret;
1309 }
1310
1311 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1312                                   handle_t *handle,
1313                                   u32 bits_wanted,
1314                                   u32 min_bits,
1315                                   u16 *bit_off,
1316                                   unsigned int *num_bits,
1317                                   u64 gd_blkno,
1318                                   u16 *bits_left)
1319 {
1320         int ret;
1321         u16 found;
1322         struct buffer_head *group_bh = NULL;
1323         struct ocfs2_group_desc *gd;
1324         struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
1325         struct inode *alloc_inode = ac->ac_inode;
1326
1327         ret = ocfs2_read_group_descriptor(alloc_inode, di, gd_blkno,
1328                                           &group_bh);
1329         if (ret < 0) {
1330                 mlog_errno(ret);
1331                 return ret;
1332         }
1333
1334         gd = (struct ocfs2_group_desc *) group_bh->b_data;
1335         ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1336                                   ac->ac_max_block, bit_off, &found);
1337         if (ret < 0) {
1338                 if (ret != -ENOSPC)
1339                         mlog_errno(ret);
1340                 goto out;
1341         }
1342
1343         *num_bits = found;
1344
1345         ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1346                                                *num_bits,
1347                                                le16_to_cpu(gd->bg_chain));
1348         if (ret < 0) {
1349                 mlog_errno(ret);
1350                 goto out;
1351         }
1352
1353         ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1354                                          *bit_off, *num_bits);
1355         if (ret < 0)
1356                 mlog_errno(ret);
1357
1358         *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1359
1360 out:
1361         brelse(group_bh);
1362
1363         return ret;
1364 }
1365
1366 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1367                               handle_t *handle,
1368                               u32 bits_wanted,
1369                               u32 min_bits,
1370                               u16 *bit_off,
1371                               unsigned int *num_bits,
1372                               u64 *bg_blkno,
1373                               u16 *bits_left)
1374 {
1375         int status;
1376         u16 chain, tmp_bits;
1377         u32 tmp_used;
1378         u64 next_group;
1379         struct inode *alloc_inode = ac->ac_inode;
1380         struct buffer_head *group_bh = NULL;
1381         struct buffer_head *prev_group_bh = NULL;
1382         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1383         struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1384         struct ocfs2_group_desc *bg;
1385
1386         chain = ac->ac_chain;
1387         mlog(0, "trying to alloc %u bits from chain %u, inode %llu\n",
1388              bits_wanted, chain,
1389              (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno);
1390
1391         status = ocfs2_read_group_descriptor(alloc_inode, fe,
1392                                              le64_to_cpu(cl->cl_recs[chain].c_blkno),
1393                                              &group_bh);
1394         if (status < 0) {
1395                 mlog_errno(status);
1396                 goto bail;
1397         }
1398         bg = (struct ocfs2_group_desc *) group_bh->b_data;
1399
1400         status = -ENOSPC;
1401         /* for now, the chain search is a bit simplistic. We just use
1402          * the 1st group with any empty bits. */
1403         while ((status = ac->ac_group_search(alloc_inode, group_bh,
1404                                              bits_wanted, min_bits,
1405                                              ac->ac_max_block, bit_off,
1406                                              &tmp_bits)) == -ENOSPC) {
1407                 if (!bg->bg_next_group)
1408                         break;
1409
1410                 brelse(prev_group_bh);
1411                 prev_group_bh = NULL;
1412
1413                 next_group = le64_to_cpu(bg->bg_next_group);
1414                 prev_group_bh = group_bh;
1415                 group_bh = NULL;
1416                 status = ocfs2_read_group_descriptor(alloc_inode, fe,
1417                                                      next_group, &group_bh);
1418                 if (status < 0) {
1419                         mlog_errno(status);
1420                         goto bail;
1421                 }
1422                 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1423         }
1424         if (status < 0) {
1425                 if (status != -ENOSPC)
1426                         mlog_errno(status);
1427                 goto bail;
1428         }
1429
1430         mlog(0, "alloc succeeds: we give %u bits from block group %llu\n",
1431              tmp_bits, (unsigned long long)le64_to_cpu(bg->bg_blkno));
1432
1433         *num_bits = tmp_bits;
1434
1435         BUG_ON(*num_bits == 0);
1436
1437         /*
1438          * Keep track of previous block descriptor read. When
1439          * we find a target, if we have read more than X
1440          * number of descriptors, and the target is reasonably
1441          * empty, relink him to top of his chain.
1442          *
1443          * We've read 0 extra blocks and only send one more to
1444          * the transaction, yet the next guy to search has a
1445          * much easier time.
