Loading...
1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * NILFS dat/inode allocator
4 *
5 * Copyright (C) 2006-2008 Nippon Telegraph and Telephone Corporation.
6 *
7 * Originally written by Koji Sato.
8 * Two allocators were unified by Ryusuke Konishi and Amagai Yoshiji.
9 */
10
11#include <linux/types.h>
12#include <linux/buffer_head.h>
13#include <linux/fs.h>
14#include <linux/bitops.h>
15#include <linux/slab.h>
16#include "mdt.h"
17#include "alloc.h"
18
19
20/**
21 * nilfs_palloc_groups_per_desc_block - get the number of groups that a group
22 * descriptor block can maintain
23 * @inode: inode of metadata file using this allocator
24 */
25static inline unsigned long
26nilfs_palloc_groups_per_desc_block(const struct inode *inode)
27{
28 return i_blocksize(inode) /
29 sizeof(struct nilfs_palloc_group_desc);
30}
31
32/**
33 * nilfs_palloc_groups_count - get maximum number of groups
34 * @inode: inode of metadata file using this allocator
35 */
36static inline unsigned long
37nilfs_palloc_groups_count(const struct inode *inode)
38{
39 return 1UL << (BITS_PER_LONG - (inode->i_blkbits + 3 /* log2(8) */));
40}
41
42/**
43 * nilfs_palloc_init_blockgroup - initialize private variables for allocator
44 * @inode: inode of metadata file using this allocator
45 * @entry_size: size of the persistent object
46 */
47int nilfs_palloc_init_blockgroup(struct inode *inode, unsigned int entry_size)
48{
49 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
50
51 mi->mi_bgl = kmalloc(sizeof(*mi->mi_bgl), GFP_NOFS);
52 if (!mi->mi_bgl)
53 return -ENOMEM;
54
55 bgl_lock_init(mi->mi_bgl);
56
57 nilfs_mdt_set_entry_size(inode, entry_size, 0);
58
59 mi->mi_blocks_per_group =
60 DIV_ROUND_UP(nilfs_palloc_entries_per_group(inode),
61 mi->mi_entries_per_block) + 1;
62 /*
63 * Number of blocks in a group including entry blocks
64 * and a bitmap block
65 */
66 mi->mi_blocks_per_desc_block =
67 nilfs_palloc_groups_per_desc_block(inode) *
68 mi->mi_blocks_per_group + 1;
69 /*
70 * Number of blocks per descriptor including the
71 * descriptor block
72 */
73 return 0;
74}
75
76/**
77 * nilfs_palloc_group - get group number and offset from an entry number
78 * @inode: inode of metadata file using this allocator
79 * @nr: serial number of the entry (e.g. inode number)
80 * @offset: pointer to store offset number in the group
81 */
82static unsigned long nilfs_palloc_group(const struct inode *inode, __u64 nr,
83 unsigned long *offset)
84{
85 __u64 group = nr;
86
87 *offset = do_div(group, nilfs_palloc_entries_per_group(inode));
88 return group;
89}
90
91/**
92 * nilfs_palloc_desc_blkoff - get block offset of a group descriptor block
93 * @inode: inode of metadata file using this allocator
94 * @group: group number
95 *
96 * nilfs_palloc_desc_blkoff() returns block offset of the descriptor
97 * block which contains a descriptor of the specified group.
98 */
99static unsigned long
100nilfs_palloc_desc_blkoff(const struct inode *inode, unsigned long group)
101{
102 unsigned long desc_block =
103 group / nilfs_palloc_groups_per_desc_block(inode);
104 return desc_block * NILFS_MDT(inode)->mi_blocks_per_desc_block;
105}
106
107/**
108 * nilfs_palloc_bitmap_blkoff - get block offset of a bitmap block
109 * @inode: inode of metadata file using this allocator
110 * @group: group number
111 *
112 * nilfs_palloc_bitmap_blkoff() returns block offset of the bitmap
113 * block used to allocate/deallocate entries in the specified group.
