xref: /openbmc/linux/fs/f2fs/segment.c (revision 26a28a0c1eb756ba18bfb1f93309c4b4406b9cd9)
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/swap.h>
18 #include <linux/timer.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 *bio_entry_slab;
30 static struct kmem_cache *sit_entry_set_slab;
31 static struct kmem_cache *inmem_entry_slab;
32 
33 static unsigned long __reverse_ulong(unsigned char *str)
34 {
35 	unsigned long tmp = 0;
36 	int shift = 24, idx = 0;
37 
38 #if BITS_PER_LONG == 64
39 	shift = 56;
40 #endif
41 	while (shift >= 0) {
42 		tmp |= (unsigned long)str[idx++] << shift;
43 		shift -= BITS_PER_BYTE;
44 	}
45 	return tmp;
46 }
47 
48 /*
49  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
50  * MSB and LSB are reversed in a byte by f2fs_set_bit.
51  */
52 static inline unsigned long __reverse_ffs(unsigned long word)
53 {
54 	int num = 0;
55 
56 #if BITS_PER_LONG == 64
57 	if ((word & 0xffffffff00000000UL) == 0)
58 		num += 32;
59 	else
60 		word >>= 32;
61 #endif
62 	if ((word & 0xffff0000) == 0)
63 		num += 16;
64 	else
65 		word >>= 16;
66 
67 	if ((word & 0xff00) == 0)
68 		num += 8;
69 	else
70 		word >>= 8;
71 
72 	if ((word & 0xf0) == 0)
73 		num += 4;
74 	else
75 		word >>= 4;
76 
77 	if ((word & 0xc) == 0)
78 		num += 2;
79 	else
80 		word >>= 2;
81 
82 	if ((word & 0x2) == 0)
83 		num += 1;
84 	return num;
85 }
86 
87 /*
88  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
89  * f2fs_set_bit makes MSB and LSB reversed in a byte.
90  * @size must be integral times of unsigned long.
91  * Example:
92  *                             MSB <--> LSB
93  *   f2fs_set_bit(0, bitmap) => 1000 0000
94  *   f2fs_set_bit(7, bitmap) => 0000 0001
95  */
96 static unsigned long __find_rev_next_bit(const unsigned long *addr,
97 			unsigned long size, unsigned long offset)
98 {
99 	const unsigned long *p = addr + BIT_WORD(offset);
100 	unsigned long result = size;
101 	unsigned long tmp;
102 
103 	if (offset >= size)
104 		return size;
105 
106 	size -= (offset & ~(BITS_PER_LONG - 1));
107 	offset %= BITS_PER_LONG;
108 
109 	while (1) {
110 		if (*p == 0)
111 			goto pass;
112 
113 		tmp = __reverse_ulong((unsigned char *)p);
114 
115 		tmp &= ~0UL >> offset;
116 		if (size < BITS_PER_LONG)
117 			tmp &= (~0UL << (BITS_PER_LONG - size));
118 		if (tmp)
119 			goto found;
120 pass:
121 		if (size <= BITS_PER_LONG)
122 			break;
123 		size -= BITS_PER_LONG;
124 		offset = 0;
125 		p++;
126 	}
127 	return result;
128 found:
129 	return result - size + __reverse_ffs(tmp);
130 }
131 
132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
133 			unsigned long size, unsigned long offset)
134 {
135 	const unsigned long *p = addr + BIT_WORD(offset);
136 	unsigned long result = size;
137 	unsigned long tmp;
138 
139 	if (offset >= size)
140 		return size;
141 
142 	size -= (offset & ~(BITS_PER_LONG - 1));
143 	offset %= BITS_PER_LONG;
144 
145 	while (1) {
146 		if (*p == ~0UL)
147 			goto pass;
148 
149 		tmp = __reverse_ulong((unsigned char *)p);
150 
151 		if (offset)
152 			tmp |= ~0UL << (BITS_PER_LONG - offset);
153 		if (size < BITS_PER_LONG)
154 			tmp |= ~0UL >> size;
155 		if (tmp != ~0UL)
156 			goto found;
157 pass:
158 		if (size <= BITS_PER_LONG)
159 			break;
160 		size -= BITS_PER_LONG;
161 		offset = 0;
162 		p++;
163 	}
164 	return result;
165 found:
166 	return result - size + __reverse_ffz(tmp);
167 }
168 
169 void register_inmem_page(struct inode *inode, struct page *page)
170 {
171 	struct f2fs_inode_info *fi = F2FS_I(inode);
172 	struct inmem_pages *new;
173 
174 	f2fs_trace_pid(page);
175 
176 	set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
177 	SetPagePrivate(page);
178 
179 	new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
180 
181 	/* add atomic page indices to the list */
182 	new->page = page;
183 	INIT_LIST_HEAD(&new->list);
184 
185 	/* increase reference count with clean state */
186 	mutex_lock(&fi->inmem_lock);
187 	get_page(page);
188 	list_add_tail(&new->list, &fi->inmem_pages);
189 	inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
190 	mutex_unlock(&fi->inmem_lock);
191 
192 	trace_f2fs_register_inmem_page(page, INMEM);
193 }
194 
195 static int __revoke_inmem_pages(struct inode *inode,
196 				struct list_head *head, bool drop, bool recover)
197 {
198 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
199 	struct inmem_pages *cur, *tmp;
200 	int err = 0;
201 
202 	list_for_each_entry_safe(cur, tmp, head, list) {
203 		struct page *page = cur->page;
204 
205 		if (drop)
206 			trace_f2fs_commit_inmem_page(page, INMEM_DROP);
207 
208 		lock_page(page);
209 
210 		if (recover) {
211 			struct dnode_of_data dn;
212 			struct node_info ni;
213 
214 			trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
215 
216 			set_new_dnode(&dn, inode, NULL, NULL, 0);
217 			if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
218 				err = -EAGAIN;
219 				goto next;
220 			}
221 			get_node_info(sbi, dn.nid, &ni);
222 			f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
223 					cur->old_addr, ni.version, true, true);
224 			f2fs_put_dnode(&dn);
225 		}
226 next:
227 		/* we don't need to invalidate this in the sccessful status */
228 		if (drop || recover)
229 			ClearPageUptodate(page);
230 		set_page_private(page, 0);
231 		ClearPagePrivate(page);
232 		f2fs_put_page(page, 1);
233 
234 		list_del(&cur->list);
235 		kmem_cache_free(inmem_entry_slab, cur);
236 		dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
237 	}
238 	return err;
239 }
240 
241 void drop_inmem_pages(struct inode *inode)
242 {
243 	struct f2fs_inode_info *fi = F2FS_I(inode);
244 
245 	clear_inode_flag(inode, FI_ATOMIC_FILE);
246 	stat_dec_atomic_write(inode);
247 
248 	mutex_lock(&fi->inmem_lock);
249 	__revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
250 	mutex_unlock(&fi->inmem_lock);
251 }
252 
253 static int __commit_inmem_pages(struct inode *inode,
254 					struct list_head *revoke_list)
255 {
256 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
257 	struct f2fs_inode_info *fi = F2FS_I(inode);
258 	struct inmem_pages *cur, *tmp;
259 	struct f2fs_io_info fio = {
260 		.sbi = sbi,
261 		.type = DATA,
262 		.op = REQ_OP_WRITE,
263 		.op_flags = REQ_SYNC | REQ_PRIO,
264 		.encrypted_page = NULL,
265 	};
266 	bool submit_bio = false;
267 	int err = 0;
268 
269 	list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
270 		struct page *page = cur->page;
271 
272 		lock_page(page);
273 		if (page->mapping == inode->i_mapping) {
274 			trace_f2fs_commit_inmem_page(page, INMEM);
275 
276 			set_page_dirty(page);
277 			f2fs_wait_on_page_writeback(page, DATA, true);
278 			if (clear_page_dirty_for_io(page)) {
279 				inode_dec_dirty_pages(inode);
280 				remove_dirty_inode(inode);
281 			}
282 
283 			fio.page = page;
284 			err = do_write_data_page(&fio);
285 			if (err) {
286 				unlock_page(page);
287 				break;
288 			}
289 
290 			/* record old blkaddr for revoking */
291 			cur->old_addr = fio.old_blkaddr;
292 
293 			submit_bio = true;
294 		}
295 		unlock_page(page);
296 		list_move_tail(&cur->list, revoke_list);
297 	}
298 
299 	if (submit_bio)
300 		f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
301 
302 	if (!err)
303 		__revoke_inmem_pages(inode, revoke_list, false, false);
304 
305 	return err;
306 }
307 
308 int commit_inmem_pages(struct inode *inode)
309 {
310 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
311 	struct f2fs_inode_info *fi = F2FS_I(inode);
312 	struct list_head revoke_list;
313 	int err;
314 
315 	INIT_LIST_HEAD(&revoke_list);
316 	f2fs_balance_fs(sbi, true);
317 	f2fs_lock_op(sbi);
318 
319 	mutex_lock(&fi->inmem_lock);
320 	err = __commit_inmem_pages(inode, &revoke_list);
321 	if (err) {
322 		int ret;
323 		/*
324 		 * try to revoke all committed pages, but still we could fail
325 		 * due to no memory or other reason, if that happened, EAGAIN
326 		 * will be returned, which means in such case, transaction is
327 		 * already not integrity, caller should use journal to do the
328 		 * recovery or rewrite & commit last transaction. For other
329 		 * error number, revoking was done by filesystem itself.
330 		 */
331 		ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
332 		if (ret)
333 			err = ret;
334 
335 		/* drop all uncommitted pages */
336 		__revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
337 	}
338 	mutex_unlock(&fi->inmem_lock);
339 
340 	f2fs_unlock_op(sbi);
341 	return err;
342 }
343 
344 /*
345  * This function balances dirty node and dentry pages.
346  * In addition, it controls garbage collection.
347  */
348 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
349 {
350 #ifdef CONFIG_F2FS_FAULT_INJECTION
351 	if (time_to_inject(sbi, FAULT_CHECKPOINT))
352 		f2fs_stop_checkpoint(sbi, false);
353 #endif
354 
355 	if (!need)
356 		return;
357 
358 	/* balance_fs_bg is able to be pending */
359 	if (excess_cached_nats(sbi))
360 		f2fs_balance_fs_bg(sbi);
361 
362 	/*
363 	 * We should do GC or end up with checkpoint, if there are so many dirty
364 	 * dir/node pages without enough free segments.
