xref: /openbmc/linux/fs/f2fs/segment.c (revision 19f106bc03e62739961249a29916ee3602ac3de9)
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.h"
24 #include <trace/events/f2fs.h>
25 
26 #define __reverse_ffz(x) __reverse_ffs(~(x))
27 
28 static struct kmem_cache *discard_entry_slab;
29 static struct kmem_cache *sit_entry_set_slab;
30 static struct kmem_cache *inmem_entry_slab;
31 
32 /*
33  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
34  * MSB and LSB are reversed in a byte by f2fs_set_bit.
35  */
36 static inline unsigned long __reverse_ffs(unsigned long word)
37 {
38 	int num = 0;
39 
40 #if BITS_PER_LONG == 64
41 	if ((word & 0xffffffff) == 0) {
42 		num += 32;
43 		word >>= 32;
44 	}
45 #endif
46 	if ((word & 0xffff) == 0) {
47 		num += 16;
48 		word >>= 16;
49 	}
50 	if ((word & 0xff) == 0) {
51 		num += 8;
52 		word >>= 8;
53 	}
54 	if ((word & 0xf0) == 0)
55 		num += 4;
56 	else
57 		word >>= 4;
58 	if ((word & 0xc) == 0)
59 		num += 2;
60 	else
61 		word >>= 2;
62 	if ((word & 0x2) == 0)
63 		num += 1;
64 	return num;
65 }
66 
67 /*
68  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
69  * f2fs_set_bit makes MSB and LSB reversed in a byte.
70  * Example:
71  *                             LSB <--> MSB
72  *   f2fs_set_bit(0, bitmap) => 0000 0001
73  *   f2fs_set_bit(7, bitmap) => 1000 0000
74  */
75 static unsigned long __find_rev_next_bit(const unsigned long *addr,
76 			unsigned long size, unsigned long offset)
77 {
78 	const unsigned long *p = addr + BIT_WORD(offset);
79 	unsigned long result = offset & ~(BITS_PER_LONG - 1);
80 	unsigned long tmp;
81 	unsigned long mask, submask;
82 	unsigned long quot, rest;
83 
84 	if (offset >= size)
85 		return size;
86 
87 	size -= result;
88 	offset %= BITS_PER_LONG;
89 	if (!offset)
90 		goto aligned;
91 
92 	tmp = *(p++);
93 	quot = (offset >> 3) << 3;
94 	rest = offset & 0x7;
95 	mask = ~0UL << quot;
96 	submask = (unsigned char)(0xff << rest) >> rest;
97 	submask <<= quot;
98 	mask &= submask;
99 	tmp &= mask;
100 	if (size < BITS_PER_LONG)
101 		goto found_first;
102 	if (tmp)
103 		goto found_middle;
104 
105 	size -= BITS_PER_LONG;
106 	result += BITS_PER_LONG;
107 aligned:
108 	while (size & ~(BITS_PER_LONG-1)) {
109 		tmp = *(p++);
110 		if (tmp)
111 			goto found_middle;
112 		result += BITS_PER_LONG;
113 		size -= BITS_PER_LONG;
114 	}
115 	if (!size)
116 		return result;
117 	tmp = *p;
118 found_first:
119 	tmp &= (~0UL >> (BITS_PER_LONG - size));
120 	if (tmp == 0UL)		/* Are any bits set? */
121 		return result + size;   /* Nope. */
122 found_middle:
123 	return result + __reverse_ffs(tmp);
124 }
125 
126 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
127 			unsigned long size, unsigned long offset)
128 {
129 	const unsigned long *p = addr + BIT_WORD(offset);
130 	unsigned long result = offset & ~(BITS_PER_LONG - 1);
131 	unsigned long tmp;
132 	unsigned long mask, submask;
133 	unsigned long quot, rest;
134 
135 	if (offset >= size)
136 		return size;
137 
138 	size -= result;
139 	offset %= BITS_PER_LONG;
140 	if (!offset)
141 		goto aligned;
142 
143 	tmp = *(p++);
144 	quot = (offset >> 3) << 3;
145 	rest = offset & 0x7;
146 	mask = ~(~0UL << quot);
147 	submask = (unsigned char)~((unsigned char)(0xff << rest) >> rest);
148 	submask <<= quot;
149 	mask += submask;
150 	tmp |= mask;
151 	if (size < BITS_PER_LONG)
152 		goto found_first;
153 	if (~tmp)
154 		goto found_middle;
155 
156 	size -= BITS_PER_LONG;
157 	result += BITS_PER_LONG;
158 aligned:
159 	while (size & ~(BITS_PER_LONG - 1)) {
160 		tmp = *(p++);
161 		if (~tmp)
162 			goto found_middle;
163 		result += BITS_PER_LONG;
164 		size -= BITS_PER_LONG;
165 	}
166 	if (!size)
167 		return result;
168 	tmp = *p;
169 
170 found_first:
171 	tmp |= ~0UL << size;
172 	if (tmp == ~0UL)        /* Are any bits zero? */
173 		return result + size;   /* Nope. */
174 found_middle:
175 	return result + __reverse_ffz(tmp);
176 }
177 
178 void register_inmem_page(struct inode *inode, struct page *page)
179 {
180 	struct f2fs_inode_info *fi = F2FS_I(inode);
181 	struct inmem_pages *new;
182 	int err;
183 
184 	SetPagePrivate(page);
185 	f2fs_trace_pid(page);
186 
187 	new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
188 
189 	/* add atomic page indices to the list */
190 	new->page = page;
191 	INIT_LIST_HEAD(&new->list);
192 retry:
193 	/* increase reference count with clean state */
194 	mutex_lock(&fi->inmem_lock);
195 	err = radix_tree_insert(&fi->inmem_root, page->index, new);
196 	if (err == -EEXIST) {
197 		mutex_unlock(&fi->inmem_lock);
198 		kmem_cache_free(inmem_entry_slab, new);
199 		return;
200 	} else if (err) {
201 		mutex_unlock(&fi->inmem_lock);
202 		goto retry;
203 	}
204 	get_page(page);
205 	list_add_tail(&new->list, &fi->inmem_pages);
206 	inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
207 	mutex_unlock(&fi->inmem_lock);
208 
209 	trace_f2fs_register_inmem_page(page, INMEM);
210 }
211 
212 void commit_inmem_pages(struct inode *inode, bool abort)
213 {
214 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
215 	struct f2fs_inode_info *fi = F2FS_I(inode);
216 	struct inmem_pages *cur, *tmp;
217 	bool submit_bio = false;
218 	struct f2fs_io_info fio = {
219 		.sbi = sbi,
220 		.type = DATA,
221 		.rw = WRITE_SYNC | REQ_PRIO,
222 	};
223 
224 	/*
225 	 * The abort is true only when f2fs_evict_inode is called.
226 	 * Basically, the f2fs_evict_inode doesn't produce any data writes, so
227 	 * that we don't need to call f2fs_balance_fs.
228 	 * Otherwise, f2fs_gc in f2fs_balance_fs can wait forever until this
229 	 * inode becomes free by iget_locked in f2fs_iget.
230 	 */
231 	if (!abort) {
232 		f2fs_balance_fs(sbi);
233 		f2fs_lock_op(sbi);
234 	}
235 
236 	mutex_lock(&fi->inmem_lock);
237 	list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
238 		if (!abort) {
239 			lock_page(cur->page);
240 			if (cur->page->mapping == inode->i_mapping) {
241 				f2fs_wait_on_page_writeback(cur->page, DATA);
242 				if (clear_page_dirty_for_io(cur->page))
243 					inode_dec_dirty_pages(inode);
244 				trace_f2fs_commit_inmem_page(cur->page, INMEM);
245 				fio.page = cur->page;
246 				do_write_data_page(&fio);
247 				submit_bio = true;
248 			}
249 			f2fs_put_page(cur->page, 1);
250 		} else {
251 			trace_f2fs_commit_inmem_page(cur->page, INMEM_DROP);
252 			put_page(cur->page);
253 		}
254 		radix_tree_delete(&fi->inmem_root, cur->page->index);
255 		list_del(&cur->list);
256 		kmem_cache_free(inmem_entry_slab, cur);
257 		dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
258 	}
259 	mutex_unlock(&fi->inmem_lock);
260 
261 	if (!abort) {
262 		f2fs_unlock_op(sbi);
263 		if (submit_bio)
264 			f2fs_submit_merged_bio(sbi, DATA, WRITE);
265 	}
266 }
267 
268 /*
269  * This function balances dirty node and dentry pages.
270  * In addition, it controls garbage collection.
271  */
272 void f2fs_balance_fs(struct f2fs_sb_info *sbi)
273 {
274 	/*
275 	 * We should do GC or end up with checkpoint, if there are so many dirty
276 	 * dir/node pages without enough free segments.
