xref: /openbmc/linux/fs/f2fs/segment.c (revision 4660f9c0fe484353b17a4b9d1cc2b036fa895f76)
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/vmalloc.h>
17 
18 #include "f2fs.h"
19 #include "segment.h"
20 #include "node.h"
21 #include <trace/events/f2fs.h>
22 
23 /*
24  * This function balances dirty node and dentry pages.
25  * In addition, it controls garbage collection.
26  */
27 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
28 {
29 	/*
30 	 * We should do GC or end up with checkpoint, if there are so many dirty
31 	 * dir/node pages without enough free segments.
32 	 */
33 	if (has_not_enough_free_secs(sbi, 0)) {
34 		mutex_lock(&sbi->gc_mutex);
35 		f2fs_gc(sbi);
36 	}
37 }
38 
39 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
40 {
41 	/* check the # of cached NAT entries and prefree segments */
42 	if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
43 				excess_prefree_segs(sbi))
44 		f2fs_sync_fs(sbi->sb, true);
45 }
46 
47 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
48 		enum dirty_type dirty_type)
49 {
50 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
51 
52 	/* need not be added */
53 	if (IS_CURSEG(sbi, segno))
54 		return;
55 
56 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
57 		dirty_i->nr_dirty[dirty_type]++;
58 
59 	if (dirty_type == DIRTY) {
60 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
61 		enum dirty_type t = DIRTY_HOT_DATA;
62 
63 		dirty_type = sentry->type;
64 
65 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
66 			dirty_i->nr_dirty[dirty_type]++;
67 
68 		/* Only one bitmap should be set */
69 		for (; t <= DIRTY_COLD_NODE; t++) {
70 			if (t == dirty_type)
71 				continue;
72 			if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
73 				dirty_i->nr_dirty[t]--;
74 		}
75 	}
76 }
77 
78 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
79 		enum dirty_type dirty_type)
80 {
81 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
82 
83 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
84 		dirty_i->nr_dirty[dirty_type]--;
85 
86 	if (dirty_type == DIRTY) {
87 		enum dirty_type t = DIRTY_HOT_DATA;
88 
89 		/* clear its dirty bitmap */
90 		for (; t <= DIRTY_COLD_NODE; t++) {
91 			if (test_and_clear_bit(segno,
92 						dirty_i->dirty_segmap[t])) {
93 				dirty_i->nr_dirty[t]--;
94 				break;
95 			}
96 		}
97 
98 		if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
99 			clear_bit(GET_SECNO(sbi, segno),
100 						dirty_i->victim_secmap);
101 	}
102 }
103 
104 /*
105  * Should not occur error such as -ENOMEM.
106  * Adding dirty entry into seglist is not critical operation.
107  * If a given segment is one of current working segments, it won't be added.
108  */
109 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
110 {
111 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
112 	unsigned short valid_blocks;
113 
114 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
115 		return;
116 
117 	mutex_lock(&dirty_i->seglist_lock);
118 
119 	valid_blocks = get_valid_blocks(sbi, segno, 0);
120 
121 	if (valid_blocks == 0) {
122 		__locate_dirty_segment(sbi, segno, PRE);
123 		__remove_dirty_segment(sbi, segno, DIRTY);
124 	} else if (valid_blocks < sbi->blocks_per_seg) {
125 		__locate_dirty_segment(sbi, segno, DIRTY);
126 	} else {
127 		/* Recovery routine with SSR needs this */
128 		__remove_dirty_segment(sbi, segno, DIRTY);
129 	}
130 
131 	mutex_unlock(&dirty_i->seglist_lock);
132 }
133 
134 /*
135  * Should call clear_prefree_segments after checkpoint is done.
136  */
137 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
138 {
139 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
140 	unsigned int segno = -1;
141 	unsigned int total_segs = TOTAL_SEGS(sbi);
142 
143 	mutex_lock(&dirty_i->seglist_lock);
144 	while (1) {
145 		segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
146 				segno + 1);
147 		if (segno >= total_segs)
148 			break;
149 		__set_test_and_free(sbi, segno);
150 	}
151 	mutex_unlock(&dirty_i->seglist_lock);
152 }
153 
154 void clear_prefree_segments(struct f2fs_sb_info *sbi)
155 {
156 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
157 	unsigned int segno = -1;
158 	unsigned int total_segs = TOTAL_SEGS(sbi);
159 
160 	mutex_lock(&dirty_i->seglist_lock);
161 	while (1) {
162 		segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
163 				segno + 1);
164 		if (segno >= total_segs)
165 			break;
166 
167 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[PRE]))
168 			dirty_i->nr_dirty[PRE]--;
169 
170 		/* Let's use trim */
171 		if (test_opt(sbi, DISCARD))
172 			blkdev_issue_discard(sbi->sb->s_bdev,
173 					START_BLOCK(sbi, segno) <<
174 					sbi->log_sectors_per_block,
175 					1 << (sbi->log_sectors_per_block +
176 						sbi->log_blocks_per_seg),
177 					GFP_NOFS, 0);
178 	}
179 	mutex_unlock(&dirty_i->seglist_lock);
180 }
181 
182 static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
183 {
184 	struct sit_info *sit_i = SIT_I(sbi);
185 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
186 		sit_i->dirty_sentries++;
187 }
188 
189 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
190 					unsigned int segno, int modified)
191 {
192 	struct seg_entry *se = get_seg_entry(sbi, segno);
193 	se->type = type;
194 	if (modified)
195 		__mark_sit_entry_dirty(sbi, segno);
196 }
197 
198 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
199 {
200 	struct seg_entry *se;
201 	unsigned int segno, offset;
202 	long int new_vblocks;
203 
204 	segno = GET_SEGNO(sbi, blkaddr);
205 
206 	se = get_seg_entry(sbi, segno);
207 	new_vblocks = se->valid_blocks + del;
208 	offset = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & (sbi->blocks_per_seg - 1);
209 
210 	BUG_ON((new_vblocks >> (sizeof(unsigned short) << 3) ||
211 				(new_vblocks > sbi->blocks_per_seg)));
212 
213 	se->valid_blocks = new_vblocks;
214 	se->mtime = get_mtime(sbi);
215 	SIT_I(sbi)->max_mtime = se->mtime;
216 
217 	/* Update valid block bitmap */
218 	if (del > 0) {
219 		if (f2fs_set_bit(offset, se->cur_valid_map))
220 			BUG();
221 	} else {
222 		if (!f2fs_clear_bit(offset, se->cur_valid_map))
223 			BUG();
224 	}
225 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
226 		se->ckpt_valid_blocks += del;
227 
228 	__mark_sit_entry_dirty(sbi, segno);
229 
230 	/* update total number of valid blocks to be written in ckpt area */
231 	SIT_I(sbi)->written_valid_blocks += del;
232 
233 	if (sbi->segs_per_sec > 1)
