]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - fs/f2fs/segment.c
f2fs: avoid reverse IO order for NODE and DATA
[mirror_ubuntu-zesty-kernel.git] / fs / f2fs / segment.c
1 /*
2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/bio.h>
14 #include <linux/blkdev.h>
15 #include <linux/prefetch.h>
16 #include <linux/kthread.h>
17 #include <linux/swap.h>
18 #include <linux/timer.h>
19
20 #include "f2fs.h"
21 #include "segment.h"
22 #include "node.h"
23 #include "trace.h"
24 #include <trace/events/f2fs.h>
25
26 #define __reverse_ffz(x) __reverse_ffs(~(x))
27
28 static struct kmem_cache *discard_entry_slab;
29 static struct kmem_cache *sit_entry_set_slab;
30 static struct kmem_cache *inmem_entry_slab;
31
32 static unsigned long __reverse_ulong(unsigned char *str)
33 {
34 unsigned long tmp = 0;
35 int shift = 24, idx = 0;
36
37 #if BITS_PER_LONG == 64
38 shift = 56;
39 #endif
40 while (shift >= 0) {
41 tmp |= (unsigned long)str[idx++] << shift;
42 shift -= BITS_PER_BYTE;
43 }
44 return tmp;
45 }
46
47 /*
48 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
49 * MSB and LSB are reversed in a byte by f2fs_set_bit.
50 */
51 static inline unsigned long __reverse_ffs(unsigned long word)
52 {
53 int num = 0;
54
55 #if BITS_PER_LONG == 64
56 if ((word & 0xffffffff00000000UL) == 0)
57 num += 32;
58 else
59 word >>= 32;
60 #endif
61 if ((word & 0xffff0000) == 0)
62 num += 16;
63 else
64 word >>= 16;
65
66 if ((word & 0xff00) == 0)
67 num += 8;
68 else
69 word >>= 8;
70
71 if ((word & 0xf0) == 0)
72 num += 4;
73 else
74 word >>= 4;
75
76 if ((word & 0xc) == 0)
77 num += 2;
78 else
79 word >>= 2;
80
81 if ((word & 0x2) == 0)
82 num += 1;
83 return num;
84 }
85
86 /*
87 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
88 * f2fs_set_bit makes MSB and LSB reversed in a byte.
89 * @size must be integral times of unsigned long.
90 * Example:
91 * MSB <--> LSB
92 * f2fs_set_bit(0, bitmap) => 1000 0000
93 * f2fs_set_bit(7, bitmap) => 0000 0001
94 */
95 static unsigned long __find_rev_next_bit(const unsigned long *addr,
96 unsigned long size, unsigned long offset)
97 {
98 const unsigned long *p = addr + BIT_WORD(offset);
99 unsigned long result = size;
100 unsigned long tmp;
101
102 if (offset >= size)
103 return size;
104
105 size -= (offset & ~(BITS_PER_LONG - 1));
106 offset %= BITS_PER_LONG;
107
108 while (1) {
109 if (*p == 0)
110 goto pass;
111
112 tmp = __reverse_ulong((unsigned char *)p);
113
114 tmp &= ~0UL >> offset;
115 if (size < BITS_PER_LONG)
116 tmp &= (~0UL << (BITS_PER_LONG - size));
117 if (tmp)
118 goto found;
119 pass:
120 if (size <= BITS_PER_LONG)
121 break;
122 size -= BITS_PER_LONG;
123 offset = 0;
124 p++;
125 }
126 return result;
127 found:
128 return result - size + __reverse_ffs(tmp);
129 }
130
131 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
132 unsigned long size, unsigned long offset)
133 {
134 const unsigned long *p = addr + BIT_WORD(offset);
135 unsigned long result = size;
136 unsigned long tmp;
137
138 if (offset >= size)
139 return size;
140
141 size -= (offset & ~(BITS_PER_LONG - 1));
142 offset %= BITS_PER_LONG;
143
144 while (1) {
145 if (*p == ~0UL)
146 goto pass;
147
148 tmp = __reverse_ulong((unsigned char *)p);
149
150 if (offset)
151 tmp |= ~0UL << (BITS_PER_LONG - offset);
152 if (size < BITS_PER_LONG)
153 tmp |= ~0UL >> size;
154 if (tmp != ~0UL)
155 goto found;
156 pass:
157 if (size <= BITS_PER_LONG)
158 break;
159 size -= BITS_PER_LONG;
160 offset = 0;
161 p++;
162 }
163 return result;
164 found:
165 return result - size + __reverse_ffz(tmp);
166 }
167
168 void register_inmem_page(struct inode *inode, struct page *page)
169 {
170 struct f2fs_inode_info *fi = F2FS_I(inode);
171 struct inmem_pages *new;
172
173 f2fs_trace_pid(page);
174
175 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
176 SetPagePrivate(page);
177
178 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
179
180 /* add atomic page indices to the list */
181 new->page = page;
182 INIT_LIST_HEAD(&new->list);
183
184 /* increase reference count with clean state */
185 mutex_lock(&fi->inmem_lock);
186 get_page(page);
187 list_add_tail(&new->list, &fi->inmem_pages);
188 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
189 mutex_unlock(&fi->inmem_lock);
190
191 trace_f2fs_register_inmem_page(page, INMEM);
192 }
193
194 static int __revoke_inmem_pages(struct inode *inode,
195 struct list_head *head, bool drop, bool recover)
196 {
197 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
198 struct inmem_pages *cur, *tmp;
199 int err = 0;
200
201 list_for_each_entry_safe(cur, tmp, head, list) {
202 struct page *page = cur->page;
203
204 if (drop)
205 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
206
207 lock_page(page);
208
209 if (recover) {
210 struct dnode_of_data dn;
211 struct node_info ni;
212
213 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
214
215 set_new_dnode(&dn, inode, NULL, NULL, 0);
216 if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
217 err = -EAGAIN;
218 goto next;
219 }
220 get_node_info(sbi, dn.nid, &ni);
221 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
222 cur->old_addr, ni.version, true, true);
223 f2fs_put_dnode(&dn);
224 }
225 next:
226 /* we don't need to invalidate this in the sccessful status */
227 if (drop || recover)
228 ClearPageUptodate(page);
229 set_page_private(page, 0);
230 ClearPagePrivate(page);
231 f2fs_put_page(page, 1);
232
233 list_del(&cur->list);
234 kmem_cache_free(inmem_entry_slab, cur);
235 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
236 }
237 return err;
238 }
239
240 void drop_inmem_pages(struct inode *inode)
241 {
242 struct f2fs_inode_info *fi = F2FS_I(inode);
243
244 clear_inode_flag(inode, FI_ATOMIC_FILE);
245
246 mutex_lock(&fi->inmem_lock);
247 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
248 mutex_unlock(&fi->inmem_lock);
249 }
250
251 static int __commit_inmem_pages(struct inode *inode,
252 struct list_head *revoke_list)
253 {
254 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
255 struct f2fs_inode_info *fi = F2FS_I(inode);
256 struct inmem_pages *cur, *tmp;
257 struct f2fs_io_info fio = {
258 .sbi = sbi,
259 .type = DATA,
260 .rw = WRITE_SYNC | REQ_PRIO,
261 .encrypted_page = NULL,
262 };
263 bool submit_bio = false;
264 int err = 0;
265
266 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
267 struct page *page = cur->page;
268
269 lock_page(page);
270 if (page->mapping == inode->i_mapping) {
271 trace_f2fs_commit_inmem_page(page, INMEM);
272
273 set_page_dirty(page);
274 f2fs_wait_on_page_writeback(page, DATA, true);
275 if (clear_page_dirty_for_io(page))
276 inode_dec_dirty_pages(inode);
277
278 fio.page = page;
279 err = do_write_data_page(&fio);
280 if (err) {
281 unlock_page(page);
282 break;
283 }
284
285 /* record old blkaddr for revoking */
286 cur->old_addr = fio.old_blkaddr;
287
288 clear_cold_data(page);
289 submit_bio = true;
290 }
291 unlock_page(page);
292 list_move_tail(&cur->list, revoke_list);
293 }
294
295 if (submit_bio)
296 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
297
298 if (!err)
299 __revoke_inmem_pages(inode, revoke_list, false, false);
300
301 return err;
302 }
303
304 int commit_inmem_pages(struct inode *inode)
305 {
306 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
307 struct f2fs_inode_info *fi = F2FS_I(inode);
308 struct list_head revoke_list;
309 int err;
310
311 INIT_LIST_HEAD(&revoke_list);
312 f2fs_balance_fs(sbi, true);
313 f2fs_lock_op(sbi);
314
315 mutex_lock(&fi->inmem_lock);
316 err = __commit_inmem_pages(inode, &revoke_list);
317 if (err) {
318 int ret;
319 /*
320 * try to revoke all committed pages, but still we could fail
321 * due to no memory or other reason, if that happened, EAGAIN
322 * will be returned, which means in such case, transaction is
323 * already not integrity, caller should use journal to do the
324 * recovery or rewrite & commit last transaction. For other
325 * error number, revoking was done by filesystem itself.
326 */
327 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
328 if (ret)
329 err = ret;
330
331 /* drop all uncommitted pages */
332 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
333 }
334 mutex_unlock(&fi->inmem_lock);
335
336 f2fs_unlock_op(sbi);
337 return err;
338 }
339
340 /*
341 * This function balances dirty node and dentry pages.
