xref: /linux/fs/f2fs/segment.c (revision 3932b9ca55b0be314a36d3e84faff3e823c081f5)
1 /*
2  * fs/f2fs/segment.c
3  *
4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5  *             http://www.samsung.com/
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/bio.h>
14 #include <linux/blkdev.h>
15 #include <linux/prefetch.h>
16 #include <linux/kthread.h>
17 #include <linux/vmalloc.h>
18 #include <linux/swap.h>
19 
20 #include "f2fs.h"
21 #include "segment.h"
22 #include "node.h"
23 #include <trace/events/f2fs.h>
24 
25 #define __reverse_ffz(x) __reverse_ffs(~(x))
26 
27 static struct kmem_cache *discard_entry_slab;
28 
29 /*
30  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
31  * MSB and LSB are reversed in a byte by f2fs_set_bit.
32  */
33 static inline unsigned long __reverse_ffs(unsigned long word)
34 {
35 	int num = 0;
36 
37 #if BITS_PER_LONG == 64
38 	if ((word & 0xffffffff) == 0) {
39 		num += 32;
40 		word >>= 32;
41 	}
42 #endif
43 	if ((word & 0xffff) == 0) {
44 		num += 16;
45 		word >>= 16;
46 	}
47 	if ((word & 0xff) == 0) {
48 		num += 8;
49 		word >>= 8;
50 	}
51 	if ((word & 0xf0) == 0)
52 		num += 4;
53 	else
54 		word >>= 4;
55 	if ((word & 0xc) == 0)
56 		num += 2;
57 	else
58 		word >>= 2;
59 	if ((word & 0x2) == 0)
60 		num += 1;
61 	return num;
62 }
63 
64 /*
65  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
66  * f2fs_set_bit makes MSB and LSB reversed in a byte.
67  * Example:
68  *                             LSB <--> MSB
69  *   f2fs_set_bit(0, bitmap) => 0000 0001
70  *   f2fs_set_bit(7, bitmap) => 1000 0000
71  */
72 static unsigned long __find_rev_next_bit(const unsigned long *addr,
73 			unsigned long size, unsigned long offset)
74 {
75 	const unsigned long *p = addr + BIT_WORD(offset);
76 	unsigned long result = offset & ~(BITS_PER_LONG - 1);
77 	unsigned long tmp;
78 	unsigned long mask, submask;
79 	unsigned long quot, rest;
80 
81 	if (offset >= size)
82 		return size;
83 
84 	size -= result;
85 	offset %= BITS_PER_LONG;
86 	if (!offset)
87 		goto aligned;
88 
89 	tmp = *(p++);
90 	quot = (offset >> 3) << 3;
91 	rest = offset & 0x7;
92 	mask = ~0UL << quot;
93 	submask = (unsigned char)(0xff << rest) >> rest;
94 	submask <<= quot;
95 	mask &= submask;
96 	tmp &= mask;
97 	if (size < BITS_PER_LONG)
98 		goto found_first;
99 	if (tmp)
100 		goto found_middle;
101 
102 	size -= BITS_PER_LONG;
103 	result += BITS_PER_LONG;
104 aligned:
105 	while (size & ~(BITS_PER_LONG-1)) {
106 		tmp = *(p++);
107 		if (tmp)
108 			goto found_middle;
109 		result += BITS_PER_LONG;
110 		size -= BITS_PER_LONG;
111 	}
112 	if (!size)
113 		return result;
114 	tmp = *p;
115 found_first:
116 	tmp &= (~0UL >> (BITS_PER_LONG - size));
117 	if (tmp == 0UL)		/* Are any bits set? */
118 		return result + size;   /* Nope. */
119 found_middle:
120 	return result + __reverse_ffs(tmp);
121 }
122 
123 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
124 			unsigned long size, unsigned long offset)
125 {
126 	const unsigned long *p = addr + BIT_WORD(offset);
127 	unsigned long result = offset & ~(BITS_PER_LONG - 1);
128 	unsigned long tmp;
129 	unsigned long mask, submask;
130 	unsigned long quot, rest;
131 
132 	if (offset >= size)
133 		return size;
134 
135 	size -= result;
136 	offset %= BITS_PER_LONG;
137 	if (!offset)
138 		goto aligned;
139 
140 	tmp = *(p++);
141 	quot = (offset >> 3) << 3;
142 	rest = offset & 0x7;
143 	mask = ~(~0UL << quot);
144 	submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
145 	submask <<= quot;
146 	mask += submask;
147 	tmp |= mask;
148 	if (size < BITS_PER_LONG)
149 		goto found_first;
150 	if (~tmp)
151 		goto found_middle;
152 
153 	size -= BITS_PER_LONG;
154 	result += BITS_PER_LONG;
155 aligned:
156 	while (size & ~(BITS_PER_LONG - 1)) {
157 		tmp = *(p++);
158 		if (~tmp)
159 			goto found_middle;
160 		result += BITS_PER_LONG;
161 		size -= BITS_PER_LONG;
162 	}
163 	if (!size)
164 		return result;
165 	tmp = *p;
166 
167 found_first:
168 	tmp |= ~0UL << size;
169 	if (tmp == ~0UL)        /* Are any bits zero? */
170 		return result + size;   /* Nope. */
171 found_middle:
172 	return result + __reverse_ffz(tmp);
173 }
174 
175 /*
176  * This function balances dirty node and dentry pages.
177  * In addition, it controls garbage collection.
178  */
179 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
180 {
181 	/*
182 	 * We should do GC or end up with checkpoint, if there are so many dirty
183 	 * dir/node pages without enough free segments.
184 	 */
185 	if (has_not_enough_free_secs(sbi, 0)) {
186 		mutex_lock(&sbi->gc_mutex);
187 		f2fs_gc(sbi);
188 	}
189 }
190 
191 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
192 {
193 	/* check the # of cached NAT entries and prefree segments */
194 	if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
195 				excess_prefree_segs(sbi))
196 		f2fs_sync_fs(sbi->sb, true);
197 }
198 
199 static int issue_flush_thread(void *data)
200 {
201 	struct f2fs_sb_info *sbi = data;
202 	struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
203 	wait_queue_head_t *q = &fcc->flush_wait_queue;
204 repeat:
205 	if (kthread_should_stop())
206 		return 0;
207 
208 	spin_lock(&fcc->issue_lock);
209 	if (fcc->issue_list) {
210 		fcc->dispatch_list = fcc->issue_list;
211 		fcc->issue_list = fcc->issue_tail = NULL;
212 	}
213 	spin_unlock(&fcc->issue_lock);
214 
215 	if (fcc->dispatch_list) {
216 		struct bio *bio = bio_alloc(GFP_NOIO, 0);
217 		struct flush_cmd *cmd, *next;
218 		int ret;
219 
220 		bio->bi_bdev = sbi->sb->s_bdev;
221 		ret = submit_bio_wait(WRITE_FLUSH, bio);
222 
223 		for (cmd = fcc->dispatch_list; cmd; cmd = next) {
224 			cmd->ret = ret;
225 			next = cmd->next;
226 			complete(&cmd->wait);
227 		}
228 		bio_put(bio);
229 		fcc->dispatch_list = NULL;
230 	}
231 
232 	wait_event_interruptible(*q,
233 			kthread_should_stop() || fcc->issue_list);
234 	goto repeat;
235 }
236 
237 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
238 {
239 	struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
240 	struct flush_cmd cmd;
241 
242 	trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
243 					test_opt(sbi, FLUSH_MERGE));
244 
245 	if (test_opt(sbi, NOBARRIER))
246 		return 0;
247 
248 	if (!test_opt(sbi, FLUSH_MERGE))
249 		return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL);
250 
251 	init_completion(&cmd.wait);
252 	cmd.next = NULL;
253 
254 	spin_lock(&fcc->issue_lock);
255 	if (fcc->issue_list)
256 		fcc->issue_tail->next = &cmd;
257 	else
258 		fcc->issue_list = &cmd;
259 	fcc->issue_tail = &cmd;
260 	spin_unlock(&fcc->issue_lock);
261 
262 	if (!fcc->dispatch_list)
263 		wake_up(&fcc->flush_wait_queue);
264 
265 	wait_for_completion(&cmd.wait);
266 
267 	return cmd.ret;
268 }
269 
270 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
271 {
272 	dev_t dev = sbi->sb->s_bdev->bd_dev;
273 	struct flush_cmd_control *fcc;
274 	int err = 0;
275 
276 	fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
277 	if (!fcc)
278 		return -ENOMEM;
279 	spin_lock_init(&fcc->issue_lock);
280 	init_waitqueue_head(&fcc->flush_wait_queue);
281 	SM_I(sbi)->cmd_control_info = fcc;
282 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
283 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
