7ac68cac041d9f7c0765a819cbec49c8b57a3b12
[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 "dlmglue.h"
39 #include "inode.h"
40 #include "journal.h"
41 #include "localalloc.h"
42 #include "suballoc.h"
43 #include "super.h"
44 #include "sysfile.h"
45 #include "uptodate.h"
46
47 #include "buffer_head_io.h"
48
49 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg);
50 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe);
51 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
52 static int ocfs2_block_group_fill(struct ocfs2_journal_handle *handle,
53                                   struct inode *alloc_inode,
54                                   struct buffer_head *bg_bh,
55                                   u64 group_blkno,
56                                   u16 my_chain,
57                                   struct ocfs2_chain_list *cl);
58 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
59                                    struct inode *alloc_inode,
60                                    struct buffer_head *bh);
61
62 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
63                                        struct ocfs2_alloc_context *ac);
64
65 static int ocfs2_cluster_group_search(struct inode *inode,
66                                       struct buffer_head *group_bh,
67                                       u32 bits_wanted, u32 min_bits,
68                                       u16 *bit_off, u16 *bits_found);
69 static int ocfs2_block_group_search(struct inode *inode,
70                                     struct buffer_head *group_bh,
71                                     u32 bits_wanted, u32 min_bits,
72                                     u16 *bit_off, u16 *bits_found);
73 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
74                               u32 bits_wanted,
75                               u32 min_bits,
76                               u16 *bit_off,
77                               unsigned int *num_bits,
78                               u64 *bg_blkno);
79 static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
80                                      struct ocfs2_alloc_context *ac,
81                                      u32 bits_wanted,
82                                      u32 min_bits,
83                                      u16 *bit_off,
84                                      unsigned int *num_bits,
85                                      u64 *bg_blkno);
86 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
87                                          int nr);
88 static inline int ocfs2_block_group_set_bits(struct ocfs2_journal_handle *handle,
89                                              struct inode *alloc_inode,
90                                              struct ocfs2_group_desc *bg,
91                                              struct buffer_head *group_bh,
92                                              unsigned int bit_off,
93                                              unsigned int num_bits);
94 static inline int ocfs2_block_group_clear_bits(struct ocfs2_journal_handle *handle,
95                                                struct inode *alloc_inode,
96                                                struct ocfs2_group_desc *bg,
97                                                struct buffer_head *group_bh,
98                                                unsigned int bit_off,
99                                                unsigned int num_bits);
100
101 static int ocfs2_relink_block_group(struct ocfs2_journal_handle *handle,
102                                     struct inode *alloc_inode,
103                                     struct buffer_head *fe_bh,
104                                     struct buffer_head *bg_bh,
105                                     struct buffer_head *prev_bg_bh,
106                                     u16 chain);
107 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
108                                                      u32 wanted);
109 static int ocfs2_free_suballoc_bits(struct ocfs2_journal_handle *handle,
110                                     struct inode *alloc_inode,
111                                     struct buffer_head *alloc_bh,
112                                     unsigned int start_bit,
113                                     u64 bg_blkno,
114                                     unsigned int count);
115 static inline u64 ocfs2_which_suballoc_group(u64 block,
116                                              unsigned int bit);
117 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
118                                                    u64 bg_blkno,
119                                                    u16 bg_bit_off);
120 static inline u64 ocfs2_which_cluster_group(struct inode *inode,
121                                             u32 cluster);
122 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
123                                                 u64 data_blkno,
124                                                 u64 *bg_blkno,
125                                                 u16 *bg_bit_off);
126
127 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
128 {
129         if (ac->ac_inode)
130                 iput(ac->ac_inode);
131         if (ac->ac_bh)
132                 brelse(ac->ac_bh);
133         kfree(ac);
134 }
135
136 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
137 {
138         return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
139 }
140
141 /* somewhat more expensive than our other checks, so use sparingly. */
142 static int ocfs2_check_group_descriptor(struct super_block *sb,
143                                         struct ocfs2_dinode *di,
144                                         struct ocfs2_group_desc *gd)
145 {
146         unsigned int max_bits;
147
148         if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
149                 OCFS2_RO_ON_INVALID_GROUP_DESC(sb, gd);
150                 return -EIO;
151         }
152
153         if (di->i_blkno != gd->bg_parent_dinode) {
154                 ocfs2_error(sb, "Group descriptor # %llu has bad parent "
155                             "pointer (%llu, expected %llu)",
156                             (unsigned long long)le64_to_cpu(gd->bg_blkno),
157                             (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
158                             (unsigned long long)le64_to_cpu(di->i_blkno));
159                 return -EIO;
160         }
161
162         max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
163         if (le16_to_cpu(gd->bg_bits) > max_bits) {
164                 ocfs2_error(sb, "Group descriptor # %llu has bit count of %u",
165                             (unsigned long long)le64_to_cpu(gd->bg_blkno),
166                             le16_to_cpu(gd->bg_bits));
167                 return -EIO;
168         }
169
170         if (le16_to_cpu(gd->bg_chain) >=
171             le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) {
172                 ocfs2_error(sb, "Group descriptor # %llu has bad chain %u",
173                             (unsigned long long)le64_to_cpu(gd->bg_blkno),
174                             le16_to_cpu(gd->bg_chain));
175                 return -EIO;
176         }
177
178         if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
179                 ocfs2_error(sb, "Group descriptor # %llu has bit count %u but "
180                             "claims that %u are free",
181                             (unsigned long long)le64_to_cpu(gd->bg_blkno),
182                             le16_to_cpu(gd->bg_bits),
183                             le16_to_cpu(gd->bg_free_bits_count));
184                 return -EIO;
185         }
186
187         if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
188                 ocfs2_error(sb, "Group descriptor # %llu has bit count %u but "
189                             "max bitmap bits of %u",
190                             (unsigned long long)le64_to_cpu(gd->bg_blkno),
191                             le16_to_cpu(gd->bg_bits),
192                             8 * le16_to_cpu(gd->bg_size));
193                 return -EIO;
194         }
195
196         return 0;
197 }
198
199 static int ocfs2_block_group_fill(struct ocfs2_journal_handle *handle,
200                                   struct inode *alloc_inode,
201                                   struct buffer_head *bg_bh,
202                                   u64 group_blkno,
203                                   u16 my_chain,
204                                   struct ocfs2_chain_list *cl)
205 {
206         int status = 0;
207         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
208         struct super_block * sb = alloc_inode->i_sb;
209
210         mlog_entry_void();
211
212         if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
213                 ocfs2_error(alloc_inode->i_sb, "group block (%llu) != "
214                             "b_blocknr (%llu)",
215                             (unsigned long long)group_blkno,
216                             (unsigned long long) bg_bh->b_blocknr);
217                 status = -EIO;
218                 goto bail;
219         }
220
221         status = ocfs2_journal_access(handle,
222                                       alloc_inode,
223                                       bg_bh,
224                                       OCFS2_JOURNAL_ACCESS_CREATE);
225         if (status < 0) {
226                 mlog_errno(status);
227                 goto bail;
228         }
229
230         memset(bg, 0, sb->s_blocksize);
231         strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
232         bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
233         bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb));
234         bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
235         bg->bg_chain = cpu_to_le16(my_chain);
236         bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
237         bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
238         bg->bg_blkno = cpu_to_le64(group_blkno);
239         /* set the 1st bit in the bitmap to account for the descriptor block */
240         ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
241         bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
242
243         status = ocfs2_journal_dirty(handle, bg_bh);
244         if (status < 0)
245                 mlog_errno(status);
246
247         /* There is no need to zero out or otherwise initialize the
248          * other blocks in a group - All valid FS metadata in a block
249          * group stores the superblock fs_generation value at
250          * allocation time. */
251
252 bail:
253         mlog_exit(status);
254         return status;
255 }
256
257 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
258 {
259         u16 curr, best;
260
261         best = curr = 0;
262         while (curr < le16_to_cpu(cl->cl_count)) {
263                 if (le32_to_cpu(cl->cl_recs[best].c_total) >
264                     le32_to_cpu(cl->cl_recs[curr].c_total))
265                         best = curr;
266                 curr++;
267         }
268         return best;
269 }
270
271 /*
272  * We expect the block group allocator to already be locked.
