Linux 2.6.28.1
[linux/fpc-iii.git] / fs / ocfs2 / suballoc.c
blobc5ff18b46b57524884d9c4b7639f8089e7d3c532
1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * suballoc.c
6 * metadata alloc and free
7 * Inspired by ext3 block groups.
9 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
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.
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.
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.
27 #include <linux/fs.h>
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
32 #define MLOG_MASK_PREFIX ML_DISK_ALLOC
33 #include <cluster/masklog.h>
35 #include "ocfs2.h"
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"
47 #include "buffer_head_io.h"
49 #define NOT_ALLOC_NEW_GROUP 0
50 #define ALLOC_NEW_GROUP 1
52 #define OCFS2_MAX_INODES_TO_STEAL 1024
54 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg);
55 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe);
56 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
57 static int ocfs2_block_group_fill(handle_t *handle,
58 struct inode *alloc_inode,
59 struct buffer_head *bg_bh,
60 u64 group_blkno,
61 u16 my_chain,
62 struct ocfs2_chain_list *cl);
63 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
64 struct inode *alloc_inode,
65 struct buffer_head *bh,
66 u64 max_block);
68 static int ocfs2_cluster_group_search(struct inode *inode,
69 struct buffer_head *group_bh,
70 u32 bits_wanted, u32 min_bits,
71 u64 max_block,
72 u16 *bit_off, u16 *bits_found);
73 static int ocfs2_block_group_search(struct inode *inode,
74 struct buffer_head *group_bh,
75 u32 bits_wanted, u32 min_bits,
76 u64 max_block,
77 u16 *bit_off, u16 *bits_found);
78 static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
79 struct ocfs2_alloc_context *ac,
80 handle_t *handle,
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(handle_t *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(handle_t *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);
101 static int ocfs2_relink_block_group(handle_t *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 inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
110 u64 bg_blkno,
111 u16 bg_bit_off);
112 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
113 u64 data_blkno,
114 u64 *bg_blkno,
115 u16 *bg_bit_off);
116 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
117 u32 bits_wanted, u64 max_block,
118 struct ocfs2_alloc_context **ac);
120 void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
122 struct inode *inode = ac->ac_inode;
124 if (inode) {
125 if (ac->ac_which != OCFS2_AC_USE_LOCAL)
126 ocfs2_inode_unlock(inode, 1);
128 mutex_unlock(&inode->i_mutex);
130 iput(inode);
131 ac->ac_inode = NULL;
133 brelse(ac->ac_bh);
134 ac->ac_bh = NULL;
137 void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
139 ocfs2_free_ac_resource(ac);
140 kfree(ac);
143 static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
145 return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
148 /* somewhat more expensive than our other checks, so use sparingly. */
149 int ocfs2_check_group_descriptor(struct super_block *sb,
150 struct ocfs2_dinode *di,
151 struct ocfs2_group_desc *gd)
153 unsigned int max_bits;
155 if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
156 OCFS2_RO_ON_INVALID_GROUP_DESC(sb, gd);
157 return -EIO;
160 if (di->i_blkno != gd->bg_parent_dinode) {
161 ocfs2_error(sb, "Group descriptor # %llu has bad parent "
162 "pointer (%llu, expected %llu)",
163 (unsigned long long)le64_to_cpu(gd->bg_blkno),
164 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
165 (unsigned long long)le64_to_cpu(di->i_blkno));
166 return -EIO;
169 max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
170 if (le16_to_cpu(gd->bg_bits) > max_bits) {
171 ocfs2_error(sb, "Group descriptor # %llu has bit count of %u",
172 (unsigned long long)le64_to_cpu(gd->bg_blkno),
173 le16_to_cpu(gd->bg_bits));
174 return -EIO;
177 if (le16_to_cpu(gd->bg_chain) >=
178 le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) {
179 ocfs2_error(sb, "Group descriptor # %llu has bad chain %u",
180 (unsigned long long)le64_to_cpu(gd->bg_blkno),
181 le16_to_cpu(gd->bg_chain));
182 return -EIO;
185 if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
186 ocfs2_error(sb, "Group descriptor # %llu has bit count %u but "
187 "claims that %u are free",
188 (unsigned long long)le64_to_cpu(gd->bg_blkno),
189 le16_to_cpu(gd->bg_bits),
190 le16_to_cpu(gd->bg_free_bits_count));
191 return -EIO;
194 if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
195 ocfs2_error(sb, "Group descriptor # %llu has bit count %u but "
196 "max bitmap bits of %u",
197 (unsigned long long)le64_to_cpu(gd->bg_blkno),
198 le16_to_cpu(gd->bg_bits),
199 8 * le16_to_cpu(gd->bg_size));
200 return -EIO;
203 return 0;
206 static int ocfs2_block_group_fill(handle_t *handle,
207 struct inode *alloc_inode,
208 struct buffer_head *bg_bh,
209 u64 group_blkno,
210 u16 my_chain,
211 struct ocfs2_chain_list *cl)
213 int status = 0;
214 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
215 struct super_block * sb = alloc_inode->i_sb;
217 mlog_entry_void();
219 if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
220 ocfs2_error(alloc_inode->i_sb, "group block (%llu) != "
221 "b_blocknr (%llu)",
222 (unsigned long long)group_blkno,
223 (unsigned long long) bg_bh->b_blocknr);
224 status = -EIO;
225 goto bail;
228 status = ocfs2_journal_access(handle,
229 alloc_inode,
230 bg_bh,
231 OCFS2_JOURNAL_ACCESS_CREATE);
232 if (status < 0) {
233 mlog_errno(status);
234 goto bail;
237 memset(bg, 0, sb->s_blocksize);
238 strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
239 bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
240 bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb));
241 bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
242 bg->bg_chain = cpu_to_le16(my_chain);
243 bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
244 bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
245 bg->bg_blkno = cpu_to_le64(group_blkno);
246 /* set the 1st bit in the bitmap to account for the descriptor block */
247 ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
248 bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
250 status = ocfs2_journal_dirty(handle, bg_bh);
251 if (status < 0)
252 mlog_errno(status);
254 /* There is no need to zero out or otherwise initialize the
255 * other blocks in a group - All valid FS metadata in a block
256 * group stores the superblock fs_generation value at
257 * allocation time. */
259 bail:
260 mlog_exit(status);
261 return status;
264 static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
266 u16 curr, best;
268 best = curr = 0;
269 while (curr < le16_to_cpu(cl->cl_count)) {
270 if (le32_to_cpu(cl->cl_recs[best].c_total) >
271 le32_to_cpu(cl->cl_recs[curr].c_total))
272 best = curr;
273 curr++;
275 return best;
279 * We expect the block group allocator to already be locked.