1446          *
1447          * Do this *after* figuring out how many bits we're taking out
1448          * of our target group.
1449          */
1450         if (ac->ac_allow_chain_relink &&
1451             (prev_group_bh) &&
1452             (ocfs2_block_group_reasonably_empty(bg, *num_bits))) {
1453                 status = ocfs2_relink_block_group(handle, alloc_inode,
1454                                                   ac->ac_bh, group_bh,
1455                                                   prev_group_bh, chain);
1456                 if (status < 0) {
1457                         mlog_errno(status);
1458                         goto bail;
1459                 }
1460         }
1461
1462         /* Ok, claim our bits now: set the info on dinode, chainlist
1463          * and then the group */
1464         status = ocfs2_journal_access_di(handle,
1465                                          INODE_CACHE(alloc_inode),
1466                                          ac->ac_bh,
1467                                          OCFS2_JOURNAL_ACCESS_WRITE);
1468         if (status < 0) {
1469                 mlog_errno(status);
1470                 goto bail;
1471         }
1472
1473         tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1474         fe->id1.bitmap1.i_used = cpu_to_le32(*num_bits + tmp_used);
1475         le32_add_cpu(&cl->cl_recs[chain].c_free, -(*num_bits));
1476
1477         status = ocfs2_journal_dirty(handle,
1478                                      ac->ac_bh);
1479         if (status < 0) {
1480                 mlog_errno(status);
1481                 goto bail;
1482         }
1483
1484         status = ocfs2_block_group_set_bits(handle,
1485                                             alloc_inode,
1486                                             bg,
1487                                             group_bh,
1488                                             *bit_off,
1489                                             *num_bits);
1490         if (status < 0) {
1491                 mlog_errno(status);
1492                 goto bail;
1493         }
1494
1495         mlog(0, "Allocated %u bits from suballocator %llu\n", *num_bits,
1496              (unsigned long long)le64_to_cpu(fe->i_blkno));
1497
1498         *bg_blkno = le64_to_cpu(bg->bg_blkno);
1499         *bits_left = le16_to_cpu(bg->bg_free_bits_count);
1500 bail:
1501         brelse(group_bh);
1502         brelse(prev_group_bh);
1503
1504         mlog_exit(status);
1505         return status;
1506 }
1507
1508 /* will give out up to bits_wanted contiguous bits. */
1509 static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
1510                                      struct ocfs2_alloc_context *ac,
1511                                      handle_t *handle,
1512                                      u32 bits_wanted,
1513                                      u32 min_bits,
1514                                      u16 *bit_off,
1515                                      unsigned int *num_bits,
1516                                      u64 *bg_blkno)
1517 {
1518         int status;
1519         u16 victim, i;
1520         u16 bits_left = 0;
1521         u64 hint_blkno = ac->ac_last_group;
1522         struct ocfs2_chain_list *cl;
1523         struct ocfs2_dinode *fe;
1524
1525         mlog_entry_void();
1526
1527         BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1528         BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1529         BUG_ON(!ac->ac_bh);
1530
1531         fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1532
1533         /* The bh was validated by the inode read during
1534          * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
1535         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1536
1537         if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1538             le32_to_cpu(fe->id1.bitmap1.i_total)) {
1539                 ocfs2_error(osb->sb, "Chain allocator dinode %llu has %u used "
1540                             "bits but only %u total.",
1541                             (unsigned long long)le64_to_cpu(fe->i_blkno),
1542                             le32_to_cpu(fe->id1.bitmap1.i_used),
1543                             le32_to_cpu(fe->id1.bitmap1.i_total));
1544                 status = -EIO;
1545                 goto bail;
1546         }
1547
1548         if (hint_blkno) {
1549                 /* Attempt to short-circuit the usual search mechanism
1550                  * by jumping straight to the most recently used
1551                  * allocation group. This helps us mantain some
1552                  * contiguousness across allocations. */
1553                 status = ocfs2_search_one_group(ac, handle, bits_wanted,
1554                                                 min_bits, bit_off, num_bits,
1555                                                 hint_blkno, &bits_left);
1556                 if (!status) {
1557                         /* Be careful to update *bg_blkno here as the