234 		get_sec_entry(sbi, segno)->valid_blocks += del;
235 }
236 
237 static void refresh_sit_entry(struct f2fs_sb_info *sbi,
238 			block_t old_blkaddr, block_t new_blkaddr)
239 {
240 	update_sit_entry(sbi, new_blkaddr, 1);
241 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
242 		update_sit_entry(sbi, old_blkaddr, -1);
243 }
244 
245 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
246 {
247 	unsigned int segno = GET_SEGNO(sbi, addr);
248 	struct sit_info *sit_i = SIT_I(sbi);
249 
250 	BUG_ON(addr == NULL_ADDR);
251 	if (addr == NEW_ADDR)
252 		return;
253 
254 	/* add it into sit main buffer */
255 	mutex_lock(&sit_i->sentry_lock);
256 
257 	update_sit_entry(sbi, addr, -1);
258 
259 	/* add it into dirty seglist */
260 	locate_dirty_segment(sbi, segno);
261 
262 	mutex_unlock(&sit_i->sentry_lock);
263 }
264 
265 /*
266  * This function should be resided under the curseg_mutex lock
267  */
268 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
269 					struct f2fs_summary *sum)
270 {
271 	struct curseg_info *curseg = CURSEG_I(sbi, type);
272 	void *addr = curseg->sum_blk;
273 	addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
274 	memcpy(addr, sum, sizeof(struct f2fs_summary));
275 }
276 
277 /*
278  * Calculate the number of current summary pages for writing
279  */
280 int npages_for_summary_flush(struct f2fs_sb_info *sbi)
281 {
282 	int total_size_bytes = 0;
283 	int valid_sum_count = 0;
284 	int i, sum_space;
285 
286 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
287 		if (sbi->ckpt->alloc_type[i] == SSR)
288 			valid_sum_count += sbi->blocks_per_seg;
289 		else
290 			valid_sum_count += curseg_blkoff(sbi, i);
291 	}
292 
293 	total_size_bytes = valid_sum_count * (SUMMARY_SIZE + 1)
294 			+ sizeof(struct nat_journal) + 2
295 			+ sizeof(struct sit_journal) + 2;
296 	sum_space = PAGE_CACHE_SIZE - SUM_FOOTER_SIZE;
297 	if (total_size_bytes < sum_space)
298 		return 1;
299 	else if (total_size_bytes < 2 * sum_space)
300 		return 2;
301 	return 3;
302 }
303 
304 /*
305  * Caller should put this summary page
306  */
307 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
308 {
309 	return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
310 }
311 
312 static void write_sum_page(struct f2fs_sb_info *sbi,
313 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
314 {
315 	struct page *page = grab_meta_page(sbi, blk_addr);
316 	void *kaddr = page_address(page);
317 	memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
318 	set_page_dirty(page);
319 	f2fs_put_page(page, 1);
320 }
321 
322 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
323 {
324 	struct curseg_info *curseg = CURSEG_I(sbi, type);
325 	unsigned int segno = curseg->segno + 1;
326 	struct free_segmap_info *free_i = FREE_I(sbi);
327 
328 	if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec)
329 		return !test_bit(segno, free_i->free_segmap);
330 	return 0;
331 }
332 
333 /*
334  * Find a new segment from the free segments bitmap to right order
335  * This function should be returned with success, otherwise BUG
336  */
337 static void get_new_segment(struct f2fs_sb_info *sbi,
338 			unsigned int *newseg, bool new_sec, int dir)
339 {
340 	struct free_segmap_info *free_i = FREE_I(sbi);
341 	unsigned int segno, secno, zoneno;
342 	unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone;
343 	unsigned int hint = *newseg / sbi->segs_per_sec;
344 	unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
345 	unsigned int left_start = hint;
346 	bool init = true;
347 	int go_left = 0;
348 	int i;
349 
350 	write_lock(&free_i->segmap_lock);
351 
352 	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
353 		segno = find_next_zero_bit(free_i->free_segmap,
354 					TOTAL_SEGS(sbi), *newseg + 1);
355 		if (segno - *newseg < sbi->segs_per_sec -
356 					(*newseg % sbi->segs_per_sec))
357 			goto got_it;
358 	}
359 find_other_zone:
360 	secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint);
361 	if (secno >= TOTAL_SECS(sbi)) {
362 		if (dir == ALLOC_RIGHT) {
363 			secno = find_next_zero_bit(free_i->free_secmap,
364 							TOTAL_SECS(sbi), 0);
365 			BUG_ON(secno >= TOTAL_SECS(sbi));
366 		} else {
367 			go_left = 1;
368 			left_start = hint - 1;
369 		}
370 	}
371 	if (go_left == 0)
372 		goto skip_left;
373 
374 	while (test_bit(left_start, free_i->free_secmap)) {
375 		if (left_start > 0) {
376 			left_start--;
377 			continue;
378 		}
379 		left_start = find_next_zero_bit(free_i->free_secmap,
380 							TOTAL_SECS(sbi), 0);
381 		BUG_ON(left_start >= TOTAL_SECS(sbi));
382 		break;
383 	}
384 	secno = left_start;
385 skip_left:
386 	hint = secno;
387 	segno = secno * sbi->segs_per_sec;
388 	zoneno = secno / sbi->secs_per_zone;
389 
390 	/* give up on finding another zone */
391 	if (!init)
392 		goto got_it;
393 	if (sbi->secs_per_zone == 1)
394 		goto got_it;
395 	if (zoneno == old_zoneno)
396 		goto got_it;
397 	if (dir == ALLOC_LEFT) {
398 		if (!go_left && zoneno + 1 >= total_zones)
399 			goto got_it;
400 		if (go_left && zoneno == 0)
401 			goto got_it;
402 	}
403 	for (i = 0; i < NR_CURSEG_TYPE; i++)
404 		if (CURSEG_I(sbi, i)->zone == zoneno)
405 			break;
406 
407 	if (i < NR_CURSEG_TYPE) {
408 		/* zone is in user, try another */
409 		if (go_left)
410 			hint = zoneno * sbi->secs_per_zone - 1;
411 		else if (zoneno + 1 >= total_zones)
412 			hint = 0;
413 		else
414 			hint = (zoneno + 1) * sbi->secs_per_zone;
415 		init = false;
416 		goto find_other_zone;
417 	}
418 got_it:
419 	/* set it as dirty segment in free segmap */
420 	BUG_ON(test_bit(segno, free_i->free_segmap));
421 	__set_inuse(sbi, segno);
422 	*newseg = segno;
423 	write_unlock(&free_i->segmap_lock);
424 }
425 
426 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
427 {
428 	struct curseg_info *curseg = CURSEG_I(sbi, type);
429 	struct summary_footer *sum_footer;
430 
431 	curseg->segno = curseg->next_segno;
432 	curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
433 	curseg->next_blkoff = 0;
434 	curseg->next_segno = NULL_SEGNO;
435 
436 	sum_footer = &(curseg->sum_blk->footer);
437 	memset(sum_footer, 0, sizeof(struct summary_footer));
438 	if (IS_DATASEG(type))
439 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
440 	if (IS_NODESEG(type))
441 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
442 	__set_sit_entry_type(sbi, type, curseg->segno, modified);
443 }
444 
445 /*
446  * Allocate a current working segment.
447  * This function always allocates a free segment in LFS manner.