284 	if (IS_ERR(fcc->f2fs_issue_flush)) {
285 		err = PTR_ERR(fcc->f2fs_issue_flush);
286 		kfree(fcc);
287 		SM_I(sbi)->cmd_control_info = NULL;
288 		return err;
289 	}
290 
291 	return err;
292 }
293 
294 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
295 {
296 	struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
297 
298 	if (fcc && fcc->f2fs_issue_flush)
299 		kthread_stop(fcc->f2fs_issue_flush);
300 	kfree(fcc);
301 	SM_I(sbi)->cmd_control_info = NULL;
302 }
303 
304 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
305 		enum dirty_type dirty_type)
306 {
307 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
308 
309 	/* need not be added */
310 	if (IS_CURSEG(sbi, segno))
311 		return;
312 
313 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
314 		dirty_i->nr_dirty[dirty_type]++;
315 
316 	if (dirty_type == DIRTY) {
317 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
318 		enum dirty_type t = sentry->type;
319 
320 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
321 			dirty_i->nr_dirty[t]++;
322 	}
323 }
324 
325 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
326 		enum dirty_type dirty_type)
327 {
328 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
329 
330 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
331 		dirty_i->nr_dirty[dirty_type]--;
332 
333 	if (dirty_type == DIRTY) {
334 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
335 		enum dirty_type t = sentry->type;
336 
337 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
338 			dirty_i->nr_dirty[t]--;
339 
340 		if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
341 			clear_bit(GET_SECNO(sbi, segno),
342 						dirty_i->victim_secmap);
343 	}
344 }
345 
346 /*
347  * Should not occur error such as -ENOMEM.
348  * Adding dirty entry into seglist is not critical operation.
349  * If a given segment is one of current working segments, it won't be added.
350  */
351 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
352 {
353 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
354 	unsigned short valid_blocks;
355 
356 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
357 		return;
358 
359 	mutex_lock(&dirty_i->seglist_lock);
360 
361 	valid_blocks = get_valid_blocks(sbi, segno, 0);
362 
363 	if (valid_blocks == 0) {
364 		__locate_dirty_segment(sbi, segno, PRE);
365 		__remove_dirty_segment(sbi, segno, DIRTY);
366 	} else if (valid_blocks < sbi->blocks_per_seg) {
367 		__locate_dirty_segment(sbi, segno, DIRTY);
368 	} else {
369 		/* Recovery routine with SSR needs this */
370 		__remove_dirty_segment(sbi, segno, DIRTY);
371 	}
372 
373 	mutex_unlock(&dirty_i->seglist_lock);
374 }
375 
376 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
377 				block_t blkstart, block_t blklen)
378 {
379 	sector_t start = SECTOR_FROM_BLOCK(sbi, blkstart);
380 	sector_t len = SECTOR_FROM_BLOCK(sbi, blklen);
381 	trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
382 	return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
383 }
384 
385 void discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
386 {
387 	if (f2fs_issue_discard(sbi, blkaddr, 1)) {
388 		struct page *page = grab_meta_page(sbi, blkaddr);
389 		/* zero-filled page */
390 		set_page_dirty(page);
391 		f2fs_put_page(page, 1);
392 	}
393 }
394 
395 static void add_discard_addrs(struct f2fs_sb_info *sbi,
396 			unsigned int segno, struct seg_entry *se)
397 {
398 	struct list_head *head = &SM_I(sbi)->discard_list;
399 	struct discard_entry *new;
400 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
401 	int max_blocks = sbi->blocks_per_seg;
402 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
403 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
404 	unsigned long dmap[entries];
405 	unsigned int start = 0, end = -1;
406 	int i;
407 
408 	if (!test_opt(sbi, DISCARD))
409 		return;
410 
411 	/* zero block will be discarded through the prefree list */
412 	if (!se->valid_blocks || se->valid_blocks == max_blocks)
413 		return;
414 
415 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
416 	for (i = 0; i < entries; i++)
417 		dmap[i] = (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
418 
419 	while (SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
420 		start = __find_rev_next_bit(dmap, max_blocks, end + 1);
421 		if (start >= max_blocks)
422 			break;
423 
424 		end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
425 
426 		new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
427 		INIT_LIST_HEAD(&new->list);
428 		new->blkaddr = START_BLOCK(sbi, segno) + start;
429 		new->len = end - start;
430 
431 		list_add_tail(&new->list, head);
432 		SM_I(sbi)->nr_discards += end - start;
433 	}
434 }
435 
436 /*
437  * Should call clear_prefree_segments after checkpoint is done.
438  */
439 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
440 {
441 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
442 	unsigned int segno;
443 	unsigned int total_segs = TOTAL_SEGS(sbi);
444 
445 	mutex_lock(&dirty_i->seglist_lock);
446 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], total_segs)
447 		__set_test_and_free(sbi, segno);
448 	mutex_unlock(&dirty_i->seglist_lock);
449 }
450 
451 void clear_prefree_segments(struct f2fs_sb_info *sbi)
452 {
453 	struct list_head *head = &(SM_I(sbi)->discard_list);
454 	struct discard_entry *entry, *this;
455 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
456 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
457 	unsigned int total_segs = TOTAL_SEGS(sbi);
458 	unsigned int start = 0, end = -1;
459 
460 	mutex_lock(&dirty_i->seglist_lock);
461 
462 	while (1) {
463 		int i;
464 		start = find_next_bit(prefree_map, total_segs, end + 1);
465 		if (start >= total_segs)
466 			break;
467 		end = find_next_zero_bit(prefree_map, total_segs, start + 1);
468 
469 		for (i = start; i < end; i++)
470 			clear_bit(i, prefree_map);
471 
472 		dirty_i->nr_dirty[PRE] -= end - start;
473 
474 		if (!test_opt(sbi, DISCARD))
475 			continue;
476 
477 		f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
478 				(end - start) << sbi->log_blocks_per_seg);
479 	}
480 	mutex_unlock(&dirty_i->seglist_lock);
481 
482 	/* send small discards */
483 	list_for_each_entry_safe(entry, this, head, list) {
484 		f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
485 		list_del(&entry->list);
486 		SM_I(sbi)->nr_discards -= entry->len;
487 		kmem_cache_free(discard_entry_slab, entry);
488 	}
489 }
490 
491 static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
492 {
493 	struct sit_info *sit_i = SIT_I(sbi);
494 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
495 		sit_i->dirty_sentries++;
496 }
497 
498 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
499 					unsigned int segno, int modified)
500 {
501 	struct seg_entry *se = get_seg_entry(sbi, segno);
502 	se->type = type;
503 	if (modified)
504 		__mark_sit_entry_dirty(sbi, segno);
505 }
506 
507 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
508 {
509 	struct seg_entry *se;
510 	unsigned int segno, offset;
511 	long int new_vblocks;
512 
513 	segno = GET_SEGNO(sbi, blkaddr);
514 
515 	se = get_seg_entry(sbi, segno);
516 	new_vblocks = se->valid_blocks + del;
517 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
518 