281 static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
282 struct inode *alloc_inode,
283 struct buffer_head *bh,
284 u64 max_block)
286 int status, credits;
287 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
288 struct ocfs2_chain_list *cl;
289 struct ocfs2_alloc_context *ac = NULL;
290 handle_t *handle = NULL;
291 u32 bit_off, num_bits;
292 u16 alloc_rec;
293 u64 bg_blkno;
294 struct buffer_head *bg_bh = NULL;
295 struct ocfs2_group_desc *bg;
297 BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
299 mlog_entry_void();
301 cl = &fe->id2.i_chain;
302 status = ocfs2_reserve_clusters_with_limit(osb,
303 le16_to_cpu(cl->cl_cpg),
304 max_block, &ac);
305 if (status < 0) {
306 if (status != -ENOSPC)
307 mlog_errno(status);
308 goto bail;
311 credits = ocfs2_calc_group_alloc_credits(osb->sb,
312 le16_to_cpu(cl->cl_cpg));
313 handle = ocfs2_start_trans(osb, credits);
314 if (IS_ERR(handle)) {
315 status = PTR_ERR(handle);
316 handle = NULL;
317 mlog_errno(status);
318 goto bail;
321 status = ocfs2_claim_clusters(osb,
322 handle,
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;
333 alloc_rec = ocfs2_find_smallest_chain(cl);
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);
340 bg_bh = sb_getblk(osb->sb, bg_blkno);
341 if (!bg_bh) {
342 status = -EIO;
343 mlog_errno(status);
344 goto bail;
346 ocfs2_set_new_buffer_uptodate(alloc_inode, bg_bh);
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;
359 bg = (struct ocfs2_group_desc *) bg_bh->b_data;
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;
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);
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));
380 status = ocfs2_journal_dirty(handle, bh);
381 if (status < 0) {
382 mlog_errno(status);
383 goto bail;
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 = ocfs2_inode_sector_count(alloc_inode);
394 status = 0;
395 bail:
396 if (handle)
397 ocfs2_commit_trans(osb, handle);
399 if (ac)
400 ocfs2_free_alloc_context(ac);
402 brelse(bg_bh);
404 mlog_exit(status);
405 return status;
408 static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
409 struct ocfs2_alloc_context *ac,
410 int type,
411 u32 slot,
412 int alloc_new_group)
414 int status;
415 u32 bits_wanted = ac->ac_bits_wanted;
416 struct inode *alloc_inode;
417 struct buffer_head *bh = NULL;
418 struct ocfs2_dinode *fe;
419 u32 free_bits;
421 mlog_entry_void();
423 alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
424 if (!alloc_inode) {
425 mlog_errno(-EINVAL);
426 return -EINVAL;
429 mutex_lock(&alloc_inode->i_mutex);
431 status = ocfs2_inode_lock(alloc_inode, &bh, 1);
432 if (status < 0) {
433 mutex_unlock(&alloc_inode->i_mutex);
434 iput(alloc_inode);
436 mlog_errno(status);
437 return status;
440 ac->ac_inode = alloc_inode;
441 ac->ac_alloc_slot = slot;
443 fe = (struct ocfs2_dinode *) bh->b_data;
444 if (!OCFS2_IS_VALID_DINODE(fe)) {
445 OCFS2_RO_ON_INVALID_DINODE(alloc_inode->i_sb, fe);
446 status = -EIO;
447 goto bail;
449 if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
450 ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
451 (unsigned long long)le64_to_cpu(fe->i_blkno));
452 status = -EIO;
453 goto bail;
456 free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
457 le32_to_cpu(fe->id1.bitmap1.i_used);
459 if (bits_wanted > free_bits) {
460 /* cluster bitmap never grows */
461 if (ocfs2_is_cluster_bitmap(alloc_inode)) {
462 mlog(0, "Disk Full: wanted=%u, free_bits=%u\n",
463 bits_wanted, free_bits);
464 status = -ENOSPC;
465 goto bail;
468 if (alloc_new_group != ALLOC_NEW_GROUP) {
469 mlog(0, "Alloc File %u Full: wanted=%u, free_bits=%u, "
470 "and we don't alloc a new group for it.\n",
471 slot, bits_wanted, free_bits);
472 status = -ENOSPC;
473 goto bail;
476 status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
477 ac->ac_max_block);
478 if (status < 0) {
479 if (status != -ENOSPC)
480 mlog_errno(status);
481 goto bail;
483 atomic_inc(&osb->alloc_stats.bg_extends);
485 /* You should never ask for this much metadata */
486 BUG_ON(bits_wanted >
487 (le32_to_cpu(fe->id1.bitmap1.i_total)
488 - le32_to_cpu(fe->id1.bitmap1.i_used)));
491 get_bh(bh);
492 ac->ac_bh = bh;
493 bail:
494 brelse(bh);
496 mlog_exit(status);
497 return status;
500 int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
501 int blocks,
502 struct ocfs2_alloc_context **ac)
504 int status;
505 u32 slot;
507 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
508 if (!(*ac)) {
509 status = -ENOMEM;
510 mlog_errno(status);
511 goto bail;
514 (*ac)->ac_bits_wanted = blocks;
515 (*ac)->ac_which = OCFS2_AC_USE_META;
516 slot = osb->slot_num;
517 (*ac)->ac_group_search = ocfs2_block_group_search;
519 status = ocfs2_reserve_suballoc_bits(osb, (*ac),
520 EXTENT_ALLOC_SYSTEM_INODE,
521 slot, ALLOC_NEW_GROUP);
522 if (status < 0) {
523 if (status != -ENOSPC)
524 mlog_errno(status);
525 goto bail;
528 status = 0;
529 bail:
530 if ((status < 0) && *ac) {
531 ocfs2_free_alloc_context(*ac);
532 *ac = NULL;
535 mlog_exit(status);
536 return status;
539 int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
540 struct ocfs2_extent_list *root_el,
541 struct ocfs2_alloc_context **ac)
543 return ocfs2_reserve_new_metadata_blocks(osb,
544 ocfs2_extend_meta_needed(root_el),
545 ac);
548 static int ocfs2_steal_inode_from_other_nodes(struct ocfs2_super *osb,
549 struct ocfs2_alloc_context *ac)
551 int i, status = -ENOSPC;
552 s16 slot = ocfs2_get_inode_steal_slot(osb);
554 /* Start to steal inodes from the first slot after ours. */
555 if (slot == OCFS2_INVALID_SLOT)
556 slot = osb->slot_num + 1;
558 for (i = 0; i < osb->max_slots; i++, slot++) {
559 if (slot == osb->max_slots)
560 slot = 0;
562 if (slot == osb->slot_num)
563 continue;
565 status = ocfs2_reserve_suballoc_bits(osb, ac,
566 INODE_ALLOC_SYSTEM_INODE,
567 slot, NOT_ALLOC_NEW_GROUP);
568 if (status >= 0) {
569 ocfs2_set_inode_steal_slot(osb, slot);
570 break;
573 ocfs2_free_ac_resource(ac);
576 return status;
579 int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
580 struct ocfs2_alloc_context **ac)
582 int status;
583 s16 slot = ocfs2_get_inode_steal_slot(osb);
585 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
586 if (!(*ac)) {
587 status = -ENOMEM;
588 mlog_errno(status);
589 goto bail;
592 (*ac)->ac_bits_wanted = 1;
593 (*ac)->ac_which = OCFS2_AC_USE_INODE;
595 (*ac)->ac_group_search = ocfs2_block_group_search;
598 * stat(2) can't handle i_ino > 32bits, so we tell the
599 * lower levels not to allocate us a block group past that
600 * limit. The 'inode64' mount option avoids this behavior.
