Loading...
Note: File does not exist in v3.1.
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
28static struct kmem_cache *discard_entry_slab;
29static struct kmem_cache *sit_entry_set_slab;
30static struct kmem_cache *inmem_entry_slab;
31
32static 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 */
51static 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 */
95static 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;
119pass:
120 if (size <= BITS_PER_LONG)
121 break;
122 size -= BITS_PER_LONG;
123 offset = 0;
124 p++;
125 }
126 return result;
127found:
128 return result - size + __reverse_ffs(tmp);
129}
130
131static 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;
156pass:
157 if (size <= BITS_PER_LONG)
158 break;
159 size -= BITS_PER_LONG;
160 offset = 0;
161 p++;
162 }
163 return result;
164found:
165 return result - size + __reverse_ffz(tmp);
166}
167
168void 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
194static 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 }
225next:
226 ClearPageUptodate(page);
227 set_page_private(page, 0);
228 ClearPageUptodate(page);
229 f2fs_put_page(page, 1);
230
231 list_del(&cur->list);
232 kmem_cache_free(inmem_entry_slab, cur);
233 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
234 }
235 return err;
236}
237
238void drop_inmem_pages(struct inode *inode)
239{
240 struct f2fs_inode_info *fi = F2FS_I(inode);
241
242 mutex_lock(&fi->inmem_lock);
243 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
244 mutex_unlock(&fi->inmem_lock);
245}
246
247static int __commit_inmem_pages(struct inode *inode,
248 struct list_head *revoke_list)
249{
250 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
251 struct f2fs_inode_info *fi = F2FS_I(inode);
252 struct inmem_pages *cur, *tmp;
253 struct f2fs_io_info fio = {
254 .sbi = sbi,
255 .type = DATA,
256 .rw = WRITE_SYNC | REQ_PRIO,
257 .encrypted_page = NULL,
258 };
259 bool submit_bio = false;
260 int err = 0;
261
262 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
263 struct page *page = cur->page;
264
265 lock_page(page);
266 if (page->mapping == inode->i_mapping) {
267 trace_f2fs_commit_inmem_page(page, INMEM);
268
269 set_page_dirty(page);
270 f2fs_wait_on_page_writeback(page, DATA, true);
271 if (clear_page_dirty_for_io(page))
272 inode_dec_dirty_pages(inode);
273
274 fio.page = page;
275 err = do_write_data_page(&fio);
276 if (err) {
277 unlock_page(page);
278 break;
279 }
280
281 /* record old blkaddr for revoking */
282 cur->old_addr = fio.old_blkaddr;
283
284 clear_cold_data(page);
285 submit_bio = true;
286 }
287 unlock_page(page);
288 list_move_tail(&cur->list, revoke_list);
289 }
290
291 if (submit_bio)
292 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
293
294 if (!err)
295 __revoke_inmem_pages(inode, revoke_list, false, false);
296
297 return err;
298}
299
300int commit_inmem_pages(struct inode *inode)
301{
302 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
303 struct f2fs_inode_info *fi = F2FS_I(inode);
304 struct list_head revoke_list;
305 int err;
306
307 INIT_LIST_HEAD(&revoke_list);
308 f2fs_balance_fs(sbi, true);
309 f2fs_lock_op(sbi);
310
311 mutex_lock(&fi->inmem_lock);
312 err = __commit_inmem_pages(inode, &revoke_list);
313 if (err) {
314 int ret;
315 /*
316 * try to revoke all committed pages, but still we could fail
317 * due to no memory or other reason, if that happened, EAGAIN
318 * will be returned, which means in such case, transaction is
319 * already not integrity, caller should use journal to do the
320 * recovery or rewrite & commit last transaction. For other
321 * error number, revoking was done by filesystem itself.
322 */
323 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
324 if (ret)
325 err = ret;
326
327 /* drop all uncommitted pages */
328 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
329 }
330 mutex_unlock(&fi->inmem_lock);
331
332 f2fs_unlock_op(sbi);
333 return err;
334}
335
336/*
337 * This function balances dirty node and dentry pages.
338 * In addition, it controls garbage collection.
339 */
340void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
341{
342 if (!need)
343 return;
344 /*
345 * We should do GC or end up with checkpoint, if there are so many dirty
346 * dir/node pages without enough free segments.
347 */
348 if (has_not_enough_free_secs(sbi, 0)) {
349 mutex_lock(&sbi->gc_mutex);
350 f2fs_gc(sbi, false);
351 }
352}
353
354void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
355{
356 /* try to shrink extent cache when there is no enough memory */
357 if (!available_free_memory(sbi, EXTENT_CACHE))
358 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
359
360 /* check the # of cached NAT entries */
361 if (!available_free_memory(sbi, NAT_ENTRIES))
362 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
363
364 if (!available_free_memory(sbi, FREE_NIDS))
365 try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES);
366
367 /* checkpoint is the only way to shrink partial cached entries */
368 if (!available_free_memory(sbi, NAT_ENTRIES) ||
369 !available_free_memory(sbi, INO_ENTRIES) ||
370 excess_prefree_segs(sbi) ||
371 excess_dirty_nats(sbi) ||
372 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
373 if (test_opt(sbi, DATA_FLUSH)) {
374 struct blk_plug plug;
375
376 blk_start_plug(&plug);
377 sync_dirty_inodes(sbi, FILE_INODE);
378 blk_finish_plug(&plug);
379 }
380 f2fs_sync_fs(sbi->sb, true);
381 stat_inc_bg_cp_count(sbi->stat_info);
382 }
383}
384
385static int issue_flush_thread(void *data)
386{
387 struct f2fs_sb_info *sbi = data;
388 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
389 wait_queue_head_t *q = &fcc->flush_wait_queue;
390repeat:
391 if (kthread_should_stop())
392 return 0;
393
394 if (!llist_empty(&fcc->issue_list)) {
395 struct bio *bio;
396 struct flush_cmd *cmd, *next;
397 int ret;
398
399 bio = f2fs_bio_alloc(0);
400
401 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
402 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
403
404 bio->bi_bdev = sbi->sb->s_bdev;
405 ret = submit_bio_wait(WRITE_FLUSH, bio);
406
407 llist_for_each_entry_safe(cmd, next,
408 fcc->dispatch_list, llnode) {
409 cmd->ret = ret;
410 complete(&cmd->wait);
411 }
412 bio_put(bio);
413 fcc->dispatch_list = NULL;
414 }
415
416 wait_event_interruptible(*q,
417 kthread_should_stop() || !llist_empty(&fcc->issue_list));
418 goto repeat;
419}
420
421int f2fs_issue_flush(struct f2fs_sb_info *sbi)
422{
423 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
424 struct flush_cmd cmd;
425
426 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
427 test_opt(sbi, FLUSH_MERGE));
428
429 if (test_opt(sbi, NOBARRIER))
430 return 0;
431
432 if (!test_opt(sbi, FLUSH_MERGE)) {
433 struct bio *bio = f2fs_bio_alloc(0);
434 int ret;
435
436 bio->bi_bdev = sbi->sb->s_bdev;
437 ret = submit_bio_wait(WRITE_FLUSH, bio);
438 bio_put(bio);
439 return ret;
440 }
441
442 init_completion(&cmd.wait);
443
444 llist_add(&cmd.llnode, &fcc->issue_list);
445
446 if (!fcc->dispatch_list)
447 wake_up(&fcc->flush_wait_queue);
448
449 wait_for_completion(&cmd.wait);
450
451 return cmd.ret;
452}
453
454int create_flush_cmd_control(struct f2fs_sb_info *sbi)
455{
456 dev_t dev = sbi->sb->s_bdev->bd_dev;
457 struct flush_cmd_control *fcc;
458 int err = 0;
459
460 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
461 if (!fcc)
462 return -ENOMEM;
463 init_waitqueue_head(&fcc->flush_wait_queue);
464 init_llist_head(&fcc->issue_list);
465 SM_I(sbi)->cmd_control_info = fcc;
466 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
467 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
468 if (IS_ERR(fcc->f2fs_issue_flush)) {
469 err = PTR_ERR(fcc->f2fs_issue_flush);
470 kfree(fcc);
471 SM_I(sbi)->cmd_control_info = NULL;
472 return err;
473 }
474
475 return err;
476}
477
478void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
479{
480 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
481
482 if (fcc && fcc->f2fs_issue_flush)
483 kthread_stop(fcc->f2fs_issue_flush);
484 kfree(fcc);
485 SM_I(sbi)->cmd_control_info = NULL;
486}
487
488static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
489 enum dirty_type dirty_type)
490{
491 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
492
493 /* need not be added */
494 if (IS_CURSEG(sbi, segno))
495 return;
496
497 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
498 dirty_i->nr_dirty[dirty_type]++;
499
500 if (dirty_type == DIRTY) {
501 struct seg_entry *sentry = get_seg_entry(sbi, segno);
502 enum dirty_type t = sentry->type;
503
504 if (unlikely(t >= DIRTY)) {
505 f2fs_bug_on(sbi, 1);
506 return;
507 }
508 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
509 dirty_i->nr_dirty[t]++;
510 }
511}
512
513static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
514 enum dirty_type dirty_type)
515{
516 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
517
518 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
519 dirty_i->nr_dirty[dirty_type]--;
520
521 if (dirty_type == DIRTY) {
522 struct seg_entry *sentry = get_seg_entry(sbi, segno);
523 enum dirty_type t = sentry->type;
524
525 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
526 dirty_i->nr_dirty[t]--;
527
528 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
529 clear_bit(GET_SECNO(sbi, segno),
530 dirty_i->victim_secmap);
531 }
532}
533
534/*
535 * Should not occur error such as -ENOMEM.