519 	f2fs_bug_on((new_vblocks >> (sizeof(unsigned short) << 3) ||
520 				(new_vblocks > sbi->blocks_per_seg)));
521 
522 	se->valid_blocks = new_vblocks;
523 	se->mtime = get_mtime(sbi);
524 	SIT_I(sbi)->max_mtime = se->mtime;
525 
526 	/* Update valid block bitmap */
527 	if (del > 0) {
528 		if (f2fs_set_bit(offset, se->cur_valid_map))
529 			BUG();
530 	} else {
531 		if (!f2fs_clear_bit(offset, se->cur_valid_map))
532 			BUG();
533 	}
534 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
535 		se->ckpt_valid_blocks += del;
536 
537 	__mark_sit_entry_dirty(sbi, segno);
538 
539 	/* update total number of valid blocks to be written in ckpt area */
540 	SIT_I(sbi)->written_valid_blocks += del;
541 
542 	if (sbi->segs_per_sec > 1)
543 		get_sec_entry(sbi, segno)->valid_blocks += del;
544 }
545 
546 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
547 {
548 	update_sit_entry(sbi, new, 1);
549 	if (GET_SEGNO(sbi, old) != NULL_SEGNO)
550 		update_sit_entry(sbi, old, -1);
551 
552 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
553 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
554 }
555 
556 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
557 {
558 	unsigned int segno = GET_SEGNO(sbi, addr);
559 	struct sit_info *sit_i = SIT_I(sbi);
560 
561 	f2fs_bug_on(addr == NULL_ADDR);
562 	if (addr == NEW_ADDR)
563 		return;
564 
565 	/* add it into sit main buffer */
566 	mutex_lock(&sit_i->sentry_lock);
567 
568 	update_sit_entry(sbi, addr, -1);
569 
570 	/* add it into dirty seglist */
571 	locate_dirty_segment(sbi, segno);
572 
573 	mutex_unlock(&sit_i->sentry_lock);
574 }
575 
576 /*
577  * This function should be resided under the curseg_mutex lock
578  */
579 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
580 					struct f2fs_summary *sum)
581 {
582 	struct curseg_info *curseg = CURSEG_I(sbi, type);
583 	void *addr = curseg->sum_blk;
584 	addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
585 	memcpy(addr, sum, sizeof(struct f2fs_summary));
586 }
587 
588 /*
589  * Calculate the number of current summary pages for writing
590  */
591 int npages_for_summary_flush(struct f2fs_sb_info *sbi)
592 {
593 	int valid_sum_count = 0;
594 	int i, sum_in_page;
595 
596 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
597 		if (sbi->ckpt->alloc_type[i] == SSR)
598 			valid_sum_count += sbi->blocks_per_seg;
599 		else
600 			valid_sum_count += curseg_blkoff(sbi, i);
601 	}
602 
603 	sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
604 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
605 	if (valid_sum_count <= sum_in_page)
606 		return 1;
607 	else if ((valid_sum_count - sum_in_page) <=
608 		(PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
609 		return 2;
610 	return 3;
611 }
612 
613 /*
614  * Caller should put this summary page
615  */
616 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
617 {
618 	return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
619 }
620 
621 static void write_sum_page(struct f2fs_sb_info *sbi,
622 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
623 {
624 	struct page *page = grab_meta_page(sbi, blk_addr);
625 	void *kaddr = page_address(page);
626 	memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
627 	set_page_dirty(page);
628 	f2fs_put_page(page, 1);
629 }
630 
631 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
632 {
633 	struct curseg_info *curseg = CURSEG_I(sbi, type);
634 	unsigned int segno = curseg->segno + 1;
635 	struct free_segmap_info *free_i = FREE_I(sbi);
636 
637 	if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec)
638 		return !test_bit(segno, free_i->free_segmap);
639 	return 0;
640 }
641 
642 /*
643  * Find a new segment from the free segments bitmap to right order
644  * This function should be returned with success, otherwise BUG
645  */
646 static void get_new_segment(struct f2fs_sb_info *sbi,
647 			unsigned int *newseg, bool new_sec, int dir)
648 {
649 	struct free_segmap_info *free_i = FREE_I(sbi);
650 	unsigned int segno, secno, zoneno;
651 	unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone;
652 	unsigned int hint = *newseg / sbi->segs_per_sec;
653 	unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
654 	unsigned int left_start = hint;
655 	bool init = true;
656 	int go_left = 0;
657 	int i;
658 
659 	write_lock(&free_i->segmap_lock);
660 
661 	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
662 		segno = find_next_zero_bit(free_i->free_segmap,
663 					TOTAL_SEGS(sbi), *newseg + 1);
664 		if (segno - *newseg < sbi->segs_per_sec -
665 					(*newseg % sbi->segs_per_sec))
666 			goto got_it;
667 	}
668 find_other_zone:
669 	secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint);
670 	if (secno >= TOTAL_SECS(sbi)) {
671 		if (dir == ALLOC_RIGHT) {
672 			secno = find_next_zero_bit(free_i->free_secmap,
673 							TOTAL_SECS(sbi), 0);
674 			f2fs_bug_on(secno >= TOTAL_SECS(sbi));
675 		} else {
676 			go_left = 1;
677 			left_start = hint - 1;
678 		}
679 	}
680 	if (go_left == 0)
681 		goto skip_left;
682 
683 	while (test_bit(left_start, free_i->free_secmap)) {
684 		if (left_start > 0) {
685 			left_start--;
686 			continue;
687 		}
688 		left_start = find_next_zero_bit(free_i->free_secmap,
689 							TOTAL_SECS(sbi), 0);
690 		f2fs_bug_on(left_start >= TOTAL_SECS(sbi));
691 		break;
692 	}
693 	secno = left_start;
694 skip_left:
695 	hint = secno;
696 	segno = secno * sbi->segs_per_sec;
697 	zoneno = secno / sbi->secs_per_zone;
698 
699 	/* give up on finding another zone */
700 	if (!init)
701 		goto got_it;
702 	if (sbi->secs_per_zone == 1)
703 		goto got_it;
704 	if (zoneno == old_zoneno)
705 		goto got_it;
706 	if (dir == ALLOC_LEFT) {
707 		if (!go_left && zoneno + 1 >= total_zones)
708 			goto got_it;
709 		if (go_left && zoneno == 0)
710 			goto got_it;
711 	}
712 	for (i = 0; i < NR_CURSEG_TYPE; i++)
713 		if (CURSEG_I(sbi, i)->zone == zoneno)
714 			break;
715 
716 	if (i < NR_CURSEG_TYPE) {
717 		/* zone is in user, try another */
718 		if (go_left)
719 			hint = zoneno * sbi->secs_per_zone - 1;
720 		else if (zoneno + 1 >= total_zones)
721 			hint = 0;
722 		else
723 			hint = (zoneno + 1) * sbi->secs_per_zone;
724 		init = false;
725 		goto find_other_zone;
726 	}
727 got_it:
728 	/* set it as dirty segment in free segmap */
729 	f2fs_bug_on(test_bit(segno, free_i->free_segmap));
730 	__set_inuse(sbi, segno);
731 	*newseg = segno;
732 	write_unlock(&free_i->segmap_lock);
733 }
734 
735 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
736 {
737 	struct curseg_info *curseg = CURSEG_I(sbi, type);
738 	struct summary_footer *sum_footer;
739 
740 	curseg->segno = curseg->next_segno;
741 	curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
742 	curseg->next_blkoff = 0;
743 	curseg->next_segno = NULL_SEGNO;
744 
745 	sum_footer = &(curseg->sum_blk->footer);
746 	memset(sum_footer, 0, sizeof(struct summary_footer));
747 	if (IS_DATASEG(type))
748 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
749 	if (IS_NODESEG(type))
750 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
751 	__set_sit_entry_type(sbi, type, curseg->segno, modified);
752 }
753 
754 /*
755  * Allocate a current working segment.
756  * This function always allocates a free segment in LFS manner.
