xref: /linux/fs/f2fs/segment.c (revision c5aec4c76af1a2d89ee2f2d4d5463b2ad2d85de5)
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 becasue
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 	if (!test_opt(sbi, FLUSH_MERGE))
243 		return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL);
244 
245 	init_completion(&cmd.wait);
246 	cmd.next = NULL;
247 
248 	spin_lock(&fcc->issue_lock);
249 	if (fcc->issue_list)
250 		fcc->issue_tail->next = &cmd;
251 	else
252 		fcc->issue_list = &cmd;
253 	fcc->issue_tail = &cmd;
254 	spin_unlock(&fcc->issue_lock);
255 
256 	if (!fcc->dispatch_list)
257 		wake_up(&fcc->flush_wait_queue);
258 
259 	wait_for_completion(&cmd.wait);
260 
261 	return cmd.ret;
262 }
263 
264 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
265 {
266 	dev_t dev = sbi->sb->s_bdev->bd_dev;
267 	struct flush_cmd_control *fcc;
268 	int err = 0;
269 
270 	fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
271 	if (!fcc)
272 		return -ENOMEM;
273 	spin_lock_init(&fcc->issue_lock);
274 	init_waitqueue_head(&fcc->flush_wait_queue);
275 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
276 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
277 	if (IS_ERR(fcc->f2fs_issue_flush)) {
278 		err = PTR_ERR(fcc->f2fs_issue_flush);
279 		kfree(fcc);
280 		return err;
281 	}
282 	sbi->sm_info->cmd_control_info = fcc;
283 
284 	return err;
285 }
286 
287 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
288 {
289 	struct flush_cmd_control *fcc =
290 				sbi->sm_info->cmd_control_info;
291 
292 	if (fcc && fcc->f2fs_issue_flush)
293 		kthread_stop(fcc->f2fs_issue_flush);
294 	kfree(fcc);
295 	sbi->sm_info->cmd_control_info = NULL;
296 }
297 
298 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
299 		enum dirty_type dirty_type)
300 {
301 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
302 
303 	/* need not be added */
304 	if (IS_CURSEG(sbi, segno))
305 		return;
306 
307 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
308 		dirty_i->nr_dirty[dirty_type]++;
309 
310 	if (dirty_type == DIRTY) {
311 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
312 		enum dirty_type t = sentry->type;
313 
314 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
315 			dirty_i->nr_dirty[t]++;
316 	}
317 }
318 
319 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
320 		enum dirty_type dirty_type)
321 {
322 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
323 
324 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
325 		dirty_i->nr_dirty[dirty_type]--;
326 
327 	if (dirty_type == DIRTY) {
328 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
329 		enum dirty_type t = sentry->type;
330 
331 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
332 			dirty_i->nr_dirty[t]--;
333 
334 		if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
335 			clear_bit(GET_SECNO(sbi, segno),
336 						dirty_i->victim_secmap);
337 	}
338 }
339 
340 /*
341  * Should not occur error such as -ENOMEM.
342  * Adding dirty entry into seglist is not critical operation.
343  * If a given segment is one of current working segments, it won't be added.
344  */
345 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
346 {
347 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
348 	unsigned short valid_blocks;
349 
350 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
351 		return;
352 
353 	mutex_lock(&dirty_i->seglist_lock);
354 
355 	valid_blocks = get_valid_blocks(sbi, segno, 0);
356 
357 	if (valid_blocks == 0) {
358 		__locate_dirty_segment(sbi, segno, PRE);
359 		__remove_dirty_segment(sbi, segno, DIRTY);
360 	} else if (valid_blocks < sbi->blocks_per_seg) {
361 		__locate_dirty_segment(sbi, segno, DIRTY);
362 	} else {
363 		/* Recovery routine with SSR needs this */
364 		__remove_dirty_segment(sbi, segno, DIRTY);
365 	}
366 
367 	mutex_unlock(&dirty_i->seglist_lock);
368 }
369 
370 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
371 				block_t blkstart, block_t blklen)
372 {
373 	sector_t start = SECTOR_FROM_BLOCK(sbi, blkstart);
374 	sector_t len = SECTOR_FROM_BLOCK(sbi, blklen);
375 	trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
376 	return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
377 }
378 
379 void discard_next_dnode(struct f2fs_sb_info *sbi)
380 {
381 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
382 	block_t blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
383 
384 	if (f2fs_issue_discard(sbi, blkaddr, 1)) {
385 		struct page *page = grab_meta_page(sbi, blkaddr);
386 		/* zero-filled page */
387 		set_page_dirty(page);
388 		f2fs_put_page(page, 1);
389 	}
390 }
391 
392 static void add_discard_addrs(struct f2fs_sb_info *sbi,
393 			unsigned int segno, struct seg_entry *se)
394 {
395 	struct list_head *head = &SM_I(sbi)->discard_list;
396 	struct discard_entry *new;
397 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
398 	int max_blocks = sbi->blocks_per_seg;
399 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
400 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
401 	unsigned long dmap[entries];
402 	unsigned int start = 0, end = -1;
403 	int i;
404 
405 	if (!test_opt(sbi, DISCARD))
406 		return;
407 
408 	/* zero block will be discarded through the prefree list */
409 	if (!se->valid_blocks || se->valid_blocks == max_blocks)
410 		return;
411 
412 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
413 	for (i = 0; i < entries; i++)
414 		dmap[i] = (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
415 
416 	while (SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
417 		start = __find_rev_next_bit(dmap, max_blocks, end + 1);
418 		if (start >= max_blocks)
419 			break;
420 
421 		end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
422 
423 		new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
424 		INIT_LIST_HEAD(&new->list);
425 		new->blkaddr = START_BLOCK(sbi, segno) + start;
426 		new->len = end - start;
427 
428 		list_add_tail(&new->list, head);
429 		SM_I(sbi)->nr_discards += end - start;
430 	}
431 }
432 
433 /*
434  * Should call clear_prefree_segments after checkpoint is done.