1558                          * caller is expecting it to be filled in, and
1559                          * ocfs2_search_one_group() won't do that for
1560                          * us. */
1561                         *bg_blkno = hint_blkno;
1562                         goto set_hint;
1563                 }
1564                 if (status < 0 && status != -ENOSPC) {
1565                         mlog_errno(status);
1566                         goto bail;
1567                 }
1568         }
1569
1570         cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1571
1572         victim = ocfs2_find_victim_chain(cl);
1573         ac->ac_chain = victim;
1574         ac->ac_allow_chain_relink = 1;
1575
1576         status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits, bit_off,
1577                                     num_bits, bg_blkno, &bits_left);
1578         if (!status)
1579                 goto set_hint;
1580         if (status < 0 && status != -ENOSPC) {
1581                 mlog_errno(status);
1582                 goto bail;
1583         }
1584
1585         mlog(0, "Search of victim chain %u came up with nothing, "
1586              "trying all chains now.\n", victim);
1587
1588         /* If we didn't pick a good victim, then just default to
1589          * searching each chain in order. Don't allow chain relinking
1590          * because we only calculate enough journal credits for one
1591          * relink per alloc. */
1592         ac->ac_allow_chain_relink = 0;
1593         for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1594                 if (i == victim)
1595                         continue;
1596                 if (!cl->cl_recs[i].c_free)
1597                         continue;
1598
1599                 ac->ac_chain = i;
1600                 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1601                                             bit_off, num_bits, bg_blkno,
1602                                             &bits_left);
1603                 if (!status)
1604                         break;
1605                 if (status < 0 && status != -ENOSPC) {
1606                         mlog_errno(status);
1607                         goto bail;
1608                 }
1609         }
1610
1611 set_hint:
1612         if (status != -ENOSPC) {
1613                 /* If the next search of this group is not likely to
1614                  * yield a suitable extent, then we reset the last
1615                  * group hint so as to not waste a disk read */
1616                 if (bits_left < min_bits)
1617                         ac->ac_last_group = 0;
1618                 else
1619                         ac->ac_last_group = *bg_blkno;
1620         }
1621
1622 bail:
1623         mlog_exit(status);
1624         return status;
1625 }
1626
1627 int ocfs2_claim_metadata(struct ocfs2_super *osb,
1628                          handle_t *handle,
1629                          struct ocfs2_alloc_context *ac,
1630                          u32 bits_wanted,
1631                          u16 *suballoc_bit_start,
1632                          unsigned int *num_bits,
1633                          u64 *blkno_start)
1634 {
1635         int status;
1636         u64 bg_blkno;
1637
1638         BUG_ON(!ac);
1639         BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1640         BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
1641
1642         status = ocfs2_claim_suballoc_bits(osb,
1643                                            ac,
1644                                            handle,
1645                                            bits_wanted,
1646                                            1,
1647                                            suballoc_bit_start,
1648                                            num_bits,
1649                                            &bg_blkno);
1650         if (status < 0) {
1651                 mlog_errno(status);
1652                 goto bail;
1653         }
1654         atomic_inc(&osb->alloc_stats.bg_allocs);
1655
1656         *blkno_start = bg_blkno + (u64) *suballoc_bit_start;
1657         ac->ac_bits_given += (*num_bits);
1658         status = 0;
1659 bail:
1660         mlog_exit(status);
1661         return status;
1662 }
1663
1664 static void ocfs2_init_inode_ac_group(struct inode *dir,
1665                                       struct buffer_head *parent_fe_bh,
1666                                       struct ocfs2_alloc_context *ac)
1667 {
1668         struct ocfs2_dinode *fe = (struct ocfs2_dinode *)parent_fe_bh->b_data;
1669         /*
1670          * Try to allocate inodes from some specific group.
1671          *
1672          * If the parent dir has recorded the last group used in allocation,
1673          * cool, use it. Otherwise if we try to allocate new inode from the
1674          * same slot the parent dir belongs to, use the same chunk.
1675          *
1676          * We are very careful here to avoid the mistake of setting
1677          * ac_last_group to a group descriptor from a different (unlocked) slot.
