Loading...
1/*
2 * alloc.c - NILFS dat/inode allocator
3 *
4 * Copyright (C) 2006-2008 Nippon Telegraph and Telephone Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * Originally written by Koji Sato.
17 * Two allocators were unified by Ryusuke Konishi and Amagai Yoshiji.
18 */
19
20#include <linux/types.h>
21#include <linux/buffer_head.h>
22#include <linux/fs.h>
23#include <linux/bitops.h>
24#include <linux/slab.h>
25#include "mdt.h"
26#include "alloc.h"
27
28
29/**
30 * nilfs_palloc_groups_per_desc_block - get the number of groups that a group
31 * descriptor block can maintain
32 * @inode: inode of metadata file using this allocator
33 */
34static inline unsigned long
35nilfs_palloc_groups_per_desc_block(const struct inode *inode)
36{
37 return (1UL << inode->i_blkbits) /
38 sizeof(struct nilfs_palloc_group_desc);
39}
40
41/**
42 * nilfs_palloc_groups_count - get maximum number of groups
43 * @inode: inode of metadata file using this allocator
44 */
45static inline unsigned long
46nilfs_palloc_groups_count(const struct inode *inode)
47{
48 return 1UL << (BITS_PER_LONG - (inode->i_blkbits + 3 /* log2(8) */));
49}
50
51/**
52 * nilfs_palloc_init_blockgroup - initialize private variables for allocator
53 * @inode: inode of metadata file using this allocator
54 * @entry_size: size of the persistent object
55 */
56int nilfs_palloc_init_blockgroup(struct inode *inode, unsigned int entry_size)
57{
58 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
59
60 mi->mi_bgl = kmalloc(sizeof(*mi->mi_bgl), GFP_NOFS);
61 if (!mi->mi_bgl)
62 return -ENOMEM;
63
64 bgl_lock_init(mi->mi_bgl);
65
66 nilfs_mdt_set_entry_size(inode, entry_size, 0);
67
68 mi->mi_blocks_per_group =
69 DIV_ROUND_UP(nilfs_palloc_entries_per_group(inode),
70 mi->mi_entries_per_block) + 1;
71 /*
72 * Number of blocks in a group including entry blocks
73 * and a bitmap block
74 */
75 mi->mi_blocks_per_desc_block =
76 nilfs_palloc_groups_per_desc_block(inode) *
77 mi->mi_blocks_per_group + 1;
78 /*
79 * Number of blocks per descriptor including the
80 * descriptor block
81 */
82 return 0;
83}
84
85/**
86 * nilfs_palloc_group - get group number and offset from an entry number
87 * @inode: inode of metadata file using this allocator
88 * @nr: serial number of the entry (e.g. inode number)
89 * @offset: pointer to store offset number in the group
90 */
91static unsigned long nilfs_palloc_group(const struct inode *inode, __u64 nr,
92 unsigned long *offset)
93{
94 __u64 group = nr;
95
96 *offset = do_div(group, nilfs_palloc_entries_per_group(inode));
97 return group;
98}
99
100/**
101 * nilfs_palloc_desc_blkoff - get block offset of a group descriptor block
102 * @inode: inode of metadata file using this allocator
103 * @group: group number
104 *
105 * nilfs_palloc_desc_blkoff() returns block offset of the descriptor
106 * block which contains a descriptor of the specified group.
107 */
108static unsigned long
109nilfs_palloc_desc_blkoff(const struct inode *inode, unsigned long group)
110{
111 unsigned long desc_block =
112 group / nilfs_palloc_groups_per_desc_block(inode);
113 return desc_block * NILFS_MDT(inode)->mi_blocks_per_desc_block;
114}
115
116/**
117 * nilfs_palloc_bitmap_blkoff - get block offset of a bitmap block
118 * @inode: inode of metadata file using this allocator
119 * @group: group number
120 *
121 * nilfs_palloc_bitmap_blkoff() returns block offset of the bitmap
122 * block used to allocate/deallocate entries in the specified group.
123 */
124static unsigned long
125nilfs_palloc_bitmap_blkoff(const struct inode *inode, unsigned long group)
126{
127 unsigned long desc_offset =
128 group % nilfs_palloc_groups_per_desc_block(inode);
129 return nilfs_palloc_desc_blkoff(inode, group) + 1 +
130 desc_offset * NILFS_MDT(inode)->mi_blocks_per_group;
131}
132
133/**
134 * nilfs_palloc_group_desc_nfrees - get the number of free entries in a group
135 * @desc: pointer to descriptor structure for the group
136 * @lock: spin lock protecting @desc
137 */
138static unsigned long
139nilfs_palloc_group_desc_nfrees(const struct nilfs_palloc_group_desc *desc,
140 spinlock_t *lock)
141{
142 unsigned long nfree;
143
144 spin_lock(lock);
145 nfree = le32_to_cpu(desc->pg_nfrees);
146 spin_unlock(lock);
147 return nfree;
148}
149
150/**
151 * nilfs_palloc_group_desc_add_entries - adjust count of free entries
152 * @desc: pointer to descriptor structure for the group
153 * @lock: spin lock protecting @desc
154 * @n: delta to be added
155 */
156static u32
157nilfs_palloc_group_desc_add_entries(struct nilfs_palloc_group_desc *desc,
158 spinlock_t *lock, u32 n)
159{
160 u32 nfree;
161
162 spin_lock(lock);
163 le32_add_cpu(&desc->pg_nfrees, n);
164 nfree = le32_to_cpu(desc->pg_nfrees);
165 spin_unlock(lock);
166 return nfree;
167}
168
169/**
170 * nilfs_palloc_entry_blkoff - get block offset of an entry block
171 * @inode: inode of metadata file using this allocator
172 * @nr: serial number of the entry (e.g. inode number)
173 */
174static unsigned long
175nilfs_palloc_entry_blkoff(const struct inode *inode, __u64 nr)
176{
177 unsigned long group, group_offset;
178
179 group = nilfs_palloc_group(inode, nr, &group_offset);
180
181 return nilfs_palloc_bitmap_blkoff(inode, group) + 1 +
182 group_offset / NILFS_MDT(inode)->mi_entries_per_block;
183}
184
185/**
186 * nilfs_palloc_desc_block_init - initialize buffer of a group descriptor block
187 * @inode: inode of metadata file
188 * @bh: buffer head of the buffer to be initialized
189 * @kaddr: kernel address mapped for the page including the buffer
190 */
191static void nilfs_palloc_desc_block_init(struct inode *inode,
192 struct buffer_head *bh, void *kaddr)
193{
194 struct nilfs_palloc_group_desc *desc = kaddr + bh_offset(bh);
195 unsigned long n = nilfs_palloc_groups_per_desc_block(inode);
196 __le32 nfrees;
197
198 nfrees = cpu_to_le32(nilfs_palloc_entries_per_group(inode));
199 while (n-- > 0) {
200 desc->pg_nfrees = nfrees;
201 desc++;
202 }
203}
204
205static int nilfs_palloc_get_block(struct inode *inode, unsigned long blkoff,
206 int create,
207 void (*init_block)(struct inode *,
208 struct buffer_head *,
209 void *),
210 struct buffer_head **bhp,
211 struct nilfs_bh_assoc *prev,
212 spinlock_t *lock)
213{
214 int ret;
215
216 spin_lock(lock);
217 if (prev->bh && blkoff == prev->blkoff) {
218 get_bh(prev->bh);
219 *bhp = prev->bh;
220 spin_unlock(lock);
221 return 0;
222 }
223 spin_unlock(lock);
224
225 ret = nilfs_mdt_get_block(inode, blkoff, create, init_block, bhp);
226 if (!ret) {
227 spin_lock(lock);
228 /*
229 * The following code must be safe for change of the
230 * cache contents during the get block call.
