xref: /openbmc/linux/fs/f2fs/segment.c (revision 81eb8d6e2869b119d4a7b8c02091c3779733a3ac)
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 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
40 		enum dirty_type dirty_type)
41 {
42 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
43 
44 	/* need not be added */
45 	if (IS_CURSEG(sbi, segno))
46 		return;
47 
48 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
49 		dirty_i->nr_dirty[dirty_type]++;
50 
51 	if (dirty_type == DIRTY) {
52 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
53 		enum dirty_type t = DIRTY_HOT_DATA;
54 
55 		dirty_type = sentry->type;
56 
57 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
58 			dirty_i->nr_dirty[dirty_type]++;
59 
60 		/* Only one bitmap should be set */
61 		for (; t <= DIRTY_COLD_NODE; t++) {
62 			if (t == dirty_type)
63 				continue;
64 			if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
65 				dirty_i->nr_dirty[t]--;
66 		}
67 	}
68 }
69 
70 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
71 		enum dirty_type dirty_type)
72 {
73 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
74 
75 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
76 		dirty_i->nr_dirty[dirty_type]--;
77 
78 	if (dirty_type == DIRTY) {
79 		enum dirty_type t = DIRTY_HOT_DATA;
80 
81 		/* clear its dirty bitmap */
82 		for (; t <= DIRTY_COLD_NODE; t++) {
83 			if (test_and_clear_bit(segno,
84 						dirty_i->dirty_segmap[t])) {
85 				dirty_i->nr_dirty[t]--;
86 				break;
87 			}
88 		}
89 
90 		if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
91 			clear_bit(GET_SECNO(sbi, segno),
92 						dirty_i->victim_secmap);
93 	}
94 }
95 
96 /*
97  * Should not occur error such as -ENOMEM.
98  * Adding dirty entry into seglist is not critical operation.
99  * If a given segment is one of current working segments, it won't be added.
100  */
101 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
102 {
103 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
104 	unsigned short valid_blocks;
105 
106 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
107 		return;
108 
109 	mutex_lock(&dirty_i->seglist_lock);
110 
111 	valid_blocks = get_valid_blocks(sbi, segno, 0);
112 
113 	if (valid_blocks == 0) {
114 		__locate_dirty_segment(sbi, segno, PRE);
115 		__remove_dirty_segment(sbi, segno, DIRTY);
116 	} else if (valid_blocks < sbi->blocks_per_seg) {
117 		__locate_dirty_segment(sbi, segno, DIRTY);
118 	} else {
119 		/* Recovery routine with SSR needs this */
120 		__remove_dirty_segment(sbi, segno, DIRTY);
121 	}
122 
123 	mutex_unlock(&dirty_i->seglist_lock);
124 }
125 
126 /*
127  * Should call clear_prefree_segments after checkpoint is done.
128  */
129 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
130 {
131 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
132 	unsigned int segno = -1;
133 	unsigned int total_segs = TOTAL_SEGS(sbi);
134 
135 	mutex_lock(&dirty_i->seglist_lock);
136 	while (1) {
137 		segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
138 				segno + 1);
139 		if (segno >= total_segs)
140 			break;
141 		__set_test_and_free(sbi, segno);
142 	}
143 	mutex_unlock(&dirty_i->seglist_lock);
144 }
145 
146 void clear_prefree_segments(struct f2fs_sb_info *sbi)
147 {
148 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
149 	unsigned int segno = -1;
150 	unsigned int total_segs = TOTAL_SEGS(sbi);
151 
152 	mutex_lock(&dirty_i->seglist_lock);
153 	while (1) {
154 		segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
155 				segno + 1);
156 		if (segno >= total_segs)
157 			break;
158 
159 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[PRE]))
160 			dirty_i->nr_dirty[PRE]--;
161 
162 		/* Let's use trim */
163 		if (test_opt(sbi, DISCARD))
164 			blkdev_issue_discard(sbi->sb->s_bdev,
165 					START_BLOCK(sbi, segno) <<
166 					sbi->log_sectors_per_block,
167 					1 << (sbi->log_sectors_per_block +
168 						sbi->log_blocks_per_seg),
169 					GFP_NOFS, 0);
170 	}
171 	mutex_unlock(&dirty_i->seglist_lock);
172 }
173 
174 static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
175 {
176 	struct sit_info *sit_i = SIT_I(sbi);
177 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
178 		sit_i->dirty_sentries++;
179 }
180 
181 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
182 					unsigned int segno, int modified)
183 {
184 	struct seg_entry *se = get_seg_entry(sbi, segno);
185 	se->type = type;
186 	if (modified)
187 		__mark_sit_entry_dirty(sbi, segno);
188 }
189 
190 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
191 {
192 	struct seg_entry *se;
193 	unsigned int segno, offset;
194 	long int new_vblocks;
195 
196 	segno = GET_SEGNO(sbi, blkaddr);
197 
198 	se = get_seg_entry(sbi, segno);
199 	new_vblocks = se->valid_blocks + del;
200 	offset = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) & (sbi->blocks_per_seg - 1);
201 
202 	BUG_ON((new_vblocks >> (sizeof(unsigned short) << 3) ||
203 				(new_vblocks > sbi->blocks_per_seg)));
204 
205 	se->valid_blocks = new_vblocks;
206 	se->mtime = get_mtime(sbi);
207 	SIT_I(sbi)->max_mtime = se->mtime;
208 
209 	/* Update valid block bitmap */
210 	if (del > 0) {
211 		if (f2fs_set_bit(offset, se->cur_valid_map))
212 			BUG();
213 	} else {
214 		if (!f2fs_clear_bit(offset, se->cur_valid_map))
215 			BUG();
216 	}
217 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
218 		se->ckpt_valid_blocks += del;
219 
220 	__mark_sit_entry_dirty(sbi, segno);
221 
222 	/* update total number of valid blocks to be written in ckpt area */
223 	SIT_I(sbi)->written_valid_blocks += del;
224 
225 	if (sbi->segs_per_sec > 1)
226 		get_sec_entry(sbi, segno)->valid_blocks += del;
227 }
228 
229 static void refresh_sit_entry(struct f2fs_sb_info *sbi,
230 			block_t old_blkaddr, block_t new_blkaddr)
231 {
232 	update_sit_entry(sbi, new_blkaddr, 1);
233 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
234 		update_sit_entry(sbi, old_blkaddr, -1);
235 }
236 
237 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
238 {
239 	unsigned int segno = GET_SEGNO(sbi, addr);