342 * In addition, it controls garbage collection.
343 */
344 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
345 {
346 if (!need)
347 return;
348
349 /* balance_fs_bg is able to be pending */
350 if (excess_cached_nats(sbi))
351 f2fs_balance_fs_bg(sbi);
352
353 /*
354 * We should do GC or end up with checkpoint, if there are so many dirty
355 * dir/node pages without enough free segments.
356 */
357 if (has_not_enough_free_secs(sbi, 0)) {
358 mutex_lock(&sbi->gc_mutex);
359 f2fs_gc(sbi, false);
360 }
361 }
362
363 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
364 {
365 /* try to shrink extent cache when there is no enough memory */
366 if (!available_free_memory(sbi, EXTENT_CACHE))
367 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
368
369 /* check the # of cached NAT entries */
370 if (!available_free_memory(sbi, NAT_ENTRIES))
371 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
372
373 if (!available_free_memory(sbi, FREE_NIDS))
374 try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES);
375
376 /* checkpoint is the only way to shrink partial cached entries */
377 if (!available_free_memory(sbi, NAT_ENTRIES) ||
378 !available_free_memory(sbi, INO_ENTRIES) ||
379 excess_prefree_segs(sbi) ||
380 excess_dirty_nats(sbi) ||
381 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
382 if (test_opt(sbi, DATA_FLUSH)) {
383 struct blk_plug plug;
384
385 blk_start_plug(&plug);
386 sync_dirty_inodes(sbi, FILE_INODE);
387 blk_finish_plug(&plug);
388 }
389 f2fs_sync_fs(sbi->sb, true);
390 stat_inc_bg_cp_count(sbi->stat_info);
391 }
392 }
393
394 static int issue_flush_thread(void *data)
395 {
396 struct f2fs_sb_info *sbi = data;
397 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
398 wait_queue_head_t *q = &fcc->flush_wait_queue;
399 repeat:
400 if (kthread_should_stop())
401 return 0;
402
403 if (!llist_empty(&fcc->issue_list)) {
404 struct bio *bio;
405 struct flush_cmd *cmd, *next;
406 int ret;
407
408 bio = f2fs_bio_alloc(0);
409
410 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
411 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
412
413 bio->bi_bdev = sbi->sb->s_bdev;
414 ret = submit_bio_wait(WRITE_FLUSH, bio);
415
416 llist_for_each_entry_safe(cmd, next,
417 fcc->dispatch_list, llnode) {
418 cmd->ret = ret;
419 complete(&cmd->wait);
420 }
421 bio_put(bio);
422 fcc->dispatch_list = NULL;
423 }
424
425 wait_event_interruptible(*q,
426 kthread_should_stop() || !llist_empty(&fcc->issue_list));
427 goto repeat;
428 }
429
430 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
431 {
432 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
433 struct flush_cmd cmd;
434
435 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
436 test_opt(sbi, FLUSH_MERGE));
437
438 if (test_opt(sbi, NOBARRIER))
439 return 0;
440
441 if (!test_opt(sbi, FLUSH_MERGE) || !atomic_read(&fcc->submit_flush)) {
442 struct bio *bio = f2fs_bio_alloc(0);
443 int ret;
444
445 atomic_inc(&fcc->submit_flush);
446 bio->bi_bdev = sbi->sb->s_bdev;
447 ret = submit_bio_wait(WRITE_FLUSH, bio);
448 atomic_dec(&fcc->submit_flush);
449 bio_put(bio);
450 return ret;
451 }
452
453 init_completion(&cmd.wait);
454
455 atomic_inc(&fcc->submit_flush);
456 llist_add(&cmd.llnode, &fcc->issue_list);
457
458 if (!fcc->dispatch_list)
459 wake_up(&fcc->flush_wait_queue);
460
461 wait_for_completion(&cmd.wait);
462 atomic_dec(&fcc->submit_flush);
463
464 return cmd.ret;
465 }
466
467 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
468 {
469 dev_t dev = sbi->sb->s_bdev->bd_dev;
470 struct flush_cmd_control *fcc;
471 int err = 0;
472
473 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
474 if (!fcc)
475 return -ENOMEM;
476 atomic_set(&fcc->submit_flush, 0);
477 init_waitqueue_head(&fcc->flush_wait_queue);
478 init_llist_head(&fcc->issue_list);
479 SM_I(sbi)->cmd_control_info = fcc;
480 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
481 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
482 if (IS_ERR(fcc->f2fs_issue_flush)) {
483 err = PTR_ERR(fcc->f2fs_issue_flush);
484 kfree(fcc);
485 SM_I(sbi)->cmd_control_info = NULL;
486 return err;
487 }
488
489 return err;
490 }
491
492 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
493 {
494 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
495
496 if (fcc && fcc->f2fs_issue_flush)
497 kthread_stop(fcc->f2fs_issue_flush);
498 kfree(fcc);
499 SM_I(sbi)->cmd_control_info = NULL;
500 }
501
502 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
503 enum dirty_type dirty_type)
504 {
505 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
506
507 /* need not be added */
508 if (IS_CURSEG(sbi, segno))
509 return;
510
511 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
512 dirty_i->nr_dirty[dirty_type]++;
513
514 if (dirty_type == DIRTY) {
515 struct seg_entry *sentry = get_seg_entry(sbi, segno);
516 enum dirty_type t = sentry->type;
517
518 if (unlikely(t >= DIRTY)) {
519 f2fs_bug_on(sbi, 1);
520 return;
521 }
522 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
523 dirty_i->nr_dirty[t]++;
524 }
525 }
526
527 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
528 enum dirty_type dirty_type)
529 {
530 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
531
532 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
533 dirty_i->nr_dirty[dirty_type]--;
534
535 if (dirty_type == DIRTY) {
536 struct seg_entry *sentry = get_seg_entry(sbi, segno);
537 enum dirty_type t = sentry->type;
538
539 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
540 dirty_i->nr_dirty[t]--;
541
542 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
543 clear_bit(GET_SECNO(sbi, segno),
544 dirty_i->victim_secmap);
545 }
546 }
547
548 /*
549 * Should not occur error such as -ENOMEM.
550 * Adding dirty entry into seglist is not critical operation.
551 * If a given segment is one of current working segments, it won't be added.
552 */
553 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
554 {
555 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
556 unsigned short valid_blocks;
557
558 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
559 return;
560
561 mutex_lock(&dirty_i->seglist_lock);
562
563 valid_blocks = get_valid_blocks(sbi, segno, 0);
564
565 if (valid_blocks == 0) {
566 __locate_dirty_segment(sbi, segno, PRE);
567 __remove_dirty_segment(sbi, segno, DIRTY);
568 } else if (valid_blocks < sbi->blocks_per_seg) {
569 __locate_dirty_segment(sbi, segno, DIRTY);
570 } else {
571 /* Recovery routine with SSR needs this */
572 __remove_dirty_segment(sbi, segno, DIRTY);
573 }
574
575 mutex_unlock(&dirty_i->seglist_lock);
576 }
577
578 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
579 block_t blkstart, block_t blklen)
580 {
581 sector_t start = SECTOR_FROM_BLOCK(blkstart);
582 sector_t len = SECTOR_FROM_BLOCK(blklen);
583 struct seg_entry *se;
584 unsigned int offset;
585 block_t i;
586
587 for (i = blkstart; i < blkstart + blklen; i++) {
588 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
589 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
590
591 if (!f2fs_test_and_set_bit(offset, se->discard_map))
592 sbi->discard_blks--;
593 }
594 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
595 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
596 }
597
598 bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
599 {
600 int err = -EOPNOTSUPP;
601
602 if (test_opt(sbi, DISCARD)) {
603 struct seg_entry *se = get_seg_entry(sbi,
604 GET_SEGNO(sbi, blkaddr));
605 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
606
607 if (f2fs_test_bit(offset, se->discard_map))
608 return false;
609
610 err = f2fs_issue_discard(sbi, blkaddr, 1);
611 }
612
613 if (err) {
614 update_meta_page(sbi, NULL, blkaddr);
615 return true;
616 }
617 return false;
618 }
619
620 static void __add_discard_entry(struct f2fs_sb_info *sbi,
621 struct cp_control *cpc, struct seg_entry *se,
622 unsigned int start, unsigned int end)
623 {
624 struct list_head *head = &SM_I(sbi)->discard_list;
625 struct discard_entry *new, *last;
626
627 if (!list_empty(head)) {
628 last = list_last_entry(head, struct discard_entry, list);
629 if (START_BLOCK(sbi, cpc->trim_start) + start ==
630 last->blkaddr + last->len) {
631 last->len += end - start;
632 goto done;
633 }
634 }
635
636 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
637 INIT_LIST_HEAD(&new->list);
638 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
639 new->len = end - start;
640 list_add_tail(&new->list, head);
641 done:
642 SM_I(sbi)->nr_discards += end - start;
643 }
644
645 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
646 {
647 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
648 int max_blocks = sbi->blocks_per_seg;
649 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
650 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
651 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
652 unsigned long *discard_map = (unsigned long *)se->discard_map;
653 unsigned long *dmap = SIT_I(sbi)->tmp_map;
654 unsigned int start = 0, end = -1;
655 bool force = (cpc->reason == CP_DISCARD);
656 int i;
657
658 if (se->valid_blocks == max_blocks)
659 return;
660
661 if (!force) {
662 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
663 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
664 return;
665 }
666
667 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
668 for (i = 0; i < entries; i++)
669 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
670 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
671
672 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
673 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
674 if (start >= max_blocks)
675 break;
676
677 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
678 __add_discard_entry(sbi, cpc, se, start, end);
679 }
680 }
681
682 void release_discard_addrs(struct f2fs_sb_info *sbi)
683 {
684 struct list_head *head = &(SM_I(sbi)->discard_list);
685 struct discard_entry *entry, *this;
686
687 /* drop caches */
688 list_for_each_entry_safe(entry, this, head, list) {
689 list_del(&entry->list);
690 kmem_cache_free(discard_entry_slab, entry);
691 }
692 }
693
694 /*
695 * Should call clear_prefree_segments after checkpoint is done.
