projects / deliverable / linux.git / blob
? search:


0ceff2891625e71275412d1608767f27c086f6c9
[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
19 #include "f2fs.h"
20 #include "segment.h"
21 #include "node.h"
22 #include "trace.h"
23 #include <trace/events/f2fs.h>
24
25 #define __reverse_ffz(x) __reverse_ffs(~(x))
26
27 static struct kmem_cache *discard_entry_slab;
28 static struct kmem_cache *sit_entry_set_slab;
29 static struct kmem_cache *inmem_entry_slab;
30
31 /*
32 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
33 * MSB and LSB are reversed in a byte by f2fs_set_bit.
34 */
35 static inline unsigned long __reverse_ffs(unsigned long word)
36 {
37 int num = 0;
38
39 #if BITS_PER_LONG == 64
40 if ((word & 0xffffffff) == 0) {
41 num += 32;
42 word >>= 32;
43 }
44 #endif
45 if ((word & 0xffff) == 0) {
46 num += 16;
47 word >>= 16;
48 }
49 if ((word & 0xff) == 0) {
50 num += 8;
51 word >>= 8;
52 }
53 if ((word & 0xf0) == 0)
54 num += 4;
55 else
56 word >>= 4;
57 if ((word & 0xc) == 0)
58 num += 2;
59 else
60 word >>= 2;
61 if ((word & 0x2) == 0)
62 num += 1;
63 return num;
64 }
65
66 /*
67 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
68 * f2fs_set_bit makes MSB and LSB reversed in a byte.
69 * Example:
70 * LSB <--> MSB
71 * f2fs_set_bit(0, bitmap) => 0000 0001
72 * f2fs_set_bit(7, bitmap) => 1000 0000
73 */
74 static unsigned long __find_rev_next_bit(const unsigned long *addr,
75 unsigned long size, unsigned long offset)
76 {
77 while (!f2fs_test_bit(offset, (unsigned char *)addr))
78 offset++;
79
80 if (offset > size)
81 offset = size;
82
83 return offset;
84 #if 0
85 const unsigned long *p = addr + BIT_WORD(offset);
86 unsigned long result = offset & ~(BITS_PER_LONG - 1);
87 unsigned long tmp;
88 unsigned long mask, submask;
89 unsigned long quot, rest;
90
91 if (offset >= size)
92 return size;
93
94 size -= result;
95 offset %= BITS_PER_LONG;
96 if (!offset)
97 goto aligned;
98
99 tmp = *(p++);
100 quot = (offset >> 3) << 3;
101 rest = offset & 0x7;
102 mask = ~0UL << quot;
103 submask = (unsigned char)(0xff << rest) >> rest;
104 submask <<= quot;
105 mask &= submask;
106 tmp &= mask;
107 if (size < BITS_PER_LONG)
108 goto found_first;
109 if (tmp)
110 goto found_middle;
111
112 size -= BITS_PER_LONG;
113 result += BITS_PER_LONG;
114 aligned:
115 while (size & ~(BITS_PER_LONG-1)) {
116 tmp = *(p++);
117 if (tmp)
118 goto found_middle;
119 result += BITS_PER_LONG;
120 size -= BITS_PER_LONG;
121 }
122 if (!size)
123 return result;
124 tmp = *p;
125 found_first:
126 tmp &= (~0UL >> (BITS_PER_LONG - size));
127 if (tmp == 0UL) /* Are any bits set? */
128 return result + size; /* Nope. */
129 found_middle:
130 return result + __reverse_ffs(tmp);
131 #endif
132 }
133
134 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
135 unsigned long size, unsigned long offset)
136 {
137 while (f2fs_test_bit(offset, (unsigned char *)addr))
138 offset++;
139
140 if (offset > size)
141 offset = size;
142
143 return offset;
144 #if 0
145 const unsigned long *p = addr + BIT_WORD(offset);
146 unsigned long result = offset & ~(BITS_PER_LONG - 1);
147 unsigned long tmp;
148 unsigned long mask, submask;
149 unsigned long quot, rest;
150
151 if (offset >= size)
152 return size;
153
154 size -= result;
155 offset %= BITS_PER_LONG;
156 if (!offset)
157 goto aligned;
158
159 tmp = *(p++);
160 quot = (offset >> 3) << 3;
161 rest = offset & 0x7;
162 mask = ~(~0UL << quot);
163 submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
164 submask <<= quot;
165 mask += submask;
166 tmp |= mask;
167 if (size < BITS_PER_LONG)
168 goto found_first;
169 if (~tmp)
170 goto found_middle;
171
172 size -= BITS_PER_LONG;
173 result += BITS_PER_LONG;
174 aligned:
175 while (size & ~(BITS_PER_LONG - 1)) {
176 tmp = *(p++);
177 if (~tmp)
178 goto found_middle;
179 result += BITS_PER_LONG;
180 size -= BITS_PER_LONG;
181 }
182 if (!size)
183 return result;
184 tmp = *p;
185
186 found_first:
187 tmp |= ~0UL << size;
188 if (tmp == ~0UL) /* Are any bits zero? */
189 return result + size; /* Nope. */
190 found_middle:
191 return result + __reverse_ffz(tmp);
192 #endif
193 }
194
195 void register_inmem_page(struct inode *inode, struct page *page)
196 {
197 struct f2fs_inode_info *fi = F2FS_I(inode);
198 struct inmem_pages *new;
199
200 f2fs_trace_pid(page);
201
202 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
203 SetPagePrivate(page);
204
205 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
206
207 /* add atomic page indices to the list */
208 new->page = page;
209 INIT_LIST_HEAD(&new->list);
210
211 /* increase reference count with clean state */
212 mutex_lock(&fi->inmem_lock);
213 get_page(page);
214 list_add_tail(&new->list, &fi->inmem_pages);
215 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
216 mutex_unlock(&fi->inmem_lock);
217
218 trace_f2fs_register_inmem_page(page, INMEM);
219 }
220
221 int commit_inmem_pages(struct inode *inode, bool abort)
222 {
223 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
224 struct f2fs_inode_info *fi = F2FS_I(inode);
225 struct inmem_pages *cur, *tmp;
226 bool submit_bio = false;
227 struct f2fs_io_info fio = {
228 .sbi = sbi,
229 .type = DATA,
230 .rw = WRITE_SYNC | REQ_PRIO,
231 .encrypted_page = NULL,
232 };
233 int err = 0;
234
235 /*
236 * The abort is true only when f2fs_evict_inode is called.