448  */
449 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
450 {
451 	struct curseg_info *curseg = CURSEG_I(sbi, type);
452 	unsigned int segno = curseg->segno;
453 	int dir = ALLOC_LEFT;
454 
455 	write_sum_page(sbi, curseg->sum_blk,
456 				GET_SUM_BLOCK(sbi, segno));
457 	if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
458 		dir = ALLOC_RIGHT;
459 
460 	if (test_opt(sbi, NOHEAP))
461 		dir = ALLOC_RIGHT;
462 
463 	get_new_segment(sbi, &segno, new_sec, dir);
464 	curseg->next_segno = segno;
465 	reset_curseg(sbi, type, 1);
466 	curseg->alloc_type = LFS;
467 }
468 
469 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
470 			struct curseg_info *seg, block_t start)
471 {
472 	struct seg_entry *se = get_seg_entry(sbi, seg->segno);
473 	block_t ofs;
474 	for (ofs = start; ofs < sbi->blocks_per_seg; ofs++) {
475 		if (!f2fs_test_bit(ofs, se->ckpt_valid_map)
476 			&& !f2fs_test_bit(ofs, se->cur_valid_map))
477 			break;
478 	}
479 	seg->next_blkoff = ofs;
480 }
481 
482 /*
483  * If a segment is written by LFS manner, next block offset is just obtained
484  * by increasing the current block offset. However, if a segment is written by
485  * SSR manner, next block offset obtained by calling __next_free_blkoff
486  */
487 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
488 				struct curseg_info *seg)
489 {
490 	if (seg->alloc_type == SSR)
491 		__next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
492 	else
493 		seg->next_blkoff++;
494 }
495 
496 /*
497  * This function always allocates a used segment (from dirty seglist) by SSR
498  * manner, so it should recover the existing segment information of valid blocks
499  */
500 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
501 {
502 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
503 	struct curseg_info *curseg = CURSEG_I(sbi, type);
504 	unsigned int new_segno = curseg->next_segno;
505 	struct f2fs_summary_block *sum_node;
506 	struct page *sum_page;
507 
508 	write_sum_page(sbi, curseg->sum_blk,
509 				GET_SUM_BLOCK(sbi, curseg->segno));
510 	__set_test_and_inuse(sbi, new_segno);
511 
512 	mutex_lock(&dirty_i->seglist_lock);
513 	__remove_dirty_segment(sbi, new_segno, PRE);
514 	__remove_dirty_segment(sbi, new_segno, DIRTY);
515 	mutex_unlock(&dirty_i->seglist_lock);
516 
517 	reset_curseg(sbi, type, 1);
518 	curseg->alloc_type = SSR;
519 	__next_free_blkoff(sbi, curseg, 0);
520 
521 	if (reuse) {
522 		sum_page = get_sum_page(sbi, new_segno);
523 		sum_node = (struct f2fs_summary_block *)page_address(sum_page);
524 		memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
525 		f2fs_put_page(sum_page, 1);
526 	}
527 }
528 
529 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
530 {
531 	struct curseg_info *curseg = CURSEG_I(sbi, type);
532 	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
533 
534 	if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
535 		return v_ops->get_victim(sbi,
536 				&(curseg)->next_segno, BG_GC, type, SSR);
537 
538 	/* For data segments, let's do SSR more intensively */
539 	for (; type >= CURSEG_HOT_DATA; type--)
540 		if (v_ops->get_victim(sbi, &(curseg)->next_segno,
541 						BG_GC, type, SSR))
542 			return 1;
543 	return 0;
544 }
545 
546 /*
547  * flush out current segment and replace it with new segment
548  * This function should be returned with success, otherwise BUG
549  */
550 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
551 						int type, bool force)
552 {
553 	struct curseg_info *curseg = CURSEG_I(sbi, type);
554 
555 	if (force)
556 		new_curseg(sbi, type, true);
557 	else if (type == CURSEG_WARM_NODE)
558 		new_curseg(sbi, type, false);
559 	else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
560 		new_curseg(sbi, type, false);
561 	else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
562 		change_curseg(sbi, type, true);
563 	else
564 		new_curseg(sbi, type, false);
565 
566 	stat_inc_seg_type(sbi, curseg);
567 }
568 
569 void allocate_new_segments(struct f2fs_sb_info *sbi)
570 {
571 	struct curseg_info *curseg;
572 	unsigned int old_curseg;
573 	int i;
574 
575 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
576 		curseg = CURSEG_I(sbi, i);
577 		old_curseg = curseg->segno;
578 		SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
579 		locate_dirty_segment(sbi, old_curseg);
580 	}
581 }
582 
583 static const struct segment_allocation default_salloc_ops = {
584 	.allocate_segment = allocate_segment_by_default,
585 };
586 
587 static void f2fs_end_io_write(struct bio *bio, int err)
588 {
589 	const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
590 	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
591 	struct bio_private *p = bio->bi_private;
592 
593 	do {
594 		struct page *page = bvec->bv_page;
595 
596 		if (--bvec >= bio->bi_io_vec)
597 			prefetchw(&bvec->bv_page->flags);
598 		if (!uptodate) {
599 			SetPageError(page);
600 			if (page->mapping)
601 				set_bit(AS_EIO, &page->mapping->flags);
602 			set_ckpt_flags(p->sbi->ckpt, CP_ERROR_FLAG);
603 			p->sbi->sb->s_flags |= MS_RDONLY;
604 		}
605 		end_page_writeback(page);
606 		dec_page_count(p->sbi, F2FS_WRITEBACK);
607 	} while (bvec >= bio->bi_io_vec);
608 
609 	if (p->is_sync)
610 		complete(p->wait);
611 
612 	if (!get_pages(p->sbi, F2FS_WRITEBACK) && p->sbi->cp_task)
613 		wake_up_process(p->sbi->cp_task);
614 
615 	kfree(p);
616 	bio_put(bio);
617 }
618 
619 struct bio *f2fs_bio_alloc(struct block_device *bdev, int npages)
620 {
621 	struct bio *bio;
622 
623 	/* No failure on bio allocation */
624 	bio = bio_alloc(GFP_NOIO, npages);
625 	bio->bi_bdev = bdev;
626 	bio->bi_private = NULL;
627 
628 	return bio;
629 }
630 
631 static void do_submit_bio(struct f2fs_sb_info *sbi,
632 				enum page_type type, bool sync)
633 {
634 	int rw = sync ? WRITE_SYNC : WRITE;
635 	enum page_type btype = type > META ? META : type;
636 
637 	if (type >= META_FLUSH)
638 		rw = WRITE_FLUSH_FUA;
639 
640 	if (btype == META)
641 		rw |= REQ_META;
642 
643 	if (sbi->bio[btype]) {
644 		struct bio_private *p = sbi->bio[btype]->bi_private;
645 		p->sbi = sbi;
646 		sbi->bio[btype]->bi_end_io = f2fs_end_io_write;
647 
648 		trace_f2fs_do_submit_bio(sbi->sb, btype, sync, sbi->bio[btype]);
649 
650 		if (type == META_FLUSH) {
651 			DECLARE_COMPLETION_ONSTACK(wait);
652 			p->is_sync = true;
653 			p->wait = &wait;
654 			submit_bio(rw, sbi->bio[btype]);
655 			wait_for_completion(&wait);
656 		} else {
657 			p->is_sync = false;
658 			submit_bio(rw, sbi->bio[btype]);
659 		}
660 		sbi->bio[btype] = NULL;
661 	}
662 }
663 
664 void f2fs_submit_bio(struct f2fs_sb_info *sbi, enum page_type type, bool sync)
665 {
666 	down_write(&sbi->bio_sem);
667 	do_submit_bio(sbi, type, sync);
668 	up_write(&sbi->bio_sem);
669 }
670 
671 static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page,
672 				block_t blk_addr, enum page_type type)
673 {
674 	struct block_device *bdev = sbi->sb->s_bdev;
675 	int bio_blocks;
676 
677 	verify_block_addr(sbi, blk_addr);
678 
679 	down_write(&sbi->bio_sem);
680 
681 	inc_page_count(sbi, F2FS_WRITEBACK);
682 
683 	if (sbi->bio[type] && sbi->last_block_in_bio[type] != blk_addr - 1)
684 		do_submit_bio(sbi, type, false);
685 alloc_new:
686 	if (sbi->bio[type] == NULL) {
687 		struct bio_private *priv;
688 retry:
689 		priv = kmalloc(sizeof(struct bio_private), GFP_NOFS);
690 		if (!priv) {
691 			cond_resched();
692 			goto retry;
693 		}
694 
695 		bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
696 		sbi->bio[type] = f2fs_bio_alloc(bdev, bio_blocks);
697 		sbi->bio[type]->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
698 		sbi->bio[type]->bi_private = priv;
699 		/*
700 		 * The end_io will be assigned at the sumbission phase.