273  */
274 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
275                                    struct inode *alloc_inode,
276                                    struct buffer_head *bh)
277 {
278         int status, credits;
279         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
280         struct ocfs2_chain_list *cl;
281         struct ocfs2_alloc_context *ac = NULL;
282         struct ocfs2_journal_handle *handle = NULL;
283         u32 bit_off, num_bits;
284         u16 alloc_rec;
285         u64 bg_blkno;
286         struct buffer_head *bg_bh = NULL;
287         struct ocfs2_group_desc *bg;
288
289         BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
290
291         mlog_entry_void();
292
293         handle = ocfs2_alloc_handle(osb);
294         if (!handle) {
295                 status = -ENOMEM;
296                 mlog_errno(status);
297                 goto bail;
298         }
299
300         cl = &fe->id2.i_chain;
301         status = ocfs2_reserve_clusters(osb,
302                                         handle,
303                                         le16_to_cpu(cl->cl_cpg),
304                                         &ac);
305         if (status < 0) {
306                 if (status != -ENOSPC)
307                         mlog_errno(status);
308                 goto bail;
309         }
310
311         credits = ocfs2_calc_group_alloc_credits(osb->sb,
312                                                  le16_to_cpu(cl->cl_cpg));
313         handle = ocfs2_start_trans(osb, handle, credits);
314         if (IS_ERR(handle)) {
315                 status = PTR_ERR(handle);
316                 handle = NULL;
317                 mlog_errno(status);
318                 goto bail;
319         }
320
321         status = ocfs2_claim_clusters(osb,
322                                       handle,
323                                       ac,
324                                       le16_to_cpu(cl->cl_cpg),
325                                       &bit_off,
326                                       &num_bits);
327         if (status < 0) {
328                 if (status != -ENOSPC)
329                         mlog_errno(status);
330                 goto bail;
331         }
332
333         alloc_rec = ocfs2_find_smallest_chain(cl);
334
335         /* setup the group */
336         bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
337         mlog(0, "new descriptor, record %u, at block %llu\n",
338              alloc_rec, (unsigned long long)bg_blkno);
339
340         bg_bh = sb_getblk(osb->sb, bg_blkno);
341         if (!bg_bh) {
342                 status = -EIO;
343                 mlog_errno(status);
344                 goto bail;
345         }
346         ocfs2_set_new_buffer_uptodate(alloc_inode, bg_bh);
347
348         status = ocfs2_block_group_fill(handle,
349                                         alloc_inode,
350                                         bg_bh,
351                                         bg_blkno,
352                                         alloc_rec,
353                                         cl);
354         if (status < 0) {
355                 mlog_errno(status);
356                 goto bail;
357         }
358
359         bg = (struct ocfs2_group_desc *) bg_bh->b_data;
360
361         status = ocfs2_journal_access(handle, alloc_inode,
362                                       bh, OCFS2_JOURNAL_ACCESS_WRITE);
363         if (status < 0) {
364                 mlog_errno(status);
365                 goto bail;
366         }
367
368         le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
369                      le16_to_cpu(bg->bg_free_bits_count));
370         le32_add_cpu(&cl->cl_recs[alloc_rec].c_total, le16_to_cpu(bg->bg_bits));
371         cl->cl_recs[alloc_rec].c_blkno  = cpu_to_le64(bg_blkno);
372         if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
373                 le16_add_cpu(&cl->cl_next_free_rec, 1);
374
375         le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
376                                         le16_to_cpu(bg->bg_free_bits_count));
377         le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
378         le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
379
380         status = ocfs2_journal_dirty(handle, bh);
381         if (status < 0) {
382                 mlog_errno(status);
383                 goto bail;
384         }
385
386         spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
387         OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
388         fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
389                                              le32_to_cpu(fe->i_clusters)));
390         spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
391         i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
392         alloc_inode->i_blocks =
393                 ocfs2_align_bytes_to_sectors(i_size_read(alloc_inode));
394
395         status = 0;
396 bail:
397         if (handle)
398                 ocfs2_commit_trans(handle);
399
400         if (ac)
401                 ocfs2_free_alloc_context(ac);
402
403         if (bg_bh)
404                 brelse(bg_bh);
405
406         mlog_exit(status);
407         return status;
408 }
409
410 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
411                                        struct ocfs2_alloc_context *ac)
412 {
413         int status;
414         u32 bits_wanted = ac->ac_bits_wanted;
415         struct inode *alloc_inode = ac->ac_inode;
416         struct buffer_head *bh = NULL;
417         struct ocfs2_journal_handle *handle = ac->ac_handle;
418         struct ocfs2_dinode *fe;
419         u32 free_bits;
420
421         mlog_entry_void();
422
423         BUG_ON(handle->flags & OCFS2_HANDLE_STARTED);
424
425         ocfs2_handle_add_inode(handle, alloc_inode);
426         status = ocfs2_meta_lock(alloc_inode, handle, &bh, 1);
427         if (status < 0) {
428                 mlog_errno(status);
429                 goto bail;
430         }
431
432         fe = (struct ocfs2_dinode *) bh->b_data;
433         if (!OCFS2_IS_VALID_DINODE(fe)) {
434                 OCFS2_RO_ON_INVALID_DINODE(alloc_inode->i_sb, fe);
435                 status = -EIO;
436                 goto bail;
437         }
438         if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
439                 ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
440                             (unsigned long long)le64_to_cpu(fe->i_blkno));
441                 status = -EIO;
442                 goto bail;
443         }
444
445         free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
446                 le32_to_cpu(fe->id1.bitmap1.i_used);
447
448         if (bits_wanted > free_bits) {
449                 /* cluster bitmap never grows */
450                 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
451                         mlog(0, "Disk Full: wanted=%u, free_bits=%u\n",
452                              bits_wanted, free_bits);
453                         status = -ENOSPC;
454                         goto bail;
455                 }
456
457                 status = ocfs2_block_group_alloc(osb, alloc_inode, bh);
458                 if (status < 0) {
459                         if (status != -ENOSPC)
460                                 mlog_errno(status);
461                         goto bail;
462                 }
463                 atomic_inc(&osb->alloc_stats.bg_extends);
464
465                 /* You should never ask for this much metadata */
466                 BUG_ON(bits_wanted >
467                        (le32_to_cpu(fe->id1.bitmap1.i_total)
468                         - le32_to_cpu(fe->id1.bitmap1.i_used)));