237 * Basically, the f2fs_evict_inode doesn't produce any data writes, so
238 * that we don't need to call f2fs_balance_fs.
239 * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
240 * inode becomes free by iget_locked in f2fs_iget.
241 */
242 if (!abort) {
243 f2fs_balance_fs(sbi);
244 f2fs_lock_op(sbi);
245 }
246
247 mutex_lock(&fi->inmem_lock);
248 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
249 lock_page(cur->page);
250 if (!abort) {
251 if (cur->page->mapping == inode->i_mapping) {
252 set_page_dirty(cur->page);
253 f2fs_wait_on_page_writeback(cur->page, DATA);
254 if (clear_page_dirty_for_io(cur->page))
255 inode_dec_dirty_pages(inode);
256 trace_f2fs_commit_inmem_page(cur->page, INMEM);
257 fio.page = cur->page;
258 err = do_write_data_page(&fio);
259 submit_bio = true;
260 if (err) {
261 unlock_page(cur->page);
262 break;
263 }
264 }
265 } else {
266 trace_f2fs_commit_inmem_page(cur->page, INMEM_DROP);
267 }
268 set_page_private(cur->page, 0);
269 ClearPagePrivate(cur->page);
270 f2fs_put_page(cur->page, 1);
271
272 list_del(&cur->list);
273 kmem_cache_free(inmem_entry_slab, cur);
274 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
275 }
276 mutex_unlock(&fi->inmem_lock);
277
278 if (!abort) {
279 f2fs_unlock_op(sbi);
280 if (submit_bio)
281 f2fs_submit_merged_bio(sbi, DATA, WRITE);
282 }
283 return err;
284 }
285
286 /*
287 * This function balances dirty node and dentry pages.
288 * In addition, it controls garbage collection.
289 */
290 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
291 {
292 /*
293 * We should do GC or end up with checkpoint, if there are so many dirty
294 * dir/node pages without enough free segments.
295 */
296 if (has_not_enough_free_secs(sbi, 0)) {
297 mutex_lock(&sbi->gc_mutex);
298 f2fs_gc(sbi);
299 }
300 }
301
302 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
303 {
304 /* try to shrink extent cache when there is no enough memory */
305 if (!available_free_memory(sbi, EXTENT_CACHE))
306 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
307
308 /* check the # of cached NAT entries */
309 if (!available_free_memory(sbi, NAT_ENTRIES))
310 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
311
312 if (!available_free_memory(sbi, FREE_NIDS))
313 try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES);
314
315 /* checkpoint is the only way to shrink partial cached entries */
316 if (!available_free_memory(sbi, NAT_ENTRIES) ||
317 excess_prefree_segs(sbi) ||
318 !available_free_memory(sbi, INO_ENTRIES))
319 f2fs_sync_fs(sbi->sb, true);
320 }
321
322 static int issue_flush_thread(void *data)
323 {
324 struct f2fs_sb_info *sbi = data;
325 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
326 wait_queue_head_t *q = &fcc->flush_wait_queue;
327 repeat:
328 if (kthread_should_stop())
329 return 0;
330
331 if (!llist_empty(&fcc->issue_list)) {
332 struct bio *bio;
333 struct flush_cmd *cmd, *next;
334 int ret;
335
336 bio = f2fs_bio_alloc(0);
337
338 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
339 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
340
341 bio->bi_bdev = sbi->sb->s_bdev;
342 ret = submit_bio_wait(WRITE_FLUSH, bio);
343
344 llist_for_each_entry_safe(cmd, next,
345 fcc->dispatch_list, llnode) {
346 cmd->ret = ret;
347 complete(&cmd->wait);
348 }
349 bio_put(bio);
350 fcc->dispatch_list = NULL;
351 }
352
353 wait_event_interruptible(*q,
354 kthread_should_stop() || !llist_empty(&fcc->issue_list));
355 goto repeat;
356 }
357
358 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
359 {
360 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
361 struct flush_cmd cmd;
362
363 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
364 test_opt(sbi, FLUSH_MERGE));
365
366 if (test_opt(sbi, NOBARRIER))
367 return 0;
368
369 if (!test_opt(sbi, FLUSH_MERGE)) {
370 struct bio *bio = f2fs_bio_alloc(0);
371 int ret;
372
373 bio->bi_bdev = sbi->sb->s_bdev;
374 ret = submit_bio_wait(WRITE_FLUSH, bio);
375 bio_put(bio);
376 return ret;
377 }
378
379 init_completion(&cmd.wait);
380
381 llist_add(&cmd.llnode, &fcc->issue_list);
382
383 if (!fcc->dispatch_list)
384 wake_up(&fcc->flush_wait_queue);
385
386 wait_for_completion(&cmd.wait);
387
388 return cmd.ret;
389 }
390
391 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
392 {
393 dev_t dev = sbi->sb->s_bdev->bd_dev;
394 struct flush_cmd_control *fcc;
395 int err = 0;
396
397 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
398 if (!fcc)
399 return -ENOMEM;
400 init_waitqueue_head(&fcc->flush_wait_queue);
401 init_llist_head(&fcc->issue_list);
402 SM_I(sbi)->cmd_control_info = fcc;
403 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
404 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
405 if (IS_ERR(fcc->f2fs_issue_flush)) {
406 err = PTR_ERR(fcc->f2fs_issue_flush);
407 kfree(fcc);
408 SM_I(sbi)->cmd_control_info = NULL;
409 return err;
410 }
411
412 return err;
413 }
414
415 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
416 {
417 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
418
419 if (fcc && fcc->f2fs_issue_flush)
420 kthread_stop(fcc->f2fs_issue_flush);
421 kfree(fcc);
422 SM_I(sbi)->cmd_control_info = NULL;
423 }
424
425 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
426 enum dirty_type dirty_type)
427 {
428 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
429
430 /* need not be added */
431 if (IS_CURSEG(sbi, segno))
432 return;
433
434 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
