projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: use correct errno
[deliverable/linux.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 int commit_inmem_pages(struct inode *inode, bool abort)
195 {
196 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
197 struct f2fs_inode_info *fi = F2FS_I(inode);
198 struct inmem_pages *cur, *tmp;
199 bool submit_bio = false;
200 struct f2fs_io_info fio = {
201 .sbi = sbi,
202 .type = DATA,
203 .rw = WRITE_SYNC | REQ_PRIO,
204 .encrypted_page = NULL,
205 };
206 int err = 0;
207
208 /*
209 * The abort is true only when f2fs_evict_inode is called.
210 * Basically, the f2fs_evict_inode doesn't produce any data writes, so
211 * that we don't need to call f2fs_balance_fs.
212 * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
213 * inode becomes free by iget_locked in f2fs_iget.
214 */
215 if (!abort) {
216 f2fs_balance_fs(sbi, true);
217 f2fs_lock_op(sbi);
218 }
219
220 mutex_lock(&fi->inmem_lock);
221 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
222 lock_page(cur->page);
223 if (!abort) {
224 if (cur->page->mapping == inode->i_mapping) {
225 set_page_dirty(cur->page);
226 f2fs_wait_on_page_writeback(cur->page, DATA,
227 true);
228 if (clear_page_dirty_for_io(cur->page))
229 inode_dec_dirty_pages(inode);
230 trace_f2fs_commit_inmem_page(cur->page, INMEM);
231 fio.page = cur->page;
232 err = do_write_data_page(&fio);
233 if (err) {
234 unlock_page(cur->page);
235 break;
236 }
237 clear_cold_data(cur->page);
238 submit_bio = true;
239 }
240 } else {
241 ClearPageUptodate(cur->page);
242 trace_f2fs_commit_inmem_page(cur->page, INMEM_DROP);
243 }
244 set_page_private(cur->page, 0);
245 ClearPagePrivate(cur->page);
246 f2fs_put_page(cur->page, 1);
247
248 list_del(&cur->list);
249 kmem_cache_free(inmem_entry_slab, cur);
250 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
251 }
252 mutex_unlock(&fi->inmem_lock);
253
254 if (!abort) {
255 f2fs_unlock_op(sbi);
256 if (submit_bio)
257 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0,
258 DATA, WRITE);
259 }
260 return err;
261 }
262
263 /*
264 * This function balances dirty node and dentry pages.
265 * In addition, it controls garbage collection.
266 */
267 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
268 {
269 if (!need)
270 return;
271 /*
272 * We should do GC or end up with checkpoint, if there are so many dirty
273 * dir/node pages without enough free segments.
274 */
275 if (has_not_enough_free_secs(sbi, 0)) {
276 mutex_lock(&sbi->gc_mutex);
277 f2fs_gc(sbi, false);
278 }
279 }
280
281 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
282 {
283 /* try to shrink extent cache when there is no enough memory */
284 if (!available_free_memory(sbi, EXTENT_CACHE))
285 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
286
287 /* check the # of cached NAT entries */
288 if (!available_free_memory(sbi, NAT_ENTRIES))
289 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
290
291 if (!available_free_memory(sbi, FREE_NIDS))
292 try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES);
293
294 /* checkpoint is the only way to shrink partial cached entries */
295 if (!available_free_memory(sbi, NAT_ENTRIES) ||
296 !available_free_memory(sbi, INO_ENTRIES) ||
297 excess_prefree_segs(sbi) ||
298 excess_dirty_nats(sbi) ||
299 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
300 if (test_opt(sbi, DATA_FLUSH))
301 sync_dirty_inodes(sbi, FILE_INODE);
302 f2fs_sync_fs(sbi->sb, true);
303 stat_inc_bg_cp_count(sbi->stat_info);
304 }
305 }
306
307 static int issue_flush_thread(void *data)
308 {
309 struct f2fs_sb_info *sbi = data;
310 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
311 wait_queue_head_t *q = &fcc->flush_wait_queue;
312 repeat:
313 if (kthread_should_stop())
314 return 0;
315
316 if (!llist_empty(&fcc->issue_list)) {
317 struct bio *bio;
318 struct flush_cmd *cmd, *next;
319 int ret;
320
321 bio = f2fs_bio_alloc(0);
322
323 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
324 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
325
326 bio->bi_bdev = sbi->sb->s_bdev;
327 ret = submit_bio_wait(WRITE_FLUSH, bio);
328
329 llist_for_each_entry_safe(cmd, next,
330 fcc->dispatch_list, llnode) {
331 cmd->ret = ret;
332 complete(&cmd->wait);
333 }
334 bio_put(bio);
335 fcc->dispatch_list = NULL;
336 }
337
338 wait_event_interruptible(*q,
339 kthread_should_stop() || !llist_empty(&fcc->issue_list));
340 goto repeat;
341 }
342
343 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
344 {
345 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
346 struct flush_cmd cmd;
347
348 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
349 test_opt(sbi, FLUSH_MERGE));
350
351 if (test_opt(sbi, NOBARRIER))
352 return 0;
353
354 if (!test_opt(sbi, FLUSH_MERGE)) {
355 struct bio *bio = f2fs_bio_alloc(0);
356 int ret;
357
358 bio->bi_bdev = sbi->sb->s_bdev;
359 ret = submit_bio_wait(WRITE_FLUSH, bio);
360 bio_put(bio);
361 return ret;
362 }
363
364 init_completion(&cmd.wait);
365
366 llist_add(&cmd.llnode, &fcc->issue_list);
367
368 if (!fcc->dispatch_list)
369 wake_up(&fcc->flush_wait_queue);
370
371 wait_for_completion(&cmd.wait);
372
373 return cmd.ret;
374 }
375
376 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
377 {
378 dev_t dev = sbi->sb->s_bdev->bd_dev;
379 struct flush_cmd_control *fcc;
380 int err = 0;
381
382 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
383 if (!fcc)
384 return -ENOMEM;
385 init_waitqueue_head(&fcc->flush_wait_queue);
386 init_llist_head(&fcc->issue_list);
387 SM_I(sbi)->cmd_control_info = fcc;
388 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
389 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
390 if (IS_ERR(fcc->f2fs_issue_flush)) {
391 err = PTR_ERR(fcc->f2fs_issue_flush);
392 kfree(fcc);
393 SM_I(sbi)->cmd_control_info = NULL;
394 return err;
395 }
396
397 return err;
398 }
399
400 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
401 {
402 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
403
404 if (fcc && fcc->f2fs_issue_flush)
405 kthread_stop(fcc->f2fs_issue_flush);
406 kfree(fcc);
407 SM_I(sbi)->cmd_control_info = NULL;
408 }
409
410 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
411 enum dirty_type dirty_type)
412 {
413 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
414
415 /* need not be added */
416 if (IS_CURSEG(sbi, segno))
417 return;
418
419 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
420 dirty_i->nr_dirty[dirty_type]++;
421
422 if (dirty_type == DIRTY) {
423 struct seg_entry *sentry = get_seg_entry(sbi, segno);
424 enum dirty_type t = sentry->type;
425
426 if (unlikely(t >= DIRTY)) {
427 f2fs_bug_on(sbi, 1);
428 return;
429 }
430 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
431 dirty_i->nr_dirty[t]++;
432 }
433 }
434
435 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
436 enum dirty_type dirty_type)
437 {
438 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
439
440 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
441 dirty_i->nr_dirty[dirty_type]--;
442
443 if (dirty_type == DIRTY) {
444 struct seg_entry *sentry = get_seg_entry(sbi, segno);
445 enum dirty_type t = sentry->type;
446
447 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
448 dirty_i->nr_dirty[t]--;
449
450 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
451 clear_bit(GET_SECNO(sbi, segno),
452 dirty_i->victim_secmap);
453 }
454 }
455
456 /*
457 * Should not occur error such as -ENOMEM.
458 * Adding dirty entry into seglist is not critical operation.
459 * If a given segment is one of current working segments, it won't be added.