231 */
232 brelse(prev->bh);
233 get_bh(*bhp);
234 prev->bh = *bhp;
235 prev->blkoff = blkoff;
236 spin_unlock(lock);
237 }
238 return ret;
239}
240
241/**
242 * nilfs_palloc_delete_block - delete a block on the persistent allocator file
243 * @inode: inode of metadata file using this allocator
244 * @blkoff: block offset
245 * @prev: nilfs_bh_assoc struct of the last used buffer
246 * @lock: spin lock protecting @prev
247 */
248static int nilfs_palloc_delete_block(struct inode *inode, unsigned long blkoff,
249 struct nilfs_bh_assoc *prev,
250 spinlock_t *lock)
251{
252 spin_lock(lock);
253 if (prev->bh && blkoff == prev->blkoff) {
254 brelse(prev->bh);
255 prev->bh = NULL;
256 }
257 spin_unlock(lock);
258 return nilfs_mdt_delete_block(inode, blkoff);
259}
260
261/**
262 * nilfs_palloc_get_desc_block - get buffer head of a group descriptor block
263 * @inode: inode of metadata file using this allocator
264 * @group: group number
265 * @create: create flag
266 * @bhp: pointer to store the resultant buffer head
267 */
268static int nilfs_palloc_get_desc_block(struct inode *inode,
269 unsigned long group,
270 int create, struct buffer_head **bhp)
271{
272 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
273
274 return nilfs_palloc_get_block(inode,
275 nilfs_palloc_desc_blkoff(inode, group),
276 create, nilfs_palloc_desc_block_init,
277 bhp, &cache->prev_desc, &cache->lock);
278}
279
280/**
281 * nilfs_palloc_get_bitmap_block - get buffer head of a bitmap block
282 * @inode: inode of metadata file using this allocator
283 * @group: group number
284 * @create: create flag
285 * @bhp: pointer to store the resultant buffer head
286 */
287static int nilfs_palloc_get_bitmap_block(struct inode *inode,
288 unsigned long group,
289 int create, struct buffer_head **bhp)
290{
291 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
292
293 return nilfs_palloc_get_block(inode,
294 nilfs_palloc_bitmap_blkoff(inode, group),
295 create, NULL, bhp,
296 &cache->prev_bitmap, &cache->lock);
297}
298
299/**
300 * nilfs_palloc_delete_bitmap_block - delete a bitmap block
301 * @inode: inode of metadata file using this allocator
302 * @group: group number
303 */
304static int nilfs_palloc_delete_bitmap_block(struct inode *inode,
305 unsigned long group)
306{
307 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
308
309 return nilfs_palloc_delete_block(inode,
310 nilfs_palloc_bitmap_blkoff(inode,
311 group),
312 &cache->prev_bitmap, &cache->lock);
313}
314
315/**
316 * nilfs_palloc_get_entry_block - get buffer head of an entry block
317 * @inode: inode of metadata file using this allocator
318 * @nr: serial number of the entry (e.g. inode number)
319 * @create: create flag
320 * @bhp: pointer to store the resultant buffer head
321 */
322int nilfs_palloc_get_entry_block(struct inode *inode, __u64 nr,
323 int create, struct buffer_head **bhp)
324{
325 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
326
327 return nilfs_palloc_get_block(inode,
328 nilfs_palloc_entry_blkoff(inode, nr),
329 create, NULL, bhp,
330 &cache->prev_entry, &cache->lock);
331}
332
333/**
334 * nilfs_palloc_delete_entry_block - delete an entry block
335 * @inode: inode of metadata file using this allocator
336 * @nr: serial number of the entry
337 */
338static int nilfs_palloc_delete_entry_block(struct inode *inode, __u64 nr)
339{
340 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
341
342 return nilfs_palloc_delete_block(inode,
343 nilfs_palloc_entry_blkoff(inode, nr),
344 &cache->prev_entry, &cache->lock);
345}
346
347/**
348 * nilfs_palloc_block_get_group_desc - get kernel address of a group descriptor
349 * @inode: inode of metadata file using this allocator
350 * @group: group number
351 * @bh: buffer head of the buffer storing the group descriptor block
352 * @kaddr: kernel address mapped for the page including the buffer
353 */
354static struct nilfs_palloc_group_desc *
355nilfs_palloc_block_get_group_desc(const struct inode *inode,
356 unsigned long group,
357 const struct buffer_head *bh, void *kaddr)
358{
359 return (struct nilfs_palloc_group_desc *)(kaddr + bh_offset(bh)) +
360 group % nilfs_palloc_groups_per_desc_block(inode);
361}
362
363/**
364 * nilfs_palloc_block_get_entry - get kernel address of an entry
365 * @inode: inode of metadata file using this allocator
366 * @nr: serial number of the entry (e.g. inode number)
367 * @bh: buffer head of the buffer storing the entry block
368 * @kaddr: kernel address mapped for the page including the buffer
369 */
370void *nilfs_palloc_block_get_entry(const struct inode *inode, __u64 nr,
371 const struct buffer_head *bh, void *kaddr)
372{
373 unsigned long entry_offset, group_offset;
374
375 nilfs_palloc_group(inode, nr, &group_offset);
376 entry_offset = group_offset % NILFS_MDT(inode)->mi_entries_per_block;
377
378 return kaddr + bh_offset(bh) +
379 entry_offset * NILFS_MDT(inode)->mi_entry_size;
380}
381
382/**
383 * nilfs_palloc_find_available_slot - find available slot in a group
384 * @bitmap: bitmap of the group
385 * @target: offset number of an entry in the group (start point)
386 * @bsize: size in bits
387 * @lock: spin lock protecting @bitmap
388 */
389static int nilfs_palloc_find_available_slot(unsigned char *bitmap,
390 unsigned long target,
391 unsigned int bsize,
392 spinlock_t *lock)
393{
394 int pos, end = bsize;
395
396 if (likely(target < bsize)) {
397 pos = target;
398 do {
399 pos = nilfs_find_next_zero_bit(bitmap, end, pos);
400 if (pos >= end)
401 break;
402 if (!nilfs_set_bit_atomic(lock, pos, bitmap))
403 return pos;
404 } while (++pos < end);
405
406 end = target;
407 }
408
409 /* wrap around */
410 for (pos = 0; pos < end; pos++) {
411 pos = nilfs_find_next_zero_bit(bitmap, end, pos);
412 if (pos >= end)
413 break;
414 if (!nilfs_set_bit_atomic(lock, pos, bitmap))
415 return pos;
416 }
417
418 return -ENOSPC;
419}
420
421/**
422 * nilfs_palloc_rest_groups_in_desc_block - get the remaining number of groups
423 * in a group descriptor block
424 * @inode: inode of metadata file using this allocator
425 * @curr: current group number
426 * @max: maximum number of groups
427 */
428static unsigned long
429nilfs_palloc_rest_groups_in_desc_block(const struct inode *inode,
430 unsigned long curr, unsigned long max)
431{
432 return min_t(unsigned long,
433 nilfs_palloc_groups_per_desc_block(inode) -
434 curr % nilfs_palloc_groups_per_desc_block(inode),
435 max - curr + 1);
436}
437
438/**
439 * nilfs_palloc_count_desc_blocks - count descriptor blocks number
440 * @inode: inode of metadata file using this allocator
441 * @desc_blocks: descriptor blocks number [out]
442 */