536 * Adding dirty entry into seglist is not critical operation.
537 * If a given segment is one of current working segments, it won't be added.
538 */
539static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
540{
541 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
542 unsigned short valid_blocks;
543
544 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
545 return;
546
547 mutex_lock(&dirty_i->seglist_lock);
548
549 valid_blocks = get_valid_blocks(sbi, segno, 0);
550
551 if (valid_blocks == 0) {
552 __locate_dirty_segment(sbi, segno, PRE);
553 __remove_dirty_segment(sbi, segno, DIRTY);
554 } else if (valid_blocks < sbi->blocks_per_seg) {
555 __locate_dirty_segment(sbi, segno, DIRTY);
556 } else {
557 /* Recovery routine with SSR needs this */
558 __remove_dirty_segment(sbi, segno, DIRTY);
559 }
560
561 mutex_unlock(&dirty_i->seglist_lock);
562}
563
564static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
565 block_t blkstart, block_t blklen)
566{
567 sector_t start = SECTOR_FROM_BLOCK(blkstart);
568 sector_t len = SECTOR_FROM_BLOCK(blklen);
569 struct seg_entry *se;
570 unsigned int offset;
571 block_t i;
572
573 for (i = blkstart; i < blkstart + blklen; i++) {
574 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
575 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
576
577 if (!f2fs_test_and_set_bit(offset, se->discard_map))
578 sbi->discard_blks--;
579 }
580 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
581 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
582}
583
584bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
585{
586 int err = -EOPNOTSUPP;
587
588 if (test_opt(sbi, DISCARD)) {
589 struct seg_entry *se = get_seg_entry(sbi,
590 GET_SEGNO(sbi, blkaddr));
591 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
592
593 if (f2fs_test_bit(offset, se->discard_map))
594 return false;
595
596 err = f2fs_issue_discard(sbi, blkaddr, 1);
597 }
598
599 if (err) {
600 update_meta_page(sbi, NULL, blkaddr);
601 return true;
602 }
603 return false;
604}
605
606static void __add_discard_entry(struct f2fs_sb_info *sbi,
607 struct cp_control *cpc, struct seg_entry *se,
608 unsigned int start, unsigned int end)
609{
610 struct list_head *head = &SM_I(sbi)->discard_list;
611 struct discard_entry *new, *last;
612
613 if (!list_empty(head)) {
614 last = list_last_entry(head, struct discard_entry, list);
615 if (START_BLOCK(sbi, cpc->trim_start) + start ==
616 last->blkaddr + last->len) {
617 last->len += end - start;
618 goto done;
619 }
620 }
621
622 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
623 INIT_LIST_HEAD(&new->list);
624 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
625 new->len = end - start;
626 list_add_tail(&new->list, head);
627done:
628 SM_I(sbi)->nr_discards += end - start;
629}
630
631static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
632{
633 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
634 int max_blocks = sbi->blocks_per_seg;
635 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
636 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
637 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
638 unsigned long *discard_map = (unsigned long *)se->discard_map;
639 unsigned long *dmap = SIT_I(sbi)->tmp_map;
640 unsigned int start = 0, end = -1;
641 bool force = (cpc->reason == CP_DISCARD);
642 int i;
643
644 if (se->valid_blocks == max_blocks)
645 return;
646
647 if (!force) {
648 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
649 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
650 return;
651 }
652
653 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
654 for (i = 0; i < entries; i++)
655 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
656 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
657
658 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
659 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
660 if (start >= max_blocks)
661 break;
662
663 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
664 __add_discard_entry(sbi, cpc, se, start, end);
665 }
666}
667
668void release_discard_addrs(struct f2fs_sb_info *sbi)
669{
670 struct list_head *head = &(SM_I(sbi)->discard_list);
671 struct discard_entry *entry, *this;
672
673 /* drop caches */
674 list_for_each_entry_safe(entry, this, head, list) {
675 list_del(&entry->list);
676 kmem_cache_free(discard_entry_slab, entry);
677 }
678}
679
680/*
681 * Should call clear_prefree_segments after checkpoint is done.
682 */
683static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
684{
685 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
686 unsigned int segno;
687
688 mutex_lock(&dirty_i->seglist_lock);
689 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
690 __set_test_and_free(sbi, segno);
691 mutex_unlock(&dirty_i->seglist_lock);
692}
693
694void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
695{
696 struct list_head *head = &(SM_I(sbi)->discard_list);
697 struct discard_entry *entry, *this;
698 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
699 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
700 unsigned int start = 0, end = -1;
701
702 mutex_lock(&dirty_i->seglist_lock);
703
704 while (1) {
705 int i;
706 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
707 if (start >= MAIN_SEGS(sbi))
708 break;
709 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
710 start + 1);
711
712 for (i = start; i < end; i++)
713 clear_bit(i, prefree_map);
714
715 dirty_i->nr_dirty[PRE] -= end - start;
716
717 if (!test_opt(sbi, DISCARD))
718 continue;
719
720 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
721 (end - start) << sbi->log_blocks_per_seg);
722 }
723 mutex_unlock(&dirty_i->seglist_lock);
724
725 /* send small discards */
726 list_for_each_entry_safe(entry, this, head, list) {
727 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
728 goto skip;
729 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
730 cpc->trimmed += entry->len;
731skip:
732 list_del(&entry->list);
733 SM_I(sbi)->nr_discards -= entry->len;
734 kmem_cache_free(discard_entry_slab, entry);
735 }
736}
737
738static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
739{
740 struct sit_info *sit_i = SIT_I(sbi);
741
742 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
743 sit_i->dirty_sentries++;
744 return false;
745 }
746
747 return true;
748}
749
750static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
751 unsigned int segno, int modified)
752{
753 struct seg_entry *se = get_seg_entry(sbi, segno);
754 se->type = type;
755 if (modified)
756 __mark_sit_entry_dirty(sbi, segno);
757}
758
759static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
760{
761 struct seg_entry *se;
762 unsigned int segno, offset;
763 long int new_vblocks;
764
765 segno = GET_SEGNO(sbi, blkaddr);
766
767 se = get_seg_entry(sbi, segno);
768 new_vblocks = se->valid_blocks + del;
769 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
770
771 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
772 (new_vblocks > sbi->blocks_per_seg)));
773
774 se->valid_blocks = new_vblocks;
775 se->mtime = get_mtime(sbi);
776 SIT_I(sbi)->max_mtime = se->mtime;
777
778 /* Update valid block bitmap */
779 if (del > 0) {
780 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
781 f2fs_bug_on(sbi, 1);
782 if (!f2fs_test_and_set_bit(offset, se->discard_map))
783 sbi->discard_blks--;
784 } else {
785 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
786 f2fs_bug_on(sbi, 1);
787 if (f2fs_test_and_clear_bit(offset, se->discard_map))
788 sbi->discard_blks++;
789 }
790 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
791 se->ckpt_valid_blocks += del;
792
793 __mark_sit_entry_dirty(sbi, segno);
794
795 /* update total number of valid blocks to be written in ckpt area */
796 SIT_I(sbi)->written_valid_blocks += del;
797
798 if (sbi->segs_per_sec > 1)
799 get_sec_entry(sbi, segno)->valid_blocks += del;
800}
801