469         }
470
471         get_bh(bh);
472         ac->ac_bh = bh;
473 bail:
474         if (bh)
475                 brelse(bh);
476
477         mlog_exit(status);
478         return status;
479 }
480
481 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
482                                struct ocfs2_journal_handle *handle,
483                                struct ocfs2_dinode *fe,
484                                struct ocfs2_alloc_context **ac)
485 {
486         int status;
487         struct inode *alloc_inode = NULL;
488
489         *ac = kcalloc(1, sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
490         if (!(*ac)) {
491                 status = -ENOMEM;
492                 mlog_errno(status);
493                 goto bail;
494         }
495
496         (*ac)->ac_bits_wanted = ocfs2_extend_meta_needed(fe);
497         (*ac)->ac_handle = handle;
498         (*ac)->ac_which = OCFS2_AC_USE_META;
499
500 #ifndef OCFS2_USE_ALL_METADATA_SUBALLOCATORS
501         alloc_inode = ocfs2_get_system_file_inode(osb,
502                                                   EXTENT_ALLOC_SYSTEM_INODE,
503                                                   0);
504 #else
505         alloc_inode = ocfs2_get_system_file_inode(osb,
506                                                   EXTENT_ALLOC_SYSTEM_INODE,
507                                                   osb->slot_num);
508 #endif
509         if (!alloc_inode) {
510                 status = -ENOMEM;
511                 mlog_errno(status);
512                 goto bail;
513         }
514
515         (*ac)->ac_inode = igrab(alloc_inode);
516         (*ac)->ac_group_search = ocfs2_block_group_search;
517
518         status = ocfs2_reserve_suballoc_bits(osb, (*ac));
519         if (status < 0) {
520                 if (status != -ENOSPC)
521                         mlog_errno(status);
522                 goto bail;
523         }
524
525         status = 0;
526 bail:
527         if ((status < 0) && *ac) {
528                 ocfs2_free_alloc_context(*ac);
529                 *ac = NULL;
530         }
531
532         if (alloc_inode)
533                 iput(alloc_inode);
534
535         mlog_exit(status);
536         return status;
537 }
538
539 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
540                             struct ocfs2_journal_handle *handle,
541                             struct ocfs2_alloc_context **ac)
542 {
543         int status;
544         struct inode *alloc_inode = NULL;
545
546         *ac = kcalloc(1, sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
547         if (!(*ac)) {
548                 status = -ENOMEM;
549                 mlog_errno(status);
550                 goto bail;
551         }
552
553         (*ac)->ac_bits_wanted = 1;
554         (*ac)->ac_handle = handle;
555         (*ac)->ac_which = OCFS2_AC_USE_INODE;
556
557         alloc_inode = ocfs2_get_system_file_inode(osb,
558                                                   INODE_ALLOC_SYSTEM_INODE,
559                                                   osb->slot_num);
560         if (!alloc_inode) {
561                 status = -ENOMEM;
562                 mlog_errno(status);
563                 goto bail;
564         }
565
566         (*ac)->ac_inode = igrab(alloc_inode);
567         (*ac)->ac_group_search = ocfs2_block_group_search;
568
569         status = ocfs2_reserve_suballoc_bits(osb, *ac);
570         if (status < 0) {
571                 if (status != -ENOSPC)
572                         mlog_errno(status);
573                 goto bail;
574         }
575
576         status = 0;
577 bail:
578         if ((status < 0) && *ac) {
579                 ocfs2_free_alloc_context(*ac);
580                 *ac = NULL;
581         }
582
583         if (alloc_inode)
584                 iput(alloc_inode);
585
586         mlog_exit(status);
587         return status;
588 }
589
590 /* local alloc code has to do the same thing, so rather than do this
591  * twice.. */
592 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
593                                       struct ocfs2_alloc_context *ac)
594 {
595         int status;
596
597         ac->ac_inode = ocfs2_get_system_file_inode(osb,
598                                                    GLOBAL_BITMAP_SYSTEM_INODE,
599                                                    OCFS2_INVALID_SLOT);
600         if (!ac->ac_inode) {
601                 status = -EINVAL;
602                 mlog(ML_ERROR, "Could not get bitmap inode!\n");
603                 goto bail;
604         }
605         ac->ac_which = OCFS2_AC_USE_MAIN;
606         ac->ac_group_search = ocfs2_cluster_group_search;
607
608         status = ocfs2_reserve_suballoc_bits(osb, ac);
609         if (status < 0 && status != -ENOSPC)
610                 mlog_errno(status);
611 bail:
612         return status;
613 }
614
615 /* Callers don't need to care which bitmap (local alloc or main) to
616  * use so we figure it out for them, but unfortunately this clutters
617  * things a bit. */
618 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
619                            struct ocfs2_journal_handle *handle,
620                            u32 bits_wanted,
621                            struct ocfs2_alloc_context **ac)
622 {
623         int status;
624
625         mlog_entry_void();
626
627         BUG_ON(!handle);
628
629         *ac = kcalloc(1, sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
630         if (!(*ac)) {
631                 status = -ENOMEM;
632                 mlog_errno(status);
633                 goto bail;
634         }
635
636         (*ac)->ac_bits_wanted = bits_wanted;
637         (*ac)->ac_handle = handle;
638
639         status = -ENOSPC;
640         if (ocfs2_alloc_should_use_local(osb, bits_wanted)) {
641                 status = ocfs2_reserve_local_alloc_bits(osb,
642                                                         handle,
643                                                         bits_wanted,
644                                                         *ac);
645                 if ((status < 0) && (status != -ENOSPC)) {
646                         mlog_errno(status);
647                         goto bail;
648                 } else if (status == -ENOSPC) {
649                         /* reserve_local_bits will return enospc with
650                          * the local alloc inode still locked, so we
651                          * can change this safely here. */
652                         mlog(0, "Disabling local alloc\n");
653                         /* We set to OCFS2_LA_DISABLED so that umount
654                          * can clean up what's left of the local
655                          * allocation */
656                         osb->local_alloc_state = OCFS2_LA_DISABLED;
657                 }
658         }
659
660         if (status == -ENOSPC) {
661                 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
662                 if (status < 0) {
663                         if (status != -ENOSPC)
664                                 mlog_errno(status);
665                         goto bail;
666                 }
667         }
668
669         status = 0;
670 bail:
671         if ((status < 0) && *ac) {
672                 ocfs2_free_alloc_context(*ac);
673                 *ac = NULL;
674         }
675
676         mlog_exit(status);
677         return status;
678 }
679
680 /*
681  * More or less lifted from ext3. I'll leave their description below:
682  *
683  * "For ext3 allocations, we must not reuse any blocks which are
684  * allocated in the bitmap buffer's "last committed data" copy.  This
685  * prevents deletes from freeing up the page for reuse until we have
686  * committed the delete transaction.