696 */
697 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
698 {
699 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
700 unsigned int segno;
701
702 mutex_lock(&dirty_i->seglist_lock);
703 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
704 __set_test_and_free(sbi, segno);
705 mutex_unlock(&dirty_i->seglist_lock);
706 }
707
708 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
709 {
710 struct list_head *head = &(SM_I(sbi)->discard_list);
711 struct discard_entry *entry, *this;
712 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
713 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
714 unsigned int start = 0, end = -1;
715
716 mutex_lock(&dirty_i->seglist_lock);
717
718 while (1) {
719 int i;
720 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
721 if (start >= MAIN_SEGS(sbi))
722 break;
723 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
724 start + 1);
725
726 for (i = start; i < end; i++)
727 clear_bit(i, prefree_map);
728
729 dirty_i->nr_dirty[PRE] -= end - start;
730
731 if (!test_opt(sbi, DISCARD))
732 continue;
733
734 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
735 (end - start) << sbi->log_blocks_per_seg);
736 }
737 mutex_unlock(&dirty_i->seglist_lock);
738
739 /* send small discards */
740 list_for_each_entry_safe(entry, this, head, list) {
741 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
742 goto skip;
743 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
744 cpc->trimmed += entry->len;
745 skip:
746 list_del(&entry->list);
747 SM_I(sbi)->nr_discards -= entry->len;
748 kmem_cache_free(discard_entry_slab, entry);
749 }
750 }
751
752 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
753 {
754 struct sit_info *sit_i = SIT_I(sbi);
755
756 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
757 sit_i->dirty_sentries++;
758 return false;
759 }
760
761 return true;
762 }
763
764 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
765 unsigned int segno, int modified)
766 {
767 struct seg_entry *se = get_seg_entry(sbi, segno);
768 se->type = type;
769 if (modified)
770 __mark_sit_entry_dirty(sbi, segno);
771 }
772
773 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
774 {
775 struct seg_entry *se;
776 unsigned int segno, offset;
777 long int new_vblocks;
778
779 segno = GET_SEGNO(sbi, blkaddr);
780
781 se = get_seg_entry(sbi, segno);
782 new_vblocks = se->valid_blocks + del;
783 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
784
785 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
786 (new_vblocks > sbi->blocks_per_seg)));
787
788 se->valid_blocks = new_vblocks;
789 se->mtime = get_mtime(sbi);
790 SIT_I(sbi)->max_mtime = se->mtime;
791
792 /* Update valid block bitmap */
793 if (del > 0) {
794 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
795 f2fs_bug_on(sbi, 1);
796 if (!f2fs_test_and_set_bit(offset, se->discard_map))
797 sbi->discard_blks--;
798 } else {
799 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
800 f2fs_bug_on(sbi, 1);
801 if (f2fs_test_and_clear_bit(offset, se->discard_map))
802 sbi->discard_blks++;
803 }
804 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
805 se->ckpt_valid_blocks += del;
806
807 __mark_sit_entry_dirty(sbi, segno);
808
809 /* update total number of valid blocks to be written in ckpt area */
810 SIT_I(sbi)->written_valid_blocks += del;
811
812 if (sbi->segs_per_sec > 1)
813 get_sec_entry(sbi, segno)->valid_blocks += del;
814 }
815
816 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
817 {
818 update_sit_entry(sbi, new, 1);
819 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
820 update_sit_entry(sbi, old, -1);
821
822 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
823 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
824 }
825
826 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
827 {
828 unsigned int segno = GET_SEGNO(sbi, addr);
829 struct sit_info *sit_i = SIT_I(sbi);
830
831 f2fs_bug_on(sbi, addr == NULL_ADDR);
832 if (addr == NEW_ADDR)
833 return;
834
835 /* add it into sit main buffer */
836 mutex_lock(&sit_i->sentry_lock);
837
838 update_sit_entry(sbi, addr, -1);
839
840 /* add it into dirty seglist */
841 locate_dirty_segment(sbi, segno);
842
843 mutex_unlock(&sit_i->sentry_lock);
844 }
845
846 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
847 {
848 struct sit_info *sit_i = SIT_I(sbi);
849 unsigned int segno, offset;
850 struct seg_entry *se;
851 bool is_cp = false;
852
853 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
854 return true;
855
856 mutex_lock(&sit_i->sentry_lock);
857
858 segno = GET_SEGNO(sbi, blkaddr);
859 se = get_seg_entry(sbi, segno);
860 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
861
862 if (f2fs_test_bit(offset, se->ckpt_valid_map))
863 is_cp = true;
864
865 mutex_unlock(&sit_i->sentry_lock);
866
867 return is_cp;
868 }
869
870 /*
871 * This function should be resided under the curseg_mutex lock
872 */
873 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
874 struct f2fs_summary *sum)
875 {
876 struct curseg_info *curseg = CURSEG_I(sbi, type);
877 void *addr = curseg->sum_blk;
878 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
879 memcpy(addr, sum, sizeof(struct f2fs_summary));
880 }
881
882 /*
883 * Calculate the number of current summary pages for writing
884 */
885 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
886 {
887 int valid_sum_count = 0;
888 int i, sum_in_page;
889
890 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
891 if (sbi->ckpt->alloc_type[i] == SSR)
892 valid_sum_count += sbi->blocks_per_seg;
893 else {
894 if (for_ra)
895 valid_sum_count += le16_to_cpu(
896 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
897 else
898 valid_sum_count += curseg_blkoff(sbi, i);
899 }
900 }
901
902 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
903 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
904 if (valid_sum_count <= sum_in_page)
905 return 1;
906 else if ((valid_sum_count - sum_in_page) <=
907 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
908 return 2;
909 return 3;
910 }
911
912 /*
913 * Caller should put this summary page
914 */
915 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
916 {
917 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
918 }
919
920 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
921 {
922 struct page *page = grab_meta_page(sbi, blk_addr);
923 void *dst = page_address(page);
924
925 if (src)
926 memcpy(dst, src, PAGE_SIZE);
927 else
928 memset(dst, 0, PAGE_SIZE);
929 set_page_dirty(page);
930 f2fs_put_page(page, 1);
931 }
932
933 static void write_sum_page(struct f2fs_sb_info *sbi,
934 struct f2fs_summary_block *sum_blk, block_t blk_addr)
935 {
936 update_meta_page(sbi, (void *)sum_blk, blk_addr);
937 }
938
939 static void write_current_sum_page(struct f2fs_sb_info *sbi,
940 int type, block_t blk_addr)
941 {
942 struct curseg_info *curseg = CURSEG_I(sbi, type);
943 struct page *page = grab_meta_page(sbi, blk_addr);
944 struct f2fs_summary_block *src = curseg->sum_blk;
945 struct f2fs_summary_block *dst;
946
947 dst = (struct f2fs_summary_block *)page_address(page);
948
949 mutex_lock(&curseg->curseg_mutex);
950
951 down_read(&curseg->journal_rwsem);
952 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
953 up_read(&curseg->journal_rwsem);
954
955 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
956 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
957
958 mutex_unlock(&curseg->curseg_mutex);
959
960 set_page_dirty(page);
961 f2fs_put_page(page, 1);
962 }
963
964 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
965 {
966 struct curseg_info *curseg = CURSEG_I(sbi, type);
967 unsigned int segno = curseg->segno + 1;
968 struct free_segmap_info *free_i = FREE_I(sbi);
969
970 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
971 return !test_bit(segno, free_i->free_segmap);
972 return 0;
973 }
974
975 /*
976 * Find a new segment from the free segments bitmap to right order
977 * This function should be returned with success, otherwise BUG
978 */
979 static void get_new_segment(struct f2fs_sb_info *sbi,
980 unsigned int *newseg, bool new_sec, int dir)
981 {
982 struct free_segmap_info *free_i = FREE_I(sbi);
983 unsigned int segno, secno, zoneno;
984 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
985 unsigned int hint = *newseg / sbi->segs_per_sec;
986 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
987 unsigned int left_start = hint;
988 bool init = true;
989 int go_left = 0;
990 int i;
991
992 spin_lock(&free_i->segmap_lock);
993
994 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
995 segno = find_next_zero_bit(free_i->free_segmap,
996 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
997 if (segno < (hint + 1) * sbi->segs_per_sec)
998 goto got_it;
999 }
1000 find_other_zone:
1001 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1002 if (secno >= MAIN_SECS(sbi)) {
1003 if (dir == ALLOC_RIGHT) {
1004 secno = find_next_zero_bit(free_i->free_secmap,
1005 MAIN_SECS(sbi), 0);
1006 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1007 } else {
1008 go_left = 1;
1009 left_start = hint - 1;
1010 }
1011 }
1012 if (go_left == 0)
1013 goto skip_left;
1014
1015 while (test_bit(left_start, free_i->free_secmap)) {
1016 if (left_start > 0) {
1017 left_start--;
1018 continue;
1019 }
1020 left_start = find_next_zero_bit(free_i->free_secmap,
1021 MAIN_SECS(sbi), 0);
1022 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1023 break;
1024 }
1025 secno = left_start;
1026 skip_left:
1027 hint = secno;
1028 segno = secno * sbi->segs_per_sec;
1029 zoneno = secno / sbi->secs_per_zone;
1030
1031 /* give up on finding another zone */
1032 if (!init)
1033 goto got_it;
1034 if (sbi->secs_per_zone == 1)
1035 goto got_it;
1036 if (zoneno == old_zoneno)
1037 goto got_it;
1038 if (dir == ALLOC_LEFT) {
1039 if (!go_left && zoneno + 1 >= total_zones)
1040 goto got_it;
1041 if (go_left && zoneno == 0)
1042 goto got_it;
1043 }
1044 for (i = 0; i < NR_CURSEG_TYPE; i++)
1045 if (CURSEG_I(sbi, i)->zone == zoneno)
1046 break;
1047
1048 if (i < NR_CURSEG_TYPE) {
1049 /* zone is in user, try another */
1050 if (go_left)
1051 hint = zoneno * sbi->secs_per_zone - 1;
1052 else if (zoneno + 1 >= total_zones)
1053 hint = 0;
1054 else
1055 hint = (zoneno + 1) * sbi->secs_per_zone;
1056 init = false;
1057 goto find_other_zone;
1058 }
1059 got_it:
1060 /* set it as dirty segment in free segmap */
1061 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1062 __set_inuse(sbi, segno);
1063 *newseg = segno;
1064 spin_unlock(&free_i->segmap_lock);
1065 }
1066
1067 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1068 {
1069 struct curseg_info *curseg = CURSEG_I(sbi, type);
1070 struct summary_footer *sum_footer;
1071
1072 curseg->segno = curseg->next_segno;
1073 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1074 curseg->next_blkoff = 0;
1075 curseg->next_segno = NULL_SEGNO;
1076
1077 sum_footer = &(curseg->sum_blk->footer);
1078 memset(sum_footer, 0, sizeof(struct summary_footer));
1079 if (IS_DATASEG(type))
1080 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1081 if (IS_NODESEG(type))
1082 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1083 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1084 }
1085
1086 /*
1087 * Allocate a current working segment.