443static int nilfs_palloc_count_desc_blocks(struct inode *inode,
444 unsigned long *desc_blocks)
445{
446 __u64 blknum;
447 int ret;
448
449 ret = nilfs_bmap_last_key(NILFS_I(inode)->i_bmap, &blknum);
450 if (likely(!ret))
451 *desc_blocks = DIV_ROUND_UP(
452 (unsigned long)blknum,
453 NILFS_MDT(inode)->mi_blocks_per_desc_block);
454 return ret;
455}
456
457/**
458 * nilfs_palloc_mdt_file_can_grow - check potential opportunity for
459 * MDT file growing
460 * @inode: inode of metadata file using this allocator
461 * @desc_blocks: known current descriptor blocks count
462 */
463static inline bool nilfs_palloc_mdt_file_can_grow(struct inode *inode,
464 unsigned long desc_blocks)
465{
466 return (nilfs_palloc_groups_per_desc_block(inode) * desc_blocks) <
467 nilfs_palloc_groups_count(inode);
468}
469
470/**
471 * nilfs_palloc_count_max_entries - count max number of entries that can be
472 * described by descriptor blocks count
473 * @inode: inode of metadata file using this allocator
474 * @nused: current number of used entries
475 * @nmaxp: max number of entries [out]
476 */
477int nilfs_palloc_count_max_entries(struct inode *inode, u64 nused, u64 *nmaxp)
478{
479 unsigned long desc_blocks = 0;
480 u64 entries_per_desc_block, nmax;
481 int err;
482
483 err = nilfs_palloc_count_desc_blocks(inode, &desc_blocks);
484 if (unlikely(err))
485 return err;
486
487 entries_per_desc_block = (u64)nilfs_palloc_entries_per_group(inode) *
488 nilfs_palloc_groups_per_desc_block(inode);
489 nmax = entries_per_desc_block * desc_blocks;
490
491 if (nused == nmax &&
492 nilfs_palloc_mdt_file_can_grow(inode, desc_blocks))
493 nmax += entries_per_desc_block;
494
495 if (nused > nmax)
496 return -ERANGE;
497
498 *nmaxp = nmax;
499 return 0;
500}
501
502/**
503 * nilfs_palloc_prepare_alloc_entry - prepare to allocate a persistent object
504 * @inode: inode of metadata file using this allocator
505 * @req: nilfs_palloc_req structure exchanged for the allocation
506 */
507int nilfs_palloc_prepare_alloc_entry(struct inode *inode,
508 struct nilfs_palloc_req *req)
509{
510 struct buffer_head *desc_bh, *bitmap_bh;
511 struct nilfs_palloc_group_desc *desc;
512 unsigned char *bitmap;
513 void *desc_kaddr, *bitmap_kaddr;
514 unsigned long group, maxgroup, ngroups;
515 unsigned long group_offset, maxgroup_offset;
516 unsigned long n, entries_per_group;
517 unsigned long i, j;
518 spinlock_t *lock;
519 int pos, ret;
520
521 ngroups = nilfs_palloc_groups_count(inode);
522 maxgroup = ngroups - 1;
523 group = nilfs_palloc_group(inode, req->pr_entry_nr, &group_offset);
524 entries_per_group = nilfs_palloc_entries_per_group(inode);
525
526 for (i = 0; i < ngroups; i += n) {
527 if (group >= ngroups) {
528 /* wrap around */
529 group = 0;
530 maxgroup = nilfs_palloc_group(inode, req->pr_entry_nr,
531 &maxgroup_offset) - 1;
532 }
533 ret = nilfs_palloc_get_desc_block(inode, group, 1, &desc_bh);
534 if (ret < 0)
535 return ret;
536 desc_kaddr = kmap(desc_bh->b_page);
537 desc = nilfs_palloc_block_get_group_desc(
538 inode, group, desc_bh, desc_kaddr);
539 n = nilfs_palloc_rest_groups_in_desc_block(inode, group,
540 maxgroup);
541 for (j = 0; j < n; j++, desc++, group++) {
542 lock = nilfs_mdt_bgl_lock(inode, group);
543 if (nilfs_palloc_group_desc_nfrees(desc, lock) > 0) {
544 ret = nilfs_palloc_get_bitmap_block(
545 inode, group, 1, &bitmap_bh);
546 if (ret < 0)
547 goto out_desc;
548 bitmap_kaddr = kmap(bitmap_bh->b_page);
549 bitmap = bitmap_kaddr + bh_offset(bitmap_bh);
550 pos = nilfs_palloc_find_available_slot(
551 bitmap, group_offset,
552 entries_per_group, lock);
553 if (pos >= 0) {
554 /* found a free entry */
555 nilfs_palloc_group_desc_add_entries(
556 desc, lock, -1);
557 req->pr_entry_nr =
558 entries_per_group * group + pos;
559 kunmap(desc_bh->b_page);
560 kunmap(bitmap_bh->b_page);
561
562 req->pr_desc_bh = desc_bh;
563 req->pr_bitmap_bh = bitmap_bh;
564 return 0;
565 }
566 kunmap(bitmap_bh->b_page);
567 brelse(bitmap_bh);
568 }
569
570 group_offset = 0;
571 }
572
573 kunmap(desc_bh->b_page);
574 brelse(desc_bh);
575 }
576
577 /* no entries left */
578 return -ENOSPC;
579
580 out_desc:
581 kunmap(desc_bh->b_page);
582 brelse(desc_bh);
583 return ret;
584}
585
586/**
587 * nilfs_palloc_commit_alloc_entry - finish allocation of a persistent object
588 * @inode: inode of metadata file using this allocator
589 * @req: nilfs_palloc_req structure exchanged for the allocation
590 */
591void nilfs_palloc_commit_alloc_entry(struct inode *inode,
592 struct nilfs_palloc_req *req)
593{
594 mark_buffer_dirty(req->pr_bitmap_bh);
595 mark_buffer_dirty(req->pr_desc_bh);
596 nilfs_mdt_mark_dirty(inode);
597
598 brelse(req->pr_bitmap_bh);
599 brelse(req->pr_desc_bh);
600}
601
602/**
603 * nilfs_palloc_commit_free_entry - finish deallocating a persistent object
604 * @inode: inode of metadata file using this allocator
605 * @req: nilfs_palloc_req structure exchanged for the removal
606 */
607void nilfs_palloc_commit_free_entry(struct inode *inode,
608 struct nilfs_palloc_req *req)
609{
610 struct nilfs_palloc_group_desc *desc;
611 unsigned long group, group_offset;
612 unsigned char *bitmap;
613 void *desc_kaddr, *bitmap_kaddr;
614 spinlock_t *lock;
615
616 group = nilfs_palloc_group(inode, req->pr_entry_nr, &group_offset);
617 desc_kaddr = kmap(req->pr_desc_bh->b_page);
618 desc = nilfs_palloc_block_get_group_desc(inode, group,
619 req->pr_desc_bh, desc_kaddr);
620 bitmap_kaddr = kmap(req->pr_bitmap_bh->b_page);
621 bitmap = bitmap_kaddr + bh_offset(req->pr_bitmap_bh);
622 lock = nilfs_mdt_bgl_lock(inode, group);
623
624 if (!nilfs_clear_bit_atomic(lock, group_offset, bitmap))
625 nilfs_msg(inode->i_sb, KERN_WARNING,
626 "%s (ino=%lu): entry number %llu already freed",
627 __func__, inode->i_ino,
628 (unsigned long long)req->pr_entry_nr);
629 else
630 nilfs_palloc_group_desc_add_entries(desc, lock, 1);
631
632 kunmap(req->pr_bitmap_bh->b_page);
633 kunmap(req->pr_desc_bh->b_page);
634
635 mark_buffer_dirty(req->pr_desc_bh);
636 mark_buffer_dirty(req->pr_bitmap_bh);
637 nilfs_mdt_mark_dirty(inode);
638
639 brelse(req->pr_bitmap_bh);
640 brelse(req->pr_desc_bh);
641}
642
643/**
644 * nilfs_palloc_abort_alloc_entry - cancel allocation of a persistent object
645 * @inode: inode of metadata file using this allocator
646 * @req: nilfs_palloc_req structure exchanged for the allocation
647 */
648void nilfs_palloc_abort_alloc_entry(struct inode *inode,
649 struct nilfs_palloc_req *req)
650{
651 struct nilfs_palloc_group_desc *desc;
652 void *desc_kaddr, *bitmap_kaddr;
653 unsigned char *bitmap;
654 unsigned long group, group_offset;
655 spinlock_t *lock;
656
657 group = nilfs_palloc_group(inode, req->pr_entry_nr, &group_offset);