802void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
803{
804 update_sit_entry(sbi, new, 1);
805 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
806 update_sit_entry(sbi, old, -1);
807
808 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
809 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
810}
811
812void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
813{
814 unsigned int segno = GET_SEGNO(sbi, addr);
815 struct sit_info *sit_i = SIT_I(sbi);
816
817 f2fs_bug_on(sbi, addr == NULL_ADDR);
818 if (addr == NEW_ADDR)
819 return;
820
821 /* add it into sit main buffer */
822 mutex_lock(&sit_i->sentry_lock);
823
824 update_sit_entry(sbi, addr, -1);
825
826 /* add it into dirty seglist */
827 locate_dirty_segment(sbi, segno);
828
829 mutex_unlock(&sit_i->sentry_lock);
830}
831
832bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
833{
834 struct sit_info *sit_i = SIT_I(sbi);
835 unsigned int segno, offset;
836 struct seg_entry *se;
837 bool is_cp = false;
838
839 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
840 return true;
841
842 mutex_lock(&sit_i->sentry_lock);
843
844 segno = GET_SEGNO(sbi, blkaddr);
845 se = get_seg_entry(sbi, segno);
846 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
847
848 if (f2fs_test_bit(offset, se->ckpt_valid_map))
849 is_cp = true;
850
851 mutex_unlock(&sit_i->sentry_lock);
852
853 return is_cp;
854}
855
856/*
857 * This function should be resided under the curseg_mutex lock
858 */
859static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
860 struct f2fs_summary *sum)
861{
862 struct curseg_info *curseg = CURSEG_I(sbi, type);
863 void *addr = curseg->sum_blk;
864 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
865 memcpy(addr, sum, sizeof(struct f2fs_summary));
866}
867
868/*
869 * Calculate the number of current summary pages for writing
870 */
871int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
872{
873 int valid_sum_count = 0;
874 int i, sum_in_page;
875
876 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
877 if (sbi->ckpt->alloc_type[i] == SSR)
878 valid_sum_count += sbi->blocks_per_seg;
879 else {
880 if (for_ra)
881 valid_sum_count += le16_to_cpu(
882 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
883 else
884 valid_sum_count += curseg_blkoff(sbi, i);
885 }
886 }
887
888 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
889 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
890 if (valid_sum_count <= sum_in_page)
891 return 1;
892 else if ((valid_sum_count - sum_in_page) <=
893 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
894 return 2;
895 return 3;
896}
897
898/*
899 * Caller should put this summary page
900 */
901struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
902{
903 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
904}
905
906void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
907{
908 struct page *page = grab_meta_page(sbi, blk_addr);
909 void *dst = page_address(page);
910
911 if (src)
912 memcpy(dst, src, PAGE_SIZE);
913 else
914 memset(dst, 0, PAGE_SIZE);
915 set_page_dirty(page);
916 f2fs_put_page(page, 1);
917}
918
919static void write_sum_page(struct f2fs_sb_info *sbi,
920 struct f2fs_summary_block *sum_blk, block_t blk_addr)
921{
922 update_meta_page(sbi, (void *)sum_blk, blk_addr);
923}
924
925static void write_current_sum_page(struct f2fs_sb_info *sbi,
926 int type, block_t blk_addr)
927{
928 struct curseg_info *curseg = CURSEG_I(sbi, type);
929 struct page *page = grab_meta_page(sbi, blk_addr);
930 struct f2fs_summary_block *src = curseg->sum_blk;
931 struct f2fs_summary_block *dst;
932
933 dst = (struct f2fs_summary_block *)page_address(page);
934
935 mutex_lock(&curseg->curseg_mutex);
936
937 down_read(&curseg->journal_rwsem);
938 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
939 up_read(&curseg->journal_rwsem);
940
941 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
942 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
943
944 mutex_unlock(&curseg->curseg_mutex);
945
946 set_page_dirty(page);
947 f2fs_put_page(page, 1);
948}
949
950static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
951{
952 struct curseg_info *curseg = CURSEG_I(sbi, type);
953 unsigned int segno = curseg->segno + 1;
954 struct free_segmap_info *free_i = FREE_I(sbi);
955
956 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
957 return !test_bit(segno, free_i->free_segmap);
958 return 0;
959}
960
961/*
962 * Find a new segment from the free segments bitmap to right order
963 * This function should be returned with success, otherwise BUG
964 */
965static void get_new_segment(struct f2fs_sb_info *sbi,
966 unsigned int *newseg, bool new_sec, int dir)
967{
968 struct free_segmap_info *free_i = FREE_I(sbi);
969 unsigned int segno, secno, zoneno;
970 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
971 unsigned int hint = *newseg / sbi->segs_per_sec;
972 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
973 unsigned int left_start = hint;
974 bool init = true;
975 int go_left = 0;
976 int i;
977
978 spin_lock(&free_i->segmap_lock);
979
980 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
981 segno = find_next_zero_bit(free_i->free_segmap,
982 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
983 if (segno < (hint + 1) * sbi->segs_per_sec)
984 goto got_it;
985 }
986find_other_zone:
987 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
988 if (secno >= MAIN_SECS(sbi)) {
989 if (dir == ALLOC_RIGHT) {
990 secno = find_next_zero_bit(free_i->free_secmap,
991 MAIN_SECS(sbi), 0);
992 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
993 } else {
994 go_left = 1;
995 left_start = hint - 1;
996 }
997 }
998 if (go_left == 0)
999 goto skip_left;
1000
1001 while (test_bit(left_start, free_i->free_secmap)) {
1002 if (left_start > 0) {
1003 left_start--;
1004 continue;
1005 }
1006 left_start = find_next_zero_bit(free_i->free_secmap,
1007 MAIN_SECS(sbi), 0);
1008 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1009 break;
1010 }
1011 secno = left_start;
1012skip_left:
1013 hint = secno;
1014 segno = secno * sbi->segs_per_sec;
1015 zoneno = secno / sbi->secs_per_zone;
1016
1017 /* give up on finding another zone */
1018 if (!init)
1019 goto got_it;
1020 if (sbi->secs_per_zone == 1)
1021 goto got_it;
1022 if (zoneno == old_zoneno)
1023 goto got_it;
1024 if (dir == ALLOC_LEFT) {
1025 if (!go_left && zoneno + 1 >= total_zones)
1026 goto got_it;
1027 if (go_left && zoneno == 0)
1028 goto got_it;
1029 }
1030 for (i = 0; i < NR_CURSEG_TYPE; i++)
1031 if (CURSEG_I(sbi, i)->zone == zoneno)
1032 break;
1033
1034 if (i < NR_CURSEG_TYPE) {
1035 /* zone is in user, try another */
1036 if (go_left)
1037 hint = zoneno * sbi->secs_per_zone - 1;
1038 else if (zoneno + 1 >= total_zones)
1039 hint = 0;
1040 else
1041 hint = (zoneno + 1) * sbi->secs_per_zone;
1042 init = false;
1043 goto find_other_zone;
1044 }
1045got_it:
1046 /* set it as dirty segment in free segmap */
1047 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1048 __set_inuse(sbi, segno);
1049 *newseg = segno;
1050 spin_unlock(&free_i->segmap_lock);
1051}
1052
1053static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1054{
1055 struct curseg_info *curseg = CURSEG_I(sbi, type);
1056 struct summary_footer *sum_footer;
1057
1058 curseg->segno = curseg->next_segno;
1059 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1060 curseg->next_blkoff = 0;
1061 curseg->next_segno = NULL_SEGNO;
1062
1063 sum_footer = &(curseg->sum_blk->footer);
1064 memset(sum_footer, 0, sizeof(struct summary_footer));
1065 if (IS_DATASEG(type))
1066 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1067 if (IS_NODESEG(type))
1068 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1069 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1070}
1071
1072/*
1073 * Allocate a current working segment.
1074 * This function always allocates a free segment in LFS manner.