460 */
461 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
462 {
463 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
464 unsigned short valid_blocks;
465
466 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
467 return;
468
469 mutex_lock(&dirty_i->seglist_lock);
470
471 valid_blocks = get_valid_blocks(sbi, segno, 0);
472
473 if (valid_blocks == 0) {
474 __locate_dirty_segment(sbi, segno, PRE);
475 __remove_dirty_segment(sbi, segno, DIRTY);
476 } else if (valid_blocks < sbi->blocks_per_seg) {
477 __locate_dirty_segment(sbi, segno, DIRTY);
478 } else {
479 /* Recovery routine with SSR needs this */
480 __remove_dirty_segment(sbi, segno, DIRTY);
481 }
482
483 mutex_unlock(&dirty_i->seglist_lock);
484 }
485
486 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
487 block_t blkstart, block_t blklen)
488 {
489 sector_t start = SECTOR_FROM_BLOCK(blkstart);
490 sector_t len = SECTOR_FROM_BLOCK(blklen);
491 struct seg_entry *se;
492 unsigned int offset;
493 block_t i;
494
495 for (i = blkstart; i < blkstart + blklen; i++) {
496 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
497 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
498
499 if (!f2fs_test_and_set_bit(offset, se->discard_map))
500 sbi->discard_blks--;
501 }
502 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
503 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
504 }
505
506 bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
507 {
508 int err = -EOPNOTSUPP;
509
510 if (test_opt(sbi, DISCARD)) {
511 struct seg_entry *se = get_seg_entry(sbi,
512 GET_SEGNO(sbi, blkaddr));
513 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
514
515 if (f2fs_test_bit(offset, se->discard_map))
516 return false;
517
518 err = f2fs_issue_discard(sbi, blkaddr, 1);
519 }
520
521 if (err) {
522 update_meta_page(sbi, NULL, blkaddr);
523 return true;
524 }
525 return false;
526 }
527
528 static void __add_discard_entry(struct f2fs_sb_info *sbi,
529 struct cp_control *cpc, struct seg_entry *se,
530 unsigned int start, unsigned int end)
531 {
532 struct list_head *head = &SM_I(sbi)->discard_list;
533 struct discard_entry *new, *last;
534
535 if (!list_empty(head)) {
536 last = list_last_entry(head, struct discard_entry, list);
537 if (START_BLOCK(sbi, cpc->trim_start) + start ==
538 last->blkaddr + last->len) {
539 last->len += end - start;
540 goto done;
541 }
542 }
543
544 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
545 INIT_LIST_HEAD(&new->list);
546 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
547 new->len = end - start;
548 list_add_tail(&new->list, head);
549 done:
550 SM_I(sbi)->nr_discards += end - start;
551 }
552
553 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
554 {
555 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
556 int max_blocks = sbi->blocks_per_seg;
557 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
558 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
559 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
560 unsigned long *discard_map = (unsigned long *)se->discard_map;
561 unsigned long *dmap = SIT_I(sbi)->tmp_map;
562 unsigned int start = 0, end = -1;
563 bool force = (cpc->reason == CP_DISCARD);
564 int i;
565
566 if (se->valid_blocks == max_blocks)
567 return;
568
569 if (!force) {
570 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
571 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
572 return;
573 }
574
575 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
576 for (i = 0; i < entries; i++)
577 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
578 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
579
580 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
581 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
582 if (start >= max_blocks)
583 break;
584
585 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
586 __add_discard_entry(sbi, cpc, se, start, end);
587 }
588 }
589
590 void release_discard_addrs(struct f2fs_sb_info *sbi)
591 {
592 struct list_head *head = &(SM_I(sbi)->discard_list);
593 struct discard_entry *entry, *this;
594
595 /* drop caches */
596 list_for_each_entry_safe(entry, this, head, list) {
597 list_del(&entry->list);
598 kmem_cache_free(discard_entry_slab, entry);
599 }
600 }
601
602 /*
603 * Should call clear_prefree_segments after checkpoint is done.
604 */
605 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
606 {
607 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
608 unsigned int segno;
609
610 mutex_lock(&dirty_i->seglist_lock);
611 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
612 __set_test_and_free(sbi, segno);
613 mutex_unlock(&dirty_i->seglist_lock);
614 }
615
616 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
617 {
618 struct list_head *head = &(SM_I(sbi)->discard_list);
619 struct discard_entry *entry, *this;
620 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
621 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
622 unsigned int start = 0, end = -1;
623
624 mutex_lock(&dirty_i->seglist_lock);
625
626 while (1) {
627 int i;
628 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
629 if (start >= MAIN_SEGS(sbi))
630 break;
631 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
632 start + 1);
633
634 for (i = start; i < end; i++)
635 clear_bit(i, prefree_map);
636
637 dirty_i->nr_dirty[PRE] -= end - start;
638
639 if (!test_opt(sbi, DISCARD))
640 continue;
641
642 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
643 (end - start) << sbi->log_blocks_per_seg);
644 }
645 mutex_unlock(&dirty_i->seglist_lock);
646
647 /* send small discards */
648 list_for_each_entry_safe(entry, this, head, list) {
649 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
650 goto skip;
651 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
652 cpc->trimmed += entry->len;
653 skip:
654 list_del(&entry->list);
655 SM_I(sbi)->nr_discards -= entry->len;
656 kmem_cache_free(discard_entry_slab, entry);
657 }
658 }
659
660 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
661 {
662 struct sit_info *sit_i = SIT_I(sbi);
663
664 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
665 sit_i->dirty_sentries++;
666 return false;
667 }
668
669 return true;
670 }
671
672 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
673 unsigned int segno, int modified)
674 {
675 struct seg_entry *se = get_seg_entry(sbi, segno);
676 se->type = type;
677 if (modified)
678 __mark_sit_entry_dirty(sbi, segno);
679 }
680
681 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
682 {
683 struct seg_entry *se;
684 unsigned int segno, offset;
685 long int new_vblocks;
686
687 segno = GET_SEGNO(sbi, blkaddr);
688
689 se = get_seg_entry(sbi, segno);
690 new_vblocks = se->valid_blocks + del;
691 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
692
693 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
694 (new_vblocks > sbi->blocks_per_seg)));
695
696 se->valid_blocks = new_vblocks;
697 se->mtime = get_mtime(sbi);
698 SIT_I(sbi)->max_mtime = se->mtime;
699
700 /* Update valid block bitmap */
701 if (del > 0) {
702 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
703 f2fs_bug_on(sbi, 1);
704 if (!f2fs_test_and_set_bit(offset, se->discard_map))
705 sbi->discard_blks--;
706 } else {
707 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
708 f2fs_bug_on(sbi, 1);
709 if (f2fs_test_and_clear_bit(offset, se->discard_map))
710 sbi->discard_blks++;
711 }
712 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
713 se->ckpt_valid_blocks += del;
714
715 __mark_sit_entry_dirty(sbi, segno);
716
717 /* update total number of valid blocks to be written in ckpt area */
718 SIT_I(sbi)->written_valid_blocks += del;
719
720 if (sbi->segs_per_sec > 1)
721 get_sec_entry(sbi, segno)->valid_blocks += del;
722 }
723
724 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
725 {
726 update_sit_entry(sbi, new, 1);
727 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
728 update_sit_entry(sbi, old, -1);
729