1088 * This function always allocates a free segment in LFS manner.
1089 */
1090 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1091 {
1092 struct curseg_info *curseg = CURSEG_I(sbi, type);
1093 unsigned int segno = curseg->segno;
1094 int dir = ALLOC_LEFT;
1095
1096 write_sum_page(sbi, curseg->sum_blk,
1097 GET_SUM_BLOCK(sbi, segno));
1098 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1099 dir = ALLOC_RIGHT;
1100
1101 if (test_opt(sbi, NOHEAP))
1102 dir = ALLOC_RIGHT;
1103
1104 get_new_segment(sbi, &segno, new_sec, dir);
1105 curseg->next_segno = segno;
1106 reset_curseg(sbi, type, 1);
1107 curseg->alloc_type = LFS;
1108 }
1109
1110 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1111 struct curseg_info *seg, block_t start)
1112 {
1113 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1114 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1115 unsigned long *target_map = SIT_I(sbi)->tmp_map;
1116 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1117 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1118 int i, pos;
1119
1120 for (i = 0; i < entries; i++)
1121 target_map[i] = ckpt_map[i] | cur_map[i];
1122
1123 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1124
1125 seg->next_blkoff = pos;
1126 }
1127
1128 /*
1129 * If a segment is written by LFS manner, next block offset is just obtained
1130 * by increasing the current block offset. However, if a segment is written by
1131 * SSR manner, next block offset obtained by calling __next_free_blkoff
1132 */
1133 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1134 struct curseg_info *seg)
1135 {
1136 if (seg->alloc_type == SSR)
1137 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1138 else
1139 seg->next_blkoff++;
1140 }
1141
1142 /*
1143 * This function always allocates a used segment(from dirty seglist) by SSR
1144 * manner, so it should recover the existing segment information of valid blocks
1145 */
1146 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1147 {
1148 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1149 struct curseg_info *curseg = CURSEG_I(sbi, type);
1150 unsigned int new_segno = curseg->next_segno;
1151 struct f2fs_summary_block *sum_node;
1152 struct page *sum_page;
1153
1154 write_sum_page(sbi, curseg->sum_blk,
1155 GET_SUM_BLOCK(sbi, curseg->segno));
1156 __set_test_and_inuse(sbi, new_segno);
1157
1158 mutex_lock(&dirty_i->seglist_lock);
1159 __remove_dirty_segment(sbi, new_segno, PRE);
1160 __remove_dirty_segment(sbi, new_segno, DIRTY);
1161 mutex_unlock(&dirty_i->seglist_lock);
1162
1163 reset_curseg(sbi, type, 1);
1164 curseg->alloc_type = SSR;
1165 __next_free_blkoff(sbi, curseg, 0);
1166
1167 if (reuse) {
1168 sum_page = get_sum_page(sbi, new_segno);
1169 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1170 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1171 f2fs_put_page(sum_page, 1);
1172 }
1173 }
1174
1175 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1176 {
1177 struct curseg_info *curseg = CURSEG_I(sbi, type);
1178 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1179
1180 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1181 return v_ops->get_victim(sbi,
1182 &(curseg)->next_segno, BG_GC, type, SSR);
1183
1184 /* For data segments, let's do SSR more intensively */
1185 for (; type >= CURSEG_HOT_DATA; type--)
1186 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1187 BG_GC, type, SSR))
1188 return 1;
1189 return 0;
1190 }
1191
1192 /*
1193 * flush out current segment and replace it with new segment
1194 * This function should be returned with success, otherwise BUG
1195 */
1196 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1197 int type, bool force)
1198 {
1199 struct curseg_info *curseg = CURSEG_I(sbi, type);
1200
1201 if (force)
1202 new_curseg(sbi, type, true);
1203 else if (type == CURSEG_WARM_NODE)
1204 new_curseg(sbi, type, false);
1205 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1206 new_curseg(sbi, type, false);
1207 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1208 change_curseg(sbi, type, true);
1209 else
1210 new_curseg(sbi, type, false);
1211
1212 stat_inc_seg_type(sbi, curseg);
1213 }
1214
1215 static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1216 {
1217 struct curseg_info *curseg = CURSEG_I(sbi, type);
1218 unsigned int old_segno;
1219
1220 old_segno = curseg->segno;
1221 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1222 locate_dirty_segment(sbi, old_segno);
1223 }
1224
1225 void allocate_new_segments(struct f2fs_sb_info *sbi)
1226 {
1227 int i;
1228
1229 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1230 __allocate_new_segments(sbi, i);
1231 }
1232
1233 static const struct segment_allocation default_salloc_ops = {
1234 .allocate_segment = allocate_segment_by_default,
1235 };
1236
1237 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1238 {
1239 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1240 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1241 unsigned int start_segno, end_segno;
1242 struct cp_control cpc;
1243 int err = 0;
1244
1245 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1246 return -EINVAL;
1247
1248 cpc.trimmed = 0;
1249 if (end <= MAIN_BLKADDR(sbi))
1250 goto out;
1251
1252 /* start/end segment number in main_area */
1253 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1254 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1255 GET_SEGNO(sbi, end);
1256 cpc.reason = CP_DISCARD;
1257 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1258
1259 /* do checkpoint to issue discard commands safely */
1260 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1261 cpc.trim_start = start_segno;
1262
1263 if (sbi->discard_blks == 0)
1264 break;
1265 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1266 cpc.trim_end = end_segno;
1267 else
1268 cpc.trim_end = min_t(unsigned int,
1269 rounddown(start_segno +
1270 BATCHED_TRIM_SEGMENTS(sbi),
1271 sbi->segs_per_sec) - 1, end_segno);
1272
1273 mutex_lock(&sbi->gc_mutex);
1274 err = write_checkpoint(sbi, &cpc);
1275 mutex_unlock(&sbi->gc_mutex);
1276 }
1277 out:
1278 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1279 return err;
1280 }
1281
1282 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1283 {
1284 struct curseg_info *curseg = CURSEG_I(sbi, type);
1285 if (curseg->next_blkoff < sbi->blocks_per_seg)
1286 return true;
1287 return false;
1288 }
1289
1290 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1291 {
1292 if (p_type == DATA)
1293 return CURSEG_HOT_DATA;
1294 else
1295 return CURSEG_HOT_NODE;
1296 }
1297
1298 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1299 {
1300 if (p_type == DATA) {
1301 struct inode *inode = page->mapping->host;
1302
1303 if (S_ISDIR(inode->i_mode))
1304 return CURSEG_HOT_DATA;
1305 else
1306 return CURSEG_COLD_DATA;
1307 } else {
1308 if (IS_DNODE(page) && is_cold_node(page))
1309 return CURSEG_WARM_NODE;
1310 else
1311 return CURSEG_COLD_NODE;
1312 }
1313 }
1314
1315 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1316 {
1317 if (p_type == DATA) {
1318 struct inode *inode = page->mapping->host;
1319
1320 if (S_ISDIR(inode->i_mode))
1321 return CURSEG_HOT_DATA;
1322 else if (is_cold_data(page) || file_is_cold(inode))
1323 return CURSEG_COLD_DATA;
1324 else
1325 return CURSEG_WARM_DATA;
1326 } else {
1327 if (IS_DNODE(page))
1328 return is_cold_node(page) ? CURSEG_WARM_NODE :
1329 CURSEG_HOT_NODE;
1330 else
1331 return CURSEG_COLD_NODE;
1332 }
1333 }
1334
1335 static int __get_segment_type(struct page *page, enum page_type p_type)
1336 {
1337 switch (F2FS_P_SB(page)->active_logs) {
1338 case 2:
1339 return __get_segment_type_2(page, p_type);
1340 case 4:
1341 return __get_segment_type_4(page, p_type);
1342 }
1343 /* NR_CURSEG_TYPE(6) logs by default */
1344 f2fs_bug_on(F2FS_P_SB(page),
1345 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1346 return __get_segment_type_6(page, p_type);
1347 }
1348
1349 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1350 block_t old_blkaddr, block_t *new_blkaddr,
1351 struct f2fs_summary *sum, int type)
1352 {
1353 struct sit_info *sit_i = SIT_I(sbi);
1354 struct curseg_info *curseg;
1355 bool direct_io = (type == CURSEG_DIRECT_IO);
1356
1357 type = direct_io ? CURSEG_WARM_DATA : type;
1358
1359 curseg = CURSEG_I(sbi, type);
1360
1361 mutex_lock(&curseg->curseg_mutex);
1362 mutex_lock(&sit_i->sentry_lock);
1363
1364 /* direct_io'ed data is aligned to the segment for better performance */
1365 if (direct_io && curseg->next_blkoff &&
1366 !has_not_enough_free_secs(sbi, 0))
1367 __allocate_new_segments(sbi, type);
1368
1369 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1370
1371 /*
1372 * __add_sum_entry should be resided under the curseg_mutex
1373 * because, this function updates a summary entry in the
1374 * current summary block.