687  *
688  * If we didn't do this, then deleting something and reallocating it as
689  * data would allow the old block to be overwritten before the
690  * transaction committed (because we force data to disk before commit).
691  * This would lead to corruption if we crashed between overwriting the
692  * data and committing the delete.
693  *
694  * @@@ We may want to make this allocation behaviour conditional on
695  * data-writes at some point, and disable it for metadata allocations or
696  * sync-data inodes."
697  *
698  * Note: OCFS2 already does this differently for metadata vs data
699  * allocations, as those bitmaps are seperate and undo access is never
700  * called on a metadata group descriptor.
701  */
702 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
703                                          int nr)
704 {
705         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
706
707         if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
708                 return 0;
709         if (!buffer_jbd(bg_bh) || !bh2jh(bg_bh)->b_committed_data)
710                 return 1;
711
712         bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
713         return !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
714 }
715
716 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
717                                              struct buffer_head *bg_bh,
718                                              unsigned int bits_wanted,
719                                              unsigned int total_bits,
720                                              u16 *bit_off,
721                                              u16 *bits_found)
722 {
723         void *bitmap;
724         u16 best_offset, best_size;
725         int offset, start, found, status = 0;
726         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
727
728         if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
729                 OCFS2_RO_ON_INVALID_GROUP_DESC(osb->sb, bg);
730                 return -EIO;
731         }
732
733         found = start = best_offset = best_size = 0;
734         bitmap = bg->bg_bitmap;
735
736         while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
737                 if (offset == total_bits)
738                         break;
739
740                 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
741                         /* We found a zero, but we can't use it as it
742                          * hasn't been put to disk yet! */
743                         found = 0;
744                         start = offset + 1;
745                 } else if (offset == start) {
746                         /* we found a zero */
747                         found++;
748                         /* move start to the next bit to test */
749                         start++;
750                 } else {
751                         /* got a zero after some ones */
752                         found = 1;
753                         start = offset + 1;
754                 }
755                 if (found > best_size) {
756                         best_size = found;
757                         best_offset = start - found;
758                 }
759                 /* we got everything we needed */
760                 if (found == bits_wanted) {
761                         /* mlog(0, "Found it all!\n"); */
762                         break;
763                 }
764         }
765
766         /* XXX: I think the first clause is equivalent to the second
767          *      - jlbec */
768         if (found == bits_wanted) {
769                 *bit_off = start - found;
770                 *bits_found = found;
771         } else if (best_size) {
772                 *bit_off = best_offset;
773                 *bits_found = best_size;
774         } else {
775                 status = -ENOSPC;
776                 /* No error log here -- see the comment above
777                  * ocfs2_test_bg_bit_allocatable */
778         }
779
780         return status;
781 }
782
783 static inline int ocfs2_block_group_set_bits(struct ocfs2_journal_handle *handle,
784                                              struct inode *alloc_inode,
785                                              struct ocfs2_group_desc *bg,
786                                              struct buffer_head *group_bh,
787                                              unsigned int bit_off,
788                                              unsigned int num_bits)
789 {
790         int status;
791         void *bitmap = bg->bg_bitmap;
792         int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
793
794         mlog_entry_void();
795
796         if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
797                 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, bg);
798                 status = -EIO;
799                 goto bail;
800         }
801         BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
802
803         mlog(0, "block_group_set_bits: off = %u, num = %u\n", bit_off,
804              num_bits);
805
806         if (ocfs2_is_cluster_bitmap(alloc_inode))
807                 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
808
809         status = ocfs2_journal_access(handle,
810                                       alloc_inode,
811                                       group_bh,
812                                       journal_type);
813         if (status < 0) {
814                 mlog_errno(status);
815                 goto bail;
816         }
817
818         le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
819
820         while(num_bits--)
821                 ocfs2_set_bit(bit_off++, bitmap);
822
823         status = ocfs2_journal_dirty(handle,
824                                      group_bh);
825         if (status < 0) {
826                 mlog_errno(status);
827                 goto bail;
828         }
829
830 bail:
831         mlog_exit(status);
832         return status;
833 }
834
835 /* find the one with the most empty bits */
836 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
837 {
838         u16 curr, best;
839
840         BUG_ON(!cl->cl_next_free_rec);
841
842         best = curr = 0;
843         while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
844                 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
845                     le32_to_cpu(cl->cl_recs[best].c_free))
846                         best = curr;
847                 curr++;
848         }
849
850         BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
851         return best;
852 }
853
854 static int ocfs2_relink_block_group(struct ocfs2_journal_handle *handle,
855                                     struct inode *alloc_inode,
856                                     struct buffer_head *fe_bh,
857                                     struct buffer_head *bg_bh,
858                                     struct buffer_head *prev_bg_bh,
859                                     u16 chain)
860 {
861         int status;
862         /* there is a really tiny chance the journal calls could fail,
863          * but we wouldn't want inconsistent blocks in *any* case. */
864         u64 fe_ptr, bg_ptr, prev_bg_ptr;
865         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
866         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
867         struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
868
869         if (!OCFS2_IS_VALID_DINODE(fe)) {
870                 OCFS2_RO_ON_INVALID_DINODE(alloc_inode->i_sb, fe);
871                 status = -EIO;
872                 goto out;
873         }
874         if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
875                 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, bg);
876                 status = -EIO;
877                 goto out;
878         }
879         if (!OCFS2_IS_VALID_GROUP_DESC(prev_bg)) {
880                 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, prev_bg);
881                 status = -EIO;
882                 goto out;
883         }
884
885         mlog(0, "Suballoc %llu, chain %u, move group %llu to top, prev = %llu\n",
886              (unsigned long long)fe->i_blkno, chain,
887              (unsigned long long)bg->bg_blkno,
888              (unsigned long long)prev_bg->bg_blkno);
889
890         fe_ptr = le64_to_cpu(fe->id2.i_chain.cl_recs[chain].c_blkno);
891         bg_ptr = le64_to_cpu(bg->bg_next_group);
892         prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
893
894         status = ocfs2_journal_access(handle, alloc_inode, prev_bg_bh,
895                                       OCFS2_JOURNAL_ACCESS_WRITE);
896         if (status < 0) {
897                 mlog_errno(status);
898                 goto out_rollback;
899         }
900
901         prev_bg->bg_next_group = bg->bg_next_group;
902
903         status = ocfs2_journal_dirty(handle, prev_bg_bh);
904         if (status < 0) {
905                 mlog_errno(status);
906                 goto out_rollback;
907         }
908
909         status = ocfs2_journal_access(handle, alloc_inode, bg_bh,
910                                       OCFS2_JOURNAL_ACCESS_WRITE);
911         if (status < 0) {
912                 mlog_errno(status);