658 desc_kaddr = kmap(req->pr_desc_bh->b_page);
659 desc = nilfs_palloc_block_get_group_desc(inode, group,
660 req->pr_desc_bh, desc_kaddr);
661 bitmap_kaddr = kmap(req->pr_bitmap_bh->b_page);
662 bitmap = bitmap_kaddr + bh_offset(req->pr_bitmap_bh);
663 lock = nilfs_mdt_bgl_lock(inode, group);
664
665 if (!nilfs_clear_bit_atomic(lock, group_offset, bitmap))
666 nilfs_msg(inode->i_sb, KERN_WARNING,
667 "%s (ino=%lu): entry number %llu already freed",
668 __func__, inode->i_ino,
669 (unsigned long long)req->pr_entry_nr);
670 else
671 nilfs_palloc_group_desc_add_entries(desc, lock, 1);
672
673 kunmap(req->pr_bitmap_bh->b_page);
674 kunmap(req->pr_desc_bh->b_page);
675
676 brelse(req->pr_bitmap_bh);
677 brelse(req->pr_desc_bh);
678
679 req->pr_entry_nr = 0;
680 req->pr_bitmap_bh = NULL;
681 req->pr_desc_bh = NULL;
682}
683
684/**
685 * nilfs_palloc_prepare_free_entry - prepare to deallocate a persistent object
686 * @inode: inode of metadata file using this allocator
687 * @req: nilfs_palloc_req structure exchanged for the removal
688 */
689int nilfs_palloc_prepare_free_entry(struct inode *inode,
690 struct nilfs_palloc_req *req)
691{
692 struct buffer_head *desc_bh, *bitmap_bh;
693 unsigned long group, group_offset;
694 int ret;
695
696 group = nilfs_palloc_group(inode, req->pr_entry_nr, &group_offset);
697 ret = nilfs_palloc_get_desc_block(inode, group, 1, &desc_bh);
698 if (ret < 0)
699 return ret;
700 ret = nilfs_palloc_get_bitmap_block(inode, group, 1, &bitmap_bh);
701 if (ret < 0) {
702 brelse(desc_bh);
703 return ret;
704 }
705
706 req->pr_desc_bh = desc_bh;
707 req->pr_bitmap_bh = bitmap_bh;
708 return 0;
709}
710
711/**
712 * nilfs_palloc_abort_free_entry - cancel deallocating a persistent object
713 * @inode: inode of metadata file using this allocator
714 * @req: nilfs_palloc_req structure exchanged for the removal
715 */
716void nilfs_palloc_abort_free_entry(struct inode *inode,
717 struct nilfs_palloc_req *req)
718{
719 brelse(req->pr_bitmap_bh);
720 brelse(req->pr_desc_bh);
721
722 req->pr_entry_nr = 0;
723 req->pr_bitmap_bh = NULL;
724 req->pr_desc_bh = NULL;
725}
726
727/**
728 * nilfs_palloc_freev - deallocate a set of persistent objects
729 * @inode: inode of metadata file using this allocator
730 * @entry_nrs: array of entry numbers to be deallocated
731 * @nitems: number of entries stored in @entry_nrs
732 */
733int nilfs_palloc_freev(struct inode *inode, __u64 *entry_nrs, size_t nitems)
734{
735 struct buffer_head *desc_bh, *bitmap_bh;
736 struct nilfs_palloc_group_desc *desc;
737 unsigned char *bitmap;
738 void *desc_kaddr, *bitmap_kaddr;
739 unsigned long group, group_offset;
740 __u64 group_min_nr, last_nrs[8];
741 const unsigned long epg = nilfs_palloc_entries_per_group(inode);
742 const unsigned int epb = NILFS_MDT(inode)->mi_entries_per_block;
743 unsigned int entry_start, end, pos;
744 spinlock_t *lock;
745 int i, j, k, ret;
746 u32 nfree;
747
748 for (i = 0; i < nitems; i = j) {
749 int change_group = false;
750 int nempties = 0, n = 0;
751
752 group = nilfs_palloc_group(inode, entry_nrs[i], &group_offset);
753 ret = nilfs_palloc_get_desc_block(inode, group, 0, &desc_bh);
754 if (ret < 0)
755 return ret;
756 ret = nilfs_palloc_get_bitmap_block(inode, group, 0,
757 &bitmap_bh);
758 if (ret < 0) {
759 brelse(desc_bh);
760 return ret;
761 }
762
763 /* Get the first entry number of the group */
764 group_min_nr = (__u64)group * epg;
765
766 bitmap_kaddr = kmap(bitmap_bh->b_page);
767 bitmap = bitmap_kaddr + bh_offset(bitmap_bh);
768 lock = nilfs_mdt_bgl_lock(inode, group);
769
770 j = i;
771 entry_start = rounddown(group_offset, epb);
772 do {
773 if (!nilfs_clear_bit_atomic(lock, group_offset,
774 bitmap)) {
775 nilfs_msg(inode->i_sb, KERN_WARNING,
776 "%s (ino=%lu): entry number %llu already freed",
777 __func__, inode->i_ino,
778 (unsigned long long)entry_nrs[j]);
779 } else {
780 n++;
781 }
782
783 j++;
784 if (j >= nitems || entry_nrs[j] < group_min_nr ||
785 entry_nrs[j] >= group_min_nr + epg) {
786 change_group = true;
787 } else {
788 group_offset = entry_nrs[j] - group_min_nr;
789 if (group_offset >= entry_start &&
790 group_offset < entry_start + epb) {
791 /* This entry is in the same block */
792 continue;
793 }
794 }
795
796 /* Test if the entry block is empty or not */
797 end = entry_start + epb;
798 pos = nilfs_find_next_bit(bitmap, end, entry_start);
799 if (pos >= end) {
800 last_nrs[nempties++] = entry_nrs[j - 1];
801 if (nempties >= ARRAY_SIZE(last_nrs))
802 break;
803 }
804
805 if (change_group)
806 break;
807
808 /* Go on to the next entry block */
809 entry_start = rounddown(group_offset, epb);
810 } while (true);
811
812 kunmap(bitmap_bh->b_page);
813 mark_buffer_dirty(bitmap_bh);
814 brelse(bitmap_bh);
815
816 for (k = 0; k < nempties; k++) {
817 ret = nilfs_palloc_delete_entry_block(inode,
818 last_nrs[k]);
819 if (ret && ret != -ENOENT)
820 nilfs_msg(inode->i_sb, KERN_WARNING,
821 "error %d deleting block that object (entry=%llu, ino=%lu) belongs to",
822 ret, (unsigned long long)last_nrs[k],
823 inode->i_ino);
824 }
825
826 desc_kaddr = kmap_atomic(desc_bh->b_page);
827 desc = nilfs_palloc_block_get_group_desc(
828 inode, group, desc_bh, desc_kaddr);
829 nfree = nilfs_palloc_group_desc_add_entries(desc, lock, n);
830 kunmap_atomic(desc_kaddr);
831 mark_buffer_dirty(desc_bh);
832 nilfs_mdt_mark_dirty(inode);
833 brelse(desc_bh);
834
835 if (nfree == nilfs_palloc_entries_per_group(inode)) {
836 ret = nilfs_palloc_delete_bitmap_block(inode, group);
837 if (ret && ret != -ENOENT)
838 nilfs_msg(inode->i_sb, KERN_WARNING,
839 "error %d deleting bitmap block of group=%lu, ino=%lu",
840 ret, group, inode->i_ino);
841 }
842 }
843 return 0;
844}
845
846void nilfs_palloc_setup_cache(struct inode *inode,
847 struct nilfs_palloc_cache *cache)
848{
849 NILFS_MDT(inode)->mi_palloc_cache = cache;
850 spin_lock_init(&cache->lock);
851}
852
853void nilfs_palloc_clear_cache(struct inode *inode)
854{
855 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
856
857 spin_lock(&cache->lock);
858 brelse(cache->prev_desc.bh);
859 brelse(cache->prev_bitmap.bh);
860 brelse(cache->prev_entry.bh);
861 cache->prev_desc.bh = NULL;
862 cache->prev_bitmap.bh = NULL;
863 cache->prev_entry.bh = NULL;
864 spin_unlock(&cache->lock);
865}
866
867void nilfs_palloc_destroy_cache(struct inode *inode)
868{
869 nilfs_palloc_clear_cache(inode);
870 NILFS_MDT(inode)->mi_palloc_cache = NULL;
871}
1/*
2 * alloc.c - NILFS dat/inode allocator
3 *
4 * Copyright (C) 2006-2008 Nippon Telegraph and Telephone Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 *
20 * Original code was written by Koji Sato <koji@osrg.net>.