240 	struct sit_info *sit_i = SIT_I(sbi);
241 
242 	BUG_ON(addr == NULL_ADDR);
243 	if (addr == NEW_ADDR)
244 		return;
245 
246 	/* add it into sit main buffer */
247 	mutex_lock(&sit_i->sentry_lock);
248 
249 	update_sit_entry(sbi, addr, -1);
250 
251 	/* add it into dirty seglist */
252 	locate_dirty_segment(sbi, segno);
253 
254 	mutex_unlock(&sit_i->sentry_lock);
255 }
256 
257 /*
258  * This function should be resided under the curseg_mutex lock
259  */
260 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
261 					struct f2fs_summary *sum)
262 {
263 	struct curseg_info *curseg = CURSEG_I(sbi, type);
264 	void *addr = curseg->sum_blk;
265 	addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
266 	memcpy(addr, sum, sizeof(struct f2fs_summary));
267 }
268 
269 /*
270  * Calculate the number of current summary pages for writing
271  */
272 int npages_for_summary_flush(struct f2fs_sb_info *sbi)
273 {
274 	int total_size_bytes = 0;
275 	int valid_sum_count = 0;
276 	int i, sum_space;
277 
278 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
279 		if (sbi->ckpt->alloc_type[i] == SSR)
280 			valid_sum_count += sbi->blocks_per_seg;
281 		else
282 			valid_sum_count += curseg_blkoff(sbi, i);
283 	}
284 
285 	total_size_bytes = valid_sum_count * (SUMMARY_SIZE + 1)
286 			+ sizeof(struct nat_journal) + 2
287 			+ sizeof(struct sit_journal) + 2;
288 	sum_space = PAGE_CACHE_SIZE - SUM_FOOTER_SIZE;
289 	if (total_size_bytes < sum_space)
290 		return 1;
291 	else if (total_size_bytes < 2 * sum_space)
292 		return 2;
293 	return 3;
294 }
295 
296 /*
297  * Caller should put this summary page
298  */
299 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
300 {
301 	return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
302 }
303 
304 static void write_sum_page(struct f2fs_sb_info *sbi,
305 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
306 {
307 	struct page *page = grab_meta_page(sbi, blk_addr);
308 	void *kaddr = page_address(page);
309 	memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
310 	set_page_dirty(page);
311 	f2fs_put_page(page, 1);
312 }
313 
314 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
315 {
316 	struct curseg_info *curseg = CURSEG_I(sbi, type);
317 	unsigned int segno = curseg->segno + 1;
318 	struct free_segmap_info *free_i = FREE_I(sbi);
319 
320 	if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec)
321 		return !test_bit(segno, free_i->free_segmap);
322 	return 0;
323 }
324 
325 /*
326  * Find a new segment from the free segments bitmap to right order
327  * This function should be returned with success, otherwise BUG
328  */
329 static void get_new_segment(struct f2fs_sb_info *sbi,
330 			unsigned int *newseg, bool new_sec, int dir)
331 {
332 	struct free_segmap_info *free_i = FREE_I(sbi);
333 	unsigned int segno, secno, zoneno;
334 	unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone;
335 	unsigned int hint = *newseg / sbi->segs_per_sec;
336 	unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
337 	unsigned int left_start = hint;
338 	bool init = true;
339 	int go_left = 0;
340 	int i;
341 
342 	write_lock(&free_i->segmap_lock);
343 
344 	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
345 		segno = find_next_zero_bit(free_i->free_segmap,
346 					TOTAL_SEGS(sbi), *newseg + 1);
347 		if (segno - *newseg < sbi->segs_per_sec -
348 					(*newseg % sbi->segs_per_sec))
349 			goto got_it;
350 	}
351 find_other_zone:
352 	secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint);
353 	if (secno >= TOTAL_SECS(sbi)) {
354 		if (dir == ALLOC_RIGHT) {
355 			secno = find_next_zero_bit(free_i->free_secmap,
356 							TOTAL_SECS(sbi), 0);
357 			BUG_ON(secno >= TOTAL_SECS(sbi));
358 		} else {
359 			go_left = 1;
360 			left_start = hint - 1;
361 		}
362 	}
363 	if (go_left == 0)
364 		goto skip_left;
365 
366 	while (test_bit(left_start, free_i->free_secmap)) {
367 		if (left_start > 0) {
368 			left_start--;
369 			continue;
370 		}
371 		left_start = find_next_zero_bit(free_i->free_secmap,
372 							TOTAL_SECS(sbi), 0);
373 		BUG_ON(left_start >= TOTAL_SECS(sbi));
374 		break;
375 	}
376 	secno = left_start;
377 skip_left:
378 	hint = secno;
379 	segno = secno * sbi->segs_per_sec;
380 	zoneno = secno / sbi->secs_per_zone;
381 
382 	/* give up on finding another zone */
383 	if (!init)
384 		goto got_it;
385 	if (sbi->secs_per_zone == 1)
386 		goto got_it;
387 	if (zoneno == old_zoneno)
388 		goto got_it;
389 	if (dir == ALLOC_LEFT) {
390 		if (!go_left && zoneno + 1 >= total_zones)
391 			goto got_it;
392 		if (go_left && zoneno == 0)
393 			goto got_it;
394 	}
395 	for (i = 0; i < NR_CURSEG_TYPE; i++)
396 		if (CURSEG_I(sbi, i)->zone == zoneno)
397 			break;
398 
399 	if (i < NR_CURSEG_TYPE) {
400 		/* zone is in user, try another */
401 		if (go_left)
402 			hint = zoneno * sbi->secs_per_zone - 1;
403 		else if (zoneno + 1 >= total_zones)
404 			hint = 0;
405 		else
406 			hint = (zoneno + 1) * sbi->secs_per_zone;
407 		init = false;
408 		goto find_other_zone;
409 	}
410 got_it:
411 	/* set it as dirty segment in free segmap */
412 	BUG_ON(test_bit(segno, free_i->free_segmap));
413 	__set_inuse(sbi, segno);
414 	*newseg = segno;
415 	write_unlock(&free_i->segmap_lock);
416 }
417 
418 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
419 {
420 	struct curseg_info *curseg = CURSEG_I(sbi, type);
421 	struct summary_footer *sum_footer;
422 
423 	curseg->segno = curseg->next_segno;
424 	curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
425 	curseg->next_blkoff = 0;
426 	curseg->next_segno = NULL_SEGNO;
427 
428 	sum_footer = &(curseg->sum_blk->footer);
429 	memset(sum_footer, 0, sizeof(struct summary_footer));
430 	if (IS_DATASEG(type))
431 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
432 	if (IS_NODESEG(type))
433 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
434 	__set_sit_entry_type(sbi, type, curseg->segno, modified);
435 }
436 
437 /*
438  * Allocate a current working segment.
439  * This function always allocates a free segment in LFS manner.