435 dirty_i->nr_dirty[dirty_type]++;
436
437 if (dirty_type == DIRTY) {
438 struct seg_entry *sentry = get_seg_entry(sbi, segno);
439 enum dirty_type t = sentry->type;
440
441 if (unlikely(t >= DIRTY)) {
442 f2fs_bug_on(sbi, 1);
443 return;
444 }
445 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
446 dirty_i->nr_dirty[t]++;
447 }
448 }
449
450 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
451 enum dirty_type dirty_type)
452 {
453 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
454
455 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
456 dirty_i->nr_dirty[dirty_type]--;
457
458 if (dirty_type == DIRTY) {
459 struct seg_entry *sentry = get_seg_entry(sbi, segno);
460 enum dirty_type t = sentry->type;
461
462 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
463 dirty_i->nr_dirty[t]--;
464
465 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
466 clear_bit(GET_SECNO(sbi, segno),
467 dirty_i->victim_secmap);
468 }
469 }
470
471 /*
472 * Should not occur error such as -ENOMEM.
473 * Adding dirty entry into seglist is not critical operation.
474 * If a given segment is one of current working segments, it won't be added.
475 */
476 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
477 {
478 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
479 unsigned short valid_blocks;
480
481 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
482 return;
483
484 mutex_lock(&dirty_i->seglist_lock);
485
486 valid_blocks = get_valid_blocks(sbi, segno, 0);
487
488 if (valid_blocks == 0) {
489 __locate_dirty_segment(sbi, segno, PRE);
490 __remove_dirty_segment(sbi, segno, DIRTY);
491 } else if (valid_blocks < sbi->blocks_per_seg) {
492 __locate_dirty_segment(sbi, segno, DIRTY);
493 } else {
494 /* Recovery routine with SSR needs this */
495 __remove_dirty_segment(sbi, segno, DIRTY);
496 }
497
498 mutex_unlock(&dirty_i->seglist_lock);
499 }
500
501 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
502 block_t blkstart, block_t blklen)
503 {
504 sector_t start = SECTOR_FROM_BLOCK(blkstart);
505 sector_t len = SECTOR_FROM_BLOCK(blklen);
506 struct seg_entry *se;
507 unsigned int offset;
508 block_t i;
509
510 for (i = blkstart; i < blkstart + blklen; i++) {
511 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
512 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
513
514 if (!f2fs_test_and_set_bit(offset, se->discard_map))
515 sbi->discard_blks--;
516 }
517 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
518 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
519 }
520
521 bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
522 {
523 int err = -ENOTSUPP;
524
525 if (test_opt(sbi, DISCARD)) {
526 struct seg_entry *se = get_seg_entry(sbi,
527 GET_SEGNO(sbi, blkaddr));
528 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
529
530 if (f2fs_test_bit(offset, se->discard_map))
531 return false;
532
533 err = f2fs_issue_discard(sbi, blkaddr, 1);
534 }
535
536 if (err) {
537 update_meta_page(sbi, NULL, blkaddr);
538 return true;
539 }
540 return false;
541 }
542
543 static void __add_discard_entry(struct f2fs_sb_info *sbi,
544 struct cp_control *cpc, struct seg_entry *se,
545 unsigned int start, unsigned int end)
546 {
547 struct list_head *head = &SM_I(sbi)->discard_list;
548 struct discard_entry *new, *last;
549
550 if (!list_empty(head)) {
551 last = list_last_entry(head, struct discard_entry, list);
552 if (START_BLOCK(sbi, cpc->trim_start) + start ==
553 last->blkaddr + last->len) {
554 last->len += end - start;
555 goto done;
556 }
557 }
558
559 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
560 INIT_LIST_HEAD(&new->list);
561 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
562 new->len = end - start;
563 list_add_tail(&new->list, head);
564 done:
565 SM_I(sbi)->nr_discards += end - start;
566 }
567
568 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
569 {
570 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
571 int max_blocks = sbi->blocks_per_seg;
572 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
573 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
574 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
575 unsigned long *discard_map = (unsigned long *)se->discard_map;
576 unsigned long *dmap = SIT_I(sbi)->tmp_map;
577 unsigned int start = 0, end = -1;
578 bool force = (cpc->reason == CP_DISCARD);
579 int i;
580
581 if (se->valid_blocks == max_blocks)
582 return;
583
584 if (!force) {
585 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
586 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
587 return;
588 }
589
590 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
591 for (i = 0; i < entries; i++)
592 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
593 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
594
595 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
596 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
597 if (start >= max_blocks)
598 break;
599
600 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
601 __add_discard_entry(sbi, cpc, se, start, end);
602 }
603 }
604
605 void release_discard_addrs(struct f2fs_sb_info *sbi)
606 {
607 struct list_head *head = &(SM_I(sbi)->discard_list);
608 struct discard_entry *entry, *this;
609
610 /* drop caches */
611 list_for_each_entry_safe(entry, this, head, list) {
612 list_del(&entry->list);
613 kmem_cache_free(discard_entry_slab, entry);
614 }
615 }
616
617 /*
618 * Should call clear_prefree_segments after checkpoint is done.