365 	 */
366 	if (has_not_enough_free_secs(sbi, 0, 0)) {
367 		mutex_lock(&sbi->gc_mutex);
368 		f2fs_gc(sbi, false, false);
369 	}
370 }
371 
372 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
373 {
374 	/* try to shrink extent cache when there is no enough memory */
375 	if (!available_free_memory(sbi, EXTENT_CACHE))
376 		f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
377 
378 	/* check the # of cached NAT entries */
379 	if (!available_free_memory(sbi, NAT_ENTRIES))
380 		try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
381 
382 	if (!available_free_memory(sbi, FREE_NIDS))
383 		try_to_free_nids(sbi, MAX_FREE_NIDS);
384 	else
385 		build_free_nids(sbi, false);
386 
387 	if (!is_idle(sbi))
388 		return;
389 
390 	/* checkpoint is the only way to shrink partial cached entries */
391 	if (!available_free_memory(sbi, NAT_ENTRIES) ||
392 			!available_free_memory(sbi, INO_ENTRIES) ||
393 			excess_prefree_segs(sbi) ||
394 			excess_dirty_nats(sbi) ||
395 			f2fs_time_over(sbi, CP_TIME)) {
396 		if (test_opt(sbi, DATA_FLUSH)) {
397 			struct blk_plug plug;
398 
399 			blk_start_plug(&plug);
400 			sync_dirty_inodes(sbi, FILE_INODE);
401 			blk_finish_plug(&plug);
402 		}
403 		f2fs_sync_fs(sbi->sb, true);
404 		stat_inc_bg_cp_count(sbi->stat_info);
405 	}
406 }
407 
408 static int __submit_flush_wait(struct block_device *bdev)
409 {
410 	struct bio *bio = f2fs_bio_alloc(0);
411 	int ret;
412 
413 	bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
414 	bio->bi_bdev = bdev;
415 	ret = submit_bio_wait(bio);
416 	bio_put(bio);
417 	return ret;
418 }
419 
420 static int submit_flush_wait(struct f2fs_sb_info *sbi)
421 {
422 	int ret = __submit_flush_wait(sbi->sb->s_bdev);
423 	int i;
424 
425 	if (sbi->s_ndevs && !ret) {
426 		for (i = 1; i < sbi->s_ndevs; i++) {
427 			ret = __submit_flush_wait(FDEV(i).bdev);
428 			if (ret)
429 				break;
430 		}
431 	}
432 	return ret;
433 }
434 
435 static int issue_flush_thread(void *data)
436 {
437 	struct f2fs_sb_info *sbi = data;
438 	struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
439 	wait_queue_head_t *q = &fcc->flush_wait_queue;
440 repeat:
441 	if (kthread_should_stop())
442 		return 0;
443 
444 	if (!llist_empty(&fcc->issue_list)) {
445 		struct flush_cmd *cmd, *next;
446 		int ret;
447 
448 		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
449 		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
450 
451 		ret = submit_flush_wait(sbi);
452 		llist_for_each_entry_safe(cmd, next,
453 					  fcc->dispatch_list, llnode) {
454 			cmd->ret = ret;
455 			complete(&cmd->wait);
456 		}
457 		fcc->dispatch_list = NULL;
458 	}
459 
460 	wait_event_interruptible(*q,
461 		kthread_should_stop() || !llist_empty(&fcc->issue_list));
462 	goto repeat;
463 }
464 
465 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
466 {
467 	struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
468 	struct flush_cmd cmd;
469 
470 	trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
471 					test_opt(sbi, FLUSH_MERGE));
472 
473 	if (test_opt(sbi, NOBARRIER))
474 		return 0;
475 
476 	if (!test_opt(sbi, FLUSH_MERGE) || !atomic_read(&fcc->submit_flush)) {
477 		int ret;
478 
479 		atomic_inc(&fcc->submit_flush);
480 		ret = submit_flush_wait(sbi);
481 		atomic_dec(&fcc->submit_flush);
482 		return ret;
483 	}
484 
485 	init_completion(&cmd.wait);
486 
487 	atomic_inc(&fcc->submit_flush);
488 	llist_add(&cmd.llnode, &fcc->issue_list);
489 
490 	if (!fcc->dispatch_list)
491 		wake_up(&fcc->flush_wait_queue);
492 
493 	if (fcc->f2fs_issue_flush) {
494 		wait_for_completion(&cmd.wait);
495 		atomic_dec(&fcc->submit_flush);
496 	} else {
497 		llist_del_all(&fcc->issue_list);
498 		atomic_set(&fcc->submit_flush, 0);
499 	}
500 
501 	return cmd.ret;
502 }
503 
504 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
505 {
506 	dev_t dev = sbi->sb->s_bdev->bd_dev;
507 	struct flush_cmd_control *fcc;
508 	int err = 0;
509 
510 	if (SM_I(sbi)->cmd_control_info) {
511 		fcc = SM_I(sbi)->cmd_control_info;
512 		goto init_thread;
513 	}
514 
515 	fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
516 	if (!fcc)
517 		return -ENOMEM;
518 	atomic_set(&fcc->submit_flush, 0);
519 	init_waitqueue_head(&fcc->flush_wait_queue);
520 	init_llist_head(&fcc->issue_list);
521 	SM_I(sbi)->cmd_control_info = fcc;
522 init_thread:
523 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
524 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
525 	if (IS_ERR(fcc->f2fs_issue_flush)) {
526 		err = PTR_ERR(fcc->f2fs_issue_flush);
527 		kfree(fcc);
528 		SM_I(sbi)->cmd_control_info = NULL;
529 		return err;
530 	}
531 
532 	return err;
533 }
534 
535 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
536 {
537 	struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
538 
539 	if (fcc && fcc->f2fs_issue_flush) {
540 		struct task_struct *flush_thread = fcc->f2fs_issue_flush;
541 
542 		fcc->f2fs_issue_flush = NULL;
543 		kthread_stop(flush_thread);
544 	}
545 	if (free) {
546 		kfree(fcc);
547 		SM_I(sbi)->cmd_control_info = NULL;
548 	}
549 }
550 
551 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
552 		enum dirty_type dirty_type)
553 {
554 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
555 
556 	/* need not be added */
557 	if (IS_CURSEG(sbi, segno))
558 		return;
559 
560 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
561 		dirty_i->nr_dirty[dirty_type]++;
562 
563 	if (dirty_type == DIRTY) {
564 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
565 		enum dirty_type t = sentry->type;
566 
567 		if (unlikely(t >= DIRTY)) {
568 			f2fs_bug_on(sbi, 1);
569 			return;
570 		}
571 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
572 			dirty_i->nr_dirty[t]++;
573 	}
574 }
575 
576 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
577 		enum dirty_type dirty_type)
578 {
579 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
580 
581 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
582 		dirty_i->nr_dirty[dirty_type]--;
583 
584 	if (dirty_type == DIRTY) {
585 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
586 		enum dirty_type t = sentry->type;
587 
588 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
589 			dirty_i->nr_dirty[t]--;
590 
591 		if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
592 			clear_bit(GET_SECNO(sbi, segno),
593 						dirty_i->victim_secmap);
594 	}
595 }
596 
597 /*
598  * Should not occur error such as -ENOMEM.
599  * Adding dirty entry into seglist is not critical operation.
600  * If a given segment is one of current working segments, it won't be added.
601  */
602 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
603 {
604 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
605 	unsigned short valid_blocks;
606 
607 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
608 		return;
609 
610 	mutex_lock(&dirty_i->seglist_lock);
611 
612 	valid_blocks = get_valid_blocks(sbi, segno, 0);
613 
614 	if (valid_blocks == 0) {
615 		__locate_dirty_segment(sbi, segno, PRE);
616 		__remove_dirty_segment(sbi, segno, DIRTY);
617 	} else if (valid_blocks < sbi->blocks_per_seg) {
618 		__locate_dirty_segment(sbi, segno, DIRTY);
619 	} else {
620 		/* Recovery routine with SSR needs this */
621 		__remove_dirty_segment(sbi, segno, DIRTY);
622 	}
623 
624 	mutex_unlock(&dirty_i->seglist_lock);
625 }
626 
627 static struct bio_entry *__add_bio_entry(struct f2fs_sb_info *sbi,
628 							struct bio *bio)
629 {
630 	struct list_head *wait_list = &(SM_I(sbi)->wait_list);
631 	struct bio_entry *be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS);
632 
633 	INIT_LIST_HEAD(&be->list);
634 	be->bio = bio;
635 	init_completion(&be->event);
636 	list_add_tail(&be->list, wait_list);
637 
638 	return be;
639 }
640 
641 void f2fs_wait_all_discard_bio(struct f2fs_sb_info *sbi)
642 {
643 	struct list_head *wait_list = &(SM_I(sbi)->wait_list);
644 	struct bio_entry *be, *tmp;
645 
646 	list_for_each_entry_safe(be, tmp, wait_list, list) {
647 		struct bio *bio = be->bio;
648 		int err;
649 
650 		wait_for_completion_io(&be->event);
651 		err = be->error;
652 		if (err == -EOPNOTSUPP)
653 			err = 0;
654 
655 		if (err)
656 			f2fs_msg(sbi->sb, KERN_INFO,
657 				"Issue discard failed, ret: %d", err);
658 
659 		bio_put(bio);
660 		list_del(&be->list);
661 		kmem_cache_free(bio_entry_slab, be);
662 	}
663 }
664 
665 static void f2fs_submit_bio_wait_endio(struct bio *bio)
666 {
667 	struct bio_entry *be = (struct bio_entry *)bio->bi_private;
668 
669 	be->error = bio->bi_error;
670 	complete(&be->event);
671 }
672 
673 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
674 static int __f2fs_issue_discard_async(struct f2fs_sb_info *sbi,
675 		struct block_device *bdev, block_t blkstart, block_t blklen)
676 {
677 	struct bio *bio = NULL;
678 	int err;
679 
680 	trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
681 
682 	if (sbi->s_ndevs) {
683 		int devi = f2fs_target_device_index(sbi, blkstart);
684 
685 		blkstart -= FDEV(devi).start_blk;
686 	}
687 	err = __blkdev_issue_discard(bdev,
688 				SECTOR_FROM_BLOCK(blkstart),
689 				SECTOR_FROM_BLOCK(blklen),
690 				GFP_NOFS, 0, &bio);
691 	if (!err && bio) {
692 		struct bio_entry *be = __add_bio_entry(sbi, bio);
693 
694 		bio->bi_private = be;
695 		bio->bi_end_io = f2fs_submit_bio_wait_endio;
696 		bio->bi_opf |= REQ_SYNC;
697 		submit_bio(bio);
698 	}
699 
700 	return err;
701 }
702 
703 #ifdef CONFIG_BLK_DEV_ZONED
704 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
705 		struct block_device *bdev, block_t blkstart, block_t blklen)
706 {
707 	sector_t nr_sects = SECTOR_FROM_BLOCK(blklen);
708 	sector_t sector;
709 	int devi = 0;
710 
711 	if (sbi->s_ndevs) {
712 		devi = f2fs_target_device_index(sbi, blkstart);
713 		blkstart -= FDEV(devi).start_blk;
714 	}
715 	sector = SECTOR_FROM_BLOCK(blkstart);
716 
717 	if (sector & (bdev_zone_sectors(bdev) - 1) ||
718 	    nr_sects != bdev_zone_sectors(bdev)) {
719 		f2fs_msg(sbi->sb, KERN_INFO,
720 			"(%d) %s: Unaligned discard attempted (block %x + %x)",
721 			devi, sbi->s_ndevs ? FDEV(devi).path: "",
722 			blkstart, blklen);
723 		return -EIO;
724 	}
725 
726 	/*
727 	 * We need to know the type of the zone: for conventional zones,
728 	 * use regular discard if the drive supports it. For sequential
729 	 * zones, reset the zone write pointer.