757  */
758 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
759 {
760 	struct curseg_info *curseg = CURSEG_I(sbi, type);
761 	unsigned int segno = curseg->segno;
762 	int dir = ALLOC_LEFT;
763 
764 	write_sum_page(sbi, curseg->sum_blk,
765 				GET_SUM_BLOCK(sbi, segno));
766 	if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
767 		dir = ALLOC_RIGHT;
768 
769 	if (test_opt(sbi, NOHEAP))
770 		dir = ALLOC_RIGHT;
771 
772 	get_new_segment(sbi, &segno, new_sec, dir);
773 	curseg->next_segno = segno;
774 	reset_curseg(sbi, type, 1);
775 	curseg->alloc_type = LFS;
776 }
777 
778 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
779 			struct curseg_info *seg, block_t start)
780 {
781 	struct seg_entry *se = get_seg_entry(sbi, seg->segno);
782 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
783 	unsigned long target_map[entries];
784 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
785 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
786 	int i, pos;
787 
788 	for (i = 0; i < entries; i++)
789 		target_map[i] = ckpt_map[i] | cur_map[i];
790 
791 	pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
792 
793 	seg->next_blkoff = pos;
794 }
795 
796 /*
797  * If a segment is written by LFS manner, next block offset is just obtained
798  * by increasing the current block offset. However, if a segment is written by
799  * SSR manner, next block offset obtained by calling __next_free_blkoff
800  */
801 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
802 				struct curseg_info *seg)
803 {
804 	if (seg->alloc_type == SSR)
805 		__next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
806 	else
807 		seg->next_blkoff++;
808 }
809 
810 /*
811  * This function always allocates a used segment(from dirty seglist) by SSR
812  * manner, so it should recover the existing segment information of valid blocks
813  */
814 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
815 {
816 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
817 	struct curseg_info *curseg = CURSEG_I(sbi, type);
818 	unsigned int new_segno = curseg->next_segno;
819 	struct f2fs_summary_block *sum_node;
820 	struct page *sum_page;
821 
822 	write_sum_page(sbi, curseg->sum_blk,
823 				GET_SUM_BLOCK(sbi, curseg->segno));
824 	__set_test_and_inuse(sbi, new_segno);
825 
826 	mutex_lock(&dirty_i->seglist_lock);
827 	__remove_dirty_segment(sbi, new_segno, PRE);
828 	__remove_dirty_segment(sbi, new_segno, DIRTY);
829 	mutex_unlock(&dirty_i->seglist_lock);
830 
831 	reset_curseg(sbi, type, 1);
832 	curseg->alloc_type = SSR;
833 	__next_free_blkoff(sbi, curseg, 0);
834 
835 	if (reuse) {
836 		sum_page = get_sum_page(sbi, new_segno);
837 		sum_node = (struct f2fs_summary_block *)page_address(sum_page);
838 		memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
839 		f2fs_put_page(sum_page, 1);
840 	}
841 }
842 
843 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
844 {
845 	struct curseg_info *curseg = CURSEG_I(sbi, type);
846 	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
847 
848 	if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
849 		return v_ops->get_victim(sbi,
850 				&(curseg)->next_segno, BG_GC, type, SSR);
851 
852 	/* For data segments, let's do SSR more intensively */
853 	for (; type >= CURSEG_HOT_DATA; type--)
854 		if (v_ops->get_victim(sbi, &(curseg)->next_segno,
855 						BG_GC, type, SSR))
856 			return 1;
857 	return 0;
858 }
859 
860 /*
861  * flush out current segment and replace it with new segment
862  * This function should be returned with success, otherwise BUG
863  */
864 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
865 						int type, bool force)
866 {
867 	struct curseg_info *curseg = CURSEG_I(sbi, type);
868 
869 	if (force)
870 		new_curseg(sbi, type, true);
871 	else if (type == CURSEG_WARM_NODE)
872 		new_curseg(sbi, type, false);
873 	else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
874 		new_curseg(sbi, type, false);
875 	else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
876 		change_curseg(sbi, type, true);
877 	else
878 		new_curseg(sbi, type, false);
879 
880 	stat_inc_seg_type(sbi, curseg);
881 }
882 
883 void allocate_new_segments(struct f2fs_sb_info *sbi)
884 {
885 	struct curseg_info *curseg;
886 	unsigned int old_curseg;
887 	int i;
888 
889 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
890 		curseg = CURSEG_I(sbi, i);
891 		old_curseg = curseg->segno;
892 		SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
893 		locate_dirty_segment(sbi, old_curseg);
894 	}
895 }
896 
897 static const struct segment_allocation default_salloc_ops = {
898 	.allocate_segment = allocate_segment_by_default,
899 };
900 
901 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
902 {
903 	struct curseg_info *curseg = CURSEG_I(sbi, type);
904 	if (curseg->next_blkoff < sbi->blocks_per_seg)
905 		return true;
906 	return false;
907 }
908 
909 static int __get_segment_type_2(struct page *page, enum page_type p_type)
910 {
911 	if (p_type == DATA)
912 		return CURSEG_HOT_DATA;
913 	else
914 		return CURSEG_HOT_NODE;
915 }
916 
917 static int __get_segment_type_4(struct page *page, enum page_type p_type)
918 {
919 	if (p_type == DATA) {
920 		struct inode *inode = page->mapping->host;
921 
922 		if (S_ISDIR(inode->i_mode))
923 			return CURSEG_HOT_DATA;
924 		else
925 			return CURSEG_COLD_DATA;
926 	} else {
927 		if (IS_DNODE(page) && !is_cold_node(page))
928 			return CURSEG_HOT_NODE;
929 		else
930 			return CURSEG_COLD_NODE;
931 	}
932 }
933 
934 static int __get_segment_type_6(struct page *page, enum page_type p_type)
935 {
936 	if (p_type == DATA) {
937 		struct inode *inode = page->mapping->host;
938 
939 		if (S_ISDIR(inode->i_mode))
940 			return CURSEG_HOT_DATA;
941 		else if (is_cold_data(page) || file_is_cold(inode))
942 			return CURSEG_COLD_DATA;
943 		else
944 			return CURSEG_WARM_DATA;
945 	} else {
946 		if (IS_DNODE(page))
947 			return is_cold_node(page) ? CURSEG_WARM_NODE :
948 						CURSEG_HOT_NODE;
949 		else
950 			return CURSEG_COLD_NODE;
951 	}
952 }
953 
954 static int __get_segment_type(struct page *page, enum page_type p_type)
955 {
956 	struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
957 	switch (sbi->active_logs) {
958 	case 2:
959 		return __get_segment_type_2(page, p_type);
960 	case 4:
961 		return __get_segment_type_4(page, p_type);
962 	}
963 	/* NR_CURSEG_TYPE(6) logs by default */
964 	f2fs_bug_on(sbi->active_logs != NR_CURSEG_TYPE);
965 	return __get_segment_type_6(page, p_type);
966 }
967 
968 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
969 		block_t old_blkaddr, block_t *new_blkaddr,
970 		struct f2fs_summary *sum, int type)
971 {
972 	struct sit_info *sit_i = SIT_I(sbi);
973 	struct curseg_info *curseg;
974 
975 	curseg = CURSEG_I(sbi, type);
976 
977 	mutex_lock(&curseg->curseg_mutex);
978 
979 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
980 
981 	/*
982 	 * __add_sum_entry should be resided under the curseg_mutex
983 	 * because, this function updates a summary entry in the
984 	 * current summary block.
985 	 */
986 	__add_sum_entry(sbi, type, sum);
987 
988 	mutex_lock(&sit_i->sentry_lock);
989 	__refresh_next_blkoff(sbi, curseg);
990 
991 	stat_inc_block_count(sbi, curseg);
992 
993 	if (!__has_curseg_space(sbi, type))
994 		sit_i->s_ops->allocate_segment(sbi, type, false);
995 	/*
996 	 * SIT information should be updated before segment allocation,
997 	 * since SSR needs latest valid block information.