730 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
731 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
732 }
733
734 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
735 {
736 unsigned int segno = GET_SEGNO(sbi, addr);
737 struct sit_info *sit_i = SIT_I(sbi);
738
739 f2fs_bug_on(sbi, addr == NULL_ADDR);
740 if (addr == NEW_ADDR)
741 return;
742
743 /* add it into sit main buffer */
744 mutex_lock(&sit_i->sentry_lock);
745
746 update_sit_entry(sbi, addr, -1);
747
748 /* add it into dirty seglist */
749 locate_dirty_segment(sbi, segno);
750
751 mutex_unlock(&sit_i->sentry_lock);
752 }
753
754 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
755 {
756 struct sit_info *sit_i = SIT_I(sbi);
757 unsigned int segno, offset;
758 struct seg_entry *se;
759 bool is_cp = false;
760
761 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
762 return true;
763
764 mutex_lock(&sit_i->sentry_lock);
765
766 segno = GET_SEGNO(sbi, blkaddr);
767 se = get_seg_entry(sbi, segno);
768 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
769
770 if (f2fs_test_bit(offset, se->ckpt_valid_map))
771 is_cp = true;
772
773 mutex_unlock(&sit_i->sentry_lock);
774
775 return is_cp;
776 }
777
778 /*
779 * This function should be resided under the curseg_mutex lock
780 */
781 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
782 struct f2fs_summary *sum)
783 {
784 struct curseg_info *curseg = CURSEG_I(sbi, type);
785 void *addr = curseg->sum_blk;
786 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
787 memcpy(addr, sum, sizeof(struct f2fs_summary));
788 }
789
790 /*
791 * Calculate the number of current summary pages for writing
792 */
793 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
794 {
795 int valid_sum_count = 0;
796 int i, sum_in_page;
797
798 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
799 if (sbi->ckpt->alloc_type[i] == SSR)
800 valid_sum_count += sbi->blocks_per_seg;
801 else {
802 if (for_ra)
803 valid_sum_count += le16_to_cpu(
804 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
805 else
806 valid_sum_count += curseg_blkoff(sbi, i);
807 }
808 }
809
810 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
811 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
812 if (valid_sum_count <= sum_in_page)
813 return 1;
814 else if ((valid_sum_count - sum_in_page) <=
815 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
816 return 2;
817 return 3;
818 }
819
820 /*
821 * Caller should put this summary page
822 */
823 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
824 {
825 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
826 }
827
828 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
829 {
830 struct page *page = grab_meta_page(sbi, blk_addr);
831 void *dst = page_address(page);
832
833 if (src)
834 memcpy(dst, src, PAGE_CACHE_SIZE);
835 else
836 memset(dst, 0, PAGE_CACHE_SIZE);
837 set_page_dirty(page);
838 f2fs_put_page(page, 1);
839 }
840
841 static void write_sum_page(struct f2fs_sb_info *sbi,
842 struct f2fs_summary_block *sum_blk, block_t blk_addr)
843 {
844 update_meta_page(sbi, (void *)sum_blk, blk_addr);
845 }
846
847 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
848 {
849 struct curseg_info *curseg = CURSEG_I(sbi, type);
850 unsigned int segno = curseg->segno + 1;
851 struct free_segmap_info *free_i = FREE_I(sbi);
852
853 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
854 return !test_bit(segno, free_i->free_segmap);
855 return 0;
856 }
857
858 /*
859 * Find a new segment from the free segments bitmap to right order
860 * This function should be returned with success, otherwise BUG
861 */
862 static void get_new_segment(struct f2fs_sb_info *sbi,
863 unsigned int *newseg, bool new_sec, int dir)
864 {
865 struct free_segmap_info *free_i = FREE_I(sbi);
866 unsigned int segno, secno, zoneno;
867 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
868 unsigned int hint = *newseg / sbi->segs_per_sec;
869 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
870 unsigned int left_start = hint;
871 bool init = true;
872 int go_left = 0;
873 int i;
874
875 spin_lock(&free_i->segmap_lock);
876
877 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
878 segno = find_next_zero_bit(free_i->free_segmap,
879 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
880 if (segno < (hint + 1) * sbi->segs_per_sec)
881 goto got_it;
882 }
883 find_other_zone:
884 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
885 if (secno >= MAIN_SECS(sbi)) {
886 if (dir == ALLOC_RIGHT) {
887 secno = find_next_zero_bit(free_i->free_secmap,
888 MAIN_SECS(sbi), 0);
889 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
890 } else {
891 go_left = 1;
892 left_start = hint - 1;
893 }
894 }
895 if (go_left == 0)
896 goto skip_left;
897
898 while (test_bit(left_start, free_i->free_secmap)) {
899 if (left_start > 0) {
900 left_start--;
901 continue;
902 }
903 left_start = find_next_zero_bit(free_i->free_secmap,
904 MAIN_SECS(sbi), 0);
905 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
906 break;
907 }
908 secno = left_start;
909 skip_left:
910 hint = secno;
911 segno = secno * sbi->segs_per_sec;
912 zoneno = secno / sbi->secs_per_zone;
913
914 /* give up on finding another zone */
915 if (!init)
916 goto got_it;
917 if (sbi->secs_per_zone == 1)
918 goto got_it;
919 if (zoneno == old_zoneno)
920 goto got_it;
921 if (dir == ALLOC_LEFT) {
922 if (!go_left && zoneno + 1 >= total_zones)
923 goto got_it;
924 if (go_left && zoneno == 0)
925 goto got_it;
926 }
927 for (i = 0; i < NR_CURSEG_TYPE; i++)
928 if (CURSEG_I(sbi, i)->zone == zoneno)
929 break;
930
931 if (i < NR_CURSEG_TYPE) {
932 /* zone is in user, try another */
933 if (go_left)
934 hint = zoneno * sbi->secs_per_zone - 1;
935 else if (zoneno + 1 >= total_zones)
936 hint = 0;
937 else
938 hint = (zoneno + 1) * sbi->secs_per_zone;
939 init = false;
940 goto find_other_zone;
941 }
942 got_it:
943 /* set it as dirty segment in free segmap */
944 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
945 __set_inuse(sbi, segno);
946 *newseg = segno;
947 spin_unlock(&free_i->segmap_lock);
948 }
949
950 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
951 {
952 struct curseg_info *curseg = CURSEG_I(sbi, type);
953 struct summary_footer *sum_footer;
954
955 curseg->segno = curseg->next_segno;
956 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
957 curseg->next_blkoff = 0;
958 curseg->next_segno = NULL_SEGNO;
959
960 sum_footer = &(curseg->sum_blk->footer);
961 memset(sum_footer, 0, sizeof(struct summary_footer));
962 if (IS_DATASEG(type))
963 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
964 if (IS_NODESEG(type))
965 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
966 __set_sit_entry_type(sbi, type, curseg->segno, modified);
967 }
968
969 /*
970 * Allocate a current working segment.
971 * This function always allocates a free segment in LFS manner.
972 */
973 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
974 {
975 struct curseg_info *curseg = CURSEG_I(sbi, type);
976 unsigned int segno = curseg->segno;
977 int dir = ALLOC_LEFT;
978
979 write_sum_page(sbi, curseg->sum_blk,
980 GET_SUM_BLOCK(sbi, segno));
981 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
982 dir = ALLOC_RIGHT;
983
984 if (test_opt(sbi, NOHEAP))
985 dir = ALLOC_RIGHT;
986
987 get_new_segment(sbi, &segno, new_sec, dir);
988 curseg->next_segno = segno;
989 reset_curseg(sbi, type, 1);
990 curseg->alloc_type = LFS;
991 }
992
993 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
994 struct curseg_info *seg, block_t start)
995 {
996 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
997 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
998 unsigned long *target_map = SIT_I(sbi)->tmp_map;
999 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1000 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1001 int i, pos;
1002
1003 for (i = 0; i < entries; i++)
1004 target_map[i] = ckpt_map[i] | cur_map[i];
1005
1006 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1007