602 if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
603 (*ac)->ac_max_block = (u32)~0U;
606 * slot is set when we successfully steal inode from other nodes.
607 * It is reset in 3 places:
608 * 1. when we flush the truncate log
609 * 2. when we complete local alloc recovery.
610 * 3. when we successfully allocate from our own slot.
611 * After it is set, we will go on stealing inodes until we find the
612 * need to check our slots to see whether there is some space for us.
614 if (slot != OCFS2_INVALID_SLOT &&
615 atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_INODES_TO_STEAL)
616 goto inode_steal;
618 atomic_set(&osb->s_num_inodes_stolen, 0);
619 status = ocfs2_reserve_suballoc_bits(osb, *ac,
620 INODE_ALLOC_SYSTEM_INODE,
621 osb->slot_num, ALLOC_NEW_GROUP);
622 if (status >= 0) {
623 status = 0;
626 * Some inodes must be freed by us, so try to allocate
627 * from our own next time.
629 if (slot != OCFS2_INVALID_SLOT)
630 ocfs2_init_inode_steal_slot(osb);
631 goto bail;
632 } else if (status < 0 && status != -ENOSPC) {
633 mlog_errno(status);
634 goto bail;
637 ocfs2_free_ac_resource(*ac);
639 inode_steal:
640 status = ocfs2_steal_inode_from_other_nodes(osb, *ac);
641 atomic_inc(&osb->s_num_inodes_stolen);
642 if (status < 0) {
643 if (status != -ENOSPC)
644 mlog_errno(status);
645 goto bail;
648 status = 0;
649 bail:
650 if ((status < 0) && *ac) {
651 ocfs2_free_alloc_context(*ac);
652 *ac = NULL;
655 mlog_exit(status);
656 return status;
659 /* local alloc code has to do the same thing, so rather than do this
660 * twice.. */
661 int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
662 struct ocfs2_alloc_context *ac)
664 int status;
666 ac->ac_which = OCFS2_AC_USE_MAIN;
667 ac->ac_group_search = ocfs2_cluster_group_search;
669 status = ocfs2_reserve_suballoc_bits(osb, ac,
670 GLOBAL_BITMAP_SYSTEM_INODE,
671 OCFS2_INVALID_SLOT,
672 ALLOC_NEW_GROUP);
673 if (status < 0 && status != -ENOSPC) {
674 mlog_errno(status);
675 goto bail;
678 bail:
679 return status;
682 /* Callers don't need to care which bitmap (local alloc or main) to
683 * use so we figure it out for them, but unfortunately this clutters
684 * things a bit. */
685 static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
686 u32 bits_wanted, u64 max_block,
687 struct ocfs2_alloc_context **ac)
689 int status;
691 mlog_entry_void();
693 *ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
694 if (!(*ac)) {
695 status = -ENOMEM;
696 mlog_errno(status);
697 goto bail;
700 (*ac)->ac_bits_wanted = bits_wanted;
701 (*ac)->ac_max_block = max_block;
703 status = -ENOSPC;
704 if (ocfs2_alloc_should_use_local(osb, bits_wanted)) {
705 status = ocfs2_reserve_local_alloc_bits(osb,
706 bits_wanted,
707 *ac);
708 if (status == -EFBIG) {
709 /* The local alloc window is outside ac_max_block.
710 * use the main bitmap. */
711 status = -ENOSPC;
712 } else if ((status < 0) && (status != -ENOSPC)) {
713 mlog_errno(status);
714 goto bail;
718 if (status == -ENOSPC) {
719 status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
720 if (status < 0) {
721 if (status != -ENOSPC)
722 mlog_errno(status);
723 goto bail;
727 status = 0;
728 bail:
729 if ((status < 0) && *ac) {
730 ocfs2_free_alloc_context(*ac);
731 *ac = NULL;
734 mlog_exit(status);
735 return status;
738 int ocfs2_reserve_clusters(struct ocfs2_super *osb,
739 u32 bits_wanted,
740 struct ocfs2_alloc_context **ac)
742 return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0, ac);
746 * More or less lifted from ext3. I'll leave their description below:
748 * "For ext3 allocations, we must not reuse any blocks which are
749 * allocated in the bitmap buffer's "last committed data" copy. This
750 * prevents deletes from freeing up the page for reuse until we have
751 * committed the delete transaction.
753 * If we didn't do this, then deleting something and reallocating it as
754 * data would allow the old block to be overwritten before the
755 * transaction committed (because we force data to disk before commit).
756 * This would lead to corruption if we crashed between overwriting the
757 * data and committing the delete.
759 * @@@ We may want to make this allocation behaviour conditional on
760 * data-writes at some point, and disable it for metadata allocations or
761 * sync-data inodes."
763 * Note: OCFS2 already does this differently for metadata vs data
764 * allocations, as those bitmaps are separate and undo access is never
765 * called on a metadata group descriptor.