440  */
441 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
442 {
443 	struct curseg_info *curseg = CURSEG_I(sbi, type);
444 	unsigned int segno = curseg->segno;
445 	int dir = ALLOC_LEFT;
446 
447 	write_sum_page(sbi, curseg->sum_blk,
448 				GET_SUM_BLOCK(sbi, segno));
449 	if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
450 		dir = ALLOC_RIGHT;
451 
452 	if (test_opt(sbi, NOHEAP))
453 		dir = ALLOC_RIGHT;
454 
455 	get_new_segment(sbi, &segno, new_sec, dir);
456 	curseg->next_segno = segno;
457 	reset_curseg(sbi, type, 1);
458 	curseg->alloc_type = LFS;
459 }
460 
461 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
462 			struct curseg_info *seg, block_t start)
463 {
464 	struct seg_entry *se = get_seg_entry(sbi, seg->segno);
465 	block_t ofs;
466 	for (ofs = start; ofs < sbi->blocks_per_seg; ofs++) {
467 		if (!f2fs_test_bit(ofs, se->ckpt_valid_map)
468 			&& !f2fs_test_bit(ofs, se->cur_valid_map))
469 			break;
470 	}
471 	seg->next_blkoff = ofs;
472 }
473 
474 /*
475  * If a segment is written by LFS manner, next block offset is just obtained
476  * by increasing the current block offset. However, if a segment is written by
477  * SSR manner, next block offset obtained by calling __next_free_blkoff
478  */
479 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
480 				struct curseg_info *seg)
481 {
482 	if (seg->alloc_type == SSR)
483 		__next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
484 	else
485 		seg->next_blkoff++;
486 }
487 
488 /*
489  * This function always allocates a used segment (from dirty seglist) by SSR
490  * manner, so it should recover the existing segment information of valid blocks
491  */
492 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
493 {
494 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
495 	struct curseg_info *curseg = CURSEG_I(sbi, type);
496 	unsigned int new_segno = curseg->next_segno;
497 	struct f2fs_summary_block *sum_node;
498 	struct page *sum_page;
499 
500 	write_sum_page(sbi, curseg->sum_blk,
501 				GET_SUM_BLOCK(sbi, curseg->segno));
502 	__set_test_and_inuse(sbi, new_segno);
503 
504 	mutex_lock(&dirty_i->seglist_lock);
505 	__remove_dirty_segment(sbi, new_segno, PRE);
506 	__remove_dirty_segment(sbi, new_segno, DIRTY);
507 	mutex_unlock(&dirty_i->seglist_lock);
508 
509 	reset_curseg(sbi, type, 1);
510 	curseg->alloc_type = SSR;
511 	__next_free_blkoff(sbi, curseg, 0);
512 
513 	if (reuse) {
514 		sum_page = get_sum_page(sbi, new_segno);
515 		sum_node = (struct f2fs_summary_block *)page_address(sum_page);
516 		memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
517 		f2fs_put_page(sum_page, 1);
518 	}
519 }
520 
521 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
522 {
523 	struct curseg_info *curseg = CURSEG_I(sbi, type);
524 	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
525 
526 	if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
527 		return v_ops->get_victim(sbi,
528 				&(curseg)->next_segno, BG_GC, type, SSR);
529 
530 	/* For data segments, let's do SSR more intensively */
531 	for (; type >= CURSEG_HOT_DATA; type--)
532 		if (v_ops->get_victim(sbi, &(curseg)->next_segno,
533 						BG_GC, type, SSR))
534 			return 1;
535 	return 0;
536 }
537 
538 /*
539  * flush out current segment and replace it with new segment
540  * This function should be returned with success, otherwise BUG
541  */
542 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
543 						int type, bool force)
544 {
545 	struct curseg_info *curseg = CURSEG_I(sbi, type);
546 
547 	if (force)
548 		new_curseg(sbi, type, true);
549 	else if (type == CURSEG_WARM_NODE)
550 		new_curseg(sbi, type, false);
551 	else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
552 		new_curseg(sbi, type, false);
553 	else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
554 		change_curseg(sbi, type, true);
555 	else
556 		new_curseg(sbi, type, false);
557 
558 	stat_inc_seg_type(sbi, curseg);
559 }
560 
561 void allocate_new_segments(struct f2fs_sb_info *sbi)
562 {
563 	struct curseg_info *curseg;
564 	unsigned int old_curseg;
565 	int i;
566 
567 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
568 		curseg = CURSEG_I(sbi, i);
569 		old_curseg = curseg->segno;
570 		SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
571 		locate_dirty_segment(sbi, old_curseg);
572 	}
573 }
574 
575 static const struct segment_allocation default_salloc_ops = {
576 	.allocate_segment = allocate_segment_by_default,
577 };
578 
579 static void f2fs_end_io_write(struct bio *bio, int err)
580 {
581 	const int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
582 	struct bio_vec *bvec = bio->bi_io_vec + bio->bi_vcnt - 1;
583 	struct bio_private *p = bio->bi_private;
584 
585 	do {
586 		struct page *page = bvec->bv_page;
587 
588 		if (--bvec >= bio->bi_io_vec)
589 			prefetchw(&bvec->bv_page->flags);
590 		if (!uptodate) {
591 			SetPageError(page);
592 			if (page->mapping)
593 				set_bit(AS_EIO, &page->mapping->flags);
594 			set_ckpt_flags(p->sbi->ckpt, CP_ERROR_FLAG);
595 			p->sbi->sb->s_flags |= MS_RDONLY;
596 		}
597 		end_page_writeback(page);
598 		dec_page_count(p->sbi, F2FS_WRITEBACK);
599 	} while (bvec >= bio->bi_io_vec);
600 
601 	if (p->is_sync)
602 		complete(p->wait);
603 
604 	if (!get_pages(p->sbi, F2FS_WRITEBACK) && p->sbi->cp_task)
605 		wake_up_process(p->sbi->cp_task);
606 
607 	kfree(p);
608 	bio_put(bio);
609 }
610 
611 struct bio *f2fs_bio_alloc(struct block_device *bdev, int npages)
612 {
613 	struct bio *bio;
614 
615 	/* No failure on bio allocation */
616 	bio = bio_alloc(GFP_NOIO, npages);
617 	bio->bi_bdev = bdev;
618 	bio->bi_private = NULL;
619 
620 	return bio;
621 }
622 
623 static void do_submit_bio(struct f2fs_sb_info *sbi,
624 				enum page_type type, bool sync)
625 {
626 	int rw = sync ? WRITE_SYNC : WRITE;
627 	enum page_type btype = type > META ? META : type;
628 
629 	if (type >= META_FLUSH)
630 		rw = WRITE_FLUSH_FUA;
631 
632 	if (btype == META)
633 		rw |= REQ_META;
634 
635 	if (sbi->bio[btype]) {
636 		struct bio_private *p = sbi->bio[btype]->bi_private;
637 		p->sbi = sbi;
638 		sbi->bio[btype]->bi_end_io = f2fs_end_io_write;
639 
640 		trace_f2fs_do_submit_bio(sbi->sb, btype, sync, sbi->bio[btype]);
641 
642 		if (type == META_FLUSH) {
643 			DECLARE_COMPLETION_ONSTACK(wait);
644 			p->is_sync = true;
645 			p->wait = &wait;
646 			submit_bio(rw, sbi->bio[btype]);
647 			wait_for_completion(&wait);
648 		} else {
649 			p->is_sync = false;
650 			submit_bio(rw, sbi->bio[btype]);
651 		}
652 		sbi->bio[btype] = NULL;
653 	}
654 }
655 
656 void f2fs_submit_bio(struct f2fs_sb_info *sbi, enum page_type type, bool sync)
657 {
658 	down_write(&sbi->bio_sem);
659 	do_submit_bio(sbi, type, sync);
660 	up_write(&sbi->bio_sem);
661 }
662 
663 static void submit_write_page(struct f2fs_sb_info *sbi, struct page *page,
664 				block_t blk_addr, enum page_type type)
665 {
666 	struct block_device *bdev = sbi->sb->s_bdev;
667 	int bio_blocks;
668 
669 	verify_block_addr(sbi, blk_addr);
670 
671 	down_write(&sbi->bio_sem);
672 
673 	inc_page_count(sbi, F2FS_WRITEBACK);
674 
675 	if (sbi->bio[type] && sbi->last_block_in_bio[type] != blk_addr - 1)
676 		do_submit_bio(sbi, type, false);
677 alloc_new:
678 	if (sbi->bio[type] == NULL) {
679 		struct bio_private *priv;
680 retry:
681 		priv = kmalloc(sizeof(struct bio_private), GFP_NOFS);
682 		if (!priv) {
683 			cond_resched();
684 			goto retry;
685 		}
686 
687 		bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
688 		sbi->bio[type] = f2fs_bio_alloc(bdev, bio_blocks);
689 		sbi->bio[type]->bi_sector = SECTOR_FROM_BLOCK(sbi, blk_addr);
690 		sbi->bio[type]->bi_private = priv;
691 		/*
692 		 * The end_io will be assigned at the sumbission phase.