435  */
436 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
437 {
438 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
439 	unsigned int segno = -1;
440 	unsigned int total_segs = TOTAL_SEGS(sbi);
441 
442 	mutex_lock(&dirty_i->seglist_lock);
443 	while (1) {
444 		segno = find_next_bit(dirty_i->dirty_segmap[PRE], total_segs,
445 				segno + 1);
446 		if (segno >= total_segs)
447 			break;
448 		__set_test_and_free(sbi, segno);
449 	}
450 	mutex_unlock(&dirty_i->seglist_lock);
451 }
452 
453 void clear_prefree_segments(struct f2fs_sb_info *sbi)
454 {
455 	struct list_head *head = &(SM_I(sbi)->discard_list);
456 	struct discard_entry *entry, *this;
457 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
458 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
459 	unsigned int total_segs = TOTAL_SEGS(sbi);
460 	unsigned int start = 0, end = -1;
461 
462 	mutex_lock(&dirty_i->seglist_lock);
463 
464 	while (1) {
465 		int i;
466 		start = find_next_bit(prefree_map, total_segs, end + 1);
467 		if (start >= total_segs)
468 			break;
469 		end = find_next_zero_bit(prefree_map, total_segs, start + 1);
470 
471 		for (i = start; i < end; i++)
472 			clear_bit(i, prefree_map);
473 
474 		dirty_i->nr_dirty[PRE] -= end - start;
475 
476 		if (!test_opt(sbi, DISCARD))
477 			continue;
478 
479 		f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
480 				(end - start) << sbi->log_blocks_per_seg);
481 	}
482 	mutex_unlock(&dirty_i->seglist_lock);
483 
484 	/* send small discards */
485 	list_for_each_entry_safe(entry, this, head, list) {
486 		f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
487 		list_del(&entry->list);
488 		SM_I(sbi)->nr_discards -= entry->len;
489 		kmem_cache_free(discard_entry_slab, entry);
490 	}
491 }
492 
493 static void __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
494 {
495 	struct sit_info *sit_i = SIT_I(sbi);
496 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap))
497 		sit_i->dirty_sentries++;
498 }
499 
500 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
501 					unsigned int segno, int modified)
502 {
503 	struct seg_entry *se = get_seg_entry(sbi, segno);
504 	se->type = type;
505 	if (modified)
506 		__mark_sit_entry_dirty(sbi, segno);
507 }
508 
509 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
510 {
511 	struct seg_entry *se;
512 	unsigned int segno, offset;
513 	long int new_vblocks;
514 
515 	segno = GET_SEGNO(sbi, blkaddr);
516 
517 	se = get_seg_entry(sbi, segno);
518 	new_vblocks = se->valid_blocks + del;
519 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
520 
521 	f2fs_bug_on((new_vblocks >> (sizeof(unsigned short) << 3) ||
522 				(new_vblocks > sbi->blocks_per_seg)));
523 
524 	se->valid_blocks = new_vblocks;
525 	se->mtime = get_mtime(sbi);
526 	SIT_I(sbi)->max_mtime = se->mtime;
527 
528 	/* Update valid block bitmap */
529 	if (del > 0) {
530 		if (f2fs_set_bit(offset, se->cur_valid_map))
531 			BUG();
532 	} else {
533 		if (!f2fs_clear_bit(offset, se->cur_valid_map))
534 			BUG();
535 	}
536 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
537 		se->ckpt_valid_blocks += del;
538 
539 	__mark_sit_entry_dirty(sbi, segno);
540 
541 	/* update total number of valid blocks to be written in ckpt area */
542 	SIT_I(sbi)->written_valid_blocks += del;
543 
544 	if (sbi->segs_per_sec > 1)
545 		get_sec_entry(sbi, segno)->valid_blocks += del;
546 }
547 
548 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
549 {
550 	update_sit_entry(sbi, new, 1);
551 	if (GET_SEGNO(sbi, old) != NULL_SEGNO)
552 		update_sit_entry(sbi, old, -1);
553 
554 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
555 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
556 }
557 
558 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
559 {
560 	unsigned int segno = GET_SEGNO(sbi, addr);
561 	struct sit_info *sit_i = SIT_I(sbi);
562 
563 	f2fs_bug_on(addr == NULL_ADDR);
564 	if (addr == NEW_ADDR)
565 		return;
566 
567 	/* add it into sit main buffer */
568 	mutex_lock(&sit_i->sentry_lock);
569 
570 	update_sit_entry(sbi, addr, -1);
571 
572 	/* add it into dirty seglist */
573 	locate_dirty_segment(sbi, segno);
574 
575 	mutex_unlock(&sit_i->sentry_lock);
576 }
577 
578 /*
579  * This function should be resided under the curseg_mutex lock
580  */
581 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
582 					struct f2fs_summary *sum)
583 {
584 	struct curseg_info *curseg = CURSEG_I(sbi, type);
585 	void *addr = curseg->sum_blk;
586 	addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
587 	memcpy(addr, sum, sizeof(struct f2fs_summary));
588 }
589 
590 /*
591  * Calculate the number of current summary pages for writing
592  */
593 int npages_for_summary_flush(struct f2fs_sb_info *sbi)
594 {
595 	int valid_sum_count = 0;
596 	int i, sum_in_page;
597 
598 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
599 		if (sbi->ckpt->alloc_type[i] == SSR)
600 			valid_sum_count += sbi->blocks_per_seg;
601 		else
602 			valid_sum_count += curseg_blkoff(sbi, i);
603 	}
604 
605 	sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
606 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
607 	if (valid_sum_count <= sum_in_page)
608 		return 1;
609 	else if ((valid_sum_count - sum_in_page) <=
610 		(PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
611 		return 2;
612 	return 3;
613 }
614 
615 /*
616  * Caller should put this summary page
617  */
618 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
619 {
620 	return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
621 }
622 
623 static void write_sum_page(struct f2fs_sb_info *sbi,
624 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
625 {
626 	struct page *page = grab_meta_page(sbi, blk_addr);
627 	void *kaddr = page_address(page);
628 	memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
629 	set_page_dirty(page);
630 	f2fs_put_page(page, 1);
631 }
632 
633 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
634 {
635 	struct curseg_info *curseg = CURSEG_I(sbi, type);
636 	unsigned int segno = curseg->segno + 1;
637 	struct free_segmap_info *free_i = FREE_I(sbi);
638 
639 	if (segno < TOTAL_SEGS(sbi) && segno % sbi->segs_per_sec)
640 		return !test_bit(segno, free_i->free_segmap);
641 	return 0;
642 }
643 
644 /*
645  * Find a new segment from the free segments bitmap to right order
646  * This function should be returned with success, otherwise BUG
647  */
648 static void get_new_segment(struct f2fs_sb_info *sbi,
649 			unsigned int *newseg, bool new_sec, int dir)
650 {
651 	struct free_segmap_info *free_i = FREE_I(sbi);
652 	unsigned int segno, secno, zoneno;
653 	unsigned int total_zones = TOTAL_SECS(sbi) / sbi->secs_per_zone;
654 	unsigned int hint = *newseg / sbi->segs_per_sec;
655 	unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
656 	unsigned int left_start = hint;
657 	bool init = true;
658 	int go_left = 0;
659 	int i;
660 
661 	write_lock(&free_i->segmap_lock);
662 
663 	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
664 		segno = find_next_zero_bit(free_i->free_segmap,
665 					TOTAL_SEGS(sbi), *newseg + 1);
666 		if (segno - *newseg < sbi->segs_per_sec -
667 					(*newseg % sbi->segs_per_sec))
668 			goto got_it;
669 	}
670 find_other_zone:
671 	secno = find_next_zero_bit(free_i->free_secmap, TOTAL_SECS(sbi), hint);
672 	if (secno >= TOTAL_SECS(sbi)) {
673 		if (dir == ALLOC_RIGHT) {
674 			secno = find_next_zero_bit(free_i->free_secmap,
675 							TOTAL_SECS(sbi), 0);
676 			f2fs_bug_on(secno >= TOTAL_SECS(sbi));
677 		} else {
678 			go_left = 1;
679 			left_start = hint - 1;
680 		}
681 	}
682 	if (go_left == 0)
683 		goto skip_left;
684 
685 	while (test_bit(left_start, free_i->free_secmap)) {
686 		if (left_start > 0) {
687 			left_start--;
688 			continue;
689 		}
690 		left_start = find_next_zero_bit(free_i->free_secmap,
691 							TOTAL_SECS(sbi), 0);
692 		f2fs_bug_on(left_start >= TOTAL_SECS(sbi));
693 		break;
694 	}
695 	secno = left_start;
696 skip_left:
697 	hint = secno;
698 	segno = secno * sbi->segs_per_sec;
699 	zoneno = secno / sbi->secs_per_zone;
700 
701 	/* give up on finding another zone */
702 	if (!init)
703 		goto got_it;
704 	if (sbi->secs_per_zone == 1)
705 		goto got_it;
706 	if (zoneno == old_zoneno)
707 		goto got_it;
708 	if (dir == ALLOC_LEFT) {
709 		if (!go_left && zoneno + 1 >= total_zones)
710 			goto got_it;
711 		if (go_left && zoneno == 0)
712 			goto got_it;
713 	}
714 	for (i = 0; i < NR_CURSEG_TYPE; i++)
715 		if (CURSEG_I(sbi, i)->zone == zoneno)
716 			break;
717 
718 	if (i < NR_CURSEG_TYPE) {
719 		/* zone is in user, try another */
720 		if (go_left)
721 			hint = zoneno * sbi->secs_per_zone - 1;
722 		else if (zoneno + 1 >= total_zones)
723 			hint = 0;
724 		else
725 			hint = (zoneno + 1) * sbi->secs_per_zone;
726 		init = false;
727 		goto find_other_zone;
728 	}
729 got_it:
730 	/* set it as dirty segment in free segmap */
731 	f2fs_bug_on(test_bit(segno, free_i->free_segmap));
732 	__set_inuse(sbi, segno);
733 	*newseg = segno;
734 	write_unlock(&free_i->segmap_lock);
735 }
736 
737 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
738 {
739 	struct curseg_info *curseg = CURSEG_I(sbi, type);
740 	struct summary_footer *sum_footer;
741 
742 	curseg->segno = curseg->next_segno;
743 	curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
744 	curseg->next_blkoff = 0;
745 	curseg->next_segno = NULL_SEGNO;
746 
747 	sum_footer = &(curseg->sum_blk->footer);
748 	memset(sum_footer, 0, sizeof(struct summary_footer));
749 	if (IS_DATASEG(type))
750 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
751 	if (IS_NODESEG(type))
752 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
753 	__set_sit_entry_type(sbi, type, curseg->segno, modified);
754 }
755 
756 /*
757  * Allocate a current working segment.