1075 */
1076static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1077{
1078 struct curseg_info *curseg = CURSEG_I(sbi, type);
1079 unsigned int segno = curseg->segno;
1080 int dir = ALLOC_LEFT;
1081
1082 write_sum_page(sbi, curseg->sum_blk,
1083 GET_SUM_BLOCK(sbi, segno));
1084 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1085 dir = ALLOC_RIGHT;
1086
1087 if (test_opt(sbi, NOHEAP))
1088 dir = ALLOC_RIGHT;
1089
1090 get_new_segment(sbi, &segno, new_sec, dir);
1091 curseg->next_segno = segno;
1092 reset_curseg(sbi, type, 1);
1093 curseg->alloc_type = LFS;
1094}
1095
1096static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1097 struct curseg_info *seg, block_t start)
1098{
1099 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1100 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1101 unsigned long *target_map = SIT_I(sbi)->tmp_map;
1102 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1103 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1104 int i, pos;
1105
1106 for (i = 0; i < entries; i++)
1107 target_map[i] = ckpt_map[i] | cur_map[i];
1108
1109 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1110
1111 seg->next_blkoff = pos;
1112}
1113
1114/*
1115 * If a segment is written by LFS manner, next block offset is just obtained
1116 * by increasing the current block offset. However, if a segment is written by
1117 * SSR manner, next block offset obtained by calling __next_free_blkoff
1118 */
1119static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1120 struct curseg_info *seg)
1121{
1122 if (seg->alloc_type == SSR)
1123 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1124 else
1125 seg->next_blkoff++;
1126}
1127
1128/*
1129 * This function always allocates a used segment(from dirty seglist) by SSR
1130 * manner, so it should recover the existing segment information of valid blocks
1131 */
1132static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1133{
1134 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1135 struct curseg_info *curseg = CURSEG_I(sbi, type);
1136 unsigned int new_segno = curseg->next_segno;
1137 struct f2fs_summary_block *sum_node;
1138 struct page *sum_page;
1139
1140 write_sum_page(sbi, curseg->sum_blk,
1141 GET_SUM_BLOCK(sbi, curseg->segno));
1142 __set_test_and_inuse(sbi, new_segno);
1143
1144 mutex_lock(&dirty_i->seglist_lock);
1145 __remove_dirty_segment(sbi, new_segno, PRE);
1146 __remove_dirty_segment(sbi, new_segno, DIRTY);
1147 mutex_unlock(&dirty_i->seglist_lock);
1148
1149 reset_curseg(sbi, type, 1);
1150 curseg->alloc_type = SSR;
1151 __next_free_blkoff(sbi, curseg, 0);
1152
1153 if (reuse) {
1154 sum_page = get_sum_page(sbi, new_segno);
1155 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1156 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1157 f2fs_put_page(sum_page, 1);
1158 }
1159}
1160
1161static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1162{
1163 struct curseg_info *curseg = CURSEG_I(sbi, type);
1164 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1165
1166 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1167 return v_ops->get_victim(sbi,
1168 &(curseg)->next_segno, BG_GC, type, SSR);
1169
1170 /* For data segments, let's do SSR more intensively */
1171 for (; type >= CURSEG_HOT_DATA; type--)
1172 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1173 BG_GC, type, SSR))
1174 return 1;
1175 return 0;
1176}
1177
1178/*
1179 * flush out current segment and replace it with new segment
1180 * This function should be returned with success, otherwise BUG
1181 */
1182static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1183 int type, bool force)
1184{
1185 struct curseg_info *curseg = CURSEG_I(sbi, type);
1186
1187 if (force)
1188 new_curseg(sbi, type, true);
1189 else if (type == CURSEG_WARM_NODE)
1190 new_curseg(sbi, type, false);
1191 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1192 new_curseg(sbi, type, false);
1193 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1194 change_curseg(sbi, type, true);
1195 else
1196 new_curseg(sbi, type, false);
1197
1198 stat_inc_seg_type(sbi, curseg);
1199}
1200
1201static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1202{
1203 struct curseg_info *curseg = CURSEG_I(sbi, type);
1204 unsigned int old_segno;
1205
1206 old_segno = curseg->segno;
1207 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1208 locate_dirty_segment(sbi, old_segno);
1209}
1210
1211void allocate_new_segments(struct f2fs_sb_info *sbi)
1212{
1213 int i;
1214
1215 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1216 __allocate_new_segments(sbi, i);
1217}
1218
1219static const struct segment_allocation default_salloc_ops = {
1220 .allocate_segment = allocate_segment_by_default,
1221};
1222
1223int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1224{
1225 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1226 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1227 unsigned int start_segno, end_segno;
1228 struct cp_control cpc;
1229 int err = 0;
1230
1231 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1232 return -EINVAL;
1233
1234 cpc.trimmed = 0;
1235 if (end <= MAIN_BLKADDR(sbi))
1236 goto out;
1237
1238 /* start/end segment number in main_area */
1239 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1240 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1241 GET_SEGNO(sbi, end);
1242 cpc.reason = CP_DISCARD;
1243 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1244
1245 /* do checkpoint to issue discard commands safely */
1246 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1247 cpc.trim_start = start_segno;
1248
1249 if (sbi->discard_blks == 0)
1250 break;
1251 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1252 cpc.trim_end = end_segno;
1253 else
1254 cpc.trim_end = min_t(unsigned int,
1255 rounddown(start_segno +
1256 BATCHED_TRIM_SEGMENTS(sbi),
1257 sbi->segs_per_sec) - 1, end_segno);
1258
1259 mutex_lock(&sbi->gc_mutex);
1260 err = write_checkpoint(sbi, &cpc);
1261 mutex_unlock(&sbi->gc_mutex);
1262 }
1263out:
1264 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1265 return err;
1266}
1267
1268static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1269{
1270 struct curseg_info *curseg = CURSEG_I(sbi, type);
1271 if (curseg->next_blkoff < sbi->blocks_per_seg)
1272 return true;
1273 return false;
1274}
1275
1276static int __get_segment_type_2(struct page *page, enum page_type p_type)
1277{
1278 if (p_type == DATA)
1279 return CURSEG_HOT_DATA;
1280 else
1281 return CURSEG_HOT_NODE;
1282}
1283
1284static int __get_segment_type_4(struct page *page, enum page_type p_type)
1285{
1286 if (p_type == DATA) {
1287 struct inode *inode = page->mapping->host;
1288
1289 if (S_ISDIR(inode->i_mode))
1290 return CURSEG_HOT_DATA;
1291 else
1292 return CURSEG_COLD_DATA;
1293 } else {
1294 if (IS_DNODE(page) && is_cold_node(page))
1295 return CURSEG_WARM_NODE;
1296 else
1297 return CURSEG_COLD_NODE;
1298 }
1299}
1300
1301static int __get_segment_type_6(struct page *page, enum page_type p_type)
1302{
1303 if (p_type == DATA) {
1304 struct inode *inode = page->mapping->host;
1305
1306 if (S_ISDIR(inode->i_mode))
1307 return CURSEG_HOT_DATA;
1308 else if (is_cold_data(page) || file_is_cold(inode))
1309 return CURSEG_COLD_DATA;
1310 else
1311 return CURSEG_WARM_DATA;
1312 } else {
1313 if (IS_DNODE(page))
1314 return is_cold_node(page) ? CURSEG_WARM_NODE :
1315 CURSEG_HOT_NODE;
1316 else
1317 return CURSEG_COLD_NODE;
1318 }
1319}
1320
1321static int __get_segment_type(struct page *page, enum page_type p_type)
1322{
1323 switch (F2FS_P_SB(page)->active_logs) {
1324 case 2:
1325 return __get_segment_type_2(page, p_type);
1326 case 4:
1327 return __get_segment_type_4(page, p_type);
1328 }
1329 /* NR_CURSEG_TYPE(6) logs by default */
1330 f2fs_bug_on(F2FS_P_SB(page),
1331 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1332 return __get_segment_type_6(page, p_type);
1333}
1334
1335void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1336 block_t old_blkaddr, block_t *new_blkaddr,
1337 struct f2fs_summary *sum, int type)
1338{
1339 struct sit_info *sit_i = SIT_I(sbi);
1340 struct curseg_info *curseg;
1341 bool direct_io = (type == CURSEG_DIRECT_IO);
1342
1343 type = direct_io ? CURSEG_WARM_DATA : type;
1344
1345 curseg = CURSEG_I(sbi, type);
1346
1347 mutex_lock(&curseg->curseg_mutex);
1348 mutex_lock(&sit_i->sentry_lock);
1349
1350 /* direct_io'ed data is aligned to the segment for better performance */
1351 if (direct_io && curseg->next_blkoff &&
1352 !has_not_enough_free_secs(sbi, 0))
1353 __allocate_new_segments(sbi, type);
1354
1355 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1356
1357 /*
1358 * __add_sum_entry should be resided under the curseg_mutex
1359 * because, this function updates a summary entry in the
1360 * current summary block.
1361 */
1362 __add_sum_entry(sbi, type, sum);
1363
1364 __refresh_next_blkoff(sbi, curseg);
1365
1366 stat_inc_block_count(sbi, curseg);
1367
1368 if (!__has_curseg_space(sbi, type))
1369 sit_i->s_ops->allocate_segment(sbi, type, false);
1370 /*
1371 * SIT information should be updated before segment allocation,
1372 * since SSR needs latest valid block information.