1008 seg->next_blkoff = pos;
1009 }
1010
1011 /*
1012 * If a segment is written by LFS manner, next block offset is just obtained
1013 * by increasing the current block offset. However, if a segment is written by
1014 * SSR manner, next block offset obtained by calling __next_free_blkoff
1015 */
1016 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1017 struct curseg_info *seg)
1018 {
1019 if (seg->alloc_type == SSR)
1020 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1021 else
1022 seg->next_blkoff++;
1023 }
1024
1025 /*
1026 * This function always allocates a used segment(from dirty seglist) by SSR
1027 * manner, so it should recover the existing segment information of valid blocks
1028 */
1029 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1030 {
1031 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1032 struct curseg_info *curseg = CURSEG_I(sbi, type);
1033 unsigned int new_segno = curseg->next_segno;
1034 struct f2fs_summary_block *sum_node;
1035 struct page *sum_page;
1036
1037 write_sum_page(sbi, curseg->sum_blk,
1038 GET_SUM_BLOCK(sbi, curseg->segno));
1039 __set_test_and_inuse(sbi, new_segno);
1040
1041 mutex_lock(&dirty_i->seglist_lock);
1042 __remove_dirty_segment(sbi, new_segno, PRE);
1043 __remove_dirty_segment(sbi, new_segno, DIRTY);
1044 mutex_unlock(&dirty_i->seglist_lock);
1045
1046 reset_curseg(sbi, type, 1);
1047 curseg->alloc_type = SSR;
1048 __next_free_blkoff(sbi, curseg, 0);
1049
1050 if (reuse) {
1051 sum_page = get_sum_page(sbi, new_segno);
1052 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1053 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1054 f2fs_put_page(sum_page, 1);
1055 }
1056 }
1057
1058 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1059 {
1060 struct curseg_info *curseg = CURSEG_I(sbi, type);
1061 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1062
1063 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1064 return v_ops->get_victim(sbi,
1065 &(curseg)->next_segno, BG_GC, type, SSR);
1066
1067 /* For data segments, let's do SSR more intensively */
1068 for (; type >= CURSEG_HOT_DATA; type--)
1069 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1070 BG_GC, type, SSR))
1071 return 1;
1072 return 0;
1073 }
1074
1075 /*
1076 * flush out current segment and replace it with new segment
1077 * This function should be returned with success, otherwise BUG
1078 */
1079 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1080 int type, bool force)
1081 {
1082 struct curseg_info *curseg = CURSEG_I(sbi, type);
1083
1084 if (force)
1085 new_curseg(sbi, type, true);
1086 else if (type == CURSEG_WARM_NODE)
1087 new_curseg(sbi, type, false);
1088 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1089 new_curseg(sbi, type, false);
1090 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1091 change_curseg(sbi, type, true);
1092 else
1093 new_curseg(sbi, type, false);
1094
1095 stat_inc_seg_type(sbi, curseg);
1096 }
1097
1098 static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1099 {
1100 struct curseg_info *curseg = CURSEG_I(sbi, type);
1101 unsigned int old_segno;
1102
1103 old_segno = curseg->segno;
1104 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1105 locate_dirty_segment(sbi, old_segno);
1106 }
1107
1108 void allocate_new_segments(struct f2fs_sb_info *sbi)
1109 {
1110 int i;
1111
1112 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1113 __allocate_new_segments(sbi, i);
1114 }
1115
1116 static const struct segment_allocation default_salloc_ops = {
1117 .allocate_segment = allocate_segment_by_default,
1118 };
1119
1120 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1121 {
1122 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1123 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1124 unsigned int start_segno, end_segno;
1125 struct cp_control cpc;
1126 int err = 0;
1127
1128 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1129 return -EINVAL;
1130
1131 cpc.trimmed = 0;
1132 if (end <= MAIN_BLKADDR(sbi))
1133 goto out;
1134
1135 /* start/end segment number in main_area */
1136 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1137 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1138 GET_SEGNO(sbi, end);
1139 cpc.reason = CP_DISCARD;
1140 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1141
1142 /* do checkpoint to issue discard commands safely */
1143 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1144 cpc.trim_start = start_segno;
1145
1146 if (sbi->discard_blks == 0)
1147 break;
1148 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1149 cpc.trim_end = end_segno;
1150 else
1151 cpc.trim_end = min_t(unsigned int,
1152 rounddown(start_segno +
1153 BATCHED_TRIM_SEGMENTS(sbi),
1154 sbi->segs_per_sec) - 1, end_segno);
1155
1156 mutex_lock(&sbi->gc_mutex);
1157 err = write_checkpoint(sbi, &cpc);
1158 mutex_unlock(&sbi->gc_mutex);
1159 }
1160 out:
1161 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1162 return err;
1163 }
1164
1165 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1166 {
1167 struct curseg_info *curseg = CURSEG_I(sbi, type);
1168 if (curseg->next_blkoff < sbi->blocks_per_seg)
1169 return true;
1170 return false;
1171 }
1172
1173 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1174 {
1175 if (p_type == DATA)
1176 return CURSEG_HOT_DATA;
1177 else
1178 return CURSEG_HOT_NODE;
1179 }
1180
1181 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1182 {
1183 if (p_type == DATA) {
1184 struct inode *inode = page->mapping->host;
1185
1186 if (S_ISDIR(inode->i_mode))
1187 return CURSEG_HOT_DATA;
1188 else
1189 return CURSEG_COLD_DATA;
1190 } else {
1191 if (IS_DNODE(page) && is_cold_node(page))
1192 return CURSEG_WARM_NODE;
1193 else
1194 return CURSEG_COLD_NODE;
1195 }
1196 }
1197
1198 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1199 {
1200 if (p_type == DATA) {
1201 struct inode *inode = page->mapping->host;
1202
1203 if (S_ISDIR(inode->i_mode))
1204 return CURSEG_HOT_DATA;
1205 else if (is_cold_data(page) || file_is_cold(inode))
1206 return CURSEG_COLD_DATA;
1207 else
1208 return CURSEG_WARM_DATA;
1209 } else {
1210 if (IS_DNODE(page))
1211 return is_cold_node(page) ? CURSEG_WARM_NODE :
1212 CURSEG_HOT_NODE;
1213 else
1214 return CURSEG_COLD_NODE;
1215 }
1216 }
1217
1218 static int __get_segment_type(struct page *page, enum page_type p_type)
1219 {
1220 switch (F2FS_P_SB(page)->active_logs) {
1221 case 2:
1222 return __get_segment_type_2(page, p_type);
1223 case 4:
1224 return __get_segment_type_4(page, p_type);
1225 }
1226 /* NR_CURSEG_TYPE(6) logs by default */
1227 f2fs_bug_on(F2FS_P_SB(page),
1228 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1229 return __get_segment_type_6(page, p_type);
1230 }
1231
1232 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1233 block_t old_blkaddr, block_t *new_blkaddr,
1234 struct f2fs_summary *sum, int type)
1235 {
1236 struct sit_info *sit_i = SIT_I(sbi);
1237 struct curseg_info *curseg;
1238 bool direct_io = (type == CURSEG_DIRECT_IO);
1239
1240 type = direct_io ? CURSEG_WARM_DATA : type;
1241
1242 curseg = CURSEG_I(sbi, type);
1243
1244 mutex_lock(&curseg->curseg_mutex);
1245 mutex_lock(&sit_i->sentry_lock);
1246
1247 /* direct_io'ed data is aligned to the segment for better performance */
1248 if (direct_io && curseg->next_blkoff &&
1249 !has_not_enough_free_secs(sbi, 0))
1250 __allocate_new_segments(sbi, type);
1251
1252 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1253
1254 /*
1255 * __add_sum_entry should be resided under the curseg_mutex
1256 * because, this function updates a summary entry in the
1257 * current summary block.
1258 */
1259 __add_sum_entry(sbi, type, sum);
1260
1261 __refresh_next_blkoff(sbi, curseg);
1262
1263 stat_inc_block_count(sbi, curseg);
1264
1265 if (!__has_curseg_space(sbi, type))
1266 sit_i->s_ops->allocate_segment(sbi, type, false);
1267 /*
1268 * SIT information should be updated before segment allocation,
1269 * since SSR needs latest valid block information.