114 */
115static unsigned long
116nilfs_palloc_bitmap_blkoff(const struct inode *inode, unsigned long group)
117{
118 unsigned long desc_offset =
119 group % nilfs_palloc_groups_per_desc_block(inode);
120 return nilfs_palloc_desc_blkoff(inode, group) + 1 +
121 desc_offset * NILFS_MDT(inode)->mi_blocks_per_group;
122}
123
124/**
125 * nilfs_palloc_group_desc_nfrees - get the number of free entries in a group
126 * @desc: pointer to descriptor structure for the group
127 * @lock: spin lock protecting @desc
128 */
129static unsigned long
130nilfs_palloc_group_desc_nfrees(const struct nilfs_palloc_group_desc *desc,
131 spinlock_t *lock)
132{
133 unsigned long nfree;
134
135 spin_lock(lock);
136 nfree = le32_to_cpu(desc->pg_nfrees);
137 spin_unlock(lock);
138 return nfree;
139}
140
141/**
142 * nilfs_palloc_group_desc_add_entries - adjust count of free entries
143 * @desc: pointer to descriptor structure for the group
144 * @lock: spin lock protecting @desc
145 * @n: delta to be added
146 */
147static u32
148nilfs_palloc_group_desc_add_entries(struct nilfs_palloc_group_desc *desc,
149 spinlock_t *lock, u32 n)
150{
151 u32 nfree;
152
153 spin_lock(lock);
154 le32_add_cpu(&desc->pg_nfrees, n);
155 nfree = le32_to_cpu(desc->pg_nfrees);
156 spin_unlock(lock);
157 return nfree;
158}
159
160/**
161 * nilfs_palloc_entry_blkoff - get block offset of an entry block
162 * @inode: inode of metadata file using this allocator
163 * @nr: serial number of the entry (e.g. inode number)
164 */
165static unsigned long
166nilfs_palloc_entry_blkoff(const struct inode *inode, __u64 nr)
167{
168 unsigned long group, group_offset;
169
170 group = nilfs_palloc_group(inode, nr, &group_offset);
171
172 return nilfs_palloc_bitmap_blkoff(inode, group) + 1 +
173 group_offset / NILFS_MDT(inode)->mi_entries_per_block;
174}
175
176/**
177 * nilfs_palloc_desc_block_init - initialize buffer of a group descriptor block
178 * @inode: inode of metadata file
179 * @bh: buffer head of the buffer to be initialized
180 * @from: kernel address mapped for a chunk of the block
181 *
182 * This function does not yet support the case where block size > PAGE_SIZE.
183 */
184static void nilfs_palloc_desc_block_init(struct inode *inode,
185 struct buffer_head *bh, void *from)
186{
187 struct nilfs_palloc_group_desc *desc = from;
188 unsigned long n = nilfs_palloc_groups_per_desc_block(inode);
189 __le32 nfrees;
190
191 nfrees = cpu_to_le32(nilfs_palloc_entries_per_group(inode));
192 while (n-- > 0) {
193 desc->pg_nfrees = nfrees;
194 desc++;
195 }
196}
197
198static int nilfs_palloc_get_block(struct inode *inode, unsigned long blkoff,
199 int create,
200 void (*init_block)(struct inode *,
201 struct buffer_head *,
202 void *),
203 struct buffer_head **bhp,
204 struct nilfs_bh_assoc *prev,
205 spinlock_t *lock)
206{
207 int ret;
208
209 spin_lock(lock);
210 if (prev->bh && blkoff == prev->blkoff &&
211 likely(buffer_uptodate(prev->bh))) {
212 get_bh(prev->bh);
213 *bhp = prev->bh;
214 spin_unlock(lock);
215 return 0;
216 }
217 spin_unlock(lock);
218
219 ret = nilfs_mdt_get_block(inode, blkoff, create, init_block, bhp);
220 if (!ret) {
221 spin_lock(lock);
222 /*
223 * The following code must be safe for change of the
224 * cache contents during the get block call.