619 */
620 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
621 {
622 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
623 unsigned int segno;
624
625 mutex_lock(&dirty_i->seglist_lock);
626 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
627 __set_test_and_free(sbi, segno);
628 mutex_unlock(&dirty_i->seglist_lock);
629 }
630
631 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
632 {
633 struct list_head *head = &(SM_I(sbi)->discard_list);
634 struct discard_entry *entry, *this;
635 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
636 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
637 unsigned int start = 0, end = -1;
638
639 mutex_lock(&dirty_i->seglist_lock);
640
641 while (1) {
642 int i;
643 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
644 if (start >= MAIN_SEGS(sbi))
645 break;
646 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
647 start + 1);
648
649 for (i = start; i < end; i++)
650 clear_bit(i, prefree_map);
651
652 dirty_i->nr_dirty[PRE] -= end - start;
653
654 if (!test_opt(sbi, DISCARD))
655 continue;
656
657 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
658 (end - start) << sbi->log_blocks_per_seg);
659 }
660 mutex_unlock(&dirty_i->seglist_lock);
661
662 /* send small discards */
663 list_for_each_entry_safe(entry, this, head, list) {
664 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
665 goto skip;
666 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
667 cpc->trimmed += entry->len;
668 skip:
669 list_del(&entry->list);
670 SM_I(sbi)->nr_discards -= entry->len;
671 kmem_cache_free(discard_entry_slab, entry);
672 }
673 }
674
675 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
676 {
677 struct sit_info *sit_i = SIT_I(sbi);
678
679 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
680 sit_i->dirty_sentries++;
681 return false;
682 }
683
684 return true;
685 }
686
687 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
688 unsigned int segno, int modified)
689 {
690 struct seg_entry *se = get_seg_entry(sbi, segno);
691 se->type = type;
692 if (modified)
693 __mark_sit_entry_dirty(sbi, segno);
694 }
695
696 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
697 {
698 struct seg_entry *se;
699 unsigned int segno, offset;
700 long int new_vblocks;
701
702 segno = GET_SEGNO(sbi, blkaddr);
703
704 se = get_seg_entry(sbi, segno);
705 new_vblocks = se->valid_blocks + del;
706 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
707
708 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
709 (new_vblocks > sbi->blocks_per_seg)));
710
711 se->valid_blocks = new_vblocks;
712 se->mtime = get_mtime(sbi);
713 SIT_I(sbi)->max_mtime = se->mtime;
714
715 /* Update valid block bitmap */
716 if (del > 0) {
717 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
718 f2fs_bug_on(sbi, 1);
719 if (!f2fs_test_and_set_bit(offset, se->discard_map))
720 sbi->discard_blks--;
721 } else {
722 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
723 f2fs_bug_on(sbi, 1);
724 if (f2fs_test_and_clear_bit(offset, se->discard_map))
725 sbi->discard_blks++;
726 }
727 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
728 se->ckpt_valid_blocks += del;
729
730 __mark_sit_entry_dirty(sbi, segno);
731
732 /* update total number of valid blocks to be written in ckpt area */
733 SIT_I(sbi)->written_valid_blocks += del;
734
735 if (sbi->segs_per_sec > 1)
736 get_sec_entry(sbi, segno)->valid_blocks += del;
737 }
738
739 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
740 {
741 update_sit_entry(sbi, new, 1);
742 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
743 update_sit_entry(sbi, old, -1);
744
745 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
746 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
747 }
748
749 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
750 {
751 unsigned int segno = GET_SEGNO(sbi, addr);
752 struct sit_info *sit_i = SIT_I(sbi);
753
754 f2fs_bug_on(sbi, addr == NULL_ADDR);
755 if (addr == NEW_ADDR)
756 return;
757
758 /* add it into sit main buffer */
759 mutex_lock(&sit_i->sentry_lock);
760
761 update_sit_entry(sbi, addr, -1);
762
763 /* add it into dirty seglist */
764 locate_dirty_segment(sbi, segno);
765
766 mutex_unlock(&sit_i->sentry_lock);
767 }
768
769 /*
770 * This function should be resided under the curseg_mutex lock
771 */
772 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
773 struct f2fs_summary *sum)
774 {
775 struct curseg_info *curseg = CURSEG_I(sbi, type);
776 void *addr = curseg->sum_blk;
777 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
778 memcpy(addr, sum, sizeof(struct f2fs_summary));
779 }
780
781 /*
782 * Calculate the number of current summary pages for writing
783 */
784 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
785 {
786 int valid_sum_count = 0;
787 int i, sum_in_page;
788
789 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
790 if (sbi->ckpt->alloc_type[i] == SSR)