913                 goto out_rollback;
914         }
915
916         bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
917
918         status = ocfs2_journal_dirty(handle, bg_bh);
919         if (status < 0) {
920                 mlog_errno(status);
921                 goto out_rollback;
922         }
923
924         status = ocfs2_journal_access(handle, alloc_inode, fe_bh,
925                                       OCFS2_JOURNAL_ACCESS_WRITE);
926         if (status < 0) {
927                 mlog_errno(status);
928                 goto out_rollback;
929         }
930
931         fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
932
933         status = ocfs2_journal_dirty(handle, fe_bh);
934         if (status < 0) {
935                 mlog_errno(status);
936                 goto out_rollback;
937         }
938
939         status = 0;
940 out_rollback:
941         if (status < 0) {
942                 fe->id2.i_chain.cl_recs[chain].c_blkno = cpu_to_le64(fe_ptr);
943                 bg->bg_next_group = cpu_to_le64(bg_ptr);
944                 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
945         }
946 out:
947         mlog_exit(status);
948         return status;
949 }
950
951 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
952                                                      u32 wanted)
953 {
954         return le16_to_cpu(bg->bg_free_bits_count) > wanted;
955 }
956
957 /* return 0 on success, -ENOSPC to keep searching and any other < 0
958  * value on error. */
959 static int ocfs2_cluster_group_search(struct inode *inode,
960                                       struct buffer_head *group_bh,
961                                       u32 bits_wanted, u32 min_bits,
962                                       u16 *bit_off, u16 *bits_found)
963 {
964         int search = -ENOSPC;
965         int ret;
966         struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
967         u16 tmp_off, tmp_found;
968         unsigned int max_bits, gd_cluster_off;
969
970         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
971
972         if (gd->bg_free_bits_count) {
973                 max_bits = le16_to_cpu(gd->bg_bits);
974
975                 /* Tail groups in cluster bitmaps which aren't cpg
976                  * aligned are prone to partial extention by a failed
977                  * fs resize. If the file system resize never got to
978                  * update the dinode cluster count, then we don't want
979                  * to trust any clusters past it, regardless of what
980                  * the group descriptor says. */
981                 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
982                                                           le64_to_cpu(gd->bg_blkno));
983                 if ((gd_cluster_off + max_bits) >
984                     OCFS2_I(inode)->ip_clusters) {
985                         max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
986                         mlog(0, "Desc %llu, bg_bits %u, clusters %u, use %u\n",
987                              (unsigned long long)le64_to_cpu(gd->bg_blkno),
988                              le16_to_cpu(gd->bg_bits),
989                              OCFS2_I(inode)->ip_clusters, max_bits);
990                 }
991
992                 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
993                                                         group_bh, bits_wanted,
994                                                         max_bits,
995                                                         &tmp_off, &tmp_found);
996                 if (ret)
997                         return ret;
998
999                 /* ocfs2_block_group_find_clear_bits() might
1000                  * return success, but we still want to return
1001                  * -ENOSPC unless it found the minimum number
1002                  * of bits. */
1003                 if (min_bits <= tmp_found) {
1004                         *bit_off = tmp_off;
1005                         *bits_found = tmp_found;
1006                         search = 0; /* success */
1007                 }
1008         }
1009
1010         return search;
1011 }
1012
1013 static int ocfs2_block_group_search(struct inode *inode,
1014                                     struct buffer_head *group_bh,
1015                                     u32 bits_wanted, u32 min_bits,
1016                                     u16 *bit_off, u16 *bits_found)
1017 {
1018         int ret = -ENOSPC;
1019         struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1020
1021         BUG_ON(min_bits != 1);
1022         BUG_ON(ocfs2_is_cluster_bitmap(inode));
1023
1024         if (bg->bg_free_bits_count)
1025                 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1026                                                         group_bh, bits_wanted,
1027                                                         le16_to_cpu(bg->bg_bits),
1028                                                         bit_off, bits_found);
1029
1030         return ret;
1031 }
1032
1033 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1034                               u32 bits_wanted,
1035                               u32 min_bits,
1036                               u16 *bit_off,
1037                               unsigned int *num_bits,
1038                               u64 *bg_blkno)
1039 {
1040         int status;
1041         u16 chain, tmp_bits;
1042         u32 tmp_used;
1043         u64 next_group;
1044         struct ocfs2_journal_handle *handle = ac->ac_handle;
1045         struct inode *alloc_inode = ac->ac_inode;
1046         struct buffer_head *group_bh = NULL;
1047         struct buffer_head *prev_group_bh = NULL;
1048         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1049         struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1050         struct ocfs2_group_desc *bg;
1051
1052         chain = ac->ac_chain;
1053         mlog(0, "trying to alloc %u bits from chain %u, inode %llu\n",
1054              bits_wanted, chain,
1055              (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno);
1056
1057         status = ocfs2_read_block(OCFS2_SB(alloc_inode->i_sb),
1058                                   le64_to_cpu(cl->cl_recs[chain].c_blkno),
1059                                   &group_bh, OCFS2_BH_CACHED, alloc_inode);
1060         if (status < 0) {
1061                 mlog_errno(status);
1062                 goto bail;
1063         }
1064         bg = (struct ocfs2_group_desc *) group_bh->b_data;
1065         status = ocfs2_check_group_descriptor(alloc_inode->i_sb, fe, bg);
1066         if (status) {
1067                 mlog_errno(status);
1068                 goto bail;
1069         }
1070
1071         status = -ENOSPC;
1072         /* for now, the chain search is a bit simplistic. We just use
1073          * the 1st group with any empty bits. */
1074         while ((status = ac->ac_group_search(alloc_inode, group_bh,
1075                                              bits_wanted, min_bits, bit_off,
1076                                              &tmp_bits)) == -ENOSPC) {
1077                 if (!bg->bg_next_group)
1078                         break;
1079
1080                 if (prev_group_bh) {
1081                         brelse(prev_group_bh);
1082                         prev_group_bh = NULL;
1083                 }
1084                 next_group = le64_to_cpu(bg->bg_next_group);
1085                 prev_group_bh = group_bh;
1086                 group_bh = NULL;
1087                 status = ocfs2_read_block(OCFS2_SB(alloc_inode->i_sb),
1088                                           next_group, &group_bh,
1089                                           OCFS2_BH_CACHED, alloc_inode);
1090                 if (status < 0) {
1091                         mlog_errno(status);
1092                         goto bail;
1093                 }
1094                 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1095                 status = ocfs2_check_group_descriptor(alloc_inode->i_sb, fe, bg);
1096                 if (status) {
1097                         mlog_errno(status);
1098                         goto bail;
1099                 }
1100         }
1101         if (status < 0) {
1102                 if (status != -ENOSPC)
1103                         mlog_errno(status);
1104                 goto bail;
1105         }
1106
1107         mlog(0, "alloc succeeds: we give %u bits from block group %llu\n",
1108              tmp_bits, (unsigned long long)bg->bg_blkno);
1109
1110         *num_bits = tmp_bits;
1111
1112         BUG_ON(*num_bits == 0);
1113
1114         /*
1115          * Keep track of previous block descriptor read. When
1116          * we find a target, if we have read more than X
1117          * number of descriptors, and the target is reasonably
1118          * empty, relink him to top of his chain.
1119          *
1120          * We've read 0 extra blocks and only send one more to
1121          * the transaction, yet the next guy to search has a
1122          * much easier time.
1123          *
1124          * Do this *after* figuring out how many bits we're taking out
1125          * of our target group.