21 * Two allocators were unified by Ryusuke Konishi <ryusuke@osrg.net>,
22 * Amagai Yoshiji <amagai@osrg.net>.
23 */
24
25#include <linux/types.h>
26#include <linux/buffer_head.h>
27#include <linux/fs.h>
28#include <linux/bitops.h>
29#include <linux/slab.h>
30#include "mdt.h"
31#include "alloc.h"
32
33
34/**
35 * nilfs_palloc_groups_per_desc_block - get the number of groups that a group
36 * descriptor block can maintain
37 * @inode: inode of metadata file using this allocator
38 */
39static inline unsigned long
40nilfs_palloc_groups_per_desc_block(const struct inode *inode)
41{
42 return (1UL << inode->i_blkbits) /
43 sizeof(struct nilfs_palloc_group_desc);
44}
45
46/**
47 * nilfs_palloc_groups_count - get maximum number of groups
48 * @inode: inode of metadata file using this allocator
49 */
50static inline unsigned long
51nilfs_palloc_groups_count(const struct inode *inode)
52{
53 return 1UL << (BITS_PER_LONG - (inode->i_blkbits + 3 /* log2(8) */));
54}
55
56/**
57 * nilfs_palloc_init_blockgroup - initialize private variables for allocator
58 * @inode: inode of metadata file using this allocator
59 * @entry_size: size of the persistent object
60 */
61int nilfs_palloc_init_blockgroup(struct inode *inode, unsigned entry_size)
62{
63 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
64
65 mi->mi_bgl = kmalloc(sizeof(*mi->mi_bgl), GFP_NOFS);
66 if (!mi->mi_bgl)
67 return -ENOMEM;
68
69 bgl_lock_init(mi->mi_bgl);
70
71 nilfs_mdt_set_entry_size(inode, entry_size, 0);
72
73 mi->mi_blocks_per_group =
74 DIV_ROUND_UP(nilfs_palloc_entries_per_group(inode),
75 mi->mi_entries_per_block) + 1;
76 /* Number of blocks in a group including entry blocks and
77 a bitmap block */
78 mi->mi_blocks_per_desc_block =
79 nilfs_palloc_groups_per_desc_block(inode) *
80 mi->mi_blocks_per_group + 1;
81 /* Number of blocks per descriptor including the
82 descriptor block */
83 return 0;
84}
85
86/**
87 * nilfs_palloc_group - get group number and offset from an entry number
88 * @inode: inode of metadata file using this allocator
89 * @nr: serial number of the entry (e.g. inode number)
90 * @offset: pointer to store offset number in the group
91 */
92static unsigned long nilfs_palloc_group(const struct inode *inode, __u64 nr,
93 unsigned long *offset)
94{
95 __u64 group = nr;
96
97 *offset = do_div(group, nilfs_palloc_entries_per_group(inode));
98 return group;
99}
100
101/**
102 * nilfs_palloc_desc_blkoff - get block offset of a group descriptor block
103 * @inode: inode of metadata file using this allocator
104 * @group: group number
105 *
106 * nilfs_palloc_desc_blkoff() returns block offset of the descriptor
107 * block which contains a descriptor of the specified group.
108 */
109static unsigned long
110nilfs_palloc_desc_blkoff(const struct inode *inode, unsigned long group)
111{
112 unsigned long desc_block =
113 group / nilfs_palloc_groups_per_desc_block(inode);
114 return desc_block * NILFS_MDT(inode)->mi_blocks_per_desc_block;
115}
116
117/**
118 * nilfs_palloc_bitmap_blkoff - get block offset of a bitmap block
119 * @inode: inode of metadata file using this allocator
120 * @group: group number
121 *
122 * nilfs_palloc_bitmap_blkoff() returns block offset of the bitmap
123 * block used to allocate/deallocate entries in the specified group.
124 */
125static unsigned long
126nilfs_palloc_bitmap_blkoff(const struct inode *inode, unsigned long group)
127{
128 unsigned long desc_offset =
129 group % nilfs_palloc_groups_per_desc_block(inode);
130 return nilfs_palloc_desc_blkoff(inode, group) + 1 +
131 desc_offset * NILFS_MDT(inode)->mi_blocks_per_group;
132}
133
134/**
135 * nilfs_palloc_group_desc_nfrees - get the number of free entries in a group
136 * @desc: pointer to descriptor structure for the group
137 * @lock: spin lock protecting @desc
138 */
139static unsigned long
140nilfs_palloc_group_desc_nfrees(const struct nilfs_palloc_group_desc *desc,
141 spinlock_t *lock)
142{
143 unsigned long nfree;
144
145 spin_lock(lock);
146 nfree = le32_to_cpu(desc->pg_nfrees);
147 spin_unlock(lock);
148 return nfree;
149}
150
151/**
152 * nilfs_palloc_group_desc_add_entries - adjust count of free entries
153 * @desc: pointer to descriptor structure for the group
154 * @lock: spin lock protecting @desc
155 * @n: delta to be added
156 */
157static u32
158nilfs_palloc_group_desc_add_entries(struct nilfs_palloc_group_desc *desc,
159 spinlock_t *lock, u32 n)
160{
161 u32 nfree;
162
163 spin_lock(lock);
164 le32_add_cpu(&desc->pg_nfrees, n);
165 nfree = le32_to_cpu(desc->pg_nfrees);
166 spin_unlock(lock);
167 return nfree;
168}
169
170/**
171 * nilfs_palloc_entry_blkoff - get block offset of an entry block
172 * @inode: inode of metadata file using this allocator
173 * @nr: serial number of the entry (e.g. inode number)
174 */
175static unsigned long
176nilfs_palloc_entry_blkoff(const struct inode *inode, __u64 nr)
177{
178 unsigned long group, group_offset;
179
180 group = nilfs_palloc_group(inode, nr, &group_offset);
181
182 return nilfs_palloc_bitmap_blkoff(inode, group) + 1 +
183 group_offset / NILFS_MDT(inode)->mi_entries_per_block;
184}
185
186/**
187 * nilfs_palloc_desc_block_init - initialize buffer of a group descriptor block
188 * @inode: inode of metadata file
189 * @bh: buffer head of the buffer to be initialized
190 * @kaddr: kernel address mapped for the page including the buffer
191 */
192static void nilfs_palloc_desc_block_init(struct inode *inode,
193 struct buffer_head *bh, void *kaddr)
194{
195 struct nilfs_palloc_group_desc *desc = kaddr + bh_offset(bh);
196 unsigned long n = nilfs_palloc_groups_per_desc_block(inode);
197 __le32 nfrees;
198
199 nfrees = cpu_to_le32(nilfs_palloc_entries_per_group(inode));
200 while (n-- > 0) {
201 desc->pg_nfrees = nfrees;
202 desc++;
203 }
204}
205
206static int nilfs_palloc_get_block(struct inode *inode, unsigned long blkoff,
207 int create,
208 void (*init_block)(struct inode *,
209 struct buffer_head *,
210 void *),
211 struct buffer_head **bhp,
212 struct nilfs_bh_assoc *prev,
213 spinlock_t *lock)
214{
215 int ret;
216
217 spin_lock(lock);
218 if (prev->bh && blkoff == prev->blkoff) {
219 get_bh(prev->bh);
220 *bhp = prev->bh;
221 spin_unlock(lock);
222 return 0;
223 }
224 spin_unlock(lock);
225
226 ret = nilfs_mdt_get_block(inode, blkoff, create, init_block, bhp);
227 if (!ret) {
228 spin_lock(lock);
229 /*
230 * The following code must be safe for change of the
231 * cache contents during the get block call.