701 		 * Until then, let bio_add_page() merge consecutive IOs as much
702 		 * as possible.
703 		 */
704 	}
705 
706 	if (bio_add_page(sbi->bio[type], page, PAGE_CACHE_SIZE, 0) <
707 							PAGE_CACHE_SIZE) {
708 		do_submit_bio(sbi, type, false);
709 		goto alloc_new;
710 	}
711 
712 	sbi->last_block_in_bio[type] = blk_addr;
713 
714 	up_write(&sbi->bio_sem);
715 	trace_f2fs_submit_write_page(page, blk_addr, type);
716 }
717 
718 void f2fs_wait_on_page_writeback(struct page *page,
719 				enum page_type type, bool sync)
720 {
721 	struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
722 	if (PageWriteback(page)) {
723 		f2fs_submit_bio(sbi, type, sync);
724 		wait_on_page_writeback(page);
725 	}
726 }
727 
728 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
729 {
730 	struct curseg_info *curseg = CURSEG_I(sbi, type);
731 	if (curseg->next_blkoff < sbi->blocks_per_seg)
732 		return true;
733 	return false;
734 }
735 
736 static int __get_segment_type_2(struct page *page, enum page_type p_type)
737 {
738 	if (p_type == DATA)
739 		return CURSEG_HOT_DATA;
740 	else
741 		return CURSEG_HOT_NODE;
742 }
743 
744 static int __get_segment_type_4(struct page *page, enum page_type p_type)
745 {
746 	if (p_type == DATA) {
747 		struct inode *inode = page->mapping->host;
748 
749 		if (S_ISDIR(inode->i_mode))
750 			return CURSEG_HOT_DATA;
751 		else
752 			return CURSEG_COLD_DATA;
753 	} else {
754 		if (IS_DNODE(page) && !is_cold_node(page))
755 			return CURSEG_HOT_NODE;
756 		else
757 			return CURSEG_COLD_NODE;
758 	}
759 }
760 
761 static int __get_segment_type_6(struct page *page, enum page_type p_type)
762 {
763 	if (p_type == DATA) {
764 		struct inode *inode = page->mapping->host;
765 
766 		if (S_ISDIR(inode->i_mode))
767 			return CURSEG_HOT_DATA;
768 		else if (is_cold_data(page) || file_is_cold(inode))
769 			return CURSEG_COLD_DATA;
770 		else
771 			return CURSEG_WARM_DATA;
772 	} else {
773 		if (IS_DNODE(page))
774 			return is_cold_node(page) ? CURSEG_WARM_NODE :
775 						CURSEG_HOT_NODE;
776 		else
777 			return CURSEG_COLD_NODE;
778 	}
779 }
780 
781 static int __get_segment_type(struct page *page, enum page_type p_type)
782 {
783 	struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
784 	switch (sbi->active_logs) {
785 	case 2:
786 		return __get_segment_type_2(page, p_type);
787 	case 4:
788 		return __get_segment_type_4(page, p_type);
789 	}
790 	/* NR_CURSEG_TYPE(6) logs by default */
791 	BUG_ON(sbi->active_logs != NR_CURSEG_TYPE);
792 	return __get_segment_type_6(page, p_type);
793 }
794 
795 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
796 			block_t old_blkaddr, block_t *new_blkaddr,
797 			struct f2fs_summary *sum, enum page_type p_type)
798 {
799 	struct sit_info *sit_i = SIT_I(sbi);
800 	struct curseg_info *curseg;
801 	unsigned int old_cursegno;
802 	int type;
803 
804 	type = __get_segment_type(page, p_type);
805 	curseg = CURSEG_I(sbi, type);
806 
807 	mutex_lock(&curseg->curseg_mutex);
808 
809 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
810 	old_cursegno = curseg->segno;
811 
812 	/*
813 	 * __add_sum_entry should be resided under the curseg_mutex
814 	 * because, this function updates a summary entry in the
815 	 * current summary block.
816 	 */
817 	__add_sum_entry(sbi, type, sum);
818 
819 	mutex_lock(&sit_i->sentry_lock);
820 	__refresh_next_blkoff(sbi, curseg);
821 
822 	stat_inc_block_count(sbi, curseg);
823 
824 	/*
825 	 * SIT information should be updated before segment allocation,
826 	 * since SSR needs latest valid block information.