767 static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
768 int nr)
770 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
772 if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
773 return 0;
774 if (!buffer_jbd(bg_bh) || !bh2jh(bg_bh)->b_committed_data)
775 return 1;
777 bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
778 return !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
781 static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
782 struct buffer_head *bg_bh,
783 unsigned int bits_wanted,
784 unsigned int total_bits,
785 u16 *bit_off,
786 u16 *bits_found)
788 void *bitmap;
789 u16 best_offset, best_size;
790 int offset, start, found, status = 0;
791 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
793 if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
794 OCFS2_RO_ON_INVALID_GROUP_DESC(osb->sb, bg);
795 return -EIO;
798 found = start = best_offset = best_size = 0;
799 bitmap = bg->bg_bitmap;
801 while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
802 if (offset == total_bits)
803 break;
805 if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
806 /* We found a zero, but we can't use it as it
807 * hasn't been put to disk yet! */
808 found = 0;
809 start = offset + 1;
810 } else if (offset == start) {
811 /* we found a zero */
812 found++;
813 /* move start to the next bit to test */
814 start++;
815 } else {
816 /* got a zero after some ones */
817 found = 1;
818 start = offset + 1;
820 if (found > best_size) {
821 best_size = found;
822 best_offset = start - found;
824 /* we got everything we needed */
825 if (found == bits_wanted) {
826 /* mlog(0, "Found it all!\n"); */
827 break;
831 /* XXX: I think the first clause is equivalent to the second
832 * - jlbec */
833 if (found == bits_wanted) {
834 *bit_off = start - found;
835 *bits_found = found;
836 } else if (best_size) {
837 *bit_off = best_offset;
838 *bits_found = best_size;
839 } else {
840 status = -ENOSPC;
841 /* No error log here -- see the comment above
842 * ocfs2_test_bg_bit_allocatable */
845 return status;
848 static inline int ocfs2_block_group_set_bits(handle_t *handle,
849 struct inode *alloc_inode,
850 struct ocfs2_group_desc *bg,
851 struct buffer_head *group_bh,
852 unsigned int bit_off,
853 unsigned int num_bits)
855 int status;
856 void *bitmap = bg->bg_bitmap;
857 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
859 mlog_entry_void();
861 if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
862 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, bg);
863 status = -EIO;
864 goto bail;
866 BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
868 mlog(0, "block_group_set_bits: off = %u, num = %u\n", bit_off,
869 num_bits);
871 if (ocfs2_is_cluster_bitmap(alloc_inode))
872 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
874 status = ocfs2_journal_access(handle,
875 alloc_inode,
876 group_bh,
877 journal_type);
878 if (status < 0) {
879 mlog_errno(status);
880 goto bail;
883 le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
885 while(num_bits--)
886 ocfs2_set_bit(bit_off++, bitmap);
888 status = ocfs2_journal_dirty(handle,
889 group_bh);
890 if (status < 0) {
891 mlog_errno(status);
892 goto bail;
895 bail:
896 mlog_exit(status);
897 return status;
900 /* find the one with the most empty bits */
901 static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
903 u16 curr, best;
905 BUG_ON(!cl->cl_next_free_rec);
907 best = curr = 0;
908 while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
909 if (le32_to_cpu(cl->cl_recs[curr].c_free) >
910 le32_to_cpu(cl->cl_recs[best].c_free))
911 best = curr;
912 curr++;
915 BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
916 return best;
919 static int ocfs2_relink_block_group(handle_t *handle,
920 struct inode *alloc_inode,
921 struct buffer_head *fe_bh,
922 struct buffer_head *bg_bh,
923 struct buffer_head *prev_bg_bh,
924 u16 chain)
926 int status;
927 /* there is a really tiny chance the journal calls could fail,
928 * but we wouldn't want inconsistent blocks in *any* case. */
929 u64 fe_ptr, bg_ptr, prev_bg_ptr;
930 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
931 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
932 struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
934 if (!OCFS2_IS_VALID_DINODE(fe)) {
935 OCFS2_RO_ON_INVALID_DINODE(alloc_inode->i_sb, fe);
936 status = -EIO;
937 goto out;
939 if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
940 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, bg);
941 status = -EIO;
942 goto out;
944 if (!OCFS2_IS_VALID_GROUP_DESC(prev_bg)) {
945 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, prev_bg);
946 status = -EIO;
947 goto out;
950 mlog(0, "Suballoc %llu, chain %u, move group %llu to top, prev = %llu\n",
951 (unsigned long long)le64_to_cpu(fe->i_blkno), chain,
952 (unsigned long long)le64_to_cpu(bg->bg_blkno),
953 (unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
955 fe_ptr = le64_to_cpu(fe->id2.i_chain.cl_recs[chain].c_blkno);
956 bg_ptr = le64_to_cpu(bg->bg_next_group);
957 prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
959 status = ocfs2_journal_access(handle, alloc_inode, prev_bg_bh,
960 OCFS2_JOURNAL_ACCESS_WRITE);
961 if (status < 0) {
962 mlog_errno(status);
963 goto out_rollback;
966 prev_bg->bg_next_group = bg->bg_next_group;
968 status = ocfs2_journal_dirty(handle, prev_bg_bh);
969 if (status < 0) {
970 mlog_errno(status);
971 goto out_rollback;
974 status = ocfs2_journal_access(handle, alloc_inode, bg_bh,
975 OCFS2_JOURNAL_ACCESS_WRITE);
976 if (status < 0) {
977 mlog_errno(status);
978 goto out_rollback;
981 bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
983 status = ocfs2_journal_dirty(handle, bg_bh);
984 if (status < 0) {
985 mlog_errno(status);
986 goto out_rollback;
989 status = ocfs2_journal_access(handle, alloc_inode, fe_bh,
990 OCFS2_JOURNAL_ACCESS_WRITE);
991 if (status < 0) {
992 mlog_errno(status);
993 goto out_rollback;
996 fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
998 status = ocfs2_journal_dirty(handle, fe_bh);
999 if (status < 0) {
1000 mlog_errno(status);
1001 goto out_rollback;
1004 status = 0;
1005 out_rollback:
1006 if (status < 0) {
1007 fe->id2.i_chain.cl_recs[chain].c_blkno = cpu_to_le64(fe_ptr);
1008 bg->bg_next_group = cpu_to_le64(bg_ptr);
1009 prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
1011 out:
1012 mlog_exit(status);
1013 return status;
1016 static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
1017 u32 wanted)
1019 return le16_to_cpu(bg->bg_free_bits_count) > wanted;
1022 /* return 0 on success, -ENOSPC to keep searching and any other < 0
1023 * value on error. */
1024 static int ocfs2_cluster_group_search(struct inode *inode,
1025 struct buffer_head *group_bh,
1026 u32 bits_wanted, u32 min_bits,
1027 u64 max_block,
1028 u16 *bit_off, u16 *bits_found)
1030 int search = -ENOSPC;
1031 int ret;
1032 u64 blkoff;
1033 struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
1034 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1035 u16 tmp_off, tmp_found;
1036 unsigned int max_bits, gd_cluster_off;
1038 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1040 if (gd->bg_free_bits_count) {
1041 max_bits = le16_to_cpu(gd->bg_bits);
1043 /* Tail groups in cluster bitmaps which aren't cpg
1044 * aligned are prone to partial extention by a failed
1045 * fs resize. If the file system resize never got to
1046 * update the dinode cluster count, then we don't want
1047 * to trust any clusters past it, regardless of what
1048 * the group descriptor says. */
1049 gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
1050 le64_to_cpu(gd->bg_blkno));
1051 if ((gd_cluster_off + max_bits) >
1052 OCFS2_I(inode)->ip_clusters) {
1053 max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
1054 mlog(0, "Desc %llu, bg_bits %u, clusters %u, use %u\n",
1055 (unsigned long long)le64_to_cpu(gd->bg_blkno),
1056 le16_to_cpu(gd->bg_bits),
1057 OCFS2_I(inode)->ip_clusters, max_bits);
1060 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1061 group_bh, bits_wanted,
1062 max_bits,
1063 &tmp_off, &tmp_found);
1064 if (ret)
1065 return ret;
1067 if (max_block) {
1068 blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
1069 gd_cluster_off +
1070 tmp_off + tmp_found);
1071 mlog(0, "Checking %llu against %llu\n",
1072 (unsigned long long)blkoff,
1073 (unsigned long long)max_block);
1074 if (blkoff > max_block)
1075 return -ENOSPC;
1078 /* ocfs2_block_group_find_clear_bits() might
1079 * return success, but we still want to return
1080 * -ENOSPC unless it found the minimum number
1081 * of bits. */
1082 if (min_bits <= tmp_found) {
1083 *bit_off = tmp_off;
1084 *bits_found = tmp_found;
1085 search = 0; /* success */
1086 } else if (tmp_found) {
1088 * Don't show bits which we'll be returning
1089 * for allocation to the local alloc bitmap.