225 */
226 brelse(prev->bh);
227 get_bh(*bhp);
228 prev->bh = *bhp;
229 prev->blkoff = blkoff;
230 spin_unlock(lock);
231 }
232 return ret;
233}
234
235/**
236 * nilfs_palloc_delete_block - delete a block on the persistent allocator file
237 * @inode: inode of metadata file using this allocator
238 * @blkoff: block offset
239 * @prev: nilfs_bh_assoc struct of the last used buffer
240 * @lock: spin lock protecting @prev
241 */
242static int nilfs_palloc_delete_block(struct inode *inode, unsigned long blkoff,
243 struct nilfs_bh_assoc *prev,
244 spinlock_t *lock)
245{
246 spin_lock(lock);
247 if (prev->bh && blkoff == prev->blkoff) {
248 brelse(prev->bh);
249 prev->bh = NULL;
250 }
251 spin_unlock(lock);
252 return nilfs_mdt_delete_block(inode, blkoff);
253}
254
255/**
256 * nilfs_palloc_get_desc_block - get buffer head of a group descriptor block
257 * @inode: inode of metadata file using this allocator
258 * @group: group number
259 * @create: create flag
260 * @bhp: pointer to store the resultant buffer head
261 */
262static int nilfs_palloc_get_desc_block(struct inode *inode,
263 unsigned long group,
264 int create, struct buffer_head **bhp)
265{
266 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
267
268 return nilfs_palloc_get_block(inode,
269 nilfs_palloc_desc_blkoff(inode, group),
270 create, nilfs_palloc_desc_block_init,
271 bhp, &cache->prev_desc, &cache->lock);
272}
273
274/**
275 * nilfs_palloc_get_bitmap_block - get buffer head of a bitmap block
276 * @inode: inode of metadata file using this allocator
277 * @group: group number
278 * @create: create flag
279 * @bhp: pointer to store the resultant buffer head
280 */
281static int nilfs_palloc_get_bitmap_block(struct inode *inode,
282 unsigned long group,
283 int create, struct buffer_head **bhp)
284{
285 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
286
287 return nilfs_palloc_get_block(inode,
288 nilfs_palloc_bitmap_blkoff(inode, group),
289 create, NULL, bhp,
290 &cache->prev_bitmap, &cache->lock);
291}
292
293/**
294 * nilfs_palloc_delete_bitmap_block - delete a bitmap block
295 * @inode: inode of metadata file using this allocator
296 * @group: group number
297 */
298static int nilfs_palloc_delete_bitmap_block(struct inode *inode,
299 unsigned long group)
300{
301 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
302
303 return nilfs_palloc_delete_block(inode,
304 nilfs_palloc_bitmap_blkoff(inode,
305 group),
306 &cache->prev_bitmap, &cache->lock);
307}
308
309/**
310 * nilfs_palloc_get_entry_block - get buffer head of an entry block
311 * @inode: inode of metadata file using this allocator
312 * @nr: serial number of the entry (e.g. inode number)
313 * @create: create flag
314 * @bhp: pointer to store the resultant buffer head
315 */
316int nilfs_palloc_get_entry_block(struct inode *inode, __u64 nr,
317 int create, struct buffer_head **bhp)
318{
319 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
320
321 return nilfs_palloc_get_block(inode,
322 nilfs_palloc_entry_blkoff(inode, nr),
323 create, NULL, bhp,
324 &cache->prev_entry, &cache->lock);
325}
326
327/**
328 * nilfs_palloc_delete_entry_block - delete an entry block
329 * @inode: inode of metadata file using this allocator
330 * @nr: serial number of the entry
331 */
332static int nilfs_palloc_delete_entry_block(struct inode *inode, __u64 nr)
333{
334 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
335
336 return nilfs_palloc_delete_block(inode,
337 nilfs_palloc_entry_blkoff(inode, nr),
338 &cache->prev_entry, &cache->lock);
339}
340
341/**
342 * nilfs_palloc_group_desc_offset - calculate the byte offset of a group
343 * descriptor in the folio containing it
344 * @inode: inode of metadata file using this allocator
345 * @group: group number
346 * @bh: buffer head of the group descriptor block
347 *
348 * Return: Byte offset in the folio of the group descriptor for @group.