693 		 * Until then, let bio_add_page() merge consecutive IOs as much
694 		 * as possible.
695 		 */
696 	}
697 
698 	if (bio_add_page(sbi->bio[type], page, PAGE_CACHE_SIZE, 0) <
699 							PAGE_CACHE_SIZE) {
700 		do_submit_bio(sbi, type, false);
701 		goto alloc_new;
702 	}
703 
704 	sbi->last_block_in_bio[type] = blk_addr;
705 
706 	up_write(&sbi->bio_sem);
707 	trace_f2fs_submit_write_page(page, blk_addr, type);
708 }
709 
710 void f2fs_wait_on_page_writeback(struct page *page,
711 				enum page_type type, bool sync)
712 {
713 	struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
714 	if (PageWriteback(page)) {
715 		f2fs_submit_bio(sbi, type, sync);
716 		wait_on_page_writeback(page);
717 	}
718 }
719 
720 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
721 {
722 	struct curseg_info *curseg = CURSEG_I(sbi, type);
723 	if (curseg->next_blkoff < sbi->blocks_per_seg)
724 		return true;
725 	return false;
726 }
727 
728 static int __get_segment_type_2(struct page *page, enum page_type p_type)
729 {
730 	if (p_type == DATA)
731 		return CURSEG_HOT_DATA;
732 	else
733 		return CURSEG_HOT_NODE;
734 }
735 
736 static int __get_segment_type_4(struct page *page, enum page_type p_type)
737 {
738 	if (p_type == DATA) {
739 		struct inode *inode = page->mapping->host;
740 
741 		if (S_ISDIR(inode->i_mode))
742 			return CURSEG_HOT_DATA;
743 		else
744 			return CURSEG_COLD_DATA;
745 	} else {
746 		if (IS_DNODE(page) && !is_cold_node(page))
747 			return CURSEG_HOT_NODE;
748 		else
749 			return CURSEG_COLD_NODE;
750 	}
751 }
752 
753 static int __get_segment_type_6(struct page *page, enum page_type p_type)
754 {
755 	if (p_type == DATA) {
756 		struct inode *inode = page->mapping->host;
757 
758 		if (S_ISDIR(inode->i_mode))
759 			return CURSEG_HOT_DATA;
760 		else if (is_cold_data(page) || file_is_cold(inode))
761 			return CURSEG_COLD_DATA;
762 		else
763 			return CURSEG_WARM_DATA;
764 	} else {
765 		if (IS_DNODE(page))
766 			return is_cold_node(page) ? CURSEG_WARM_NODE :
767 						CURSEG_HOT_NODE;
768 		else
769 			return CURSEG_COLD_NODE;
770 	}
771 }
772 
773 static int __get_segment_type(struct page *page, enum page_type p_type)
774 {
775 	struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
776 	switch (sbi->active_logs) {
777 	case 2:
778 		return __get_segment_type_2(page, p_type);
779 	case 4:
780 		return __get_segment_type_4(page, p_type);
781 	}
782 	/* NR_CURSEG_TYPE(6) logs by default */
783 	BUG_ON(sbi->active_logs != NR_CURSEG_TYPE);
784 	return __get_segment_type_6(page, p_type);
785 }
786 
787 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
788 			block_t old_blkaddr, block_t *new_blkaddr,
789 			struct f2fs_summary *sum, enum page_type p_type)
790 {
791 	struct sit_info *sit_i = SIT_I(sbi);
792 	struct curseg_info *curseg;
793 	unsigned int old_cursegno;
794 	int type;
795 
796 	type = __get_segment_type(page, p_type);
797 	curseg = CURSEG_I(sbi, type);
798 
799 	mutex_lock(&curseg->curseg_mutex);
800 
801 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
802 	old_cursegno = curseg->segno;
803 
804 	/*
805 	 * __add_sum_entry should be resided under the curseg_mutex
806 	 * because, this function updates a summary entry in the
807 	 * current summary block.
808 	 */
809 	__add_sum_entry(sbi, type, sum);
810 
811 	mutex_lock(&sit_i->sentry_lock);
812 	__refresh_next_blkoff(sbi, curseg);
813 
814 	stat_inc_block_count(sbi, curseg);
815 
816 	/*
817 	 * SIT information should be updated before segment allocation,
818 	 * since SSR needs latest valid block information.
