xref: /openbmc/linux/fs/f2fs/segment.c (revision aeb64ff3)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * fs/f2fs/segment.c
4  *
5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6  *             http://www.samsung.com/
7  */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/bio.h>
11 #include <linux/blkdev.h>
12 #include <linux/prefetch.h>
13 #include <linux/kthread.h>
14 #include <linux/swap.h>
15 #include <linux/timer.h>
16 #include <linux/freezer.h>
17 #include <linux/sched/signal.h>
18 
19 #include "f2fs.h"
20 #include "segment.h"
21 #include "node.h"
22 #include "gc.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 *discard_cmd_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 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
170 {
171 	int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
172 	int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
173 	int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
174 
175 	if (test_opt(sbi, LFS))
176 		return false;
177 	if (sbi->gc_mode == GC_URGENT)
178 		return true;
179 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
180 		return true;
181 
182 	return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
183 			SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
184 }
185 
186 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
187 {
188 	struct inmem_pages *new;
189 
190 	f2fs_trace_pid(page);
191 
192 	f2fs_set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
193 
194 	new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
195 
196 	/* add atomic page indices to the list */
197 	new->page = page;
198 	INIT_LIST_HEAD(&new->list);
199 
200 	/* increase reference count with clean state */
201 	get_page(page);
202 	mutex_lock(&F2FS_I(inode)->inmem_lock);
203 	list_add_tail(&new->list, &F2FS_I(inode)->inmem_pages);
204 	inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
205 	mutex_unlock(&F2FS_I(inode)->inmem_lock);
206 
207 	trace_f2fs_register_inmem_page(page, INMEM);
208 }
209 
210 static int __revoke_inmem_pages(struct inode *inode,
211 				struct list_head *head, bool drop, bool recover,
212 				bool trylock)
213 {
214 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
215 	struct inmem_pages *cur, *tmp;
216 	int err = 0;
217 
218 	list_for_each_entry_safe(cur, tmp, head, list) {
219 		struct page *page = cur->page;
220 
221 		if (drop)
222 			trace_f2fs_commit_inmem_page(page, INMEM_DROP);
223 
224 		if (trylock) {
225 			/*
226 			 * to avoid deadlock in between page lock and
227 			 * inmem_lock.
228 			 */
229 			if (!trylock_page(page))
230 				continue;
231 		} else {
232 			lock_page(page);
233 		}
234 
235 		f2fs_wait_on_page_writeback(page, DATA, true, true);
236 
237 		if (recover) {
238 			struct dnode_of_data dn;
239 			struct node_info ni;
240 
241 			trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
242 retry:
243 			set_new_dnode(&dn, inode, NULL, NULL, 0);
244 			err = f2fs_get_dnode_of_data(&dn, page->index,
245 								LOOKUP_NODE);
246 			if (err) {
247 				if (err == -ENOMEM) {
248 					congestion_wait(BLK_RW_ASYNC, HZ/50);
249 					cond_resched();
250 					goto retry;
251 				}
252 				err = -EAGAIN;
253 				goto next;
254 			}
255 
256 			err = f2fs_get_node_info(sbi, dn.nid, &ni);
257 			if (err) {
258 				f2fs_put_dnode(&dn);
259 				return err;
260 			}
261 
262 			if (cur->old_addr == NEW_ADDR) {
263 				f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
264 				f2fs_update_data_blkaddr(&dn, NEW_ADDR);
265 			} else
266 				f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
267 					cur->old_addr, ni.version, true, true);
268 			f2fs_put_dnode(&dn);
269 		}
270 next:
271 		/* we don't need to invalidate this in the sccessful status */
272 		if (drop || recover) {
273 			ClearPageUptodate(page);
274 			clear_cold_data(page);
275 		}
276 		f2fs_clear_page_private(page);
277 		f2fs_put_page(page, 1);
278 
279 		list_del(&cur->list);
280 		kmem_cache_free(inmem_entry_slab, cur);
281 		dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
282 	}
283 	return err;
284 }
285 
286 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
287 {
288 	struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
289 	struct inode *inode;
290 	struct f2fs_inode_info *fi;
291 	unsigned int count = sbi->atomic_files;
292 	unsigned int looped = 0;
293 next:
294 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
295 	if (list_empty(head)) {