758  * This function always allocates a free segment in LFS manner.
759  */
760 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
761 {
762 	struct curseg_info *curseg = CURSEG_I(sbi, type);
763 	unsigned int segno = curseg->segno;
764 	int dir = ALLOC_LEFT;
765 
766 	write_sum_page(sbi, curseg->sum_blk,
767 				GET_SUM_BLOCK(sbi, segno));
768 	if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
769 		dir = ALLOC_RIGHT;
770 
771 	if (test_opt(sbi, NOHEAP))
772 		dir = ALLOC_RIGHT;
773 
774 	get_new_segment(sbi, &segno, new_sec, dir);
775 	curseg->next_segno = segno;
776 	reset_curseg(sbi, type, 1);
777 	curseg->alloc_type = LFS;
778 }
779 
780 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
781 			struct curseg_info *seg, block_t start)
782 {
783 	struct seg_entry *se = get_seg_entry(sbi, seg->segno);
784 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
785 	unsigned long target_map[entries];
786 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
787 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
788 	int i, pos;
789 
790 	for (i = 0; i < entries; i++)
791 		target_map[i] = ckpt_map[i] | cur_map[i];
792 
793 	pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
794 
795 	seg->next_blkoff = pos;
796 }
797 
798 /*
799  * If a segment is written by LFS manner, next block offset is just obtained
800  * by increasing the current block offset. However, if a segment is written by
801  * SSR manner, next block offset obtained by calling __next_free_blkoff
802  */
803 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
804 				struct curseg_info *seg)
805 {
806 	if (seg->alloc_type == SSR)
807 		__next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
808 	else
809 		seg->next_blkoff++;
810 }
811 
812 /*
813  * This function always allocates a used segment (from dirty seglist) by SSR
814  * manner, so it should recover the existing segment information of valid blocks
815  */
816 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
817 {
818 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
819 	struct curseg_info *curseg = CURSEG_I(sbi, type);
820 	unsigned int new_segno = curseg->next_segno;
821 	struct f2fs_summary_block *sum_node;
822 	struct page *sum_page;
823 
824 	write_sum_page(sbi, curseg->sum_blk,
825 				GET_SUM_BLOCK(sbi, curseg->segno));
826 	__set_test_and_inuse(sbi, new_segno);
827 
828 	mutex_lock(&dirty_i->seglist_lock);
829 	__remove_dirty_segment(sbi, new_segno, PRE);
830 	__remove_dirty_segment(sbi, new_segno, DIRTY);
831 	mutex_unlock(&dirty_i->seglist_lock);
832 
833 	reset_curseg(sbi, type, 1);
834 	curseg->alloc_type = SSR;
835 	__next_free_blkoff(sbi, curseg, 0);
836 
837 	if (reuse) {
838 		sum_page = get_sum_page(sbi, new_segno);
839 		sum_node = (struct f2fs_summary_block *)page_address(sum_page);
840 		memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
841 		f2fs_put_page(sum_page, 1);
842 	}
843 }
844 
845 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
846 {
847 	struct curseg_info *curseg = CURSEG_I(sbi, type);
848 	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
849 
850 	if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
851 		return v_ops->get_victim(sbi,
852 				&(curseg)->next_segno, BG_GC, type, SSR);
853 
854 	/* For data segments, let's do SSR more intensively */
855 	for (; type >= CURSEG_HOT_DATA; type--)
856 		if (v_ops->get_victim(sbi, &(curseg)->next_segno,
857 						BG_GC, type, SSR))
858 			return 1;
859 	return 0;
860 }
861 
862 /*
863  * flush out current segment and replace it with new segment
864  * This function should be returned with success, otherwise BUG
865  */
866 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
867 						int type, bool force)
868 {
869 	struct curseg_info *curseg = CURSEG_I(sbi, type);
870 
871 	if (force)
872 		new_curseg(sbi, type, true);
873 	else if (type == CURSEG_WARM_NODE)
874 		new_curseg(sbi, type, false);
875 	else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
876 		new_curseg(sbi, type, false);
877 	else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
878 		change_curseg(sbi, type, true);
879 	else
880 		new_curseg(sbi, type, false);
881 
882 	stat_inc_seg_type(sbi, curseg);
883 }
884 
885 void allocate_new_segments(struct f2fs_sb_info *sbi)
886 {
887 	struct curseg_info *curseg;
888 	unsigned int old_curseg;
889 	int i;
890 
891 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
892 		curseg = CURSEG_I(sbi, i);
893 		old_curseg = curseg->segno;
894 		SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
895 		locate_dirty_segment(sbi, old_curseg);
896 	}
897 }
898 
899 static const struct segment_allocation default_salloc_ops = {
900 	.allocate_segment = allocate_segment_by_default,
901 };
902 
903 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
904 {
905 	struct curseg_info *curseg = CURSEG_I(sbi, type);
906 	if (curseg->next_blkoff < sbi->blocks_per_seg)
907 		return true;
908 	return false;
909 }
910 
911 static int __get_segment_type_2(struct page *page, enum page_type p_type)
912 {
913 	if (p_type == DATA)
914 		return CURSEG_HOT_DATA;
915 	else
916 		return CURSEG_HOT_NODE;
917 }
918 
919 static int __get_segment_type_4(struct page *page, enum page_type p_type)
920 {
921 	if (p_type == DATA) {
922 		struct inode *inode = page->mapping->host;
923 
924 		if (S_ISDIR(inode->i_mode))
925 			return CURSEG_HOT_DATA;
926 		else
927 			return CURSEG_COLD_DATA;
928 	} else {
929 		if (IS_DNODE(page) && !is_cold_node(page))
930 			return CURSEG_HOT_NODE;
931 		else
932 			return CURSEG_COLD_NODE;
933 	}
934 }
935 
936 static int __get_segment_type_6(struct page *page, enum page_type p_type)
937 {
938 	if (p_type == DATA) {
939 		struct inode *inode = page->mapping->host;
940 
941 		if (S_ISDIR(inode->i_mode))
942 			return CURSEG_HOT_DATA;
943 		else if (is_cold_data(page) || file_is_cold(inode))
944 			return CURSEG_COLD_DATA;
945 		else
946 			return CURSEG_WARM_DATA;
947 	} else {
948 		if (IS_DNODE(page))
949 			return is_cold_node(page) ? CURSEG_WARM_NODE :
950 						CURSEG_HOT_NODE;
951 		else
952 			return CURSEG_COLD_NODE;
953 	}
954 }
955 
956 static int __get_segment_type(struct page *page, enum page_type p_type)
957 {
958 	struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
959 	switch (sbi->active_logs) {
960 	case 2:
961 		return __get_segment_type_2(page, p_type);
962 	case 4:
963 		return __get_segment_type_4(page, p_type);
964 	}
965 	/* NR_CURSEG_TYPE(6) logs by default */
966 	f2fs_bug_on(sbi->active_logs != NR_CURSEG_TYPE);
967 	return __get_segment_type_6(page, p_type);
968 }
969 
970 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
971 		block_t old_blkaddr, block_t *new_blkaddr,
972 		struct f2fs_summary *sum, int type)
973 {
974 	struct sit_info *sit_i = SIT_I(sbi);
975 	struct curseg_info *curseg;
976 	unsigned int old_cursegno;
977 
978 	curseg = CURSEG_I(sbi, type);
979 
980 	mutex_lock(&curseg->curseg_mutex);
981 
982 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
983 	old_cursegno = curseg->segno;
984 
985 	/*
986 	 * __add_sum_entry should be resided under the curseg_mutex
987 	 * because, this function updates a summary entry in the
988 	 * current summary block.