998 	 */
999 	refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1000 
1001 	mutex_unlock(&sit_i->sentry_lock);
1002 
1003 	if (page && IS_NODESEG(type))
1004 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1005 
1006 	mutex_unlock(&curseg->curseg_mutex);
1007 }
1008 
1009 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
1010 			block_t old_blkaddr, block_t *new_blkaddr,
1011 			struct f2fs_summary *sum, struct f2fs_io_info *fio)
1012 {
1013 	int type = __get_segment_type(page, fio->type);
1014 
1015 	allocate_data_block(sbi, page, old_blkaddr, new_blkaddr, sum, type);
1016 
1017 	/* writeout dirty page into bdev */
1018 	f2fs_submit_page_mbio(sbi, page, *new_blkaddr, fio);
1019 }
1020 
1021 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1022 {
1023 	struct f2fs_io_info fio = {
1024 		.type = META,
1025 		.rw = WRITE_SYNC | REQ_META | REQ_PRIO
1026 	};
1027 
1028 	set_page_writeback(page);
1029 	f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1030 }
1031 
1032 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
1033 		struct f2fs_io_info *fio,
1034 		unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
1035 {
1036 	struct f2fs_summary sum;
1037 	set_summary(&sum, nid, 0, 0);
1038 	do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, fio);
1039 }
1040 
1041 void write_data_page(struct page *page, struct dnode_of_data *dn,
1042 		block_t *new_blkaddr, struct f2fs_io_info *fio)
1043 {
1044 	struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
1045 	struct f2fs_summary sum;
1046 	struct node_info ni;
1047 
1048 	f2fs_bug_on(dn->data_blkaddr == NULL_ADDR);
1049 	get_node_info(sbi, dn->nid, &ni);
1050 	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1051 
1052 	do_write_page(sbi, page, dn->data_blkaddr, new_blkaddr, &sum, fio);
1053 }
1054 
1055 void rewrite_data_page(struct page *page, block_t old_blkaddr,
1056 					struct f2fs_io_info *fio)
1057 {
1058 	struct inode *inode = page->mapping->host;
1059 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1060 	f2fs_submit_page_mbio(sbi, page, old_blkaddr, fio);
1061 }
1062 
1063 void recover_data_page(struct f2fs_sb_info *sbi,
1064 			struct page *page, struct f2fs_summary *sum,
1065 			block_t old_blkaddr, block_t new_blkaddr)
1066 {
1067 	struct sit_info *sit_i = SIT_I(sbi);
1068 	struct curseg_info *curseg;
1069 	unsigned int segno, old_cursegno;
1070 	struct seg_entry *se;
1071 	int type;
1072 
1073 	segno = GET_SEGNO(sbi, new_blkaddr);
1074 	se = get_seg_entry(sbi, segno);
1075 	type = se->type;
1076 
1077 	if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1078 		if (old_blkaddr == NULL_ADDR)
1079 			type = CURSEG_COLD_DATA;
1080 		else
1081 			type = CURSEG_WARM_DATA;
1082 	}
1083 	curseg = CURSEG_I(sbi, type);
1084 
1085 	mutex_lock(&curseg->curseg_mutex);
1086 	mutex_lock(&sit_i->sentry_lock);
1087 
1088 	old_cursegno = curseg->segno;
1089 
1090 	/* change the current segment */
1091 	if (segno != curseg->segno) {
1092 		curseg->next_segno = segno;
1093 		change_curseg(sbi, type, true);
1094 	}
1095 
1096 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1097 	__add_sum_entry(sbi, type, sum);
1098 
1099 	refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1100 	locate_dirty_segment(sbi, old_cursegno);
1101 
1102 	mutex_unlock(&sit_i->sentry_lock);
1103 	mutex_unlock(&curseg->curseg_mutex);
1104 }
1105 
1106 static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1107 					struct page *page, enum page_type type)
1108 {
1109 	enum page_type btype = PAGE_TYPE_OF_BIO(type);
1110 	struct f2fs_bio_info *io = &sbi->write_io[btype];
1111 	struct bio_vec *bvec;
1112 	int i;
1113 
1114 	down_read(&io->io_rwsem);
1115 	if (!io->bio)
1116 		goto out;
1117 
1118 	bio_for_each_segment_all(bvec, io->bio, i) {
1119 		if (page == bvec->bv_page) {
1120 			up_read(&io->io_rwsem);
1121 			return true;
1122 		}
1123 	}
1124 
1125 out:
1126 	up_read(&io->io_rwsem);
1127 	return false;
1128 }
1129 
1130 void f2fs_wait_on_page_writeback(struct page *page,
1131 				enum page_type type)
1132 {
1133 	struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1134 	if (PageWriteback(page)) {
1135 		if (is_merged_page(sbi, page, type))
1136 			f2fs_submit_merged_bio(sbi, type, WRITE);
1137 		wait_on_page_writeback(page);
1138 	}
1139 }
1140 
1141 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1142 {
1143 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1144 	struct curseg_info *seg_i;
1145 	unsigned char *kaddr;
1146 	struct page *page;
1147 	block_t start;
1148 	int i, j, offset;
1149 
1150 	start = start_sum_block(sbi);
1151 
1152 	page = get_meta_page(sbi, start++);
1153 	kaddr = (unsigned char *)page_address(page);
1154 
1155 	/* Step 1: restore nat cache */
1156 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1157 	memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1158 
1159 	/* Step 2: restore sit cache */
1160 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1161 	memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1162 						SUM_JOURNAL_SIZE);
1163 	offset = 2 * SUM_JOURNAL_SIZE;
1164 
1165 	/* Step 3: restore summary entries */
1166 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1167 		unsigned short blk_off;
1168 		unsigned int segno;
1169 
1170 		seg_i = CURSEG_I(sbi, i);
1171 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1172 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1173 		seg_i->next_segno = segno;
1174 		reset_curseg(sbi, i, 0);
1175 		seg_i->alloc_type = ckpt->alloc_type[i];
1176 		seg_i->next_blkoff = blk_off;
1177 
1178 		if (seg_i->alloc_type == SSR)
1179 			blk_off = sbi->blocks_per_seg;
1180 
1181 		for (j = 0; j < blk_off; j++) {
1182 			struct f2fs_summary *s;
1183 			s = (struct f2fs_summary *)(kaddr + offset);
1184 			seg_i->sum_blk->entries[j] = *s;
1185 			offset += SUMMARY_SIZE;
1186 			if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1187 						SUM_FOOTER_SIZE)
1188 				continue;
1189 
1190 			f2fs_put_page(page, 1);
1191 			page = NULL;
1192 
1193 			page = get_meta_page(sbi, start++);
1194 			kaddr = (unsigned char *)page_address(page);
1195 			offset = 0;
1196 		}
1197 	}
1198 	f2fs_put_page(page, 1);
1199 	return 0;
1200 }
1201 
1202 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1203 {
1204 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1205 	struct f2fs_summary_block *sum;
1206 	struct curseg_info *curseg;
1207 	struct page *new;
1208 	unsigned short blk_off;
1209 	unsigned int segno = 0;
1210 	block_t blk_addr = 0;
1211 
1212 	/* get segment number and block addr */
1213 	if (IS_DATASEG(type)) {
1214 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1215 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1216 							CURSEG_HOT_DATA]);
1217 		if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1218 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1219 		else
1220 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1221 	} else {
1222 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
1223 							CURSEG_HOT_NODE]);
1224 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1225 							CURSEG_HOT_NODE]);
1226 		if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1227 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1228 							type - CURSEG_HOT_NODE);
1229 		else
1230 			blk_addr = GET_SUM_BLOCK(sbi, segno);
1231 	}
1232 
1233 	new = get_meta_page(sbi, blk_addr);
1234 	sum = (struct f2fs_summary_block *)page_address(new);
1235 
1236 	if (IS_NODESEG(type)) {
1237 		if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1238 			struct f2fs_summary *ns = &sum->entries[0];
1239 			int i;