1270 */
1271 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1272
1273 mutex_unlock(&sit_i->sentry_lock);
1274
1275 if (page && IS_NODESEG(type))
1276 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1277
1278 mutex_unlock(&curseg->curseg_mutex);
1279 }
1280
1281 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1282 {
1283 int type = __get_segment_type(fio->page, fio->type);
1284
1285 allocate_data_block(fio->sbi, fio->page, fio->blk_addr,
1286 &fio->blk_addr, sum, type);
1287
1288 /* writeout dirty page into bdev */
1289 f2fs_submit_page_mbio(fio);
1290 }
1291
1292 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1293 {
1294 struct f2fs_io_info fio = {
1295 .sbi = sbi,
1296 .type = META,
1297 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1298 .blk_addr = page->index,
1299 .page = page,
1300 .encrypted_page = NULL,
1301 };
1302
1303 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1304 fio.rw &= ~REQ_META;
1305
1306 set_page_writeback(page);
1307 f2fs_submit_page_mbio(&fio);
1308 }
1309
1310 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1311 {
1312 struct f2fs_summary sum;
1313
1314 set_summary(&sum, nid, 0, 0);
1315 do_write_page(&sum, fio);
1316 }
1317
1318 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1319 {
1320 struct f2fs_sb_info *sbi = fio->sbi;
1321 struct f2fs_summary sum;
1322 struct node_info ni;
1323
1324 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1325 get_node_info(sbi, dn->nid, &ni);
1326 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1327 do_write_page(&sum, fio);
1328 dn->data_blkaddr = fio->blk_addr;
1329 }
1330
1331 void rewrite_data_page(struct f2fs_io_info *fio)
1332 {
1333 stat_inc_inplace_blocks(fio->sbi);
1334 f2fs_submit_page_mbio(fio);
1335 }
1336
1337 static void __f2fs_replace_block(struct f2fs_sb_info *sbi,
1338 struct f2fs_summary *sum,
1339 block_t old_blkaddr, block_t new_blkaddr,
1340 bool recover_curseg)
1341 {
1342 struct sit_info *sit_i = SIT_I(sbi);
1343 struct curseg_info *curseg;
1344 unsigned int segno, old_cursegno;
1345 struct seg_entry *se;
1346 int type;
1347 unsigned short old_blkoff;
1348
1349 segno = GET_SEGNO(sbi, new_blkaddr);
1350 se = get_seg_entry(sbi, segno);
1351 type = se->type;
1352
1353 if (!recover_curseg) {
1354 /* for recovery flow */
1355 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1356 if (old_blkaddr == NULL_ADDR)
1357 type = CURSEG_COLD_DATA;
1358 else
1359 type = CURSEG_WARM_DATA;
1360 }
1361 } else {
1362 if (!IS_CURSEG(sbi, segno))
1363 type = CURSEG_WARM_DATA;
1364 }
1365
1366 curseg = CURSEG_I(sbi, type);
1367
1368 mutex_lock(&curseg->curseg_mutex);
1369 mutex_lock(&sit_i->sentry_lock);
1370
1371 old_cursegno = curseg->segno;
1372 old_blkoff = curseg->next_blkoff;
1373
1374 /* change the current segment */
1375 if (segno != curseg->segno) {
1376 curseg->next_segno = segno;
1377 change_curseg(sbi, type, true);
1378 }
1379
1380 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1381 __add_sum_entry(sbi, type, sum);
1382
1383 if (!recover_curseg)
1384 update_sit_entry(sbi, new_blkaddr, 1);
1385 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1386 update_sit_entry(sbi, old_blkaddr, -1);
1387
1388 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1389 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1390
1391 locate_dirty_segment(sbi, old_cursegno);
1392
1393 if (recover_curseg) {
1394 if (old_cursegno != curseg->segno) {
1395 curseg->next_segno = old_cursegno;
1396 change_curseg(sbi, type, true);
1397 }
1398 curseg->next_blkoff = old_blkoff;
1399 }
1400
1401 mutex_unlock(&sit_i->sentry_lock);
1402 mutex_unlock(&curseg->curseg_mutex);
1403 }
1404
1405 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1406 block_t old_addr, block_t new_addr,
1407 unsigned char version, bool recover_curseg)
1408 {
1409 struct f2fs_summary sum;
1410
1411 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1412
1413 __f2fs_replace_block(sbi, &sum, old_addr, new_addr, recover_curseg);
1414
1415 dn->data_blkaddr = new_addr;
1416 set_data_blkaddr(dn);
1417 f2fs_update_extent_cache(dn);
1418 }
1419
1420 void f2fs_wait_on_page_writeback(struct page *page,
1421 enum page_type type, bool ordered)
1422 {
1423 if (PageWriteback(page)) {
1424 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1425
1426 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
1427 if (ordered)
1428 wait_on_page_writeback(page);
1429 else
1430 wait_for_stable_page(page);
1431 }
1432 }
1433
1434 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1435 block_t blkaddr)
1436 {
1437 struct page *cpage;
1438
1439 if (blkaddr == NEW_ADDR)
1440 return;
1441
1442 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1443
1444 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1445 if (cpage) {
1446 f2fs_wait_on_page_writeback(cpage, DATA, true);
1447 f2fs_put_page(cpage, 1);
1448 }
1449 }
1450
1451 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1452 {
1453 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1454 struct curseg_info *seg_i;
1455 unsigned char *kaddr;
1456 struct page *page;
1457 block_t start;
1458 int i, j, offset;
1459
1460 start = start_sum_block(sbi);
1461
1462 page = get_meta_page(sbi, start++);
1463 kaddr = (unsigned char *)page_address(page);
1464
1465 /* Step 1: restore nat cache */
1466 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1467 memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1468
1469 /* Step 2: restore sit cache */
1470 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1471 memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1472 SUM_JOURNAL_SIZE);
1473 offset = 2 * SUM_JOURNAL_SIZE;
1474
1475 /* Step 3: restore summary entries */
1476 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1477 unsigned short blk_off;
1478 unsigned int segno;
1479
1480 seg_i = CURSEG_I(sbi, i);
1481 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1482 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1483 seg_i->next_segno = segno;
1484 reset_curseg(sbi, i, 0);
1485 seg_i->alloc_type = ckpt->alloc_type[i];
1486 seg_i->next_blkoff = blk_off;
1487
1488 if (seg_i->alloc_type == SSR)
1489 blk_off = sbi->blocks_per_seg;
1490
1491 for (j = 0; j < blk_off; j++) {
1492 struct f2fs_summary *s;
1493 s = (struct f2fs_summary *)(kaddr + offset);
1494 seg_i->sum_blk->entries[j] = *s;
1495 offset += SUMMARY_SIZE;
1496 if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1497 SUM_FOOTER_SIZE)
1498 continue;
1499
1500 f2fs_put_page(page, 1);
1501 page = NULL;
1502
1503 page = get_meta_page(sbi, start++);
1504 kaddr = (unsigned char *)page_address(page);
1505 offset = 0;
1506 }
1507 }
1508 f2fs_put_page(page, 1);
1509 return 0;
1510 }
1511
1512 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1513 {
1514 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1515 struct f2fs_summary_block *sum;
1516 struct curseg_info *curseg;
1517 struct page *new;
1518 unsigned short blk_off;
1519 unsigned int segno = 0;
1520 block_t blk_addr = 0;
1521
1522 /* get segment number and block addr */
1523 if (IS_DATASEG(type)) {
1524 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1525 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1526 CURSEG_HOT_DATA]);
1527 if (__exist_node_summaries(sbi))
1528 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1529 else
1530 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1531 } else {
1532 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1533 CURSEG_HOT_NODE]);
1534 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1535 CURSEG_HOT_NODE]);
1536 if (__exist_node_summaries(sbi))
1537 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1538 type - CURSEG_HOT_NODE);
1539 else
1540 blk_addr = GET_SUM_BLOCK(sbi, segno);
1541 }
1542
1543 new = get_meta_page(sbi, blk_addr);
1544 sum = (struct f2fs_summary_block *)page_address(new);
1545
1546 if (IS_NODESEG(type)) {
1547 if (__exist_node_summaries(sbi)) {
1548 struct f2fs_summary *ns = &sum->entries[0];
1549 int i;
1550 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1551 ns->version = 0;
1552 ns->ofs_in_node = 0;
1553 }
1554 } else {
1555 int err;
1556
1557 err = restore_node_summary(sbi, segno, sum);
1558 if (err) {
1559 f2fs_put_page(new, 1);
1560 return err;
1561 }
1562 }
1563 }
1564
1565 /* set uncompleted segment to curseg */
1566 curseg = CURSEG_I(sbi, type);
1567 mutex_lock(&curseg->curseg_mutex);
1568 memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1569 curseg->next_segno = segno;
1570 reset_curseg(sbi, type, 0);
1571 curseg->alloc_type = ckpt->alloc_type[type];
1572 curseg->next_blkoff = blk_off;