1126          */
1127         if (ac->ac_allow_chain_relink &&
1128             (prev_group_bh) &&
1129             (ocfs2_block_group_reasonably_empty(bg, *num_bits))) {
1130                 status = ocfs2_relink_block_group(handle, alloc_inode,
1131                                                   ac->ac_bh, group_bh,
1132                                                   prev_group_bh, chain);
1133                 if (status < 0) {
1134                         mlog_errno(status);
1135                         goto bail;
1136                 }
1137         }
1138
1139         /* Ok, claim our bits now: set the info on dinode, chainlist
1140          * and then the group */
1141         status = ocfs2_journal_access(handle,
1142                                       alloc_inode,
1143                                       ac->ac_bh,
1144                                       OCFS2_JOURNAL_ACCESS_WRITE);
1145         if (status < 0) {
1146                 mlog_errno(status);
1147                 goto bail;
1148         }
1149
1150         tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1151         fe->id1.bitmap1.i_used = cpu_to_le32(*num_bits + tmp_used);
1152         le32_add_cpu(&cl->cl_recs[chain].c_free, -(*num_bits));
1153
1154         status = ocfs2_journal_dirty(handle,
1155                                      ac->ac_bh);
1156         if (status < 0) {
1157                 mlog_errno(status);
1158                 goto bail;
1159         }
1160
1161         status = ocfs2_block_group_set_bits(handle,
1162                                             alloc_inode,
1163                                             bg,
1164                                             group_bh,
1165                                             *bit_off,
1166                                             *num_bits);
1167         if (status < 0) {
1168                 mlog_errno(status);
1169                 goto bail;
1170         }
1171
1172         mlog(0, "Allocated %u bits from suballocator %llu\n", *num_bits,
1173              (unsigned long long)fe->i_blkno);
1174
1175         *bg_blkno = le64_to_cpu(bg->bg_blkno);
1176 bail:
1177         if (group_bh)
1178                 brelse(group_bh);
1179         if (prev_group_bh)
1180                 brelse(prev_group_bh);
1181
1182         mlog_exit(status);
1183         return status;
1184 }
1185
1186 /* will give out up to bits_wanted contiguous bits. */
1187 static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
1188                                      struct ocfs2_alloc_context *ac,
1189                                      u32 bits_wanted,
1190                                      u32 min_bits,
1191                                      u16 *bit_off,
1192                                      unsigned int *num_bits,
1193                                      u64 *bg_blkno)
1194 {
1195         int status;
1196         u16 victim, i;
1197         struct ocfs2_chain_list *cl;
1198         struct ocfs2_dinode *fe;
1199
1200         mlog_entry_void();
1201
1202         BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1203         BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1204         BUG_ON(!ac->ac_bh);
1205
1206         fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1207         if (!OCFS2_IS_VALID_DINODE(fe)) {
1208                 OCFS2_RO_ON_INVALID_DINODE(osb->sb, fe);
1209                 status = -EIO;
1210                 goto bail;
1211         }
1212         if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1213             le32_to_cpu(fe->id1.bitmap1.i_total)) {
1214                 ocfs2_error(osb->sb, "Chain allocator dinode %llu has %u used "
1215                             "bits but only %u total.",
1216                             (unsigned long long)le64_to_cpu(fe->i_blkno),
1217                             le32_to_cpu(fe->id1.bitmap1.i_used),
1218                             le32_to_cpu(fe->id1.bitmap1.i_total));
1219                 status = -EIO;
1220                 goto bail;
1221         }
1222
1223         cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1224
1225         victim = ocfs2_find_victim_chain(cl);
1226         ac->ac_chain = victim;
1227         ac->ac_allow_chain_relink = 1;
1228
1229         status = ocfs2_search_chain(ac, bits_wanted, min_bits, bit_off,
1230                                     num_bits, bg_blkno);
1231         if (!status)
1232                 goto bail;
1233         if (status < 0 && status != -ENOSPC) {
1234                 mlog_errno(status);
1235                 goto bail;
1236         }
1237
1238         mlog(0, "Search of victim chain %u came up with nothing, "
1239              "trying all chains now.\n", victim);
1240
1241         /* If we didn't pick a good victim, then just default to
1242          * searching each chain in order. Don't allow chain relinking
1243          * because we only calculate enough journal credits for one
1244          * relink per alloc. */
1245         ac->ac_allow_chain_relink = 0;
1246         for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1247                 if (i == victim)
1248                         continue;
1249                 if (!cl->cl_recs[i].c_free)
1250                         continue;
1251
1252                 ac->ac_chain = i;
1253                 status = ocfs2_search_chain(ac, bits_wanted, min_bits,
1254                                             bit_off, num_bits,
1255                                             bg_blkno);
1256                 if (!status)
1257                         break;
1258                 if (status < 0 && status != -ENOSPC) {
1259                         mlog_errno(status);
1260                         goto bail;
1261                 }
1262         }
1263 bail:
1264
1265         mlog_exit(status);
1266         return status;
1267 }
1268
1269 int ocfs2_claim_metadata(struct ocfs2_super *osb,
1270                          struct ocfs2_journal_handle *handle,
1271                          struct ocfs2_alloc_context *ac,
1272                          u32 bits_wanted,
1273                          u16 *suballoc_bit_start,
1274                          unsigned int *num_bits,
1275                          u64 *blkno_start)
1276 {
1277         int status;
1278         u64 bg_blkno;
1279
1280         BUG_ON(!ac);
1281         BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1282         BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
1283         BUG_ON(ac->ac_handle != handle);
1284
1285         status = ocfs2_claim_suballoc_bits(osb,
1286                                            ac,
1287                                            bits_wanted,
1288                                            1,
1289                                            suballoc_bit_start,
1290                                            num_bits,
1291                                            &bg_blkno);
1292         if (status < 0) {
1293                 mlog_errno(status);
1294                 goto bail;
1295         }
1296         atomic_inc(&osb->alloc_stats.bg_allocs);
1297
1298         *blkno_start = bg_blkno + (u64) *suballoc_bit_start;
1299         ac->ac_bits_given += (*num_bits);
1300         status = 0;
1301 bail:
1302         mlog_exit(status);
1303         return status;
1304 }
1305
1306 int ocfs2_claim_new_inode(struct ocfs2_super *osb,
1307                           struct ocfs2_journal_handle *handle,
1308                           struct ocfs2_alloc_context *ac,
1309                           u16 *suballoc_bit,
1310                           u64 *fe_blkno)
1311 {
1312         int status;
1313         unsigned int num_bits;
1314         u64 bg_blkno;
1315
1316         mlog_entry_void();
1317
1318         BUG_ON(!ac);
1319         BUG_ON(ac->ac_bits_given != 0);
1320         BUG_ON(ac->ac_bits_wanted != 1);
1321         BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
1322         BUG_ON(ac->ac_handle != handle);
1323
1324         status = ocfs2_claim_suballoc_bits(osb,
1325                                            ac,
1326                                            1,
1327                                            1,
1328                                            suballoc_bit,
1329                                            &num_bits,
1330                                            &bg_blkno);
1331         if (status < 0) {
1332                 mlog_errno(status);
1333                 goto bail;
1334         }
1335         atomic_inc(&osb->alloc_stats.bg_allocs);
1336