1678          */
1679         if (OCFS2_I(dir)->ip_last_used_group &&
1680             OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
1681                 ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
1682         else if (le16_to_cpu(fe->i_suballoc_slot) == ac->ac_alloc_slot)
1683                 ac->ac_last_group = ocfs2_which_suballoc_group(
1684                                         le64_to_cpu(fe->i_blkno),
1685                                         le16_to_cpu(fe->i_suballoc_bit));
1686 }
1687
1688 static inline void ocfs2_save_inode_ac_group(struct inode *dir,
1689                                              struct ocfs2_alloc_context *ac)
1690 {
1691         OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
1692         OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
1693 }
1694
1695 int ocfs2_claim_new_inode(struct ocfs2_super *osb,
1696                           handle_t *handle,
1697                           struct inode *dir,
1698                           struct buffer_head *parent_fe_bh,
1699                           struct ocfs2_alloc_context *ac,
1700                           u16 *suballoc_bit,
1701                           u64 *fe_blkno)
1702 {
1703         int status;
1704         unsigned int num_bits;
1705         u64 bg_blkno;
1706
1707         mlog_entry_void();
1708
1709         BUG_ON(!ac);
1710         BUG_ON(ac->ac_bits_given != 0);
1711         BUG_ON(ac->ac_bits_wanted != 1);
1712         BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
1713
1714         ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
1715
1716         status = ocfs2_claim_suballoc_bits(osb,
1717                                            ac,
1718                                            handle,
1719                                            1,
1720                                            1,
1721                                            suballoc_bit,
1722                                            &num_bits,
1723                                            &bg_blkno);
1724         if (status < 0) {
1725                 mlog_errno(status);
1726                 goto bail;
1727         }
1728         atomic_inc(&osb->alloc_stats.bg_allocs);
1729
1730         BUG_ON(num_bits != 1);
1731
1732         *fe_blkno = bg_blkno + (u64) (*suballoc_bit);
1733         ac->ac_bits_given++;
1734         ocfs2_save_inode_ac_group(dir, ac);
1735         status = 0;
1736 bail:
1737         mlog_exit(status);
1738         return status;
1739 }
1740
1741 /* translate a group desc. blkno and it's bitmap offset into
1742  * disk cluster offset. */
1743 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
1744                                                    u64 bg_blkno,
1745                                                    u16 bg_bit_off)
1746 {
1747         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1748         u32 cluster = 0;
1749
1750         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1751
1752         if (bg_blkno != osb->first_cluster_group_blkno)
1753                 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
1754         cluster += (u32) bg_bit_off;
1755         return cluster;
1756 }
1757
1758 /* given a cluster offset, calculate which block group it belongs to
1759  * and return that block offset. */
1760 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
1761 {
1762         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1763         u32 group_no;
1764
1765         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1766
1767         group_no = cluster / osb->bitmap_cpg;
1768         if (!group_no)
1769                 return osb->first_cluster_group_blkno;
1770         return ocfs2_clusters_to_blocks(inode->i_sb,
1771                                         group_no * osb->bitmap_cpg);
1772 }
1773
1774 /* given the block number of a cluster start, calculate which cluster
1775  * group and descriptor bitmap offset that corresponds to. */
1776 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
1777                                                 u64 data_blkno,
1778                                                 u64 *bg_blkno,
1779                                                 u16 *bg_bit_off)
1780 {
1781         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1782         u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
1783
1784         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1785
1786         *bg_blkno = ocfs2_which_cluster_group(inode,
1787                                               data_cluster);
1788
1789         if (*bg_blkno == osb->first_cluster_group_blkno)
1790                 *bg_bit_off = (u16) data_cluster;
1791         else
1792                 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
1793                                                              data_blkno - *bg_blkno);
1794 }
1795
1796 /*
1797  * min_bits - minimum contiguous chunk from this total allocation we
1798  * can handle. set to what we asked for originally for a full
1799  * contig. allocation, set to '1' to indicate we can deal with extents
1800  * of any size.