277 	 */
278 	if (has_not_enough_free_secs(sbi, 0)) {
279 		mutex_lock(&sbi->gc_mutex);
280 		f2fs_gc(sbi);
281 	}
282 }
283 
284 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
285 {
286 	/* try to shrink extent cache when there is no enough memory */
287 	f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
288 
289 	/* check the # of cached NAT entries and prefree segments */
290 	if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK) ||
291 			excess_prefree_segs(sbi) ||
292 			!available_free_memory(sbi, INO_ENTRIES))
293 		f2fs_sync_fs(sbi->sb, true);
294 }
295 
296 static int issue_flush_thread(void *data)
297 {
298 	struct f2fs_sb_info *sbi = data;
299 	struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
300 	wait_queue_head_t *q = &fcc->flush_wait_queue;
301 repeat:
302 	if (kthread_should_stop())
303 		return 0;
304 
305 	if (!llist_empty(&fcc->issue_list)) {
306 		struct bio *bio = bio_alloc(GFP_NOIO, 0);
307 		struct flush_cmd *cmd, *next;
308 		int ret;
309 
310 		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
311 		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
312 
313 		bio->bi_bdev = sbi->sb->s_bdev;
314 		ret = submit_bio_wait(WRITE_FLUSH, bio);
315 
316 		llist_for_each_entry_safe(cmd, next,
317 					  fcc->dispatch_list, llnode) {
318 			cmd->ret = ret;
319 			complete(&cmd->wait);
320 		}
321 		bio_put(bio);
322 		fcc->dispatch_list = NULL;
323 	}
324 
325 	wait_event_interruptible(*q,
326 		kthread_should_stop() || !llist_empty(&fcc->issue_list));
327 	goto repeat;
328 }
329 
330 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
331 {
332 	struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
333 	struct flush_cmd cmd;
334 
335 	trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
336 					test_opt(sbi, FLUSH_MERGE));
337 
338 	if (test_opt(sbi, NOBARRIER))
339 		return 0;
340 
341 	if (!test_opt(sbi, FLUSH_MERGE))
342 		return blkdev_issue_flush(sbi->sb->s_bdev, GFP_KERNEL, NULL);
343 
344 	init_completion(&cmd.wait);
345 
346 	llist_add(&cmd.llnode, &fcc->issue_list);
347 
348 	if (!fcc->dispatch_list)
349 		wake_up(&fcc->flush_wait_queue);
350 
351 	wait_for_completion(&cmd.wait);
352 
353 	return cmd.ret;
354 }
355 
356 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
357 {
358 	dev_t dev = sbi->sb->s_bdev->bd_dev;
359 	struct flush_cmd_control *fcc;
360 	int err = 0;
361 
362 	fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
363 	if (!fcc)
364 		return -ENOMEM;
365 	init_waitqueue_head(&fcc->flush_wait_queue);
366 	init_llist_head(&fcc->issue_list);
367 	SM_I(sbi)->cmd_control_info = fcc;
368 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
369 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
370 	if (IS_ERR(fcc->f2fs_issue_flush)) {
371 		err = PTR_ERR(fcc->f2fs_issue_flush);
372 		kfree(fcc);
373 		SM_I(sbi)->cmd_control_info = NULL;
374 		return err;
375 	}
376 
377 	return err;
378 }
379 
380 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
381 {
382 	struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
383 
384 	if (fcc && fcc->f2fs_issue_flush)
385 		kthread_stop(fcc->f2fs_issue_flush);
386 	kfree(fcc);
387 	SM_I(sbi)->cmd_control_info = NULL;
388 }
389 
390 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
391 		enum dirty_type dirty_type)
392 {
393 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
394 
395 	/* need not be added */
396 	if (IS_CURSEG(sbi, segno))
397 		return;
398 
399 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
400 		dirty_i->nr_dirty[dirty_type]++;
401 
402 	if (dirty_type == DIRTY) {
403 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
404 		enum dirty_type t = sentry->type;
405 
406 		if (unlikely(t >= DIRTY)) {
407 			f2fs_bug_on(sbi, 1);
408 			return;
409 		}
410 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
411 			dirty_i->nr_dirty[t]++;
412 	}
413 }
414 
415 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
416 		enum dirty_type dirty_type)
417 {
418 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
419 
420 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
421 		dirty_i->nr_dirty[dirty_type]--;
422 
423 	if (dirty_type == DIRTY) {
424 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
425 		enum dirty_type t = sentry->type;
426 
427 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
428 			dirty_i->nr_dirty[t]--;
429 
430 		if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
431 			clear_bit(GET_SECNO(sbi, segno),
432 						dirty_i->victim_secmap);
433 	}
434 }
435 
436 /*
437  * Should not occur error such as -ENOMEM.
438  * Adding dirty entry into seglist is not critical operation.
439  * If a given segment is one of current working segments, it won't be added.
440  */
441 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
442 {
443 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
444 	unsigned short valid_blocks;
445 
446 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
447 		return;
448 
449 	mutex_lock(&dirty_i->seglist_lock);
450 
451 	valid_blocks = get_valid_blocks(sbi, segno, 0);
452 
453 	if (valid_blocks == 0) {
454 		__locate_dirty_segment(sbi, segno, PRE);
455 		__remove_dirty_segment(sbi, segno, DIRTY);
456 	} else if (valid_blocks < sbi->blocks_per_seg) {
457 		__locate_dirty_segment(sbi, segno, DIRTY);
458 	} else {
459 		/* Recovery routine with SSR needs this */
460 		__remove_dirty_segment(sbi, segno, DIRTY);
461 	}
462 
463 	mutex_unlock(&dirty_i->seglist_lock);
464 }
465 
466 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
467 				block_t blkstart, block_t blklen)
468 {
469 	sector_t start = SECTOR_FROM_BLOCK(blkstart);
470 	sector_t len = SECTOR_FROM_BLOCK(blklen);
471 	struct seg_entry *se;
472 	unsigned int offset;
473 	block_t i;
474 
475 	for (i = blkstart; i < blkstart + blklen; i++) {
476 		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
477 		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
478 
479 		if (!f2fs_test_and_set_bit(offset, se->discard_map))
480 			sbi->discard_blks--;
481 	}
482 	trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
483 	return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
484 }
485 
486 void discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
487 {
488 	if (f2fs_issue_discard(sbi, blkaddr, 1)) {
489 		struct page *page = grab_meta_page(sbi, blkaddr);
490 		/* zero-filled page */
491 		set_page_dirty(page);
492 		f2fs_put_page(page, 1);
493 	}
494 }
495 
496 static void __add_discard_entry(struct f2fs_sb_info *sbi,
497 		struct cp_control *cpc, struct seg_entry *se,
498 		unsigned int start, unsigned int end)
499 {
500 	struct list_head *head = &SM_I(sbi)->discard_list;
501 	struct discard_entry *new, *last;
502 
503 	if (!list_empty(head)) {
504 		last = list_last_entry(head, struct discard_entry, list);
505 		if (START_BLOCK(sbi, cpc->trim_start) + start ==
506 						last->blkaddr + last->len) {
507 			last->len += end - start;
508 			goto done;
509 		}
510 	}
511 
512 	new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
513 	INIT_LIST_HEAD(&new->list);
514 	new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
515 	new->len = end - start;
516 	list_add_tail(&new->list, head);
517 done:
518 	SM_I(sbi)->nr_discards += end - start;
519 	cpc->trimmed += end - start;
520 }
521 
522 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
523 {
524 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
525 	int max_blocks = sbi->blocks_per_seg;
526 	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
527 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
528 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
529 	unsigned long *discard_map = (unsigned long *)se->discard_map;
530 	unsigned long *dmap = SIT_I(sbi)->tmp_map;
531 	unsigned int start = 0, end = -1;
532 	bool force = (cpc->reason == CP_DISCARD);
533 	int i;
534 
535 	if (se->valid_blocks == max_blocks)
536 		return;
537 
538 	if (!force) {
539 		if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
540 			SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
541 		return;
542 	}
543 
544 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
545 	for (i = 0; i < entries; i++)
546 		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
547 				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
548 
549 	while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
550 		start = __find_rev_next_bit(dmap, max_blocks, end + 1);
551 		if (start >= max_blocks)
552 			break;
553 
554 		end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
555 		__add_discard_entry(sbi, cpc, se, start, end);
556 	}
557 }
558 
559 void release_discard_addrs(struct f2fs_sb_info *sbi)
560 {
561 	struct list_head *head = &(SM_I(sbi)->discard_list);
562 	struct discard_entry *entry, *this;
563 
564 	/* drop caches */
565 	list_for_each_entry_safe(entry, this, head, list) {
566 		list_del(&entry->list);
567 		kmem_cache_free(discard_entry_slab, entry);
568 	}
569 }
570 
571 /*
572  * Should call clear_prefree_segments after checkpoint is done.