989 	 */
990 	__add_sum_entry(sbi, type, sum);
991 
992 	mutex_lock(&sit_i->sentry_lock);
993 	__refresh_next_blkoff(sbi, curseg);
994 
995 	stat_inc_block_count(sbi, curseg);
996 
997 	if (!__has_curseg_space(sbi, type))
998 		sit_i->s_ops->allocate_segment(sbi, type, false);
999 	/*
1000 	 * SIT information should be updated before segment allocation,
1001 	 * since SSR needs latest valid block information.
1002 	 */
1003 	refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1004 	locate_dirty_segment(sbi, old_cursegno);
1005 
1006 	mutex_unlock(&sit_i->sentry_lock);
1007 
1008 	if (page && IS_NODESEG(type))
1009 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1010 
1011 	mutex_unlock(&curseg->curseg_mutex);
1012 }
1013 
1014 static void do_write_page(struct f2fs_sb_info *sbi, struct page *page,
1015 			block_t old_blkaddr, block_t *new_blkaddr,
1016 			struct f2fs_summary *sum, struct f2fs_io_info *fio)
1017 {
1018 	int type = __get_segment_type(page, fio->type);
1019 
1020 	allocate_data_block(sbi, page, old_blkaddr, new_blkaddr, sum, type);
1021 
1022 	/* writeout dirty page into bdev */
1023 	f2fs_submit_page_mbio(sbi, page, *new_blkaddr, fio);
1024 }
1025 
1026 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1027 {
1028 	struct f2fs_io_info fio = {
1029 		.type = META,
1030 		.rw = WRITE_SYNC | REQ_META | REQ_PRIO
1031 	};
1032 
1033 	set_page_writeback(page);
1034 	f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1035 }
1036 
1037 void write_node_page(struct f2fs_sb_info *sbi, struct page *page,
1038 		struct f2fs_io_info *fio,
1039 		unsigned int nid, block_t old_blkaddr, block_t *new_blkaddr)
1040 {
1041 	struct f2fs_summary sum;
1042 	set_summary(&sum, nid, 0, 0);
1043 	do_write_page(sbi, page, old_blkaddr, new_blkaddr, &sum, fio);
1044 }
1045 
1046 void write_data_page(struct page *page, struct dnode_of_data *dn,
1047 		block_t *new_blkaddr, struct f2fs_io_info *fio)
1048 {
1049 	struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
1050 	struct f2fs_summary sum;
1051 	struct node_info ni;
1052 
1053 	f2fs_bug_on(dn->data_blkaddr == NULL_ADDR);
1054 	get_node_info(sbi, dn->nid, &ni);
1055 	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1056 
1057 	do_write_page(sbi, page, dn->data_blkaddr, new_blkaddr, &sum, fio);
1058 }
1059 
1060 void rewrite_data_page(struct page *page, block_t old_blkaddr,
1061 					struct f2fs_io_info *fio)
1062 {
1063 	struct inode *inode = page->mapping->host;
1064 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1065 	f2fs_submit_page_mbio(sbi, page, old_blkaddr, fio);
1066 }
1067 
1068 void recover_data_page(struct f2fs_sb_info *sbi,
1069 			struct page *page, struct f2fs_summary *sum,
1070 			block_t old_blkaddr, block_t new_blkaddr)
1071 {
1072 	struct sit_info *sit_i = SIT_I(sbi);
1073 	struct curseg_info *curseg;
1074 	unsigned int segno, old_cursegno;
1075 	struct seg_entry *se;
1076 	int type;
1077 
1078 	segno = GET_SEGNO(sbi, new_blkaddr);
1079 	se = get_seg_entry(sbi, segno);
1080 	type = se->type;
1081 
1082 	if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1083 		if (old_blkaddr == NULL_ADDR)
1084 			type = CURSEG_COLD_DATA;
1085 		else
1086 			type = CURSEG_WARM_DATA;
1087 	}
1088 	curseg = CURSEG_I(sbi, type);
1089 
1090 	mutex_lock(&curseg->curseg_mutex);
1091 	mutex_lock(&sit_i->sentry_lock);
1092 
1093 	old_cursegno = curseg->segno;
1094 
1095 	/* change the current segment */
1096 	if (segno != curseg->segno) {
1097 		curseg->next_segno = segno;
1098 		change_curseg(sbi, type, true);
1099 	}
1100 
1101 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1102 	__add_sum_entry(sbi, type, sum);
1103 
1104 	refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1105 	locate_dirty_segment(sbi, old_cursegno);
1106 
1107 	mutex_unlock(&sit_i->sentry_lock);
1108 	mutex_unlock(&curseg->curseg_mutex);
1109 }
1110 
1111 void rewrite_node_page(struct f2fs_sb_info *sbi,
1112 			struct page *page, struct f2fs_summary *sum,
1113 			block_t old_blkaddr, block_t new_blkaddr)
1114 {
1115 	struct sit_info *sit_i = SIT_I(sbi);
1116 	int type = CURSEG_WARM_NODE;
1117 	struct curseg_info *curseg;
1118 	unsigned int segno, old_cursegno;
1119 	block_t next_blkaddr = next_blkaddr_of_node(page);
1120 	unsigned int next_segno = GET_SEGNO(sbi, next_blkaddr);
1121 	struct f2fs_io_info fio = {
1122 		.type = NODE,
1123 		.rw = WRITE_SYNC,
1124 	};
1125 
1126 	curseg = CURSEG_I(sbi, type);
1127 
1128 	mutex_lock(&curseg->curseg_mutex);
1129 	mutex_lock(&sit_i->sentry_lock);
1130 
1131 	segno = GET_SEGNO(sbi, new_blkaddr);
1132 	old_cursegno = curseg->segno;
1133 
1134 	/* change the current segment */
1135 	if (segno != curseg->segno) {
1136 		curseg->next_segno = segno;
1137 		change_curseg(sbi, type, true);
1138 	}
1139 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1140 	__add_sum_entry(sbi, type, sum);
1141 
1142 	/* change the current log to the next block addr in advance */
1143 	if (next_segno != segno) {
1144 		curseg->next_segno = next_segno;
1145 		change_curseg(sbi, type, true);
1146 	}
1147 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, next_blkaddr);
1148 
1149 	/* rewrite node page */
1150 	set_page_writeback(page);
1151 	f2fs_submit_page_mbio(sbi, page, new_blkaddr, &fio);
1152 	f2fs_submit_merged_bio(sbi, NODE, WRITE);
1153 	refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1154 	locate_dirty_segment(sbi, old_cursegno);
1155 
1156 	mutex_unlock(&sit_i->sentry_lock);
1157 	mutex_unlock(&curseg->curseg_mutex);
1158 }
1159 
1160 static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1161 					struct page *page, enum page_type type)
1162 {
1163 	enum page_type btype = PAGE_TYPE_OF_BIO(type);
1164 	struct f2fs_bio_info *io = &sbi->write_io[btype];
1165 	struct bio_vec *bvec;
1166 	int i;
1167 
1168 	down_read(&io->io_rwsem);