1240 			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1241 				ns->version = 0;
1242 				ns->ofs_in_node = 0;
1243 			}
1244 		} else {
1245 			int err;
1246 
1247 			err = restore_node_summary(sbi, segno, sum);
1248 			if (err) {
1249 				f2fs_put_page(new, 1);
1250 				return err;
1251 			}
1252 		}
1253 	}
1254 
1255 	/* set uncompleted segment to curseg */
1256 	curseg = CURSEG_I(sbi, type);
1257 	mutex_lock(&curseg->curseg_mutex);
1258 	memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1259 	curseg->next_segno = segno;
1260 	reset_curseg(sbi, type, 0);
1261 	curseg->alloc_type = ckpt->alloc_type[type];
1262 	curseg->next_blkoff = blk_off;
1263 	mutex_unlock(&curseg->curseg_mutex);
1264 	f2fs_put_page(new, 1);
1265 	return 0;
1266 }
1267 
1268 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1269 {
1270 	int type = CURSEG_HOT_DATA;
1271 	int err;
1272 
1273 	if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1274 		/* restore for compacted data summary */
1275 		if (read_compacted_summaries(sbi))
1276 			return -EINVAL;
1277 		type = CURSEG_HOT_NODE;
1278 	}
1279 
1280 	for (; type <= CURSEG_COLD_NODE; type++) {
1281 		err = read_normal_summaries(sbi, type);
1282 		if (err)
1283 			return err;
1284 	}
1285 
1286 	return 0;
1287 }
1288 
1289 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1290 {
1291 	struct page *page;
1292 	unsigned char *kaddr;
1293 	struct f2fs_summary *summary;
1294 	struct curseg_info *seg_i;
1295 	int written_size = 0;
1296 	int i, j;
1297 
1298 	page = grab_meta_page(sbi, blkaddr++);
1299 	kaddr = (unsigned char *)page_address(page);
1300 
1301 	/* Step 1: write nat cache */
1302 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1303 	memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1304 	written_size += SUM_JOURNAL_SIZE;
1305 
1306 	/* Step 2: write sit cache */
1307 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1308 	memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1309 						SUM_JOURNAL_SIZE);
1310 	written_size += SUM_JOURNAL_SIZE;
1311 
1312 	/* Step 3: write summary entries */
1313 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1314 		unsigned short blkoff;
1315 		seg_i = CURSEG_I(sbi, i);
1316 		if (sbi->ckpt->alloc_type[i] == SSR)
1317 			blkoff = sbi->blocks_per_seg;
1318 		else
1319 			blkoff = curseg_blkoff(sbi, i);
1320 
1321 		for (j = 0; j < blkoff; j++) {
1322 			if (!page) {
1323 				page = grab_meta_page(sbi, blkaddr++);
1324 				kaddr = (unsigned char *)page_address(page);
1325 				written_size = 0;
1326 			}
1327 			summary = (struct f2fs_summary *)(kaddr + written_size);
1328 			*summary = seg_i->sum_blk->entries[j];
1329 			written_size += SUMMARY_SIZE;
1330 
1331 			if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1332 							SUM_FOOTER_SIZE)
1333 				continue;
1334 
1335 			set_page_dirty(page);
1336 			f2fs_put_page(page, 1);
1337 			page = NULL;
1338 		}
1339 	}
1340 	if (page) {
1341 		set_page_dirty(page);
1342 		f2fs_put_page(page, 1);
1343 	}
1344 }
1345 
1346 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1347 					block_t blkaddr, int type)
1348 {
1349 	int i, end;
1350 	if (IS_DATASEG(type))
1351 		end = type + NR_CURSEG_DATA_TYPE;
1352 	else
1353 		end = type + NR_CURSEG_NODE_TYPE;
1354 
1355 	for (i = type; i < end; i++) {
1356 		struct curseg_info *sum = CURSEG_I(sbi, i);
1357 		mutex_lock(&sum->curseg_mutex);
1358 		write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1359 		mutex_unlock(&sum->curseg_mutex);
1360 	}
1361 }
1362 
1363 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1364 {
1365 	if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1366 		write_compacted_summaries(sbi, start_blk);
1367 	else
1368 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1369 }
1370 
1371 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1372 {
1373 	if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1374 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1375 }
1376 
1377 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1378 					unsigned int val, int alloc)
1379 {
1380 	int i;
1381 
1382 	if (type == NAT_JOURNAL) {
1383 		for (i = 0; i < nats_in_cursum(sum); i++) {
1384 			if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1385 				return i;
1386 		}
1387 		if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1388 			return update_nats_in_cursum(sum, 1);
1389 	} else if (type == SIT_JOURNAL) {
1390 		for (i = 0; i < sits_in_cursum(sum); i++)
1391 			if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1392 				return i;
1393 		if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1394 			return update_sits_in_cursum(sum, 1);
1395 	}
1396 	return -1;
1397 }
1398 
1399 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1400 					unsigned int segno)
1401 {
1402 	struct sit_info *sit_i = SIT_I(sbi);
1403 	unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1404 	block_t blk_addr = sit_i->sit_base_addr + offset;
1405 
1406 	check_seg_range(sbi, segno);
1407 
1408 	/* calculate sit block address */
1409 	if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1410 		blk_addr += sit_i->sit_blocks;
1411 
1412 	return get_meta_page(sbi, blk_addr);
1413 }
1414 
1415 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1416 					unsigned int start)
1417 {
1418 	struct sit_info *sit_i = SIT_I(sbi);
1419 	struct page *src_page, *dst_page;
1420 	pgoff_t src_off, dst_off;
1421 	void *src_addr, *dst_addr;
1422 
1423 	src_off = current_sit_addr(sbi, start);
1424 	dst_off = next_sit_addr(sbi, src_off);
1425 
1426 	/* get current sit block page without lock */
1427 	src_page = get_meta_page(sbi, src_off);
1428 	dst_page = grab_meta_page(sbi, dst_off);
1429 	f2fs_bug_on(PageDirty(src_page));
1430 
1431 	src_addr = page_address(src_page);
1432 	dst_addr = page_address(dst_page);
1433 	memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1434 
1435 	set_page_dirty(dst_page);
1436 	f2fs_put_page(src_page, 1);
1437 
1438 	set_to_next_sit(sit_i, start);
1439 
1440 	return dst_page;
1441 }
1442 
1443 static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1444 {
1445 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1446 	struct f2fs_summary_block *sum = curseg->sum_blk;
1447 	int i;
1448 
1449 	/*
1450 	 * If the journal area in the current summary is full of sit entries,
1451 	 * all the sit entries will be flushed. Otherwise the sit entries
1452 	 * are not able to replace with newly hot sit entries.
1453 	 */
1454 	if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1455 		for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1456 			unsigned int segno;
1457 			segno = le32_to_cpu(segno_in_journal(sum, i));
1458 			__mark_sit_entry_dirty(sbi, segno);
1459 		}
1460 		update_sits_in_cursum(sum, -sits_in_cursum(sum));
1461 		return true;
1462 	}
1463 	return false;
1464 }
1465 
1466 /*
1467  * CP calls this function, which flushes SIT entries including sit_journal,
1468  * and moves prefree segs to free segs.
1469  */
1470 void flush_sit_entries(struct f2fs_sb_info *sbi)
1471 {
1472 	struct sit_info *sit_i = SIT_I(sbi);
1473 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1474 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1475 	struct f2fs_summary_block *sum = curseg->sum_blk;
1476 	unsigned long nsegs = TOTAL_SEGS(sbi);
1477 	struct page *page = NULL;
1478 	struct f2fs_sit_block *raw_sit = NULL;
1479 	unsigned int start = 0, end = 0;
1480 	unsigned int segno;
1481 	bool flushed;
1482 
1483 	mutex_lock(&curseg->curseg_mutex);
1484 	mutex_lock(&sit_i->sentry_lock);
1485 
1486 	/*
1487 	 * "flushed" indicates whether sit entries in journal are flushed
1488 	 * to the SIT area or not.