573  */
574 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
575 {
576 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
577 	unsigned int segno;
578 
579 	mutex_lock(&dirty_i->seglist_lock);
580 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
581 		__set_test_and_free(sbi, segno);
582 	mutex_unlock(&dirty_i->seglist_lock);
583 }
584 
585 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
586 {
587 	struct list_head *head = &(SM_I(sbi)->discard_list);
588 	struct discard_entry *entry, *this;
589 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
590 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
591 	unsigned int start = 0, end = -1;
592 
593 	mutex_lock(&dirty_i->seglist_lock);
594 
595 	while (1) {
596 		int i;
597 		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
598 		if (start >= MAIN_SEGS(sbi))
599 			break;
600 		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
601 								start + 1);
602 
603 		for (i = start; i < end; i++)
604 			clear_bit(i, prefree_map);
605 
606 		dirty_i->nr_dirty[PRE] -= end - start;
607 
608 		if (!test_opt(sbi, DISCARD))
609 			continue;
610 
611 		f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
612 				(end - start) << sbi->log_blocks_per_seg);
613 	}
614 	mutex_unlock(&dirty_i->seglist_lock);
615 
616 	/* send small discards */
617 	list_for_each_entry_safe(entry, this, head, list) {
618 		if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
619 			goto skip;
620 		f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
621 skip:
622 		list_del(&entry->list);
623 		SM_I(sbi)->nr_discards -= entry->len;
624 		kmem_cache_free(discard_entry_slab, entry);
625 	}
626 }
627 
628 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
629 {
630 	struct sit_info *sit_i = SIT_I(sbi);
631 
632 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
633 		sit_i->dirty_sentries++;
634 		return false;
635 	}
636 
637 	return true;
638 }
639 
640 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
641 					unsigned int segno, int modified)
642 {
643 	struct seg_entry *se = get_seg_entry(sbi, segno);
644 	se->type = type;
645 	if (modified)
646 		__mark_sit_entry_dirty(sbi, segno);
647 }
648 
649 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
650 {
651 	struct seg_entry *se;
652 	unsigned int segno, offset;
653 	long int new_vblocks;
654 
655 	segno = GET_SEGNO(sbi, blkaddr);
656 
657 	se = get_seg_entry(sbi, segno);
658 	new_vblocks = se->valid_blocks + del;
659 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
660 
661 	f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
662 				(new_vblocks > sbi->blocks_per_seg)));
663 
664 	se->valid_blocks = new_vblocks;
665 	se->mtime = get_mtime(sbi);
666 	SIT_I(sbi)->max_mtime = se->mtime;
667 
668 	/* Update valid block bitmap */
669 	if (del > 0) {
670 		if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
671 			f2fs_bug_on(sbi, 1);
672 		if (!f2fs_test_and_set_bit(offset, se->discard_map))
673 			sbi->discard_blks--;
674 	} else {
675 		if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
676 			f2fs_bug_on(sbi, 1);
677 		if (f2fs_test_and_clear_bit(offset, se->discard_map))
678 			sbi->discard_blks++;
679 	}
680 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
681 		se->ckpt_valid_blocks += del;
682 
683 	__mark_sit_entry_dirty(sbi, segno);
684 
685 	/* update total number of valid blocks to be written in ckpt area */
686 	SIT_I(sbi)->written_valid_blocks += del;
687 
688 	if (sbi->segs_per_sec > 1)
689 		get_sec_entry(sbi, segno)->valid_blocks += del;
690 }
691 
692 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
693 {
694 	update_sit_entry(sbi, new, 1);
695 	if (GET_SEGNO(sbi, old) != NULL_SEGNO)
696 		update_sit_entry(sbi, old, -1);
697 
698 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
699 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
700 }
701 
702 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
703 {
704 	unsigned int segno = GET_SEGNO(sbi, addr);
705 	struct sit_info *sit_i = SIT_I(sbi);
706 
707 	f2fs_bug_on(sbi, addr == NULL_ADDR);
708 	if (addr == NEW_ADDR)
709 		return;
710 
711 	/* add it into sit main buffer */
712 	mutex_lock(&sit_i->sentry_lock);
713 
714 	update_sit_entry(sbi, addr, -1);
715 
716 	/* add it into dirty seglist */
717 	locate_dirty_segment(sbi, segno);
718 
719 	mutex_unlock(&sit_i->sentry_lock);
720 }
721 
722 /*
723  * This function should be resided under the curseg_mutex lock
724  */
725 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
726 					struct f2fs_summary *sum)
727 {
728 	struct curseg_info *curseg = CURSEG_I(sbi, type);
729 	void *addr = curseg->sum_blk;
730 	addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
731 	memcpy(addr, sum, sizeof(struct f2fs_summary));
732 }
733 
734 /*
735  * Calculate the number of current summary pages for writing
736  */
737 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
738 {
739 	int valid_sum_count = 0;
740 	int i, sum_in_page;
741 
742 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
743 		if (sbi->ckpt->alloc_type[i] == SSR)
744 			valid_sum_count += sbi->blocks_per_seg;
745 		else {
746 			if (for_ra)
747 				valid_sum_count += le16_to_cpu(
748 					F2FS_CKPT(sbi)->cur_data_blkoff[i]);
749 			else
750 				valid_sum_count += curseg_blkoff(sbi, i);
751 		}
752 	}
753 
754 	sum_in_page = (PAGE_CACHE_SIZE - 2 * SUM_JOURNAL_SIZE -
755 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
756 	if (valid_sum_count <= sum_in_page)
757 		return 1;
758 	else if ((valid_sum_count - sum_in_page) <=
759 		(PAGE_CACHE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
760 		return 2;
761 	return 3;
762 }
763 
764 /*
765  * Caller should put this summary page
766  */
767 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
768 {
769 	return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
770 }
771 
772 static void write_sum_page(struct f2fs_sb_info *sbi,
773 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
774 {
775 	struct page *page = grab_meta_page(sbi, blk_addr);
776 	void *kaddr = page_address(page);
777 	memcpy(kaddr, sum_blk, PAGE_CACHE_SIZE);
778 	set_page_dirty(page);
779 	f2fs_put_page(page, 1);
780 }
781 
782 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
783 {
784 	struct curseg_info *curseg = CURSEG_I(sbi, type);
785 	unsigned int segno = curseg->segno + 1;
786 	struct free_segmap_info *free_i = FREE_I(sbi);
787 
788 	if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
789 		return !test_bit(segno, free_i->free_segmap);
790 	return 0;
791 }
792 
793 /*
794  * Find a new segment from the free segments bitmap to right order
795  * This function should be returned with success, otherwise BUG
796  */
797 static void get_new_segment(struct f2fs_sb_info *sbi,
798 			unsigned int *newseg, bool new_sec, int dir)
799 {
800 	struct free_segmap_info *free_i = FREE_I(sbi);
801 	unsigned int segno, secno, zoneno;
802 	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
803 	unsigned int hint = *newseg / sbi->segs_per_sec;
804 	unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
805 	unsigned int left_start = hint;
806 	bool init = true;
807 	int go_left = 0;
808 	int i;
809 
810 	spin_lock(&free_i->segmap_lock);
811 
812 	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
813 		segno = find_next_zero_bit(free_i->free_segmap,
814 					MAIN_SEGS(sbi), *newseg + 1);
815 		if (segno - *newseg < sbi->segs_per_sec -
816 					(*newseg % sbi->segs_per_sec))
817 			goto got_it;
818 	}
819 find_other_zone:
820 	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
821 	if (secno >= MAIN_SECS(sbi)) {
822 		if (dir == ALLOC_RIGHT) {
823 			secno = find_next_zero_bit(free_i->free_secmap,
824 							MAIN_SECS(sbi), 0);
825 			f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
826 		} else {
827 			go_left = 1;
828 			left_start = hint - 1;
829 		}
830 	}
831 	if (go_left == 0)
832 		goto skip_left;
833 
834 	while (test_bit(left_start, free_i->free_secmap)) {
835 		if (left_start > 0) {
836 			left_start--;
837 			continue;
838 		}
839 		left_start = find_next_zero_bit(free_i->free_secmap,
840 							MAIN_SECS(sbi), 0);
841 		f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
842 		break;
843 	}
844 	secno = left_start;
845 skip_left:
846 	hint = secno;
847 	segno = secno * sbi->segs_per_sec;
848 	zoneno = secno / sbi->secs_per_zone;
849 
850 	/* give up on finding another zone */
851 	if (!init)
852 		goto got_it;
853 	if (sbi->secs_per_zone == 1)
854 		goto got_it;
855 	if (zoneno == old_zoneno)
856 		goto got_it;
857 	if (dir == ALLOC_LEFT) {
858 		if (!go_left && zoneno + 1 >= total_zones)
859 			goto got_it;
860 		if (go_left && zoneno == 0)
861 			goto got_it;
862 	}
863 	for (i = 0; i < NR_CURSEG_TYPE; i++)
864 		if (CURSEG_I(sbi, i)->zone == zoneno)
865 			break;
866 
867 	if (i < NR_CURSEG_TYPE) {
868 		/* zone is in user, try another */
869 		if (go_left)
870 			hint = zoneno * sbi->secs_per_zone - 1;
871 		else if (zoneno + 1 >= total_zones)
872 			hint = 0;
873 		else
874 			hint = (zoneno + 1) * sbi->secs_per_zone;
875 		init = false;
876 		goto find_other_zone;
877 	}
878 got_it:
879 	/* set it as dirty segment in free segmap */
880 	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
881 	__set_inuse(sbi, segno);
882 	*newseg = segno;
883 	spin_unlock(&free_i->segmap_lock);
884 }
885 
886 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
887 {
888 	struct curseg_info *curseg = CURSEG_I(sbi, type);
889 	struct summary_footer *sum_footer;
890 
891 	curseg->segno = curseg->next_segno;
892 	curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
893 	curseg->next_blkoff = 0;
894 	curseg->next_segno = NULL_SEGNO;
895 
896 	sum_footer = &(curseg->sum_blk->footer);
897 	memset(sum_footer, 0, sizeof(struct summary_footer));
898 	if (IS_DATASEG(type))
899 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
900 	if (IS_NODESEG(type))
901 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
902 	__set_sit_entry_type(sbi, type, curseg->segno, modified);
903 }
904 
905 /*
906  * Allocate a current working segment.
907  * This function always allocates a free segment in LFS manner.