349 */
350static size_t nilfs_palloc_group_desc_offset(const struct inode *inode,
351 unsigned long group,
352 const struct buffer_head *bh)
353{
354 return offset_in_folio(bh->b_folio, bh->b_data) +
355 sizeof(struct nilfs_palloc_group_desc) *
356 (group % nilfs_palloc_groups_per_desc_block(inode));
357}
358
359/**
360 * nilfs_palloc_bitmap_offset - calculate the byte offset of a bitmap block
361 * in the folio containing it
362 * @bh: buffer head of the bitmap block
363 *
364 * Return: Byte offset in the folio of the bitmap block for @bh.
365 */
366static size_t nilfs_palloc_bitmap_offset(const struct buffer_head *bh)
367{
368 return offset_in_folio(bh->b_folio, bh->b_data);
369}
370
371/**
372 * nilfs_palloc_entry_offset - calculate the byte offset of an entry in the
373 * folio containing it
374 * @inode: inode of metadata file using this allocator
375 * @nr: serial number of the entry (e.g. inode number)
376 * @bh: buffer head of the entry block
377 *
378 * Return: Byte offset in the folio of the entry @nr.
379 */
380size_t nilfs_palloc_entry_offset(const struct inode *inode, __u64 nr,
381 const struct buffer_head *bh)
382{
383 unsigned long entry_index_in_group, entry_index_in_block;
384
385 nilfs_palloc_group(inode, nr, &entry_index_in_group);
386 entry_index_in_block = entry_index_in_group %
387 NILFS_MDT(inode)->mi_entries_per_block;
388
389 return offset_in_folio(bh->b_folio, bh->b_data) +
390 entry_index_in_block * NILFS_MDT(inode)->mi_entry_size;
391}
392
393/**
394 * nilfs_palloc_find_available_slot - find available slot in a group
395 * @bitmap: bitmap of the group
396 * @target: offset number of an entry in the group (start point)
397 * @bsize: size in bits
398 * @lock: spin lock protecting @bitmap
399 * @wrap: whether to wrap around
400 */
401static int nilfs_palloc_find_available_slot(unsigned char *bitmap,
402 unsigned long target,
403 unsigned int bsize,
404 spinlock_t *lock, bool wrap)
405{
406 int pos, end = bsize;
407
408 if (likely(target < bsize)) {
409 pos = target;
410 do {
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 } while (++pos < end);
417
418 end = target;
419 }
420 if (!wrap)
421 return -ENOSPC;
422
423 /* wrap around */
424 for (pos = 0; pos < end; pos++) {
425 pos = nilfs_find_next_zero_bit(bitmap, end, pos);
426 if (pos >= end)
427 break;
428 if (!nilfs_set_bit_atomic(lock, pos, bitmap))
429 return pos;
430 }
431
432 return -ENOSPC;
433}
434
435/**
436 * nilfs_palloc_rest_groups_in_desc_block - get the remaining number of groups
437 * in a group descriptor block
438 * @inode: inode of metadata file using this allocator
439 * @curr: current group number
440 * @max: maximum number of groups
441 */
442static unsigned long
443nilfs_palloc_rest_groups_in_desc_block(const struct inode *inode,
444 unsigned long curr, unsigned long max)
445{
446 return min_t(unsigned long,
447 nilfs_palloc_groups_per_desc_block(inode) -
448 curr % nilfs_palloc_groups_per_desc_block(inode),
449 max - curr + 1);
450}
451
452/**
453 * nilfs_palloc_count_desc_blocks - count descriptor blocks number
454 * @inode: inode of metadata file using this allocator
455 * @desc_blocks: descriptor blocks number [out]
456 */
457static int nilfs_palloc_count_desc_blocks(struct inode *inode,
458 unsigned long *desc_blocks)
459{
460 __u64 blknum;
461 int ret;
462
463 ret = nilfs_bmap_last_key(NILFS_I(inode)->i_bmap, &blknum);
464 if (likely(!ret))
465 *desc_blocks = DIV_ROUND_UP(
466 (unsigned long)blknum,
467 NILFS_MDT(inode)->mi_blocks_per_desc_block);
468 return ret;
469}
470
471/**
472 * nilfs_palloc_mdt_file_can_grow - check potential opportunity for
473 * MDT file growing