1573 mutex_unlock(&curseg->curseg_mutex);
1574 f2fs_put_page(new, 1);
1575 return 0;
1576 }
1577
1578 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1579 {
1580 int type = CURSEG_HOT_DATA;
1581 int err;
1582
1583 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1584 int npages = npages_for_summary_flush(sbi, true);
1585
1586 if (npages >= 2)
1587 ra_meta_pages(sbi, start_sum_block(sbi), npages,
1588 META_CP, true);
1589
1590 /* restore for compacted data summary */
1591 if (read_compacted_summaries(sbi))
1592 return -EINVAL;
1593 type = CURSEG_HOT_NODE;
1594 }
1595
1596 if (__exist_node_summaries(sbi))
1597 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1598 NR_CURSEG_TYPE - type, META_CP, true);
1599
1600 for (; type <= CURSEG_COLD_NODE; type++) {
1601 err = read_normal_summaries(sbi, type);
1602 if (err)
1603 return err;
1604 }
1605
1606 return 0;
1607 }
1608
1609 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1610 {
1611 struct page *page;
1612 unsigned char *kaddr;
1613 struct f2fs_summary *summary;
1614 struct curseg_info *seg_i;
1615 int written_size = 0;
1616 int i, j;
1617
1618 page = grab_meta_page(sbi, blkaddr++);
1619 kaddr = (unsigned char *)page_address(page);
1620
1621 /* Step 1: write nat cache */
1622 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1623 memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1624 written_size += SUM_JOURNAL_SIZE;
1625
1626 /* Step 2: write sit cache */
1627 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1628 memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1629 SUM_JOURNAL_SIZE);
1630 written_size += SUM_JOURNAL_SIZE;
1631
1632 /* Step 3: write summary entries */
1633 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1634 unsigned short blkoff;
1635 seg_i = CURSEG_I(sbi, i);
1636 if (sbi->ckpt->alloc_type[i] == SSR)
1637 blkoff = sbi->blocks_per_seg;
1638 else
1639 blkoff = curseg_blkoff(sbi, i);
1640
1641 for (j = 0; j < blkoff; j++) {
1642 if (!page) {
1643 page = grab_meta_page(sbi, blkaddr++);
1644 kaddr = (unsigned char *)page_address(page);
1645 written_size = 0;
1646 }
1647 summary = (struct f2fs_summary *)(kaddr + written_size);
1648 *summary = seg_i->sum_blk->entries[j];
1649 written_size += SUMMARY_SIZE;
1650
1651 if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1652 SUM_FOOTER_SIZE)
1653 continue;
1654
1655 set_page_dirty(page);
1656 f2fs_put_page(page, 1);
1657 page = NULL;
1658 }
1659 }
1660 if (page) {
1661 set_page_dirty(page);
1662 f2fs_put_page(page, 1);
1663 }
1664 }
1665
1666 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1667 block_t blkaddr, int type)
1668 {
1669 int i, end;
1670 if (IS_DATASEG(type))
1671 end = type + NR_CURSEG_DATA_TYPE;
1672 else
1673 end = type + NR_CURSEG_NODE_TYPE;
1674
1675 for (i = type; i < end; i++) {
1676 struct curseg_info *sum = CURSEG_I(sbi, i);
1677 mutex_lock(&sum->curseg_mutex);
1678 write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1679 mutex_unlock(&sum->curseg_mutex);
1680 }
1681 }
1682
1683 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1684 {
1685 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1686 write_compacted_summaries(sbi, start_blk);
1687 else
1688 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1689 }
1690
1691 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1692 {
1693 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1694 }
1695
1696 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1697 unsigned int val, int alloc)
1698 {
1699 int i;
1700
1701 if (type == NAT_JOURNAL) {
1702 for (i = 0; i < nats_in_cursum(sum); i++) {
1703 if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1704 return i;
1705 }
1706 if (alloc && __has_cursum_space(sum, 1, NAT_JOURNAL))
1707 return update_nats_in_cursum(sum, 1);
1708 } else if (type == SIT_JOURNAL) {
1709 for (i = 0; i < sits_in_cursum(sum); i++)
1710 if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1711 return i;
1712 if (alloc && __has_cursum_space(sum, 1, SIT_JOURNAL))
1713 return update_sits_in_cursum(sum, 1);
1714 }
1715 return -1;
1716 }
1717
1718 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1719 unsigned int segno)
1720 {
1721 return get_meta_page(sbi, current_sit_addr(sbi, segno));
1722 }
1723
1724 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1725 unsigned int start)
1726 {
1727 struct sit_info *sit_i = SIT_I(sbi);
1728 struct page *src_page, *dst_page;
1729 pgoff_t src_off, dst_off;
1730 void *src_addr, *dst_addr;
1731
1732 src_off = current_sit_addr(sbi, start);
1733 dst_off = next_sit_addr(sbi, src_off);
1734
1735 /* get current sit block page without lock */
1736 src_page = get_meta_page(sbi, src_off);
1737 dst_page = grab_meta_page(sbi, dst_off);
1738 f2fs_bug_on(sbi, PageDirty(src_page));
1739
1740 src_addr = page_address(src_page);
1741 dst_addr = page_address(dst_page);
1742 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1743
1744 set_page_dirty(dst_page);
1745 f2fs_put_page(src_page, 1);
1746
1747 set_to_next_sit(sit_i, start);
1748
1749 return dst_page;
1750 }
1751
1752 static struct sit_entry_set *grab_sit_entry_set(void)
1753 {
1754 struct sit_entry_set *ses =
1755 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
1756
1757 ses->entry_cnt = 0;
1758 INIT_LIST_HEAD(&ses->set_list);
1759 return ses;
1760 }
1761
1762 static void release_sit_entry_set(struct sit_entry_set *ses)
1763 {
1764 list_del(&ses->set_list);
1765 kmem_cache_free(sit_entry_set_slab, ses);
1766 }
1767
1768 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1769 struct list_head *head)
1770 {
1771 struct sit_entry_set *next = ses;
1772
1773 if (list_is_last(&ses->set_list, head))
1774 return;
1775
1776 list_for_each_entry_continue(next, head, set_list)
1777 if (ses->entry_cnt <= next->entry_cnt)
1778 break;
1779
1780 list_move_tail(&ses->set_list, &next->set_list);
1781 }
1782
1783 static void add_sit_entry(unsigned int segno, struct list_head *head)
1784 {
1785 struct sit_entry_set *ses;
1786 unsigned int start_segno = START_SEGNO(segno);
1787
1788 list_for_each_entry(ses, head, set_list) {
1789 if (ses->start_segno == start_segno) {
1790 ses->entry_cnt++;
1791 adjust_sit_entry_set(ses, head);
1792 return;
1793 }
1794 }
1795
1796 ses = grab_sit_entry_set();
1797
1798 ses->start_segno = start_segno;
1799 ses->entry_cnt++;
1800 list_add(&ses->set_list, head);
1801 }
1802
1803 static void add_sits_in_set(struct f2fs_sb_info *sbi)
1804 {
1805 struct f2fs_sm_info *sm_info = SM_I(sbi);
1806 struct list_head *set_list = &sm_info->sit_entry_set;
1807 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
1808 unsigned int segno;
1809
1810 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
1811 add_sit_entry(segno, set_list);
1812 }
1813
1814 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
1815 {
1816 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1817 struct f2fs_summary_block *sum = curseg->sum_blk;
1818 int i;
1819
1820 for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1821 unsigned int segno;
1822 bool dirtied;
1823
1824 segno = le32_to_cpu(segno_in_journal(sum, i));
1825 dirtied = __mark_sit_entry_dirty(sbi, segno);
1826
1827 if (!dirtied)
1828 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
1829 }
1830 update_sits_in_cursum(sum, -sits_in_cursum(sum));
1831 }
1832
1833 /*
1834 * CP calls this function, which flushes SIT entries including sit_journal,
1835 * and moves prefree segs to free segs.
1836 */
1837 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1838 {
1839 struct sit_info *sit_i = SIT_I(sbi);
1840 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1841 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1842 struct f2fs_summary_block *sum = curseg->sum_blk;
1843 struct sit_entry_set *ses, *tmp;
1844 struct list_head *head = &SM_I(sbi)->sit_entry_set;
1845 bool to_journal = true;
1846 struct seg_entry *se;
1847
1848 mutex_lock(&curseg->curseg_mutex);
1849 mutex_lock(&sit_i->sentry_lock);
1850
1851 if (!sit_i->dirty_sentries)
1852 goto out;
1853
1854 /*
1855 * add and account sit entries of dirty bitmap in sit entry
1856 * set temporarily
1857 */
1858 add_sits_in_set(sbi);
1859
1860 /*
1861 * if there are no enough space in journal to store dirty sit
1862 * entries, remove all entries from journal and add and account
1863 * them in sit entry set.
1864 */
1865 if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1866 remove_sits_in_journal(sbi);
1867
1868 /*
1869 * there are two steps to flush sit entries:
1870 * #1, flush sit entries to journal in current cold data summary block.
1871 * #2, flush sit entries to sit page.