827 	 */
828 	refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
829 
830 	if (!__has_curseg_space(sbi, type))
831 		sit_i->s_ops->allocate_segment(sbi, type, false);
832 
833 	locate_dirty_segment(sbi, old_cursegno);
834 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
835 	mutex_unlock(&sit_i->sentry_lock);
836 
837 	if (p_type == NODE)
838 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
839 
840 	/* writeout dirty page into bdev */
841 	submit_write_page(sbi, page, *new_blkaddr, p_type);
842 
843 	mutex_unlock(&curseg->curseg_mutex);
844 }
845 
846 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
847 {
848 	set_page_writeback(page);
849 	submit_write_page(sbi, page, page->index, META);
850 }
851 
852 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
853 		unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
854 {
855 	struct f2fs_summary sum;
856 	set_summary(&sum, nid, 0, 0);
857 	do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, NODE);
858 }
859 
860 void write_data_page(struct inode *inode, struct page *page,
861 		struct dnode_of_data *dn, block_t old_blkaddr,
862 		block_t *new_blkaddr)
863 {
864 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
865 	struct f2fs_summary sum;
866 	struct node_info ni;
867 
868 	BUG_ON(old_blkaddr == NULL_ADDR);
869 	get_node_info(sbi, dn->nid, &ni);
870 	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
871 
872 	do_write_page(sbi, page, old_blkaddr,
873 			new_blkaddr, &sum, DATA);
874 }
875 
876 void rewrite_data_page(struct f2fs_sb_info *sbi, struct page *page,
877 					block_t old_blk_addr)
878 {
879 	submit_write_page(sbi, page, old_blk_addr, DATA);
880 }
881 
882 void recover_data_page(struct f2fs_sb_info *sbi,
883 			struct page *page, struct f2fs_summary *sum,
884 			block_t old_blkaddr, block_t new_blkaddr)
885 {
886 	struct sit_info *sit_i = SIT_I(sbi);
887 	struct curseg_info *curseg;
888 	unsigned int segno, old_cursegno;
889 	struct seg_entry *se;
890 	int type;
891 
892 	segno = GET_SEGNO(sbi, new_blkaddr);
893 	se = get_seg_entry(sbi, segno);
894 	type = se->type;
895 
896 	if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
897 		if (old_blkaddr == NULL_ADDR)
898 			type = CURSEG_COLD_DATA;
899 		else
900 			type = CURSEG_WARM_DATA;
901 	}
902 	curseg = CURSEG_I(sbi, type);
903 
904 	mutex_lock(&curseg->curseg_mutex);
905 	mutex_lock(&sit_i->sentry_lock);
906 
907 	old_cursegno = curseg->segno;
908 
909 	/* change the current segment */
910 	if (segno != curseg->segno) {
911 		curseg->next_segno = segno;
912 		change_curseg(sbi, type, true);
913 	}
914 
915 	curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
916 					(sbi->blocks_per_seg - 1);
917 	__add_sum_entry(sbi, type, sum);
918 
919 	refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
920 
921 	locate_dirty_segment(sbi, old_cursegno);
922 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
923 
924 	mutex_unlock(&sit_i->sentry_lock);
925 	mutex_unlock(&curseg->curseg_mutex);
926 }
927 
928 void rewrite_node_page(struct f2fs_sb_info *sbi,
929 			struct page *page, struct f2fs_summary *sum,
930 			block_t old_blkaddr, block_t new_blkaddr)
931 {
932 	struct sit_info *sit_i = SIT_I(sbi);
933 	int type = CURSEG_WARM_NODE;
934 	struct curseg_info *curseg;
935 	unsigned int segno, old_cursegno;
936 	block_t next_blkaddr = next_blkaddr_of_node(page);
937 	unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
938 
939 	curseg = CURSEG_I(sbi, type);
940 
941 	mutex_lock(&curseg->curseg_mutex);
942 	mutex_lock(&sit_i->sentry_lock);
943 
944 	segno = GET_SEGNO(sbi, new_blkaddr);
945 	old_cursegno = curseg->segno;
946 
947 	/* change the current segment */
948 	if (segno != curseg->segno) {
949 		curseg->next_segno = segno;
950 		change_curseg(sbi, type, true);
951 	}
952 	curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
953 					(sbi->blocks_per_seg - 1);
954 	__add_sum_entry(sbi, type, sum);
955 
956 	/* change the current log to the next block addr in advance */
957 	if (next_segno != segno) {
958 		curseg->next_segno = next_segno;
959 		change_curseg(sbi, type, true);
960 	}
961 	curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) &
962 					(sbi->blocks_per_seg - 1);
963 
964 	/* rewrite node page */
965 	set_page_writeback(page);
966 	submit_write_page(sbi, page, new_blkaddr, NODE);
967 	f2fs_submit_bio(sbi, NODE, true);
968 	refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
969 
970 	locate_dirty_segment(sbi, old_cursegno);
971 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
972 
973 	mutex_unlock(&sit_i->sentry_lock);
974 	mutex_unlock(&curseg->curseg_mutex);
975 }
976 
977 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
978 {
979 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
980 	struct curseg_info *seg_i;
981 	unsigned char *kaddr;
982 	struct page *page;
983 	block_t start;
984 	int i, j, offset;
985 
986 	start = start_sum_block(sbi);
987 
988 	page = get_meta_page(sbi, start++);
989 	kaddr = (unsigned char *)page_address(page);
990 
991 	/* Step 1: restore nat cache */
992 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
993 	memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
994 
995 	/* Step 2: restore sit cache */
996 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
997 	memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
998 						SUM_JOURNAL_SIZE);
999 	offset = 2 * SUM_JOURNAL_SIZE;
1000 
1001 	/* Step 3: restore summary entries */
1002 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1003 		unsigned short blk_off;
1004 		unsigned int segno;
1005 
1006 		seg_i = CURSEG_I(sbi, i);
1007 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1008 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1009 		seg_i->next_segno = segno;
1010 		reset_curseg(sbi, i, 0);
1011 		seg_i->alloc_type = ckpt->alloc_type[i];
1012 		seg_i->next_blkoff = blk_off;
1013 
1014 		if (seg_i->alloc_type == SSR)
1015 			blk_off = sbi->blocks_per_seg;
1016 
1017 		for (j = 0; j < blk_off; j++) {
1018 			struct f2fs_summary *s;
1019 			s = (struct f2fs_summary *)(kaddr + offset);
1020 			seg_i->sum_blk->entries[j] = *s;
1021 			offset += SUMMARY_SIZE;
1022 			if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1023 						SUM_FOOTER_SIZE)
1024 				continue;
1025 
1026 			f2fs_put_page(page, 1);
1027 			page = NULL;
1028 
1029 			page = get_meta_page(sbi, start++);
1030 			kaddr = (unsigned char *)page_address(page);
1031 			offset = 0;
1032 		}
1033 	}
1034 	f2fs_put_page(page, 1);
1035 	return 0;
1036 }
1037 
1038 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1039 {
1040 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1041 	struct f2fs_summary_block *sum;
1042 	struct curseg_info *curseg;
1043 	struct page *new;
1044 	unsigned short blk_off;
1045 	unsigned int segno = 0;
1046 	block_t blk_addr = 0;
1047 
1048 	/* get segment number and block addr */
1049 	if (IS_DATASEG(type)) {
1050 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1051 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1052 							CURSEG_HOT_DATA]);
1053 		if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1054 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1055 		else
1056 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1057 	} else {
1058 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
1059 							CURSEG_HOT_NODE]);
1060 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1061 							CURSEG_HOT_NODE]);
1062 		if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1063 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1064 							type - CURSEG_HOT_NODE);
1065 		else
1066 			blk_addr = GET_SUM_BLOCK(sbi, segno);
1067 	}
1068 
1069 	new = get_meta_page(sbi, blk_addr);
1070 	sum = (struct f2fs_summary_block *)page_address(new);