474 * @inode: inode of metadata file using this allocator
475 * @desc_blocks: known current descriptor blocks count
476 */
477static inline bool nilfs_palloc_mdt_file_can_grow(struct inode *inode,
478 unsigned long desc_blocks)
479{
480 return (nilfs_palloc_groups_per_desc_block(inode) * desc_blocks) <
481 nilfs_palloc_groups_count(inode);
482}
483
484/**
485 * nilfs_palloc_count_max_entries - count max number of entries that can be
486 * described by descriptor blocks count
487 * @inode: inode of metadata file using this allocator
488 * @nused: current number of used entries
489 * @nmaxp: max number of entries [out]
490 */
491int nilfs_palloc_count_max_entries(struct inode *inode, u64 nused, u64 *nmaxp)
492{
493 unsigned long desc_blocks = 0;
494 u64 entries_per_desc_block, nmax;
495 int err;
496
497 err = nilfs_palloc_count_desc_blocks(inode, &desc_blocks);
498 if (unlikely(err))
499 return err;
500
501 entries_per_desc_block = (u64)nilfs_palloc_entries_per_group(inode) *
502 nilfs_palloc_groups_per_desc_block(inode);
503 nmax = entries_per_desc_block * desc_blocks;
504
505 if (nused == nmax &&
506 nilfs_palloc_mdt_file_can_grow(inode, desc_blocks))
507 nmax += entries_per_desc_block;
508
509 if (nused > nmax)
510 return -ERANGE;
511
512 *nmaxp = nmax;
513 return 0;
514}
515
516/**
517 * nilfs_palloc_prepare_alloc_entry - prepare to allocate a persistent object
518 * @inode: inode of metadata file using this allocator
519 * @req: nilfs_palloc_req structure exchanged for the allocation
520 * @wrap: whether to wrap around
521 */
522int nilfs_palloc_prepare_alloc_entry(struct inode *inode,
523 struct nilfs_palloc_req *req, bool wrap)
524{
525 struct buffer_head *desc_bh, *bitmap_bh;
526 struct nilfs_palloc_group_desc *desc;
527 unsigned char *bitmap;
528 size_t doff, boff;
529 unsigned long group, maxgroup, ngroups;
530 unsigned long group_offset, maxgroup_offset;
531 unsigned long n, entries_per_group;
532 unsigned long i, j;
533 spinlock_t *lock;
534 int pos, ret;
535
536 ngroups = nilfs_palloc_groups_count(inode);
537 maxgroup = ngroups - 1;
538 group = nilfs_palloc_group(inode, req->pr_entry_nr, &group_offset);
539 entries_per_group = nilfs_palloc_entries_per_group(inode);
540
541 for (i = 0; i < ngroups; i += n) {
542 if (group >= ngroups && wrap) {
543 /* wrap around */
544 group = 0;
545 maxgroup = nilfs_palloc_group(inode, req->pr_entry_nr,
546 &maxgroup_offset) - 1;
547 }
548 ret = nilfs_palloc_get_desc_block(inode, group, 1, &desc_bh);
549 if (ret < 0)
550 return ret;
551
552 doff = nilfs_palloc_group_desc_offset(inode, group, desc_bh);
553 desc = kmap_local_folio(desc_bh->b_folio, doff);
554 n = nilfs_palloc_rest_groups_in_desc_block(inode, group,
555 maxgroup);
556 for (j = 0; j < n; j++, group++, group_offset = 0) {
557 lock = nilfs_mdt_bgl_lock(inode, group);
558 if (nilfs_palloc_group_desc_nfrees(&desc[j], lock) == 0)
559 continue;
560
561 kunmap_local(desc);
562 ret = nilfs_palloc_get_bitmap_block(inode, group, 1,
563 &bitmap_bh);
564 if (unlikely(ret < 0)) {
565 brelse(desc_bh);
566 return ret;
567 }
568
569 /*
570 * Re-kmap the folio containing the first (and
571 * subsequent) group descriptors.
572 */
573 desc = kmap_local_folio(desc_bh->b_folio, doff);
574
575 boff = nilfs_palloc_bitmap_offset(bitmap_bh);
576 bitmap = kmap_local_folio(bitmap_bh->b_folio, boff);
577 pos = nilfs_palloc_find_available_slot(
578 bitmap, group_offset, entries_per_group, lock,
579 wrap);
580 /*
581 * Since the search for a free slot in the second and
582 * subsequent bitmap blocks always starts from the
583 * beginning, the wrap flag only has an effect on the
584 * first search.