791 valid_sum_count += sbi->blocks_per_seg;
792 else {
793 if (for_ra)
794 valid_sum_count += le16_to_cpu(
795 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
796 else
797 valid_sum_count += curseg_blkoff(sbi, i);
798 }
799 }
800
801 sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
802 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
803 if (valid_sum_count <= sum_in_page)
804 return 1;
805 else if ((valid_sum_count - sum_in_page) <=
806 (PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
807 return 2;
808 return 3;
809 }
810
811 /*
812 * Caller should put this summary page
813 */
814 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
815 {
816 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
817 }
818
819 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
820 {
821 struct page *page = grab_meta_page(sbi, blk_addr);
822 void *dst = page_address(page);
823
824 if (src)
825 memcpy(dst, src, PAGE_CACHE_SIZE);
826 else
827 memset(dst, 0, PAGE_CACHE_SIZE);
828 set_page_dirty(page);
829 f2fs_put_page(page, 1);
830 }
831
832 static void write_sum_page(struct f2fs_sb_info *sbi,
833 struct f2fs_summary_block *sum_blk, block_t blk_addr)
834 {
835 update_meta_page(sbi, (void *)sum_blk, blk_addr);
836 }
837
838 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
839 {
840 struct curseg_info *curseg = CURSEG_I(sbi, type);
841 unsigned int segno = curseg->segno + 1;
842 struct free_segmap_info *free_i = FREE_I(sbi);
843
844 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
845 return !test_bit(segno, free_i->free_segmap);
846 return 0;
847 }
848
849 /*
850 * Find a new segment from the free segments bitmap to right order
851 * This function should be returned with success, otherwise BUG
852 */
853 static void get_new_segment(struct f2fs_sb_info *sbi,
854 unsigned int *newseg, bool new_sec, int dir)
855 {
856 struct free_segmap_info *free_i = FREE_I(sbi);
857 unsigned int segno, secno, zoneno;
858 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
859 unsigned int hint = *newseg / sbi->segs_per_sec;
860 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
861 unsigned int left_start = hint;
862 bool init = true;
863 int go_left = 0;
864 int i;
865
866 spin_lock(&free_i->segmap_lock);
867
868 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
869 segno = find_next_zero_bit(free_i->free_segmap,
870 MAIN_SEGS(sbi), *newseg + 1);
871 if (segno - *newseg < sbi->segs_per_sec -
872 (*newseg % sbi->segs_per_sec))
873 goto got_it;
874 }
875 find_other_zone:
876 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
877 if (secno >= MAIN_SECS(sbi)) {
878 if (dir == ALLOC_RIGHT) {
879 secno = find_next_zero_bit(free_i->free_secmap,
880 MAIN_SECS(sbi), 0);
881 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
882 } else {
883 go_left = 1;
884 left_start = hint - 1;
885 }
886 }
887 if (go_left == 0)
888 goto skip_left;
889
890 while (test_bit(left_start, free_i->free_secmap)) {
891 if (left_start > 0) {
892 left_start--;
893 continue;
894 }
895 left_start = find_next_zero_bit(free_i->free_secmap,
896 MAIN_SECS(sbi), 0);
897 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
898 break;
899 }
900 secno = left_start;
901 skip_left:
902 hint = secno;
903 segno = secno * sbi->segs_per_sec;
904 zoneno = secno / sbi->secs_per_zone;
905
906 /* give up on finding another zone */
907 if (!init)
908 goto got_it;
909 if (sbi->secs_per_zone == 1)
910 goto got_it;
911 if (zoneno == old_zoneno)
912 goto got_it;
913 if (dir == ALLOC_LEFT) {
914 if (!go_left && zoneno + 1 >= total_zones)
915 goto got_it;
916 if (go_left && zoneno == 0)
917 goto got_it;
918 }
919 for (i = 0; i < NR_CURSEG_TYPE; i++)
920 if (CURSEG_I(sbi, i)->zone == zoneno)
921 break;
922
923 if (i < NR_CURSEG_TYPE) {
924 /* zone is in user, try another */
925 if (go_left)
926 hint = zoneno * sbi->secs_per_zone - 1;
927 else if (zoneno + 1 >= total_zones)
928 hint = 0;
929 else
930 hint = (zoneno + 1) * sbi->secs_per_zone;
931 init = false;
932 goto find_other_zone;
933 }
934 got_it:
935 /* set it as dirty segment in free segmap */
936 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
937 __set_inuse(sbi, segno);
938 *newseg = segno;
939 spin_unlock(&free_i->segmap_lock);
940 }
941
942 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
943 {
944 struct curseg_info *curseg = CURSEG_I(sbi, type);
945 struct summary_footer *sum_footer;
946
947 curseg->segno = curseg->next_segno;
948 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
949 curseg->next_blkoff = 0;
950 curseg->next_segno = NULL_SEGNO;
951
952 sum_footer = &(curseg->sum_blk->footer);
953 memset(sum_footer, 0, sizeof(struct summary_footer));
954 if (IS_DATASEG(type))
955 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
956 if (IS_NODESEG(type))
957 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
958 __set_sit_entry_type(sbi, type, curseg->segno, modified);
959 }
960
961 /*
962 * Allocate a current working segment.