1071 
1072 	if (IS_NODESEG(type)) {
1073 		if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1074 			struct f2fs_summary *ns = &sum->entries[0];
1075 			int i;
1076 			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1077 				ns->version = 0;
1078 				ns->ofs_in_node = 0;
1079 			}
1080 		} else {
1081 			if (restore_node_summary(sbi, segno, sum)) {
1082 				f2fs_put_page(new, 1);
1083 				return -EINVAL;
1084 			}
1085 		}
1086 	}
1087 
1088 	/* set uncompleted segment to curseg */
1089 	curseg = CURSEG_I(sbi, type);
1090 	mutex_lock(&curseg->curseg_mutex);
1091 	memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1092 	curseg->next_segno = segno;
1093 	reset_curseg(sbi, type, 0);
1094 	curseg->alloc_type = ckpt->alloc_type[type];
1095 	curseg->next_blkoff = blk_off;
1096 	mutex_unlock(&curseg->curseg_mutex);
1097 	f2fs_put_page(new, 1);
1098 	return 0;
1099 }
1100 
1101 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1102 {
1103 	int type = CURSEG_HOT_DATA;
1104 
1105 	if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1106 		/* restore for compacted data summary */
1107 		if (read_compacted_summaries(sbi))
1108 			return -EINVAL;
1109 		type = CURSEG_HOT_NODE;
1110 	}
1111 
1112 	for (; type <= CURSEG_COLD_NODE; type++)
1113 		if (read_normal_summaries(sbi, type))
1114 			return -EINVAL;
1115 	return 0;
1116 }
1117 
1118 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1119 {
1120 	struct page *page;
1121 	unsigned char *kaddr;
1122 	struct f2fs_summary *summary;
1123 	struct curseg_info *seg_i;
1124 	int written_size = 0;
1125 	int i, j;
1126 
1127 	page = grab_meta_page(sbi, blkaddr++);
1128 	kaddr = (unsigned char *)page_address(page);
1129 
1130 	/* Step 1: write nat cache */
1131 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1132 	memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1133 	written_size += SUM_JOURNAL_SIZE;
1134 
1135 	/* Step 2: write sit cache */
1136 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1137 	memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1138 						SUM_JOURNAL_SIZE);
1139 	written_size += SUM_JOURNAL_SIZE;
1140 
1141 	set_page_dirty(page);
1142 
1143 	/* Step 3: write summary entries */
1144 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1145 		unsigned short blkoff;
1146 		seg_i = CURSEG_I(sbi, i);
1147 		if (sbi->ckpt->alloc_type[i] == SSR)
1148 			blkoff = sbi->blocks_per_seg;
1149 		else
1150 			blkoff = curseg_blkoff(sbi, i);
1151 
1152 		for (j = 0; j < blkoff; j++) {
1153 			if (!page) {
1154 				page = grab_meta_page(sbi, blkaddr++);
1155 				kaddr = (unsigned char *)page_address(page);
1156 				written_size = 0;
1157 			}
1158 			summary = (struct f2fs_summary *)(kaddr + written_size);
1159 			*summary = seg_i->sum_blk->entries[j];
1160 			written_size += SUMMARY_SIZE;
1161 			set_page_dirty(page);
1162 
1163 			if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1164 							SUM_FOOTER_SIZE)
1165 				continue;
1166 
1167 			f2fs_put_page(page, 1);
1168 			page = NULL;
1169 		}
1170 	}
1171 	if (page)
1172 		f2fs_put_page(page, 1);
1173 }
1174 
1175 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1176 					block_t blkaddr, int type)
1177 {
1178 	int i, end;
1179 	if (IS_DATASEG(type))
1180 		end = type + NR_CURSEG_DATA_TYPE;
1181 	else
1182 		end = type + NR_CURSEG_NODE_TYPE;
1183 
1184 	for (i = type; i < end; i++) {
1185 		struct curseg_info *sum = CURSEG_I(sbi, i);
1186 		mutex_lock(&sum->curseg_mutex);
1187 		write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1188 		mutex_unlock(&sum->curseg_mutex);
1189 	}
1190 }
1191 
1192 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1193 {
1194 	if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1195 		write_compacted_summaries(sbi, start_blk);
1196 	else
1197 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1198 }
1199 
1200 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1201 {
1202 	if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1203 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1204 }
1205 
1206 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1207 					unsigned int val, int alloc)
1208 {
1209 	int i;
1210 
1211 	if (type == NAT_JOURNAL) {
1212 		for (i = 0; i < nats_in_cursum(sum); i++) {
1213 			if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1214 				return i;
1215 		}
1216 		if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1217 			return update_nats_in_cursum(sum, 1);
1218 	} else if (type == SIT_JOURNAL) {
1219 		for (i = 0; i < sits_in_cursum(sum); i++)
1220 			if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1221 				return i;
1222 		if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1223 			return update_sits_in_cursum(sum, 1);
1224 	}
1225 	return -1;
1226 }
1227 
1228 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1229 					unsigned int segno)
1230 {
1231 	struct sit_info *sit_i = SIT_I(sbi);
1232 	unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1233 	block_t blk_addr = sit_i->sit_base_addr + offset;
1234 
1235 	check_seg_range(sbi, segno);
1236 
1237 	/* calculate sit block address */
1238 	if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1239 		blk_addr += sit_i->sit_blocks;
1240 
1241 	return get_meta_page(sbi, blk_addr);
1242 }
1243 
1244 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1245 					unsigned int start)
1246 {
1247 	struct sit_info *sit_i = SIT_I(sbi);
1248 	struct page *src_page, *dst_page;
1249 	pgoff_t src_off, dst_off;
1250 	void *src_addr, *dst_addr;
1251 
1252 	src_off = current_sit_addr(sbi, start);
1253 	dst_off = next_sit_addr(sbi, src_off);
1254 
1255 	/* get current sit block page without lock */
1256 	src_page = get_meta_page(sbi, src_off);
1257 	dst_page = grab_meta_page(sbi, dst_off);
1258 	BUG_ON(PageDirty(src_page));
1259 
1260 	src_addr = page_address(src_page);
1261 	dst_addr = page_address(dst_page);
1262 	memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1263 
1264 	set_page_dirty(dst_page);
1265 	f2fs_put_page(src_page, 1);
1266 
1267 	set_to_next_sit(sit_i, start);
1268 
1269 	return dst_page;
1270 }
1271 
1272 static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1273 {
1274 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1275 	struct f2fs_summary_block *sum = curseg->sum_blk;
1276 	int i;
1277 
1278 	/*
1279 	 * If the journal area in the current summary is full of sit entries,
1280 	 * all the sit entries will be flushed. Otherwise the sit entries
1281 	 * are not able to replace with newly hot sit entries.
1282 	 */
1283 	if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1284 		for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1285 			unsigned int segno;
1286 			segno = le32_to_cpu(segno_in_journal(sum, i));
1287 			__mark_sit_entry_dirty(sbi, segno);
1288 		}
1289 		update_sits_in_cursum(sum, -sits_in_cursum(sum));
1290 		return true;
1291 	}
1292 	return false;
1293 }
1294 
1295 /*
1296  * CP calls this function, which flushes SIT entries including sit_journal,
1297  * and moves prefree segs to free segs.
1298  */
1299 void flush_sit_entries(struct f2fs_sb_info *sbi)
1300 {
1301 	struct sit_info *sit_i = SIT_I(sbi);
1302 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1303 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1304 	struct f2fs_summary_block *sum = curseg->sum_blk;
1305 	unsigned long nsegs = TOTAL_SEGS(sbi);
1306 	struct page *page = NULL;
1307 	struct f2fs_sit_block *raw_sit = NULL;
1308 	unsigned int start = 0, end = 0;
1309 	unsigned int segno = -1;
1310 	bool flushed;
1311 
1312 	mutex_lock(&curseg->curseg_mutex);
1313 	mutex_lock(&sit_i->sentry_lock);
1314 
1315 	/*
1316 	 * "flushed" indicates whether sit entries in journal are flushed
1317 	 * to the SIT area or not.