585 */
586 kunmap_local(bitmap);
587 if (pos >= 0)
588 goto found;
589
590 brelse(bitmap_bh);
591 }
592
593 kunmap_local(desc);
594 brelse(desc_bh);
595 }
596
597 /* no entries left */
598 return -ENOSPC;
599
600found:
601 /* found a free entry */
602 nilfs_palloc_group_desc_add_entries(&desc[j], lock, -1);
603 req->pr_entry_nr = entries_per_group * group + pos;
604 kunmap_local(desc);
605
606 req->pr_desc_bh = desc_bh;
607 req->pr_bitmap_bh = bitmap_bh;
608 return 0;
609}
610
611/**
612 * nilfs_palloc_commit_alloc_entry - finish allocation of a persistent object
613 * @inode: inode of metadata file using this allocator
614 * @req: nilfs_palloc_req structure exchanged for the allocation
615 */
616void nilfs_palloc_commit_alloc_entry(struct inode *inode,
617 struct nilfs_palloc_req *req)
618{
619 mark_buffer_dirty(req->pr_bitmap_bh);
620 mark_buffer_dirty(req->pr_desc_bh);
621 nilfs_mdt_mark_dirty(inode);
622
623 brelse(req->pr_bitmap_bh);
624 brelse(req->pr_desc_bh);
625}
626
627/**
628 * nilfs_palloc_commit_free_entry - finish deallocating a persistent object
629 * @inode: inode of metadata file using this allocator
630 * @req: nilfs_palloc_req structure exchanged for the removal
631 */
632void nilfs_palloc_commit_free_entry(struct inode *inode,
633 struct nilfs_palloc_req *req)
634{
635 unsigned long group, group_offset;
636 size_t doff, boff;
637 struct nilfs_palloc_group_desc *desc;
638 unsigned char *bitmap;
639 spinlock_t *lock;
640
641 group = nilfs_palloc_group(inode, req->pr_entry_nr, &group_offset);
642 doff = nilfs_palloc_group_desc_offset(inode, group, req->pr_desc_bh);
643 desc = kmap_local_folio(req->pr_desc_bh->b_folio, doff);
644
645 boff = nilfs_palloc_bitmap_offset(req->pr_bitmap_bh);
646 bitmap = kmap_local_folio(req->pr_bitmap_bh->b_folio, boff);
647 lock = nilfs_mdt_bgl_lock(inode, group);
648
649 if (!nilfs_clear_bit_atomic(lock, group_offset, bitmap))
650 nilfs_warn(inode->i_sb,
651 "%s (ino=%lu): entry number %llu already freed",
652 __func__, inode->i_ino,
653 (unsigned long long)req->pr_entry_nr);
654 else
655 nilfs_palloc_group_desc_add_entries(desc, lock, 1);
656
657 kunmap_local(bitmap);
658 kunmap_local(desc);
659
660 mark_buffer_dirty(req->pr_desc_bh);
661 mark_buffer_dirty(req->pr_bitmap_bh);
662 nilfs_mdt_mark_dirty(inode);
663
664 brelse(req->pr_bitmap_bh);
665 brelse(req->pr_desc_bh);
666}
667
668/**
669 * nilfs_palloc_abort_alloc_entry - cancel allocation of a persistent object
670 * @inode: inode of metadata file using this allocator
671 * @req: nilfs_palloc_req structure exchanged for the allocation
672 */
673void nilfs_palloc_abort_alloc_entry(struct inode *inode,
674 struct nilfs_palloc_req *req)
675{
676 struct nilfs_palloc_group_desc *desc;
677 size_t doff, boff;
678 unsigned char *bitmap;
679 unsigned long group, group_offset;
680 spinlock_t *lock;
681
682 group = nilfs_palloc_group(inode, req->pr_entry_nr, &group_offset);
683 doff = nilfs_palloc_group_desc_offset(inode, group, req->pr_desc_bh);
684 desc = kmap_local_folio(req->pr_desc_bh->b_folio, doff);
685
686 boff = nilfs_palloc_bitmap_offset(req->pr_bitmap_bh);