908  */
909 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
910 {
911 	struct curseg_info *curseg = CURSEG_I(sbi, type);
912 	unsigned int segno = curseg->segno;
913 	int dir = ALLOC_LEFT;
914 
915 	write_sum_page(sbi, curseg->sum_blk,
916 				GET_SUM_BLOCK(sbi, segno));
917 	if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
918 		dir = ALLOC_RIGHT;
919 
920 	if (test_opt(sbi, NOHEAP))
921 		dir = ALLOC_RIGHT;
922 
923 	get_new_segment(sbi, &segno, new_sec, dir);
924 	curseg->next_segno = segno;
925 	reset_curseg(sbi, type, 1);
926 	curseg->alloc_type = LFS;
927 }
928 
929 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
930 			struct curseg_info *seg, block_t start)
931 {
932 	struct seg_entry *se = get_seg_entry(sbi, seg->segno);
933 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
934 	unsigned long *target_map = SIT_I(sbi)->tmp_map;
935 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
936 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
937 	int i, pos;
938 
939 	for (i = 0; i < entries; i++)
940 		target_map[i] = ckpt_map[i] | cur_map[i];
941 
942 	pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
943 
944 	seg->next_blkoff = pos;
945 }
946 
947 /*
948  * If a segment is written by LFS manner, next block offset is just obtained
949  * by increasing the current block offset. However, if a segment is written by
950  * SSR manner, next block offset obtained by calling __next_free_blkoff
951  */
952 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
953 				struct curseg_info *seg)
954 {
955 	if (seg->alloc_type == SSR)
956 		__next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
957 	else
958 		seg->next_blkoff++;
959 }
960 
961 /*
962  * This function always allocates a used segment(from dirty seglist) by SSR
963  * manner, so it should recover the existing segment information of valid blocks
964  */
965 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
966 {
967 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
968 	struct curseg_info *curseg = CURSEG_I(sbi, type);
969 	unsigned int new_segno = curseg->next_segno;
970 	struct f2fs_summary_block *sum_node;
971 	struct page *sum_page;
972 
973 	write_sum_page(sbi, curseg->sum_blk,
974 				GET_SUM_BLOCK(sbi, curseg->segno));
975 	__set_test_and_inuse(sbi, new_segno);
976 
977 	mutex_lock(&dirty_i->seglist_lock);
978 	__remove_dirty_segment(sbi, new_segno, PRE);
979 	__remove_dirty_segment(sbi, new_segno, DIRTY);
980 	mutex_unlock(&dirty_i->seglist_lock);
981 
982 	reset_curseg(sbi, type, 1);
983 	curseg->alloc_type = SSR;
984 	__next_free_blkoff(sbi, curseg, 0);
985 
986 	if (reuse) {
987 		sum_page = get_sum_page(sbi, new_segno);
988 		sum_node = (struct f2fs_summary_block *)page_address(sum_page);
989 		memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
990 		f2fs_put_page(sum_page, 1);
991 	}
992 }
993 
994 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
995 {
996 	struct curseg_info *curseg = CURSEG_I(sbi, type);
997 	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
998 
999 	if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1000 		return v_ops->get_victim(sbi,
1001 				&(curseg)->next_segno, BG_GC, type, SSR);
1002 
1003 	/* For data segments, let's do SSR more intensively */
1004 	for (; type >= CURSEG_HOT_DATA; type--)
1005 		if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1006 						BG_GC, type, SSR))
1007 			return 1;
1008 	return 0;
1009 }
1010 
1011 /*
1012  * flush out current segment and replace it with new segment
1013  * This function should be returned with success, otherwise BUG
1014  */
1015 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1016 						int type, bool force)
1017 {
1018 	struct curseg_info *curseg = CURSEG_I(sbi, type);
1019 
1020 	if (force)
1021 		new_curseg(sbi, type, true);
1022 	else if (type == CURSEG_WARM_NODE)
1023 		new_curseg(sbi, type, false);
1024 	else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1025 		new_curseg(sbi, type, false);
1026 	else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1027 		change_curseg(sbi, type, true);
1028 	else
1029 		new_curseg(sbi, type, false);
1030 
1031 	stat_inc_seg_type(sbi, curseg);
1032 }
1033 
1034 static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1035 {
1036 	struct curseg_info *curseg = CURSEG_I(sbi, type);
1037 	unsigned int old_segno;
1038 
1039 	old_segno = curseg->segno;
1040 	SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1041 	locate_dirty_segment(sbi, old_segno);
1042 }
1043 
1044 void allocate_new_segments(struct f2fs_sb_info *sbi)
1045 {
1046 	int i;
1047 
1048 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1049 		__allocate_new_segments(sbi, i);
1050 }
1051 
1052 static const struct segment_allocation default_salloc_ops = {
1053 	.allocate_segment = allocate_segment_by_default,
1054 };
1055 
1056 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1057 {
1058 	__u64 start = F2FS_BYTES_TO_BLK(range->start);
1059 	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1060 	unsigned int start_segno, end_segno;
1061 	struct cp_control cpc;
1062 
1063 	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1064 		return -EINVAL;
1065 
1066 	cpc.trimmed = 0;
1067 	if (end <= MAIN_BLKADDR(sbi))
1068 		goto out;
1069 
1070 	/* start/end segment number in main_area */
1071 	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1072 	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1073 						GET_SEGNO(sbi, end);
1074 	cpc.reason = CP_DISCARD;
1075 	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1076 
1077 	/* do checkpoint to issue discard commands safely */
1078 	for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1079 		cpc.trim_start = start_segno;
1080 
1081 		if (sbi->discard_blks == 0)
1082 			break;
1083 		else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1084 			cpc.trim_end = end_segno;
1085 		else
1086 			cpc.trim_end = min_t(unsigned int,
1087 				rounddown(start_segno +
1088 				BATCHED_TRIM_SEGMENTS(sbi),
1089 				sbi->segs_per_sec) - 1, end_segno);
1090 
1091 		mutex_lock(&sbi->gc_mutex);
1092 		write_checkpoint(sbi, &cpc);
1093 		mutex_unlock(&sbi->gc_mutex);
1094 	}
1095 out:
1096 	range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1097 	return 0;
1098 }
1099 
1100 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1101 {
1102 	struct curseg_info *curseg = CURSEG_I(sbi, type);
1103 	if (curseg->next_blkoff < sbi->blocks_per_seg)
1104 		return true;
1105 	return false;
1106 }
1107 
1108 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1109 {
1110 	if (p_type == DATA)
1111 		return CURSEG_HOT_DATA;
1112 	else
1113 		return CURSEG_HOT_NODE;
1114 }
1115 
1116 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1117 {
1118 	if (p_type == DATA) {
1119 		struct inode *inode = page->mapping->host;
1120 
1121 		if (S_ISDIR(inode->i_mode))
1122 			return CURSEG_HOT_DATA;
1123 		else
1124 			return CURSEG_COLD_DATA;
1125 	} else {
1126 		if (IS_DNODE(page) && is_cold_node(page))
1127 			return CURSEG_WARM_NODE;
1128 		else
1129 			return CURSEG_COLD_NODE;
1130 	}
1131 }
1132 
1133 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1134 {
1135 	if (p_type == DATA) {
1136 		struct inode *inode = page->mapping->host;
1137 
1138 		if (S_ISDIR(inode->i_mode))
1139 			return CURSEG_HOT_DATA;
1140 		else if (is_cold_data(page) || file_is_cold(inode))
1141 			return CURSEG_COLD_DATA;
1142 		else
1143 			return CURSEG_WARM_DATA;
1144 	} else {
1145 		if (IS_DNODE(page))
1146 			return is_cold_node(page) ? CURSEG_WARM_NODE :
1147 						CURSEG_HOT_NODE;
1148 		else
1149 			return CURSEG_COLD_NODE;
1150 	}
1151 }
1152 
1153 static int __get_segment_type(struct page *page, enum page_type p_type)
1154 {
1155 	switch (F2FS_P_SB(page)->active_logs) {
1156 	case 2:
1157 		return __get_segment_type_2(page, p_type);
1158 	case 4:
1159 		return __get_segment_type_4(page, p_type);
1160 	}
1161 	/* NR_CURSEG_TYPE(6) logs by default */
1162 	f2fs_bug_on(F2FS_P_SB(page),
1163 		F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1164 	return __get_segment_type_6(page, p_type);
1165 }
1166 
1167 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1168 		block_t old_blkaddr, block_t *new_blkaddr,
1169 		struct f2fs_summary *sum, int type)
1170 {
1171 	struct sit_info *sit_i = SIT_I(sbi);
1172 	struct curseg_info *curseg;
1173 	bool direct_io = (type == CURSEG_DIRECT_IO);
1174 
1175 	type = direct_io ? CURSEG_WARM_DATA : type;
1176 
1177 	curseg = CURSEG_I(sbi, type);
1178 
1179 	mutex_lock(&curseg->curseg_mutex);
1180 	mutex_lock(&sit_i->sentry_lock);
1181 
1182 	/* direct_io'ed data is aligned to the segment for better performance */
1183 	if (direct_io && curseg->next_blkoff)
1184 		__allocate_new_segments(sbi, type);
1185 
1186 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1187 
1188 	/*
1189 	 * __add_sum_entry should be resided under the curseg_mutex
1190 	 * because, this function updates a summary entry in the
1191 	 * current summary block.
1192 	 */
1193 	__add_sum_entry(sbi, type, sum);
1194 
1195 	__refresh_next_blkoff(sbi, curseg);
1196 
1197 	stat_inc_block_count(sbi, curseg);
1198 
1199 	if (!__has_curseg_space(sbi, type))
1200 		sit_i->s_ops->allocate_segment(sbi, type, false);
1201 	/*
1202 	 * SIT information should be updated before segment allocation,
1203 	 * since SSR needs latest valid block information.