232 */
233 brelse(prev->bh);
234 get_bh(*bhp);
235 prev->bh = *bhp;
236 prev->blkoff = blkoff;
237 spin_unlock(lock);
238 }
239 return ret;
240}
241
242/**
243 * nilfs_palloc_delete_block - delete a block on the persistent allocator file
244 * @inode: inode of metadata file using this allocator
245 * @blkoff: block offset
246 * @prev: nilfs_bh_assoc struct of the last used buffer
247 * @lock: spin lock protecting @prev
248 */
249static int nilfs_palloc_delete_block(struct inode *inode, unsigned long blkoff,
250 struct nilfs_bh_assoc *prev,
251 spinlock_t *lock)
252{
253 spin_lock(lock);
254 if (prev->bh && blkoff == prev->blkoff) {
255 brelse(prev->bh);
256 prev->bh = NULL;
257 }
258 spin_unlock(lock);
259 return nilfs_mdt_delete_block(inode, blkoff);
260}
261
262/**
263 * nilfs_palloc_get_desc_block - get buffer head of a group descriptor block
264 * @inode: inode of metadata file using this allocator
265 * @group: group number
266 * @create: create flag
267 * @bhp: pointer to store the resultant buffer head
268 */
269static int nilfs_palloc_get_desc_block(struct inode *inode,
270 unsigned long group,
271 int create, struct buffer_head **bhp)
272{
273 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
274
275 return nilfs_palloc_get_block(inode,
276 nilfs_palloc_desc_blkoff(inode, group),
277 create, nilfs_palloc_desc_block_init,
278 bhp, &cache->prev_desc, &cache->lock);
279}
280
281/**
282 * nilfs_palloc_get_bitmap_block - get buffer head of a bitmap block
283 * @inode: inode of metadata file using this allocator
284 * @group: group number
285 * @create: create flag
286 * @bhp: pointer to store the resultant buffer head
287 */
288static int nilfs_palloc_get_bitmap_block(struct inode *inode,
289 unsigned long group,
290 int create, struct buffer_head **bhp)
291{
292 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
293
294 return nilfs_palloc_get_block(inode,
295 nilfs_palloc_bitmap_blkoff(inode, group),
296 create, NULL, bhp,
297 &cache->prev_bitmap, &cache->lock);
298}
299
300/**
301 * nilfs_palloc_delete_bitmap_block - delete a bitmap block
302 * @inode: inode of metadata file using this allocator
303 * @group: group number
304 */
305static int nilfs_palloc_delete_bitmap_block(struct inode *inode,
306 unsigned long group)
307{
308 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
309
310 return nilfs_palloc_delete_block(inode,
311 nilfs_palloc_bitmap_blkoff(inode,
312 group),
313 &cache->prev_bitmap, &cache->lock);
314}
315
316/**
317 * nilfs_palloc_get_entry_block - get buffer head of an entry block
318 * @inode: inode of metadata file using this allocator
319 * @nr: serial number of the entry (e.g. inode number)
320 * @create: create flag
321 * @bhp: pointer to store the resultant buffer head
322 */
323int nilfs_palloc_get_entry_block(struct inode *inode, __u64 nr,
324 int create, struct buffer_head **bhp)
325{
326 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
327
328 return nilfs_palloc_get_block(inode,
329 nilfs_palloc_entry_blkoff(inode, nr),
330 create, NULL, bhp,
331 &cache->prev_entry, &cache->lock);
332}
333
334/**
335 * nilfs_palloc_delete_entry_block - delete an entry block
336 * @inode: inode of metadata file using this allocator
337 * @nr: serial number of the entry
338 */
339static int nilfs_palloc_delete_entry_block(struct inode *inode, __u64 nr)
340{
341 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
342
343 return nilfs_palloc_delete_block(inode,
344 nilfs_palloc_entry_blkoff(inode, nr),
345 &cache->prev_entry, &cache->lock);
346}
347
348/**
349 * nilfs_palloc_block_get_group_desc - get kernel address of a group descriptor
350 * @inode: inode of metadata file using this allocator
351 * @group: group number
352 * @bh: buffer head of the buffer storing the group descriptor block
353 * @kaddr: kernel address mapped for the page including the buffer
354 */
355static struct nilfs_palloc_group_desc *
356nilfs_palloc_block_get_group_desc(const struct inode *inode,
357 unsigned long group,
358 const struct buffer_head *bh, void *kaddr)
359{
360 return (struct nilfs_palloc_group_desc *)(kaddr + bh_offset(bh)) +
361 group % nilfs_palloc_groups_per_desc_block(inode);
362}
363
364/**
365 * nilfs_palloc_block_get_entry - get kernel address of an entry
366 * @inode: inode of metadata file using this allocator
367 * @nr: serial number of the entry (e.g. inode number)
368 * @bh: buffer head of the buffer storing the entry block
369 * @kaddr: kernel address mapped for the page including the buffer
370 */
371void *nilfs_palloc_block_get_entry(const struct inode *inode, __u64 nr,
372 const struct buffer_head *bh, void *kaddr)
373{
374 unsigned long entry_offset, group_offset;
375
376 nilfs_palloc_group(inode, nr, &group_offset);
377 entry_offset = group_offset % NILFS_MDT(inode)->mi_entries_per_block;
378
379 return kaddr + bh_offset(bh) +
380 entry_offset * NILFS_MDT(inode)->mi_entry_size;
381}
382
383/**
384 * nilfs_palloc_find_available_slot - find available slot in a group
385 * @bitmap: bitmap of the group
386 * @target: offset number of an entry in the group (start point)
387 * @bsize: size in bits
388 * @lock: spin lock protecting @bitmap
389 */
390static int nilfs_palloc_find_available_slot(unsigned char *bitmap,
391 unsigned long target,
392 unsigned bsize,
393 spinlock_t *lock)
394{
395 int pos, end = bsize;
396
397 if (likely(target < bsize)) {
398 pos = target;
399 do {
400 pos = nilfs_find_next_zero_bit(bitmap, end, pos);
401 if (pos >= end)
402 break;
403 if (!nilfs_set_bit_atomic(lock, pos, bitmap))
404 return pos;
405 } while (++pos < end);
406
407 end = target;
408 }
409
410 /* wrap around */
411 for (pos = 0; pos < end; pos++) {
412 pos = nilfs_find_next_zero_bit(bitmap, end, pos);
413 if (pos >= end)
414 break;
415 if (!nilfs_set_bit_atomic(lock, pos, bitmap))
416 return pos;
417 }
418
419 return -ENOSPC;
420}
421
422/**
423 * nilfs_palloc_rest_groups_in_desc_block - get the remaining number of groups
424 * in a group descriptor block
425 * @inode: inode of metadata file using this allocator
426 * @curr: current group number
427 * @max: maximum number of groups
428 */
429static unsigned long
430nilfs_palloc_rest_groups_in_desc_block(const struct inode *inode,
431 unsigned long curr, unsigned long max)
432{
433 return min_t(unsigned long,
434 nilfs_palloc_groups_per_desc_block(inode) -
435 curr % nilfs_palloc_groups_per_desc_block(inode),
436 max - curr + 1);
437}
438
439/**
440 * nilfs_palloc_count_desc_blocks - count descriptor blocks number
441 * @inode: inode of metadata file using this allocator
442 * @desc_blocks: descriptor blocks number [out]