687 bitmap = kmap_local_folio(req->pr_bitmap_bh->b_folio, boff);
688 lock = nilfs_mdt_bgl_lock(inode, group);
689
690 if (!nilfs_clear_bit_atomic(lock, group_offset, bitmap))
691 nilfs_warn(inode->i_sb,
692 "%s (ino=%lu): entry number %llu already freed",
693 __func__, inode->i_ino,
694 (unsigned long long)req->pr_entry_nr);
695 else
696 nilfs_palloc_group_desc_add_entries(desc, lock, 1);
697
698 kunmap_local(bitmap);
699 kunmap_local(desc);
700
701 brelse(req->pr_bitmap_bh);
702 brelse(req->pr_desc_bh);
703
704 req->pr_entry_nr = 0;
705 req->pr_bitmap_bh = NULL;
706 req->pr_desc_bh = NULL;
707}
708
709/**
710 * nilfs_palloc_prepare_free_entry - prepare to deallocate a persistent object
711 * @inode: inode of metadata file using this allocator
712 * @req: nilfs_palloc_req structure exchanged for the removal
713 */
714int nilfs_palloc_prepare_free_entry(struct inode *inode,
715 struct nilfs_palloc_req *req)
716{
717 struct buffer_head *desc_bh, *bitmap_bh;
718 unsigned long group, group_offset;
719 int ret;
720
721 group = nilfs_palloc_group(inode, req->pr_entry_nr, &group_offset);
722 ret = nilfs_palloc_get_desc_block(inode, group, 1, &desc_bh);
723 if (ret < 0)
724 return ret;
725 ret = nilfs_palloc_get_bitmap_block(inode, group, 1, &bitmap_bh);
726 if (ret < 0) {
727 brelse(desc_bh);
728 return ret;
729 }
730
731 req->pr_desc_bh = desc_bh;
732 req->pr_bitmap_bh = bitmap_bh;
733 return 0;
734}
735
736/**
737 * nilfs_palloc_abort_free_entry - cancel deallocating a persistent object
738 * @inode: inode of metadata file using this allocator
739 * @req: nilfs_palloc_req structure exchanged for the removal
740 */
741void nilfs_palloc_abort_free_entry(struct inode *inode,
742 struct nilfs_palloc_req *req)
743{
744 brelse(req->pr_bitmap_bh);
745 brelse(req->pr_desc_bh);
746
747 req->pr_entry_nr = 0;
748 req->pr_bitmap_bh = NULL;
749 req->pr_desc_bh = NULL;
750}
751
752/**
753 * nilfs_palloc_freev - deallocate a set of persistent objects
754 * @inode: inode of metadata file using this allocator
755 * @entry_nrs: array of entry numbers to be deallocated
756 * @nitems: number of entries stored in @entry_nrs
757 */
758int nilfs_palloc_freev(struct inode *inode, __u64 *entry_nrs, size_t nitems)
759{
760 struct buffer_head *desc_bh, *bitmap_bh;
761 struct nilfs_palloc_group_desc *desc;
762 unsigned char *bitmap;
763 size_t doff, boff;
764 unsigned long group, group_offset;
765 __u64 group_min_nr, last_nrs[8];
766 const unsigned long epg = nilfs_palloc_entries_per_group(inode);
767 const unsigned int epb = NILFS_MDT(inode)->mi_entries_per_block;
768 unsigned int entry_start, end, pos;
769 spinlock_t *lock;
770 int i, j, k, ret;
771 u32 nfree;
772
773 for (i = 0; i < nitems; i = j) {
774 int change_group = false;
775 int nempties = 0, n = 0;
776
777 group = nilfs_palloc_group(inode, entry_nrs[i], &group_offset);
778 ret = nilfs_palloc_get_desc_block(inode, group, 0, &desc_bh);
779 if (ret < 0)
780 return ret;
781 ret = nilfs_palloc_get_bitmap_block(inode, group, 0,
782 &bitmap_bh);
783 if (ret < 0) {
784 brelse(desc_bh);
785 return ret;
786 }
787