1204 	 */
1205 	refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1206 
1207 	mutex_unlock(&sit_i->sentry_lock);
1208 
1209 	if (page && IS_NODESEG(type))
1210 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1211 
1212 	mutex_unlock(&curseg->curseg_mutex);
1213 }
1214 
1215 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1216 {
1217 	int type = __get_segment_type(fio->page, fio->type);
1218 
1219 	allocate_data_block(fio->sbi, fio->page, fio->blk_addr,
1220 					&fio->blk_addr, sum, type);
1221 
1222 	/* writeout dirty page into bdev */
1223 	f2fs_submit_page_mbio(fio);
1224 }
1225 
1226 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1227 {
1228 	struct f2fs_io_info fio = {
1229 		.sbi = sbi,
1230 		.type = META,
1231 		.rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1232 		.blk_addr = page->index,
1233 		.page = page,
1234 	};
1235 
1236 	set_page_writeback(page);
1237 	f2fs_submit_page_mbio(&fio);
1238 }
1239 
1240 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1241 {
1242 	struct f2fs_summary sum;
1243 
1244 	set_summary(&sum, nid, 0, 0);
1245 	do_write_page(&sum, fio);
1246 }
1247 
1248 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1249 {
1250 	struct f2fs_sb_info *sbi = fio->sbi;
1251 	struct f2fs_summary sum;
1252 	struct node_info ni;
1253 
1254 	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1255 	get_node_info(sbi, dn->nid, &ni);
1256 	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1257 	do_write_page(&sum, fio);
1258 	dn->data_blkaddr = fio->blk_addr;
1259 }
1260 
1261 void rewrite_data_page(struct f2fs_io_info *fio)
1262 {
1263 	stat_inc_inplace_blocks(fio->sbi);
1264 	f2fs_submit_page_mbio(fio);
1265 }
1266 
1267 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1268 				block_t old_blkaddr, block_t new_blkaddr,
1269 				bool recover_curseg)
1270 {
1271 	struct sit_info *sit_i = SIT_I(sbi);
1272 	struct curseg_info *curseg;
1273 	unsigned int segno, old_cursegno;
1274 	struct seg_entry *se;
1275 	int type;
1276 	unsigned short old_blkoff;
1277 
1278 	segno = GET_SEGNO(sbi, new_blkaddr);
1279 	se = get_seg_entry(sbi, segno);
1280 	type = se->type;
1281 
1282 	if (!recover_curseg) {
1283 		/* for recovery flow */
1284 		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1285 			if (old_blkaddr == NULL_ADDR)
1286 				type = CURSEG_COLD_DATA;
1287 			else
1288 				type = CURSEG_WARM_DATA;
1289 		}
1290 	} else {
1291 		if (!IS_CURSEG(sbi, segno))
1292 			type = CURSEG_WARM_DATA;
1293 	}
1294 
1295 	curseg = CURSEG_I(sbi, type);
1296 
1297 	mutex_lock(&curseg->curseg_mutex);
1298 	mutex_lock(&sit_i->sentry_lock);
1299 
1300 	old_cursegno = curseg->segno;
1301 	old_blkoff = curseg->next_blkoff;
1302 
1303 	/* change the current segment */
1304 	if (segno != curseg->segno) {
1305 		curseg->next_segno = segno;
1306 		change_curseg(sbi, type, true);
1307 	}
1308 
1309 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1310 	__add_sum_entry(sbi, type, sum);
1311 
1312 	refresh_sit_entry(sbi, old_blkaddr, new_blkaddr);
1313 	locate_dirty_segment(sbi, old_cursegno);
1314 
1315 	if (recover_curseg) {
1316 		if (old_cursegno != curseg->segno) {
1317 			curseg->next_segno = old_cursegno;
1318 			change_curseg(sbi, type, true);
1319 		}
1320 		curseg->next_blkoff = old_blkoff;
1321 	}
1322 
1323 	mutex_unlock(&sit_i->sentry_lock);
1324 	mutex_unlock(&curseg->curseg_mutex);
1325 }
1326 
1327 static inline bool is_merged_page(struct f2fs_sb_info *sbi,
1328 					struct page *page, enum page_type type)
1329 {
1330 	enum page_type btype = PAGE_TYPE_OF_BIO(type);
1331 	struct f2fs_bio_info *io = &sbi->write_io[btype];
1332 	struct bio_vec *bvec;
1333 	int i;
1334 
1335 	down_read(&io->io_rwsem);
1336 	if (!io->bio)
1337 		goto out;
1338 
1339 	bio_for_each_segment_all(bvec, io->bio, i) {
1340 		if (page == bvec->bv_page) {
1341 			up_read(&io->io_rwsem);
1342 			return true;
1343 		}
1344 	}
1345 
1346 out:
1347 	up_read(&io->io_rwsem);
1348 	return false;
1349 }
1350 
1351 void f2fs_wait_on_page_writeback(struct page *page,
1352 				enum page_type type)
1353 {
1354 	if (PageWriteback(page)) {
1355 		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1356 
1357 		if (is_merged_page(sbi, page, type))
1358 			f2fs_submit_merged_bio(sbi, type, WRITE);
1359 		wait_on_page_writeback(page);
1360 	}
1361 }
1362 
1363 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1364 {
1365 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1366 	struct curseg_info *seg_i;
1367 	unsigned char *kaddr;
1368 	struct page *page;
1369 	block_t start;
1370 	int i, j, offset;
1371 
1372 	start = start_sum_block(sbi);
1373 
1374 	page = get_meta_page(sbi, start++);
1375 	kaddr = (unsigned char *)page_address(page);
1376 
1377 	/* Step 1: restore nat cache */
1378 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1379 	memcpy(&seg_i->sum_blk->n_nats, kaddr, SUM_JOURNAL_SIZE);
1380 
1381 	/* Step 2: restore sit cache */
1382 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1383 	memcpy(&seg_i->sum_blk->n_sits, kaddr + SUM_JOURNAL_SIZE,
1384 						SUM_JOURNAL_SIZE);
1385 	offset = 2 * SUM_JOURNAL_SIZE;
1386 
1387 	/* Step 3: restore summary entries */
1388 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1389 		unsigned short blk_off;
1390 		unsigned int segno;
1391 
1392 		seg_i = CURSEG_I(sbi, i);
1393 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1394 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1395 		seg_i->next_segno = segno;
1396 		reset_curseg(sbi, i, 0);
1397 		seg_i->alloc_type = ckpt->alloc_type[i];
1398 		seg_i->next_blkoff = blk_off;
1399 
1400 		if (seg_i->alloc_type == SSR)
1401 			blk_off = sbi->blocks_per_seg;
1402 
1403 		for (j = 0; j < blk_off; j++) {
1404 			struct f2fs_summary *s;
1405 			s = (struct f2fs_summary *)(kaddr + offset);
1406 			seg_i->sum_blk->entries[j] = *s;
1407 			offset += SUMMARY_SIZE;
1408 			if (offset + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1409 						SUM_FOOTER_SIZE)
1410 				continue;
1411 
1412 			f2fs_put_page(page, 1);
1413 			page = NULL;
1414 
1415 			page = get_meta_page(sbi, start++);
1416 			kaddr = (unsigned char *)page_address(page);
1417 			offset = 0;
1418 		}
1419 	}
1420 	f2fs_put_page(page, 1);
1421 	return 0;
1422 }
1423 
1424 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1425 {
1426 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1427 	struct f2fs_summary_block *sum;
1428 	struct curseg_info *curseg;
1429 	struct page *new;
1430 	unsigned short blk_off;
1431 	unsigned int segno = 0;
1432 	block_t blk_addr = 0;
1433 
1434 	/* get segment number and block addr */
1435 	if (IS_DATASEG(type)) {
1436 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1437 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1438 							CURSEG_HOT_DATA]);
1439 		if (__exist_node_summaries(sbi))
1440 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1441 		else
1442 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1443 	} else {
1444 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
1445 							CURSEG_HOT_NODE]);
1446 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1447 							CURSEG_HOT_NODE]);
1448 		if (__exist_node_summaries(sbi))
1449 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1450 							type - CURSEG_HOT_NODE);
1451 		else
1452 			blk_addr = GET_SUM_BLOCK(sbi, segno);
1453 	}
1454 
1455 	new = get_meta_page(sbi, blk_addr);
1456 	sum = (struct f2fs_summary_block *)page_address(new);
1457 
1458 	if (IS_NODESEG(type)) {
1459 		if (__exist_node_summaries(sbi)) {
1460 			struct f2fs_summary *ns = &sum->entries[0];
1461 			int i;
1462 			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1463 				ns->version = 0;
1464 				ns->ofs_in_node = 0;
1465 			}
1466 		} else {
1467 			int err;
1468 
1469 			err = restore_node_summary(sbi, segno, sum);
1470 			if (err) {
1471 				f2fs_put_page(new, 1);
1472 				return err;
1473 			}
1474 		}
1475 	}
1476 
1477 	/* set uncompleted segment to curseg */
1478 	curseg = CURSEG_I(sbi, type);
1479 	mutex_lock(&curseg->curseg_mutex);
1480 	memcpy(curseg->sum_blk, sum, PAGE_CACHE_SIZE);
1481 	curseg->next_segno = segno;
1482 	reset_curseg(sbi, type, 0);
1483 	curseg->alloc_type = ckpt->alloc_type[type];
1484 	curseg->next_blkoff = blk_off;
1485 	mutex_unlock(&curseg->curseg_mutex);
1486 	f2fs_put_page(new, 1);
1487 	return 0;
1488 }
1489 
1490 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1491 {
1492 	int type = CURSEG_HOT_DATA;
1493 	int err;
1494 