819 	 */
820 	refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
821 
822 	if (!__has_curseg_space(sbi, type))
823 		sit_i->s_ops->allocate_segment(sbi, type, false);
824 
825 	locate_dirty_segment(sbi, old_cursegno);
826 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
827 	mutex_unlock(&sit_i->sentry_lock);
828 
829 	if (p_type == NODE)
830 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
831 
832 	/* writeout dirty page into bdev */
833 	submit_write_page(sbi, page, *new_blkaddr, p_type);
834 
835 	mutex_unlock(&curseg->curseg_mutex);
836 }
837 
838 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
839 {
840 	set_page_writeback(page);
841 	submit_write_page(sbi, page, page->index, META);
842 }
843 
844 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
845 		unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
846 {
847 	struct f2fs_summary sum;
848 	set_summary(&sum, nid, 0, 0);
849 	do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, NODE);
850 }
851 
852 void write_data_page(struct inode *inode, struct page *page,
853 		struct dnode_of_data *dn, block_t old_blkaddr,
854 		block_t *new_blkaddr)
855 {
856 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
857 	struct f2fs_summary sum;
858 	struct node_info ni;
859 
860 	BUG_ON(old_blkaddr == NULL_ADDR);
861 	get_node_info(sbi, dn->nid, &ni);
862 	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
863 
864 	do_write_page(sbi, page, old_blkaddr,
865 			new_blkaddr, &sum, DATA);
866 }
867 
868 void rewrite_data_page(struct f2fs_sb_info *sbi, struct page *page,
869 					block_t old_blk_addr)
870 {
871 	submit_write_page(sbi, page, old_blk_addr, DATA);
872 }
873 
874 void recover_data_page(struct f2fs_sb_info *sbi,
875 			struct page *page, struct f2fs_summary *sum,
876 			block_t old_blkaddr, block_t new_blkaddr)
877 {
878 	struct sit_info *sit_i = SIT_I(sbi);
879 	struct curseg_info *curseg;
880 	unsigned int segno, old_cursegno;
881 	struct seg_entry *se;
882 	int type;
883 
884 	segno = GET_SEGNO(sbi, new_blkaddr);
885 	se = get_seg_entry(sbi, segno);
886 	type = se->type;
887 
888 	if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
889 		if (old_blkaddr == NULL_ADDR)
890 			type = CURSEG_COLD_DATA;
891 		else
892 			type = CURSEG_WARM_DATA;
893 	}
894 	curseg = CURSEG_I(sbi, type);
895 
896 	mutex_lock(&curseg->curseg_mutex);
897 	mutex_lock(&sit_i->sentry_lock);
898 
899 	old_cursegno = curseg->segno;
900 
901 	/* change the current segment */
902 	if (segno != curseg->segno) {
903 		curseg->next_segno = segno;
904 		change_curseg(sbi, type, true);
905 	}
906 
907 	curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
908 					(sbi->blocks_per_seg - 1);
909 	__add_sum_entry(sbi, type, sum);
910 
911 	refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
912 
913 	locate_dirty_segment(sbi, old_cursegno);
914 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
915 
916 	mutex_unlock(&sit_i->sentry_lock);
917 	mutex_unlock(&curseg->curseg_mutex);
918 }
919 
920 void rewrite_node_page(struct f2fs_sb_info *sbi,
921 			struct page *page, struct f2fs_summary *sum,
922 			block_t old_blkaddr, block_t new_blkaddr)
923 {
924 	struct sit_info *sit_i = SIT_I(sbi);
925 	int type = CURSEG_WARM_NODE;
926 	struct curseg_info *curseg;
927 	unsigned int segno, old_cursegno;
928 	block_t next_blkaddr = next_blkaddr_of_node(page);
929 	unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
930 
931 	curseg = CURSEG_I(sbi, type);
932 
933 	mutex_lock(&curseg->curseg_mutex);
934 	mutex_lock(&sit_i->sentry_lock);
935 
936 	segno = GET_SEGNO(sbi, new_blkaddr);
937 	old_cursegno = curseg->segno;
938 
939 	/* change the current segment */
940 	if (segno != curseg->segno) {
941 		curseg->next_segno = segno;
942 		change_curseg(sbi, type, true);
943 	}
944 	curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, new_blkaddr) &
945 					(sbi->blocks_per_seg - 1);
946 	__add_sum_entry(sbi, type, sum);
947 
948 	/* change the current log to the next block addr in advance */
949 	if (next_segno != segno) {
950 		curseg->next_segno = next_segno;
951 		change_curseg(sbi, type, true);
952 	}
953 	curseg->next_blkoff = GET_SEGOFF_FROM_SEG0(sbi, next_blkaddr) &
954 					(sbi->blocks_per_seg - 1);
955 
956 	/* rewrite node page */
957 	set_page_writeback(page);
958 	submit_write_page(sbi, page, new_blkaddr, NODE);
959 	f2fs_submit_bio(sbi, NODE, true);
960 	refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
961 
962 	locate_dirty_segment(sbi, old_cursegno);
963 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
964 
965 	mutex_unlock(&sit_i->sentry_lock);
966 	mutex_unlock(&curseg->curseg_mutex);
967 }
968 
969 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
970 {
971 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
972 	struct curseg_info *seg_i;
973 	unsigned char *kaddr;
974 	struct page *page;
975 	block_t start;
976 	int i, j, offset;
977 
978 	start = start_sum_block(sbi);
979 
980 	page = get_meta_page(sbi, start++);
981 	kaddr = (unsigned char *)page_address(page);
982 
983 	/* Step 1: restore nat cache */
984 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
985 	memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
986 
987 	/* Step 2: restore sit cache */
988 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
989 	memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
990 						SUM_JOURNAL_SIZE);
991 	offset = 2 * SUM_JOURNAL_SIZE;
992 
993 	/* Step 3: restore summary entries */
994 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
995 		unsigned short blk_off;
996 		unsigned int segno;
997 
998 		seg_i = CURSEG_I(sbi, i);
999 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1000 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1001 		seg_i->next_segno = segno;
1002 		reset_curseg(sbi, i, 0);
1003 		seg_i->alloc_type = ckpt->alloc_type[i];
1004 		seg_i->next_blkoff = blk_off;
1005 
1006 		if (seg_i->alloc_type == SSR)
1007 			blk_off = sbi->blocks_per_seg;
1008 
1009 		for (j = 0; j < blk_off; j++) {
1010 			struct f2fs_summary *s;
1011 			s = (struct f2fs_summary *)(kaddr + offset);
1012 			seg_i->sum_blk->entries[j] = *s;
1013 			offset += SUMMARY_SIZE;
1014 			if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1015 						SUM_FOOTER_SIZE)
1016 				continue;
1017 
1018 			f2fs_put_page(page, 1);
1019 			page = NULL;
1020 
1021 			page = get_meta_page(sbi, start++);
1022 			kaddr = (unsigned char *)page_address(page);
1023 			offset = 0;
1024 		}
1025 	}
1026 	f2fs_put_page(page, 1);
1027 	return 0;
1028 }
1029 
1030 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1031 {
1032 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1033 	struct f2fs_summary_block *sum;
1034 	struct curseg_info *curseg;
1035 	struct page *new;
1036 	unsigned short blk_off;
1037 	unsigned int segno = 0;
1038 	block_t blk_addr = 0;
1039 
1040 	/* get segment number and block addr */
1041 	if (IS_DATASEG(type)) {
1042 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1043 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1044 							CURSEG_HOT_DATA]);
1045 		if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1046 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1047 		else
1048 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1049 	} else {
1050 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
1051 							CURSEG_HOT_NODE]);
1052 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1053 							CURSEG_HOT_NODE]);
1054 		if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1055 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1056 							type - CURSEG_HOT_NODE);
1057 		else
1058 			blk_addr = GET_SUM_BLOCK(sbi, segno);
1059 	}
1060 
1061 	new = get_meta_page(sbi, blk_addr);
1062 	sum = (struct f2fs_summary_block *)page_address(new);