730 	 */
731 	switch (get_blkz_type(sbi, bdev, blkstart)) {
732 
733 	case BLK_ZONE_TYPE_CONVENTIONAL:
734 		if (!blk_queue_discard(bdev_get_queue(bdev)))
735 			return 0;
736 		return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen);
737 	case BLK_ZONE_TYPE_SEQWRITE_REQ:
738 	case BLK_ZONE_TYPE_SEQWRITE_PREF:
739 		trace_f2fs_issue_reset_zone(sbi->sb, blkstart);
740 		return blkdev_reset_zones(bdev, sector,
741 					  nr_sects, GFP_NOFS);
742 	default:
743 		/* Unknown zone type: broken device ? */
744 		return -EIO;
745 	}
746 }
747 #endif
748 
749 static int __issue_discard_async(struct f2fs_sb_info *sbi,
750 		struct block_device *bdev, block_t blkstart, block_t blklen)
751 {
752 #ifdef CONFIG_BLK_DEV_ZONED
753 	if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
754 				bdev_zoned_model(bdev) != BLK_ZONED_NONE)
755 		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
756 #endif
757 	return __f2fs_issue_discard_async(sbi, bdev, blkstart, blklen);
758 }
759 
760 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
761 				block_t blkstart, block_t blklen)
762 {
763 	sector_t start = blkstart, len = 0;
764 	struct block_device *bdev;
765 	struct seg_entry *se;
766 	unsigned int offset;
767 	block_t i;
768 	int err = 0;
769 
770 	bdev = f2fs_target_device(sbi, blkstart, NULL);
771 
772 	for (i = blkstart; i < blkstart + blklen; i++, len++) {
773 		if (i != start) {
774 			struct block_device *bdev2 =
775 				f2fs_target_device(sbi, i, NULL);
776 
777 			if (bdev2 != bdev) {
778 				err = __issue_discard_async(sbi, bdev,
779 						start, len);
780 				if (err)
781 					return err;
782 				bdev = bdev2;
783 				start = i;
784 				len = 0;
785 			}
786 		}
787 
788 		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
789 		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
790 
791 		if (!f2fs_test_and_set_bit(offset, se->discard_map))
792 			sbi->discard_blks--;
793 	}
794 
795 	if (len)
796 		err = __issue_discard_async(sbi, bdev, start, len);
797 	return err;
798 }
799 
800 static void __add_discard_entry(struct f2fs_sb_info *sbi,
801 		struct cp_control *cpc, struct seg_entry *se,
802 		unsigned int start, unsigned int end)
803 {
804 	struct list_head *head = &SM_I(sbi)->discard_list;
805 	struct discard_entry *new, *last;
806 
807 	if (!list_empty(head)) {
808 		last = list_last_entry(head, struct discard_entry, list);
809 		if (START_BLOCK(sbi, cpc->trim_start) + start ==
810 						last->blkaddr + last->len) {
811 			last->len += end - start;
812 			goto done;
813 		}
814 	}
815 
816 	new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
817 	INIT_LIST_HEAD(&new->list);
818 	new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
819 	new->len = end - start;
820 	list_add_tail(&new->list, head);
821 done:
822 	SM_I(sbi)->nr_discards += end - start;
823 }
824 
825 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
826 {
827 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
828 	int max_blocks = sbi->blocks_per_seg;
829 	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
830 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
831 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
832 	unsigned long *discard_map = (unsigned long *)se->discard_map;
833 	unsigned long *dmap = SIT_I(sbi)->tmp_map;
834 	unsigned int start = 0, end = -1;
835 	bool force = (cpc->reason == CP_DISCARD);
836 	int i;
837 
838 	if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
839 		return;
840 
841 	if (!force) {
842 		if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
843 		    SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
844 			return;
845 	}
846 
847 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
848 	for (i = 0; i < entries; i++)
849 		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
850 				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
851 
852 	while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
853 		start = __find_rev_next_bit(dmap, max_blocks, end + 1);
854 		if (start >= max_blocks)
855 			break;
856 
857 		end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
858 		if (force && start && end != max_blocks
859 					&& (end - start) < cpc->trim_minlen)
860 			continue;
861 
862 		__add_discard_entry(sbi, cpc, se, start, end);
863 	}
864 }
865 
866 void release_discard_addrs(struct f2fs_sb_info *sbi)
867 {
868 	struct list_head *head = &(SM_I(sbi)->discard_list);
869 	struct discard_entry *entry, *this;
870 
871 	/* drop caches */
872 	list_for_each_entry_safe(entry, this, head, list) {
873 		list_del(&entry->list);
874 		kmem_cache_free(discard_entry_slab, entry);
875 	}
876 }
877 
878 /*
879  * Should call clear_prefree_segments after checkpoint is done.
880  */
881 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
882 {
883 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
884 	unsigned int segno;
885 
886 	mutex_lock(&dirty_i->seglist_lock);
887 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
888 		__set_test_and_free(sbi, segno);
889 	mutex_unlock(&dirty_i->seglist_lock);
890 }
891 
892 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
893 {
894 	struct list_head *head = &(SM_I(sbi)->discard_list);
895 	struct discard_entry *entry, *this;
896 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
897 	struct blk_plug plug;
898 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
899 	unsigned int start = 0, end = -1;
900 	unsigned int secno, start_segno;
901 	bool force = (cpc->reason == CP_DISCARD);
902 
903 	blk_start_plug(&plug);
904 
905 	mutex_lock(&dirty_i->seglist_lock);
906 
907 	while (1) {
908 		int i;
909 		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
910 		if (start >= MAIN_SEGS(sbi))
911 			break;
912 		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
913 								start + 1);
914 
915 		for (i = start; i < end; i++)
916 			clear_bit(i, prefree_map);
917 
918 		dirty_i->nr_dirty[PRE] -= end - start;
919 
920 		if (!test_opt(sbi, DISCARD))
921 			continue;
922 
923 		if (force && start >= cpc->trim_start &&
924 					(end - 1) <= cpc->trim_end)
925 				continue;
926 
927 		if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
928 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
929 				(end - start) << sbi->log_blocks_per_seg);
930 			continue;
931 		}
932 next:
933 		secno = GET_SECNO(sbi, start);
934 		start_segno = secno * sbi->segs_per_sec;
935 		if (!IS_CURSEC(sbi, secno) &&
936 			!get_valid_blocks(sbi, start, sbi->segs_per_sec))
937 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
938 				sbi->segs_per_sec << sbi->log_blocks_per_seg);
939 
940 		start = start_segno + sbi->segs_per_sec;
941 		if (start < end)
942 			goto next;
943 	}
944 	mutex_unlock(&dirty_i->seglist_lock);
945 
946 	/* send small discards */
947 	list_for_each_entry_safe(entry, this, head, list) {
948 		if (force && entry->len < cpc->trim_minlen)
949 			goto skip;
950 		f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
951 		cpc->trimmed += entry->len;
952 skip:
953 		list_del(&entry->list);
954 		SM_I(sbi)->nr_discards -= entry->len;
955 		kmem_cache_free(discard_entry_slab, entry);
956 	}
957 
958 	blk_finish_plug(&plug);
959 }
960 
961 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
962 {
963 	struct sit_info *sit_i = SIT_I(sbi);
964 
965 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
966 		sit_i->dirty_sentries++;
967 		return false;
968 	}
969 
970 	return true;
971 }
972 
973 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
974 					unsigned int segno, int modified)
975 {
976 	struct seg_entry *se = get_seg_entry(sbi, segno);
977 	se->type = type;
978 	if (modified)
979 		__mark_sit_entry_dirty(sbi, segno);
980 }
981 
982 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
983 {
984 	struct seg_entry *se;
985 	unsigned int segno, offset;
986 	long int new_vblocks;
987 
988 	segno = GET_SEGNO(sbi, blkaddr);
989 
990 	se = get_seg_entry(sbi, segno);
991 	new_vblocks = se->valid_blocks + del;
992 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
993 
994 	f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
995 				(new_vblocks > sbi->blocks_per_seg)));
996 
997 	se->valid_blocks = new_vblocks;
998 	se->mtime = get_mtime(sbi);
999 	SIT_I(sbi)->max_mtime = se->mtime;
1000 
1001 	/* Update valid block bitmap */
1002 	if (del > 0) {
1003 		if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
1004 			f2fs_bug_on(sbi, 1);
1005 		if (f2fs_discard_en(sbi) &&
1006 			!f2fs_test_and_set_bit(offset, se->discard_map))
1007 			sbi->discard_blks--;
1008 	} else {
1009 		if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
1010 			f2fs_bug_on(sbi, 1);
1011 		if (f2fs_discard_en(sbi) &&
1012 			f2fs_test_and_clear_bit(offset, se->discard_map))
1013 			sbi->discard_blks++;
1014 	}
1015 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1016 		se->ckpt_valid_blocks += del;
1017 
1018 	__mark_sit_entry_dirty(sbi, segno);
1019 
1020 	/* update total number of valid blocks to be written in ckpt area */
1021 	SIT_I(sbi)->written_valid_blocks += del;
1022 
1023 	if (sbi->segs_per_sec > 1)
1024 		get_sec_entry(sbi, segno)->valid_blocks += del;
1025 }
1026 
1027 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
1028 {
1029 	update_sit_entry(sbi, new, 1);
1030 	if (GET_SEGNO(sbi, old) != NULL_SEGNO)
1031 		update_sit_entry(sbi, old, -1);
1032 
1033 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
1034 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
1035 }
1036 
1037 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1038 {
1039 	unsigned int segno = GET_SEGNO(sbi, addr);
1040 	struct sit_info *sit_i = SIT_I(sbi);
1041 
1042 	f2fs_bug_on(sbi, addr == NULL_ADDR);
1043 	if (addr == NEW_ADDR)
1044 		return;
1045 
1046 	/* add it into sit main buffer */
1047 	mutex_lock(&sit_i->sentry_lock);
1048 
1049 	update_sit_entry(sbi, addr, -1);
1050 
1051 	/* add it into dirty seglist */
1052 	locate_dirty_segment(sbi, segno);
1053 
1054 	mutex_unlock(&sit_i->sentry_lock);
1055 }
1056 
1057 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1058 {
1059 	struct sit_info *sit_i = SIT_I(sbi);
1060 	unsigned int segno, offset;
1061 	struct seg_entry *se;
1062 	bool is_cp = false;
1063 
1064 	if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1065 		return true;
1066 
1067 	mutex_lock(&sit_i->sentry_lock);
1068 
1069 	segno = GET_SEGNO(sbi, blkaddr);
1070 	se = get_seg_entry(sbi, segno);
1071 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1072 
1073 	if (f2fs_test_bit(offset, se->ckpt_valid_map))
1074 		is_cp = true;
1075 
1076 	mutex_unlock(&sit_i->sentry_lock);
1077 
1078 	return is_cp;
1079 }
1080 
1081 /*
1082  * This function should be resided under the curseg_mutex lock
1083  */
1084 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1085 					struct f2fs_summary *sum)
1086 {
1087 	struct curseg_info *curseg = CURSEG_I(sbi, type);
1088 	void *addr = curseg->sum_blk;
1089 	addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1090 	memcpy(addr, sum, sizeof(struct f2fs_summary));
1091 }
1092 
1093 /*