1318 	 */
1319 	flushed = flush_sits_in_journal(sbi);
1320 
1321 	while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1322 		struct seg_entry *se = get_seg_entry(sbi, segno);
1323 		int sit_offset, offset;
1324 
1325 		sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1326 
1327 		if (flushed)
1328 			goto to_sit_page;
1329 
1330 		offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1331 		if (offset >= 0) {
1332 			segno_in_journal(sum, offset) = cpu_to_le32(segno);
1333 			seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1334 			goto flush_done;
1335 		}
1336 to_sit_page:
1337 		if (!page || (start > segno) || (segno > end)) {
1338 			if (page) {
1339 				f2fs_put_page(page, 1);
1340 				page = NULL;
1341 			}
1342 
1343 			start = START_SEGNO(sit_i, segno);
1344 			end = start + SIT_ENTRY_PER_BLOCK - 1;
1345 
1346 			/* read sit block that will be updated */
1347 			page = get_next_sit_page(sbi, start);
1348 			raw_sit = page_address(page);
1349 		}
1350 
1351 		/* udpate entry in SIT block */
1352 		seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1353 flush_done:
1354 		__clear_bit(segno, bitmap);
1355 		sit_i->dirty_sentries--;
1356 	}
1357 	mutex_unlock(&sit_i->sentry_lock);
1358 	mutex_unlock(&curseg->curseg_mutex);
1359 
1360 	/* writeout last modified SIT block */
1361 	f2fs_put_page(page, 1);
1362 
1363 	set_prefree_as_free_segments(sbi);
1364 }
1365 
1366 static int build_sit_info(struct f2fs_sb_info *sbi)
1367 {
1368 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1369 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1370 	struct sit_info *sit_i;
1371 	unsigned int sit_segs, start;
1372 	char *src_bitmap, *dst_bitmap;
1373 	unsigned int bitmap_size;
1374 
1375 	/* allocate memory for SIT information */
1376 	sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1377 	if (!sit_i)
1378 		return -ENOMEM;
1379 
1380 	SM_I(sbi)->sit_info = sit_i;
1381 
1382 	sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1383 	if (!sit_i->sentries)
1384 		return -ENOMEM;
1385 
1386 	bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1387 	sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1388 	if (!sit_i->dirty_sentries_bitmap)
1389 		return -ENOMEM;
1390 
1391 	for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1392 		sit_i->sentries[start].cur_valid_map
1393 			= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1394 		sit_i->sentries[start].ckpt_valid_map
1395 			= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1396 		if (!sit_i->sentries[start].cur_valid_map
1397 				|| !sit_i->sentries[start].ckpt_valid_map)
1398 			return -ENOMEM;
1399 	}
1400 
1401 	if (sbi->segs_per_sec > 1) {
1402 		sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
1403 					sizeof(struct sec_entry));
1404 		if (!sit_i->sec_entries)
1405 			return -ENOMEM;
1406 	}
1407 
1408 	/* get information related with SIT */
1409 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1410 
1411 	/* setup SIT bitmap from ckeckpoint pack */
1412 	bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1413 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1414 
1415 	dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1416 	if (!dst_bitmap)
1417 		return -ENOMEM;
1418 
1419 	/* init SIT information */
1420 	sit_i->s_ops = &default_salloc_ops;
1421 
1422 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1423 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1424 	sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1425 	sit_i->sit_bitmap = dst_bitmap;
1426 	sit_i->bitmap_size = bitmap_size;
1427 	sit_i->dirty_sentries = 0;
1428 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1429 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1430 	sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1431 	mutex_init(&sit_i->sentry_lock);
1432 	return 0;
1433 }
1434 
1435 static int build_free_segmap(struct f2fs_sb_info *sbi)
1436 {
1437 	struct f2fs_sm_info *sm_info = SM_I(sbi);
1438 	struct free_segmap_info *free_i;
1439 	unsigned int bitmap_size, sec_bitmap_size;
1440 
1441 	/* allocate memory for free segmap information */
1442 	free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1443 	if (!free_i)
1444 		return -ENOMEM;
1445 
1446 	SM_I(sbi)->free_info = free_i;
1447 
1448 	bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1449 	free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1450 	if (!free_i->free_segmap)
1451 		return -ENOMEM;
1452 
1453 	sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1454 	free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1455 	if (!free_i->free_secmap)
1456 		return -ENOMEM;
1457 
1458 	/* set all segments as dirty temporarily */
1459 	memset(free_i->free_segmap, 0xff, bitmap_size);
1460 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1461 
1462 	/* init free segmap information */
1463 	free_i->start_segno =
1464 		(unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1465 	free_i->free_segments = 0;
1466 	free_i->free_sections = 0;
1467 	rwlock_init(&free_i->segmap_lock);
1468 	return 0;
1469 }
1470 
1471 static int build_curseg(struct f2fs_sb_info *sbi)
1472 {
1473 	struct curseg_info *array;
1474 	int i;
1475 
1476 	array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
1477 	if (!array)
1478 		return -ENOMEM;
1479 
1480 	SM_I(sbi)->curseg_array = array;
1481 
1482 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
1483 		mutex_init(&array[i].curseg_mutex);
1484 		array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1485 		if (!array[i].sum_blk)
1486 			return -ENOMEM;
1487 		array[i].segno = NULL_SEGNO;
1488 		array[i].next_blkoff = 0;
1489 	}
1490 	return restore_curseg_summaries(sbi);
1491 }
1492 
1493 static void build_sit_entries(struct f2fs_sb_info *sbi)
1494 {
1495 	struct sit_info *sit_i = SIT_I(sbi);
1496 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1497 	struct f2fs_summary_block *sum = curseg->sum_blk;
1498 	unsigned int start;
1499 
1500 	for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1501 		struct seg_entry *se = &sit_i->sentries[start];
1502 		struct f2fs_sit_block *sit_blk;
1503 		struct f2fs_sit_entry sit;
1504 		struct page *page;
1505 		int i;
1506 
1507 		mutex_lock(&curseg->curseg_mutex);
1508 		for (i = 0; i < sits_in_cursum(sum); i++) {
1509 			if (le32_to_cpu(segno_in_journal(sum, i)) == start) {
1510 				sit = sit_in_journal(sum, i);
1511 				mutex_unlock(&curseg->curseg_mutex);
1512 				goto got_it;
1513 			}
1514 		}
1515 		mutex_unlock(&curseg->curseg_mutex);
1516 		page = get_current_sit_page(sbi, start);
1517 		sit_blk = (struct f2fs_sit_block *)page_address(page);
1518 		sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1519 		f2fs_put_page(page, 1);
1520 got_it:
1521 		check_block_count(sbi, start, &sit);
1522 		seg_info_from_raw_sit(se, &sit);
1523 		if (sbi->segs_per_sec > 1) {
1524 			struct sec_entry *e = get_sec_entry(sbi, start);
1525 			e->valid_blocks += se->valid_blocks;
1526 		}
1527 	}
1528 }
1529 
1530 static void init_free_segmap(struct f2fs_sb_info *sbi)