1375 */
1376 __add_sum_entry(sbi, type, sum);
1377
1378 __refresh_next_blkoff(sbi, curseg);
1379
1380 stat_inc_block_count(sbi, curseg);
1381
1382 if (!__has_curseg_space(sbi, type))
1383 sit_i->s_ops->allocate_segment(sbi, type, false);
1384 /*
1385 * SIT information should be updated before segment allocation,
1386 * since SSR needs latest valid block information.
1387 */
1388 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1389
1390 mutex_unlock(&sit_i->sentry_lock);
1391
1392 if (page && IS_NODESEG(type))
1393 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1394
1395 mutex_unlock(&curseg->curseg_mutex);
1396 }
1397
1398 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1399 {
1400 int type = __get_segment_type(fio->page, fio->type);
1401
1402 if (fio->type == NODE || fio->type == DATA)
1403 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1404
1405 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1406 &fio->new_blkaddr, sum, type);
1407
1408 /* writeout dirty page into bdev */
1409 f2fs_submit_page_mbio(fio);
1410
1411 if (fio->type == NODE || fio->type == DATA)
1412 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
1413 }
1414
1415 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1416 {
1417 struct f2fs_io_info fio = {
1418 .sbi = sbi,
1419 .type = META,
1420 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1421 .old_blkaddr = page->index,
1422 .new_blkaddr = page->index,
1423 .page = page,
1424 .encrypted_page = NULL,
1425 };
1426
1427 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1428 fio.rw &= ~REQ_META;
1429
1430 set_page_writeback(page);
1431 f2fs_submit_page_mbio(&fio);
1432 }
1433
1434 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1435 {
1436 struct f2fs_summary sum;
1437
1438 set_summary(&sum, nid, 0, 0);
1439 do_write_page(&sum, fio);
1440 }
1441
1442 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1443 {
1444 struct f2fs_sb_info *sbi = fio->sbi;
1445 struct f2fs_summary sum;
1446 struct node_info ni;
1447
1448 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1449 get_node_info(sbi, dn->nid, &ni);
1450 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1451 do_write_page(&sum, fio);
1452 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1453 }
1454
1455 void rewrite_data_page(struct f2fs_io_info *fio)
1456 {
1457 fio->new_blkaddr = fio->old_blkaddr;
1458 stat_inc_inplace_blocks(fio->sbi);
1459 f2fs_submit_page_mbio(fio);
1460 }
1461
1462 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1463 block_t old_blkaddr, block_t new_blkaddr,
1464 bool recover_curseg, bool recover_newaddr)
1465 {
1466 struct sit_info *sit_i = SIT_I(sbi);
1467 struct curseg_info *curseg;
1468 unsigned int segno, old_cursegno;
1469 struct seg_entry *se;
1470 int type;
1471 unsigned short old_blkoff;
1472
1473 segno = GET_SEGNO(sbi, new_blkaddr);
1474 se = get_seg_entry(sbi, segno);
1475 type = se->type;
1476
1477 if (!recover_curseg) {
1478 /* for recovery flow */
1479 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1480 if (old_blkaddr == NULL_ADDR)
1481 type = CURSEG_COLD_DATA;
1482 else
1483 type = CURSEG_WARM_DATA;
1484 }
1485 } else {
1486 if (!IS_CURSEG(sbi, segno))
1487 type = CURSEG_WARM_DATA;
1488 }
1489
1490 curseg = CURSEG_I(sbi, type);
1491
1492 mutex_lock(&curseg->curseg_mutex);
1493 mutex_lock(&sit_i->sentry_lock);
1494
1495 old_cursegno = curseg->segno;
1496 old_blkoff = curseg->next_blkoff;
1497
1498 /* change the current segment */
1499 if (segno != curseg->segno) {
1500 curseg->next_segno = segno;
1501 change_curseg(sbi, type, true);
1502 }
1503
1504 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1505 __add_sum_entry(sbi, type, sum);
1506
1507 if (!recover_curseg || recover_newaddr)
1508 update_sit_entry(sbi, new_blkaddr, 1);
1509 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1510 update_sit_entry(sbi, old_blkaddr, -1);
1511
1512 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1513 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1514
1515 locate_dirty_segment(sbi, old_cursegno);
1516
1517 if (recover_curseg) {
1518 if (old_cursegno != curseg->segno) {
1519 curseg->next_segno = old_cursegno;
1520 change_curseg(sbi, type, true);
1521 }
1522 curseg->next_blkoff = old_blkoff;
1523 }
1524
1525 mutex_unlock(&sit_i->sentry_lock);
1526 mutex_unlock(&curseg->curseg_mutex);
1527 }
1528
1529 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1530 block_t old_addr, block_t new_addr,
1531 unsigned char version, bool recover_curseg,
1532 bool recover_newaddr)
1533 {
1534 struct f2fs_summary sum;
1535
1536 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1537
1538 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1539 recover_curseg, recover_newaddr);
1540
1541 f2fs_update_data_blkaddr(dn, new_addr);
1542 }
1543
1544 void f2fs_wait_on_page_writeback(struct page *page,
1545 enum page_type type, bool ordered)
1546 {
1547 if (PageWriteback(page)) {
1548 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1549
1550 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
1551 if (ordered)
1552 wait_on_page_writeback(page);
1553 else
1554 wait_for_stable_page(page);
1555 }
1556 }
1557
1558 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1559 block_t blkaddr)
1560 {
1561 struct page *cpage;
1562
1563 if (blkaddr == NEW_ADDR)
1564 return;
1565
1566 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1567
1568 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1569 if (cpage) {
1570 f2fs_wait_on_page_writeback(cpage, DATA, true);
1571 f2fs_put_page(cpage, 1);
1572 }
1573 }
1574
1575 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1576 {
1577 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1578 struct curseg_info *seg_i;
1579 unsigned char *kaddr;
1580 struct page *page;
1581 block_t start;
1582 int i, j, offset;
1583
1584 start = start_sum_block(sbi);
1585
1586 page = get_meta_page(sbi, start++);
1587 kaddr = (unsigned char *)page_address(page);
1588
1589 /* Step 1: restore nat cache */
1590 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1591 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
1592
1593 /* Step 2: restore sit cache */
1594 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1595 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
1596 offset = 2 * SUM_JOURNAL_SIZE;
1597
1598 /* Step 3: restore summary entries */
1599 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1600 unsigned short blk_off;
1601 unsigned int segno;
1602
1603 seg_i = CURSEG_I(sbi, i);
1604 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1605 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1606 seg_i->next_segno = segno;
1607 reset_curseg(sbi, i, 0);
1608 seg_i->alloc_type = ckpt->alloc_type[i];
1609 seg_i->next_blkoff = blk_off;
1610
1611 if (seg_i->alloc_type == SSR)
1612 blk_off = sbi->blocks_per_seg;
1613
1614 for (j = 0; j < blk_off; j++) {
1615 struct f2fs_summary *s;
1616 s = (struct f2fs_summary *)(kaddr + offset);
1617 seg_i->sum_blk->entries[j] = *s;
1618 offset += SUMMARY_SIZE;
1619 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
1620 SUM_FOOTER_SIZE)
1621 continue;
1622
1623 f2fs_put_page(page, 1);
1624 page = NULL;
1625
1626 page = get_meta_page(sbi, start++);
1627 kaddr = (unsigned char *)page_address(page);
1628 offset = 0;
1629 }
1630 }
1631 f2fs_put_page(page, 1);
1632 return 0;
1633 }
1634
1635 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1636 {
1637 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1638 struct f2fs_summary_block *sum;
1639 struct curseg_info *curseg;
1640 struct page *new;
1641 unsigned short blk_off;
1642 unsigned int segno = 0;
1643 block_t blk_addr = 0;
1644
1645 /* get segment number and block addr */
1646 if (IS_DATASEG(type)) {
1647 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1648 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1649 CURSEG_HOT_DATA]);
1650 if (__exist_node_summaries(sbi))
1651 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1652 else
1653 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1654 } else {
1655 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1656 CURSEG_HOT_NODE]);
1657 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1658 CURSEG_HOT_NODE]);
1659 if (__exist_node_summaries(sbi))
1660 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1661 type - CURSEG_HOT_NODE);
1662 else
1663 blk_addr = GET_SUM_BLOCK(sbi, segno);
1664 }