1063 
1064 	if (IS_NODESEG(type)) {
1065 		if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1066 			struct f2fs_summary *ns = &sum->entries[0];
1067 			int i;
1068 			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1069 				ns->version = 0;
1070 				ns->ofs_in_node = 0;
1071 			}
1072 		} else {
1073 			if (restore_node_summary(sbi, segno, sum)) {
1074 				f2fs_put_page(new, 1);
1075 				return -EINVAL;
1076 			}
1077 		}
1078 	}
1079 
1080 	/* set uncompleted segment to curseg */
1081 	curseg = CURSEG_I(sbi, type);
1082 	mutex_lock(&curseg->curseg_mutex);
1083 	memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1084 	curseg->next_segno = segno;
1085 	reset_curseg(sbi, type, 0);
1086 	curseg->alloc_type = ckpt->alloc_type[type];
1087 	curseg->next_blkoff = blk_off;
1088 	mutex_unlock(&curseg->curseg_mutex);
1089 	f2fs_put_page(new, 1);
1090 	return 0;
1091 }
1092 
1093 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1094 {
1095 	int type = CURSEG_HOT_DATA;
1096 
1097 	if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1098 		/* restore for compacted data summary */
1099 		if (read_compacted_summaries(sbi))
1100 			return -EINVAL;
1101 		type = CURSEG_HOT_NODE;
1102 	}
1103 
1104 	for (; type <= CURSEG_COLD_NODE; type++)
1105 		if (read_normal_summaries(sbi, type))
1106 			return -EINVAL;
1107 	return 0;
1108 }
1109 
1110 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1111 {
1112 	struct page *page;
1113 	unsigned char *kaddr;
1114 	struct f2fs_summary *summary;
1115 	struct curseg_info *seg_i;
1116 	int written_size = 0;
1117 	int i, j;
1118 
1119 	page = grab_meta_page(sbi, blkaddr++);
1120 	kaddr = (unsigned char *)page_address(page);
1121 
1122 	/* Step 1: write nat cache */
1123 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1124 	memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1125 	written_size += SUM_JOURNAL_SIZE;
1126 
1127 	/* Step 2: write sit cache */
1128 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1129 	memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1130 						SUM_JOURNAL_SIZE);
1131 	written_size += SUM_JOURNAL_SIZE;
1132 
1133 	set_page_dirty(page);
1134 
1135 	/* Step 3: write summary entries */
1136 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1137 		unsigned short blkoff;
1138 		seg_i = CURSEG_I(sbi, i);
1139 		if (sbi->ckpt->alloc_type[i] == SSR)
1140 			blkoff = sbi->blocks_per_seg;
1141 		else
1142 			blkoff = curseg_blkoff(sbi, i);
1143 
1144 		for (j = 0; j < blkoff; j++) {
1145 			if (!page) {
1146 				page = grab_meta_page(sbi, blkaddr++);
1147 				kaddr = (unsigned char *)page_address(page);
1148 				written_size = 0;
1149 			}
1150 			summary = (struct f2fs_summary *)(kaddr + written_size);
1151 			*summary = seg_i->sum_blk->entries[j];
1152 			written_size += SUMMARY_SIZE;
1153 			set_page_dirty(page);
1154 
1155 			if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1156 							SUM_FOOTER_SIZE)
1157 				continue;
1158 
1159 			f2fs_put_page(page, 1);
1160 			page = NULL;
1161 		}
1162 	}
1163 	if (page)
1164 		f2fs_put_page(page, 1);
1165 }
1166 
1167 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1168 					block_t blkaddr, int type)
1169 {
1170 	int i, end;
1171 	if (IS_DATASEG(type))
1172 		end = type + NR_CURSEG_DATA_TYPE;
1173 	else
1174 		end = type + NR_CURSEG_NODE_TYPE;
1175 
1176 	for (i = type; i < end; i++) {
1177 		struct curseg_info *sum = CURSEG_I(sbi, i);
1178 		mutex_lock(&sum->curseg_mutex);
1179 		write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1180 		mutex_unlock(&sum->curseg_mutex);
1181 	}
1182 }
1183 
1184 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1185 {
1186 	if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1187 		write_compacted_summaries(sbi, start_blk);
1188 	else
1189 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1190 }
1191 
1192 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1193 {
1194 	if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1195 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1196 }
1197 
1198 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1199 					unsigned int val, int alloc)
1200 {
1201 	int i;
1202 
1203 	if (type == NAT_JOURNAL) {
1204 		for (i = 0; i < nats_in_cursum(sum); i++) {
1205 			if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1206 				return i;
1207 		}
1208 		if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1209 			return update_nats_in_cursum(sum, 1);
1210 	} else if (type == SIT_JOURNAL) {
1211 		for (i = 0; i < sits_in_cursum(sum); i++)
1212 			if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1213 				return i;
1214 		if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1215 			return update_sits_in_cursum(sum, 1);
1216 	}
1217 	return -1;
1218 }
1219 
1220 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1221 					unsigned int segno)
1222 {
1223 	struct sit_info *sit_i = SIT_I(sbi);
1224 	unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1225 	block_t blk_addr = sit_i->sit_base_addr + offset;
1226 
1227 	check_seg_range(sbi, segno);
1228 
1229 	/* calculate sit block address */
1230 	if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1231 		blk_addr += sit_i->sit_blocks;
1232 
1233 	return get_meta_page(sbi, blk_addr);
1234 }
1235 
1236 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1237 					unsigned int start)
1238 {
1239 	struct sit_info *sit_i = SIT_I(sbi);
1240 	struct page *src_page, *dst_page;
1241 	pgoff_t src_off, dst_off;
1242 	void *src_addr, *dst_addr;
1243 
1244 	src_off = current_sit_addr(sbi, start);
1245 	dst_off = next_sit_addr(sbi, src_off);
1246 
1247 	/* get current sit block page without lock */
1248 	src_page = get_meta_page(sbi, src_off);
1249 	dst_page = grab_meta_page(sbi, dst_off);
1250 	BUG_ON(PageDirty(src_page));
1251 
1252 	src_addr = page_address(src_page);
1253 	dst_addr = page_address(dst_page);
1254 	memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1255 
1256 	set_page_dirty(dst_page);
1257 	f2fs_put_page(src_page, 1);
1258 
1259 	set_to_next_sit(sit_i, start);
1260 
1261 	return dst_page;
1262 }
1263 
1264 static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1265 {
1266 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1267 	struct f2fs_summary_block *sum = curseg->sum_blk;
1268 	int i;
1269 
1270 	/*
1271 	 * If the journal area in the current summary is full of sit entries,
1272 	 * all the sit entries will be flushed. Otherwise the sit entries
1273 	 * are not able to replace with newly hot sit entries.
1274 	 */
1275 	if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1276 		for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1277 			unsigned int segno;
1278 			segno = le32_to_cpu(segno_in_journal(sum, i));
1279 			__mark_sit_entry_dirty(sbi, segno);
1280 		}
1281 		update_sits_in_cursum(sum, -sits_in_cursum(sum));
1282 		return true;
1283 	}
1284 	return false;
1285 }
1286 
1287 /*
1288  * CP calls this function, which flushes SIT entries including sit_journal,
1289  * and moves prefree segs to free segs.
1290  */
1291 void flush_sit_entries(struct f2fs_sb_info *sbi)
1292 {
1293 	struct sit_info *sit_i = SIT_I(sbi);
1294 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1295 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1296 	struct f2fs_summary_block *sum = curseg->sum_blk;
1297 	unsigned long nsegs = TOTAL_SEGS(sbi);
1298 	struct page *page = NULL;
1299 	struct f2fs_sit_block *raw_sit = NULL;
1300 	unsigned int start = 0, end = 0;
1301 	unsigned int segno = -1;
1302 	bool flushed;
1303 
1304 	mutex_lock(&curseg->curseg_mutex);
1305 	mutex_lock(&sit_i->sentry_lock);
1306 
1307 	/*
1308 	 * "flushed" indicates whether sit entries in journal are flushed
1309 	 * to the SIT area or not.