1489 	 */
1490 	flushed = flush_sits_in_journal(sbi);
1491 
1492 	for_each_set_bit(segno, bitmap, nsegs) {
1493 		struct seg_entry *se = get_seg_entry(sbi, segno);
1494 		int sit_offset, offset;
1495 
1496 		sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1497 
1498 		/* add discard candidates */
1499 		if (SM_I(sbi)->nr_discards < SM_I(sbi)->max_discards)
1500 			add_discard_addrs(sbi, segno, se);
1501 
1502 		if (flushed)
1503 			goto to_sit_page;
1504 
1505 		offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1506 		if (offset >= 0) {
1507 			segno_in_journal(sum, offset) = cpu_to_le32(segno);
1508 			seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1509 			goto flush_done;
1510 		}
1511 to_sit_page:
1512 		if (!page || (start > segno) || (segno > end)) {
1513 			if (page) {
1514 				f2fs_put_page(page, 1);
1515 				page = NULL;
1516 			}
1517 
1518 			start = START_SEGNO(sit_i, segno);
1519 			end = start + SIT_ENTRY_PER_BLOCK - 1;
1520 
1521 			/* read sit block that will be updated */
1522 			page = get_next_sit_page(sbi, start);
1523 			raw_sit = page_address(page);
1524 		}
1525 
1526 		/* udpate entry in SIT block */
1527 		seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1528 flush_done:
1529 		__clear_bit(segno, bitmap);
1530 		sit_i->dirty_sentries--;
1531 	}
1532 	mutex_unlock(&sit_i->sentry_lock);
1533 	mutex_unlock(&curseg->curseg_mutex);
1534 
1535 	/* writeout last modified SIT block */
1536 	f2fs_put_page(page, 1);
1537 
1538 	set_prefree_as_free_segments(sbi);
1539 }
1540 
1541 static int build_sit_info(struct f2fs_sb_info *sbi)
1542 {
1543 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1544 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1545 	struct sit_info *sit_i;
1546 	unsigned int sit_segs, start;
1547 	char *src_bitmap, *dst_bitmap;
1548 	unsigned int bitmap_size;
1549 
1550 	/* allocate memory for SIT information */
1551 	sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1552 	if (!sit_i)
1553 		return -ENOMEM;
1554 
1555 	SM_I(sbi)->sit_info = sit_i;
1556 
1557 	sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1558 	if (!sit_i->sentries)
1559 		return -ENOMEM;
1560 
1561 	bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1562 	sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1563 	if (!sit_i->dirty_sentries_bitmap)
1564 		return -ENOMEM;
1565 
1566 	for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1567 		sit_i->sentries[start].cur_valid_map
1568 			= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1569 		sit_i->sentries[start].ckpt_valid_map
1570 			= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1571 		if (!sit_i->sentries[start].cur_valid_map
1572 				|| !sit_i->sentries[start].ckpt_valid_map)
1573 			return -ENOMEM;
1574 	}
1575 
1576 	if (sbi->segs_per_sec > 1) {
1577 		sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
1578 					sizeof(struct sec_entry));
1579 		if (!sit_i->sec_entries)
1580 			return -ENOMEM;
1581 	}
1582 
1583 	/* get information related with SIT */
1584 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1585 
1586 	/* setup SIT bitmap from ckeckpoint pack */
1587 	bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1588 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1589 
1590 	dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1591 	if (!dst_bitmap)
1592 		return -ENOMEM;
1593 
1594 	/* init SIT information */
1595 	sit_i->s_ops = &default_salloc_ops;
1596 
1597 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1598 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1599 	sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1600 	sit_i->sit_bitmap = dst_bitmap;
1601 	sit_i->bitmap_size = bitmap_size;
1602 	sit_i->dirty_sentries = 0;
1603 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1604 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1605 	sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1606 	mutex_init(&sit_i->sentry_lock);
1607 	return 0;
1608 }
1609 
1610 static int build_free_segmap(struct f2fs_sb_info *sbi)
1611 {
1612 	struct f2fs_sm_info *sm_info = SM_I(sbi);
1613 	struct free_segmap_info *free_i;
1614 	unsigned int bitmap_size, sec_bitmap_size;
1615 
1616 	/* allocate memory for free segmap information */
1617 	free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1618 	if (!free_i)
1619 		return -ENOMEM;
1620 
1621 	SM_I(sbi)->free_info = free_i;
1622 
1623 	bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1624 	free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1625 	if (!free_i->free_segmap)
1626 		return -ENOMEM;
1627 
1628 	sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1629 	free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1630 	if (!free_i->free_secmap)
1631 		return -ENOMEM;
1632 
1633 	/* set all segments as dirty temporarily */
1634 	memset(free_i->free_segmap, 0xff, bitmap_size);
1635 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1636 
1637 	/* init free segmap information */
1638 	free_i->start_segno =
1639 		(unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1640 	free_i->free_segments = 0;
1641 	free_i->free_sections = 0;
1642 	rwlock_init(&free_i->segmap_lock);
1643 	return 0;
1644 }
1645 
1646 static int build_curseg(struct f2fs_sb_info *sbi)
1647 {
1648 	struct curseg_info *array;
1649 	int i;
1650 
1651 	array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
1652 	if (!array)
1653 		return -ENOMEM;
1654 
1655 	SM_I(sbi)->curseg_array = array;
1656 
1657 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
1658 		mutex_init(&array[i].curseg_mutex);
1659 		array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1660 		if (!array[i].sum_blk)
1661 			return -ENOMEM;
1662 		array[i].segno = NULL_SEGNO;
1663 		array[i].next_blkoff = 0;
1664 	}
1665 	return restore_curseg_summaries(sbi);
1666 }
1667 
1668 static void build_sit_entries(struct f2fs_sb_info *sbi)
1669 {
1670 	struct sit_info *sit_i = SIT_I(sbi);
1671 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1672 	struct f2fs_summary_block *sum = curseg->sum_blk;
1673 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
1674 	unsigned int i, start, end;
1675 	unsigned int readed, start_blk = 0;
1676 	int nrpages = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1677 
1678 	do {
1679 		readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
1680 
1681 		start = start_blk * sit_i->sents_per_block;
1682 		end = (start_blk + readed) * sit_i->sents_per_block;
1683 
1684 		for (; start < end && start < TOTAL_SEGS(sbi); start++) {
1685 			struct seg_entry *se = &sit_i->sentries[start];
1686 			struct f2fs_sit_block *sit_blk;
1687 			struct f2fs_sit_entry sit;
1688 			struct page *page;
1689 
1690 			mutex_lock(&curseg->curseg_mutex);
1691 			for (i = 0; i < sits_in_cursum(sum); i++) {
1692 				if (le32_to_cpu(segno_in_journal(sum, i))
1693 								== start) {
1694 					sit = sit_in_journal(sum, i);
1695 					mutex_unlock(&curseg->curseg_mutex);
1696 					goto got_it;
1697 				}
1698 			}
1699 			mutex_unlock(&curseg->curseg_mutex);
1700 
1701 			page = get_current_sit_page(sbi, start);
1702 			sit_blk = (struct f2fs_sit_block *)page_address(page);
1703 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1704 			f2fs_put_page(page, 1);
1705 got_it:
1706 			check_block_count(sbi, start, &sit);
1707 			seg_info_from_raw_sit(se, &sit);
1708 			if (sbi->segs_per_sec > 1) {
1709 				struct sec_entry *e = get_sec_entry(sbi, start);
1710 				e->valid_blocks += se->valid_blocks;
1711 			}
1712 		}
1713 		start_blk += readed;
1714 	} while (start_blk < sit_blk_cnt);
1715 }
1716 
1717 static void init_free_segmap(struct f2fs_sb_info *sbi)
1718 {
1719 	unsigned int start;
1720 	int type;
1721 