1665
1666 new = get_meta_page(sbi, blk_addr);
1667 sum = (struct f2fs_summary_block *)page_address(new);
1668
1669 if (IS_NODESEG(type)) {
1670 if (__exist_node_summaries(sbi)) {
1671 struct f2fs_summary *ns = &sum->entries[0];
1672 int i;
1673 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1674 ns->version = 0;
1675 ns->ofs_in_node = 0;
1676 }
1677 } else {
1678 int err;
1679
1680 err = restore_node_summary(sbi, segno, sum);
1681 if (err) {
1682 f2fs_put_page(new, 1);
1683 return err;
1684 }
1685 }
1686 }
1687
1688 /* set uncompleted segment to curseg */
1689 curseg = CURSEG_I(sbi, type);
1690 mutex_lock(&curseg->curseg_mutex);
1691
1692 /* update journal info */
1693 down_write(&curseg->journal_rwsem);
1694 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1695 up_write(&curseg->journal_rwsem);
1696
1697 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1698 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
1699 curseg->next_segno = segno;
1700 reset_curseg(sbi, type, 0);
1701 curseg->alloc_type = ckpt->alloc_type[type];
1702 curseg->next_blkoff = blk_off;
1703 mutex_unlock(&curseg->curseg_mutex);
1704 f2fs_put_page(new, 1);
1705 return 0;
1706 }
1707
1708 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1709 {
1710 int type = CURSEG_HOT_DATA;
1711 int err;
1712
1713 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1714 int npages = npages_for_summary_flush(sbi, true);
1715
1716 if (npages >= 2)
1717 ra_meta_pages(sbi, start_sum_block(sbi), npages,
1718 META_CP, true);
1719
1720 /* restore for compacted data summary */
1721 if (read_compacted_summaries(sbi))
1722 return -EINVAL;
1723 type = CURSEG_HOT_NODE;
1724 }
1725
1726 if (__exist_node_summaries(sbi))
1727 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1728 NR_CURSEG_TYPE - type, META_CP, true);
1729
1730 for (; type <= CURSEG_COLD_NODE; type++) {
1731 err = read_normal_summaries(sbi, type);
1732 if (err)
1733 return err;
1734 }
1735
1736 return 0;
1737 }
1738
1739 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1740 {
1741 struct page *page;
1742 unsigned char *kaddr;
1743 struct f2fs_summary *summary;
1744 struct curseg_info *seg_i;
1745 int written_size = 0;
1746 int i, j;
1747
1748 page = grab_meta_page(sbi, blkaddr++);
1749 kaddr = (unsigned char *)page_address(page);
1750
1751 /* Step 1: write nat cache */
1752 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1753 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
1754 written_size += SUM_JOURNAL_SIZE;
1755
1756 /* Step 2: write sit cache */
1757 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1758 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
1759 written_size += SUM_JOURNAL_SIZE;
1760
1761 /* Step 3: write summary entries */
1762 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1763 unsigned short blkoff;
1764 seg_i = CURSEG_I(sbi, i);
1765 if (sbi->ckpt->alloc_type[i] == SSR)
1766 blkoff = sbi->blocks_per_seg;
1767 else
1768 blkoff = curseg_blkoff(sbi, i);
1769
1770 for (j = 0; j < blkoff; j++) {
1771 if (!page) {
1772 page = grab_meta_page(sbi, blkaddr++);
1773 kaddr = (unsigned char *)page_address(page);
1774 written_size = 0;
1775 }
1776 summary = (struct f2fs_summary *)(kaddr + written_size);
1777 *summary = seg_i->sum_blk->entries[j];
1778 written_size += SUMMARY_SIZE;
1779
1780 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
1781 SUM_FOOTER_SIZE)
1782 continue;
1783
1784 set_page_dirty(page);
1785 f2fs_put_page(page, 1);
1786 page = NULL;
1787 }
1788 }
1789 if (page) {
1790 set_page_dirty(page);
1791 f2fs_put_page(page, 1);
1792 }
1793 }
1794
1795 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1796 block_t blkaddr, int type)
1797 {
1798 int i, end;
1799 if (IS_DATASEG(type))
1800 end = type + NR_CURSEG_DATA_TYPE;
1801 else
1802 end = type + NR_CURSEG_NODE_TYPE;
1803
1804 for (i = type; i < end; i++)
1805 write_current_sum_page(sbi, i, blkaddr + (i - type));
1806 }
1807
1808 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1809 {
1810 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1811 write_compacted_summaries(sbi, start_blk);
1812 else
1813 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1814 }
1815
1816 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1817 {
1818 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1819 }
1820
1821 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
1822 unsigned int val, int alloc)
1823 {
1824 int i;
1825
1826 if (type == NAT_JOURNAL) {
1827 for (i = 0; i < nats_in_cursum(journal); i++) {
1828 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
1829 return i;
1830 }
1831 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1832 return update_nats_in_cursum(journal, 1);
1833 } else if (type == SIT_JOURNAL) {
1834 for (i = 0; i < sits_in_cursum(journal); i++)
1835 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
1836 return i;
1837 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1838 return update_sits_in_cursum(journal, 1);
1839 }
1840 return -1;
1841 }
1842
1843 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1844 unsigned int segno)
1845 {
1846 return get_meta_page(sbi, current_sit_addr(sbi, segno));
1847 }
1848
1849 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1850 unsigned int start)
1851 {
1852 struct sit_info *sit_i = SIT_I(sbi);
1853 struct page *src_page, *dst_page;
1854 pgoff_t src_off, dst_off;
1855 void *src_addr, *dst_addr;
1856
1857 src_off = current_sit_addr(sbi, start);
1858 dst_off = next_sit_addr(sbi, src_off);
1859
1860 /* get current sit block page without lock */
1861 src_page = get_meta_page(sbi, src_off);
1862 dst_page = grab_meta_page(sbi, dst_off);
1863 f2fs_bug_on(sbi, PageDirty(src_page));
1864
1865 src_addr = page_address(src_page);
1866 dst_addr = page_address(dst_page);
1867 memcpy(dst_addr, src_addr, PAGE_SIZE);
1868
1869 set_page_dirty(dst_page);
1870 f2fs_put_page(src_page, 1);
1871
1872 set_to_next_sit(sit_i, start);
1873
1874 return dst_page;
1875 }
1876
1877 static struct sit_entry_set *grab_sit_entry_set(void)
1878 {
1879 struct sit_entry_set *ses =
1880 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
1881
1882 ses->entry_cnt = 0;
1883 INIT_LIST_HEAD(&ses->set_list);
1884 return ses;
1885 }
1886
1887 static void release_sit_entry_set(struct sit_entry_set *ses)
1888 {
1889 list_del(&ses->set_list);
1890 kmem_cache_free(sit_entry_set_slab, ses);
1891 }
1892
1893 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1894 struct list_head *head)
1895 {
1896 struct sit_entry_set *next = ses;
1897
1898 if (list_is_last(&ses->set_list, head))
1899 return;
1900
1901 list_for_each_entry_continue(next, head, set_list)
1902 if (ses->entry_cnt <= next->entry_cnt)
1903 break;
1904
1905 list_move_tail(&ses->set_list, &next->set_list);
1906 }
1907
1908 static void add_sit_entry(unsigned int segno, struct list_head *head)
1909 {
1910 struct sit_entry_set *ses;
1911 unsigned int start_segno = START_SEGNO(segno);
1912
1913 list_for_each_entry(ses, head, set_list) {
1914 if (ses->start_segno == start_segno) {
1915 ses->entry_cnt++;
1916 adjust_sit_entry_set(ses, head);
1917 return;
1918 }
1919 }
1920
1921 ses = grab_sit_entry_set();
1922
1923 ses->start_segno = start_segno;
1924 ses->entry_cnt++;
1925 list_add(&ses->set_list, head);
1926 }
1927
1928 static void add_sits_in_set(struct f2fs_sb_info *sbi)
1929 {
1930 struct f2fs_sm_info *sm_info = SM_I(sbi);
1931 struct list_head *set_list = &sm_info->sit_entry_set;
1932 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
1933 unsigned int segno;
1934
1935 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
1936 add_sit_entry(segno, set_list);
1937 }
1938
1939 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
1940 {
1941 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1942 struct f2fs_journal *journal = curseg->journal;
1943 int i;
1944
1945 down_write(&curseg->journal_rwsem);
1946 for (i = 0; i < sits_in_cursum(journal); i++) {
1947 unsigned int segno;
1948 bool dirtied;
1949
1950 segno = le32_to_cpu(segno_in_journal(journal, i));
1951 dirtied = __mark_sit_entry_dirty(sbi, segno);
1952
1953 if (!dirtied)
1954 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
1955 }
1956 update_sits_in_cursum(journal, -i);
1957 up_write(&curseg->journal_rwsem);
1958 }
1959
1960 /*
1961 * CP calls this function, which flushes SIT entries including sit_journal,
1962 * and moves prefree segs to free segs.