1310 	 */
1311 	flushed = flush_sits_in_journal(sbi);
1312 
1313 	while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1314 		struct seg_entry *se = get_seg_entry(sbi, segno);
1315 		int sit_offset, offset;
1316 
1317 		sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1318 
1319 		if (flushed)
1320 			goto to_sit_page;
1321 
1322 		offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1323 		if (offset >= 0) {
1324 			segno_in_journal(sum, offset) = cpu_to_le32(segno);
1325 			seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1326 			goto flush_done;
1327 		}
1328 to_sit_page:
1329 		if (!page || (start > segno) || (segno > end)) {
1330 			if (page) {
1331 				f2fs_put_page(page, 1);
1332 				page = NULL;
1333 			}
1334 
1335 			start = START_SEGNO(sit_i, segno);
1336 			end = start + SIT_ENTRY_PER_BLOCK - 1;
1337 
1338 			/* read sit block that will be updated */
1339 			page = get_next_sit_page(sbi, start);
1340 			raw_sit = page_address(page);
1341 		}
1342 
1343 		/* udpate entry in SIT block */
1344 		seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1345 flush_done:
1346 		__clear_bit(segno, bitmap);
1347 		sit_i->dirty_sentries--;
1348 	}
1349 	mutex_unlock(&sit_i->sentry_lock);
1350 	mutex_unlock(&curseg->curseg_mutex);
1351 
1352 	/* writeout last modified SIT block */
1353 	f2fs_put_page(page, 1);
1354 
1355 	set_prefree_as_free_segments(sbi);
1356 }
1357 
1358 static int build_sit_info(struct f2fs_sb_info *sbi)
1359 {
1360 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1361 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1362 	struct sit_info *sit_i;
1363 	unsigned int sit_segs, start;
1364 	char *src_bitmap, *dst_bitmap;
1365 	unsigned int bitmap_size;
1366 
1367 	/* allocate memory for SIT information */
1368 	sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1369 	if (!sit_i)
1370 		return -ENOMEM;
1371 
1372 	SM_I(sbi)->sit_info = sit_i;
1373 
1374 	sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1375 	if (!sit_i->sentries)
1376 		return -ENOMEM;
1377 
1378 	bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1379 	sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1380 	if (!sit_i->dirty_sentries_bitmap)
1381 		return -ENOMEM;
1382 
1383 	for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1384 		sit_i->sentries[start].cur_valid_map
1385 			= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1386 		sit_i->sentries[start].ckpt_valid_map
1387 			= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1388 		if (!sit_i->sentries[start].cur_valid_map
1389 				|| !sit_i->sentries[start].ckpt_valid_map)
1390 			return -ENOMEM;
1391 	}
1392 
1393 	if (sbi->segs_per_sec > 1) {
1394 		sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
1395 					sizeof(struct sec_entry));
1396 		if (!sit_i->sec_entries)
1397 			return -ENOMEM;
1398 	}
1399 
1400 	/* get information related with SIT */
1401 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1402 
1403 	/* setup SIT bitmap from ckeckpoint pack */
1404 	bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1405 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1406 
1407 	dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1408 	if (!dst_bitmap)
1409 		return -ENOMEM;
1410 
1411 	/* init SIT information */
1412 	sit_i->s_ops = &default_salloc_ops;
1413 
1414 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1415 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1416 	sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1417 	sit_i->sit_bitmap = dst_bitmap;
1418 	sit_i->bitmap_size = bitmap_size;
1419 	sit_i->dirty_sentries = 0;
1420 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1421 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1422 	sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1423 	mutex_init(&sit_i->sentry_lock);
1424 	return 0;
1425 }
1426 
1427 static int build_free_segmap(struct f2fs_sb_info *sbi)
1428 {
1429 	struct f2fs_sm_info *sm_info = SM_I(sbi);
1430 	struct free_segmap_info *free_i;
1431 	unsigned int bitmap_size, sec_bitmap_size;
1432 
1433 	/* allocate memory for free segmap information */
1434 	free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1435 	if (!free_i)
1436 		return -ENOMEM;
1437 
1438 	SM_I(sbi)->free_info = free_i;
1439 
1440 	bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1441 	free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1442 	if (!free_i->free_segmap)
1443 		return -ENOMEM;
1444 
1445 	sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1446 	free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1447 	if (!free_i->free_secmap)
1448 		return -ENOMEM;
1449 
1450 	/* set all segments as dirty temporarily */
1451 	memset(free_i->free_segmap, 0xff, bitmap_size);
1452 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1453 
1454 	/* init free segmap information */
1455 	free_i->start_segno =
1456 		(unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1457 	free_i->free_segments = 0;
1458 	free_i->free_sections = 0;
1459 	rwlock_init(&free_i->segmap_lock);
1460 	return 0;
1461 }
1462 
1463 static int build_curseg(struct f2fs_sb_info *sbi)
1464 {
1465 	struct curseg_info *array;
1466 	int i;
1467 
1468 	array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
1469 	if (!array)
1470 		return -ENOMEM;
1471 
1472 	SM_I(sbi)->curseg_array = array;
1473 
1474 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
1475 		mutex_init(&array[i].curseg_mutex);
1476 		array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1477 		if (!array[i].sum_blk)
1478 			return -ENOMEM;
1479 		array[i].segno = NULL_SEGNO;
1480 		array[i].next_blkoff = 0;
1481 	}
1482 	return restore_curseg_summaries(sbi);
1483 }
1484 
1485 static void build_sit_entries(struct f2fs_sb_info *sbi)
1486 {
1487 	struct sit_info *sit_i = SIT_I(sbi);
1488 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1489 	struct f2fs_summary_block *sum = curseg->sum_blk;
1490 	unsigned int start;
1491 
1492 	for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1493 		struct seg_entry *se = &sit_i->sentries[start];
1494 		struct f2fs_sit_block *sit_blk;
1495 		struct f2fs_sit_entry sit;
1496 		struct page *page;
1497 		int i;
1498 
1499 		mutex_lock(&curseg->curseg_mutex);
1500 		for (i = 0; i < sits_in_cursum(sum); i++) {
1501 			if (le32_to_cpu(segno_in_journal(sum, i)) == start) {
1502 				sit = sit_in_journal(sum, i);
1503 				mutex_unlock(&curseg->curseg_mutex);
1504 				goto got_it;
1505 			}
1506 		}
1507 		mutex_unlock(&curseg->curseg_mutex);
1508 		page = get_current_sit_page(sbi, start);
1509 		sit_blk = (struct f2fs_sit_block *)page_address(page);
1510 		sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1511 		f2fs_put_page(page, 1);
1512 got_it:
1513 		check_block_count(sbi, start, &sit);
1514 		seg_info_from_raw_sit(se, &sit);
1515 		if (sbi->segs_per_sec > 1) {
1516 			struct sec_entry *e = get_sec_entry(sbi, start);
1517 			e->valid_blocks += se->valid_blocks;
1518 		}
1519 	}
1520 }
1521 
1522 static void init_free_segmap(struct f2fs_sb_info *sbi)