443 */
444static int nilfs_palloc_count_desc_blocks(struct inode *inode,
445 unsigned long *desc_blocks)
446{
447 __u64 blknum;
448 int ret;
449
450 ret = nilfs_bmap_last_key(NILFS_I(inode)->i_bmap, &blknum);
451 if (likely(!ret))
452 *desc_blocks = DIV_ROUND_UP(
453 (unsigned long)blknum,
454 NILFS_MDT(inode)->mi_blocks_per_desc_block);
455 return ret;
456}
457
458/**
459 * nilfs_palloc_mdt_file_can_grow - check potential opportunity for
460 * MDT file growing
461 * @inode: inode of metadata file using this allocator
462 * @desc_blocks: known current descriptor blocks count
463 */
464static inline bool nilfs_palloc_mdt_file_can_grow(struct inode *inode,
465 unsigned long desc_blocks)
466{
467 return (nilfs_palloc_groups_per_desc_block(inode) * desc_blocks) <
468 nilfs_palloc_groups_count(inode);
469}
470
471/**
472 * nilfs_palloc_count_max_entries - count max number of entries that can be
473 * described by descriptor blocks count
474 * @inode: inode of metadata file using this allocator
475 * @nused: current number of used entries
476 * @nmaxp: max number of entries [out]
477 */
478int nilfs_palloc_count_max_entries(struct inode *inode, u64 nused, u64 *nmaxp)
479{
480 unsigned long desc_blocks = 0;
481 u64 entries_per_desc_block, nmax;
482 int err;
483
484 err = nilfs_palloc_count_desc_blocks(inode, &desc_blocks);
485 if (unlikely(err))
486 return err;
487
488 entries_per_desc_block = (u64)nilfs_palloc_entries_per_group(inode) *
489 nilfs_palloc_groups_per_desc_block(inode);
490 nmax = entries_per_desc_block * desc_blocks;
491
492 if (nused == nmax &&
493 nilfs_palloc_mdt_file_can_grow(inode, desc_blocks))
494 nmax += entries_per_desc_block;
495
496 if (nused > nmax)
497 return -ERANGE;
498
499 *nmaxp = nmax;
500 return 0;
501}
502
503/**
504 * nilfs_palloc_prepare_alloc_entry - prepare to allocate a persistent object
505 * @inode: inode of metadata file using this allocator
506 * @req: nilfs_palloc_req structure exchanged for the allocation
507 */
508int nilfs_palloc_prepare_alloc_entry(struct inode *inode,
509 struct nilfs_palloc_req *req)
510{
511 struct buffer_head *desc_bh, *bitmap_bh;
512 struct nilfs_palloc_group_desc *desc;
513 unsigned char *bitmap;
514 void *desc_kaddr, *bitmap_kaddr;
515 unsigned long group, maxgroup, ngroups;
516 unsigned long group_offset, maxgroup_offset;
517 unsigned long n, entries_per_group;
518 unsigned long i, j;
519 spinlock_t *lock;
520 int pos, ret;
521
522 ngroups = nilfs_palloc_groups_count(inode);
523 maxgroup = ngroups - 1;
524 group = nilfs_palloc_group(inode, req->pr_entry_nr, &group_offset);
525 entries_per_group = nilfs_palloc_entries_per_group(inode);
526
527 for (i = 0; i < ngroups; i += n) {
528 if (group >= ngroups) {
529 /* wrap around */
530 group = 0;
531 maxgroup = nilfs_palloc_group(inode, req->pr_entry_nr,
532 &maxgroup_offset) - 1;
533 }
534 ret = nilfs_palloc_get_desc_block(inode, group, 1, &desc_bh);
535 if (ret < 0)
536 return ret;
537 desc_kaddr = kmap(desc_bh->b_page);
538 desc = nilfs_palloc_block_get_group_desc(
539 inode, group, desc_bh, desc_kaddr);
540 n = nilfs_palloc_rest_groups_in_desc_block(inode, group,
541 maxgroup);
542 for (j = 0; j < n; j++, desc++, group++) {
543 lock = nilfs_mdt_bgl_lock(inode, group);
544 if (nilfs_palloc_group_desc_nfrees(desc, lock) > 0) {
545 ret = nilfs_palloc_get_bitmap_block(
546 inode, group, 1, &bitmap_bh);
547 if (ret < 0)
548 goto out_desc;
549 bitmap_kaddr = kmap(bitmap_bh->b_page);
550 bitmap = bitmap_kaddr + bh_offset(bitmap_bh);
551 pos = nilfs_palloc_find_available_slot(
552 bitmap, group_offset,
553 entries_per_group, lock);
554 if (pos >= 0) {
555 /* found a free entry */
556 nilfs_palloc_group_desc_add_entries(
557 desc, lock, -1);
558 req->pr_entry_nr =
559 entries_per_group * group + pos;
560 kunmap(desc_bh->b_page);
561 kunmap(bitmap_bh->b_page);
562
563 req->pr_desc_bh = desc_bh;
564 req->pr_bitmap_bh = bitmap_bh;
565 return 0;
566 }
567 kunmap(bitmap_bh->b_page);
568 brelse(bitmap_bh);
569 }
570
571 group_offset = 0;
572 }
573
574 kunmap(desc_bh->b_page);
575 brelse(desc_bh);
576 }
577
578 /* no entries left */
579 return -ENOSPC;
580
581 out_desc:
582 kunmap(desc_bh->b_page);
583 brelse(desc_bh);
584 return ret;
585}
586
587/**
588 * nilfs_palloc_commit_alloc_entry - finish allocation of a persistent object
589 * @inode: inode of metadata file using this allocator
590 * @req: nilfs_palloc_req structure exchanged for the allocation
591 */
592void nilfs_palloc_commit_alloc_entry(struct inode *inode,
593 struct nilfs_palloc_req *req)
594{
595 mark_buffer_dirty(req->pr_bitmap_bh);
596 mark_buffer_dirty(req->pr_desc_bh);
597 nilfs_mdt_mark_dirty(inode);
598
599 brelse(req->pr_bitmap_bh);
600 brelse(req->pr_desc_bh);
601}
602
603/**
604 * nilfs_palloc_commit_free_entry - finish deallocating a persistent object
605 * @inode: inode of metadata file using this allocator
606 * @req: nilfs_palloc_req structure exchanged for the removal
607 */
608void nilfs_palloc_commit_free_entry(struct inode *inode,
609 struct nilfs_palloc_req *req)
610{
611 struct nilfs_palloc_group_desc *desc;
612 unsigned long group, group_offset;
613 unsigned char *bitmap;
614 void *desc_kaddr, *bitmap_kaddr;
615 spinlock_t *lock;
616
617 group = nilfs_palloc_group(inode, req->pr_entry_nr, &group_offset);
618 desc_kaddr = kmap(req->pr_desc_bh->b_page);
619 desc = nilfs_palloc_block_get_group_desc(inode, group,
620 req->pr_desc_bh, desc_kaddr);
621 bitmap_kaddr = kmap(req->pr_bitmap_bh->b_page);
622 bitmap = bitmap_kaddr + bh_offset(req->pr_bitmap_bh);
623 lock = nilfs_mdt_bgl_lock(inode, group);
624
625 if (!nilfs_clear_bit_atomic(lock, group_offset, bitmap))
626 nilfs_warning(inode->i_sb, __func__,
627 "entry number %llu already freed: ino=%lu\n",
628 (unsigned long long)req->pr_entry_nr,
629 (unsigned long)inode->i_ino);
630 else
631 nilfs_palloc_group_desc_add_entries(desc, lock, 1);
632
633 kunmap(req->pr_bitmap_bh->b_page);
634 kunmap(req->pr_desc_bh->b_page);
635
636 mark_buffer_dirty(req->pr_desc_bh);
637 mark_buffer_dirty(req->pr_bitmap_bh);
638 nilfs_mdt_mark_dirty(inode);
639
640 brelse(req->pr_bitmap_bh);
641 brelse(req->pr_desc_bh);
642}
643
644/**
645 * nilfs_palloc_abort_alloc_entry - cancel allocation of a persistent object
646 * @inode: inode of metadata file using this allocator
647 * @req: nilfs_palloc_req structure exchanged for the allocation
648 */
649void nilfs_palloc_abort_alloc_entry(struct inode *inode,
650 struct nilfs_palloc_req *req)
651{
652 struct nilfs_palloc_group_desc *desc;
653 void *desc_kaddr, *bitmap_kaddr;
654 unsigned char *bitmap;