296 		spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
297 		return;
298 	}
299 	fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
300 	inode = igrab(&fi->vfs_inode);
301 	if (inode)
302 		list_move_tail(&fi->inmem_ilist, head);
303 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
304 
305 	if (inode) {
306 		if (gc_failure) {
307 			if (!fi->i_gc_failures[GC_FAILURE_ATOMIC])
308 				goto skip;
309 		}
310 		set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
311 		f2fs_drop_inmem_pages(inode);
312 skip:
313 		iput(inode);
314 	}
315 	congestion_wait(BLK_RW_ASYNC, HZ/50);
316 	cond_resched();
317 	if (gc_failure) {
318 		if (++looped >= count)
319 			return;
320 	}
321 	goto next;
322 }
323 
324 void f2fs_drop_inmem_pages(struct inode *inode)
325 {
326 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
327 	struct f2fs_inode_info *fi = F2FS_I(inode);
328 
329 	while (!list_empty(&fi->inmem_pages)) {
330 		mutex_lock(&fi->inmem_lock);
331 		__revoke_inmem_pages(inode, &fi->inmem_pages,
332 						true, false, true);
333 		mutex_unlock(&fi->inmem_lock);
334 	}
335 
336 	fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
337 	stat_dec_atomic_write(inode);
338 
339 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
340 	if (!list_empty(&fi->inmem_ilist))
341 		list_del_init(&fi->inmem_ilist);
342 	if (f2fs_is_atomic_file(inode)) {
343 		clear_inode_flag(inode, FI_ATOMIC_FILE);
344 		sbi->atomic_files--;
345 	}
346 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
347 }
348 
349 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
350 {
351 	struct f2fs_inode_info *fi = F2FS_I(inode);
352 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
353 	struct list_head *head = &fi->inmem_pages;
354 	struct inmem_pages *cur = NULL;
355 
356 	f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
357 
358 	mutex_lock(&fi->inmem_lock);
359 	list_for_each_entry(cur, head, list) {
360 		if (cur->page == page)
361 			break;
362 	}
363 
364 	f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
365 	list_del(&cur->list);
366 	mutex_unlock(&fi->inmem_lock);
367 
368 	dec_page_count(sbi, F2FS_INMEM_PAGES);
369 	kmem_cache_free(inmem_entry_slab, cur);
370 
371 	ClearPageUptodate(page);
372 	f2fs_clear_page_private(page);
373 	f2fs_put_page(page, 0);
374 
375 	trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
376 }
377 
378 static int __f2fs_commit_inmem_pages(struct inode *inode)
379 {
380 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
381 	struct f2fs_inode_info *fi = F2FS_I(inode);
382 	struct inmem_pages *cur, *tmp;
383 	struct f2fs_io_info fio = {
384 		.sbi = sbi,
385 		.ino = inode->i_ino,
386 		.type = DATA,
387 		.op = REQ_OP_WRITE,
388 		.op_flags = REQ_SYNC | REQ_PRIO,
389 		.io_type = FS_DATA_IO,
390 	};
391 	struct list_head revoke_list;
392 	bool submit_bio = false;
393 	int err = 0;
394 
395 	INIT_LIST_HEAD(&revoke_list);
396 
397 	list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
398 		struct page *page = cur->page;
399 
400 		lock_page(page);
401 		if (page->mapping == inode->i_mapping) {
402 			trace_f2fs_commit_inmem_page(page, INMEM);
403 
404 			f2fs_wait_on_page_writeback(page, DATA, true, true);
405 
406 			set_page_dirty(page);
407 			if (clear_page_dirty_for_io(page)) {
408 				inode_dec_dirty_pages(inode);
409 				f2fs_remove_dirty_inode(inode);
410 			}
411 retry:
412 			fio.page = page;
413 			fio.old_blkaddr = NULL_ADDR;
414 			fio.encrypted_page = NULL;
415 			fio.need_lock = LOCK_DONE;
416 			err = f2fs_do_write_data_page(&fio);
417 			if (err) {
418 				if (err == -ENOMEM) {
419 					congestion_wait(BLK_RW_ASYNC, HZ/50);
420 					cond_resched();
421 					goto retry;
422 				}
423 				unlock_page(page);
424 				break;
425 			}
426 			/* record old blkaddr for revoking */
427 			cur->old_addr = fio.old_blkaddr;
428 			submit_bio = true;
429 		}
430 		unlock_page(page);
431 		list_move_tail(&cur->list, &revoke_list);
432 	}
433 
434 	if (submit_bio)
435 		f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
436 
437 	if (err) {
438 		/*
439 		 * try to revoke all committed pages, but still we could fail
440 		 * due to no memory or other reason, if that happened, EAGAIN
441 		 * will be returned, which means in such case, transaction is
442 		 * already not integrity, caller should use journal to do the
443 		 * recovery or rewrite & commit last transaction. For other
444 		 * error number, revoking was done by filesystem itself.