1872 */
1873 list_for_each_entry_safe(ses, tmp, head, set_list) {
1874 struct page *page = NULL;
1875 struct f2fs_sit_block *raw_sit = NULL;
1876 unsigned int start_segno = ses->start_segno;
1877 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
1878 (unsigned long)MAIN_SEGS(sbi));
1879 unsigned int segno = start_segno;
1880
1881 if (to_journal &&
1882 !__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1883 to_journal = false;
1884
1885 if (!to_journal) {
1886 page = get_next_sit_page(sbi, start_segno);
1887 raw_sit = page_address(page);
1888 }
1889
1890 /* flush dirty sit entries in region of current sit set */
1891 for_each_set_bit_from(segno, bitmap, end) {
1892 int offset, sit_offset;
1893
1894 se = get_seg_entry(sbi, segno);
1895
1896 /* add discard candidates */
1897 if (cpc->reason != CP_DISCARD) {
1898 cpc->trim_start = segno;
1899 add_discard_addrs(sbi, cpc);
1900 }
1901
1902 if (to_journal) {
1903 offset = lookup_journal_in_cursum(sum,
1904 SIT_JOURNAL, segno, 1);
1905 f2fs_bug_on(sbi, offset < 0);
1906 segno_in_journal(sum, offset) =
1907 cpu_to_le32(segno);
1908 seg_info_to_raw_sit(se,
1909 &sit_in_journal(sum, offset));
1910 } else {
1911 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1912 seg_info_to_raw_sit(se,
1913 &raw_sit->entries[sit_offset]);
1914 }
1915
1916 __clear_bit(segno, bitmap);
1917 sit_i->dirty_sentries--;
1918 ses->entry_cnt--;
1919 }
1920
1921 if (!to_journal)
1922 f2fs_put_page(page, 1);
1923
1924 f2fs_bug_on(sbi, ses->entry_cnt);
1925 release_sit_entry_set(ses);
1926 }
1927
1928 f2fs_bug_on(sbi, !list_empty(head));
1929 f2fs_bug_on(sbi, sit_i->dirty_sentries);
1930 out:
1931 if (cpc->reason == CP_DISCARD) {
1932 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1933 add_discard_addrs(sbi, cpc);
1934 }
1935 mutex_unlock(&sit_i->sentry_lock);
1936 mutex_unlock(&curseg->curseg_mutex);
1937
1938 set_prefree_as_free_segments(sbi);
1939 }
1940
1941 static int build_sit_info(struct f2fs_sb_info *sbi)
1942 {
1943 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1944 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1945 struct sit_info *sit_i;
1946 unsigned int sit_segs, start;
1947 char *src_bitmap, *dst_bitmap;
1948 unsigned int bitmap_size;
1949
1950 /* allocate memory for SIT information */
1951 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1952 if (!sit_i)
1953 return -ENOMEM;
1954
1955 SM_I(sbi)->sit_info = sit_i;
1956
1957 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
1958 sizeof(struct seg_entry), GFP_KERNEL);
1959 if (!sit_i->sentries)
1960 return -ENOMEM;
1961
1962 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
1963 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
1964 if (!sit_i->dirty_sentries_bitmap)
1965 return -ENOMEM;
1966
1967 for (start = 0; start < MAIN_SEGS(sbi); start++) {
1968 sit_i->sentries[start].cur_valid_map
1969 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1970 sit_i->sentries[start].ckpt_valid_map
1971 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1972 sit_i->sentries[start].discard_map
1973 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1974 if (!sit_i->sentries[start].cur_valid_map ||
1975 !sit_i->sentries[start].ckpt_valid_map ||
1976 !sit_i->sentries[start].discard_map)
1977 return -ENOMEM;
1978 }
1979
1980 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1981 if (!sit_i->tmp_map)
1982 return -ENOMEM;
1983
1984 if (sbi->segs_per_sec > 1) {
1985 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
1986 sizeof(struct sec_entry), GFP_KERNEL);
1987 if (!sit_i->sec_entries)
1988 return -ENOMEM;
1989 }
1990
1991 /* get information related with SIT */
1992 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1993
1994 /* setup SIT bitmap from ckeckpoint pack */
1995 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1996 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1997
1998 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1999 if (!dst_bitmap)
2000 return -ENOMEM;
2001
2002 /* init SIT information */
2003 sit_i->s_ops = &default_salloc_ops;
2004
2005 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2006 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2007 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2008 sit_i->sit_bitmap = dst_bitmap;
2009 sit_i->bitmap_size = bitmap_size;
2010 sit_i->dirty_sentries = 0;
2011 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2012 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2013 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2014 mutex_init(&sit_i->sentry_lock);
2015 return 0;
2016 }
2017
2018 static int build_free_segmap(struct f2fs_sb_info *sbi)
2019 {
2020 struct free_segmap_info *free_i;
2021 unsigned int bitmap_size, sec_bitmap_size;
2022
2023 /* allocate memory for free segmap information */
2024 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2025 if (!free_i)
2026 return -ENOMEM;
2027
2028 SM_I(sbi)->free_info = free_i;
2029
2030 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2031 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2032 if (!free_i->free_segmap)
2033 return -ENOMEM;
2034
2035 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2036 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2037 if (!free_i->free_secmap)
2038 return -ENOMEM;
2039
2040 /* set all segments as dirty temporarily */
2041 memset(free_i->free_segmap, 0xff, bitmap_size);
2042 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2043
2044 /* init free segmap information */
2045 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2046 free_i->free_segments = 0;
2047 free_i->free_sections = 0;
2048 spin_lock_init(&free_i->segmap_lock);
2049 return 0;
2050 }
2051
2052 static int build_curseg(struct f2fs_sb_info *sbi)
2053 {
2054 struct curseg_info *array;
2055 int i;
2056
2057 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2058 if (!array)
2059 return -ENOMEM;
2060
2061 SM_I(sbi)->curseg_array = array;
2062
2063 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2064 mutex_init(&array[i].curseg_mutex);
2065 array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
2066 if (!array[i].sum_blk)
2067 return -ENOMEM;
2068 array[i].segno = NULL_SEGNO;
2069 array[i].next_blkoff = 0;
2070 }
2071 return restore_curseg_summaries(sbi);
2072 }
2073
2074 static void build_sit_entries(struct f2fs_sb_info *sbi)
2075 {
2076 struct sit_info *sit_i = SIT_I(sbi);
2077 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2078 struct f2fs_summary_block *sum = curseg->sum_blk;
2079 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2080 unsigned int i, start, end;
2081 unsigned int readed, start_blk = 0;
2082 int nrpages = MAX_BIO_BLOCKS(sbi);
2083
2084 do {
2085 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
2086
2087 start = start_blk * sit_i->sents_per_block;
2088 end = (start_blk + readed) * sit_i->sents_per_block;
2089
2090 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2091 struct seg_entry *se = &sit_i->sentries[start];
2092 struct f2fs_sit_block *sit_blk;
2093 struct f2fs_sit_entry sit;
2094 struct page *page;
2095
2096 mutex_lock(&curseg->curseg_mutex);
2097 for (i = 0; i < sits_in_cursum(sum); i++) {
2098 if (le32_to_cpu(segno_in_journal(sum, i))
2099 == start) {
2100 sit = sit_in_journal(sum, i);
2101 mutex_unlock(&curseg->curseg_mutex);
2102 goto got_it;
2103 }
2104 }
2105 mutex_unlock(&curseg->curseg_mutex);
2106
2107 page = get_current_sit_page(sbi, start);
2108 sit_blk = (struct f2fs_sit_block *)page_address(page);
2109 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2110 f2fs_put_page(page, 1);
2111 got_it:
2112 check_block_count(sbi, start, &sit);
2113 seg_info_from_raw_sit(se, &sit);
2114
2115 /* build discard map only one time */
2116 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2117 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2118
2119 if (sbi->segs_per_sec > 1) {
2120 struct sec_entry *e = get_sec_entry(sbi, start);
2121 e->valid_blocks += se->valid_blocks;
2122 }
2123 }
2124 start_blk += readed;
2125 } while (start_blk < sit_blk_cnt);
2126 }
2127
2128 static void init_free_segmap(struct f2fs_sb_info *sbi)
2129 {
2130 unsigned int start;
2131 int type;
2132
2133 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2134 struct seg_entry *sentry = get_seg_entry(sbi, start);
2135 if (!sentry->valid_blocks)