1094  * Calculate the number of current summary pages for writing
1095  */
1096 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1097 {
1098 	int valid_sum_count = 0;
1099 	int i, sum_in_page;
1100 
1101 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1102 		if (sbi->ckpt->alloc_type[i] == SSR)
1103 			valid_sum_count += sbi->blocks_per_seg;
1104 		else {
1105 			if (for_ra)
1106 				valid_sum_count += le16_to_cpu(
1107 					F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1108 			else
1109 				valid_sum_count += curseg_blkoff(sbi, i);
1110 		}
1111 	}
1112 
1113 	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1114 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1115 	if (valid_sum_count <= sum_in_page)
1116 		return 1;
1117 	else if ((valid_sum_count - sum_in_page) <=
1118 		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1119 		return 2;
1120 	return 3;
1121 }
1122 
1123 /*
1124  * Caller should put this summary page
1125  */
1126 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1127 {
1128 	return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1129 }
1130 
1131 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1132 {
1133 	struct page *page = grab_meta_page(sbi, blk_addr);
1134 	void *dst = page_address(page);
1135 
1136 	if (src)
1137 		memcpy(dst, src, PAGE_SIZE);
1138 	else
1139 		memset(dst, 0, PAGE_SIZE);
1140 	set_page_dirty(page);
1141 	f2fs_put_page(page, 1);
1142 }
1143 
1144 static void write_sum_page(struct f2fs_sb_info *sbi,
1145 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
1146 {
1147 	update_meta_page(sbi, (void *)sum_blk, blk_addr);
1148 }
1149 
1150 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1151 						int type, block_t blk_addr)
1152 {
1153 	struct curseg_info *curseg = CURSEG_I(sbi, type);
1154 	struct page *page = grab_meta_page(sbi, blk_addr);
1155 	struct f2fs_summary_block *src = curseg->sum_blk;
1156 	struct f2fs_summary_block *dst;
1157 
1158 	dst = (struct f2fs_summary_block *)page_address(page);
1159 
1160 	mutex_lock(&curseg->curseg_mutex);
1161 
1162 	down_read(&curseg->journal_rwsem);
1163 	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1164 	up_read(&curseg->journal_rwsem);
1165 
1166 	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1167 	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1168 
1169 	mutex_unlock(&curseg->curseg_mutex);
1170 
1171 	set_page_dirty(page);
1172 	f2fs_put_page(page, 1);
1173 }
1174 
1175 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1176 {
1177 	struct curseg_info *curseg = CURSEG_I(sbi, type);
1178 	unsigned int segno = curseg->segno + 1;
1179 	struct free_segmap_info *free_i = FREE_I(sbi);
1180 
1181 	if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1182 		return !test_bit(segno, free_i->free_segmap);
1183 	return 0;
1184 }
1185 
1186 /*
1187  * Find a new segment from the free segments bitmap to right order
1188  * This function should be returned with success, otherwise BUG
1189  */
1190 static void get_new_segment(struct f2fs_sb_info *sbi,
1191 			unsigned int *newseg, bool new_sec, int dir)
1192 {
1193 	struct free_segmap_info *free_i = FREE_I(sbi);
1194 	unsigned int segno, secno, zoneno;
1195 	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1196 	unsigned int hint = *newseg / sbi->segs_per_sec;
1197 	unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1198 	unsigned int left_start = hint;
1199 	bool init = true;
1200 	int go_left = 0;
1201 	int i;
1202 
1203 	spin_lock(&free_i->segmap_lock);
1204 
1205 	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1206 		segno = find_next_zero_bit(free_i->free_segmap,
1207 				(hint + 1) * sbi->segs_per_sec, *newseg + 1);
1208 		if (segno < (hint + 1) * sbi->segs_per_sec)
1209 			goto got_it;
1210 	}
1211 find_other_zone:
1212 	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1213 	if (secno >= MAIN_SECS(sbi)) {
1214 		if (dir == ALLOC_RIGHT) {
1215 			secno = find_next_zero_bit(free_i->free_secmap,
1216 							MAIN_SECS(sbi), 0);
1217 			f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1218 		} else {
1219 			go_left = 1;
1220 			left_start = hint - 1;
1221 		}
1222 	}
1223 	if (go_left == 0)
1224 		goto skip_left;
1225 
1226 	while (test_bit(left_start, free_i->free_secmap)) {
1227 		if (left_start > 0) {
1228 			left_start--;
1229 			continue;
1230 		}
1231 		left_start = find_next_zero_bit(free_i->free_secmap,
1232 							MAIN_SECS(sbi), 0);
1233 		f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1234 		break;
1235 	}
1236 	secno = left_start;
1237 skip_left:
1238 	hint = secno;
1239 	segno = secno * sbi->segs_per_sec;
1240 	zoneno = secno / sbi->secs_per_zone;
1241 
1242 	/* give up on finding another zone */
1243 	if (!init)
1244 		goto got_it;
1245 	if (sbi->secs_per_zone == 1)
1246 		goto got_it;
1247 	if (zoneno == old_zoneno)
1248 		goto got_it;
1249 	if (dir == ALLOC_LEFT) {
1250 		if (!go_left && zoneno + 1 >= total_zones)
1251 			goto got_it;
1252 		if (go_left && zoneno == 0)
1253 			goto got_it;
1254 	}
1255 	for (i = 0; i < NR_CURSEG_TYPE; i++)
1256 		if (CURSEG_I(sbi, i)->zone == zoneno)
1257 			break;
1258 
1259 	if (i < NR_CURSEG_TYPE) {
1260 		/* zone is in user, try another */
1261 		if (go_left)
1262 			hint = zoneno * sbi->secs_per_zone - 1;
1263 		else if (zoneno + 1 >= total_zones)
1264 			hint = 0;
1265 		else
1266 			hint = (zoneno + 1) * sbi->secs_per_zone;
1267 		init = false;
1268 		goto find_other_zone;
1269 	}
1270 got_it:
1271 	/* set it as dirty segment in free segmap */
1272 	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1273 	__set_inuse(sbi, segno);
1274 	*newseg = segno;
1275 	spin_unlock(&free_i->segmap_lock);
1276 }
1277 
1278 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1279 {
1280 	struct curseg_info *curseg = CURSEG_I(sbi, type);
1281 	struct summary_footer *sum_footer;
1282 
1283 	curseg->segno = curseg->next_segno;
1284 	curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1285 	curseg->next_blkoff = 0;
1286 	curseg->next_segno = NULL_SEGNO;
1287 
1288 	sum_footer = &(curseg->sum_blk->footer);
1289 	memset(sum_footer, 0, sizeof(struct summary_footer));
1290 	if (IS_DATASEG(type))
1291 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1292 	if (IS_NODESEG(type))
1293 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1294 	__set_sit_entry_type(sbi, type, curseg->segno, modified);
1295 }
1296 
1297 /*
1298  * Allocate a current working segment.
1299  * This function always allocates a free segment in LFS manner.
1300  */
1301 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1302 {
1303 	struct curseg_info *curseg = CURSEG_I(sbi, type);
1304 	unsigned int segno = curseg->segno;
1305 	int dir = ALLOC_LEFT;
1306 
1307 	write_sum_page(sbi, curseg->sum_blk,
1308 				GET_SUM_BLOCK(sbi, segno));
1309 	if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1310 		dir = ALLOC_RIGHT;
1311 
1312 	if (test_opt(sbi, NOHEAP))
1313 		dir = ALLOC_RIGHT;
1314 
1315 	get_new_segment(sbi, &segno, new_sec, dir);
1316 	curseg->next_segno = segno;
1317 	reset_curseg(sbi, type, 1);
1318 	curseg->alloc_type = LFS;
1319 }
1320 
1321 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1322 			struct curseg_info *seg, block_t start)
1323 {
1324 	struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1325 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1326 	unsigned long *target_map = SIT_I(sbi)->tmp_map;
1327 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1328 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1329 	int i, pos;
1330 
1331 	for (i = 0; i < entries; i++)
1332 		target_map[i] = ckpt_map[i] | cur_map[i];
1333 
1334 	pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1335 
1336 	seg->next_blkoff = pos;
1337 }
1338 
1339 /*
1340  * If a segment is written by LFS manner, next block offset is just obtained
1341  * by increasing the current block offset. However, if a segment is written by
1342  * SSR manner, next block offset obtained by calling __next_free_blkoff
1343  */
1344 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1345 				struct curseg_info *seg)
1346 {
1347 	if (seg->alloc_type == SSR)
1348 		__next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1349 	else
1350 		seg->next_blkoff++;
1351 }
1352 
1353 /*
1354  * This function always allocates a used segment(from dirty seglist) by SSR
1355  * manner, so it should recover the existing segment information of valid blocks
1356  */
1357 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1358 {
1359 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1360 	struct curseg_info *curseg = CURSEG_I(sbi, type);
1361 	unsigned int new_segno = curseg->next_segno;
1362 	struct f2fs_summary_block *sum_node;
1363 	struct page *sum_page;
1364 
1365 	write_sum_page(sbi, curseg->sum_blk,
1366 				GET_SUM_BLOCK(sbi, curseg->segno));
1367 	__set_test_and_inuse(sbi, new_segno);
1368 
1369 	mutex_lock(&dirty_i->seglist_lock);
1370 	__remove_dirty_segment(sbi, new_segno, PRE);
1371 	__remove_dirty_segment(sbi, new_segno, DIRTY);
1372 	mutex_unlock(&dirty_i->seglist_lock);
1373 
1374 	reset_curseg(sbi, type, 1);
1375 	curseg->alloc_type = SSR;
1376 	__next_free_blkoff(sbi, curseg, 0);
1377 
1378 	if (reuse) {
1379 		sum_page = get_sum_page(sbi, new_segno);
1380 		sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1381 		memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1382 		f2fs_put_page(sum_page, 1);
1383 	}
1384 }
1385 
1386 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1387 {
1388 	struct curseg_info *curseg = CURSEG_I(sbi, type);
1389 	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1390 
1391 	if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0, 0))
1392 		return v_ops->get_victim(sbi,
1393 				&(curseg)->next_segno, BG_GC, type, SSR);
1394 
1395 	/* For data segments, let's do SSR more intensively */
1396 	for (; type >= CURSEG_HOT_DATA; type--)
1397 		if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1398 						BG_GC, type, SSR))
1399 			return 1;
1400 	return 0;
1401 }
1402 
1403 /*
1404  * flush out current segment and replace it with new segment
1405  * This function should be returned with success, otherwise BUG
1406  */
1407 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1408 						int type, bool force)
1409 {
1410 	struct curseg_info *curseg = CURSEG_I(sbi, type);
1411 
1412 	if (force)
1413 		new_curseg(sbi, type, true);
1414 	else if (type == CURSEG_WARM_NODE)
1415 		new_curseg(sbi, type, false);
1416 	else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1417 		new_curseg(sbi, type, false);
1418 	else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1419 		change_curseg(sbi, type, true);
1420 	else
1421 		new_curseg(sbi, type, false);
1422 
1423 	stat_inc_seg_type(sbi, curseg);
1424 }
1425 
1426 void allocate_new_segments(struct f2fs_sb_info *sbi)
1427 {
1428 	struct curseg_info *curseg;
1429 	unsigned int old_segno;
1430 	int i;
1431 
1432 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1433 		curseg = CURSEG_I(sbi, i);
1434 		old_segno = curseg->segno;