1963 */
1964 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1965 {
1966 struct sit_info *sit_i = SIT_I(sbi);
1967 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1968 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1969 struct f2fs_journal *journal = curseg->journal;
1970 struct sit_entry_set *ses, *tmp;
1971 struct list_head *head = &SM_I(sbi)->sit_entry_set;
1972 bool to_journal = true;
1973 struct seg_entry *se;
1974
1975 mutex_lock(&sit_i->sentry_lock);
1976
1977 if (!sit_i->dirty_sentries)
1978 goto out;
1979
1980 /*
1981 * add and account sit entries of dirty bitmap in sit entry
1982 * set temporarily
1983 */
1984 add_sits_in_set(sbi);
1985
1986 /*
1987 * if there are no enough space in journal to store dirty sit
1988 * entries, remove all entries from journal and add and account
1989 * them in sit entry set.
1990 */
1991 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
1992 remove_sits_in_journal(sbi);
1993
1994 /*
1995 * there are two steps to flush sit entries:
1996 * #1, flush sit entries to journal in current cold data summary block.
1997 * #2, flush sit entries to sit page.
1998 */
1999 list_for_each_entry_safe(ses, tmp, head, set_list) {
2000 struct page *page = NULL;
2001 struct f2fs_sit_block *raw_sit = NULL;
2002 unsigned int start_segno = ses->start_segno;
2003 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2004 (unsigned long)MAIN_SEGS(sbi));
2005 unsigned int segno = start_segno;
2006
2007 if (to_journal &&
2008 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2009 to_journal = false;
2010
2011 if (to_journal) {
2012 down_write(&curseg->journal_rwsem);
2013 } else {
2014 page = get_next_sit_page(sbi, start_segno);
2015 raw_sit = page_address(page);
2016 }
2017
2018 /* flush dirty sit entries in region of current sit set */
2019 for_each_set_bit_from(segno, bitmap, end) {
2020 int offset, sit_offset;
2021
2022 se = get_seg_entry(sbi, segno);
2023
2024 /* add discard candidates */
2025 if (cpc->reason != CP_DISCARD) {
2026 cpc->trim_start = segno;
2027 add_discard_addrs(sbi, cpc);
2028 }
2029
2030 if (to_journal) {
2031 offset = lookup_journal_in_cursum(journal,
2032 SIT_JOURNAL, segno, 1);
2033 f2fs_bug_on(sbi, offset < 0);
2034 segno_in_journal(journal, offset) =
2035 cpu_to_le32(segno);
2036 seg_info_to_raw_sit(se,
2037 &sit_in_journal(journal, offset));
2038 } else {
2039 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2040 seg_info_to_raw_sit(se,
2041 &raw_sit->entries[sit_offset]);
2042 }
2043
2044 __clear_bit(segno, bitmap);
2045 sit_i->dirty_sentries--;
2046 ses->entry_cnt--;
2047 }
2048
2049 if (to_journal)
2050 up_write(&curseg->journal_rwsem);
2051 else
2052 f2fs_put_page(page, 1);
2053
2054 f2fs_bug_on(sbi, ses->entry_cnt);
2055 release_sit_entry_set(ses);
2056 }
2057
2058 f2fs_bug_on(sbi, !list_empty(head));
2059 f2fs_bug_on(sbi, sit_i->dirty_sentries);
2060 out:
2061 if (cpc->reason == CP_DISCARD) {
2062 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2063 add_discard_addrs(sbi, cpc);
2064 }
2065 mutex_unlock(&sit_i->sentry_lock);
2066
2067 set_prefree_as_free_segments(sbi);
2068 }
2069
2070 static int build_sit_info(struct f2fs_sb_info *sbi)
2071 {
2072 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2073 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2074 struct sit_info *sit_i;
2075 unsigned int sit_segs, start;
2076 char *src_bitmap, *dst_bitmap;
2077 unsigned int bitmap_size;
2078
2079 /* allocate memory for SIT information */
2080 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2081 if (!sit_i)
2082 return -ENOMEM;
2083
2084 SM_I(sbi)->sit_info = sit_i;
2085
2086 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2087 sizeof(struct seg_entry), GFP_KERNEL);
2088 if (!sit_i->sentries)
2089 return -ENOMEM;
2090
2091 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2092 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2093 if (!sit_i->dirty_sentries_bitmap)
2094 return -ENOMEM;
2095
2096 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2097 sit_i->sentries[start].cur_valid_map
2098 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2099 sit_i->sentries[start].ckpt_valid_map
2100 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2101 sit_i->sentries[start].discard_map
2102 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2103 if (!sit_i->sentries[start].cur_valid_map ||
2104 !sit_i->sentries[start].ckpt_valid_map ||
2105 !sit_i->sentries[start].discard_map)
2106 return -ENOMEM;
2107 }
2108
2109 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2110 if (!sit_i->tmp_map)
2111 return -ENOMEM;
2112
2113 if (sbi->segs_per_sec > 1) {
2114 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2115 sizeof(struct sec_entry), GFP_KERNEL);
2116 if (!sit_i->sec_entries)
2117 return -ENOMEM;
2118 }
2119
2120 /* get information related with SIT */
2121 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2122
2123 /* setup SIT bitmap from ckeckpoint pack */
2124 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2125 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2126
2127 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2128 if (!dst_bitmap)
2129 return -ENOMEM;
2130
2131 /* init SIT information */
2132 sit_i->s_ops = &default_salloc_ops;
2133
2134 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2135 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2136 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2137 sit_i->sit_bitmap = dst_bitmap;
2138 sit_i->bitmap_size = bitmap_size;
2139 sit_i->dirty_sentries = 0;
2140 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2141 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2142 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2143 mutex_init(&sit_i->sentry_lock);
2144 return 0;
2145 }
2146
2147 static int build_free_segmap(struct f2fs_sb_info *sbi)
2148 {
2149 struct free_segmap_info *free_i;
2150 unsigned int bitmap_size, sec_bitmap_size;
2151
2152 /* allocate memory for free segmap information */
2153 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2154 if (!free_i)
2155 return -ENOMEM;
2156
2157 SM_I(sbi)->free_info = free_i;
2158
2159 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2160 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2161 if (!free_i->free_segmap)
2162 return -ENOMEM;
2163
2164 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2165 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2166 if (!free_i->free_secmap)
2167 return -ENOMEM;
2168
2169 /* set all segments as dirty temporarily */
2170 memset(free_i->free_segmap, 0xff, bitmap_size);
2171 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2172
2173 /* init free segmap information */
2174 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2175 free_i->free_segments = 0;
2176 free_i->free_sections = 0;
2177 spin_lock_init(&free_i->segmap_lock);
2178 return 0;
2179 }
2180
2181 static int build_curseg(struct f2fs_sb_info *sbi)
2182 {
2183 struct curseg_info *array;
2184 int i;
2185
2186 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2187 if (!array)
2188 return -ENOMEM;
2189
2190 SM_I(sbi)->curseg_array = array;
2191
2192 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2193 mutex_init(&array[i].curseg_mutex);
2194 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2195 if (!array[i].sum_blk)
2196 return -ENOMEM;
2197 init_rwsem(&array[i].journal_rwsem);
2198 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2199 GFP_KERNEL);
2200 if (!array[i].journal)
2201 return -ENOMEM;
2202 array[i].segno = NULL_SEGNO;
2203 array[i].next_blkoff = 0;
2204 }
2205 return restore_curseg_summaries(sbi);
2206 }
2207
2208 static void build_sit_entries(struct f2fs_sb_info *sbi)
2209 {
2210 struct sit_info *sit_i = SIT_I(sbi);
2211 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2212 struct f2fs_journal *journal = curseg->journal;
2213 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2214 unsigned int i, start, end;
2215 unsigned int readed, start_blk = 0;
2216 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
2217
2218 do {
2219 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
2220
2221 start = start_blk * sit_i->sents_per_block;
2222 end = (start_blk + readed) * sit_i->sents_per_block;
2223
2224 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2225 struct seg_entry *se = &sit_i->sentries[start];
2226 struct f2fs_sit_block *sit_blk;
2227 struct f2fs_sit_entry sit;
2228 struct page *page;
2229
2230 down_read(&curseg->journal_rwsem);
2231 for (i = 0; i < sits_in_cursum(journal); i++) {
2232 if (le32_to_cpu(segno_in_journal(journal, i))
2233 == start) {
2234 sit = sit_in_journal(journal, i);
2235 up_read(&curseg->journal_rwsem);
2236 goto got_it;
2237 }
2238 }
2239 up_read(&curseg->journal_rwsem);
2240
2241 page = get_current_sit_page(sbi, start);
2242 sit_blk = (struct f2fs_sit_block *)page_address(page);
2243 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2244 f2fs_put_page(page, 1);
2245 got_it:
2246 check_block_count(sbi, start, &sit);