963 * This function always allocates a free segment in LFS manner.
964 */
965 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
966 {
967 struct curseg_info *curseg = CURSEG_I(sbi, type);
968 unsigned int segno = curseg->segno;
969 int dir = ALLOC_LEFT;
970
971 write_sum_page(sbi, curseg->sum_blk,
972 GET_SUM_BLOCK(sbi, segno));
973 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
974 dir = ALLOC_RIGHT;
975
976 if (test_opt(sbi, NOHEAP))
977 dir = ALLOC_RIGHT;
978
979 get_new_segment(sbi, &segno, new_sec, dir);
980 curseg->next_segno = segno;
981 reset_curseg(sbi, type, 1);
982 curseg->alloc_type = LFS;
983 }
984
985 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
986 struct curseg_info *seg, block_t start)
987 {
988 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
989 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
990 unsigned long *target_map = SIT_I(sbi)->tmp_map;
991 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
992 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
993 int i, pos;
994
995 for (i = 0; i < entries; i++)
996 target_map[i] = ckpt_map[i] | cur_map[i];
997
998 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
999
1000 seg->next_blkoff = pos;
1001 }
1002
1003 /*
1004 * If a segment is written by LFS manner, next block offset is just obtained
1005 * by increasing the current block offset. However, if a segment is written by
1006 * SSR manner, next block offset obtained by calling __next_free_blkoff
1007 */
1008 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1009 struct curseg_info *seg)
1010 {
1011 if (seg->alloc_type == SSR)
1012 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1013 else
1014 seg->next_blkoff++;
1015 }
1016
1017 /*
1018 * This function always allocates a used segment(from dirty seglist) by SSR
1019 * manner, so it should recover the existing segment information of valid blocks
1020 */
1021 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1022 {
1023 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1024 struct curseg_info *curseg = CURSEG_I(sbi, type);
1025 unsigned int new_segno = curseg->next_segno;
1026 struct f2fs_summary_block *sum_node;
1027 struct page *sum_page;
1028
1029 write_sum_page(sbi, curseg->sum_blk,
1030 GET_SUM_BLOCK(sbi, curseg->segno));
1031 __set_test_and_inuse(sbi, new_segno);
1032
1033 mutex_lock(&dirty_i->seglist_lock);
1034 __remove_dirty_segment(sbi, new_segno, PRE);
1035 __remove_dirty_segment(sbi, new_segno, DIRTY);
1036 mutex_unlock(&dirty_i->seglist_lock);
1037
1038 reset_curseg(sbi, type, 1);
1039 curseg->alloc_type = SSR;
1040 __next_free_blkoff(sbi, curseg, 0);
1041
1042 if (reuse) {
1043 sum_page = get_sum_page(sbi, new_segno);
1044 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1045 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1046 f2fs_put_page(sum_page, 1);
1047 }
1048 }
1049
1050 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1051 {
1052 struct curseg_info *curseg = CURSEG_I(sbi, type);
1053 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1054
1055 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1056 return v_ops->get_victim(sbi,
1057 &(curseg)->next_segno, BG_GC, type, SSR);
1058
1059 /* For data segments, let's do SSR more intensively */
1060 for (; type >= CURSEG_HOT_DATA; type--)
1061 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1062 BG_GC, type, SSR))
1063 return 1;
1064 return 0;
1065 }
1066
1067 /*
1068 * flush out current segment and replace it with new segment
1069 * This function should be returned with success, otherwise BUG
1070 */
1071 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1072 int type, bool force)
1073 {
1074 struct curseg_info *curseg = CURSEG_I(sbi, type);
1075
1076 if (force)
1077 new_curseg(sbi, type, true);
1078 else if (type == CURSEG_WARM_NODE)
1079 new_curseg(sbi, type, false);
1080 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1081 new_curseg(sbi, type, false);
1082 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1083 change_curseg(sbi, type, true);
1084 else
1085 new_curseg(sbi, type, false);
1086
1087 stat_inc_seg_type(sbi, curseg);
1088 }
1089
1090 static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1091 {
1092 struct curseg_info *curseg = CURSEG_I(sbi, type);
1093 unsigned int old_segno;
1094
1095 old_segno = curseg->segno;
1096 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1097 locate_dirty_segment(sbi, old_segno);
1098 }
1099
1100 void allocate_new_segments(struct f2fs_sb_info *sbi)
1101 {
1102 int i;
1103