1801  */
1802 int __ocfs2_claim_clusters(struct ocfs2_super *osb,
1803                            handle_t *handle,
1804                            struct ocfs2_alloc_context *ac,
1805                            u32 min_clusters,
1806                            u32 max_clusters,
1807                            u32 *cluster_start,
1808                            u32 *num_clusters)
1809 {
1810         int status;
1811         unsigned int bits_wanted = max_clusters;
1812         u64 bg_blkno = 0;
1813         u16 bg_bit_off;
1814
1815         mlog_entry_void();
1816
1817         BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1818
1819         BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
1820                && ac->ac_which != OCFS2_AC_USE_MAIN);
1821
1822         if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
1823                 status = ocfs2_claim_local_alloc_bits(osb,
1824                                                       handle,
1825                                                       ac,
1826                                                       bits_wanted,
1827                                                       cluster_start,
1828                                                       num_clusters);
1829                 if (!status)
1830                         atomic_inc(&osb->alloc_stats.local_data);
1831         } else {
1832                 if (min_clusters > (osb->bitmap_cpg - 1)) {
1833                         /* The only paths asking for contiguousness
1834                          * should know about this already. */
1835                         mlog(ML_ERROR, "minimum allocation requested %u exceeds "
1836                              "group bitmap size %u!\n", min_clusters,
1837                              osb->bitmap_cpg);
1838                         status = -ENOSPC;
1839                         goto bail;
1840                 }
1841                 /* clamp the current request down to a realistic size. */
1842                 if (bits_wanted > (osb->bitmap_cpg - 1))
1843                         bits_wanted = osb->bitmap_cpg - 1;
1844
1845                 status = ocfs2_claim_suballoc_bits(osb,
1846                                                    ac,
1847                                                    handle,
1848                                                    bits_wanted,
1849                                                    min_clusters,
1850                                                    &bg_bit_off,
1851                                                    num_clusters,
1852                                                    &bg_blkno);
1853                 if (!status) {
1854                         *cluster_start =
1855                                 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
1856                                                                  bg_blkno,
1857                                                                  bg_bit_off);
1858                         atomic_inc(&osb->alloc_stats.bitmap_data);
1859                 }
1860         }
1861         if (status < 0) {
1862                 if (status != -ENOSPC)
1863                         mlog_errno(status);
1864                 goto bail;
1865         }
1866
1867         ac->ac_bits_given += *num_clusters;
1868
1869 bail:
1870         mlog_exit(status);
1871         return status;
1872 }
1873
1874 int ocfs2_claim_clusters(struct ocfs2_super *osb,
1875                          handle_t *handle,
1876                          struct ocfs2_alloc_context *ac,
1877                          u32 min_clusters,
1878                          u32 *cluster_start,
1879                          u32 *num_clusters)
1880 {
1881         unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
1882
1883         return __ocfs2_claim_clusters(osb, handle, ac, min_clusters,
1884                                       bits_wanted, cluster_start, num_clusters);
1885 }
1886
1887 static inline int ocfs2_block_group_clear_bits(handle_t *handle,
1888                                                struct inode *alloc_inode,
1889                                                struct ocfs2_group_desc *bg,
1890                                                struct buffer_head *group_bh,
1891                                                unsigned int bit_off,
1892                                                unsigned int num_bits)
1893 {
1894         int status;
1895         unsigned int tmp;
1896         int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1897         struct ocfs2_group_desc *undo_bg = NULL;
1898         int cluster_bitmap = 0;
1899
1900         mlog_entry_void();
1901
1902         /* The caller got this descriptor from
1903          * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
1904         BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
1905
1906         mlog(0, "off = %u, num = %u\n", bit_off, num_bits);
1907
1908         if (ocfs2_is_cluster_bitmap(alloc_inode))
1909                 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1910
1911         status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
1912                                          group_bh, journal_type);
1913         if (status < 0) {
1914                 mlog_errno(status);
1915                 goto bail;
1916         }
1917
1918         if (ocfs2_is_cluster_bitmap(alloc_inode))
1919                 cluster_bitmap = 1;
1920
1921         if (cluster_bitmap) {
1922                 jbd_lock_bh_state(group_bh);
1923                 undo_bg = (struct ocfs2_group_desc *)
1924                                         bh2jh(group_bh)->b_committed_data;
1925                 BUG_ON(!undo_bg);
1926         }
1927
1928         tmp = num_bits;
1929         while(tmp--) {
1930                 ocfs2_clear_bit((bit_off + tmp),
1931                                 (unsigned long *) bg->bg_bitmap);
1932                 if (cluster_bitmap)
1933                         ocfs2_set_bit(bit_off + tmp,
1934                                       (unsigned long *) undo_bg->bg_bitmap);
1935         }
1936         le16_add_cpu(&bg->bg_free_bits_count, num_bits);
1937
1938         if (cluster_bitmap)
1939                 jbd_unlock_bh_state(group_bh);
1940
1941         status = ocfs2_journal_dirty(handle, group_bh);
1942         if (status < 0)
1943                 mlog_errno(status);
1944 bail:
1945         return status;
1946 }
1947
1948 /*
1949  * expects the suballoc inode to already be locked.