1495 	if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1496 		int npages = npages_for_summary_flush(sbi, true);
1497 
1498 		if (npages >= 2)
1499 			ra_meta_pages(sbi, start_sum_block(sbi), npages,
1500 								META_CP);
1501 
1502 		/* restore for compacted data summary */
1503 		if (read_compacted_summaries(sbi))
1504 			return -EINVAL;
1505 		type = CURSEG_HOT_NODE;
1506 	}
1507 
1508 	if (__exist_node_summaries(sbi))
1509 		ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1510 					NR_CURSEG_TYPE - type, META_CP);
1511 
1512 	for (; type <= CURSEG_COLD_NODE; type++) {
1513 		err = read_normal_summaries(sbi, type);
1514 		if (err)
1515 			return err;
1516 	}
1517 
1518 	return 0;
1519 }
1520 
1521 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1522 {
1523 	struct page *page;
1524 	unsigned char *kaddr;
1525 	struct f2fs_summary *summary;
1526 	struct curseg_info *seg_i;
1527 	int written_size = 0;
1528 	int i, j;
1529 
1530 	page = grab_meta_page(sbi, blkaddr++);
1531 	kaddr = (unsigned char *)page_address(page);
1532 
1533 	/* Step 1: write nat cache */
1534 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1535 	memcpy(kaddr, &seg_i->sum_blk->n_nats, SUM_JOURNAL_SIZE);
1536 	written_size += SUM_JOURNAL_SIZE;
1537 
1538 	/* Step 2: write sit cache */
1539 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1540 	memcpy(kaddr + written_size, &seg_i->sum_blk->n_sits,
1541 						SUM_JOURNAL_SIZE);
1542 	written_size += SUM_JOURNAL_SIZE;
1543 
1544 	/* Step 3: write summary entries */
1545 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1546 		unsigned short blkoff;
1547 		seg_i = CURSEG_I(sbi, i);
1548 		if (sbi->ckpt->alloc_type[i] == SSR)
1549 			blkoff = sbi->blocks_per_seg;
1550 		else
1551 			blkoff = curseg_blkoff(sbi, i);
1552 
1553 		for (j = 0; j < blkoff; j++) {
1554 			if (!page) {
1555 				page = grab_meta_page(sbi, blkaddr++);
1556 				kaddr = (unsigned char *)page_address(page);
1557 				written_size = 0;
1558 			}
1559 			summary = (struct f2fs_summary *)(kaddr + written_size);
1560 			*summary = seg_i->sum_blk->entries[j];
1561 			written_size += SUMMARY_SIZE;
1562 
1563 			if (written_size + SUMMARY_SIZE <= PAGE_CACHE_SIZE -
1564 							SUM_FOOTER_SIZE)
1565 				continue;
1566 
1567 			set_page_dirty(page);
1568 			f2fs_put_page(page, 1);
1569 			page = NULL;
1570 		}
1571 	}
1572 	if (page) {
1573 		set_page_dirty(page);
1574 		f2fs_put_page(page, 1);
1575 	}
1576 }
1577 
1578 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1579 					block_t blkaddr, int type)
1580 {
1581 	int i, end;
1582 	if (IS_DATASEG(type))
1583 		end = type + NR_CURSEG_DATA_TYPE;
1584 	else
1585 		end = type + NR_CURSEG_NODE_TYPE;
1586 
1587 	for (i = type; i < end; i++) {
1588 		struct curseg_info *sum = CURSEG_I(sbi, i);
1589 		mutex_lock(&sum->curseg_mutex);
1590 		write_sum_page(sbi, sum->sum_blk, blkaddr + (i - type));
1591 		mutex_unlock(&sum->curseg_mutex);
1592 	}
1593 }
1594 
1595 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1596 {
1597 	if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1598 		write_compacted_summaries(sbi, start_blk);
1599 	else
1600 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1601 }
1602 
1603 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1604 {
1605 	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1606 }
1607 
1608 int lookup_journal_in_cursum(struct f2fs_summary_block *sum, int type,
1609 					unsigned int val, int alloc)
1610 {
1611 	int i;
1612 
1613 	if (type == NAT_JOURNAL) {
1614 		for (i = 0; i < nats_in_cursum(sum); i++) {
1615 			if (le32_to_cpu(nid_in_journal(sum, i)) == val)
1616 				return i;
1617 		}
1618 		if (alloc && nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES)
1619 			return update_nats_in_cursum(sum, 1);
1620 	} else if (type == SIT_JOURNAL) {
1621 		for (i = 0; i < sits_in_cursum(sum); i++)
1622 			if (le32_to_cpu(segno_in_journal(sum, i)) == val)
1623 				return i;
1624 		if (alloc && sits_in_cursum(sum) < SIT_JOURNAL_ENTRIES)
1625 			return update_sits_in_cursum(sum, 1);
1626 	}
1627 	return -1;
1628 }
1629 
1630 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1631 					unsigned int segno)
1632 {
1633 	return get_meta_page(sbi, current_sit_addr(sbi, segno));
1634 }
1635 
1636 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1637 					unsigned int start)
1638 {
1639 	struct sit_info *sit_i = SIT_I(sbi);
1640 	struct page *src_page, *dst_page;
1641 	pgoff_t src_off, dst_off;
1642 	void *src_addr, *dst_addr;
1643 
1644 	src_off = current_sit_addr(sbi, start);
1645 	dst_off = next_sit_addr(sbi, src_off);
1646 
1647 	/* get current sit block page without lock */
1648 	src_page = get_meta_page(sbi, src_off);
1649 	dst_page = grab_meta_page(sbi, dst_off);
1650 	f2fs_bug_on(sbi, PageDirty(src_page));
1651 
1652 	src_addr = page_address(src_page);
1653 	dst_addr = page_address(dst_page);
1654 	memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
1655 
1656 	set_page_dirty(dst_page);
1657 	f2fs_put_page(src_page, 1);
1658 
1659 	set_to_next_sit(sit_i, start);
1660 
1661 	return dst_page;
1662 }
1663 
1664 static struct sit_entry_set *grab_sit_entry_set(void)
1665 {
1666 	struct sit_entry_set *ses =
1667 			f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_ATOMIC);
1668 
1669 	ses->entry_cnt = 0;
1670 	INIT_LIST_HEAD(&ses->set_list);
1671 	return ses;
1672 }
1673 
1674 static void release_sit_entry_set(struct sit_entry_set *ses)
1675 {
1676 	list_del(&ses->set_list);
1677 	kmem_cache_free(sit_entry_set_slab, ses);
1678 }
1679 
1680 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1681 						struct list_head *head)
1682 {
1683 	struct sit_entry_set *next = ses;
1684 
1685 	if (list_is_last(&ses->set_list, head))
1686 		return;
1687 
1688 	list_for_each_entry_continue(next, head, set_list)
1689 		if (ses->entry_cnt <= next->entry_cnt)
1690 			break;
1691 
1692 	list_move_tail(&ses->set_list, &next->set_list);
1693 }
1694 
1695 static void add_sit_entry(unsigned int segno, struct list_head *head)
1696 {
1697 	struct sit_entry_set *ses;
1698 	unsigned int start_segno = START_SEGNO(segno);
1699 
1700 	list_for_each_entry(ses, head, set_list) {
1701 		if (ses->start_segno == start_segno) {
1702 			ses->entry_cnt++;
1703 			adjust_sit_entry_set(ses, head);
1704 			return;
1705 		}
1706 	}
1707 
1708 	ses = grab_sit_entry_set();
1709 
1710 	ses->start_segno = start_segno;
1711 	ses->entry_cnt++;
1712 	list_add(&ses->set_list, head);
1713 }
1714 
1715 static void add_sits_in_set(struct f2fs_sb_info *sbi)
1716 {
1717 	struct f2fs_sm_info *sm_info = SM_I(sbi);
1718 	struct list_head *set_list = &sm_info->sit_entry_set;
1719 	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
1720 	unsigned int segno;
1721 
1722 	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
1723 		add_sit_entry(segno, set_list);
1724 }
1725 
1726 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
1727 {
1728 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1729 	struct f2fs_summary_block *sum = curseg->sum_blk;
1730 	int i;
1731 
1732 	for (i = sits_in_cursum(sum) - 1; i >= 0; i--) {
1733 		unsigned int segno;
1734 		bool dirtied;
1735 
1736 		segno = le32_to_cpu(segno_in_journal(sum, i));
1737 		dirtied = __mark_sit_entry_dirty(sbi, segno);
1738 
1739 		if (!dirtied)
1740 			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
1741 	}
1742 	update_sits_in_cursum(sum, -sits_in_cursum(sum));
1743 }
1744 
1745 /*
1746  * CP calls this function, which flushes SIT entries including sit_journal,
1747  * and moves prefree segs to free segs.
1748  */
1749 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1750 {
1751 	struct sit_info *sit_i = SIT_I(sbi);
1752 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1753 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1754 	struct f2fs_summary_block *sum = curseg->sum_blk;
1755 	struct sit_entry_set *ses, *tmp;
1756 	struct list_head *head = &SM_I(sbi)->sit_entry_set;
1757 	bool to_journal = true;
1758 	struct seg_entry *se;
1759 
1760 	mutex_lock(&curseg->curseg_mutex);
1761 	mutex_lock(&sit_i->sentry_lock);
1762 
1763 	if (!sit_i->dirty_sentries)
1764 		goto out;
1765 
1766 	/*
1767 	 * add and account sit entries of dirty bitmap in sit entry
1768 	 * set temporarily
1769 	 */
1770 	add_sits_in_set(sbi);
1771 
1772 	/*
1773 	 * if there are no enough space in journal to store dirty sit
1774 	 * entries, remove all entries from journal and add and account
1775 	 * them in sit entry set.