1104 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1105 __allocate_new_segments(sbi, i);
1106 }
1107
1108 static const struct segment_allocation default_salloc_ops = {
1109 .allocate_segment = allocate_segment_by_default,
1110 };
1111
1112 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1113 {
1114 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1115 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1116 unsigned int start_segno, end_segno;
1117 struct cp_control cpc;
1118
1119 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1120 return -EINVAL;
1121
1122 cpc.trimmed = 0;
1123 if (end <= MAIN_BLKADDR(sbi))
1124 goto out;
1125
1126 /* start/end segment number in main_area */
1127 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1128 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1129 GET_SEGNO(sbi, end);
1130 cpc.reason = CP_DISCARD;
1131 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1132
1133 /* do checkpoint to issue discard commands safely */
1134 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1135 cpc.trim_start = start_segno;
1136
1137 if (sbi->discard_blks == 0)
1138 break;
1139 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1140 cpc.trim_end = end_segno;
1141 else
1142 cpc.trim_end = min_t(unsigned int,
1143 rounddown(start_segno +
1144 BATCHED_TRIM_SEGMENTS(sbi),
1145 sbi->segs_per_sec) - 1, end_segno);
1146
1147 mutex_lock(&sbi->gc_mutex);
1148 write_checkpoint(sbi, &cpc);
1149 mutex_unlock(&sbi->gc_mutex);
1150 }
1151 out:
1152 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1153 return 0;
1154 }
1155
1156 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1157 {
1158 struct curseg_info *curseg = CURSEG_I(sbi, type);
1159 if (curseg->next_blkoff < sbi->blocks_per_seg)
1160 return true;
1161 return false;
1162 }
1163
1164 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1165 {
1166 if (p_type == DATA)
1167 return CURSEG_HOT_DATA;
1168 else
1169 return CURSEG_HOT_NODE;
1170 }
1171
1172 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1173 {
1174 if (p_type == DATA) {
1175 struct inode *inode = page->mapping->host;
1176
1177 if (S_ISDIR(inode->i_mode))
1178 return CURSEG_HOT_DATA;
1179 else
1180 return CURSEG_COLD_DATA;
1181 } else {
1182 if (IS_DNODE(page) && is_cold_node(page))
1183 return CURSEG_WARM_NODE;
1184 else
1185 return CURSEG_COLD_NODE;
1186 }
1187 }
1188
1189 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1190 {
1191 if (p_type == DATA) {
1192 struct inode *inode = page->mapping->host;
1193
1194 if (S_ISDIR(inode->i_mode))
1195 return CURSEG_HOT_DATA;
1196 else if (is_cold_data(page) || file_is_cold(inode))
1197 return CURSEG_COLD_DATA;
1198 else
1199 return CURSEG_WARM_DATA;
1200 } else {
1201 if (IS_DNODE(page))
1202 return is_cold_node(page) ? CURSEG_WARM_NODE :
1203 CURSEG_HOT_NODE;
1204 else
1205 return CURSEG_COLD_NODE;
1206 }
1207 }
1208
1209 static int __get_segment_type(struct page *page, enum page_type p_type)
1210 {
1211 switch (F2FS_P_SB(page)->active_logs) {
1212 case 2:
1213 return __get_segment_type_2(page, p_type);
1214 case 4:
1215 return __get_segment_type_4(page, p_type);
1216 }
1217 /* NR_CURSEG_TYPE(6) logs by default */
1218 f2fs_bug_on(F2FS_P_SB(page),
1219 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1220 return __get_segment_type_6(page, p_type);
1221 }
1222
1223 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1224 block_t old_blkaddr, block_t *new_blkaddr,
1225 struct f2fs_summary *sum, int type)
1226 {
1227 struct sit_info *sit_i = SIT_I(sbi);
1228 struct curseg_info *curseg;
1229 bool direct_io = (type == CURSEG_DIRECT_IO);
1230
1231 type = direct_io ? CURSEG_WARM_DATA : type;
1232
1233 curseg = CURSEG_I(sbi, type);
1234
1235 mutex_lock(&curseg->curseg_mutex);
1236 mutex_lock(&sit_i->sentry_lock);
1237
1238 /* direct_io'ed data is aligned to the segment for better performance */
1239 if (direct_io && curseg->next_blkoff &&
1240 !has_not_enough_free_secs(sbi, 0))
1241 __allocate_new_segments(sbi, type);
1242
1243 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1244
1245 /*
1246 * __add_sum_entry should be resided under the curseg_mutex
1247 * because, this function updates a summary entry in the
1248 * current summary block.
1249 */
1250 __add_sum_entry(sbi, type, sum);
1251
1252 __refresh_next_blkoff(sbi, curseg);
1253
1254 stat_inc_block_count(sbi, curseg);
1255
1256 if (!__has_curseg_space(sbi, type))
1257 sit_i->s_ops->allocate_segment(sbi, type, false);
1258 /*
1259 * SIT information should be updated before segment allocation,
1260 * since SSR needs latest valid block information.