1523 {
1524 	unsigned int start;
1525 	int type;
1526 
1527 	for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1528 		struct seg_entry *sentry = get_seg_entry(sbi, start);
1529 		if (!sentry->valid_blocks)
1530 			__set_free(sbi, start);
1531 	}
1532 
1533 	/* set use the current segments */
1534 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1535 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1536 		__set_test_and_inuse(sbi, curseg_t->segno);
1537 	}
1538 }
1539 
1540 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1541 {
1542 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1543 	struct free_segmap_info *free_i = FREE_I(sbi);
1544 	unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi);
1545 	unsigned short valid_blocks;
1546 
1547 	while (1) {
1548 		/* find dirty segment based on free segmap */
1549 		segno = find_next_inuse(free_i, total_segs, offset);
1550 		if (segno >= total_segs)
1551 			break;
1552 		offset = segno + 1;
1553 		valid_blocks = get_valid_blocks(sbi, segno, 0);
1554 		if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1555 			continue;
1556 		mutex_lock(&dirty_i->seglist_lock);
1557 		__locate_dirty_segment(sbi, segno, DIRTY);
1558 		mutex_unlock(&dirty_i->seglist_lock);
1559 	}
1560 }
1561 
1562 static int init_victim_secmap(struct f2fs_sb_info *sbi)
1563 {
1564 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1565 	unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1566 
1567 	dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1568 	if (!dirty_i->victim_secmap)
1569 		return -ENOMEM;
1570 	return 0;
1571 }
1572 
1573 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1574 {
1575 	struct dirty_seglist_info *dirty_i;
1576 	unsigned int bitmap_size, i;
1577 
1578 	/* allocate memory for dirty segments list information */
1579 	dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1580 	if (!dirty_i)
1581 		return -ENOMEM;
1582 
1583 	SM_I(sbi)->dirty_info = dirty_i;
1584 	mutex_init(&dirty_i->seglist_lock);
1585 
1586 	bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1587 
1588 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
1589 		dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1590 		if (!dirty_i->dirty_segmap[i])
1591 			return -ENOMEM;
1592 	}
1593 
1594 	init_dirty_segmap(sbi);
1595 	return init_victim_secmap(sbi);
1596 }
1597 
1598 /*
1599  * Update min, max modified time for cost-benefit GC algorithm
1600  */
1601 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1602 {
1603 	struct sit_info *sit_i = SIT_I(sbi);
1604 	unsigned int segno;
1605 
1606 	mutex_lock(&sit_i->sentry_lock);
1607 
1608 	sit_i->min_mtime = LLONG_MAX;
1609 
1610 	for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1611 		unsigned int i;
1612 		unsigned long long mtime = 0;
1613 
1614 		for (i = 0; i < sbi->segs_per_sec; i++)
1615 			mtime += get_seg_entry(sbi, segno + i)->mtime;
1616 
1617 		mtime = div_u64(mtime, sbi->segs_per_sec);
1618 
1619 		if (sit_i->min_mtime > mtime)
1620 			sit_i->min_mtime = mtime;
1621 	}
1622 	sit_i->max_mtime = get_mtime(sbi);
1623 	mutex_unlock(&sit_i->sentry_lock);
1624 }
1625 
1626 int build_segment_manager(struct f2fs_sb_info *sbi)
1627 {
1628 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1629 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1630 	struct f2fs_sm_info *sm_info;
1631 	int err;
1632 
1633 	sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1634 	if (!sm_info)
1635 		return -ENOMEM;
1636 
1637 	/* init sm info */
1638 	sbi->sm_info = sm_info;
1639 	INIT_LIST_HEAD(&sm_info->wblist_head);
1640 	spin_lock_init(&sm_info->wblist_lock);
1641 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1642 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1643 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1644 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1645 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1646 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1647 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1648 	sm_info->rec_prefree_segments = DEF_RECLAIM_PREFREE_SEGMENTS;
1649 
1650 	err = build_sit_info(sbi);
1651 	if (err)
1652 		return err;
1653 	err = build_free_segmap(sbi);
1654 	if (err)
1655 		return err;
1656 	err = build_curseg(sbi);
1657 	if (err)
1658 		return err;
1659 
1660 	/* reinit free segmap based on SIT */
1661 	build_sit_entries(sbi);
1662 
1663 	init_free_segmap(sbi);
1664 	err = build_dirty_segmap(sbi);
1665 	if (err)
1666 		return err;
1667 
1668 	init_min_max_mtime(sbi);
1669 	return 0;
1670 }
1671 
1672 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1673 		enum dirty_type dirty_type)
1674 {
1675 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1676 
1677 	mutex_lock(&dirty_i->seglist_lock);
1678 	kfree(dirty_i->dirty_segmap[dirty_type]);
1679 	dirty_i->nr_dirty[dirty_type] = 0;
1680 	mutex_unlock(&dirty_i->seglist_lock);
1681 }
1682 
1683 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
1684 {
1685 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1686 	kfree(dirty_i->victim_secmap);
1687 }
1688 
1689 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1690 {
1691 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1692 	int i;
1693 
1694 	if (!dirty_i)
1695 		return;
1696 
1697 	/* discard pre-free/dirty segments list */
1698 	for (i = 0; i < NR_DIRTY_TYPE; i++)
1699 		discard_dirty_segmap(sbi, i);
1700 
1701 	destroy_victim_secmap(sbi);
1702 	SM_I(sbi)->dirty_info = NULL;
1703 	kfree(dirty_i);
1704 }
1705 
1706 static void destroy_curseg(struct f2fs_sb_info *sbi)
1707 {
1708 	struct curseg_info *array = SM_I(sbi)->curseg_array;
1709 	int i;
1710 
1711 	if (!array)
1712 		return;
1713 	SM_I(sbi)->curseg_array = NULL;
1714 	for (i = 0; i < NR_CURSEG_TYPE; i++)
1715 		kfree(array[i].sum_blk);
1716 	kfree(array);
1717 }
1718 
1719 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1720 {
1721 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
1722 	if (!free_i)
1723 		return;
1724 	SM_I(sbi)->free_info = NULL;
1725 	kfree(free_i->free_segmap);
1726 	kfree(free_i->free_secmap);
1727 	kfree(free_i);
1728 }
1729 
1730 static void destroy_sit_info(struct f2fs_sb_info *sbi)
1731 {
1732 	struct sit_info *sit_i = SIT_I(sbi);
1733 	unsigned int start;
1734 
1735 	if (!sit_i)
1736 		return;
1737 
1738 	if (sit_i->sentries) {
1739 		for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1740 			kfree(sit_i->sentries[start].cur_valid_map);
1741 			kfree(sit_i->sentries[start].ckpt_valid_map);
1742 		}
1743 	}
1744 	vfree(sit_i->sentries);
1745 	vfree(sit_i->sec_entries);
1746 	kfree(sit_i->dirty_sentries_bitmap);
1747 
1748 	SM_I(sbi)->sit_info = NULL;
1749 	kfree(sit_i->sit_bitmap);
1750 	kfree(sit_i);
1751 }
1752 
1753 void destroy_segment_manager(struct f2fs_sb_info *sbi)
1754 {
1755 	struct f2fs_sm_info *sm_info = SM_I(sbi);
1756 	destroy_dirty_segmap(sbi);
1757 	destroy_curseg(sbi);
1758 	destroy_free_segmap(sbi);
1759 	destroy_sit_info(sbi);
1760 	sbi->sm_info = NULL;
1761 	kfree(sm_info);
1762 }
1763