1531 {
1532 	unsigned int start;
1533 	int type;
1534 
1535 	for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1536 		struct seg_entry *sentry = get_seg_entry(sbi, start);
1537 		if (!sentry->valid_blocks)
1538 			__set_free(sbi, start);
1539 	}
1540 
1541 	/* set use the current segments */
1542 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1543 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1544 		__set_test_and_inuse(sbi, curseg_t->segno);
1545 	}
1546 }
1547 
1548 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1549 {
1550 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1551 	struct free_segmap_info *free_i = FREE_I(sbi);
1552 	unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi);
1553 	unsigned short valid_blocks;
1554 
1555 	while (1) {
1556 		/* find dirty segment based on free segmap */
1557 		segno = find_next_inuse(free_i, total_segs, offset);
1558 		if (segno >= total_segs)
1559 			break;
1560 		offset = segno + 1;
1561 		valid_blocks = get_valid_blocks(sbi, segno, 0);
1562 		if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1563 			continue;
1564 		mutex_lock(&dirty_i->seglist_lock);
1565 		__locate_dirty_segment(sbi, segno, DIRTY);
1566 		mutex_unlock(&dirty_i->seglist_lock);
1567 	}
1568 }
1569 
1570 static int init_victim_secmap(struct f2fs_sb_info *sbi)
1571 {
1572 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1573 	unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1574 
1575 	dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1576 	if (!dirty_i->victim_secmap)
1577 		return -ENOMEM;
1578 	return 0;
1579 }
1580 
1581 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1582 {
1583 	struct dirty_seglist_info *dirty_i;
1584 	unsigned int bitmap_size, i;
1585 
1586 	/* allocate memory for dirty segments list information */
1587 	dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1588 	if (!dirty_i)
1589 		return -ENOMEM;
1590 
1591 	SM_I(sbi)->dirty_info = dirty_i;
1592 	mutex_init(&dirty_i->seglist_lock);
1593 
1594 	bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1595 
1596 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
1597 		dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1598 		if (!dirty_i->dirty_segmap[i])
1599 			return -ENOMEM;
1600 	}
1601 
1602 	init_dirty_segmap(sbi);
1603 	return init_victim_secmap(sbi);
1604 }
1605 
1606 /*
1607  * Update min, max modified time for cost-benefit GC algorithm
1608  */
1609 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1610 {
1611 	struct sit_info *sit_i = SIT_I(sbi);
1612 	unsigned int segno;
1613 
1614 	mutex_lock(&sit_i->sentry_lock);
1615 
1616 	sit_i->min_mtime = LLONG_MAX;
1617 
1618 	for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1619 		unsigned int i;
1620 		unsigned long long mtime = 0;
1621 
1622 		for (i = 0; i < sbi->segs_per_sec; i++)
1623 			mtime += get_seg_entry(sbi, segno + i)->mtime;
1624 
1625 		mtime = div_u64(mtime, sbi->segs_per_sec);
1626 
1627 		if (sit_i->min_mtime > mtime)
1628 			sit_i->min_mtime = mtime;
1629 	}
1630 	sit_i->max_mtime = get_mtime(sbi);
1631 	mutex_unlock(&sit_i->sentry_lock);
1632 }
1633 
1634 int build_segment_manager(struct f2fs_sb_info *sbi)
1635 {
1636 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1637 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1638 	struct f2fs_sm_info *sm_info;
1639 	int err;
1640 
1641 	sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1642 	if (!sm_info)
1643 		return -ENOMEM;
1644 
1645 	/* init sm info */
1646 	sbi->sm_info = sm_info;
1647 	INIT_LIST_HEAD(&sm_info->wblist_head);
1648 	spin_lock_init(&sm_info->wblist_lock);
1649 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1650 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1651 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1652 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1653 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1654 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1655 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1656 	sm_info->rec_prefree_segments = DEF_RECLAIM_PREFREE_SEGMENTS;
1657 
1658 	err = build_sit_info(sbi);
1659 	if (err)
1660 		return err;
1661 	err = build_free_segmap(sbi);
1662 	if (err)
1663 		return err;
1664 	err = build_curseg(sbi);
1665 	if (err)
1666 		return err;
1667 
1668 	/* reinit free segmap based on SIT */
1669 	build_sit_entries(sbi);
1670 
1671 	init_free_segmap(sbi);
1672 	err = build_dirty_segmap(sbi);
1673 	if (err)
1674 		return err;
1675 
1676 	init_min_max_mtime(sbi);
1677 	return 0;
1678 }
1679 
1680 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1681 		enum dirty_type dirty_type)
1682 {
1683 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1684 
1685 	mutex_lock(&dirty_i->seglist_lock);
1686 	kfree(dirty_i->dirty_segmap[dirty_type]);
1687 	dirty_i->nr_dirty[dirty_type] = 0;
1688 	mutex_unlock(&dirty_i->seglist_lock);
1689 }
1690 
1691 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
1692 {
1693 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1694 	kfree(dirty_i->victim_secmap);
1695 }
1696 
1697 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1698 {
1699 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1700 	int i;
1701 
1702 	if (!dirty_i)
1703 		return;
1704 
1705 	/* discard pre-free/dirty segments list */
1706 	for (i = 0; i < NR_DIRTY_TYPE; i++)
1707 		discard_dirty_segmap(sbi, i);
1708 
1709 	destroy_victim_secmap(sbi);
1710 	SM_I(sbi)->dirty_info = NULL;
1711 	kfree(dirty_i);
1712 }
1713 
1714 static void destroy_curseg(struct f2fs_sb_info *sbi)
1715 {
1716 	struct curseg_info *array = SM_I(sbi)->curseg_array;
1717 	int i;
1718 
1719 	if (!array)
1720 		return;
1721 	SM_I(sbi)->curseg_array = NULL;
1722 	for (i = 0; i < NR_CURSEG_TYPE; i++)
1723 		kfree(array[i].sum_blk);
1724 	kfree(array);
1725 }
1726 
1727 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1728 {
1729 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
1730 	if (!free_i)
1731 		return;
1732 	SM_I(sbi)->free_info = NULL;
1733 	kfree(free_i->free_segmap);
1734 	kfree(free_i->free_secmap);
1735 	kfree(free_i);
1736 }
1737 
1738 static void destroy_sit_info(struct f2fs_sb_info *sbi)
1739 {
1740 	struct sit_info *sit_i = SIT_I(sbi);
1741 	unsigned int start;
1742 
1743 	if (!sit_i)
1744 		return;
1745 
1746 	if (sit_i->sentries) {
1747 		for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1748 			kfree(sit_i->sentries[start].cur_valid_map);
1749 			kfree(sit_i->sentries[start].ckpt_valid_map);
1750 		}
1751 	}
1752 	vfree(sit_i->sentries);
1753 	vfree(sit_i->sec_entries);
1754 	kfree(sit_i->dirty_sentries_bitmap);
1755 
1756 	SM_I(sbi)->sit_info = NULL;
1757 	kfree(sit_i->sit_bitmap);
1758 	kfree(sit_i);
1759 }
1760 
1761 void destroy_segment_manager(struct f2fs_sb_info *sbi)
1762 {
1763 	struct f2fs_sm_info *sm_info = SM_I(sbi);
1764 	destroy_dirty_segmap(sbi);
1765 	destroy_curseg(sbi);
1766 	destroy_free_segmap(sbi);
1767 	destroy_sit_info(sbi);
1768 	sbi->sm_info = NULL;
1769 	kfree(sm_info);
1770 }
1771