1169 	if (!io->bio)
1170 		goto out;
1171 
1172 	bio_for_each_segment_all(bvec, io->bio, i) {
1173 		if (page == bvec->bv_page) {
1174 			up_read(&io->io_rwsem);
1175 			return true;
1176 		}
1177 	}
1178 
1179 out:
1180 	up_read(&io->io_rwsem);
1181 	return false;
1182 }
1183 
1184 void f2fs_wait_on_page_writeback(struct page *page,
1185 				enum page_type type)
1186 {
1187 	struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1188 	if (PageWriteback(page)) {
1189 		if (is_merged_page(sbi, page, type))
1190 			f2fs_submit_merged_bio(sbi, type, WRITE);
1191 		wait_on_page_writeback(page);
1192 	}
1193 }
1194 
1195 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1196 {
1197 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1198 	struct curseg_info *seg_i;
1199 	unsigned char *kaddr;
1200 	struct page *page;
1201 	block_t start;
1202 	int i, j, offset;
1203 
1204 	start = start_sum_block(sbi);
1205 
1206 	page = get_meta_page(sbi, start++);
1207 	kaddr = (unsigned char *)page_address(page);
1208 
1209 	/* Step 1: restore nat cache */
1210 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1211 	memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1212 
1213 	/* Step 2: restore sit cache */
1214 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1215 	memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1216 						SUM_JOURNAL_SIZE);
1217 	offset = 2 * SUM_JOURNAL_SIZE;
1218 
1219 	/* Step 3: restore summary entries */
1220 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1221 		unsigned short blk_off;
1222 		unsigned int segno;
1223 
1224 		seg_i = CURSEG_I(sbi, i);
1225 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1226 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1227 		seg_i->next_segno = segno;
1228 		reset_curseg(sbi, i, 0);
1229 		seg_i->alloc_type = ckpt->alloc_type[i];
1230 		seg_i->next_blkoff = blk_off;
1231 
1232 		if (seg_i->alloc_type == SSR)
1233 			blk_off = sbi->blocks_per_seg;
1234 
1235 		for (j = 0; j < blk_off; j++) {
1236 			struct f2fs_summary *s;
1237 			s = (struct f2fs_summary *)(kaddr + offset);
1238 			seg_i->sum_blk->entries[j] = *s;
1239 			offset += SUMMARY_SIZE;
1240 			if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1241 						SUM_FOOTER_SIZE)
1242 				continue;
1243 
1244 			f2fs_put_page(page, 1);
1245 			page = NULL;
1246 
1247 			page = get_meta_page(sbi, start++);
1248 			kaddr = (unsigned char *)page_address(page);
1249 			offset = 0;
1250 		}
1251 	}
1252 	f2fs_put_page(page, 1);
1253 	return 0;
1254 }
1255 
1256 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1257 {
1258 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1259 	struct f2fs_summary_block *sum;
1260 	struct curseg_info *curseg;
1261 	struct page *new;
1262 	unsigned short blk_off;
1263 	unsigned int segno = 0;
1264 	block_t blk_addr = 0;
1265 
1266 	/* get segment number and block addr */
1267 	if (IS_DATASEG(type)) {
1268 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1269 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1270 							CURSEG_HOT_DATA]);
1271 		if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1272 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1273 		else
1274 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1275 	} else {
1276 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
1277 							CURSEG_HOT_NODE]);
1278 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1279 							CURSEG_HOT_NODE]);
1280 		if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG))
1281 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1282 							type - CURSEG_HOT_NODE);
1283 		else
1284 			blk_addr = GET_SUM_BLOCK(sbi, segno);
1285 	}
1286 
1287 	new = get_meta_page(sbi, blk_addr);
1288 	sum = (struct f2fs_summary_block *)page_address(new);
1289 
1290 	if (IS_NODESEG(type)) {
1291 		if (is_set_ckpt_flags(ckpt, CP_UMOUNT_FLAG)) {
1292 			struct f2fs_summary *ns = &sum->entries[0];
1293 			int i;
1294 			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1295 				ns->version = 0;
1296 				ns->ofs_in_node = 0;
1297 			}
1298 		} else {
1299 			int err;
1300 
1301 			err = restore_node_summary(sbi, segno, sum);
1302 			if (err) {
1303 				f2fs_put_page(new, 1);
1304 				return err;
1305 			}
1306 		}
1307 	}
1308 
1309 	/* set uncompleted segment to curseg */
1310 	curseg = CURSEG_I(sbi, type);
1311 	mutex_lock(&curseg->curseg_mutex);
1312 	memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1313 	curseg->next_segno = segno;
1314 	reset_curseg(sbi, type, 0);
1315 	curseg->alloc_type = ckpt->alloc_type[type];
1316 	curseg->next_blkoff = blk_off;
1317 	mutex_unlock(&curseg->curseg_mutex);
1318 	f2fs_put_page(new, 1);
1319 	return 0;
1320 }
1321 
1322 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1323 {
1324 	int type = CURSEG_HOT_DATA;
1325 	int err;
1326 
1327 	if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1328 		/* restore for compacted data summary */
1329 		if (read_compacted_summaries(sbi))
1330 			return -EINVAL;
1331 		type = CURSEG_HOT_NODE;
1332 	}
1333 
1334 	for (; type <= CURSEG_COLD_NODE; type++) {
1335 		err = read_normal_summaries(sbi, type);
1336 		if (err)
1337 			return err;
1338 	}
1339 
1340 	return 0;
1341 }
1342 
1343 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1344 {
1345 	struct page *page;
1346 	unsigned char *kaddr;
1347 	struct f2fs_summary *summary;
1348 	struct curseg_info *seg_i;
1349 	int written_size = 0;
1350 	int i, j;
1351 
1352 	page = grab_meta_page(sbi, blkaddr++);
1353 	kaddr = (unsigned char *)page_address(page);
1354 
1355 	/* Step 1: write nat cache */
1356 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1357 	memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1358 	written_size += SUM_JOURNAL_SIZE;
1359 
1360 	/* Step 2: write sit cache */