2136 __set_free(sbi, start);
2137 }
2138
2139 /* set use the current segments */
2140 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2141 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2142 __set_test_and_inuse(sbi, curseg_t->segno);
2143 }
2144 }
2145
2146 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2147 {
2148 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2149 struct free_segmap_info *free_i = FREE_I(sbi);
2150 unsigned int segno = 0, offset = 0;
2151 unsigned short valid_blocks;
2152
2153 while (1) {
2154 /* find dirty segment based on free segmap */
2155 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2156 if (segno >= MAIN_SEGS(sbi))
2157 break;
2158 offset = segno + 1;
2159 valid_blocks = get_valid_blocks(sbi, segno, 0);
2160 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2161 continue;
2162 if (valid_blocks > sbi->blocks_per_seg) {
2163 f2fs_bug_on(sbi, 1);
2164 continue;
2165 }
2166 mutex_lock(&dirty_i->seglist_lock);
2167 __locate_dirty_segment(sbi, segno, DIRTY);
2168 mutex_unlock(&dirty_i->seglist_lock);
2169 }
2170 }
2171
2172 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2173 {
2174 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2175 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2176
2177 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2178 if (!dirty_i->victim_secmap)
2179 return -ENOMEM;
2180 return 0;
2181 }
2182
2183 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2184 {
2185 struct dirty_seglist_info *dirty_i;
2186 unsigned int bitmap_size, i;
2187
2188 /* allocate memory for dirty segments list information */
2189 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2190 if (!dirty_i)
2191 return -ENOMEM;
2192
2193 SM_I(sbi)->dirty_info = dirty_i;
2194 mutex_init(&dirty_i->seglist_lock);
2195
2196 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2197
2198 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2199 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2200 if (!dirty_i->dirty_segmap[i])
2201 return -ENOMEM;
2202 }
2203
2204 init_dirty_segmap(sbi);
2205 return init_victim_secmap(sbi);
2206 }
2207
2208 /*
2209 * Update min, max modified time for cost-benefit GC algorithm
2210 */
2211 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2212 {
2213 struct sit_info *sit_i = SIT_I(sbi);
2214 unsigned int segno;
2215
2216 mutex_lock(&sit_i->sentry_lock);
2217
2218 sit_i->min_mtime = LLONG_MAX;
2219
2220 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2221 unsigned int i;
2222 unsigned long long mtime = 0;
2223
2224 for (i = 0; i < sbi->segs_per_sec; i++)
2225 mtime += get_seg_entry(sbi, segno + i)->mtime;
2226
2227 mtime = div_u64(mtime, sbi->segs_per_sec);
2228
2229 if (sit_i->min_mtime > mtime)
2230 sit_i->min_mtime = mtime;
2231 }
2232 sit_i->max_mtime = get_mtime(sbi);
2233 mutex_unlock(&sit_i->sentry_lock);
2234 }
2235
2236 int build_segment_manager(struct f2fs_sb_info *sbi)
2237 {
2238 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2239 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2240 struct f2fs_sm_info *sm_info;
2241 int err;
2242
2243 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2244 if (!sm_info)
2245 return -ENOMEM;
2246
2247 /* init sm info */
2248 sbi->sm_info = sm_info;
2249 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2250 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2251 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2252 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2253 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2254 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2255 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2256 sm_info->rec_prefree_segments = sm_info->main_segments *
2257 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2258 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2259 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2260 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2261
2262 INIT_LIST_HEAD(&sm_info->discard_list);
2263 sm_info->nr_discards = 0;
2264 sm_info->max_discards = 0;
2265
2266 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2267
2268 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2269
2270 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2271 err = create_flush_cmd_control(sbi);
2272 if (err)
2273 return err;
2274 }
2275
2276 err = build_sit_info(sbi);
2277 if (err)
2278 return err;
2279 err = build_free_segmap(sbi);
2280 if (err)
2281 return err;
2282 err = build_curseg(sbi);
2283 if (err)
2284 return err;
2285
2286 /* reinit free segmap based on SIT */
2287 build_sit_entries(sbi);
2288
2289 init_free_segmap(sbi);
2290 err = build_dirty_segmap(sbi);
2291 if (err)
2292 return err;
2293
2294 init_min_max_mtime(sbi);
2295 return 0;
2296 }
2297
2298 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2299 enum dirty_type dirty_type)
2300 {
2301 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2302
2303 mutex_lock(&dirty_i->seglist_lock);
2304 kvfree(dirty_i->dirty_segmap[dirty_type]);
2305 dirty_i->nr_dirty[dirty_type] = 0;
2306 mutex_unlock(&dirty_i->seglist_lock);
2307 }
2308
2309 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2310 {
2311 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2312 kvfree(dirty_i->victim_secmap);
2313 }
2314
2315 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2316 {
2317 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2318 int i;
2319
2320 if (!dirty_i)
2321 return;
2322
2323 /* discard pre-free/dirty segments list */
2324 for (i = 0; i < NR_DIRTY_TYPE; i++)
2325 discard_dirty_segmap(sbi, i);
2326
2327 destroy_victim_secmap(sbi);
2328 SM_I(sbi)->dirty_info = NULL;
2329 kfree(dirty_i);
2330 }
2331
2332 static void destroy_curseg(struct f2fs_sb_info *sbi)
2333 {
2334 struct curseg_info *array = SM_I(sbi)->curseg_array;
2335 int i;
2336
2337 if (!array)
2338 return;
2339 SM_I(sbi)->curseg_array = NULL;
2340 for (i = 0; i < NR_CURSEG_TYPE; i++)
2341 kfree(array[i].sum_blk);
2342 kfree(array);
2343 }
2344
2345 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2346 {
2347 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2348 if (!free_i)
2349 return;
2350 SM_I(sbi)->free_info = NULL;
2351 kvfree(free_i->free_segmap);
2352 kvfree(free_i->free_secmap);
2353 kfree(free_i);
2354 }
2355
2356 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2357 {
2358 struct sit_info *sit_i = SIT_I(sbi);
2359 unsigned int start;
2360
2361 if (!sit_i)
2362 return;
2363
2364 if (sit_i->sentries) {
2365 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2366 kfree(sit_i->sentries[start].cur_valid_map);
2367 kfree(sit_i->sentries[start].ckpt_valid_map);
2368 kfree(sit_i->sentries[start].discard_map);
2369 }
2370 }
2371 kfree(sit_i->tmp_map);
2372
2373 kvfree(sit_i->sentries);
2374 kvfree(sit_i->sec_entries);
2375 kvfree(sit_i->dirty_sentries_bitmap);
2376
2377 SM_I(sbi)->sit_info = NULL;
2378 kfree(sit_i->sit_bitmap);
2379 kfree(sit_i);
2380 }
2381
2382 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2383 {
2384 struct f2fs_sm_info *sm_info = SM_I(sbi);
2385
2386 if (!sm_info)
2387 return;
2388 destroy_flush_cmd_control(sbi);
2389 destroy_dirty_segmap(sbi);
2390 destroy_curseg(sbi);
2391 destroy_free_segmap(sbi);
2392 destroy_sit_info(sbi);
2393 sbi->sm_info = NULL;
2394 kfree(sm_info);
2395 }
2396
2397 int __init create_segment_manager_caches(void)
2398 {
2399 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2400 sizeof(struct discard_entry));
2401 if (!discard_entry_slab)
2402 goto fail;
2403
2404 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2405 sizeof(struct sit_entry_set));
2406 if (!sit_entry_set_slab)
2407 goto destory_discard_entry;
2408
2409 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2410 sizeof(struct inmem_pages));
2411 if (!inmem_entry_slab)
2412 goto destroy_sit_entry_set;
2413 return 0;
2414
2415 destroy_sit_entry_set:
2416 kmem_cache_destroy(sit_entry_set_slab);
2417 destory_discard_entry:
2418 kmem_cache_destroy(discard_entry_slab);
2419 fail:
2420 return -ENOMEM;
2421 }
2422
2423 void destroy_segment_manager_caches(void)
2424 {
2425 kmem_cache_destroy(sit_entry_set_slab);
2426 kmem_cache_destroy(discard_entry_slab);
2427 kmem_cache_destroy(inmem_entry_slab);
2428 }
Linux kernel - Deliverables
This page took 0.083515 seconds and 5 git commands to generate.