1261 */
1262 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1263
1264 mutex_unlock(&sit_i->sentry_lock);
1265
1266 if (page && IS_NODESEG(type))
1267 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1268
1269 mutex_unlock(&curseg->curseg_mutex);
1270 }
1271
1272 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1273 {
1274 int type = __get_segment_type(fio->page, fio->type);
1275
1276 allocate_data_block(fio->sbi, fio->page, fio->blk_addr,
1277 &fio->blk_addr, sum, type);
1278
1279 /* writeout dirty page into bdev */
1280 f2fs_submit_page_mbio(fio);
1281 }
1282
1283 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1284 {
1285 struct f2fs_io_info fio = {
1286 .sbi = sbi,
1287 .type = META,
1288 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1289 .blk_addr = page->index,
1290 .page = page,
1291 .encrypted_page = NULL,
1292 };
1293
1294 set_page_writeback(page);
1295 f2fs_submit_page_mbio(&fio);
1296 }
1297
1298 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1299 {
1300 struct f2fs_summary sum;
1301
1302 set_summary(&sum, nid, 0, 0);
1303 do_write_page(&sum, fio);
1304 }
1305
1306 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1307 {
1308 struct f2fs_sb_info *sbi = fio->sbi;
1309 struct f2fs_summary sum;
1310 struct node_info ni;
1311
1312 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1313 get_node_info(sbi, dn->nid, &ni);
1314 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1315 do_write_page(&sum, fio);
1316 dn->data_blkaddr = fio->blk_addr;
1317 }
1318
1319 void rewrite_data_page(struct f2fs_io_info *fio)
1320 {
1321 stat_inc_inplace_blocks(fio->sbi);
1322 f2fs_submit_page_mbio(fio);
1323 }
1324
1325 static void __f2fs_replace_block(struct f2fs_sb_info *sbi,
1326 struct f2fs_summary *sum,
1327 block_t old_blkaddr, block_t new_blkaddr,
1328 bool recover_curseg)
1329 {
1330 struct sit_info *sit_i = SIT_I(sbi);
1331 struct curseg_info *curseg;
1332 unsigned int segno, old_cursegno;
1333 struct seg_entry *se;
1334 int type;
1335 unsigned short old_blkoff;
1336
1337 segno = GET_SEGNO(sbi, new_blkaddr);
1338 se = get_seg_entry(sbi, segno);
1339 type = se->type;
1340
1341 if (!recover_curseg) {
1342 /* for recovery flow */
1343 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1344 if (old_blkaddr == NULL_ADDR)
1345 type = CURSEG_COLD_DATA;
1346 else
1347 type = CURSEG_WARM_DATA;
1348 }
1349 } else {
1350 if (!IS_CURSEG(sbi, segno))
1351 type = CURSEG_WARM_DATA;
1352 }
1353
1354 curseg = CURSEG_I(sbi, type);
1355
1356 mutex_lock(&curseg->curseg_mutex);
1357 mutex_lock(&sit_i->sentry_lock);
1358
1359 old_cursegno = curseg->segno;
1360 old_blkoff = curseg->next_blkoff;
1361
1362 /* change the current segment */
1363 if (segno != curseg->segno) {
1364 curseg->next_segno = segno;
1365 change_curseg(sbi, type, true);
1366 }
1367
1368 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1369 __add_sum_entry(sbi, type, sum);
1370
1371 refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1372 locate_dirty_segment(sbi, old_cursegno);
1373
1374 if (recover_curseg) {
1375 if (old_cursegno != curseg->segno) {
1376 curseg->next_segno = old_cursegno;
1377 change_curseg(sbi, type, true);
1378 }
1379 curseg->next_blkoff = old_blkoff;
1380 }
1381
1382 mutex_unlock(&sit_i->sentry_lock);
1383 mutex_unlock(&curseg->curseg_mutex);
1384 }
1385
1386 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1387 block_t old_addr, block_t new_addr,
1388 unsigned char version, bool recover_curseg)
1389 {
1390 struct f2fs_summary sum;
1391
1392 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1393
1394 __f2fs_replace_block(sbi, &sum, old_addr, new_addr, recover_curseg);
1395
1396 dn->data_blkaddr = new_addr;
1397 set_data_blkaddr(dn);
1398 f2fs_update_extent_cache(dn);
1399 }
1400
1401 static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1402 struct page *page, enum page_type type)
1403 {
1404 enum page_type btype = PAGE_TYPE_OF_BIO(type);
1405 struct f2fs_bio_info *io = &sbi->write_io[btype];
1406 struct bio_vec *bvec;
1407 struct page *target;
1408 int i;
1409
1410 down_read(&io->io_rwsem);
1411 if (!io->bio) {
1412 up_read(&io->io_rwsem);
1413 return false;
1414 }
1415
1416 bio_for_each_segment_all(bvec, io->bio, i) {
1417
1418 if (bvec->bv_page->mapping) {
1419 target = bvec->bv_page;
1420 } else {
1421 struct f2fs_crypto_ctx *ctx;
1422
1423 /* encrypted page */
1424 ctx = (struct f2fs_crypto_ctx *)page_private(
1425 bvec->bv_page);
1426 target = ctx->w.control_page;
1427 }
1428
1429 if (page == target) {
1430 up_read(&io->io_rwsem);
1431 return true;
1432 }
1433 }
1434
1435 up_read(&io->io_rwsem);
1436 return false;
1437 }
1438
1439 void f2fs_wait_on_page_writeback(struct page *page,
1440 enum page_type type)
1441 {
1442 if (PageWriteback(page)) {
1443 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1444
1445 if (is_merged_page(sbi, page, type))
1446 f2fs_submit_merged_bio(sbi, type, WRITE);
1447 wait_on_page_writeback(page);
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);
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);
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 && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
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 && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
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);
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 0.077482 seconds and 4 git commands to generate.