1091 ocfs2_local_alloc_seen_free_bits(osb, tmp_found);
1095 return search;
1098 static int ocfs2_block_group_search(struct inode *inode,
1099 struct buffer_head *group_bh,
1100 u32 bits_wanted, u32 min_bits,
1101 u64 max_block,
1102 u16 *bit_off, u16 *bits_found)
1104 int ret = -ENOSPC;
1105 u64 blkoff;
1106 struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
1108 BUG_ON(min_bits != 1);
1109 BUG_ON(ocfs2_is_cluster_bitmap(inode));
1111 if (bg->bg_free_bits_count) {
1112 ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
1113 group_bh, bits_wanted,
1114 le16_to_cpu(bg->bg_bits),
1115 bit_off, bits_found);
1116 if (!ret && max_block) {
1117 blkoff = le64_to_cpu(bg->bg_blkno) + *bit_off +
1118 *bits_found;
1119 mlog(0, "Checking %llu against %llu\n",
1120 (unsigned long long)blkoff,
1121 (unsigned long long)max_block);
1122 if (blkoff > max_block)
1123 ret = -ENOSPC;
1127 return ret;
1130 static int ocfs2_alloc_dinode_update_counts(struct inode *inode,
1131 handle_t *handle,
1132 struct buffer_head *di_bh,
1133 u32 num_bits,
1134 u16 chain)
1136 int ret;
1137 u32 tmp_used;
1138 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
1139 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
1141 ret = ocfs2_journal_access(handle, inode, di_bh,
1142 OCFS2_JOURNAL_ACCESS_WRITE);
1143 if (ret < 0) {
1144 mlog_errno(ret);
1145 goto out;
1148 tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
1149 di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
1150 le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
1152 ret = ocfs2_journal_dirty(handle, di_bh);
1153 if (ret < 0)
1154 mlog_errno(ret);
1156 out:
1157 return ret;
1160 static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
1161 handle_t *handle,
1162 u32 bits_wanted,
1163 u32 min_bits,
1164 u16 *bit_off,
1165 unsigned int *num_bits,
1166 u64 gd_blkno,
1167 u16 *bits_left)
1169 int ret;
1170 u16 found;
1171 struct buffer_head *group_bh = NULL;
1172 struct ocfs2_group_desc *gd;
1173 struct inode *alloc_inode = ac->ac_inode;
1175 ret = ocfs2_read_block(alloc_inode, gd_blkno, &group_bh);
1176 if (ret < 0) {
1177 mlog_errno(ret);
1178 return ret;
1181 gd = (struct ocfs2_group_desc *) group_bh->b_data;
1182 if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
1183 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, gd);
1184 ret = -EIO;
1185 goto out;
1188 ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
1189 ac->ac_max_block, bit_off, &found);
1190 if (ret < 0) {
1191 if (ret != -ENOSPC)
1192 mlog_errno(ret);
1193 goto out;
1196 *num_bits = found;
1198 ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
1199 *num_bits,
1200 le16_to_cpu(gd->bg_chain));
1201 if (ret < 0) {
1202 mlog_errno(ret);
1203 goto out;
1206 ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
1207 *bit_off, *num_bits);
1208 if (ret < 0)
1209 mlog_errno(ret);
1211 *bits_left = le16_to_cpu(gd->bg_free_bits_count);
1213 out:
1214 brelse(group_bh);
1216 return ret;
1219 static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
1220 handle_t *handle,
1221 u32 bits_wanted,
1222 u32 min_bits,
1223 u16 *bit_off,
1224 unsigned int *num_bits,
1225 u64 *bg_blkno,
1226 u16 *bits_left)
1228 int status;
1229 u16 chain, tmp_bits;
1230 u32 tmp_used;
1231 u64 next_group;
1232 struct inode *alloc_inode = ac->ac_inode;
1233 struct buffer_head *group_bh = NULL;
1234 struct buffer_head *prev_group_bh = NULL;
1235 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1236 struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1237 struct ocfs2_group_desc *bg;
1239 chain = ac->ac_chain;
1240 mlog(0, "trying to alloc %u bits from chain %u, inode %llu\n",
1241 bits_wanted, chain,
1242 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno);
1244 status = ocfs2_read_block(alloc_inode,
1245 le64_to_cpu(cl->cl_recs[chain].c_blkno),
1246 &group_bh);
1247 if (status < 0) {
1248 mlog_errno(status);
1249 goto bail;
1251 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1252 status = ocfs2_check_group_descriptor(alloc_inode->i_sb, fe, bg);
1253 if (status) {
1254 mlog_errno(status);
1255 goto bail;
1258 status = -ENOSPC;
1259 /* for now, the chain search is a bit simplistic. We just use
1260 * the 1st group with any empty bits. */
1261 while ((status = ac->ac_group_search(alloc_inode, group_bh,
1262 bits_wanted, min_bits,
1263 ac->ac_max_block, bit_off,
1264 &tmp_bits)) == -ENOSPC) {
1265 if (!bg->bg_next_group)
1266 break;
1268 brelse(prev_group_bh);
1269 prev_group_bh = NULL;
1271 next_group = le64_to_cpu(bg->bg_next_group);
1272 prev_group_bh = group_bh;
1273 group_bh = NULL;
1274 status = ocfs2_read_block(alloc_inode,
1275 next_group, &group_bh);
1276 if (status < 0) {
1277 mlog_errno(status);
1278 goto bail;
1280 bg = (struct ocfs2_group_desc *) group_bh->b_data;
1281 status = ocfs2_check_group_descriptor(alloc_inode->i_sb, fe, bg);
1282 if (status) {
1283 mlog_errno(status);
1284 goto bail;
1287 if (status < 0) {
1288 if (status != -ENOSPC)
1289 mlog_errno(status);
1290 goto bail;
1293 mlog(0, "alloc succeeds: we give %u bits from block group %llu\n",
1294 tmp_bits, (unsigned long long)le64_to_cpu(bg->bg_blkno));
1296 *num_bits = tmp_bits;
1298 BUG_ON(*num_bits == 0);
1301 * Keep track of previous block descriptor read. When
1302 * we find a target, if we have read more than X
1303 * number of descriptors, and the target is reasonably
1304 * empty, relink him to top of his chain.
1306 * We've read 0 extra blocks and only send one more to
1307 * the transaction, yet the next guy to search has a
1308 * much easier time.
1310 * Do this *after* figuring out how many bits we're taking out
1311 * of our target group.