1950  */
1951 int ocfs2_free_suballoc_bits(handle_t *handle,
1952                              struct inode *alloc_inode,
1953                              struct buffer_head *alloc_bh,
1954                              unsigned int start_bit,
1955                              u64 bg_blkno,
1956                              unsigned int count)
1957 {
1958         int status = 0;
1959         u32 tmp_used;
1960         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
1961         struct ocfs2_chain_list *cl = &fe->id2.i_chain;
1962         struct buffer_head *group_bh = NULL;
1963         struct ocfs2_group_desc *group;
1964
1965         mlog_entry_void();
1966
1967         /* The alloc_bh comes from ocfs2_free_dinode() or
1968          * ocfs2_free_clusters().  The callers have all locked the
1969          * allocator and gotten alloc_bh from the lock call.  This
1970          * validates the dinode buffer.  Any corruption that has happended
1971          * is a code bug. */
1972         BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
1973         BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
1974
1975         mlog(0, "%llu: freeing %u bits from group %llu, starting at %u\n",
1976              (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno, count,
1977              (unsigned long long)bg_blkno, start_bit);
1978
1979         status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
1980                                              &group_bh);
1981         if (status < 0) {
1982                 mlog_errno(status);
1983                 goto bail;
1984         }
1985         group = (struct ocfs2_group_desc *) group_bh->b_data;
1986
1987         BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
1988
1989         status = ocfs2_block_group_clear_bits(handle, alloc_inode,
1990                                               group, group_bh,
1991                                               start_bit, count);
1992         if (status < 0) {
1993                 mlog_errno(status);
1994                 goto bail;
1995         }
1996
1997         status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
1998                                          alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
1999         if (status < 0) {
2000                 mlog_errno(status);
2001                 goto bail;
2002         }
2003
2004         le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
2005                      count);
2006         tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
2007         fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
2008
2009         status = ocfs2_journal_dirty(handle, alloc_bh);
2010         if (status < 0) {
2011                 mlog_errno(status);
2012                 goto bail;
2013         }
2014
2015 bail:
2016         brelse(group_bh);
2017
2018         mlog_exit(status);
2019         return status;
2020 }
2021
2022 int ocfs2_free_dinode(handle_t *handle,
2023                       struct inode *inode_alloc_inode,
2024                       struct buffer_head *inode_alloc_bh,
2025                       struct ocfs2_dinode *di)
2026 {
2027         u64 blk = le64_to_cpu(di->i_blkno);
2028         u16 bit = le16_to_cpu(di->i_suballoc_bit);
2029         u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2030
2031         return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
2032                                         inode_alloc_bh, bit, bg_blkno, 1);
2033 }
2034
2035 int ocfs2_free_clusters(handle_t *handle,
2036                        struct inode *bitmap_inode,
2037                        struct buffer_head *bitmap_bh,
2038                        u64 start_blk,
2039                        unsigned int num_clusters)
2040 {
2041         int status;
2042         u16 bg_start_bit;
2043         u64 bg_blkno;
2044         struct ocfs2_dinode *fe;
2045
2046         /* You can't ever have a contiguous set of clusters
2047          * bigger than a block group bitmap so we never have to worry
2048          * about looping on them. */
2049
2050         mlog_entry_void();
2051
2052         /* This is expensive. We can safely remove once this stuff has
2053          * gotten tested really well. */
2054         BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
2055
2056         fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
2057
2058         ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
2059                                      &bg_start_bit);
2060
2061         mlog(0, "want to free %u clusters starting at block %llu\n",
2062              num_clusters, (unsigned long long)start_blk);
2063         mlog(0, "bg_blkno = %llu, bg_start_bit = %u\n",
2064              (unsigned long long)bg_blkno, bg_start_bit);
2065
2066         status = ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
2067                                           bg_start_bit, bg_blkno,
2068                                           num_clusters);
2069         if (status < 0) {
2070                 mlog_errno(status);
2071                 goto out;
2072         }
2073
2074         ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
2075                                          num_clusters);
2076
2077 out:
2078         mlog_exit(status);
2079         return status;
2080 }
2081
2082 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
2083 {
2084         printk("Block Group:\n");
2085         printk("bg_signature:       %s\n", bg->bg_signature);