1435 		SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
1436 		locate_dirty_segment(sbi, old_segno);
1437 	}
1438 }
1439 
1440 static const struct segment_allocation default_salloc_ops = {
1441 	.allocate_segment = allocate_segment_by_default,
1442 };
1443 
1444 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1445 {
1446 	__u64 start = F2FS_BYTES_TO_BLK(range->start);
1447 	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1448 	unsigned int start_segno, end_segno;
1449 	struct cp_control cpc;
1450 	int err = 0;
1451 
1452 	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1453 		return -EINVAL;
1454 
1455 	cpc.trimmed = 0;
1456 	if (end <= MAIN_BLKADDR(sbi))
1457 		goto out;
1458 
1459 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1460 		f2fs_msg(sbi->sb, KERN_WARNING,
1461 			"Found FS corruption, run fsck to fix.");
1462 		goto out;
1463 	}
1464 
1465 	/* start/end segment number in main_area */
1466 	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1467 	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1468 						GET_SEGNO(sbi, end);
1469 	cpc.reason = CP_DISCARD;
1470 	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1471 
1472 	/* do checkpoint to issue discard commands safely */
1473 	for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1474 		cpc.trim_start = start_segno;
1475 
1476 		if (sbi->discard_blks == 0)
1477 			break;
1478 		else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1479 			cpc.trim_end = end_segno;
1480 		else
1481 			cpc.trim_end = min_t(unsigned int,
1482 				rounddown(start_segno +
1483 				BATCHED_TRIM_SEGMENTS(sbi),
1484 				sbi->segs_per_sec) - 1, end_segno);
1485 
1486 		mutex_lock(&sbi->gc_mutex);
1487 		err = write_checkpoint(sbi, &cpc);
1488 		mutex_unlock(&sbi->gc_mutex);
1489 		if (err)
1490 			break;
1491 
1492 		schedule();
1493 	}
1494 out:
1495 	range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1496 	return err;
1497 }
1498 
1499 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1500 {
1501 	struct curseg_info *curseg = CURSEG_I(sbi, type);
1502 	if (curseg->next_blkoff < sbi->blocks_per_seg)
1503 		return true;
1504 	return false;
1505 }
1506 
1507 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1508 {
1509 	if (p_type == DATA)
1510 		return CURSEG_HOT_DATA;
1511 	else
1512 		return CURSEG_HOT_NODE;
1513 }
1514 
1515 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1516 {
1517 	if (p_type == DATA) {
1518 		struct inode *inode = page->mapping->host;
1519 
1520 		if (S_ISDIR(inode->i_mode))
1521 			return CURSEG_HOT_DATA;
1522 		else
1523 			return CURSEG_COLD_DATA;
1524 	} else {
1525 		if (IS_DNODE(page) && is_cold_node(page))
1526 			return CURSEG_WARM_NODE;
1527 		else
1528 			return CURSEG_COLD_NODE;
1529 	}
1530 }
1531 
1532 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1533 {
1534 	if (p_type == DATA) {
1535 		struct inode *inode = page->mapping->host;
1536 
1537 		if (S_ISDIR(inode->i_mode))
1538 			return CURSEG_HOT_DATA;
1539 		else if (is_cold_data(page) || file_is_cold(inode))
1540 			return CURSEG_COLD_DATA;
1541 		else
1542 			return CURSEG_WARM_DATA;
1543 	} else {
1544 		if (IS_DNODE(page))
1545 			return is_cold_node(page) ? CURSEG_WARM_NODE :
1546 						CURSEG_HOT_NODE;
1547 		else
1548 			return CURSEG_COLD_NODE;
1549 	}
1550 }
1551 
1552 static int __get_segment_type(struct page *page, enum page_type p_type)
1553 {
1554 	switch (F2FS_P_SB(page)->active_logs) {
1555 	case 2:
1556 		return __get_segment_type_2(page, p_type);
1557 	case 4:
1558 		return __get_segment_type_4(page, p_type);
1559 	}
1560 	/* NR_CURSEG_TYPE(6) logs by default */
1561 	f2fs_bug_on(F2FS_P_SB(page),
1562 		F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1563 	return __get_segment_type_6(page, p_type);
1564 }
1565 
1566 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1567 		block_t old_blkaddr, block_t *new_blkaddr,
1568 		struct f2fs_summary *sum, int type)
1569 {
1570 	struct sit_info *sit_i = SIT_I(sbi);
1571 	struct curseg_info *curseg = CURSEG_I(sbi, type);
1572 
1573 	mutex_lock(&curseg->curseg_mutex);
1574 	mutex_lock(&sit_i->sentry_lock);
1575 
1576 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1577 
1578 	/*
1579 	 * __add_sum_entry should be resided under the curseg_mutex
1580 	 * because, this function updates a summary entry in the
1581 	 * current summary block.
1582 	 */
1583 	__add_sum_entry(sbi, type, sum);
1584 
1585 	__refresh_next_blkoff(sbi, curseg);
1586 
1587 	stat_inc_block_count(sbi, curseg);
1588 
1589 	if (!__has_curseg_space(sbi, type))
1590 		sit_i->s_ops->allocate_segment(sbi, type, false);
1591 	/*
1592 	 * SIT information should be updated before segment allocation,
1593 	 * since SSR needs latest valid block information.
1594 	 */
1595 	refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1596 
1597 	mutex_unlock(&sit_i->sentry_lock);
1598 
1599 	if (page && IS_NODESEG(type))
1600 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1601 
1602 	mutex_unlock(&curseg->curseg_mutex);
1603 }
1604 
1605 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1606 {
1607 	int type = __get_segment_type(fio->page, fio->type);
1608 	int err;
1609 
1610 	if (fio->type == NODE || fio->type == DATA)
1611 		mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1612 reallocate:
1613 	allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1614 					&fio->new_blkaddr, sum, type);
1615 
1616 	/* writeout dirty page into bdev */
1617 	err = f2fs_submit_page_mbio(fio);
1618 	if (err == -EAGAIN) {
1619 		fio->old_blkaddr = fio->new_blkaddr;
1620 		goto reallocate;
1621 	}
1622 
1623 	if (fio->type == NODE || fio->type == DATA)
1624 		mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
1625 }
1626 
1627 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1628 {
1629 	struct f2fs_io_info fio = {
1630 		.sbi = sbi,
1631 		.type = META,
1632 		.op = REQ_OP_WRITE,
1633 		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
1634 		.old_blkaddr = page->index,
1635 		.new_blkaddr = page->index,
1636 		.page = page,
1637 		.encrypted_page = NULL,
1638 	};
1639 
1640 	if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1641 		fio.op_flags &= ~REQ_META;
1642 
1643 	set_page_writeback(page);
1644 	f2fs_submit_page_mbio(&fio);
1645 }
1646 
1647 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1648 {
1649 	struct f2fs_summary sum;
1650 
1651 	set_summary(&sum, nid, 0, 0);
1652 	do_write_page(&sum, fio);
1653 }
1654 
1655 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1656 {
1657 	struct f2fs_sb_info *sbi = fio->sbi;
1658 	struct f2fs_summary sum;
1659 	struct node_info ni;
1660 
1661 	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1662 	get_node_info(sbi, dn->nid, &ni);
1663 	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1664 	do_write_page(&sum, fio);
1665 	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1666 }
1667 
1668 void rewrite_data_page(struct f2fs_io_info *fio)
1669 {
1670 	fio->new_blkaddr = fio->old_blkaddr;
1671 	stat_inc_inplace_blocks(fio->sbi);
1672 	f2fs_submit_page_mbio(fio);
1673 }
1674 
1675 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1676 				block_t old_blkaddr, block_t new_blkaddr,
1677 				bool recover_curseg, bool recover_newaddr)
1678 {
1679 	struct sit_info *sit_i = SIT_I(sbi);
1680 	struct curseg_info *curseg;
1681 	unsigned int segno, old_cursegno;
1682 	struct seg_entry *se;
1683 	int type;
1684 	unsigned short old_blkoff;
1685 
1686 	segno = GET_SEGNO(sbi, new_blkaddr);
1687 	se = get_seg_entry(sbi, segno);
1688 	type = se->type;
1689 
1690 	if (!recover_curseg) {
1691 		/* for recovery flow */
1692 		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1693 			if (old_blkaddr == NULL_ADDR)
1694 				type = CURSEG_COLD_DATA;
1695 			else
1696 				type = CURSEG_WARM_DATA;
1697 		}
1698 	} else {
1699 		if (!IS_CURSEG(sbi, segno))
1700 			type = CURSEG_WARM_DATA;
1701 	}
1702 
1703 	curseg = CURSEG_I(sbi, type);
1704 
1705 	mutex_lock(&curseg->curseg_mutex);
1706 	mutex_lock(&sit_i->sentry_lock);
1707 
1708 	old_cursegno = curseg->segno;
1709 	old_blkoff = curseg->next_blkoff;
1710 
1711 	/* change the current segment */
1712 	if (segno != curseg->segno) {
1713 		curseg->next_segno = segno;
1714 		change_curseg(sbi, type, true);
1715 	}
1716 
1717 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1718 	__add_sum_entry(sbi, type, sum);
1719 
1720 	if (!recover_curseg || recover_newaddr)
1721 		update_sit_entry(sbi, new_blkaddr, 1);
1722 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1723 		update_sit_entry(sbi, old_blkaddr, -1);
1724 
1725 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1726 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1727 
1728 	locate_dirty_segment(sbi, old_cursegno);
1729 
1730 	if (recover_curseg) {
1731 		if (old_cursegno != curseg->segno) {
1732 			curseg->next_segno = old_cursegno;
1733 			change_curseg(sbi, type, true);
1734 		}
1735 		curseg->next_blkoff = old_blkoff;
1736 	}
1737 
1738 	mutex_unlock(&sit_i->sentry_lock);
1739 	mutex_unlock(&curseg->curseg_mutex);
1740 }
1741 
1742 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1743 				block_t old_addr, block_t new_addr,
1744 				unsigned char version, bool recover_curseg,
1745 				bool recover_newaddr)
1746 {
1747 	struct f2fs_summary sum;
1748 
1749 	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1750 
1751 	__f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1752 					recover_curseg, recover_newaddr);
1753 
1754 	f2fs_update_data_blkaddr(dn, new_addr);
1755 }
1756 
1757 void f2fs_wait_on_page_writeback(struct page *page,
1758 				enum page_type type, bool ordered)
1759 {
1760 	if (PageWriteback(page)) {
1761 		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1762 
1763 		f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
1764 		if (ordered)
1765 			wait_on_page_writeback(page);
1766 		else
1767 			wait_for_stable_page(page);
1768 	}
1769 }
1770 
1771 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1772 							block_t blkaddr)
1773 {
1774 	struct page *cpage;
1775 
1776 	if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1777 		return;
1778 
1779 	cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1780 	if (cpage) {
1781 		f2fs_wait_on_page_writeback(cpage, DATA, true);
1782 		f2fs_put_page(cpage, 1);
1783 	}
1784 }
1785 
1786 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1787 {
1788 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1789 	struct curseg_info *seg_i;
1790 	unsigned char *kaddr;
1791 	struct page *page;
1792 	block_t start;
1793 	int i, j, offset;
1794 
1795 	start = start_sum_block(sbi);
1796 
1797 	page = get_meta_page(sbi, start++);
1798 	kaddr = (unsigned char *)page_address(page);
1799 
1800 	/* Step 1: restore nat cache */
1801 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1802 	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
1803 
1804 	/* Step 2: restore sit cache */
1805 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1806 	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
1807 	offset = 2 * SUM_JOURNAL_SIZE;