1361 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1362 	memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1363 						SUM_JOURNAL_SIZE);
1364 	written_size += SUM_JOURNAL_SIZE;
1365 
1366 	/* Step 3: write summary entries */
1367 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1368 		unsigned short blkoff;
1369 		seg_i = CURSEG_I(sbi, i);
1370 		if (sbi->ckpt->alloc_type[i] == SSR)
1371 			blkoff = sbi->blocks_per_seg;
1372 		else
1373 			blkoff = curseg_blkoff(sbi, i);
1374 
1375 		for (j = 0; j < blkoff; j++) {
1376 			if (!page) {
1377 				page = grab_meta_page(sbi, blkaddr++);
1378 				kaddr = (unsigned char *)page_address(page);
1379 				written_size = 0;
1380 			}
1381 			summary = (struct f2fs_summary *)(kaddr + written_size);
1382 			*summary = seg_i->sum_blk->entries[j];
1383 			written_size += SUMMARY_SIZE;
1384 
1385 			if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1386 							SUM_FOOTER_SIZE)
1387 				continue;
1388 
1389 			set_page_dirty(page);
1390 			f2fs_put_page(page, 1);
1391 			page = NULL;
1392 		}
1393 	}
1394 	if (page) {
1395 		set_page_dirty(page);
1396 		f2fs_put_page(page, 1);
1397 	}
1398 }
1399 
1400 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1401 					block_t blkaddr, int type)
1402 {
1403 	int i, end;
1404 	if (IS_DATASEG(type))
1405 		end = type + NR_CURSEG_DATA_TYPE;
1406 	else
1407 		end = type + NR_CURSEG_NODE_TYPE;
1408 
1409 	for (i = type; i < end; i++) {
1410 		struct curseg_info *sum = CURSEG_I(sbi, i);
1411 		mutex_lock(&sum->curseg_mutex);
1412 		write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1413 		mutex_unlock(&sum->curseg_mutex);
1414 	}
1415 }
1416 
1417 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1418 {
1419 	if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1420 		write_compacted_summaries(sbi, start_blk);
1421 	else
1422 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1423 }
1424 
1425 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1426 {
1427 	if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_UMOUNT_FLAG))
1428 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1429 }
1430 
1431 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1432 					unsigned int val, int alloc)
1433 {
1434 	int i;
1435 
1436 	if (type == NAT_JOURNAL) {
1437 		for (i = 0; i < nats_in_cursum(sum); i++) {
1438 			if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1439 				return i;
1440 		}
1441 		if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1442 			return update_nats_in_cursum(sum, 1);
1443 	} else if (type == SIT_JOURNAL) {
1444 		for (i = 0; i < sits_in_cursum(sum); i++)
1445 			if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1446 				return i;
1447 		if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1448 			return update_sits_in_cursum(sum, 1);
1449 	}
1450 	return -1;
1451 }
1452 
1453 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1454 					unsigned int segno)
1455 {
1456 	struct sit_info *sit_i = SIT_I(sbi);
1457 	unsigned int offset = SIT_BLOCK_OFFSET(sit_i, segno);
1458 	block_t blk_addr = sit_i->sit_base_addr + offset;
1459 
1460 	check_seg_range(sbi, segno);
1461 
1462 	/* calculate sit block address */
1463 	if (f2fs_test_bit(offset, sit_i->sit_bitmap))
1464 		blk_addr += sit_i->sit_blocks;
1465 
1466 	return get_meta_page(sbi, blk_addr);
1467 }
1468 
1469 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1470 					unsigned int start)
1471 {
1472 	struct sit_info *sit_i = SIT_I(sbi);
1473 	struct page *src_page, *dst_page;
1474 	pgoff_t src_off, dst_off;
1475 	void *src_addr, *dst_addr;
1476 
1477 	src_off = current_sit_addr(sbi, start);
1478 	dst_off = next_sit_addr(sbi, src_off);
1479 
1480 	/* get current sit block page without lock */
1481 	src_page = get_meta_page(sbi, src_off);
1482 	dst_page = grab_meta_page(sbi, dst_off);
1483 	f2fs_bug_on(PageDirty(src_page));
1484 
1485 	src_addr = page_address(src_page);
1486 	dst_addr = page_address(dst_page);
1487 	memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1488 
1489 	set_page_dirty(dst_page);
1490 	f2fs_put_page(src_page, 1);
1491 
1492 	set_to_next_sit(sit_i, start);
1493 
1494 	return dst_page;
1495 }
1496 
1497 static bool flush_sits_in_journal(struct f2fs_sb_info *sbi)
1498 {
1499 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1500 	struct f2fs_summary_block *sum = curseg->sum_blk;
1501 	int i;
1502 
1503 	/*
1504 	 * If the journal area in the current summary is full of sit entries,
1505 	 * all the sit entries will be flushed. Otherwise the sit entries
1506 	 * are not able to replace with newly hot sit entries.
1507 	 */
1508 	if (sits_in_cursum(sum) >= SIT_JOURNAL_ENTRIES) {
1509 		for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1510 			unsigned int segno;
1511 			segno = le32_to_cpu(segno_in_journal(sum, i));
1512 			__mark_sit_entry_dirty(sbi, segno);
1513 		}
1514 		update_sits_in_cursum(sum, -sits_in_cursum(sum));
1515 		return true;
1516 	}
1517 	return false;
1518 }
1519 
1520 /*
1521  * CP calls this function, which flushes SIT entries including sit_journal,
1522  * and moves prefree segs to free segs.
1523  */
1524 void flush_sit_entries(struct f2fs_sb_info *sbi)
1525 {
1526 	struct sit_info *sit_i = SIT_I(sbi);
1527 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1528 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1529 	struct f2fs_summary_block *sum = curseg->sum_blk;
1530 	unsigned long nsegs = TOTAL_SEGS(sbi);
1531 	struct page *page = NULL;
1532 	struct f2fs_sit_block *raw_sit = NULL;
1533 	unsigned int start = 0, end = 0;
1534 	unsigned int segno = -1;
1535 	bool flushed;
1536 
1537 	mutex_lock(&curseg->curseg_mutex);
1538 	mutex_lock(&sit_i->sentry_lock);
1539 
1540 	/*
1541 	 * "flushed" indicates whether sit entries in journal are flushed
1542 	 * to the SIT area or not.