1373 */
1374 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1375
1376 mutex_unlock(&sit_i->sentry_lock);
1377
1378 if (page && IS_NODESEG(type))
1379 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1380
1381 mutex_unlock(&curseg->curseg_mutex);
1382}
1383
1384static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1385{
1386 int type = __get_segment_type(fio->page, fio->type);
1387
1388 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1389 &fio->new_blkaddr, sum, type);
1390
1391 /* writeout dirty page into bdev */
1392 f2fs_submit_page_mbio(fio);
1393}
1394
1395void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1396{
1397 struct f2fs_io_info fio = {
1398 .sbi = sbi,
1399 .type = META,
1400 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1401 .old_blkaddr = page->index,
1402 .new_blkaddr = page->index,
1403 .page = page,
1404 .encrypted_page = NULL,
1405 };
1406
1407 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1408 fio.rw &= ~REQ_META;
1409
1410 set_page_writeback(page);
1411 f2fs_submit_page_mbio(&fio);
1412}
1413
1414void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1415{
1416 struct f2fs_summary sum;
1417
1418 set_summary(&sum, nid, 0, 0);
1419 do_write_page(&sum, fio);
1420}
1421
1422void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1423{
1424 struct f2fs_sb_info *sbi = fio->sbi;
1425 struct f2fs_summary sum;
1426 struct node_info ni;
1427
1428 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1429 get_node_info(sbi, dn->nid, &ni);
1430 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1431 do_write_page(&sum, fio);
1432 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1433}
1434
1435void rewrite_data_page(struct f2fs_io_info *fio)
1436{
1437 fio->new_blkaddr = fio->old_blkaddr;
1438 stat_inc_inplace_blocks(fio->sbi);
1439 f2fs_submit_page_mbio(fio);
1440}
1441
1442void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1443 block_t old_blkaddr, block_t new_blkaddr,
1444 bool recover_curseg, bool recover_newaddr)
1445{
1446 struct sit_info *sit_i = SIT_I(sbi);
1447 struct curseg_info *curseg;
1448 unsigned int segno, old_cursegno;
1449 struct seg_entry *se;
1450 int type;
1451 unsigned short old_blkoff;
1452
1453 segno = GET_SEGNO(sbi, new_blkaddr);
1454 se = get_seg_entry(sbi, segno);
1455 type = se->type;
1456
1457 if (!recover_curseg) {
1458 /* for recovery flow */
1459 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1460 if (old_blkaddr == NULL_ADDR)
1461 type = CURSEG_COLD_DATA;
1462 else
1463 type = CURSEG_WARM_DATA;
1464 }
1465 } else {
1466 if (!IS_CURSEG(sbi, segno))
1467 type = CURSEG_WARM_DATA;
1468 }
1469
1470 curseg = CURSEG_I(sbi, type);
1471
1472 mutex_lock(&curseg->curseg_mutex);
1473 mutex_lock(&sit_i->sentry_lock);
1474
1475 old_cursegno = curseg->segno;
1476 old_blkoff = curseg->next_blkoff;
1477
1478 /* change the current segment */
1479 if (segno != curseg->segno) {
1480 curseg->next_segno = segno;
1481 change_curseg(sbi, type, true);
1482 }
1483
1484 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1485 __add_sum_entry(sbi, type, sum);
1486
1487 if (!recover_curseg || recover_newaddr)
1488 update_sit_entry(sbi, new_blkaddr, 1);
1489 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1490 update_sit_entry(sbi, old_blkaddr, -1);
1491
1492 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1493 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1494
1495 locate_dirty_segment(sbi, old_cursegno);
1496
1497 if (recover_curseg) {
1498 if (old_cursegno != curseg->segno) {
1499 curseg->next_segno = old_cursegno;
1500 change_curseg(sbi, type, true);
1501 }
1502 curseg->next_blkoff = old_blkoff;
1503 }
1504
1505 mutex_unlock(&sit_i->sentry_lock);
1506 mutex_unlock(&curseg->curseg_mutex);
1507}
1508
1509void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1510 block_t old_addr, block_t new_addr,
1511 unsigned char version, bool recover_curseg,
1512 bool recover_newaddr)
1513{
1514 struct f2fs_summary sum;
1515
1516 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1517
1518 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1519 recover_curseg, recover_newaddr);
1520
1521 f2fs_update_data_blkaddr(dn, new_addr);
1522}
1523
1524void f2fs_wait_on_page_writeback(struct page *page,
1525 enum page_type type, bool ordered)
1526{
1527 if (PageWriteback(page)) {
1528 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1529
1530 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
1531 if (ordered)
1532 wait_on_page_writeback(page);
1533 else
1534 wait_for_stable_page(page);
1535 }
1536}
1537
1538void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1539 block_t blkaddr)
1540{
1541 struct page *cpage;
1542
1543 if (blkaddr == NEW_ADDR)
1544 return;
1545
1546 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1547
1548 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1549 if (cpage) {
1550 f2fs_wait_on_page_writeback(cpage, DATA, true);
1551 f2fs_put_page(cpage, 1);
1552 }
1553}
1554
1555static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1556{
1557 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1558 struct curseg_info *seg_i;
1559 unsigned char *kaddr;
1560 struct page *page;
1561 block_t start;
1562 int i, j, offset;
1563
1564 start = start_sum_block(sbi);
1565
1566 page = get_meta_page(sbi, start++);
1567 kaddr = (unsigned char *)page_address(page);
1568
1569 /* Step 1: restore nat cache */
1570 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1571 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
1572
1573 /* Step 2: restore sit cache */
1574 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1575 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
1576 offset = 2 * SUM_JOURNAL_SIZE;
1577
1578 /* Step 3: restore summary entries */
1579 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1580 unsigned short blk_off;
1581 unsigned int segno;
1582
1583 seg_i = CURSEG_I(sbi, i);
1584 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1585 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1586 seg_i->next_segno = segno;
1587 reset_curseg(sbi, i, 0);
1588 seg_i->alloc_type = ckpt->alloc_type[i];
1589 seg_i->next_blkoff = blk_off;
1590
1591 if (seg_i->alloc_type == SSR)
1592 blk_off = sbi->blocks_per_seg;
1593
1594 for (j = 0; j < blk_off; j++) {
1595 struct f2fs_summary *s;
1596 s = (struct f2fs_summary *)(kaddr + offset);
1597 seg_i->sum_blk->entries[j] = *s;
1598 offset += SUMMARY_SIZE;
1599 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
1600 SUM_FOOTER_SIZE)
1601 continue;
1602
1603 f2fs_put_page(page, 1);
1604 page = NULL;
1605
1606 page = get_meta_page(sbi, start++);
1607 kaddr = (unsigned char *)page_address(page);
1608 offset = 0;
1609 }
1610 }
1611 f2fs_put_page(page, 1);
1612 return 0;
1613}
1614
1615static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1616{
1617 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1618 struct f2fs_summary_block *sum;
1619 struct curseg_info *curseg;
1620 struct page *new;
1621 unsigned short blk_off;
1622 unsigned int segno = 0;
1623 block_t blk_addr = 0;
1624
1625 /* get segment number and block addr */
1626 if (IS_DATASEG(type)) {
1627 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1628 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1629 CURSEG_HOT_DATA]);
1630 if (__exist_node_summaries(sbi))
1631 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1632 else
1633 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1634 } else {
1635 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1636 CURSEG_HOT_NODE]);
1637 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1638 CURSEG_HOT_NODE]);
1639 if (__exist_node_summaries(sbi))
1640 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1641 type - CURSEG_HOT_NODE);
1642 else
1643 blk_addr = GET_SUM_BLOCK(sbi, segno);
1644 }
1645
1646 new = get_meta_page(sbi, blk_addr);
1647 sum = (struct f2fs_summary_block *)page_address(new);
1648
1649 if (IS_NODESEG(type)) {
1650 if (__exist_node_summaries(sbi)) {
1651 struct f2fs_summary *ns = &sum->entries[0];
1652 int i;
1653 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1654 ns->version = 0;
1655 ns->ofs_in_node = 0;
1656 }
1657 } else {
1658 int err;
1659
1660 err = restore_node_summary(sbi, segno, sum);
1661 if (err) {
1662 f2fs_put_page(new, 1);
1663 return err;
1664 }
1665 }
1666 }
1667
1668 /* set uncompleted segment to curseg */
1669 curseg = CURSEG_I(sbi, type);
1670 mutex_lock(&curseg->curseg_mutex);
1671
1672 /* update journal info */
1673 down_write(&curseg->journal_rwsem);