655 unsigned long group, group_offset;
656 spinlock_t *lock;
657
658 group = nilfs_palloc_group(inode, req->pr_entry_nr, &group_offset);
659 desc_kaddr = kmap(req->pr_desc_bh->b_page);
660 desc = nilfs_palloc_block_get_group_desc(inode, group,
661 req->pr_desc_bh, desc_kaddr);
662 bitmap_kaddr = kmap(req->pr_bitmap_bh->b_page);
663 bitmap = bitmap_kaddr + bh_offset(req->pr_bitmap_bh);
664 lock = nilfs_mdt_bgl_lock(inode, group);
665
666 if (!nilfs_clear_bit_atomic(lock, group_offset, bitmap))
667 nilfs_warning(inode->i_sb, __func__,
668 "entry number %llu already freed: ino=%lu\n",
669 (unsigned long long)req->pr_entry_nr,
670 (unsigned long)inode->i_ino);
671 else
672 nilfs_palloc_group_desc_add_entries(desc, lock, 1);
673
674 kunmap(req->pr_bitmap_bh->b_page);
675 kunmap(req->pr_desc_bh->b_page);
676
677 brelse(req->pr_bitmap_bh);
678 brelse(req->pr_desc_bh);
679
680 req->pr_entry_nr = 0;
681 req->pr_bitmap_bh = NULL;
682 req->pr_desc_bh = NULL;
683}
684
685/**
686 * nilfs_palloc_prepare_free_entry - prepare to deallocate a persistent object
687 * @inode: inode of metadata file using this allocator
688 * @req: nilfs_palloc_req structure exchanged for the removal
689 */
690int nilfs_palloc_prepare_free_entry(struct inode *inode,
691 struct nilfs_palloc_req *req)
692{
693 struct buffer_head *desc_bh, *bitmap_bh;
694 unsigned long group, group_offset;
695 int ret;
696
697 group = nilfs_palloc_group(inode, req->pr_entry_nr, &group_offset);
698 ret = nilfs_palloc_get_desc_block(inode, group, 1, &desc_bh);
699 if (ret < 0)
700 return ret;
701 ret = nilfs_palloc_get_bitmap_block(inode, group, 1, &bitmap_bh);
702 if (ret < 0) {
703 brelse(desc_bh);
704 return ret;
705 }
706
707 req->pr_desc_bh = desc_bh;
708 req->pr_bitmap_bh = bitmap_bh;
709 return 0;
710}
711
712/**
713 * nilfs_palloc_abort_free_entry - cancel deallocating a persistent object
714 * @inode: inode of metadata file using this allocator
715 * @req: nilfs_palloc_req structure exchanged for the removal
716 */
717void nilfs_palloc_abort_free_entry(struct inode *inode,
718 struct nilfs_palloc_req *req)
719{
720 brelse(req->pr_bitmap_bh);
721 brelse(req->pr_desc_bh);
722
723 req->pr_entry_nr = 0;
724 req->pr_bitmap_bh = NULL;
725 req->pr_desc_bh = NULL;
726}
727
728/**
729 * nilfs_palloc_freev - deallocate a set of persistent objects
730 * @inode: inode of metadata file using this allocator
731 * @entry_nrs: array of entry numbers to be deallocated
732 * @nitems: number of entries stored in @entry_nrs
733 */
734int nilfs_palloc_freev(struct inode *inode, __u64 *entry_nrs, size_t nitems)
735{
736 struct buffer_head *desc_bh, *bitmap_bh;
737 struct nilfs_palloc_group_desc *desc;
738 unsigned char *bitmap;
739 void *desc_kaddr, *bitmap_kaddr;
740 unsigned long group, group_offset;
741 __u64 group_min_nr, last_nrs[8];
742 const unsigned long epg = nilfs_palloc_entries_per_group(inode);
743 const unsigned epb = NILFS_MDT(inode)->mi_entries_per_block;
744 unsigned entry_start, end, pos;
745 spinlock_t *lock;
746 int i, j, k, ret;
747 u32 nfree;
748
749 for (i = 0; i < nitems; i = j) {
750 int change_group = false;
751 int nempties = 0, n = 0;
752
753 group = nilfs_palloc_group(inode, entry_nrs[i], &group_offset);
754 ret = nilfs_palloc_get_desc_block(inode, group, 0, &desc_bh);
755 if (ret < 0)
756 return ret;
757 ret = nilfs_palloc_get_bitmap_block(inode, group, 0,
758 &bitmap_bh);
759 if (ret < 0) {
760 brelse(desc_bh);
761 return ret;
762 }
763
764 /* Get the first entry number of the group */
765 group_min_nr = (__u64)group * epg;
766
767 bitmap_kaddr = kmap(bitmap_bh->b_page);
768 bitmap = bitmap_kaddr + bh_offset(bitmap_bh);
769 lock = nilfs_mdt_bgl_lock(inode, group);
770
771 j = i;
772 entry_start = rounddown(group_offset, epb);
773 do {
774 if (!nilfs_clear_bit_atomic(lock, group_offset,
775 bitmap)) {
776 nilfs_warning(inode->i_sb, __func__,
777 "entry number %llu already freed: ino=%lu\n",
778 (unsigned long long)entry_nrs[j],
779 (unsigned long)inode->i_ino);
780 } else {
781 n++;
782 }
783
784 j++;
785 if (j >= nitems || entry_nrs[j] < group_min_nr ||
786 entry_nrs[j] >= group_min_nr + epg) {
787 change_group = true;
788 } else {
789 group_offset = entry_nrs[j] - group_min_nr;
790 if (group_offset >= entry_start &&
791 group_offset < entry_start + epb) {
792 /* This entry is in the same block */
793 continue;
794 }
795 }
796
797 /* Test if the entry block is empty or not */
798 end = entry_start + epb;
799 pos = nilfs_find_next_bit(bitmap, end, entry_start);
800 if (pos >= end) {
801 last_nrs[nempties++] = entry_nrs[j - 1];
802 if (nempties >= ARRAY_SIZE(last_nrs))
803 break;
804 }
805
806 if (change_group)
807 break;
808
809 /* Go on to the next entry block */
810 entry_start = rounddown(group_offset, epb);
811 } while (true);
812
813 kunmap(bitmap_bh->b_page);
814 mark_buffer_dirty(bitmap_bh);
815 brelse(bitmap_bh);
816
817 for (k = 0; k < nempties; k++) {
818 ret = nilfs_palloc_delete_entry_block(inode,
819 last_nrs[k]);
820 if (ret && ret != -ENOENT) {
821 nilfs_warning(inode->i_sb, __func__,
822 "failed to delete block of entry %llu: ino=%lu, err=%d\n",
823 (unsigned long long)last_nrs[k],
824 (unsigned long)inode->i_ino, ret);
825 }
826 }
827
828 desc_kaddr = kmap_atomic(desc_bh->b_page);
829 desc = nilfs_palloc_block_get_group_desc(
830 inode, group, desc_bh, desc_kaddr);
831 nfree = nilfs_palloc_group_desc_add_entries(desc, lock, n);
832 kunmap_atomic(desc_kaddr);
833 mark_buffer_dirty(desc_bh);
834 nilfs_mdt_mark_dirty(inode);
835 brelse(desc_bh);
836
837 if (nfree == nilfs_palloc_entries_per_group(inode)) {
838 ret = nilfs_palloc_delete_bitmap_block(inode, group);
839 if (ret && ret != -ENOENT) {
840 nilfs_warning(inode->i_sb, __func__,
841 "failed to delete bitmap block of group %lu: ino=%lu, err=%d\n",
842 group,
843 (unsigned long)inode->i_ino, ret);
844 }
845 }
846 }
847 return 0;
848}
849
850void nilfs_palloc_setup_cache(struct inode *inode,
851 struct nilfs_palloc_cache *cache)
852{
853 NILFS_MDT(inode)->mi_palloc_cache = cache;
854 spin_lock_init(&cache->lock);
855}
856
857void nilfs_palloc_clear_cache(struct inode *inode)
858{
859 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
860
861 spin_lock(&cache->lock);
862 brelse(cache->prev_desc.bh);
863 brelse(cache->prev_bitmap.bh);
864 brelse(cache->prev_entry.bh);
865 cache->prev_desc.bh = NULL;
866 cache->prev_bitmap.bh = NULL;
867 cache->prev_entry.bh = NULL;
868 spin_unlock(&cache->lock);
869}
870
871void nilfs_palloc_destroy_cache(struct inode *inode)
872{
873 nilfs_palloc_clear_cache(inode);
874 NILFS_MDT(inode)->mi_palloc_cache = NULL;
875}