2247 seg_info_from_raw_sit(se, &sit);
2248
2249 /* build discard map only one time */
2250 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2251 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2252
2253 if (sbi->segs_per_sec > 1) {
2254 struct sec_entry *e = get_sec_entry(sbi, start);
2255 e->valid_blocks += se->valid_blocks;
2256 }
2257 }
2258 start_blk += readed;
2259 } while (start_blk < sit_blk_cnt);
2260 }
2261
2262 static void init_free_segmap(struct f2fs_sb_info *sbi)
2263 {
2264 unsigned int start;
2265 int type;
2266
2267 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2268 struct seg_entry *sentry = get_seg_entry(sbi, start);
2269 if (!sentry->valid_blocks)
2270 __set_free(sbi, start);
2271 }
2272
2273 /* set use the current segments */
2274 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2275 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2276 __set_test_and_inuse(sbi, curseg_t->segno);
2277 }
2278 }
2279
2280 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2281 {
2282 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2283 struct free_segmap_info *free_i = FREE_I(sbi);
2284 unsigned int segno = 0, offset = 0;
2285 unsigned short valid_blocks;
2286
2287 while (1) {
2288 /* find dirty segment based on free segmap */
2289 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2290 if (segno >= MAIN_SEGS(sbi))
2291 break;
2292 offset = segno + 1;
2293 valid_blocks = get_valid_blocks(sbi, segno, 0);
2294 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2295 continue;
2296 if (valid_blocks > sbi->blocks_per_seg) {
2297 f2fs_bug_on(sbi, 1);
2298 continue;
2299 }
2300 mutex_lock(&dirty_i->seglist_lock);
2301 __locate_dirty_segment(sbi, segno, DIRTY);
2302 mutex_unlock(&dirty_i->seglist_lock);
2303 }
2304 }
2305
2306 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2307 {
2308 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2309 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2310
2311 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2312 if (!dirty_i->victim_secmap)
2313 return -ENOMEM;
2314 return 0;
2315 }
2316
2317 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2318 {
2319 struct dirty_seglist_info *dirty_i;
2320 unsigned int bitmap_size, i;
2321
2322 /* allocate memory for dirty segments list information */
2323 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2324 if (!dirty_i)
2325 return -ENOMEM;
2326
2327 SM_I(sbi)->dirty_info = dirty_i;
2328 mutex_init(&dirty_i->seglist_lock);
2329
2330 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2331
2332 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2333 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2334 if (!dirty_i->dirty_segmap[i])
2335 return -ENOMEM;
2336 }
2337
2338 init_dirty_segmap(sbi);
2339 return init_victim_secmap(sbi);
2340 }
2341
2342 /*
2343 * Update min, max modified time for cost-benefit GC algorithm
2344 */
2345 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2346 {
2347 struct sit_info *sit_i = SIT_I(sbi);
2348 unsigned int segno;
2349
2350 mutex_lock(&sit_i->sentry_lock);
2351
2352 sit_i->min_mtime = LLONG_MAX;
2353
2354 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2355 unsigned int i;
2356 unsigned long long mtime = 0;
2357
2358 for (i = 0; i < sbi->segs_per_sec; i++)
2359 mtime += get_seg_entry(sbi, segno + i)->mtime;
2360
2361 mtime = div_u64(mtime, sbi->segs_per_sec);
2362
2363 if (sit_i->min_mtime > mtime)
2364 sit_i->min_mtime = mtime;
2365 }
2366 sit_i->max_mtime = get_mtime(sbi);
2367 mutex_unlock(&sit_i->sentry_lock);
2368 }
2369
2370 int build_segment_manager(struct f2fs_sb_info *sbi)
2371 {
2372 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2373 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2374 struct f2fs_sm_info *sm_info;
2375 int err;
2376
2377 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2378 if (!sm_info)
2379 return -ENOMEM;
2380
2381 /* init sm info */
2382 sbi->sm_info = sm_info;
2383 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2384 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2385 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2386 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2387 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2388 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2389 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2390 sm_info->rec_prefree_segments = sm_info->main_segments *
2391 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2392 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2393 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2394 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2395
2396 INIT_LIST_HEAD(&sm_info->discard_list);
2397 sm_info->nr_discards = 0;
2398 sm_info->max_discards = 0;
2399
2400 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2401
2402 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2403
2404 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2405 err = create_flush_cmd_control(sbi);
2406 if (err)
2407 return err;
2408 }
2409
2410 err = build_sit_info(sbi);
2411 if (err)
2412 return err;
2413 err = build_free_segmap(sbi);
2414 if (err)
2415 return err;
2416 err = build_curseg(sbi);
2417 if (err)
2418 return err;
2419
2420 /* reinit free segmap based on SIT */
2421 build_sit_entries(sbi);
2422
2423 init_free_segmap(sbi);
2424 err = build_dirty_segmap(sbi);
2425 if (err)
2426 return err;
2427
2428 init_min_max_mtime(sbi);
2429 return 0;
2430 }
2431
2432 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2433 enum dirty_type dirty_type)
2434 {
2435 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2436
2437 mutex_lock(&dirty_i->seglist_lock);
2438 kvfree(dirty_i->dirty_segmap[dirty_type]);
2439 dirty_i->nr_dirty[dirty_type] = 0;
2440 mutex_unlock(&dirty_i->seglist_lock);
2441 }
2442
2443 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2444 {
2445 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2446 kvfree(dirty_i->victim_secmap);
2447 }
2448
2449 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2450 {
2451 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2452 int i;
2453
2454 if (!dirty_i)
2455 return;
2456
2457 /* discard pre-free/dirty segments list */
2458 for (i = 0; i < NR_DIRTY_TYPE; i++)
2459 discard_dirty_segmap(sbi, i);
2460
2461 destroy_victim_secmap(sbi);
2462 SM_I(sbi)->dirty_info = NULL;
2463 kfree(dirty_i);
2464 }
2465
2466 static void destroy_curseg(struct f2fs_sb_info *sbi)
2467 {
2468 struct curseg_info *array = SM_I(sbi)->curseg_array;
2469 int i;
2470
2471 if (!array)
2472 return;
2473 SM_I(sbi)->curseg_array = NULL;
2474 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2475 kfree(array[i].sum_blk);
2476 kfree(array[i].journal);
2477 }
2478 kfree(array);
2479 }
2480
2481 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2482 {
2483 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2484 if (!free_i)
2485 return;
2486 SM_I(sbi)->free_info = NULL;
2487 kvfree(free_i->free_segmap);
2488 kvfree(free_i->free_secmap);
2489 kfree(free_i);
2490 }
2491
2492 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2493 {
2494 struct sit_info *sit_i = SIT_I(sbi);
2495 unsigned int start;
2496
2497 if (!sit_i)
2498 return;
2499
2500 if (sit_i->sentries) {
2501 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2502 kfree(sit_i->sentries[start].cur_valid_map);
2503 kfree(sit_i->sentries[start].ckpt_valid_map);
2504 kfree(sit_i->sentries[start].discard_map);
2505 }
2506 }
2507 kfree(sit_i->tmp_map);
2508
2509 kvfree(sit_i->sentries);
2510 kvfree(sit_i->sec_entries);
2511 kvfree(sit_i->dirty_sentries_bitmap);
2512
2513 SM_I(sbi)->sit_info = NULL;
2514 kfree(sit_i->sit_bitmap);
2515 kfree(sit_i);
2516 }
2517
2518 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2519 {
2520 struct f2fs_sm_info *sm_info = SM_I(sbi);
2521
2522 if (!sm_info)
2523 return;
2524 destroy_flush_cmd_control(sbi);
2525 destroy_dirty_segmap(sbi);
2526 destroy_curseg(sbi);
2527 destroy_free_segmap(sbi);
2528 destroy_sit_info(sbi);
2529 sbi->sm_info = NULL;
2530 kfree(sm_info);
2531 }
2532
2533 int __init create_segment_manager_caches(void)
2534 {
2535 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2536 sizeof(struct discard_entry));
2537 if (!discard_entry_slab)
2538 goto fail;
2539
2540 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2541 sizeof(struct sit_entry_set));
2542 if (!sit_entry_set_slab)
2543 goto destory_discard_entry;
2544
2545 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2546 sizeof(struct inmem_pages));
2547 if (!inmem_entry_slab)
2548 goto destroy_sit_entry_set;
2549 return 0;
2550
2551 destroy_sit_entry_set:
2552 kmem_cache_destroy(sit_entry_set_slab);
2553 destory_discard_entry:
2554 kmem_cache_destroy(discard_entry_slab);
2555 fail:
2556 return -ENOMEM;
2557 }
2558
2559 void destroy_segment_manager_caches(void)
2560 {
2561 kmem_cache_destroy(sit_entry_set_slab);
2562 kmem_cache_destroy(discard_entry_slab);
2563 kmem_cache_destroy(inmem_entry_slab);
2564 }