1313 if (ac->ac_allow_chain_relink &&
1314 (prev_group_bh) &&
1315 (ocfs2_block_group_reasonably_empty(bg, *num_bits))) {
1316 status = ocfs2_relink_block_group(handle, alloc_inode,
1317 ac->ac_bh, group_bh,
1318 prev_group_bh, chain);
1319 if (status < 0) {
1320 mlog_errno(status);
1321 goto bail;
1325 /* Ok, claim our bits now: set the info on dinode, chainlist
1326 * and then the group */
1327 status = ocfs2_journal_access(handle,
1328 alloc_inode,
1329 ac->ac_bh,
1330 OCFS2_JOURNAL_ACCESS_WRITE);
1331 if (status < 0) {
1332 mlog_errno(status);
1333 goto bail;
1336 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1337 fe->id1.bitmap1.i_used = cpu_to_le32(*num_bits + tmp_used);
1338 le32_add_cpu(&cl->cl_recs[chain].c_free, -(*num_bits));
1340 status = ocfs2_journal_dirty(handle,
1341 ac->ac_bh);
1342 if (status < 0) {
1343 mlog_errno(status);
1344 goto bail;
1347 status = ocfs2_block_group_set_bits(handle,
1348 alloc_inode,
1350 group_bh,
1351 *bit_off,
1352 *num_bits);
1353 if (status < 0) {
1354 mlog_errno(status);
1355 goto bail;
1358 mlog(0, "Allocated %u bits from suballocator %llu\n", *num_bits,
1359 (unsigned long long)le64_to_cpu(fe->i_blkno));
1361 *bg_blkno = le64_to_cpu(bg->bg_blkno);
1362 *bits_left = le16_to_cpu(bg->bg_free_bits_count);
1363 bail:
1364 brelse(group_bh);
1365 brelse(prev_group_bh);
1367 mlog_exit(status);
1368 return status;
1371 /* will give out up to bits_wanted contiguous bits. */
1372 static int ocfs2_claim_suballoc_bits(struct ocfs2_super *osb,
1373 struct ocfs2_alloc_context *ac,
1374 handle_t *handle,
1375 u32 bits_wanted,
1376 u32 min_bits,
1377 u16 *bit_off,
1378 unsigned int *num_bits,
1379 u64 *bg_blkno)
1381 int status;
1382 u16 victim, i;
1383 u16 bits_left = 0;
1384 u64 hint_blkno = ac->ac_last_group;
1385 struct ocfs2_chain_list *cl;
1386 struct ocfs2_dinode *fe;
1388 mlog_entry_void();
1390 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1391 BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
1392 BUG_ON(!ac->ac_bh);
1394 fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
1395 if (!OCFS2_IS_VALID_DINODE(fe)) {
1396 OCFS2_RO_ON_INVALID_DINODE(osb->sb, fe);
1397 status = -EIO;
1398 goto bail;
1400 if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
1401 le32_to_cpu(fe->id1.bitmap1.i_total)) {
1402 ocfs2_error(osb->sb, "Chain allocator dinode %llu has %u used "
1403 "bits but only %u total.",
1404 (unsigned long long)le64_to_cpu(fe->i_blkno),
1405 le32_to_cpu(fe->id1.bitmap1.i_used),
1406 le32_to_cpu(fe->id1.bitmap1.i_total));
1407 status = -EIO;
1408 goto bail;
1411 if (hint_blkno) {
1412 /* Attempt to short-circuit the usual search mechanism
1413 * by jumping straight to the most recently used
1414 * allocation group. This helps us mantain some
1415 * contiguousness across allocations. */
1416 status = ocfs2_search_one_group(ac, handle, bits_wanted,
1417 min_bits, bit_off, num_bits,
1418 hint_blkno, &bits_left);
1419 if (!status) {
1420 /* Be careful to update *bg_blkno here as the
1421 * caller is expecting it to be filled in, and
1422 * ocfs2_search_one_group() won't do that for
1423 * us. */
1424 *bg_blkno = hint_blkno;
1425 goto set_hint;
1427 if (status < 0 && status != -ENOSPC) {
1428 mlog_errno(status);
1429 goto bail;
1433 cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
1435 victim = ocfs2_find_victim_chain(cl);
1436 ac->ac_chain = victim;
1437 ac->ac_allow_chain_relink = 1;
1439 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits, bit_off,
1440 num_bits, bg_blkno, &bits_left);
1441 if (!status)
1442 goto set_hint;
1443 if (status < 0 && status != -ENOSPC) {
1444 mlog_errno(status);
1445 goto bail;
1448 mlog(0, "Search of victim chain %u came up with nothing, "
1449 "trying all chains now.\n", victim);
1451 /* If we didn't pick a good victim, then just default to
1452 * searching each chain in order. Don't allow chain relinking
1453 * because we only calculate enough journal credits for one
1454 * relink per alloc. */
1455 ac->ac_allow_chain_relink = 0;
1456 for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
1457 if (i == victim)
1458 continue;
1459 if (!cl->cl_recs[i].c_free)
1460 continue;
1462 ac->ac_chain = i;
1463 status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
1464 bit_off, num_bits, bg_blkno,
1465 &bits_left);
1466 if (!status)
1467 break;
1468 if (status < 0 && status != -ENOSPC) {
1469 mlog_errno(status);
1470 goto bail;
1474 set_hint:
1475 if (status != -ENOSPC) {
1476 /* If the next search of this group is not likely to
1477 * yield a suitable extent, then we reset the last
1478 * group hint so as to not waste a disk read */
1479 if (bits_left < min_bits)
1480 ac->ac_last_group = 0;
1481 else
1482 ac->ac_last_group = *bg_blkno;
1485 bail:
1486 mlog_exit(status);
1487 return status;
1490 int ocfs2_claim_metadata(struct ocfs2_super *osb,
1491 handle_t *handle,
1492 struct ocfs2_alloc_context *ac,
1493 u32 bits_wanted,
1494 u16 *suballoc_bit_start,
1495 unsigned int *num_bits,
1496 u64 *blkno_start)
1498 int status;
1499 u64 bg_blkno;
1501 BUG_ON(!ac);
1502 BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
1503 BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
1505 status = ocfs2_claim_suballoc_bits(osb,
1507 handle,
1508 bits_wanted,
1510 suballoc_bit_start,
1511 num_bits,
1512 &bg_blkno);
1513 if (status < 0) {
1514 mlog_errno(status);
1515 goto bail;
1517 atomic_inc(&osb->alloc_stats.bg_allocs);
1519 *blkno_start = bg_blkno + (u64) *suballoc_bit_start;
1520 ac->ac_bits_given += (*num_bits);
1521 status = 0;
1522 bail:
1523 mlog_exit(status);
1524 return status;
1527 int ocfs2_claim_new_inode(struct ocfs2_super *osb,
1528 handle_t *handle,
1529 struct ocfs2_alloc_context *ac,
1530 u16 *suballoc_bit,
1531 u64 *fe_blkno)
1533 int status;
1534 unsigned int num_bits;
1535 u64 bg_blkno;
1537 mlog_entry_void();
1539 BUG_ON(!ac);
1540 BUG_ON(ac->ac_bits_given != 0);
1541 BUG_ON(ac->ac_bits_wanted != 1);
1542 BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
1544 status = ocfs2_claim_suballoc_bits(osb,
1546 handle,
1549 suballoc_bit,
1550 &num_bits,
1551 &bg_blkno);
1552 if (status < 0) {
1553 mlog_errno(status);
1554 goto bail;
1556 atomic_inc(&osb->alloc_stats.bg_allocs);
1558 BUG_ON(num_bits != 1);
1560 *fe_blkno = bg_blkno + (u64) (*suballoc_bit);
1561 ac->ac_bits_given++;
1562 status = 0;
1563 bail:
1564 mlog_exit(status);
1565 return status;
1568 /* translate a group desc. blkno and it's bitmap offset into
1569 * disk cluster offset. */
1570 static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
1571 u64 bg_blkno,
1572 u16 bg_bit_off)
1574 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1575 u32 cluster = 0;
1577 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1579 if (bg_blkno != osb->first_cluster_group_blkno)
1580 cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
1581 cluster += (u32) bg_bit_off;
1582 return cluster;
1585 /* given a cluster offset, calculate which block group it belongs to
1586 * and return that block offset. */
1587 u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
1589 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1590 u32 group_no;
1592 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1594 group_no = cluster / osb->bitmap_cpg;
1595 if (!group_no)
1596 return osb->first_cluster_group_blkno;
1597 return ocfs2_clusters_to_blocks(inode->i_sb,
1598 group_no * osb->bitmap_cpg);
1601 /* given the block number of a cluster start, calculate which cluster
1602 * group and descriptor bitmap offset that corresponds to. */
1603 static inline void ocfs2_block_to_cluster_group(struct inode *inode,
1604 u64 data_blkno,
1605 u64 *bg_blkno,
1606 u16 *bg_bit_off)
1608 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1609 u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
1611 BUG_ON(!ocfs2_is_cluster_bitmap(inode));
1613 *bg_blkno = ocfs2_which_cluster_group(inode,
1614 data_cluster);
1616 if (*bg_blkno == osb->first_cluster_group_blkno)
1617 *bg_bit_off = (u16) data_cluster;
1618 else
1619 *bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
1620 data_blkno - *bg_blkno);
1624 * min_bits - minimum contiguous chunk from this total allocation we
1625 * can handle. set to what we asked for originally for a full
1626 * contig. allocation, set to '1' to indicate we can deal with extents
1627 * of any size.