445 		 */
446 		err = __revoke_inmem_pages(inode, &revoke_list,
447 						false, true, false);
448 
449 		/* drop all uncommitted pages */
450 		__revoke_inmem_pages(inode, &fi->inmem_pages,
451 						true, false, false);
452 	} else {
453 		__revoke_inmem_pages(inode, &revoke_list,
454 						false, false, false);
455 	}
456 
457 	return err;
458 }
459 
460 int f2fs_commit_inmem_pages(struct inode *inode)
461 {
462 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
463 	struct f2fs_inode_info *fi = F2FS_I(inode);
464 	int err;
465 
466 	f2fs_balance_fs(sbi, true);
467 
468 	down_write(&fi->i_gc_rwsem[WRITE]);
469 
470 	f2fs_lock_op(sbi);
471 	set_inode_flag(inode, FI_ATOMIC_COMMIT);
472 
473 	mutex_lock(&fi->inmem_lock);
474 	err = __f2fs_commit_inmem_pages(inode);
475 	mutex_unlock(&fi->inmem_lock);
476 
477 	clear_inode_flag(inode, FI_ATOMIC_COMMIT);
478 
479 	f2fs_unlock_op(sbi);
480 	up_write(&fi->i_gc_rwsem[WRITE]);
481 
482 	return err;
483 }
484 
485 /*
486  * This function balances dirty node and dentry pages.
487  * In addition, it controls garbage collection.
488  */
489 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
490 {
491 	if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
492 		f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
493 		f2fs_stop_checkpoint(sbi, false);
494 	}
495 
496 	/* balance_fs_bg is able to be pending */
497 	if (need && excess_cached_nats(sbi))
498 		f2fs_balance_fs_bg(sbi);
499 
500 	if (!f2fs_is_checkpoint_ready(sbi))
501 		return;
502 
503 	/*
504 	 * We should do GC or end up with checkpoint, if there are so many dirty
505 	 * dir/node pages without enough free segments.
506 	 */
507 	if (has_not_enough_free_secs(sbi, 0, 0)) {
508 		mutex_lock(&sbi->gc_mutex);
509 		f2fs_gc(sbi, false, false, NULL_SEGNO);
510 	}
511 }
512 
513 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
514 {
515 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
516 		return;
517 
518 	/* try to shrink extent cache when there is no enough memory */
519 	if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
520 		f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
521 
522 	/* check the # of cached NAT entries */
523 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
524 		f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
525 
526 	if (!f2fs_available_free_memory(sbi, FREE_NIDS))
527 		f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
528 	else
529 		f2fs_build_free_nids(sbi, false, false);
530 
531 	if (!is_idle(sbi, REQ_TIME) &&
532 		(!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
533 		return;
534 
535 	/* checkpoint is the only way to shrink partial cached entries */
536 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
537 			!f2fs_available_free_memory(sbi, INO_ENTRIES) ||
538 			excess_prefree_segs(sbi) ||
539 			excess_dirty_nats(sbi) ||
540 			excess_dirty_nodes(sbi) ||
541 			f2fs_time_over(sbi, CP_TIME)) {
542 		if (test_opt(sbi, DATA_FLUSH)) {
543 			struct blk_plug plug;
544 
545 			mutex_lock(&sbi->flush_lock);
546 
547 			blk_start_plug(&plug);
548 			f2fs_sync_dirty_inodes(sbi, FILE_INODE);
549 			blk_finish_plug(&plug);
550 
551 			mutex_unlock(&sbi->flush_lock);
552 		}
553 		f2fs_sync_fs(sbi->sb, true);
554 		stat_inc_bg_cp_count(sbi->stat_info);
555 	}
556 }
557 
558 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
559 				struct block_device *bdev)
560 {
561 	struct bio *bio;
562 	int ret;
563 
564 	bio = f2fs_bio_alloc(sbi, 0, false);
565 	if (!bio)
566 		return -ENOMEM;
567 
568 	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
569 	bio_set_dev(bio, bdev);
570 	ret = submit_bio_wait(bio);
571 	bio_put(bio);
572 
573 	trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
574 				test_opt(sbi, FLUSH_MERGE), ret);
575 	return ret;
576 }
577 
578 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
579 {
580 	int ret = 0;
581 	int i;
582 
583 	if (!f2fs_is_multi_device(sbi))
584 		return __submit_flush_wait(sbi, sbi->sb->s_bdev);
585 
586 	for (i = 0; i < sbi->s_ndevs; i++) {
587 		if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
588 			continue;
589 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
590 		if (ret)
591 			break;
592 	}
593 	return ret;
594 }
595 
596 static int issue_flush_thread(void *data)
597 {
598 	struct f2fs_sb_info *sbi = data;
599 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