1543 	 */
1544 	flushed = flush_sits_in_journal(sbi);
1545 
1546 	while ((segno = find_next_bit(bitmap, nsegs, segno + 1)) < nsegs) {
1547 		struct seg_entry *se = get_seg_entry(sbi, segno);
1548 		int sit_offset, offset;
1549 
1550 		sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1551 
1552 		/* add discard candidates */
1553 		if (SM_I(sbi)->nr_discards < SM_I(sbi)->max_discards)
1554 			add_discard_addrs(sbi, segno, se);
1555 
1556 		if (flushed)
1557 			goto to_sit_page;
1558 
1559 		offset = lookup_journal_in_cursum(sum, SIT_JOURNAL, segno, 1);
1560 		if (offset >= 0) {
1561 			segno_in_journal(sum, offset) = cpu_to_le32(segno);
1562 			seg_info_to_raw_sit(se, &sit_in_journal(sum, offset));
1563 			goto flush_done;
1564 		}
1565 to_sit_page:
1566 		if (!page || (start > segno) || (segno > end)) {
1567 			if (page) {
1568 				f2fs_put_page(page, 1);
1569 				page = NULL;
1570 			}
1571 
1572 			start = START_SEGNO(sit_i, segno);
1573 			end = start + SIT_ENTRY_PER_BLOCK - 1;
1574 
1575 			/* read sit block that will be updated */
1576 			page = get_next_sit_page(sbi, start);
1577 			raw_sit = page_address(page);
1578 		}
1579 
1580 		/* udpate entry in SIT block */
1581 		seg_info_to_raw_sit(se, &raw_sit->entries[sit_offset]);
1582 flush_done:
1583 		__clear_bit(segno, bitmap);
1584 		sit_i->dirty_sentries--;
1585 	}
1586 	mutex_unlock(&sit_i->sentry_lock);
1587 	mutex_unlock(&curseg->curseg_mutex);
1588 
1589 	/* writeout last modified SIT block */
1590 	f2fs_put_page(page, 1);
1591 
1592 	set_prefree_as_free_segments(sbi);
1593 }
1594 
1595 static int build_sit_info(struct f2fs_sb_info *sbi)
1596 {
1597 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1598 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1599 	struct sit_info *sit_i;
1600 	unsigned int sit_segs, start;
1601 	char *src_bitmap, *dst_bitmap;
1602 	unsigned int bitmap_size;
1603 
1604 	/* allocate memory for SIT information */
1605 	sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1606 	if (!sit_i)
1607 		return -ENOMEM;
1608 
1609 	SM_I(sbi)->sit_info = sit_i;
1610 
1611 	sit_i->sentries = vzalloc(TOTAL_SEGS(sbi) * sizeof(struct seg_entry));
1612 	if (!sit_i->sentries)
1613 		return -ENOMEM;
1614 
1615 	bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1616 	sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1617 	if (!sit_i->dirty_sentries_bitmap)
1618 		return -ENOMEM;
1619 
1620 	for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1621 		sit_i->sentries[start].cur_valid_map
1622 			= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1623 		sit_i->sentries[start].ckpt_valid_map
1624 			= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1625 		if (!sit_i->sentries[start].cur_valid_map
1626 				|| !sit_i->sentries[start].ckpt_valid_map)
1627 			return -ENOMEM;
1628 	}
1629 
1630 	if (sbi->segs_per_sec > 1) {
1631 		sit_i->sec_entries = vzalloc(TOTAL_SECS(sbi) *
1632 					sizeof(struct sec_entry));
1633 		if (!sit_i->sec_entries)
1634 			return -ENOMEM;
1635 	}
1636 
1637 	/* get information related with SIT */
1638 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1639 
1640 	/* setup SIT bitmap from ckeckpoint pack */
1641 	bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1642 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1643 
1644 	dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1645 	if (!dst_bitmap)
1646 		return -ENOMEM;
1647 
1648 	/* init SIT information */
1649 	sit_i->s_ops = &default_salloc_ops;
1650 
1651 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1652 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1653 	sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1654 	sit_i->sit_bitmap = dst_bitmap;
1655 	sit_i->bitmap_size = bitmap_size;
1656 	sit_i->dirty_sentries = 0;
1657 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1658 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1659 	sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1660 	mutex_init(&sit_i->sentry_lock);
1661 	return 0;
1662 }
1663 
1664 static int build_free_segmap(struct f2fs_sb_info *sbi)
1665 {
1666 	struct f2fs_sm_info *sm_info = SM_I(sbi);
1667 	struct free_segmap_info *free_i;
1668 	unsigned int bitmap_size, sec_bitmap_size;
1669 
1670 	/* allocate memory for free segmap information */
1671 	free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1672 	if (!free_i)
1673 		return -ENOMEM;
1674 
1675 	SM_I(sbi)->free_info = free_i;
1676 
1677 	bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1678 	free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1679 	if (!free_i->free_segmap)
1680 		return -ENOMEM;
1681 
1682 	sec_bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1683 	free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1684 	if (!free_i->free_secmap)
1685 		return -ENOMEM;
1686 
1687 	/* set all segments as dirty temporarily */
1688 	memset(free_i->free_segmap, 0xff, bitmap_size);
1689 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1690 
1691 	/* init free segmap information */
1692 	free_i->start_segno =
1693 		(unsigned int) GET_SEGNO_FROM_SEG0(sbi, sm_info->main_blkaddr);
1694 	free_i->free_segments = 0;
1695 	free_i->free_sections = 0;
1696 	rwlock_init(&free_i->segmap_lock);
1697 	return 0;
1698 }
1699 
1700 static int build_curseg(struct f2fs_sb_info *sbi)
1701 {
1702 	struct curseg_info *array;
1703 	int i;
1704 
1705 	array = kzalloc(sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
1706 	if (!array)
1707 		return -ENOMEM;
1708 
1709 	SM_I(sbi)->curseg_array = array;
1710 
1711 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
1712 		mutex_init(&array[i].curseg_mutex);
1713 		array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1714 		if (!array[i].sum_blk)
1715 			return -ENOMEM;
1716 		array[i].segno = NULL_SEGNO;
1717 		array[i].next_blkoff = 0;
1718 	}
1719 	return restore_curseg_summaries(sbi);
1720 }
1721 
1722 static void build_sit_entries(struct f2fs_sb_info *sbi)
1723 {
1724 	struct sit_info *sit_i = SIT_I(sbi);
1725 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1726 	struct f2fs_summary_block *sum = curseg->sum_blk;
1727 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
1728 	unsigned int i, start, end;
1729 	unsigned int readed, start_blk = 0;
1730 	int nrpages = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1731 
1732 	do {
1733 		readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
1734 
1735 		start = start_blk * sit_i->sents_per_block;
1736 		end = (start_blk + readed) * sit_i->sents_per_block;
1737 
1738 		for (; start < end && start < TOTAL_SEGS(sbi); start++) {
1739 			struct seg_entry *se = &sit_i->sentries[start];
1740 			struct f2fs_sit_block *sit_blk;
1741 			struct f2fs_sit_entry sit;
1742 			struct page *page;
1743 
1744 			mutex_lock(&curseg->curseg_mutex);
1745 			for (i = 0; i < sits_in_cursum(sum); i++) {
1746 				if (le32_to_cpu(segno_in_journal(sum, i))
1747 								== start) {
1748 					sit = sit_in_journal(sum, i);
1749 					mutex_unlock(&curseg->curseg_mutex);
1750 					goto got_it;
1751 				}
1752 			}
1753 			mutex_unlock(&curseg->curseg_mutex);
1754 
1755 			page = get_current_sit_page(sbi, start);
1756 			sit_blk = (struct f2fs_sit_block *)page_address(page);
1757 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
1758 			f2fs_put_page(page, 1);
1759 got_it:
1760 			check_block_count(sbi, start, &sit);
1761 			seg_info_from_raw_sit(se, &sit);
1762 			if (sbi->segs_per_sec > 1) {
1763 				struct sec_entry *e = get_sec_entry(sbi, start);
1764 				e->valid_blocks += se->valid_blocks;
1765 			}
1766 		}
1767 		start_blk += readed;
1768 	} while (start_blk < sit_blk_cnt);
1769 }
1770 
1771 static void init_free_segmap(struct f2fs_sb_info *sbi)
1772 {
1773 	unsigned int start;
1774 	int type;
1775 
1776 	for (start = 0; start < TOTAL_SEGS(sbi); start++) {