1674 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1675 up_write(&curseg->journal_rwsem);
1676
1677 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1678 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
1679 curseg->next_segno = segno;
1680 reset_curseg(sbi, type, 0);
1681 curseg->alloc_type = ckpt->alloc_type[type];
1682 curseg->next_blkoff = blk_off;
1683 mutex_unlock(&curseg->curseg_mutex);
1684 f2fs_put_page(new, 1);
1685 return 0;
1686}
1687
1688static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1689{
1690 int type = CURSEG_HOT_DATA;
1691 int err;
1692
1693 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1694 int npages = npages_for_summary_flush(sbi, true);
1695
1696 if (npages >= 2)
1697 ra_meta_pages(sbi, start_sum_block(sbi), npages,
1698 META_CP, true);
1699
1700 /* restore for compacted data summary */
1701 if (read_compacted_summaries(sbi))
1702 return -EINVAL;
1703 type = CURSEG_HOT_NODE;
1704 }
1705
1706 if (__exist_node_summaries(sbi))
1707 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1708 NR_CURSEG_TYPE - type, META_CP, true);
1709
1710 for (; type <= CURSEG_COLD_NODE; type++) {
1711 err = read_normal_summaries(sbi, type);
1712 if (err)
1713 return err;
1714 }
1715
1716 return 0;
1717}
1718
1719static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1720{
1721 struct page *page;
1722 unsigned char *kaddr;
1723 struct f2fs_summary *summary;
1724 struct curseg_info *seg_i;
1725 int written_size = 0;
1726 int i, j;
1727
1728 page = grab_meta_page(sbi, blkaddr++);
1729 kaddr = (unsigned char *)page_address(page);
1730
1731 /* Step 1: write nat cache */
1732 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1733 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
1734 written_size += SUM_JOURNAL_SIZE;
1735
1736 /* Step 2: write sit cache */
1737 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1738 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
1739 written_size += SUM_JOURNAL_SIZE;
1740
1741 /* Step 3: write summary entries */
1742 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1743 unsigned short blkoff;
1744 seg_i = CURSEG_I(sbi, i);
1745 if (sbi->ckpt->alloc_type[i] == SSR)
1746 blkoff = sbi->blocks_per_seg;
1747 else
1748 blkoff = curseg_blkoff(sbi, i);
1749
1750 for (j = 0; j < blkoff; j++) {
1751 if (!page) {
1752 page = grab_meta_page(sbi, blkaddr++);
1753 kaddr = (unsigned char *)page_address(page);
1754 written_size = 0;
1755 }
1756 summary = (struct f2fs_summary *)(kaddr + written_size);
1757 *summary = seg_i->sum_blk->entries[j];
1758 written_size += SUMMARY_SIZE;
1759
1760 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
1761 SUM_FOOTER_SIZE)
1762 continue;
1763
1764 set_page_dirty(page);
1765 f2fs_put_page(page, 1);
1766 page = NULL;
1767 }
1768 }
1769 if (page) {
1770 set_page_dirty(page);
1771 f2fs_put_page(page, 1);
1772 }
1773}
1774
1775static void write_normal_summaries(struct f2fs_sb_info *sbi,
1776 block_t blkaddr, int type)
1777{
1778 int i, end;
1779 if (IS_DATASEG(type))
1780 end = type + NR_CURSEG_DATA_TYPE;
1781 else
1782 end = type + NR_CURSEG_NODE_TYPE;
1783
1784 for (i = type; i < end; i++)
1785 write_current_sum_page(sbi, i, blkaddr + (i - type));
1786}
1787
1788void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1789{
1790 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1791 write_compacted_summaries(sbi, start_blk);
1792 else
1793 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1794}
1795
1796void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1797{
1798 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1799}
1800
1801int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
1802 unsigned int val, int alloc)
1803{
1804 int i;
1805
1806 if (type == NAT_JOURNAL) {
1807 for (i = 0; i < nats_in_cursum(journal); i++) {
1808 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
1809 return i;
1810 }
1811 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1812 return update_nats_in_cursum(journal, 1);
1813 } else if (type == SIT_JOURNAL) {
1814 for (i = 0; i < sits_in_cursum(journal); i++)
1815 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
1816 return i;
1817 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1818 return update_sits_in_cursum(journal, 1);
1819 }
1820 return -1;
1821}
1822
1823static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1824 unsigned int segno)
1825{
1826 return get_meta_page(sbi, current_sit_addr(sbi, segno));
1827}
1828
1829static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1830 unsigned int start)
1831{
1832 struct sit_info *sit_i = SIT_I(sbi);
1833 struct page *src_page, *dst_page;
1834 pgoff_t src_off, dst_off;
1835 void *src_addr, *dst_addr;
1836
1837 src_off = current_sit_addr(sbi, start);
1838 dst_off = next_sit_addr(sbi, src_off);
1839
1840 /* get current sit block page without lock */
1841 src_page = get_meta_page(sbi, src_off);
1842 dst_page = grab_meta_page(sbi, dst_off);
1843 f2fs_bug_on(sbi, PageDirty(src_page));
1844
1845 src_addr = page_address(src_page);
1846 dst_addr = page_address(dst_page);
1847 memcpy(dst_addr, src_addr, PAGE_SIZE);
1848
1849 set_page_dirty(dst_page);
1850 f2fs_put_page(src_page, 1);
1851
1852 set_to_next_sit(sit_i, start);
1853
1854 return dst_page;
1855}
1856
1857static struct sit_entry_set *grab_sit_entry_set(void)
1858{
1859 struct sit_entry_set *ses =
1860 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
1861
1862 ses->entry_cnt = 0;
1863 INIT_LIST_HEAD(&ses->set_list);
1864 return ses;
1865}
1866
1867static void release_sit_entry_set(struct sit_entry_set *ses)
1868{
1869 list_del(&ses->set_list);
1870 kmem_cache_free(sit_entry_set_slab, ses);
1871}
1872
1873static void adjust_sit_entry_set(struct sit_entry_set *ses,
1874 struct list_head *head)
1875{
1876 struct sit_entry_set *next = ses;
1877
1878 if (list_is_last(&ses->set_list, head))
1879 return;
1880
1881 list_for_each_entry_continue(next, head, set_list)
1882 if (ses->entry_cnt <= next->entry_cnt)
1883 break;
1884
1885 list_move_tail(&ses->set_list, &next->set_list);
1886}
1887
1888static void add_sit_entry(unsigned int segno, struct list_head *head)
1889{
1890 struct sit_entry_set *ses;
1891 unsigned int start_segno = START_SEGNO(segno);
1892
1893 list_for_each_entry(ses, head, set_list) {
1894 if (ses->start_segno == start_segno) {
1895 ses->entry_cnt++;
1896 adjust_sit_entry_set(ses, head);
1897 return;
1898 }
1899 }
1900
1901 ses = grab_sit_entry_set();
1902
1903 ses->start_segno = start_segno;
1904 ses->entry_cnt++;
1905 list_add(&ses->set_list, head);
1906}
1907
1908static void add_sits_in_set(struct f2fs_sb_info *sbi)
1909{
1910 struct f2fs_sm_info *sm_info = SM_I(sbi);
1911 struct list_head *set_list = &sm_info->sit_entry_set;
1912 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
1913 unsigned int segno;
1914
1915 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
1916 add_sit_entry(segno, set_list);
1917}
1918
1919static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
1920{
1921 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1922 struct f2fs_journal *journal = curseg->journal;
1923 int i;
1924
1925 down_write(&curseg->journal_rwsem);
1926 for (i = 0; i < sits_in_cursum(journal); i++) {
1927 unsigned int segno;
1928 bool dirtied;
1929
1930 segno = le32_to_cpu(segno_in_journal(journal, i));
1931 dirtied = __mark_sit_entry_dirty(sbi, segno);
1932
1933 if (!dirtied)
1934 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
1935 }
1936 update_sits_in_cursum(journal, -i);
1937 up_write(&curseg->journal_rwsem);
1938}
1939
1940/*
1941 * CP calls this function, which flushes SIT entries including sit_journal,
1942 * and moves prefree segs to free segs.
1943 */
1944void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1945{
1946 struct sit_info *sit_i = SIT_I(sbi);
1947 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1948 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1949 struct f2fs_journal *journal = curseg->journal;
1950 struct sit_entry_set *ses, *tmp;
1951 struct list_head *head = &SM_I(sbi)->sit_entry_set;
1952 bool to_journal = true;
1953 struct seg_entry *se;
1954
1955 mutex_lock(&sit_i->sentry_lock);
1956
1957 if (!sit_i->dirty_sentries)
1958 goto out;
1959
1960 /*
1961 * add and account sit entries of dirty bitmap in sit entry
1962 * set temporarily
1963 */
1964 add_sits_in_set(sbi);
1965
1966 /*
1967 * if there are no enough space in journal to store dirty sit
1968 * entries, remove all entries from journal and add and account
1969 * them in sit entry set.