600 	wait_queue_head_t *q = &fcc->flush_wait_queue;
601 repeat:
602 	if (kthread_should_stop())
603 		return 0;
604 
605 	sb_start_intwrite(sbi->sb);
606 
607 	if (!llist_empty(&fcc->issue_list)) {
608 		struct flush_cmd *cmd, *next;
609 		int ret;
610 
611 		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
612 		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
613 
614 		cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
615 
616 		ret = submit_flush_wait(sbi, cmd->ino);
617 		atomic_inc(&fcc->issued_flush);
618 
619 		llist_for_each_entry_safe(cmd, next,
620 					  fcc->dispatch_list, llnode) {
621 			cmd->ret = ret;
622 			complete(&cmd->wait);
623 		}
624 		fcc->dispatch_list = NULL;
625 	}
626 
627 	sb_end_intwrite(sbi->sb);
628 
629 	wait_event_interruptible(*q,
630 		kthread_should_stop() || !llist_empty(&fcc->issue_list));
631 	goto repeat;
632 }
633 
634 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
635 {
636 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
637 	struct flush_cmd cmd;
638 	int ret;
639 
640 	if (test_opt(sbi, NOBARRIER))
641 		return 0;
642 
643 	if (!test_opt(sbi, FLUSH_MERGE)) {
644 		atomic_inc(&fcc->queued_flush);
645 		ret = submit_flush_wait(sbi, ino);
646 		atomic_dec(&fcc->queued_flush);
647 		atomic_inc(&fcc->issued_flush);
648 		return ret;
649 	}
650 
651 	if (atomic_inc_return(&fcc->queued_flush) == 1 ||
652 	    f2fs_is_multi_device(sbi)) {
653 		ret = submit_flush_wait(sbi, ino);
654 		atomic_dec(&fcc->queued_flush);
655 
656 		atomic_inc(&fcc->issued_flush);
657 		return ret;
658 	}
659 
660 	cmd.ino = ino;
661 	init_completion(&cmd.wait);
662 
663 	llist_add(&cmd.llnode, &fcc->issue_list);
664 
665 	/* update issue_list before we wake up issue_flush thread */
666 	smp_mb();
667 
668 	if (waitqueue_active(&fcc->flush_wait_queue))
669 		wake_up(&fcc->flush_wait_queue);
670 
671 	if (fcc->f2fs_issue_flush) {
672 		wait_for_completion(&cmd.wait);
673 		atomic_dec(&fcc->queued_flush);
674 	} else {
675 		struct llist_node *list;
676 
677 		list = llist_del_all(&fcc->issue_list);
678 		if (!list) {
679 			wait_for_completion(&cmd.wait);
680 			atomic_dec(&fcc->queued_flush);
681 		} else {
682 			struct flush_cmd *tmp, *next;
683 
684 			ret = submit_flush_wait(sbi, ino);
685 
686 			llist_for_each_entry_safe(tmp, next, list, llnode) {
687 				if (tmp == &cmd) {
688 					cmd.ret = ret;
689 					atomic_dec(&fcc->queued_flush);
690 					continue;
691 				}
692 				tmp->ret = ret;
693 				complete(&tmp->wait);
694 			}
695 		}
696 	}
697 
698 	return cmd.ret;
699 }
700 
701 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
702 {
703 	dev_t dev = sbi->sb->s_bdev->bd_dev;
704 	struct flush_cmd_control *fcc;
705 	int err = 0;
706 
707 	if (SM_I(sbi)->fcc_info) {
708 		fcc = SM_I(sbi)->fcc_info;
709 		if (fcc->f2fs_issue_flush)
710 			return err;
711 		goto init_thread;
712 	}
713 
714 	fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
715 	if (!fcc)
716 		return -ENOMEM;
717 	atomic_set(&fcc->issued_flush, 0);
718 	atomic_set(&fcc->queued_flush, 0);
719 	init_waitqueue_head(&fcc->flush_wait_queue);
720 	init_llist_head(&fcc->issue_list);
721 	SM_I(sbi)->fcc_info = fcc;
722 	if (!test_opt(sbi, FLUSH_MERGE))
723 		return err;
724 
725 init_thread:
726 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
727 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
728 	if (IS_ERR(fcc->f2fs_issue_flush)) {
729 		err = PTR_ERR(fcc->f2fs_issue_flush);
730 		kvfree(fcc);
731 		SM_I(sbi)->fcc_info = NULL;
732 		return err;
733 	}
734 
735 	return err;
736 }
737 
738 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
739 {
740 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
741 
742 	if (fcc && fcc->f2fs_issue_flush) {
743 		struct task_struct *flush_thread = fcc->f2fs_issue_flush;
744 
745 		fcc->f2fs_issue_flush = NULL;
746 		kthread_stop(flush_thread);
747 	}
748 	if (free) {
749 		kvfree(fcc);
750 		SM_I(sbi)->fcc_info = NULL;
751 	}
752 }
753 
754 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
755 {
756 	int ret = 0, i;
757 
758 	if (!f2fs_is_multi_device(sbi))
759 		return 0;
760 
761 	for (i = 1; i < sbi->s_ndevs; i++) {
762 		if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
763 			continue;
764 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
765 		if (ret)
766 			break;
767 
768 		spin_lock(&sbi->dev_lock);
769 		f2fs_clear_bit(i, (char *)&sbi->dirty_device);