1722 	for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1723 		struct seg_entry *sentry = get_seg_entry(sbi, start);
1724 		if (!sentry->valid_blocks)
1725 			__set_free(sbi, start);
1726 	}
1727 
1728 	/* set use the current segments */
1729 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1730 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1731 		__set_test_and_inuse(sbi, curseg_t->segno);
1732 	}
1733 }
1734 
1735 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1736 {
1737 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1738 	struct free_segmap_info *free_i = FREE_I(sbi);
1739 	unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi);
1740 	unsigned short valid_blocks;
1741 
1742 	while (1) {
1743 		/* find dirty segment based on free segmap */
1744 		segno = find_next_inuse(free_i, total_segs, offset);
1745 		if (segno >= total_segs)
1746 			break;
1747 		offset = segno + 1;
1748 		valid_blocks = get_valid_blocks(sbi, segno, 0);
1749 		if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1750 			continue;
1751 		mutex_lock(&dirty_i->seglist_lock);
1752 		__locate_dirty_segment(sbi, segno, DIRTY);
1753 		mutex_unlock(&dirty_i->seglist_lock);
1754 	}
1755 }
1756 
1757 static int init_victim_secmap(struct f2fs_sb_info *sbi)
1758 {
1759 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1760 	unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1761 
1762 	dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1763 	if (!dirty_i->victim_secmap)
1764 		return -ENOMEM;
1765 	return 0;
1766 }
1767 
1768 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1769 {
1770 	struct dirty_seglist_info *dirty_i;
1771 	unsigned int bitmap_size, i;
1772 
1773 	/* allocate memory for dirty segments list information */
1774 	dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1775 	if (!dirty_i)
1776 		return -ENOMEM;
1777 
1778 	SM_I(sbi)->dirty_info = dirty_i;
1779 	mutex_init(&dirty_i->seglist_lock);
1780 
1781 	bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1782 
1783 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
1784 		dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1785 		if (!dirty_i->dirty_segmap[i])
1786 			return -ENOMEM;
1787 	}
1788 
1789 	init_dirty_segmap(sbi);
1790 	return init_victim_secmap(sbi);
1791 }
1792 
1793 /*
1794  * Update min, max modified time for cost-benefit GC algorithm
1795  */
1796 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1797 {
1798 	struct sit_info *sit_i = SIT_I(sbi);
1799 	unsigned int segno;
1800 
1801 	mutex_lock(&sit_i->sentry_lock);
1802 
1803 	sit_i->min_mtime = LLONG_MAX;
1804 
1805 	for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1806 		unsigned int i;
1807 		unsigned long long mtime = 0;
1808 
1809 		for (i = 0; i < sbi->segs_per_sec; i++)
1810 			mtime += get_seg_entry(sbi, segno + i)->mtime;
1811 
1812 		mtime = div_u64(mtime, sbi->segs_per_sec);
1813 
1814 		if (sit_i->min_mtime > mtime)
1815 			sit_i->min_mtime = mtime;
1816 	}
1817 	sit_i->max_mtime = get_mtime(sbi);
1818 	mutex_unlock(&sit_i->sentry_lock);
1819 }
1820 
1821 int build_segment_manager(struct f2fs_sb_info *sbi)
1822 {
1823 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1824 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1825 	struct f2fs_sm_info *sm_info;
1826 	int err;
1827 
1828 	sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1829 	if (!sm_info)
1830 		return -ENOMEM;
1831 
1832 	/* init sm info */
1833 	sbi->sm_info = sm_info;
1834 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1835 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1836 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1837 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1838 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1839 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1840 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1841 	sm_info->rec_prefree_segments = sm_info->main_segments *
1842 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
1843 	sm_info->ipu_policy = F2FS_IPU_DISABLE;
1844 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
1845 
1846 	INIT_LIST_HEAD(&sm_info->discard_list);
1847 	sm_info->nr_discards = 0;
1848 	sm_info->max_discards = 0;
1849 
1850 	if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
1851 		err = create_flush_cmd_control(sbi);
1852 		if (err)
1853 			return err;
1854 	}
1855 
1856 	err = build_sit_info(sbi);
1857 	if (err)
1858 		return err;
1859 	err = build_free_segmap(sbi);
1860 	if (err)
1861 		return err;
1862 	err = build_curseg(sbi);
1863 	if (err)
1864 		return err;
1865 
1866 	/* reinit free segmap based on SIT */
1867 	build_sit_entries(sbi);
1868 
1869 	init_free_segmap(sbi);
1870 	err = build_dirty_segmap(sbi);
1871 	if (err)
1872 		return err;
1873 
1874 	init_min_max_mtime(sbi);
1875 	return 0;
1876 }
1877 
1878 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1879 		enum dirty_type dirty_type)
1880 {
1881 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1882 
1883 	mutex_lock(&dirty_i->seglist_lock);
1884 	kfree(dirty_i->dirty_segmap[dirty_type]);
1885 	dirty_i->nr_dirty[dirty_type] = 0;
1886 	mutex_unlock(&dirty_i->seglist_lock);
1887 }
1888 
1889 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
1890 {
1891 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1892 	kfree(dirty_i->victim_secmap);
1893 }
1894 
1895 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1896 {
1897 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1898 	int i;
1899 
1900 	if (!dirty_i)
1901 		return;
1902 
1903 	/* discard pre-free/dirty segments list */
1904 	for (i = 0; i < NR_DIRTY_TYPE; i++)
1905 		discard_dirty_segmap(sbi, i);
1906 
1907 	destroy_victim_secmap(sbi);
1908 	SM_I(sbi)->dirty_info = NULL;
1909 	kfree(dirty_i);
1910 }
1911 
1912 static void destroy_curseg(struct f2fs_sb_info *sbi)
1913 {
1914 	struct curseg_info *array = SM_I(sbi)->curseg_array;
1915 	int i;
1916 
1917 	if (!array)
1918 		return;
1919 	SM_I(sbi)->curseg_array = NULL;
1920 	for (i = 0; i < NR_CURSEG_TYPE; i++)
1921 		kfree(array[i].sum_blk);
1922 	kfree(array);
1923 }
1924 
1925 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1926 {
1927 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
1928 	if (!free_i)
1929 		return;
1930 	SM_I(sbi)->free_info = NULL;
1931 	kfree(free_i->free_segmap);
1932 	kfree(free_i->free_secmap);
1933 	kfree(free_i);
1934 }
1935 
1936 static void destroy_sit_info(struct f2fs_sb_info *sbi)
1937 {
1938 	struct sit_info *sit_i = SIT_I(sbi);
1939 	unsigned int start;
1940 
1941 	if (!sit_i)
1942 		return;
1943 
1944 	if (sit_i->sentries) {
1945 		for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1946 			kfree(sit_i->sentries[start].cur_valid_map);
1947 			kfree(sit_i->sentries[start].ckpt_valid_map);
1948 		}
1949 	}
1950 	vfree(sit_i->sentries);
1951 	vfree(sit_i->sec_entries);
1952 	kfree(sit_i->dirty_sentries_bitmap);
1953 
1954 	SM_I(sbi)->sit_info = NULL;
1955 	kfree(sit_i->sit_bitmap);
1956 	kfree(sit_i);
1957 }
1958 
1959 void destroy_segment_manager(struct f2fs_sb_info *sbi)
1960 {
1961 	struct f2fs_sm_info *sm_info = SM_I(sbi);
1962 
1963 	if (!sm_info)
1964 		return;
1965 	destroy_flush_cmd_control(sbi);
1966 	destroy_dirty_segmap(sbi);
1967 	destroy_curseg(sbi);
1968 	destroy_free_segmap(sbi);
1969 	destroy_sit_info(sbi);
1970 	sbi->sm_info = NULL;
1971 	kfree(sm_info);
1972 }
1973 
1974 int __init create_segment_manager_caches(void)
1975 {
1976 	discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
1977 			sizeof(struct discard_entry));
1978 	if (!discard_entry_slab)
1979 		return -ENOMEM;
1980 	return 0;
1981 }
1982 
1983 void destroy_segment_manager_caches(void)
1984 {
1985 	kmem_cache_destroy(discard_entry_slab);
1986 }
1987