1776 	 */
1777 	if (!__has_cursum_space(sum, sit_i->dirty_sentries, SIT_JOURNAL))
1778 		remove_sits_in_journal(sbi);
1779 
1780 	/*
1781 	 * there are two steps to flush sit entries:
1782 	 * #1, flush sit entries to journal in current cold data summary block.
1783 	 * #2, flush sit entries to sit page.
1784 	 */
1785 	list_for_each_entry_safe(ses, tmp, head, set_list) {
1786 		struct page *page = NULL;
1787 		struct f2fs_sit_block *raw_sit = NULL;
1788 		unsigned int start_segno = ses->start_segno;
1789 		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
1790 						(unsigned long)MAIN_SEGS(sbi));
1791 		unsigned int segno = start_segno;
1792 
1793 		if (to_journal &&
1794 			!__has_cursum_space(sum, ses->entry_cnt, SIT_JOURNAL))
1795 			to_journal = false;
1796 
1797 		if (!to_journal) {
1798 			page = get_next_sit_page(sbi, start_segno);
1799 			raw_sit = page_address(page);
1800 		}
1801 
1802 		/* flush dirty sit entries in region of current sit set */
1803 		for_each_set_bit_from(segno, bitmap, end) {
1804 			int offset, sit_offset;
1805 
1806 			se = get_seg_entry(sbi, segno);
1807 
1808 			/* add discard candidates */
1809 			if (cpc->reason != CP_DISCARD) {
1810 				cpc->trim_start = segno;
1811 				add_discard_addrs(sbi, cpc);
1812 			}
1813 
1814 			if (to_journal) {
1815 				offset = lookup_journal_in_cursum(sum,
1816 							SIT_JOURNAL, segno, 1);
1817 				f2fs_bug_on(sbi, offset < 0);
1818 				segno_in_journal(sum, offset) =
1819 							cpu_to_le32(segno);
1820 				seg_info_to_raw_sit(se,
1821 						&sit_in_journal(sum, offset));
1822 			} else {
1823 				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
1824 				seg_info_to_raw_sit(se,
1825 						&raw_sit->entries[sit_offset]);
1826 			}
1827 
1828 			__clear_bit(segno, bitmap);
1829 			sit_i->dirty_sentries--;
1830 			ses->entry_cnt--;
1831 		}
1832 
1833 		if (!to_journal)
1834 			f2fs_put_page(page, 1);
1835 
1836 		f2fs_bug_on(sbi, ses->entry_cnt);
1837 		release_sit_entry_set(ses);
1838 	}
1839 
1840 	f2fs_bug_on(sbi, !list_empty(head));
1841 	f2fs_bug_on(sbi, sit_i->dirty_sentries);
1842 out:
1843 	if (cpc->reason == CP_DISCARD) {
1844 		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
1845 			add_discard_addrs(sbi, cpc);
1846 	}
1847 	mutex_unlock(&sit_i->sentry_lock);
1848 	mutex_unlock(&curseg->curseg_mutex);
1849 
1850 	set_prefree_as_free_segments(sbi);
1851 }
1852 
1853 static int build_sit_info(struct f2fs_sb_info *sbi)
1854 {
1855 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
1856 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1857 	struct sit_info *sit_i;
1858 	unsigned int sit_segs, start;
1859 	char *src_bitmap, *dst_bitmap;
1860 	unsigned int bitmap_size;
1861 
1862 	/* allocate memory for SIT information */
1863 	sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
1864 	if (!sit_i)
1865 		return -ENOMEM;
1866 
1867 	SM_I(sbi)->sit_info = sit_i;
1868 
1869 	sit_i->sentries = vzalloc(MAIN_SEGS(sbi) * sizeof(struct seg_entry));
1870 	if (!sit_i->sentries)
1871 		return -ENOMEM;
1872 
1873 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
1874 	sit_i->dirty_sentries_bitmap = kzalloc(bitmap_size, GFP_KERNEL);
1875 	if (!sit_i->dirty_sentries_bitmap)
1876 		return -ENOMEM;
1877 
1878 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
1879 		sit_i->sentries[start].cur_valid_map
1880 			= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1881 		sit_i->sentries[start].ckpt_valid_map
1882 			= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1883 		sit_i->sentries[start].discard_map
1884 			= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1885 		if (!sit_i->sentries[start].cur_valid_map ||
1886 				!sit_i->sentries[start].ckpt_valid_map ||
1887 				!sit_i->sentries[start].discard_map)
1888 			return -ENOMEM;
1889 	}
1890 
1891 	sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
1892 	if (!sit_i->tmp_map)
1893 		return -ENOMEM;
1894 
1895 	if (sbi->segs_per_sec > 1) {
1896 		sit_i->sec_entries = vzalloc(MAIN_SECS(sbi) *
1897 					sizeof(struct sec_entry));
1898 		if (!sit_i->sec_entries)
1899 			return -ENOMEM;
1900 	}
1901 
1902 	/* get information related with SIT */
1903 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
1904 
1905 	/* setup SIT bitmap from ckeckpoint pack */
1906 	bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
1907 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
1908 
1909 	dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
1910 	if (!dst_bitmap)
1911 		return -ENOMEM;
1912 
1913 	/* init SIT information */
1914 	sit_i->s_ops = &default_salloc_ops;
1915 
1916 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
1917 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
1918 	sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
1919 	sit_i->sit_bitmap = dst_bitmap;
1920 	sit_i->bitmap_size = bitmap_size;
1921 	sit_i->dirty_sentries = 0;
1922 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
1923 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
1924 	sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
1925 	mutex_init(&sit_i->sentry_lock);
1926 	return 0;
1927 }
1928 
1929 static int build_free_segmap(struct f2fs_sb_info *sbi)
1930 {
1931 	struct free_segmap_info *free_i;
1932 	unsigned int bitmap_size, sec_bitmap_size;
1933 
1934 	/* allocate memory for free segmap information */
1935 	free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
1936 	if (!free_i)
1937 		return -ENOMEM;
1938 
1939 	SM_I(sbi)->free_info = free_i;
1940 
1941 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
1942 	free_i->free_segmap = kmalloc(bitmap_size, GFP_KERNEL);
1943 	if (!free_i->free_segmap)
1944 		return -ENOMEM;
1945 
1946 	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
1947 	free_i->free_secmap = kmalloc(sec_bitmap_size, GFP_KERNEL);
1948 	if (!free_i->free_secmap)
1949 		return -ENOMEM;
1950 
1951 	/* set all segments as dirty temporarily */
1952 	memset(free_i->free_segmap, 0xff, bitmap_size);
1953 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
1954 
1955 	/* init free segmap information */
1956 	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
1957 	free_i->free_segments = 0;
1958 	free_i->free_sections = 0;
1959 	spin_lock_init(&free_i->segmap_lock);
1960 	return 0;
1961 }
1962 
1963 static int build_curseg(struct f2fs_sb_info *sbi)
1964 {
1965 	struct curseg_info *array;
1966 	int i;
1967 
1968 	array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
1969 	if (!array)
1970 		return -ENOMEM;
1971 
1972 	SM_I(sbi)->curseg_array = array;
1973 
1974 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
1975 		mutex_init(&array[i].curseg_mutex);
1976 		array[i].sum_blk = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
1977 		if (!array[i].sum_blk)
1978 			return -ENOMEM;
1979 		array[i].segno = NULL_SEGNO;
1980 		array[i].next_blkoff = 0;
1981 	}
1982 	return restore_curseg_summaries(sbi);
1983 }
1984 
1985 static void build_sit_entries(struct f2fs_sb_info *sbi)
1986 {
1987 	struct sit_info *sit_i = SIT_I(sbi);
1988 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1989 	struct f2fs_summary_block *sum = curseg->sum_blk;
1990 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
1991 	unsigned int i, start, end;
1992 	unsigned int readed, start_blk = 0;
1993 	int nrpages = MAX_BIO_BLOCKS(sbi);
1994 
1995 	do {
1996 		readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT);
1997 
1998 		start = start_blk * sit_i->sents_per_block;
1999 		end = (start_blk + readed) * sit_i->sents_per_block;
2000 
2001 		for (; start < end && start < MAIN_SEGS(sbi); start++) {
2002 			struct seg_entry *se = &sit_i->sentries[start];
2003 			struct f2fs_sit_block *sit_blk;
2004 			struct f2fs_sit_entry sit;
2005 			struct page *page;
2006 
2007 			mutex_lock(&curseg->curseg_mutex);
2008 			for (i = 0; i < sits_in_cursum(sum); i++) {
2009 				if (le32_to_cpu(segno_in_journal(sum, i))
2010 								== start) {
2011 					sit = sit_in_journal(sum, i);
2012 					mutex_unlock(&curseg->curseg_mutex);
2013 					goto got_it;
2014 				}
2015 			}
2016 			mutex_unlock(&curseg->curseg_mutex);
2017 
2018 			page = get_current_sit_page(sbi, start);
2019 			sit_blk = (struct f2fs_sit_block *)page_address(page);
2020 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2021 			f2fs_put_page(page, 1);
2022 got_it:
2023 			check_block_count(sbi, start, &sit);
2024 			seg_info_from_raw_sit(se, &sit);
2025 
2026 			/* build discard map only one time */
2027 			memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2028 			sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2029 
2030 			if (sbi->segs_per_sec > 1) {
2031 				struct sec_entry *e = get_sec_entry(sbi, start);
2032 				e->valid_blocks += se->valid_blocks;
2033 			}
2034 		}
2035 		start_blk += readed;
2036 	} while (start_blk < sit_blk_cnt);