1337         BUG_ON(num_bits != 1);
1338
1339         *fe_blkno = bg_blkno + (u64) (*suballoc_bit);
1340         ac->ac_bits_given++;
1341         status = 0;
1342 bail:
1343         mlog_exit(status);
1344         return status;
1345 }
1346
1347 /* translate a group desc. blkno and it's bitmap offset into
1348  * disk cluster offset. */
1349 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
1350                                                    u64 bg_blkno,
1351                                                    u16 bg_bit_off)
1352 {
1353         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1354         u32 cluster = 0;
1355
1356         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1357
1358         if (bg_blkno != osb->first_cluster_group_blkno)
1359                 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
1360         cluster += (u32) bg_bit_off;
1361         return cluster;
1362 }
1363
1364 /* given a cluster offset, calculate which block group it belongs to
1365  * and return that block offset. */
1366 static inline u64 ocfs2_which_cluster_group(struct inode *inode,
1367                                             u32 cluster)
1368 {
1369         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1370         u32 group_no;
1371
1372         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1373
1374         group_no = cluster / osb->bitmap_cpg;
1375         if (!group_no)
1376                 return osb->first_cluster_group_blkno;
1377         return ocfs2_clusters_to_blocks(inode->i_sb,
1378                                         group_no * osb->bitmap_cpg);
1379 }
1380
1381 /* given the block number of a cluster start, calculate which cluster
1382  * group and descriptor bitmap offset that corresponds to. */
1383 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
1384                                                 u64 data_blkno,
1385                                                 u64 *bg_blkno,
1386                                                 u16 *bg_bit_off)
1387 {
1388         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1389         u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
1390
1391         BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1392
1393         *bg_blkno = ocfs2_which_cluster_group(inode,
1394                                               data_cluster);
1395
1396         if (*bg_blkno == osb->first_cluster_group_blkno)
1397                 *bg_bit_off = (u16) data_cluster;
1398         else
1399                 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
1400                                                              data_blkno - *bg_blkno);
1401 }
1402
1403 /*
1404  * min_bits - minimum contiguous chunk from this total allocation we
1405  * can handle. set to what we asked for originally for a full
1406  * contig. allocation, set to '1' to indicate we can deal with extents
1407  * of any size.
1408  */
1409 int ocfs2_claim_clusters(struct ocfs2_super *osb,
1410                          struct ocfs2_journal_handle *handle,
1411                          struct ocfs2_alloc_context *ac,
1412                          u32 min_clusters,
1413                          u32 *cluster_start,
1414                          u32 *num_clusters)
1415 {
1416         int status;
1417         unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
1418         u64 bg_blkno;
1419         u16 bg_bit_off;
1420
1421         mlog_entry_void();
1422
1423         BUG_ON(!ac);
1424         BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1425
1426         BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
1427                && ac->ac_which != OCFS2_AC_USE_MAIN);
1428         BUG_ON(ac->ac_handle != handle);
1429
1430         if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
1431                 status = ocfs2_claim_local_alloc_bits(osb,
1432                                                       handle,
1433                                                       ac,
1434                                                       bits_wanted,
1435                                                       cluster_start,
1436                                                       num_clusters);
1437                 if (!status)
1438                         atomic_inc(&osb->alloc_stats.local_data);
1439         } else {
1440                 if (min_clusters > (osb->bitmap_cpg - 1)) {
1441                         /* The only paths asking for contiguousness
1442                          * should know about this already. */
1443                         mlog(ML_ERROR, "minimum allocation requested exceeds "
1444                                        "group bitmap size!");
1445                         status = -ENOSPC;
1446                         goto bail;
1447                 }
1448                 /* clamp the current request down to a realistic size. */
1449                 if (bits_wanted > (osb->bitmap_cpg - 1))
1450                         bits_wanted = osb->bitmap_cpg - 1;
1451
1452                 status = ocfs2_claim_suballoc_bits(osb,
1453                                                    ac,
1454                                                    bits_wanted,
1455                                                    min_clusters,
1456                                                    &bg_bit_off,
1457                                                    num_clusters,
1458                                                    &bg_blkno);
1459                 if (!status) {
1460                         *cluster_start =
1461                                 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
1462                                                                  bg_blkno,
1463                                                                  bg_bit_off);
1464                         atomic_inc(&osb->alloc_stats.bitmap_data);
1465                 }
1466         }
1467         if (status < 0) {
1468                 if (status != -ENOSPC)
1469                         mlog_errno(status);
1470                 goto bail;
1471         }
1472
1473         ac->ac_bits_given += *num_clusters;
1474
1475 bail:
1476         mlog_exit(status);
1477         return status;
1478 }
1479
1480 static inline int ocfs2_block_group_clear_bits(struct ocfs2_journal_handle *handle,
1481                                                struct inode *alloc_inode,
1482                                                struct ocfs2_group_desc *bg,
1483                                                struct buffer_head *group_bh,
1484                                                unsigned int bit_off,
1485                                                unsigned int num_bits)
1486 {
1487         int status;
1488         unsigned int tmp;
1489         int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1490         struct ocfs2_group_desc *undo_bg = NULL;
1491
1492         mlog_entry_void();
1493
1494         if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
1495                 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, bg);
1496                 status = -EIO;
1497                 goto bail;
1498         }
1499
1500         mlog(0, "off = %u, num = %u\n", bit_off, num_bits);
1501
1502         if (ocfs2_is_cluster_bitmap(alloc_inode))
1503                 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1504
1505         status = ocfs2_journal_access(handle, alloc_inode, group_bh,
1506                                       journal_type);
1507         if (status < 0) {
1508                 mlog_errno(status);
1509                 goto bail;
1510         }
1511
1512         if (ocfs2_is_cluster_bitmap(alloc_inode))
1513                 undo_bg = (struct ocfs2_group_desc *) bh2jh(group_bh)->b_committed_data;
1514
1515         tmp = num_bits;
1516         while(tmp--) {
1517                 ocfs2_clear_bit((bit_off + tmp),
1518                                 (unsigned long *) bg->bg_bitmap);
1519                 if (ocfs2_is_cluster_bitmap(alloc_inode))
1520                         ocfs2_set_bit(bit_off + tmp,
1521                                       (unsigned long *) undo_bg->bg_bitmap);
1522         }
1523         le16_add_cpu(&bg->bg_free_bits_count, num_bits);
1524
1525         status = ocfs2_journal_dirty(handle, group_bh);
1526         if (status < 0)
1527                 mlog_errno(status);
1528 bail:
1529         return status;
1530 }
1531
1532 /*
1533  * expects the suballoc inode to already be locked.