1777 		struct seg_entry *sentry = get_seg_entry(sbi, start);
1778 		if (!sentry->valid_blocks)
1779 			__set_free(sbi, start);
1780 	}
1781 
1782 	/* set use the current segments */
1783 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
1784 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
1785 		__set_test_and_inuse(sbi, curseg_t->segno);
1786 	}
1787 }
1788 
1789 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
1790 {
1791 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1792 	struct free_segmap_info *free_i = FREE_I(sbi);
1793 	unsigned int segno = 0, offset = 0, total_segs = TOTAL_SEGS(sbi);
1794 	unsigned short valid_blocks;
1795 
1796 	while (1) {
1797 		/* find dirty segment based on free segmap */
1798 		segno = find_next_inuse(free_i, total_segs, offset);
1799 		if (segno >= total_segs)
1800 			break;
1801 		offset = segno + 1;
1802 		valid_blocks = get_valid_blocks(sbi, segno, 0);
1803 		if (valid_blocks >= sbi->blocks_per_seg || !valid_blocks)
1804 			continue;
1805 		mutex_lock(&dirty_i->seglist_lock);
1806 		__locate_dirty_segment(sbi, segno, DIRTY);
1807 		mutex_unlock(&dirty_i->seglist_lock);
1808 	}
1809 }
1810 
1811 static int init_victim_secmap(struct f2fs_sb_info *sbi)
1812 {
1813 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1814 	unsigned int bitmap_size = f2fs_bitmap_size(TOTAL_SECS(sbi));
1815 
1816 	dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
1817 	if (!dirty_i->victim_secmap)
1818 		return -ENOMEM;
1819 	return 0;
1820 }
1821 
1822 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
1823 {
1824 	struct dirty_seglist_info *dirty_i;
1825 	unsigned int bitmap_size, i;
1826 
1827 	/* allocate memory for dirty segments list information */
1828 	dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
1829 	if (!dirty_i)
1830 		return -ENOMEM;
1831 
1832 	SM_I(sbi)->dirty_info = dirty_i;
1833 	mutex_init(&dirty_i->seglist_lock);
1834 
1835 	bitmap_size = f2fs_bitmap_size(TOTAL_SEGS(sbi));
1836 
1837 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
1838 		dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
1839 		if (!dirty_i->dirty_segmap[i])
1840 			return -ENOMEM;
1841 	}
1842 
1843 	init_dirty_segmap(sbi);
1844 	return init_victim_secmap(sbi);
1845 }
1846 
1847 /*
1848  * Update min, max modified time for cost-benefit GC algorithm
1849  */
1850 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
1851 {
1852 	struct sit_info *sit_i = SIT_I(sbi);
1853 	unsigned int segno;
1854 
1855 	mutex_lock(&sit_i->sentry_lock);
1856 
1857 	sit_i->min_mtime = LLONG_MAX;
1858 
1859 	for (segno = 0; segno < TOTAL_SEGS(sbi); segno += sbi->segs_per_sec) {
1860 		unsigned int i;
1861 		unsigned long long mtime = 0;
1862 
1863 		for (i = 0; i < sbi->segs_per_sec; i++)
1864 			mtime += get_seg_entry(sbi, segno + i)->mtime;
1865 
1866 		mtime = div_u64(mtime, sbi->segs_per_sec);
1867 
1868 		if (sit_i->min_mtime > mtime)
1869 			sit_i->min_mtime = mtime;
1870 	}
1871 	sit_i->max_mtime = get_mtime(sbi);
1872 	mutex_unlock(&sit_i->sentry_lock);
1873 }
1874 
1875 int build_segment_manager(struct f2fs_sb_info *sbi)
1876 {
1877 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1878 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1879 	struct f2fs_sm_info *sm_info;
1880 	int err;
1881 
1882 	sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
1883 	if (!sm_info)
1884 		return -ENOMEM;
1885 
1886 	/* init sm info */
1887 	sbi->sm_info = sm_info;
1888 	INIT_LIST_HEAD(&sm_info->wblist_head);
1889 	spin_lock_init(&sm_info->wblist_lock);
1890 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
1891 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
1892 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
1893 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
1894 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
1895 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
1896 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
1897 	sm_info->rec_prefree_segments = sm_info->main_segments *
1898 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
1899 	sm_info->ipu_policy = F2FS_IPU_DISABLE;
1900 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
1901 
1902 	INIT_LIST_HEAD(&sm_info->discard_list);
1903 	sm_info->nr_discards = 0;
1904 	sm_info->max_discards = 0;
1905 
1906 	if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
1907 		err = create_flush_cmd_control(sbi);
1908 		if (err)
1909 			return err;
1910 	}
1911 
1912 	err = build_sit_info(sbi);
1913 	if (err)
1914 		return err;
1915 	err = build_free_segmap(sbi);
1916 	if (err)
1917 		return err;
1918 	err = build_curseg(sbi);
1919 	if (err)
1920 		return err;
1921 
1922 	/* reinit free segmap based on SIT */
1923 	build_sit_entries(sbi);
1924 
1925 	init_free_segmap(sbi);
1926 	err = build_dirty_segmap(sbi);
1927 	if (err)
1928 		return err;
1929 
1930 	init_min_max_mtime(sbi);
1931 	return 0;
1932 }
1933 
1934 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
1935 		enum dirty_type dirty_type)
1936 {
1937 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1938 
1939 	mutex_lock(&dirty_i->seglist_lock);
1940 	kfree(dirty_i->dirty_segmap[dirty_type]);
1941 	dirty_i->nr_dirty[dirty_type] = 0;
1942 	mutex_unlock(&dirty_i->seglist_lock);
1943 }
1944 
1945 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
1946 {
1947 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1948 	kfree(dirty_i->victim_secmap);
1949 }
1950 
1951 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
1952 {
1953 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1954 	int i;
1955 
1956 	if (!dirty_i)
1957 		return;
1958 
1959 	/* discard pre-free/dirty segments list */
1960 	for (i = 0; i < NR_DIRTY_TYPE; i++)
1961 		discard_dirty_segmap(sbi, i);
1962 
1963 	destroy_victim_secmap(sbi);
1964 	SM_I(sbi)->dirty_info = NULL;
1965 	kfree(dirty_i);
1966 }
1967 
1968 static void destroy_curseg(struct f2fs_sb_info *sbi)
1969 {
1970 	struct curseg_info *array = SM_I(sbi)->curseg_array;
1971 	int i;
1972 
1973 	if (!array)
1974 		return;
1975 	SM_I(sbi)->curseg_array = NULL;
1976 	for (i = 0; i < NR_CURSEG_TYPE; i++)
1977 		kfree(array[i].sum_blk);
1978 	kfree(array);
1979 }
1980 
1981 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
1982 {
1983 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
1984 	if (!free_i)
1985 		return;
1986 	SM_I(sbi)->free_info = NULL;
1987 	kfree(free_i->free_segmap);
1988 	kfree(free_i->free_secmap);
1989 	kfree(free_i);
1990 }
1991 
1992 static void destroy_sit_info(struct f2fs_sb_info *sbi)
1993 {
1994 	struct sit_info *sit_i = SIT_I(sbi);
1995 	unsigned int start;
1996 
1997 	if (!sit_i)
1998 		return;
1999 
2000 	if (sit_i->sentries) {
2001 		for (start = 0; start < TOTAL_SEGS(sbi); start++) {
2002 			kfree(sit_i->sentries[start].cur_valid_map);
2003 			kfree(sit_i->sentries[start].ckpt_valid_map);
2004 		}
2005 	}
2006 	vfree(sit_i->sentries);
2007 	vfree(sit_i->sec_entries);
2008 	kfree(sit_i->dirty_sentries_bitmap);
2009 
2010 	SM_I(sbi)->sit_info = NULL;
2011 	kfree(sit_i->sit_bitmap);
2012 	kfree(sit_i);
2013 }
2014 
2015 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2016 {
2017 	struct f2fs_sm_info *sm_info = SM_I(sbi);
2018 
2019 	if (!sm_info)
2020 		return;
2021 	destroy_flush_cmd_control(sbi);
2022 	destroy_dirty_segmap(sbi);
2023 	destroy_curseg(sbi);
2024 	destroy_free_segmap(sbi);
2025 	destroy_sit_info(sbi);
2026 	sbi->sm_info = NULL;
2027 	kfree(sm_info);
2028 }
2029 
2030 int __init create_segment_manager_caches(void)
2031 {
2032 	discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2033 			sizeof(struct discard_entry));
2034 	if (!discard_entry_slab)
2035 		return -ENOMEM;
2036 	return 0;
2037 }
2038 
2039 void destroy_segment_manager_caches(void)
2040 {
2041 	kmem_cache_destroy(discard_entry_slab);
2042 }
2043