1970 */
1971 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
1972 remove_sits_in_journal(sbi);
1973
1974 /*
1975 * there are two steps to flush sit entries:
1976 * #1, flush sit entries to journal in current cold data summary block.
1977 * #2, flush sit entries to sit page.
1978 */
1979 list_for_each_entry_safe(ses, tmp, head, set_list) {
1980 struct page *page = NULL;
1981 struct f2fs_sit_block *raw_sit = NULL;
1982 unsigned int start_segno = ses->start_segno;
1983 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
1984 (unsigned long)MAIN_SEGS(sbi));
1985 unsigned int segno = start_segno;
1986
1987 if (to_journal &&
1988 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
1989 to_journal = false;
1990
1991 if (to_journal) {
1992 down_write(&curseg->journal_rwsem);
1993 } else {
1994 page = get_next_sit_page(sbi, start_segno);
1995 raw_sit = page_address(page);
1996 }
1997
1998 /* flush dirty sit entries in region of current sit set */
1999 for_each_set_bit_from(segno, bitmap, end) {
2000 int offset, sit_offset;
2001
2002 se = get_seg_entry(sbi, segno);
2003
2004 /* add discard candidates */
2005 if (cpc->reason != CP_DISCARD) {
2006 cpc->trim_start = segno;
2007 add_discard_addrs(sbi, cpc);
2008 }
2009
2010 if (to_journal) {
2011 offset = lookup_journal_in_cursum(journal,
2012 SIT_JOURNAL, segno, 1);
2013 f2fs_bug_on(sbi, offset < 0);
2014 segno_in_journal(journal, offset) =
2015 cpu_to_le32(segno);
2016 seg_info_to_raw_sit(se,
2017 &sit_in_journal(journal, offset));
2018 } else {
2019 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2020 seg_info_to_raw_sit(se,
2021 &raw_sit->entries[sit_offset]);
2022 }
2023
2024 __clear_bit(segno, bitmap);
2025 sit_i->dirty_sentries--;
2026 ses->entry_cnt--;
2027 }
2028
2029 if (to_journal)
2030 up_write(&curseg->journal_rwsem);
2031 else
2032 f2fs_put_page(page, 1);
2033
2034 f2fs_bug_on(sbi, ses->entry_cnt);
2035 release_sit_entry_set(ses);
2036 }
2037
2038 f2fs_bug_on(sbi, !list_empty(head));
2039 f2fs_bug_on(sbi, sit_i->dirty_sentries);
2040out:
2041 if (cpc->reason == CP_DISCARD) {
2042 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2043 add_discard_addrs(sbi, cpc);
2044 }
2045 mutex_unlock(&sit_i->sentry_lock);
2046
2047 set_prefree_as_free_segments(sbi);
2048}
2049
2050static int build_sit_info(struct f2fs_sb_info *sbi)
2051{
2052 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2053 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2054 struct sit_info *sit_i;
2055 unsigned int sit_segs, start;
2056 char *src_bitmap, *dst_bitmap;
2057 unsigned int bitmap_size;
2058
2059 /* allocate memory for SIT information */
2060 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2061 if (!sit_i)
2062 return -ENOMEM;
2063
2064 SM_I(sbi)->sit_info = sit_i;
2065
2066 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2067 sizeof(struct seg_entry), GFP_KERNEL);
2068 if (!sit_i->sentries)
2069 return -ENOMEM;
2070
2071 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2072 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2073 if (!sit_i->dirty_sentries_bitmap)
2074 return -ENOMEM;
2075
2076 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2077 sit_i->sentries[start].cur_valid_map
2078 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2079 sit_i->sentries[start].ckpt_valid_map
2080 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2081 sit_i->sentries[start].discard_map
2082 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2083 if (!sit_i->sentries[start].cur_valid_map ||
2084 !sit_i->sentries[start].ckpt_valid_map ||
2085 !sit_i->sentries[start].discard_map)
2086 return -ENOMEM;
2087 }
2088
2089 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2090 if (!sit_i->tmp_map)
2091 return -ENOMEM;
2092
2093 if (sbi->segs_per_sec > 1) {
2094 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2095 sizeof(struct sec_entry), GFP_KERNEL);
2096 if (!sit_i->sec_entries)
2097 return -ENOMEM;
2098 }
2099
2100 /* get information related with SIT */
2101 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2102
2103 /* setup SIT bitmap from ckeckpoint pack */
2104 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2105 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2106
2107 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2108 if (!dst_bitmap)
2109 return -ENOMEM;
2110
2111 /* init SIT information */
2112 sit_i->s_ops = &default_salloc_ops;
2113
2114 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2115 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2116 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2117 sit_i->sit_bitmap = dst_bitmap;
2118 sit_i->bitmap_size = bitmap_size;
2119 sit_i->dirty_sentries = 0;
2120 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2121 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2122 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2123 mutex_init(&sit_i->sentry_lock);
2124 return 0;
2125}
2126
2127static int build_free_segmap(struct f2fs_sb_info *sbi)
2128{
2129 struct free_segmap_info *free_i;
2130 unsigned int bitmap_size, sec_bitmap_size;
2131
2132 /* allocate memory for free segmap information */
2133 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2134 if (!free_i)
2135 return -ENOMEM;
2136
2137 SM_I(sbi)->free_info = free_i;
2138
2139 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2140 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2141 if (!free_i->free_segmap)
2142 return -ENOMEM;
2143
2144 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2145 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2146 if (!free_i->free_secmap)
2147 return -ENOMEM;
2148
2149 /* set all segments as dirty temporarily */
2150 memset(free_i->free_segmap, 0xff, bitmap_size);
2151 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2152
2153 /* init free segmap information */
2154 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2155 free_i->free_segments = 0;
2156 free_i->free_sections = 0;
2157 spin_lock_init(&free_i->segmap_lock);
2158 return 0;
2159}
2160
2161static int build_curseg(struct f2fs_sb_info *sbi)
2162{
2163 struct curseg_info *array;
2164 int i;
2165
2166 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2167 if (!array)
2168 return -ENOMEM;
2169
2170 SM_I(sbi)->curseg_array = array;
2171
2172 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2173 mutex_init(&array[i].curseg_mutex);
2174 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2175 if (!array[i].sum_blk)
2176 return -ENOMEM;
2177 init_rwsem(&array[i].journal_rwsem);
2178 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2179 GFP_KERNEL);
2180 if (!array[i].journal)
2181 return -ENOMEM;
2182 array[i].segno = NULL_SEGNO;
2183 array[i].next_blkoff = 0;
2184 }
2185 return restore_curseg_summaries(sbi);
2186}
2187
2188static void build_sit_entries(struct f2fs_sb_info *sbi)
2189{
2190 struct sit_info *sit_i = SIT_I(sbi);
2191 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2192 struct f2fs_journal *journal = curseg->journal;
2193 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2194 unsigned int i, start, end;
2195 unsigned int readed, start_blk = 0;
2196 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
2197
2198 do {
2199 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
2200
2201 start = start_blk * sit_i->sents_per_block;
2202 end = (start_blk + readed) * sit_i->sents_per_block;
2203
2204 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2205 struct seg_entry *se = &sit_i->sentries[start];
2206 struct f2fs_sit_block *sit_blk;
2207 struct f2fs_sit_entry sit;
2208 struct page *page;
2209
2210 down_read(&curseg->journal_rwsem);
2211 for (i = 0; i < sits_in_cursum(journal); i++) {
2212 if (le32_to_cpu(segno_in_journal(journal, i))
2213 == start) {
2214 sit = sit_in_journal(journal, i);
2215 up_read(&curseg->journal_rwsem);
2216 goto got_it;
2217 }
2218 }
2219 up_read(&curseg->journal_rwsem);
2220
2221 page = get_current_sit_page(sbi, start);
2222 sit_blk = (struct f2fs_sit_block *)page_address(page);
2223 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2224 f2fs_put_page(page, 1);
2225got_it:
2226 check_block_count(sbi, start, &sit);
2227 seg_info_from_raw_sit(se, &sit);
2228
2229 /* build discard map only one time */
2230 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2231 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2232
2233 if (sbi->segs_per_sec > 1) {
2234 struct sec_entry *e = get_sec_entry(sbi, start);
2235 e->valid_blocks += se->valid_blocks;
2236 }
2237 }
2238 start_blk += readed;
2239 } while (start_blk < sit_blk_cnt);
2240}