770 		spin_unlock(&sbi->dev_lock);
771 	}
772 
773 	return ret;
774 }
775 
776 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
777 		enum dirty_type dirty_type)
778 {
779 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
780 
781 	/* need not be added */
782 	if (IS_CURSEG(sbi, segno))
783 		return;
784 
785 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
786 		dirty_i->nr_dirty[dirty_type]++;
787 
788 	if (dirty_type == DIRTY) {
789 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
790 		enum dirty_type t = sentry->type;
791 
792 		if (unlikely(t >= DIRTY)) {
793 			f2fs_bug_on(sbi, 1);
794 			return;
795 		}
796 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
797 			dirty_i->nr_dirty[t]++;
798 	}
799 }
800 
801 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
802 		enum dirty_type dirty_type)
803 {
804 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
805 
806 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
807 		dirty_i->nr_dirty[dirty_type]--;
808 
809 	if (dirty_type == DIRTY) {
810 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
811 		enum dirty_type t = sentry->type;
812 
813 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
814 			dirty_i->nr_dirty[t]--;
815 
816 		if (get_valid_blocks(sbi, segno, true) == 0) {
817 			clear_bit(GET_SEC_FROM_SEG(sbi, segno),
818 						dirty_i->victim_secmap);
819 #ifdef CONFIG_F2FS_CHECK_FS
820 			clear_bit(segno, SIT_I(sbi)->invalid_segmap);
821 #endif
822 		}
823 	}
824 }
825 
826 /*
827  * Should not occur error such as -ENOMEM.
828  * Adding dirty entry into seglist is not critical operation.
829  * If a given segment is one of current working segments, it won't be added.
830  */
831 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
832 {
833 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
834 	unsigned short valid_blocks, ckpt_valid_blocks;
835 
836 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
837 		return;
838 
839 	mutex_lock(&dirty_i->seglist_lock);
840 
841 	valid_blocks = get_valid_blocks(sbi, segno, false);
842 	ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
843 
844 	if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
845 				ckpt_valid_blocks == sbi->blocks_per_seg)) {
846 		__locate_dirty_segment(sbi, segno, PRE);
847 		__remove_dirty_segment(sbi, segno, DIRTY);
848 	} else if (valid_blocks < sbi->blocks_per_seg) {
849 		__locate_dirty_segment(sbi, segno, DIRTY);
850 	} else {
851 		/* Recovery routine with SSR needs this */
852 		__remove_dirty_segment(sbi, segno, DIRTY);
853 	}
854 
855 	mutex_unlock(&dirty_i->seglist_lock);
856 }
857 
858 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
859 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
860 {
861 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
862 	unsigned int segno;
863 
864 	mutex_lock(&dirty_i->seglist_lock);
865 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
866 		if (get_valid_blocks(sbi, segno, false))
867 			continue;
868 		if (IS_CURSEG(sbi, segno))
869 			continue;
870 		__locate_dirty_segment(sbi, segno, PRE);
871 		__remove_dirty_segment(sbi, segno, DIRTY);
872 	}
873 	mutex_unlock(&dirty_i->seglist_lock);
874 }
875 
876 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
877 {
878 	int ovp_hole_segs =
879 		(overprovision_segments(sbi) - reserved_segments(sbi));
880 	block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
881 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
882 	block_t holes[2] = {0, 0};	/* DATA and NODE */
883 	block_t unusable;
884 	struct seg_entry *se;
885 	unsigned int segno;
886 
887 	mutex_lock(&dirty_i->seglist_lock);
888 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
889 		se = get_seg_entry(sbi, segno);
890 		if (IS_NODESEG(se->type))
891 			holes[NODE] += sbi->blocks_per_seg - se->valid_blocks;
892 		else
893 			holes[DATA] += sbi->blocks_per_seg - se->valid_blocks;
894 	}
895 	mutex_unlock(&dirty_i->seglist_lock);
896 
897 	unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
898 	if (unusable > ovp_holes)
899 		return unusable - ovp_holes;
900 	return 0;
901 }
902 
903 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
904 {
905 	int ovp_hole_segs =
906 		(overprovision_segments(sbi) - reserved_segments(sbi));
907 	if (unusable > F2FS_OPTION(sbi).unusable_cap)
908 		return -EAGAIN;
909 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