1629 int __ocfs2_claim_clusters(struct ocfs2_super *osb,
1630 handle_t *handle,
1631 struct ocfs2_alloc_context *ac,
1632 u32 min_clusters,
1633 u32 max_clusters,
1634 u32 *cluster_start,
1635 u32 *num_clusters)
1637 int status;
1638 unsigned int bits_wanted = max_clusters;
1639 u64 bg_blkno = 0;
1640 u16 bg_bit_off;
1642 mlog_entry_void();
1644 BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
1646 BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
1647 && ac->ac_which != OCFS2_AC_USE_MAIN);
1649 if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
1650 status = ocfs2_claim_local_alloc_bits(osb,
1651 handle,
1653 bits_wanted,
1654 cluster_start,
1655 num_clusters);
1656 if (!status)
1657 atomic_inc(&osb->alloc_stats.local_data);
1658 } else {
1659 if (min_clusters > (osb->bitmap_cpg - 1)) {
1660 /* The only paths asking for contiguousness
1661 * should know about this already. */
1662 mlog(ML_ERROR, "minimum allocation requested %u exceeds "
1663 "group bitmap size %u!\n", min_clusters,
1664 osb->bitmap_cpg);
1665 status = -ENOSPC;
1666 goto bail;
1668 /* clamp the current request down to a realistic size. */
1669 if (bits_wanted > (osb->bitmap_cpg - 1))
1670 bits_wanted = osb->bitmap_cpg - 1;
1672 status = ocfs2_claim_suballoc_bits(osb,
1674 handle,
1675 bits_wanted,
1676 min_clusters,
1677 &bg_bit_off,
1678 num_clusters,
1679 &bg_blkno);
1680 if (!status) {
1681 *cluster_start =
1682 ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
1683 bg_blkno,
1684 bg_bit_off);
1685 atomic_inc(&osb->alloc_stats.bitmap_data);
1688 if (status < 0) {
1689 if (status != -ENOSPC)
1690 mlog_errno(status);
1691 goto bail;
1694 ac->ac_bits_given += *num_clusters;
1696 bail:
1697 mlog_exit(status);
1698 return status;
1701 int ocfs2_claim_clusters(struct ocfs2_super *osb,
1702 handle_t *handle,
1703 struct ocfs2_alloc_context *ac,
1704 u32 min_clusters,
1705 u32 *cluster_start,
1706 u32 *num_clusters)
1708 unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
1710 return __ocfs2_claim_clusters(osb, handle, ac, min_clusters,
1711 bits_wanted, cluster_start, num_clusters);
1714 static inline int ocfs2_block_group_clear_bits(handle_t *handle,
1715 struct inode *alloc_inode,
1716 struct ocfs2_group_desc *bg,
1717 struct buffer_head *group_bh,
1718 unsigned int bit_off,
1719 unsigned int num_bits)
1721 int status;
1722 unsigned int tmp;
1723 int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
1724 struct ocfs2_group_desc *undo_bg = NULL;
1726 mlog_entry_void();
1728 if (!OCFS2_IS_VALID_GROUP_DESC(bg)) {
1729 OCFS2_RO_ON_INVALID_GROUP_DESC(alloc_inode->i_sb, bg);
1730 status = -EIO;
1731 goto bail;
1734 mlog(0, "off = %u, num = %u\n", bit_off, num_bits);
1736 if (ocfs2_is_cluster_bitmap(alloc_inode))
1737 journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
1739 status = ocfs2_journal_access(handle, alloc_inode, group_bh,
1740 journal_type);
1741 if (status < 0) {
1742 mlog_errno(status);
1743 goto bail;
1746 if (ocfs2_is_cluster_bitmap(alloc_inode))
1747 undo_bg = (struct ocfs2_group_desc *) bh2jh(group_bh)->b_committed_data;
1749 tmp = num_bits;
1750 while(tmp--) {
1751 ocfs2_clear_bit((bit_off + tmp),
1752 (unsigned long *) bg->bg_bitmap);
1753 if (ocfs2_is_cluster_bitmap(alloc_inode))
1754 ocfs2_set_bit(bit_off + tmp,
1755 (unsigned long *) undo_bg->bg_bitmap);
1757 le16_add_cpu(&bg->bg_free_bits_count, num_bits);
1759 status = ocfs2_journal_dirty(handle, group_bh);
1760 if (status < 0)
1761 mlog_errno(status);
1762 bail:
1763 return status;
1767 * expects the suballoc inode to already be locked.