1808 
1809 	/* Step 3: restore summary entries */
1810 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1811 		unsigned short blk_off;
1812 		unsigned int segno;
1813 
1814 		seg_i = CURSEG_I(sbi, i);
1815 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1816 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1817 		seg_i->next_segno = segno;
1818 		reset_curseg(sbi, i, 0);
1819 		seg_i->alloc_type = ckpt->alloc_type[i];
1820 		seg_i->next_blkoff = blk_off;
1821 
1822 		if (seg_i->alloc_type == SSR)
1823 			blk_off = sbi->blocks_per_seg;
1824 
1825 		for (j = 0; j < blk_off; j++) {
1826 			struct f2fs_summary *s;
1827 			s = (struct f2fs_summary *)(kaddr + offset);
1828 			seg_i->sum_blk->entries[j] = *s;
1829 			offset += SUMMARY_SIZE;
1830 			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
1831 						SUM_FOOTER_SIZE)
1832 				continue;
1833 
1834 			f2fs_put_page(page, 1);
1835 			page = NULL;
1836 
1837 			page = get_meta_page(sbi, start++);
1838 			kaddr = (unsigned char *)page_address(page);
1839 			offset = 0;
1840 		}
1841 	}
1842 	f2fs_put_page(page, 1);
1843 	return 0;
1844 }
1845 
1846 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1847 {
1848 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1849 	struct f2fs_summary_block *sum;
1850 	struct curseg_info *curseg;
1851 	struct page *new;
1852 	unsigned short blk_off;
1853 	unsigned int segno = 0;
1854 	block_t blk_addr = 0;
1855 
1856 	/* get segment number and block addr */
1857 	if (IS_DATASEG(type)) {
1858 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1859 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1860 							CURSEG_HOT_DATA]);
1861 		if (__exist_node_summaries(sbi))
1862 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1863 		else
1864 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1865 	} else {
1866 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
1867 							CURSEG_HOT_NODE]);
1868 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1869 							CURSEG_HOT_NODE]);
1870 		if (__exist_node_summaries(sbi))
1871 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1872 							type - CURSEG_HOT_NODE);
1873 		else
1874 			blk_addr = GET_SUM_BLOCK(sbi, segno);
1875 	}
1876 
1877 	new = get_meta_page(sbi, blk_addr);
1878 	sum = (struct f2fs_summary_block *)page_address(new);
1879 
1880 	if (IS_NODESEG(type)) {
1881 		if (__exist_node_summaries(sbi)) {
1882 			struct f2fs_summary *ns = &sum->entries[0];
1883 			int i;
1884 			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1885 				ns->version = 0;
1886 				ns->ofs_in_node = 0;
1887 			}
1888 		} else {
1889 			int err;
1890 
1891 			err = restore_node_summary(sbi, segno, sum);
1892 			if (err) {
1893 				f2fs_put_page(new, 1);
1894 				return err;
1895 			}
1896 		}
1897 	}
1898 
1899 	/* set uncompleted segment to curseg */
1900 	curseg = CURSEG_I(sbi, type);
1901 	mutex_lock(&curseg->curseg_mutex);
1902 
1903 	/* update journal info */
1904 	down_write(&curseg->journal_rwsem);
1905 	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1906 	up_write(&curseg->journal_rwsem);
1907 
1908 	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1909 	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
1910 	curseg->next_segno = segno;
1911 	reset_curseg(sbi, type, 0);
1912 	curseg->alloc_type = ckpt->alloc_type[type];
1913 	curseg->next_blkoff = blk_off;
1914 	mutex_unlock(&curseg->curseg_mutex);
1915 	f2fs_put_page(new, 1);
1916 	return 0;
1917 }
1918 
1919 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1920 {
1921 	int type = CURSEG_HOT_DATA;
1922 	int err;
1923 
1924 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
1925 		int npages = npages_for_summary_flush(sbi, true);
1926 
1927 		if (npages >= 2)
1928 			ra_meta_pages(sbi, start_sum_block(sbi), npages,
1929 							META_CP, true);
1930 
1931 		/* restore for compacted data summary */
1932 		if (read_compacted_summaries(sbi))
1933 			return -EINVAL;
1934 		type = CURSEG_HOT_NODE;
1935 	}
1936 
1937 	if (__exist_node_summaries(sbi))
1938 		ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1939 					NR_CURSEG_TYPE - type, META_CP, true);
1940 
1941 	for (; type <= CURSEG_COLD_NODE; type++) {
1942 		err = read_normal_summaries(sbi, type);
1943 		if (err)
1944 			return err;
1945 	}
1946 
1947 	return 0;
1948 }
1949 
1950 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1951 {
1952 	struct page *page;
1953 	unsigned char *kaddr;
1954 	struct f2fs_summary *summary;
1955 	struct curseg_info *seg_i;
1956 	int written_size = 0;
1957 	int i, j;
1958 
1959 	page = grab_meta_page(sbi, blkaddr++);
1960 	kaddr = (unsigned char *)page_address(page);
1961 
1962 	/* Step 1: write nat cache */
1963 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1964 	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
1965 	written_size += SUM_JOURNAL_SIZE;
1966 
1967 	/* Step 2: write sit cache */
1968 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1969 	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
1970 	written_size += SUM_JOURNAL_SIZE;
1971 
1972 	/* Step 3: write summary entries */
1973 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1974 		unsigned short blkoff;
1975 		seg_i = CURSEG_I(sbi, i);
1976 		if (sbi->ckpt->alloc_type[i] == SSR)
1977 			blkoff = sbi->blocks_per_seg;
1978 		else
1979 			blkoff = curseg_blkoff(sbi, i);
1980 
1981 		for (j = 0; j < blkoff; j++) {
1982 			if (!page) {
1983 				page = grab_meta_page(sbi, blkaddr++);
1984 				kaddr = (unsigned char *)page_address(page);
1985 				written_size = 0;
1986 			}
1987 			summary = (struct f2fs_summary *)(kaddr + written_size);
1988 			*summary = seg_i->sum_blk->entries[j];
1989 			written_size += SUMMARY_SIZE;
1990 
1991 			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
1992 							SUM_FOOTER_SIZE)
1993 				continue;
1994 
1995 			set_page_dirty(page);
1996 			f2fs_put_page(page, 1);
1997 			page = NULL;
1998 		}
1999 	}
2000 	if (page) {
2001 		set_page_dirty(page);
2002 		f2fs_put_page(page, 1);
2003 	}
2004 }
2005 
2006 static void write_normal_summaries(struct f2fs_sb_info *sbi,
2007 					block_t blkaddr, int type)
2008 {
2009 	int i, end;
2010 	if (IS_DATASEG(type))
2011 		end = type + NR_CURSEG_DATA_TYPE;
2012 	else
2013 		end = type + NR_CURSEG_NODE_TYPE;
2014 
2015 	for (i = type; i < end; i++)
2016 		write_current_sum_page(sbi, i, blkaddr + (i - type));
2017 }
2018 
2019 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2020 {
2021 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
2022 		write_compacted_summaries(sbi, start_blk);
2023 	else
2024 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
2025 }
2026 
2027 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
2028 {
2029 	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
2030 }
2031 
2032 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
2033 					unsigned int val, int alloc)
2034 {
2035 	int i;
2036 
2037 	if (type == NAT_JOURNAL) {
2038 		for (i = 0; i < nats_in_cursum(journal); i++) {
2039 			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
2040 				return i;
2041 		}
2042 		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
2043 			return update_nats_in_cursum(journal, 1);
2044 	} else if (type == SIT_JOURNAL) {
2045 		for (i = 0; i < sits_in_cursum(journal); i++)
2046 			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
2047 				return i;
2048 		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
2049 			return update_sits_in_cursum(journal, 1);
2050 	}
2051 	return -1;
2052 }
2053 
2054 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
2055 					unsigned int segno)
2056 {
2057 	return get_meta_page(sbi, current_sit_addr(sbi, segno));
2058 }
2059 
2060 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
2061 					unsigned int start)
2062 {
2063 	struct sit_info *sit_i = SIT_I(sbi);
2064 	struct page *src_page, *dst_page;
2065 	pgoff_t src_off, dst_off;
2066 	void *src_addr, *dst_addr;
2067 
2068 	src_off = current_sit_addr(sbi, start);
2069 	dst_off = next_sit_addr(sbi, src_off);
2070 
2071 	/* get current sit block page without lock */
2072 	src_page = get_meta_page(sbi, src_off);
2073 	dst_page = grab_meta_page(sbi, dst_off);
2074 	f2fs_bug_on(sbi, PageDirty(src_page));
2075 
2076 	src_addr = page_address(src_page);
2077 	dst_addr = page_address(dst_page);
2078 	memcpy(dst_addr, src_addr, PAGE_SIZE);
2079 
2080 	set_page_dirty(dst_page);
2081 	f2fs_put_page(src_page, 1);
2082 
2083 	set_to_next_sit(sit_i, start);
2084 
2085 	return dst_page;
2086 }
2087 
2088 static struct sit_entry_set *grab_sit_entry_set(void)
2089 {
2090 	struct sit_entry_set *ses =
2091 			f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
2092 
2093 	ses->entry_cnt = 0;
2094 	INIT_LIST_HEAD(&ses->set_list);
2095 	return ses;
2096 }
2097 
2098 static void release_sit_entry_set(struct sit_entry_set *ses)
2099 {
2100 	list_del(&ses->set_list);
2101 	kmem_cache_free(sit_entry_set_slab, ses);
2102 }
2103 
2104 static void adjust_sit_entry_set(struct sit_entry_set *ses,
2105 						struct list_head *head)
2106 {
2107 	struct sit_entry_set *next = ses;
2108 
2109 	if (list_is_last(&ses->set_list, head))
2110 		return;
2111 
2112 	list_for_each_entry_continue(next, head, set_list)
2113 		if (ses->entry_cnt <= next->entry_cnt)
2114 			break;
2115 
2116 	list_move_tail(&ses->set_list, &next->set_list);
2117 }
2118 
2119 static void add_sit_entry(unsigned int segno, struct list_head *head)
2120 {
2121 	struct sit_entry_set *ses;
2122 	unsigned int start_segno = START_SEGNO(segno);
2123 
2124 	list_for_each_entry(ses, head, set_list) {
2125 		if (ses->start_segno == start_segno) {
2126 			ses->entry_cnt++;
2127 			adjust_sit_entry_set(ses, head);
2128 			return;
2129 		}
2130 	}
2131 
2132 	ses = grab_sit_entry_set();
2133 
2134 	ses->start_segno = start_segno;
2135 	ses->entry_cnt++;
2136 	list_add(&ses->set_list, head);
2137 }
2138 
2139 static void add_sits_in_set(struct f2fs_sb_info *sbi)
2140 {
2141 	struct f2fs_sm_info *sm_info = SM_I(sbi);
2142 	struct list_head *set_list = &sm_info->sit_entry_set;
2143 	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2144 	unsigned int segno;
2145 
2146 	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2147 		add_sit_entry(segno, set_list);
2148 }
2149 
2150 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2151 {
2152 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2153 	struct f2fs_journal *journal = curseg->journal;
2154 	int i;
2155 
2156 	down_write(&curseg->journal_rwsem);
2157 	for (i = 0; i < sits_in_cursum(journal); i++) {
2158 		unsigned int segno;
2159 		bool dirtied;
2160 
2161 		segno = le32_to_cpu(segno_in_journal(journal, i));
2162 		dirtied = __mark_sit_entry_dirty(sbi, segno);
2163 
2164 		if (!dirtied)
2165 			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2166 	}
2167 	update_sits_in_cursum(journal, -i);
2168 	up_write(&curseg->journal_rwsem);
2169 }
2170 
2171 /*
2172  * CP calls this function, which flushes SIT entries including sit_journal,
2173  * and moves prefree segs to free segs.