910 		dirty_segments(sbi) > ovp_hole_segs)
911 		return -EAGAIN;
912 	return 0;
913 }
914 
915 /* This is only used by SBI_CP_DISABLED */
916 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
917 {
918 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
919 	unsigned int segno = 0;
920 
921 	mutex_lock(&dirty_i->seglist_lock);
922 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
923 		if (get_valid_blocks(sbi, segno, false))
924 			continue;
925 		if (get_ckpt_valid_blocks(sbi, segno))
926 			continue;
927 		mutex_unlock(&dirty_i->seglist_lock);
928 		return segno;
929 	}
930 	mutex_unlock(&dirty_i->seglist_lock);
931 	return NULL_SEGNO;
932 }
933 
934 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
935 		struct block_device *bdev, block_t lstart,
936 		block_t start, block_t len)
937 {
938 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
939 	struct list_head *pend_list;
940 	struct discard_cmd *dc;
941 
942 	f2fs_bug_on(sbi, !len);
943 
944 	pend_list = &dcc->pend_list[plist_idx(len)];
945 
946 	dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
947 	INIT_LIST_HEAD(&dc->list);
948 	dc->bdev = bdev;
949 	dc->lstart = lstart;
950 	dc->start = start;
951 	dc->len = len;
952 	dc->ref = 0;
953 	dc->state = D_PREP;
954 	dc->queued = 0;
955 	dc->error = 0;
956 	init_completion(&dc->wait);
957 	list_add_tail(&dc->list, pend_list);
958 	spin_lock_init(&dc->lock);
959 	dc->bio_ref = 0;
960 	atomic_inc(&dcc->discard_cmd_cnt);
961 	dcc->undiscard_blks += len;
962 
963 	return dc;
964 }
965 
966 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
967 				struct block_device *bdev, block_t lstart,
968 				block_t start, block_t len,
969 				struct rb_node *parent, struct rb_node **p,
970 				bool leftmost)
971 {
972 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
973 	struct discard_cmd *dc;
974 
975 	dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
976 
977 	rb_link_node(&dc->rb_node, parent, p);
978 	rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
979 
980 	return dc;
981 }
982 
983 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
984 							struct discard_cmd *dc)
985 {
986 	if (dc->state == D_DONE)
987 		atomic_sub(dc->queued, &dcc->queued_discard);
988 
989 	list_del(&dc->list);
990 	rb_erase_cached(&dc->rb_node, &dcc->root);
991 	dcc->undiscard_blks -= dc->len;
992 
993 	kmem_cache_free(discard_cmd_slab, dc);
994 
995 	atomic_dec(&dcc->discard_cmd_cnt);
996 }
997 
998 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
999 							struct discard_cmd *dc)
1000 {
1001 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1002 	unsigned long flags;
1003 
1004 	trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
1005 
1006 	spin_lock_irqsave(&dc->lock, flags);
1007 	if (dc->bio_ref) {
1008 		spin_unlock_irqrestore(&dc->lock, flags);
1009 		return;
1010 	}
1011 	spin_unlock_irqrestore(&dc->lock, flags);
1012 
1013 	f2fs_bug_on(sbi, dc->ref);
1014 
1015 	if (dc->error == -EOPNOTSUPP)
1016 		dc->error = 0;
1017 
1018 	if (dc->error)
1019 		printk_ratelimited(
1020 			"%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1021 			KERN_INFO, sbi->sb->s_id,
1022 			dc->lstart, dc->start, dc->len, dc->error);
1023 	__detach_discard_cmd(dcc, dc);
1024 }
1025 
1026 static void f2fs_submit_discard_endio(struct bio *bio)
1027 {
1028 	struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1029 	unsigned long flags;
1030 
1031 	dc->error = blk_status_to_errno(bio->bi_status);
1032 
1033 	spin_lock_irqsave(&dc->lock, flags);
1034 	dc->bio_ref--;
1035 	if (!dc->bio_ref && dc->state == D_SUBMIT) {
1036 		dc->state = D_DONE;
1037 		complete_all(&dc->wait);
1038 	}
1039 	spin_unlock_irqrestore(&dc->lock, flags);
1040 	bio_put(bio);
1041 }
1042 
1043 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1044 				block_t start, block_t end)
1045 {
1046 #ifdef CONFIG_F2FS_CHECK_FS
1047 	struct seg_entry *sentry;
1048 	unsigned int segno;
1049 	block_t blk = start;
1050 	unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1051 	unsigned long *map;
1052 
1053 	while (blk < end) {
1054 		segno = GET_SEGNO(sbi, blk);
1055 		sentry = get_seg_entry(sbi, segno);
1056 		offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1057 
1058 		if (end < START_BLOCK(sbi, segno + 1))
1059 			size = GET_BLKOFF_FROM_SEG0(sbi, end);
1060 		else
1061 			size = max_blocks;