2086         printk("bg_size:            %u\n", bg->bg_size);
2087         printk("bg_bits:            %u\n", bg->bg_bits);
2088         printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
2089         printk("bg_chain:           %u\n", bg->bg_chain);
2090         printk("bg_generation:      %u\n", le32_to_cpu(bg->bg_generation));
2091         printk("bg_next_group:      %llu\n",
2092                (unsigned long long)bg->bg_next_group);
2093         printk("bg_parent_dinode:   %llu\n",
2094                (unsigned long long)bg->bg_parent_dinode);
2095         printk("bg_blkno:           %llu\n",
2096                (unsigned long long)bg->bg_blkno);
2097 }
2098
2099 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
2100 {
2101         int i;
2102
2103         printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
2104         printk("i_signature:                  %s\n", fe->i_signature);
2105         printk("i_size:                       %llu\n",
2106                (unsigned long long)fe->i_size);
2107         printk("i_clusters:                   %u\n", fe->i_clusters);
2108         printk("i_generation:                 %u\n",
2109                le32_to_cpu(fe->i_generation));
2110         printk("id1.bitmap1.i_used:           %u\n",
2111                le32_to_cpu(fe->id1.bitmap1.i_used));
2112         printk("id1.bitmap1.i_total:          %u\n",
2113                le32_to_cpu(fe->id1.bitmap1.i_total));
2114         printk("id2.i_chain.cl_cpg:           %u\n", fe->id2.i_chain.cl_cpg);
2115         printk("id2.i_chain.cl_bpc:           %u\n", fe->id2.i_chain.cl_bpc);
2116         printk("id2.i_chain.cl_count:         %u\n", fe->id2.i_chain.cl_count);
2117         printk("id2.i_chain.cl_next_free_rec: %u\n",
2118                fe->id2.i_chain.cl_next_free_rec);
2119         for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
2120                 printk("fe->id2.i_chain.cl_recs[%d].c_free:  %u\n", i,
2121                        fe->id2.i_chain.cl_recs[i].c_free);
2122                 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
2123                        fe->id2.i_chain.cl_recs[i].c_total);
2124                 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
2125                        (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
2126         }
2127 }
2128
2129 /*
2130  * For a given allocation, determine which allocators will need to be
2131  * accessed, and lock them, reserving the appropriate number of bits.
2132  *
2133  * Sparse file systems call this from ocfs2_write_begin_nolock()
2134  * and ocfs2_allocate_unwritten_extents().
2135  *
2136  * File systems which don't support holes call this from
2137  * ocfs2_extend_allocation().
2138  */
2139 int ocfs2_lock_allocators(struct inode *inode,
2140                           struct ocfs2_extent_tree *et,
2141                           u32 clusters_to_add, u32 extents_to_split,
2142                           struct ocfs2_alloc_context **data_ac,
2143                           struct ocfs2_alloc_context **meta_ac)
2144 {
2145         int ret = 0, num_free_extents;
2146         unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
2147         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2148
2149         *meta_ac = NULL;
2150         if (data_ac)
2151                 *data_ac = NULL;
2152
2153         BUG_ON(clusters_to_add != 0 && data_ac == NULL);
2154
2155         num_free_extents = ocfs2_num_free_extents(osb, et);
2156         if (num_free_extents < 0) {
2157                 ret = num_free_extents;
2158                 mlog_errno(ret);
2159                 goto out;
2160         }
2161
2162         /*
2163          * Sparse allocation file systems need to be more conservative
2164          * with reserving room for expansion - the actual allocation
2165          * happens while we've got a journal handle open so re-taking
2166          * a cluster lock (because we ran out of room for another
2167          * extent) will violate ordering rules.
2168          *
2169          * Most of the time we'll only be seeing this 1 cluster at a time
2170          * anyway.
2171          *
2172          * Always lock for any unwritten extents - we might want to
2173          * add blocks during a split.
2174          */
2175         if (!num_free_extents ||
2176             (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
2177                 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
2178                 if (ret < 0) {
2179                         if (ret != -ENOSPC)
2180                                 mlog_errno(ret);
2181                         goto out;
2182                 }
2183         }
2184
2185         if (clusters_to_add == 0)
2186                 goto out;
2187
2188         ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2189         if (ret < 0) {
2190                 if (ret != -ENOSPC)
2191                         mlog_errno(ret);
2192                 goto out;
2193         }
2194
2195 out:
2196         if (ret) {
2197                 if (*meta_ac) {
2198                         ocfs2_free_alloc_context(*meta_ac);
2199                         *meta_ac = NULL;
2200                 }
2201
2202                 /*
2203                  * We cannot have an error and a non null *data_ac.
2204                  */
2205         }
2206
2207         return ret;
2208 }
2209
2210 /*
2211  * Read the inode specified by blkno to get suballoc_slot and
2212  * suballoc_bit.