2037 }
2038 
2039 static void init_free_segmap(struct f2fs_sb_info *sbi)
2040 {
2041 	unsigned int start;
2042 	int type;
2043 
2044 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
2045 		struct seg_entry *sentry = get_seg_entry(sbi, start);
2046 		if (!sentry->valid_blocks)
2047 			__set_free(sbi, start);
2048 	}
2049 
2050 	/* set use the current segments */
2051 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2052 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2053 		__set_test_and_inuse(sbi, curseg_t->segno);
2054 	}
2055 }
2056 
2057 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2058 {
2059 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2060 	struct free_segmap_info *free_i = FREE_I(sbi);
2061 	unsigned int segno = 0, offset = 0;
2062 	unsigned short valid_blocks;
2063 
2064 	while (1) {
2065 		/* find dirty segment based on free segmap */
2066 		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2067 		if (segno >= MAIN_SEGS(sbi))
2068 			break;
2069 		offset = segno + 1;
2070 		valid_blocks = get_valid_blocks(sbi, segno, 0);
2071 		if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2072 			continue;
2073 		if (valid_blocks > sbi->blocks_per_seg) {
2074 			f2fs_bug_on(sbi, 1);
2075 			continue;
2076 		}
2077 		mutex_lock(&dirty_i->seglist_lock);
2078 		__locate_dirty_segment(sbi, segno, DIRTY);
2079 		mutex_unlock(&dirty_i->seglist_lock);
2080 	}
2081 }
2082 
2083 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2084 {
2085 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2086 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2087 
2088 	dirty_i->victim_secmap = kzalloc(bitmap_size, GFP_KERNEL);
2089 	if (!dirty_i->victim_secmap)
2090 		return -ENOMEM;
2091 	return 0;
2092 }
2093 
2094 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2095 {
2096 	struct dirty_seglist_info *dirty_i;
2097 	unsigned int bitmap_size, i;
2098 
2099 	/* allocate memory for dirty segments list information */
2100 	dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2101 	if (!dirty_i)
2102 		return -ENOMEM;
2103 
2104 	SM_I(sbi)->dirty_info = dirty_i;
2105 	mutex_init(&dirty_i->seglist_lock);
2106 
2107 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2108 
2109 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
2110 		dirty_i->dirty_segmap[i] = kzalloc(bitmap_size, GFP_KERNEL);
2111 		if (!dirty_i->dirty_segmap[i])
2112 			return -ENOMEM;
2113 	}
2114 
2115 	init_dirty_segmap(sbi);
2116 	return init_victim_secmap(sbi);
2117 }
2118 
2119 /*
2120  * Update min, max modified time for cost-benefit GC algorithm
2121  */
2122 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2123 {
2124 	struct sit_info *sit_i = SIT_I(sbi);
2125 	unsigned int segno;
2126 
2127 	mutex_lock(&sit_i->sentry_lock);
2128 
2129 	sit_i->min_mtime = LLONG_MAX;
2130 
2131 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2132 		unsigned int i;
2133 		unsigned long long mtime = 0;
2134 
2135 		for (i = 0; i < sbi->segs_per_sec; i++)
2136 			mtime += get_seg_entry(sbi, segno + i)->mtime;
2137 
2138 		mtime = div_u64(mtime, sbi->segs_per_sec);
2139 
2140 		if (sit_i->min_mtime > mtime)
2141 			sit_i->min_mtime = mtime;
2142 	}
2143 	sit_i->max_mtime = get_mtime(sbi);
2144 	mutex_unlock(&sit_i->sentry_lock);
2145 }
2146 
2147 int build_segment_manager(struct f2fs_sb_info *sbi)
2148 {
2149 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2150 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2151 	struct f2fs_sm_info *sm_info;
2152 	int err;
2153 
2154 	sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2155 	if (!sm_info)
2156 		return -ENOMEM;
2157 
2158 	/* init sm info */
2159 	sbi->sm_info = sm_info;
2160 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2161 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2162 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2163 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2164 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2165 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2166 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2167 	sm_info->rec_prefree_segments = sm_info->main_segments *
2168 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2169 	sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2170 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2171 	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2172 
2173 	INIT_LIST_HEAD(&sm_info->discard_list);
2174 	sm_info->nr_discards = 0;
2175 	sm_info->max_discards = 0;
2176 
2177 	sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2178 
2179 	INIT_LIST_HEAD(&sm_info->sit_entry_set);
2180 
2181 	if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2182 		err = create_flush_cmd_control(sbi);
2183 		if (err)
2184 			return err;
2185 	}
2186 
2187 	err = build_sit_info(sbi);
2188 	if (err)
2189 		return err;
2190 	err = build_free_segmap(sbi);
2191 	if (err)
2192 		return err;
2193 	err = build_curseg(sbi);
2194 	if (err)
2195 		return err;
2196 
2197 	/* reinit free segmap based on SIT */
2198 	build_sit_entries(sbi);
2199 
2200 	init_free_segmap(sbi);
2201 	err = build_dirty_segmap(sbi);
2202 	if (err)
2203 		return err;
2204 
2205 	init_min_max_mtime(sbi);
2206 	return 0;
2207 }
2208 
2209 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2210 		enum dirty_type dirty_type)
2211 {
2212 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2213 
2214 	mutex_lock(&dirty_i->seglist_lock);
2215 	kfree(dirty_i->dirty_segmap[dirty_type]);
2216 	dirty_i->nr_dirty[dirty_type] = 0;
2217 	mutex_unlock(&dirty_i->seglist_lock);
2218 }
2219 
2220 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2221 {
2222 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2223 	kfree(dirty_i->victim_secmap);
2224 }
2225 
2226 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2227 {
2228 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2229 	int i;
2230 
2231 	if (!dirty_i)
2232 		return;
2233 
2234 	/* discard pre-free/dirty segments list */
2235 	for (i = 0; i < NR_DIRTY_TYPE; i++)
2236 		discard_dirty_segmap(sbi, i);
2237 
2238 	destroy_victim_secmap(sbi);
2239 	SM_I(sbi)->dirty_info = NULL;
2240 	kfree(dirty_i);
2241 }
2242 
2243 static void destroy_curseg(struct f2fs_sb_info *sbi)
2244 {
2245 	struct curseg_info *array = SM_I(sbi)->curseg_array;
2246 	int i;
2247 
2248 	if (!array)
2249 		return;
2250 	SM_I(sbi)->curseg_array = NULL;
2251 	for (i = 0; i < NR_CURSEG_TYPE; i++)
2252 		kfree(array[i].sum_blk);
2253 	kfree(array);
2254 }
2255 
2256 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2257 {
2258 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2259 	if (!free_i)
2260 		return;
2261 	SM_I(sbi)->free_info = NULL;
2262 	kfree(free_i->free_segmap);
2263 	kfree(free_i->free_secmap);
2264 	kfree(free_i);
2265 }
2266 
2267 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2268 {
2269 	struct sit_info *sit_i = SIT_I(sbi);
2270 	unsigned int start;
2271 
2272 	if (!sit_i)
2273 		return;
2274 
2275 	if (sit_i->sentries) {
2276 		for (start = 0; start < MAIN_SEGS(sbi); start++) {
2277 			kfree(sit_i->sentries[start].cur_valid_map);
2278 			kfree(sit_i->sentries[start].ckpt_valid_map);
2279 			kfree(sit_i->sentries[start].discard_map);
2280 		}
2281 	}
2282 	kfree(sit_i->tmp_map);
2283 
2284 	vfree(sit_i->sentries);
2285 	vfree(sit_i->sec_entries);
2286 	kfree(sit_i->dirty_sentries_bitmap);
2287 
2288 	SM_I(sbi)->sit_info = NULL;
2289 	kfree(sit_i->sit_bitmap);
2290 	kfree(sit_i);
2291 }
2292 
2293 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2294 {
2295 	struct f2fs_sm_info *sm_info = SM_I(sbi);
2296 
2297 	if (!sm_info)
2298 		return;
2299 	destroy_flush_cmd_control(sbi);
2300 	destroy_dirty_segmap(sbi);
2301 	destroy_curseg(sbi);
2302 	destroy_free_segmap(sbi);
2303 	destroy_sit_info(sbi);
2304 	sbi->sm_info = NULL;
2305 	kfree(sm_info);
2306 }
2307 
2308 int __init create_segment_manager_caches(void)
2309 {
2310 	discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2311 			sizeof(struct discard_entry));
2312 	if (!discard_entry_slab)
2313 		goto fail;
2314 
2315 	sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2316 			sizeof(struct sit_entry_set));
2317 	if (!sit_entry_set_slab)
2318 		goto destory_discard_entry;
2319 
2320 	inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2321 			sizeof(struct inmem_pages));
2322 	if (!inmem_entry_slab)
2323 		goto destroy_sit_entry_set;
2324 	return 0;
2325 
2326 destroy_sit_entry_set:
2327 	kmem_cache_destroy(sit_entry_set_slab);
2328 destory_discard_entry:
2329 	kmem_cache_destroy(discard_entry_slab);
2330 fail:
2331 	return -ENOMEM;
2332 }
2333 
2334 void destroy_segment_manager_caches(void)
2335 {
2336 	kmem_cache_destroy(sit_entry_set_slab);
2337 	kmem_cache_destroy(discard_entry_slab);
2338 	kmem_cache_destroy(inmem_entry_slab);
2339 }
2340