1062 		map = (unsigned long *)(sentry->cur_valid_map);
1063 		offset = __find_rev_next_bit(map, size, offset);
1064 		f2fs_bug_on(sbi, offset != size);
1065 		blk = START_BLOCK(sbi, segno + 1);
1066 	}
1067 #endif
1068 }
1069 
1070 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1071 				struct discard_policy *dpolicy,
1072 				int discard_type, unsigned int granularity)
1073 {
1074 	/* common policy */
1075 	dpolicy->type = discard_type;
1076 	dpolicy->sync = true;
1077 	dpolicy->ordered = false;
1078 	dpolicy->granularity = granularity;
1079 
1080 	dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1081 	dpolicy->io_aware_gran = MAX_PLIST_NUM;
1082 	dpolicy->timeout = 0;
1083 
1084 	if (discard_type == DPOLICY_BG) {
1085 		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1086 		dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1087 		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1088 		dpolicy->io_aware = true;
1089 		dpolicy->sync = false;
1090 		dpolicy->ordered = true;
1091 		if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1092 			dpolicy->granularity = 1;
1093 			dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1094 		}
1095 	} else if (discard_type == DPOLICY_FORCE) {
1096 		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1097 		dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1098 		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1099 		dpolicy->io_aware = false;
1100 	} else if (discard_type == DPOLICY_FSTRIM) {
1101 		dpolicy->io_aware = false;
1102 	} else if (discard_type == DPOLICY_UMOUNT) {
1103 		dpolicy->max_requests = UINT_MAX;
1104 		dpolicy->io_aware = false;
1105 		/* we need to issue all to keep CP_TRIMMED_FLAG */
1106 		dpolicy->granularity = 1;
1107 	}
1108 }
1109 
1110 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1111 				struct block_device *bdev, block_t lstart,
1112 				block_t start, block_t len);
1113 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1114 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1115 						struct discard_policy *dpolicy,
1116 						struct discard_cmd *dc,
1117 						unsigned int *issued)
1118 {
1119 	struct block_device *bdev = dc->bdev;
1120 	struct request_queue *q = bdev_get_queue(bdev);
1121 	unsigned int max_discard_blocks =
1122 			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1123 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1124 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1125 					&(dcc->fstrim_list) : &(dcc->wait_list);
1126 	int flag = dpolicy->sync ? REQ_SYNC : 0;
1127 	block_t lstart, start, len, total_len;
1128 	int err = 0;
1129 
1130 	if (dc->state != D_PREP)
1131 		return 0;
1132 
1133 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1134 		return 0;
1135 
1136 	trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1137 
1138 	lstart = dc->lstart;
1139 	start = dc->start;
1140 	len = dc->len;
1141 	total_len = len;
1142 
1143 	dc->len = 0;
1144 
1145 	while (total_len && *issued < dpolicy->max_requests && !err) {
1146 		struct bio *bio = NULL;
1147 		unsigned long flags;
1148 		bool last = true;
1149 
1150 		if (len > max_discard_blocks) {
1151 			len = max_discard_blocks;
1152 			last = false;
1153 		}
1154 
1155 		(*issued)++;
1156 		if (*issued == dpolicy->max_requests)
1157 			last = true;
1158 
1159 		dc->len += len;
1160 
1161 		if (time_to_inject(sbi, FAULT_DISCARD)) {
1162 			f2fs_show_injection_info(sbi, FAULT_DISCARD);
1163 			err = -EIO;
1164 			goto submit;
1165 		}
1166 		err = __blkdev_issue_discard(bdev,
1167 					SECTOR_FROM_BLOCK(start),
1168 					SECTOR_FROM_BLOCK(len),
1169 					GFP_NOFS, 0, &bio);
1170 submit:
1171 		if (err) {
1172 			spin_lock_irqsave(&dc->lock, flags);
1173 			if (dc->state == D_PARTIAL)
1174 				dc->state = D_SUBMIT;
1175 			spin_unlock_irqrestore(&dc->lock, flags);
1176 
1177 			break;
1178 		}
1179 
1180 		f2fs_bug_on(sbi, !bio);
1181 
1182 		/*
1183 		 * should keep before submission to avoid D_DONE
1184 		 * right away
1185 		 */
1186 		spin_lock_irqsave(&dc->lock, flags);
1187 		if (last)
1188 			dc->state = D_SUBMIT;
1189 		else
1190 			dc->state = D_PARTIAL;
1191 		dc->bio_ref++;
1192 		spin_unlock_irqrestore(&dc->lock, flags);
1193 
1194 		atomic_inc(&dcc->queued_discard);
1195 		dc->queued++;
1196 		list_move_tail(&dc->list, wait_list);
1197 
1198 		/* sanity check on discard range */
1199 		__check_sit_bitmap(sbi, lstart, lstart + len);