2213  */
2214 static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
2215                                        u16 *suballoc_slot, u16 *suballoc_bit)
2216 {
2217         int status;
2218         struct buffer_head *inode_bh = NULL;
2219         struct ocfs2_dinode *inode_fe;
2220
2221         mlog_entry("blkno: %llu\n", (unsigned long long)blkno);
2222
2223         /* dirty read disk */
2224         status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
2225         if (status < 0) {
2226                 mlog(ML_ERROR, "read block %llu failed %d\n",
2227                      (unsigned long long)blkno, status);
2228                 goto bail;
2229         }
2230
2231         inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
2232         if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
2233                 mlog(ML_ERROR, "invalid inode %llu requested\n",
2234                      (unsigned long long)blkno);
2235                 status = -EINVAL;
2236                 goto bail;
2237         }
2238
2239         if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
2240             (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
2241                 mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
2242                      (unsigned long long)blkno,
2243                      (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
2244                 status = -EINVAL;
2245                 goto bail;
2246         }
2247
2248         if (suballoc_slot)
2249                 *suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
2250         if (suballoc_bit)
2251                 *suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
2252
2253 bail:
2254         brelse(inode_bh);
2255
2256         mlog_exit(status);
2257         return status;
2258 }
2259
2260 /*
2261  * test whether bit is SET in allocator bitmap or not.  on success, 0
2262  * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
2263  * is returned and *res is meaningless.  Call this after you have
2264  * cluster locked against suballoc, or you may get a result based on
2265  * non-up2date contents
2266  */
2267 static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
2268                                    struct inode *suballoc,
2269                                    struct buffer_head *alloc_bh, u64 blkno,
2270                                    u16 bit, int *res)
2271 {
2272         struct ocfs2_dinode *alloc_fe;
2273         struct ocfs2_group_desc *group;
2274         struct buffer_head *group_bh = NULL;
2275         u64 bg_blkno;
2276         int status;
2277
2278         mlog_entry("blkno: %llu bit: %u\n", (unsigned long long)blkno,
2279                    (unsigned int)bit);
2280
2281         alloc_fe = (struct ocfs2_dinode *)alloc_bh->b_data;
2282         if ((bit + 1) > ocfs2_bits_per_group(&alloc_fe->id2.i_chain)) {
2283                 mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
2284                      (unsigned int)bit,
2285                      ocfs2_bits_per_group(&alloc_fe->id2.i_chain));
2286                 status = -EINVAL;
2287                 goto bail;
2288         }
2289
2290         bg_blkno = ocfs2_which_suballoc_group(blkno, bit);
2291         status = ocfs2_read_group_descriptor(suballoc, alloc_fe, bg_blkno,
2292                                              &group_bh);
2293         if (status < 0) {
2294                 mlog(ML_ERROR, "read group %llu failed %d\n",
2295                      (unsigned long long)bg_blkno, status);
2296                 goto bail;
2297         }
2298
2299         group = (struct ocfs2_group_desc *) group_bh->b_data;
2300         *res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
2301
2302 bail:
2303         brelse(group_bh);
2304
2305         mlog_exit(status);
2306         return status;
2307 }
2308
2309 /*
2310  * Test if the bit representing this inode (blkno) is set in the
2311  * suballocator.
2312  *
2313  * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
2314  *
2315  * In the event of failure, a negative value is returned and *res is
2316  * meaningless.
2317  *
2318  * Callers must make sure to hold nfs_sync_lock to prevent
2319  * ocfs2_delete_inode() on another node from accessing the same
2320  * suballocator concurrently.
2321  */
2322 int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
2323 {
2324         int status;
2325         u16 suballoc_bit = 0, suballoc_slot = 0;
2326         struct inode *inode_alloc_inode;
2327         struct buffer_head *alloc_bh = NULL;
2328
2329         mlog_entry("blkno: %llu", (unsigned long long)blkno);
2330
2331         status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
2332                                              &suballoc_bit);
2333         if (status < 0) {
2334                 mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
2335                 goto bail;
2336         }
2337
2338         inode_alloc_inode =
2339                 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
2340                                             suballoc_slot);
2341         if (!inode_alloc_inode) {
2342                 /* the error code could be inaccurate, but we are not able to
2343                  * get the correct one. */
2344                 status = -EINVAL;
2345                 mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
2346                      (u32)suballoc_slot);
2347                 goto bail;
2348         }
2349
2350         mutex_lock(&inode_alloc_inode->i_mutex);
2351         status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
2352         if (status < 0) {
2353                 mutex_unlock(&inode_alloc_inode->i_mutex);
2354                 mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
2355                      (u32)suballoc_slot, status);
2356                 goto bail;
2357         }
2358
2359         status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
2360                                          blkno, suballoc_bit, res);
2361         if (status < 0)
2362                 mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
2363
2364         ocfs2_inode_unlock(inode_alloc_inode, 0);
2365         mutex_unlock(&inode_alloc_inode->i_mutex);
2366
2367         iput(inode_alloc_inode);
2368         brelse(alloc_bh);
2369 bail:
2370         mlog_exit(status);
2371         return status;
2372 }