2174  */
2175 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2176 {
2177 	struct sit_info *sit_i = SIT_I(sbi);
2178 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2179 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2180 	struct f2fs_journal *journal = curseg->journal;
2181 	struct sit_entry_set *ses, *tmp;
2182 	struct list_head *head = &SM_I(sbi)->sit_entry_set;
2183 	bool to_journal = true;
2184 	struct seg_entry *se;
2185 
2186 	mutex_lock(&sit_i->sentry_lock);
2187 
2188 	if (!sit_i->dirty_sentries)
2189 		goto out;
2190 
2191 	/*
2192 	 * add and account sit entries of dirty bitmap in sit entry
2193 	 * set temporarily
2194 	 */
2195 	add_sits_in_set(sbi);
2196 
2197 	/*
2198 	 * if there are no enough space in journal to store dirty sit
2199 	 * entries, remove all entries from journal and add and account
2200 	 * them in sit entry set.
2201 	 */
2202 	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2203 		remove_sits_in_journal(sbi);
2204 
2205 	/*
2206 	 * there are two steps to flush sit entries:
2207 	 * #1, flush sit entries to journal in current cold data summary block.
2208 	 * #2, flush sit entries to sit page.
2209 	 */
2210 	list_for_each_entry_safe(ses, tmp, head, set_list) {
2211 		struct page *page = NULL;
2212 		struct f2fs_sit_block *raw_sit = NULL;
2213 		unsigned int start_segno = ses->start_segno;
2214 		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2215 						(unsigned long)MAIN_SEGS(sbi));
2216 		unsigned int segno = start_segno;
2217 
2218 		if (to_journal &&
2219 			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2220 			to_journal = false;
2221 
2222 		if (to_journal) {
2223 			down_write(&curseg->journal_rwsem);
2224 		} else {
2225 			page = get_next_sit_page(sbi, start_segno);
2226 			raw_sit = page_address(page);
2227 		}
2228 
2229 		/* flush dirty sit entries in region of current sit set */
2230 		for_each_set_bit_from(segno, bitmap, end) {
2231 			int offset, sit_offset;
2232 
2233 			se = get_seg_entry(sbi, segno);
2234 
2235 			/* add discard candidates */
2236 			if (cpc->reason != CP_DISCARD) {
2237 				cpc->trim_start = segno;
2238 				add_discard_addrs(sbi, cpc);
2239 			}
2240 
2241 			if (to_journal) {
2242 				offset = lookup_journal_in_cursum(journal,
2243 							SIT_JOURNAL, segno, 1);
2244 				f2fs_bug_on(sbi, offset < 0);
2245 				segno_in_journal(journal, offset) =
2246 							cpu_to_le32(segno);
2247 				seg_info_to_raw_sit(se,
2248 					&sit_in_journal(journal, offset));
2249 			} else {
2250 				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2251 				seg_info_to_raw_sit(se,
2252 						&raw_sit->entries[sit_offset]);
2253 			}
2254 
2255 			__clear_bit(segno, bitmap);
2256 			sit_i->dirty_sentries--;
2257 			ses->entry_cnt--;
2258 		}
2259 
2260 		if (to_journal)
2261 			up_write(&curseg->journal_rwsem);
2262 		else
2263 			f2fs_put_page(page, 1);
2264 
2265 		f2fs_bug_on(sbi, ses->entry_cnt);
2266 		release_sit_entry_set(ses);
2267 	}
2268 
2269 	f2fs_bug_on(sbi, !list_empty(head));
2270 	f2fs_bug_on(sbi, sit_i->dirty_sentries);
2271 out:
2272 	if (cpc->reason == CP_DISCARD) {
2273 		__u64 trim_start = cpc->trim_start;
2274 
2275 		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2276 			add_discard_addrs(sbi, cpc);
2277 
2278 		cpc->trim_start = trim_start;
2279 	}
2280 	mutex_unlock(&sit_i->sentry_lock);
2281 
2282 	set_prefree_as_free_segments(sbi);
2283 }
2284 
2285 static int build_sit_info(struct f2fs_sb_info *sbi)
2286 {
2287 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2288 	struct sit_info *sit_i;
2289 	unsigned int sit_segs, start;
2290 	char *src_bitmap, *dst_bitmap;
2291 	unsigned int bitmap_size;
2292 
2293 	/* allocate memory for SIT information */
2294 	sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2295 	if (!sit_i)
2296 		return -ENOMEM;
2297 
2298 	SM_I(sbi)->sit_info = sit_i;
2299 
2300 	sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2301 					sizeof(struct seg_entry), GFP_KERNEL);
2302 	if (!sit_i->sentries)
2303 		return -ENOMEM;
2304 
2305 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2306 	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2307 	if (!sit_i->dirty_sentries_bitmap)
2308 		return -ENOMEM;
2309 
2310 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
2311 		sit_i->sentries[start].cur_valid_map
2312 			= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2313 		sit_i->sentries[start].ckpt_valid_map
2314 			= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2315 		if (!sit_i->sentries[start].cur_valid_map ||
2316 				!sit_i->sentries[start].ckpt_valid_map)
2317 			return -ENOMEM;
2318 
2319 		if (f2fs_discard_en(sbi)) {
2320 			sit_i->sentries[start].discard_map
2321 				= kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2322 			if (!sit_i->sentries[start].discard_map)
2323 				return -ENOMEM;
2324 		}
2325 	}
2326 
2327 	sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2328 	if (!sit_i->tmp_map)
2329 		return -ENOMEM;
2330 
2331 	if (sbi->segs_per_sec > 1) {
2332 		sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2333 					sizeof(struct sec_entry), GFP_KERNEL);
2334 		if (!sit_i->sec_entries)
2335 			return -ENOMEM;
2336 	}
2337 
2338 	/* get information related with SIT */
2339 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2340 
2341 	/* setup SIT bitmap from ckeckpoint pack */
2342 	bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2343 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2344 
2345 	dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2346 	if (!dst_bitmap)
2347 		return -ENOMEM;
2348 
2349 	/* init SIT information */
2350 	sit_i->s_ops = &default_salloc_ops;
2351 
2352 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2353 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2354 	sit_i->written_valid_blocks = 0;
2355 	sit_i->sit_bitmap = dst_bitmap;
2356 	sit_i->bitmap_size = bitmap_size;
2357 	sit_i->dirty_sentries = 0;
2358 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2359 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2360 	sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2361 	mutex_init(&sit_i->sentry_lock);
2362 	return 0;
2363 }
2364 
2365 static int build_free_segmap(struct f2fs_sb_info *sbi)
2366 {
2367 	struct free_segmap_info *free_i;
2368 	unsigned int bitmap_size, sec_bitmap_size;
2369 
2370 	/* allocate memory for free segmap information */
2371 	free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2372 	if (!free_i)
2373 		return -ENOMEM;
2374 
2375 	SM_I(sbi)->free_info = free_i;
2376 
2377 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2378 	free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2379 	if (!free_i->free_segmap)
2380 		return -ENOMEM;
2381 
2382 	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2383 	free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2384 	if (!free_i->free_secmap)
2385 		return -ENOMEM;
2386 
2387 	/* set all segments as dirty temporarily */
2388 	memset(free_i->free_segmap, 0xff, bitmap_size);
2389 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2390 
2391 	/* init free segmap information */
2392 	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2393 	free_i->free_segments = 0;
2394 	free_i->free_sections = 0;
2395 	spin_lock_init(&free_i->segmap_lock);
2396 	return 0;
2397 }
2398 
2399 static int build_curseg(struct f2fs_sb_info *sbi)
2400 {
2401 	struct curseg_info *array;
2402 	int i;
2403 
2404 	array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2405 	if (!array)
2406 		return -ENOMEM;
2407 
2408 	SM_I(sbi)->curseg_array = array;
2409 
2410 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
2411 		mutex_init(&array[i].curseg_mutex);
2412 		array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2413 		if (!array[i].sum_blk)
2414 			return -ENOMEM;
2415 		init_rwsem(&array[i].journal_rwsem);
2416 		array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2417 							GFP_KERNEL);
2418 		if (!array[i].journal)
2419 			return -ENOMEM;
2420 		array[i].segno = NULL_SEGNO;
2421 		array[i].next_blkoff = 0;
2422 	}
2423 	return restore_curseg_summaries(sbi);
2424 }
2425 
2426 static void build_sit_entries(struct f2fs_sb_info *sbi)
2427 {
2428 	struct sit_info *sit_i = SIT_I(sbi);
2429 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2430 	struct f2fs_journal *journal = curseg->journal;
2431 	struct seg_entry *se;
2432 	struct f2fs_sit_entry sit;
2433 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
2434 	unsigned int i, start, end;
2435 	unsigned int readed, start_blk = 0;
2436 
2437 	do {
2438 		readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
2439 							META_SIT, true);
2440 
2441 		start = start_blk * sit_i->sents_per_block;
2442 		end = (start_blk + readed) * sit_i->sents_per_block;
2443 
2444 		for (; start < end && start < MAIN_SEGS(sbi); start++) {
2445 			struct f2fs_sit_block *sit_blk;
2446 			struct page *page;
2447 
2448 			se = &sit_i->sentries[start];
2449 			page = get_current_sit_page(sbi, start);
2450 			sit_blk = (struct f2fs_sit_block *)page_address(page);
2451 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2452 			f2fs_put_page(page, 1);
2453 
2454 			check_block_count(sbi, start, &sit);
2455 			seg_info_from_raw_sit(se, &sit);
2456 
2457 			/* build discard map only one time */
2458 			if (f2fs_discard_en(sbi)) {
2459 				memcpy(se->discard_map, se->cur_valid_map,
2460 							SIT_VBLOCK_MAP_SIZE);
2461 				sbi->discard_blks += sbi->blocks_per_seg -
2462 							se->valid_blocks;
2463 			}
2464 
2465 			if (sbi->segs_per_sec > 1)
2466 				get_sec_entry(sbi, start)->valid_blocks +=
2467 							se->valid_blocks;
2468 		}
2469 		start_blk += readed;
2470 	} while (start_blk < sit_blk_cnt);
2471 
2472 	down_read(&curseg->journal_rwsem);
2473 	for (i = 0; i < sits_in_cursum(journal); i++) {
2474 		unsigned int old_valid_blocks;
2475 
2476 		start = le32_to_cpu(segno_in_journal(journal, i));
2477 		se = &sit_i->sentries[start];
2478 		sit = sit_in_journal(journal, i);