788 /* Get the first entry number of the group */
789 group_min_nr = (__u64)group * epg;
790
791 boff = nilfs_palloc_bitmap_offset(bitmap_bh);
792 bitmap = kmap_local_folio(bitmap_bh->b_folio, boff);
793 lock = nilfs_mdt_bgl_lock(inode, group);
794
795 j = i;
796 entry_start = rounddown(group_offset, epb);
797 do {
798 if (!nilfs_clear_bit_atomic(lock, group_offset,
799 bitmap)) {
800 nilfs_warn(inode->i_sb,
801 "%s (ino=%lu): entry number %llu already freed",
802 __func__, inode->i_ino,
803 (unsigned long long)entry_nrs[j]);
804 } else {
805 n++;
806 }
807
808 j++;
809 if (j >= nitems || entry_nrs[j] < group_min_nr ||
810 entry_nrs[j] >= group_min_nr + epg) {
811 change_group = true;
812 } else {
813 group_offset = entry_nrs[j] - group_min_nr;
814 if (group_offset >= entry_start &&
815 group_offset < entry_start + epb) {
816 /* This entry is in the same block */
817 continue;
818 }
819 }
820
821 /* Test if the entry block is empty or not */
822 end = entry_start + epb;
823 pos = nilfs_find_next_bit(bitmap, end, entry_start);
824 if (pos >= end) {
825 last_nrs[nempties++] = entry_nrs[j - 1];
826 if (nempties >= ARRAY_SIZE(last_nrs))
827 break;
828 }
829
830 if (change_group)
831 break;
832
833 /* Go on to the next entry block */
834 entry_start = rounddown(group_offset, epb);
835 } while (true);
836
837 kunmap_local(bitmap);
838 mark_buffer_dirty(bitmap_bh);
839 brelse(bitmap_bh);
840
841 for (k = 0; k < nempties; k++) {
842 ret = nilfs_palloc_delete_entry_block(inode,
843 last_nrs[k]);
844 if (ret && ret != -ENOENT)
845 nilfs_warn(inode->i_sb,
846 "error %d deleting block that object (entry=%llu, ino=%lu) belongs to",
847 ret, (unsigned long long)last_nrs[k],
848 inode->i_ino);
849 }
850
851 doff = nilfs_palloc_group_desc_offset(inode, group, desc_bh);
852 desc = kmap_local_folio(desc_bh->b_folio, doff);
853 nfree = nilfs_palloc_group_desc_add_entries(desc, lock, n);
854 kunmap_local(desc);
855 mark_buffer_dirty(desc_bh);
856 nilfs_mdt_mark_dirty(inode);
857 brelse(desc_bh);
858
859 if (nfree == nilfs_palloc_entries_per_group(inode)) {
860 ret = nilfs_palloc_delete_bitmap_block(inode, group);
861 if (ret && ret != -ENOENT)
862 nilfs_warn(inode->i_sb,
863 "error %d deleting bitmap block of group=%lu, ino=%lu",
864 ret, group, inode->i_ino);
865 }
866 }
867 return 0;
868}
869
870void nilfs_palloc_setup_cache(struct inode *inode,
871 struct nilfs_palloc_cache *cache)
872{
873 NILFS_MDT(inode)->mi_palloc_cache = cache;
874 spin_lock_init(&cache->lock);
875}
876
877void nilfs_palloc_clear_cache(struct inode *inode)
878{
879 struct nilfs_palloc_cache *cache = NILFS_MDT(inode)->mi_palloc_cache;
880
881 spin_lock(&cache->lock);
882 brelse(cache->prev_desc.bh);
883 brelse(cache->prev_bitmap.bh);
884 brelse(cache->prev_entry.bh);
885 cache->prev_desc.bh = NULL;
886 cache->prev_bitmap.bh = NULL;
887 cache->prev_entry.bh = NULL;
888 spin_unlock(&cache->lock);
889}
890
891void nilfs_palloc_destroy_cache(struct inode *inode)
892{
893 nilfs_palloc_clear_cache(inode);
894 NILFS_MDT(inode)->mi_palloc_cache = NULL;
895}
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}