1769 int ocfs2_free_suballoc_bits(handle_t *handle,
1770 struct inode *alloc_inode,
1771 struct buffer_head *alloc_bh,
1772 unsigned int start_bit,
1773 u64 bg_blkno,
1774 unsigned int count)
1776 int status = 0;
1777 u32 tmp_used;
1778 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
1779 struct ocfs2_chain_list *cl = &fe->id2.i_chain;
1780 struct buffer_head *group_bh = NULL;
1781 struct ocfs2_group_desc *group;
1783 mlog_entry_void();
1785 if (!OCFS2_IS_VALID_DINODE(fe)) {
1786 OCFS2_RO_ON_INVALID_DINODE(alloc_inode->i_sb, fe);
1787 status = -EIO;
1788 goto bail;
1790 BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
1792 mlog(0, "%llu: freeing %u bits from group %llu, starting at %u\n",
1793 (unsigned long long)OCFS2_I(alloc_inode)->ip_blkno, count,
1794 (unsigned long long)bg_blkno, start_bit);
1796 status = ocfs2_read_block(alloc_inode, bg_blkno, &group_bh);
1797 if (status < 0) {
1798 mlog_errno(status);
1799 goto bail;
1802 group = (struct ocfs2_group_desc *) group_bh->b_data;
1803 status = ocfs2_check_group_descriptor(alloc_inode->i_sb, fe, group);
1804 if (status) {
1805 mlog_errno(status);
1806 goto bail;
1808 BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
1810 status = ocfs2_block_group_clear_bits(handle, alloc_inode,
1811 group, group_bh,
1812 start_bit, count);
1813 if (status < 0) {
1814 mlog_errno(status);
1815 goto bail;
1818 status = ocfs2_journal_access(handle, alloc_inode, alloc_bh,
1819 OCFS2_JOURNAL_ACCESS_WRITE);
1820 if (status < 0) {
1821 mlog_errno(status);
1822 goto bail;
1825 le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
1826 count);
1827 tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
1828 fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
1830 status = ocfs2_journal_dirty(handle, alloc_bh);
1831 if (status < 0) {
1832 mlog_errno(status);
1833 goto bail;
1836 bail:
1837 brelse(group_bh);
1839 mlog_exit(status);
1840 return status;
1843 int ocfs2_free_dinode(handle_t *handle,
1844 struct inode *inode_alloc_inode,
1845 struct buffer_head *inode_alloc_bh,
1846 struct ocfs2_dinode *di)
1848 u64 blk = le64_to_cpu(di->i_blkno);
1849 u16 bit = le16_to_cpu(di->i_suballoc_bit);
1850 u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1852 return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
1853 inode_alloc_bh, bit, bg_blkno, 1);
1856 int ocfs2_free_clusters(handle_t *handle,
1857 struct inode *bitmap_inode,
1858 struct buffer_head *bitmap_bh,
1859 u64 start_blk,
1860 unsigned int num_clusters)
1862 int status;
1863 u16 bg_start_bit;
1864 u64 bg_blkno;
1865 struct ocfs2_dinode *fe;
1867 /* You can't ever have a contiguous set of clusters
1868 * bigger than a block group bitmap so we never have to worry
1869 * about looping on them. */
1871 mlog_entry_void();
1873 /* This is expensive. We can safely remove once this stuff has
1874 * gotten tested really well. */
1875 BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
1877 fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
1879 ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
1880 &bg_start_bit);
1882 mlog(0, "want to free %u clusters starting at block %llu\n",
1883 num_clusters, (unsigned long long)start_blk);
1884 mlog(0, "bg_blkno = %llu, bg_start_bit = %u\n",
1885 (unsigned long long)bg_blkno, bg_start_bit);
1887 status = ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
1888 bg_start_bit, bg_blkno,
1889 num_clusters);
1890 if (status < 0) {
1891 mlog_errno(status);
1892 goto out;
1895 ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
1896 num_clusters);
1898 out:
1899 mlog_exit(status);
1900 return status;
1903 static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
1905 printk("Block Group:\n");
1906 printk("bg_signature: %s\n", bg->bg_signature);
1907 printk("bg_size: %u\n", bg->bg_size);
1908 printk("bg_bits: %u\n", bg->bg_bits);
1909 printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
1910 printk("bg_chain: %u\n", bg->bg_chain);
1911 printk("bg_generation: %u\n", le32_to_cpu(bg->bg_generation));
1912 printk("bg_next_group: %llu\n",
1913 (unsigned long long)bg->bg_next_group);
1914 printk("bg_parent_dinode: %llu\n",
1915 (unsigned long long)bg->bg_parent_dinode);
1916 printk("bg_blkno: %llu\n",
1917 (unsigned long long)bg->bg_blkno);
1920 static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
1922 int i;
1924 printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
1925 printk("i_signature: %s\n", fe->i_signature);
1926 printk("i_size: %llu\n",
1927 (unsigned long long)fe->i_size);
1928 printk("i_clusters: %u\n", fe->i_clusters);
1929 printk("i_generation: %u\n",
1930 le32_to_cpu(fe->i_generation));
1931 printk("id1.bitmap1.i_used: %u\n",
1932 le32_to_cpu(fe->id1.bitmap1.i_used));
1933 printk("id1.bitmap1.i_total: %u\n",
1934 le32_to_cpu(fe->id1.bitmap1.i_total));
1935 printk("id2.i_chain.cl_cpg: %u\n", fe->id2.i_chain.cl_cpg);
1936 printk("id2.i_chain.cl_bpc: %u\n", fe->id2.i_chain.cl_bpc);
1937 printk("id2.i_chain.cl_count: %u\n", fe->id2.i_chain.cl_count);
1938 printk("id2.i_chain.cl_next_free_rec: %u\n",
1939 fe->id2.i_chain.cl_next_free_rec);
1940 for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
1941 printk("fe->id2.i_chain.cl_recs[%d].c_free: %u\n", i,
1942 fe->id2.i_chain.cl_recs[i].c_free);
1943 printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
1944 fe->id2.i_chain.cl_recs[i].c_total);
1945 printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
1946 (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
1951 * For a given allocation, determine which allocators will need to be
1952 * accessed, and lock them, reserving the appropriate number of bits.
1954 * Sparse file systems call this from ocfs2_write_begin_nolock()
1955 * and ocfs2_allocate_unwritten_extents().
1957 * File systems which don't support holes call this from
1958 * ocfs2_extend_allocation().
1960 int ocfs2_lock_allocators(struct inode *inode,
1961 struct ocfs2_extent_tree *et,
1962 u32 clusters_to_add, u32 extents_to_split,
1963 struct ocfs2_alloc_context **data_ac,
1964 struct ocfs2_alloc_context **meta_ac)
1966 int ret = 0, num_free_extents;
1967 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
1968 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1970 *meta_ac = NULL;
1971 if (data_ac)
1972 *data_ac = NULL;
1974 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
1976 num_free_extents = ocfs2_num_free_extents(osb, inode, et);
1977 if (num_free_extents < 0) {
1978 ret = num_free_extents;
1979 mlog_errno(ret);
1980 goto out;
1984 * Sparse allocation file systems need to be more conservative
1985 * with reserving room for expansion - the actual allocation
1986 * happens while we've got a journal handle open so re-taking
1987 * a cluster lock (because we ran out of room for another
1988 * extent) will violate ordering rules.
1990 * Most of the time we'll only be seeing this 1 cluster at a time
1991 * anyway.
1993 * Always lock for any unwritten extents - we might want to
1994 * add blocks during a split.
1996 if (!num_free_extents ||
1997 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
1998 ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
1999 if (ret < 0) {
2000 if (ret != -ENOSPC)
2001 mlog_errno(ret);
2002 goto out;
2006 if (clusters_to_add == 0)
2007 goto out;
2009 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
2010 if (ret < 0) {
2011 if (ret != -ENOSPC)
2012 mlog_errno(ret);
2013 goto out;
2016 out:
2017 if (ret) {
2018 if (*meta_ac) {
2019 ocfs2_free_alloc_context(*meta_ac);
2020 *meta_ac = NULL;
2024 * We cannot have an error and a non null *data_ac.
2028 return ret;