2241
2242static void init_free_segmap(struct f2fs_sb_info *sbi)
2243{
2244 unsigned int start;
2245 int type;
2246
2247 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2248 struct seg_entry *sentry = get_seg_entry(sbi, start);
2249 if (!sentry->valid_blocks)
2250 __set_free(sbi, start);
2251 }
2252
2253 /* set use the current segments */
2254 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2255 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2256 __set_test_and_inuse(sbi, curseg_t->segno);
2257 }
2258}
2259
2260static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2261{
2262 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2263 struct free_segmap_info *free_i = FREE_I(sbi);
2264 unsigned int segno = 0, offset = 0;
2265 unsigned short valid_blocks;
2266
2267 while (1) {
2268 /* find dirty segment based on free segmap */
2269 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2270 if (segno >= MAIN_SEGS(sbi))
2271 break;
2272 offset = segno + 1;
2273 valid_blocks = get_valid_blocks(sbi, segno, 0);
2274 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2275 continue;
2276 if (valid_blocks > sbi->blocks_per_seg) {
2277 f2fs_bug_on(sbi, 1);
2278 continue;
2279 }
2280 mutex_lock(&dirty_i->seglist_lock);
2281 __locate_dirty_segment(sbi, segno, DIRTY);
2282 mutex_unlock(&dirty_i->seglist_lock);
2283 }
2284}
2285
2286static int init_victim_secmap(struct f2fs_sb_info *sbi)
2287{
2288 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2289 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2290
2291 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2292 if (!dirty_i->victim_secmap)
2293 return -ENOMEM;
2294 return 0;
2295}
2296
2297static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2298{
2299 struct dirty_seglist_info *dirty_i;
2300 unsigned int bitmap_size, i;
2301
2302 /* allocate memory for dirty segments list information */
2303 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2304 if (!dirty_i)
2305 return -ENOMEM;
2306
2307 SM_I(sbi)->dirty_info = dirty_i;
2308 mutex_init(&dirty_i->seglist_lock);
2309
2310 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2311
2312 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2313 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2314 if (!dirty_i->dirty_segmap[i])
2315 return -ENOMEM;
2316 }
2317
2318 init_dirty_segmap(sbi);
2319 return init_victim_secmap(sbi);
2320}
2321
2322/*
2323 * Update min, max modified time for cost-benefit GC algorithm
2324 */
2325static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2326{
2327 struct sit_info *sit_i = SIT_I(sbi);
2328 unsigned int segno;
2329
2330 mutex_lock(&sit_i->sentry_lock);
2331
2332 sit_i->min_mtime = LLONG_MAX;
2333
2334 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2335 unsigned int i;
2336 unsigned long long mtime = 0;
2337
2338 for (i = 0; i < sbi->segs_per_sec; i++)
2339 mtime += get_seg_entry(sbi, segno + i)->mtime;
2340
2341 mtime = div_u64(mtime, sbi->segs_per_sec);
2342
2343 if (sit_i->min_mtime > mtime)
2344 sit_i->min_mtime = mtime;
2345 }
2346 sit_i->max_mtime = get_mtime(sbi);
2347 mutex_unlock(&sit_i->sentry_lock);
2348}
2349
2350int build_segment_manager(struct f2fs_sb_info *sbi)
2351{
2352 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2353 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2354 struct f2fs_sm_info *sm_info;
2355 int err;
2356
2357 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2358 if (!sm_info)
2359 return -ENOMEM;
2360
2361 /* init sm info */
2362 sbi->sm_info = sm_info;
2363 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2364 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2365 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2366 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2367 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2368 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2369 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2370 sm_info->rec_prefree_segments = sm_info->main_segments *
2371 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2372 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2373 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2374 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2375
2376 INIT_LIST_HEAD(&sm_info->discard_list);
2377 sm_info->nr_discards = 0;
2378 sm_info->max_discards = 0;
2379
2380 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2381
2382 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2383
2384 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2385 err = create_flush_cmd_control(sbi);
2386 if (err)
2387 return err;
2388 }
2389
2390 err = build_sit_info(sbi);
2391 if (err)
2392 return err;
2393 err = build_free_segmap(sbi);
2394 if (err)
2395 return err;
2396 err = build_curseg(sbi);
2397 if (err)
2398 return err;
2399
2400 /* reinit free segmap based on SIT */
2401 build_sit_entries(sbi);
2402
2403 init_free_segmap(sbi);
2404 err = build_dirty_segmap(sbi);
2405 if (err)
2406 return err;
2407
2408 init_min_max_mtime(sbi);
2409 return 0;
2410}
2411
2412static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2413 enum dirty_type dirty_type)
2414{
2415 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2416
2417 mutex_lock(&dirty_i->seglist_lock);
2418 kvfree(dirty_i->dirty_segmap[dirty_type]);
2419 dirty_i->nr_dirty[dirty_type] = 0;
2420 mutex_unlock(&dirty_i->seglist_lock);
2421}
2422
2423static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2424{
2425 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2426 kvfree(dirty_i->victim_secmap);
2427}
2428
2429static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2430{
2431 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2432 int i;
2433
2434 if (!dirty_i)
2435 return;
2436
2437 /* discard pre-free/dirty segments list */
2438 for (i = 0; i < NR_DIRTY_TYPE; i++)
2439 discard_dirty_segmap(sbi, i);
2440
2441 destroy_victim_secmap(sbi);
2442 SM_I(sbi)->dirty_info = NULL;
2443 kfree(dirty_i);
2444}
2445
2446static void destroy_curseg(struct f2fs_sb_info *sbi)
2447{
2448 struct curseg_info *array = SM_I(sbi)->curseg_array;
2449 int i;
2450
2451 if (!array)
2452 return;
2453 SM_I(sbi)->curseg_array = NULL;
2454 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2455 kfree(array[i].sum_blk);
2456 kfree(array[i].journal);
2457 }
2458 kfree(array);
2459}
2460
2461static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2462{
2463 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2464 if (!free_i)
2465 return;
2466 SM_I(sbi)->free_info = NULL;
2467 kvfree(free_i->free_segmap);
2468 kvfree(free_i->free_secmap);
2469 kfree(free_i);
2470}
2471
2472static void destroy_sit_info(struct f2fs_sb_info *sbi)
2473{
2474 struct sit_info *sit_i = SIT_I(sbi);
2475 unsigned int start;
2476
2477 if (!sit_i)
2478 return;
2479
2480 if (sit_i->sentries) {
2481 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2482 kfree(sit_i->sentries[start].cur_valid_map);
2483 kfree(sit_i->sentries[start].ckpt_valid_map);
2484 kfree(sit_i->sentries[start].discard_map);
2485 }
2486 }
2487 kfree(sit_i->tmp_map);
2488
2489 kvfree(sit_i->sentries);
2490 kvfree(sit_i->sec_entries);
2491 kvfree(sit_i->dirty_sentries_bitmap);
2492
2493 SM_I(sbi)->sit_info = NULL;
2494 kfree(sit_i->sit_bitmap);
2495 kfree(sit_i);
2496}
2497
2498void destroy_segment_manager(struct f2fs_sb_info *sbi)
2499{
2500 struct f2fs_sm_info *sm_info = SM_I(sbi);
2501
2502 if (!sm_info)
2503 return;
2504 destroy_flush_cmd_control(sbi);
2505 destroy_dirty_segmap(sbi);
2506 destroy_curseg(sbi);
2507 destroy_free_segmap(sbi);
2508 destroy_sit_info(sbi);
2509 sbi->sm_info = NULL;
2510 kfree(sm_info);
2511}
2512
2513int __init create_segment_manager_caches(void)
2514{
2515 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2516 sizeof(struct discard_entry));
2517 if (!discard_entry_slab)
2518 goto fail;
2519
2520 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2521 sizeof(struct sit_entry_set));
2522 if (!sit_entry_set_slab)
2523 goto destory_discard_entry;
2524
2525 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2526 sizeof(struct inmem_pages));
2527 if (!inmem_entry_slab)
2528 goto destroy_sit_entry_set;
2529 return 0;
2530
2531destroy_sit_entry_set:
2532 kmem_cache_destroy(sit_entry_set_slab);
2533destory_discard_entry:
2534 kmem_cache_destroy(discard_entry_slab);
2535fail:
2536 return -ENOMEM;
2537}
2538
2539void destroy_segment_manager_caches(void)
2540{
2541 kmem_cache_destroy(sit_entry_set_slab);
2542 kmem_cache_destroy(discard_entry_slab);
2543 kmem_cache_destroy(inmem_entry_slab);
2544}