2479 
2480 		old_valid_blocks = se->valid_blocks;
2481 
2482 		check_block_count(sbi, start, &sit);
2483 		seg_info_from_raw_sit(se, &sit);
2484 
2485 		if (f2fs_discard_en(sbi)) {
2486 			memcpy(se->discard_map, se->cur_valid_map,
2487 						SIT_VBLOCK_MAP_SIZE);
2488 			sbi->discard_blks += old_valid_blocks -
2489 						se->valid_blocks;
2490 		}
2491 
2492 		if (sbi->segs_per_sec > 1)
2493 			get_sec_entry(sbi, start)->valid_blocks +=
2494 				se->valid_blocks - old_valid_blocks;
2495 	}
2496 	up_read(&curseg->journal_rwsem);
2497 }
2498 
2499 static void init_free_segmap(struct f2fs_sb_info *sbi)
2500 {
2501 	unsigned int start;
2502 	int type;
2503 
2504 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
2505 		struct seg_entry *sentry = get_seg_entry(sbi, start);
2506 		if (!sentry->valid_blocks)
2507 			__set_free(sbi, start);
2508 		else
2509 			SIT_I(sbi)->written_valid_blocks +=
2510 						sentry->valid_blocks;
2511 	}
2512 
2513 	/* set use the current segments */
2514 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2515 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2516 		__set_test_and_inuse(sbi, curseg_t->segno);
2517 	}
2518 }
2519 
2520 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2521 {
2522 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2523 	struct free_segmap_info *free_i = FREE_I(sbi);
2524 	unsigned int segno = 0, offset = 0;
2525 	unsigned short valid_blocks;
2526 
2527 	while (1) {
2528 		/* find dirty segment based on free segmap */
2529 		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2530 		if (segno >= MAIN_SEGS(sbi))
2531 			break;
2532 		offset = segno + 1;
2533 		valid_blocks = get_valid_blocks(sbi, segno, 0);
2534 		if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2535 			continue;
2536 		if (valid_blocks > sbi->blocks_per_seg) {
2537 			f2fs_bug_on(sbi, 1);
2538 			continue;
2539 		}
2540 		mutex_lock(&dirty_i->seglist_lock);
2541 		__locate_dirty_segment(sbi, segno, DIRTY);
2542 		mutex_unlock(&dirty_i->seglist_lock);
2543 	}
2544 }
2545 
2546 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2547 {
2548 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2549 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2550 
2551 	dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2552 	if (!dirty_i->victim_secmap)
2553 		return -ENOMEM;
2554 	return 0;
2555 }
2556 
2557 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2558 {
2559 	struct dirty_seglist_info *dirty_i;
2560 	unsigned int bitmap_size, i;
2561 
2562 	/* allocate memory for dirty segments list information */
2563 	dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2564 	if (!dirty_i)
2565 		return -ENOMEM;
2566 
2567 	SM_I(sbi)->dirty_info = dirty_i;
2568 	mutex_init(&dirty_i->seglist_lock);
2569 
2570 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2571 
2572 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
2573 		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2574 		if (!dirty_i->dirty_segmap[i])
2575 			return -ENOMEM;
2576 	}
2577 
2578 	init_dirty_segmap(sbi);
2579 	return init_victim_secmap(sbi);
2580 }
2581 
2582 /*
2583  * Update min, max modified time for cost-benefit GC algorithm
2584  */
2585 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2586 {
2587 	struct sit_info *sit_i = SIT_I(sbi);
2588 	unsigned int segno;
2589 
2590 	mutex_lock(&sit_i->sentry_lock);
2591 
2592 	sit_i->min_mtime = LLONG_MAX;
2593 
2594 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2595 		unsigned int i;
2596 		unsigned long long mtime = 0;
2597 
2598 		for (i = 0; i < sbi->segs_per_sec; i++)
2599 			mtime += get_seg_entry(sbi, segno + i)->mtime;
2600 
2601 		mtime = div_u64(mtime, sbi->segs_per_sec);
2602 
2603 		if (sit_i->min_mtime > mtime)
2604 			sit_i->min_mtime = mtime;
2605 	}
2606 	sit_i->max_mtime = get_mtime(sbi);
2607 	mutex_unlock(&sit_i->sentry_lock);
2608 }
2609 
2610 int build_segment_manager(struct f2fs_sb_info *sbi)
2611 {
2612 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2613 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2614 	struct f2fs_sm_info *sm_info;
2615 	int err;
2616 
2617 	sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2618 	if (!sm_info)
2619 		return -ENOMEM;
2620 
2621 	/* init sm info */
2622 	sbi->sm_info = sm_info;
2623 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2624 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2625 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2626 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2627 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2628 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2629 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2630 	sm_info->rec_prefree_segments = sm_info->main_segments *
2631 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2632 	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2633 		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2634 
2635 	if (!test_opt(sbi, LFS))
2636 		sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2637 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2638 	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2639 
2640 	INIT_LIST_HEAD(&sm_info->discard_list);
2641 	INIT_LIST_HEAD(&sm_info->wait_list);
2642 	sm_info->nr_discards = 0;
2643 	sm_info->max_discards = 0;
2644 
2645 	sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2646 
2647 	INIT_LIST_HEAD(&sm_info->sit_entry_set);
2648 
2649 	if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2650 		err = create_flush_cmd_control(sbi);
2651 		if (err)
2652 			return err;
2653 	}
2654 
2655 	err = build_sit_info(sbi);
2656 	if (err)
2657 		return err;
2658 	err = build_free_segmap(sbi);
2659 	if (err)
2660 		return err;
2661 	err = build_curseg(sbi);
2662 	if (err)
2663 		return err;
2664 
2665 	/* reinit free segmap based on SIT */
2666 	build_sit_entries(sbi);
2667 
2668 	init_free_segmap(sbi);
2669 	err = build_dirty_segmap(sbi);
2670 	if (err)
2671 		return err;
2672 
2673 	init_min_max_mtime(sbi);
2674 	return 0;
2675 }
2676 
2677 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2678 		enum dirty_type dirty_type)
2679 {
2680 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2681 
2682 	mutex_lock(&dirty_i->seglist_lock);
2683 	kvfree(dirty_i->dirty_segmap[dirty_type]);
2684 	dirty_i->nr_dirty[dirty_type] = 0;
2685 	mutex_unlock(&dirty_i->seglist_lock);
2686 }
2687 
2688 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2689 {
2690 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2691 	kvfree(dirty_i->victim_secmap);
2692 }
2693 
2694 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2695 {
2696 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2697 	int i;
2698 
2699 	if (!dirty_i)
2700 		return;
2701 
2702 	/* discard pre-free/dirty segments list */
2703 	for (i = 0; i < NR_DIRTY_TYPE; i++)
2704 		discard_dirty_segmap(sbi, i);
2705 
2706 	destroy_victim_secmap(sbi);
2707 	SM_I(sbi)->dirty_info = NULL;
2708 	kfree(dirty_i);
2709 }
2710 
2711 static void destroy_curseg(struct f2fs_sb_info *sbi)
2712 {
2713 	struct curseg_info *array = SM_I(sbi)->curseg_array;
2714 	int i;
2715 
2716 	if (!array)
2717 		return;
2718 	SM_I(sbi)->curseg_array = NULL;
2719 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
2720 		kfree(array[i].sum_blk);
2721 		kfree(array[i].journal);
2722 	}
2723 	kfree(array);
2724 }
2725 
2726 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2727 {
2728 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2729 	if (!free_i)
2730 		return;
2731 	SM_I(sbi)->free_info = NULL;
2732 	kvfree(free_i->free_segmap);
2733 	kvfree(free_i->free_secmap);
2734 	kfree(free_i);
2735 }
2736 
2737 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2738 {
2739 	struct sit_info *sit_i = SIT_I(sbi);
2740 	unsigned int start;
2741 
2742 	if (!sit_i)
2743 		return;
2744 
2745 	if (sit_i->sentries) {
2746 		for (start = 0; start < MAIN_SEGS(sbi); start++) {
2747 			kfree(sit_i->sentries[start].cur_valid_map);
2748 			kfree(sit_i->sentries[start].ckpt_valid_map);
2749 			kfree(sit_i->sentries[start].discard_map);
2750 		}
2751 	}
2752 	kfree(sit_i->tmp_map);
2753 
2754 	kvfree(sit_i->sentries);
2755 	kvfree(sit_i->sec_entries);
2756 	kvfree(sit_i->dirty_sentries_bitmap);
2757 
2758 	SM_I(sbi)->sit_info = NULL;
2759 	kfree(sit_i->sit_bitmap);
2760 	kfree(sit_i);
2761 }
2762 
2763 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2764 {
2765 	struct f2fs_sm_info *sm_info = SM_I(sbi);
2766 
2767 	if (!sm_info)
2768 		return;
2769 	destroy_flush_cmd_control(sbi, true);
2770 	destroy_dirty_segmap(sbi);
2771 	destroy_curseg(sbi);
2772 	destroy_free_segmap(sbi);
2773 	destroy_sit_info(sbi);
2774 	sbi->sm_info = NULL;
2775 	kfree(sm_info);
2776 }
2777 
2778 int __init create_segment_manager_caches(void)
2779 {
2780 	discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2781 			sizeof(struct discard_entry));
2782 	if (!discard_entry_slab)
2783 		goto fail;
2784 
2785 	bio_entry_slab = f2fs_kmem_cache_create("bio_entry",
2786 			sizeof(struct bio_entry));
2787 	if (!bio_entry_slab)
2788 		goto destroy_discard_entry;
2789 
2790 	sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2791 			sizeof(struct sit_entry_set));
2792 	if (!sit_entry_set_slab)
2793 		goto destroy_bio_entry;
2794 
2795 	inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2796 			sizeof(struct inmem_pages));
2797 	if (!inmem_entry_slab)
2798 		goto destroy_sit_entry_set;
2799 	return 0;
2800 
2801 destroy_sit_entry_set:
2802 	kmem_cache_destroy(sit_entry_set_slab);
2803 destroy_bio_entry:
2804 	kmem_cache_destroy(bio_entry_slab);
2805 destroy_discard_entry:
2806 	kmem_cache_destroy(discard_entry_slab);
2807 fail:
2808 	return -ENOMEM;
2809 }
2810 
2811 void destroy_segment_manager_caches(void)
2812 {
2813 	kmem_cache_destroy(sit_entry_set_slab);
2814 	kmem_cache_destroy(bio_entry_slab);
2815 	kmem_cache_destroy(discard_entry_slab);
2816 	kmem_cache_destroy(inmem_entry_slab);
2817 }
2818