1534  */
1535 static int ocfs2_free_suballoc_bits(struct ocfs2_journal_handle *handle,
1536                                     struct inode *alloc_inode,
1537                                     struct buffer_head *alloc_bh,
1538                                     unsigned int start_bit,
1539                                     u64 bg_blkno,
1540                                     unsigned int count)
1541 {
1542         int status = 0;
1543         u32 tmp_used;
1544         struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
1545         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
1546         struct ocfs2_chain_list *cl = &fe->id2.i_chain;
1547         struct buffer_head *group_bh = NULL;
1548         struct ocfs2_group_desc *group;
1549
1550         mlog_entry_void();
1551
1552         if (!OCFS2_IS_VALID_DINODE(fe)) {
1553                 OCFS2_RO_ON_INVALID_DINODE(alloc_inode->i_sb, fe);
1554                 status = -EIO;
1555                 goto bail;
1556         }
1557         BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
1558
1559         mlog(0, "%llu: freeing %u bits from group %llu, starting at %u\n",
1560              (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno, count,
1561              (unsigned long long)bg_blkno, start_bit);
1562
1563         status = ocfs2_read_block(osb, bg_blkno, &group_bh, OCFS2_BH_CACHED,
1564                                   alloc_inode);
1565         if (status < 0) {
1566                 mlog_errno(status);
1567                 goto bail;
1568         }
1569
1570         group = (struct ocfs2_group_desc *) group_bh->b_data;
1571         status = ocfs2_check_group_descriptor(alloc_inode->i_sb, fe, group);
1572         if (status) {
1573                 mlog_errno(status);
1574                 goto bail;
1575         }
1576         BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
1577
1578         status = ocfs2_block_group_clear_bits(handle, alloc_inode,
1579                                               group, group_bh,
1580                                               start_bit, count);
1581         if (status < 0) {
1582                 mlog_errno(status);
1583                 goto bail;
1584         }
1585
1586         status = ocfs2_journal_access(handle, alloc_inode, alloc_bh,
1587                                       OCFS2_JOURNAL_ACCESS_WRITE);
1588         if (status < 0) {
1589                 mlog_errno(status);
1590                 goto bail;
1591         }
1592
1593         le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
1594                      count);
1595         tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1596         fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
1597
1598         status = ocfs2_journal_dirty(handle, alloc_bh);
1599         if (status < 0) {
1600                 mlog_errno(status);
1601                 goto bail;
1602         }
1603
1604 bail:
1605         if (group_bh)
1606                 brelse(group_bh);
1607
1608         mlog_exit(status);
1609         return status;
1610 }
1611
1612 static inline u64 ocfs2_which_suballoc_group(u64 block, unsigned int bit)
1613 {
1614         u64 group = block - (u64) bit;
1615
1616         return group;
1617 }
1618
1619 int ocfs2_free_dinode(struct ocfs2_journal_handle *handle,
1620                       struct inode *inode_alloc_inode,
1621                       struct buffer_head *inode_alloc_bh,
1622                       struct ocfs2_dinode *di)
1623 {
1624         u64 blk = le64_to_cpu(di->i_blkno);
1625         u16 bit = le16_to_cpu(di->i_suballoc_bit);
1626         u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1627
1628         return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
1629                                         inode_alloc_bh, bit, bg_blkno, 1);
1630 }
1631
1632 int ocfs2_free_extent_block(struct ocfs2_journal_handle *handle,
1633                             struct inode *eb_alloc_inode,
1634                             struct buffer_head *eb_alloc_bh,
1635                             struct ocfs2_extent_block *eb)
1636 {
1637         u64 blk = le64_to_cpu(eb->h_blkno);
1638         u16 bit = le16_to_cpu(eb->h_suballoc_bit);
1639         u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1640
1641         return ocfs2_free_suballoc_bits(handle, eb_alloc_inode, eb_alloc_bh,
1642                                         bit, bg_blkno, 1);
1643 }
1644
1645 int ocfs2_free_clusters(struct ocfs2_journal_handle *handle,
1646                        struct inode *bitmap_inode,
1647                        struct buffer_head *bitmap_bh,
1648                        u64 start_blk,
1649                        unsigned int num_clusters)
1650 {
1651         int status;
1652         u16 bg_start_bit;
1653         u64 bg_blkno;
1654         struct ocfs2_dinode *fe;
1655
1656         /* You can't ever have a contiguous set of clusters
1657          * bigger than a block group bitmap so we never have to worry
1658          * about looping on them. */
1659
1660         mlog_entry_void();
1661
1662         /* This is expensive. We can safely remove once this stuff has
1663          * gotten tested really well. */
1664         BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
1665
1666         fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
1667
1668         ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
1669                                      &bg_start_bit);
1670
1671         mlog(0, "want to free %u clusters starting at block %llu\n",
1672              num_clusters, (unsigned long long)start_blk);
1673         mlog(0, "bg_blkno = %llu, bg_start_bit = %u\n",
1674              (unsigned long long)bg_blkno, bg_start_bit);
1675
1676         status = ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
1677                                           bg_start_bit, bg_blkno,
1678                                           num_clusters);
1679         if (status < 0)
1680                 mlog_errno(status);
1681
1682         mlog_exit(status);
1683         return status;
1684 }
1685
1686 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
1687 {
1688         printk("Block Group:\n");
1689         printk("bg_signature:       %s\n", bg->bg_signature);
1690         printk("bg_size:            %u\n", bg->bg_size);
1691         printk("bg_bits:            %u\n", bg->bg_bits);
1692         printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
1693         printk("bg_chain:           %u\n", bg->bg_chain);
1694         printk("bg_generation:      %u\n", le32_to_cpu(bg->bg_generation));
1695         printk("bg_next_group:      %llu\n",
1696                (unsigned long long)bg->bg_next_group);
1697         printk("bg_parent_dinode:   %llu\n",
1698                (unsigned long long)bg->bg_parent_dinode);
1699         printk("bg_blkno:           %llu\n",
1700                (unsigned long long)bg->bg_blkno);
1701 }
1702
1703 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
1704 {
1705         int i;
1706
1707         printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
1708         printk("i_signature:                  %s\n", fe->i_signature);
1709         printk("i_size:                       %llu\n",
1710                (unsigned long long)fe->i_size);
1711         printk("i_clusters:                   %u\n", fe->i_clusters);
1712         printk("i_generation:                 %u\n",
1713                le32_to_cpu(fe->i_generation));
1714         printk("id1.bitmap1.i_used:           %u\n",
1715                le32_to_cpu(fe->id1.bitmap1.i_used));
1716         printk("id1.bitmap1.i_total:          %u\n",
1717                le32_to_cpu(fe->id1.bitmap1.i_total));
1718         printk("id2.i_chain.cl_cpg:           %u\n", fe->id2.i_chain.cl_cpg);
1719         printk("id2.i_chain.cl_bpc:           %u\n", fe->id2.i_chain.cl_bpc);
1720         printk("id2.i_chain.cl_count:         %u\n", fe->id2.i_chain.cl_count);
1721         printk("id2.i_chain.cl_next_free_rec: %u\n",
1722                fe->id2.i_chain.cl_next_free_rec);
1723         for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
1724                 printk("fe->id2.i_chain.cl_recs[%d].c_free:  %u\n", i,
1725                        fe->id2.i_chain.cl_recs[i].c_free);
1726                 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
1727                        fe->id2.i_chain.cl_recs[i].c_total);
1728                 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
1729                        (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
1730         }
1731 }