1200 
1201 		bio->bi_private = dc;
1202 		bio->bi_end_io = f2fs_submit_discard_endio;
1203 		bio->bi_opf |= flag;
1204 		submit_bio(bio);
1205 
1206 		atomic_inc(&dcc->issued_discard);
1207 
1208 		f2fs_update_iostat(sbi, FS_DISCARD, 1);
1209 
1210 		lstart += len;
1211 		start += len;
1212 		total_len -= len;
1213 		len = total_len;
1214 	}
1215 
1216 	if (!err && len)
1217 		__update_discard_tree_range(sbi, bdev, lstart, start, len);
1218 	return err;
1219 }
1220 
1221 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1222 				struct block_device *bdev, block_t lstart,
1223 				block_t start, block_t len,
1224 				struct rb_node **insert_p,
1225 				struct rb_node *insert_parent)
1226 {
1227 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1228 	struct rb_node **p;
1229 	struct rb_node *parent = NULL;
1230 	struct discard_cmd *dc = NULL;
1231 	bool leftmost = true;
1232 
1233 	if (insert_p && insert_parent) {
1234 		parent = insert_parent;
1235 		p = insert_p;
1236 		goto do_insert;
1237 	}
1238 
1239 	p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1240 							lstart, &leftmost);
1241 do_insert:
1242 	dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1243 								p, leftmost);
1244 	if (!dc)
1245 		return NULL;
1246 
1247 	return dc;
1248 }
1249 
1250 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1251 						struct discard_cmd *dc)
1252 {
1253 	list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1254 }
1255 
1256 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1257 				struct discard_cmd *dc, block_t blkaddr)
1258 {
1259 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1260 	struct discard_info di = dc->di;
1261 	bool modified = false;
1262 
1263 	if (dc->state == D_DONE || dc->len == 1) {
1264 		__remove_discard_cmd(sbi, dc);
1265 		return;
1266 	}
1267 
1268 	dcc->undiscard_blks -= di.len;
1269 
1270 	if (blkaddr > di.lstart) {
1271 		dc->len = blkaddr - dc->lstart;
1272 		dcc->undiscard_blks += dc->len;
1273 		__relocate_discard_cmd(dcc, dc);
1274 		modified = true;
1275 	}
1276 
1277 	if (blkaddr < di.lstart + di.len - 1) {
1278 		if (modified) {
1279 			__insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1280 					di.start + blkaddr + 1 - di.lstart,
1281 					di.lstart + di.len - 1 - blkaddr,
1282 					NULL, NULL);
1283 		} else {
1284 			dc->lstart++;
1285 			dc->len--;
1286 			dc->start++;
1287 			dcc->undiscard_blks += dc->len;
1288 			__relocate_discard_cmd(dcc, dc);
1289 		}
1290 	}
1291 }
1292 
1293 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1294 				struct block_device *bdev, block_t lstart,
1295 				block_t start, block_t len)
1296 {
1297 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1298 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1299 	struct discard_cmd *dc;
1300 	struct discard_info di = {0};
1301 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1302 	struct request_queue *q = bdev_get_queue(bdev);
1303 	unsigned int max_discard_blocks =
1304 			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1305 	block_t end = lstart + len;
1306 
1307 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1308 					NULL, lstart,
1309 					(struct rb_entry **)&prev_dc,
1310 					(struct rb_entry **)&next_dc,
1311 					&insert_p, &insert_parent, true, NULL);
1312 	if (dc)
1313 		prev_dc = dc;
1314 
1315 	if (!prev_dc) {
1316 		di.lstart = lstart;
1317 		di.len = next_dc ? next_dc->lstart - lstart : len;
1318 		di.len = min(di.len, len);
1319 		di.start = start;
1320 	}
1321 
1322 	while (1) {
1323 		struct rb_node *node;
1324 		bool merged = false;
1325 		struct discard_cmd *tdc = NULL;
1326 
1327 		if (prev_dc) {
1328 			di.lstart = prev_dc->lstart + prev_dc->len;
1329 			if (di.lstart < lstart)
1330 				di.lstart = lstart;
1331 			if (di.lstart >= end)
1332 				break;
1333 
1334 			if (!next_dc || next_dc->lstart > end)
1335 				di.len = end - di.lstart;
1336 			else
1337 				di.len = next_dc->lstart - di.lstart;
1338 			di.start = start + di.lstart - lstart;
1339 		}
1340 
1341 		if (!di.len)
1342 			goto next;
1343 
1344 		if (prev_dc && prev_dc->state == D_PREP &&
1345 			prev_dc->bdev == bdev &&
1346 			__is_discard_back_mergeable(&di, &prev_dc->di,
1347 							max_discard_blocks)) {
1348 			prev_dc->di.len += di.len;
1349 			dcc->undiscard_blks += di.len;
1350 			__relocate_discard_cmd(dcc, prev_dc);
1351 			di = prev_dc->di;
1352 			tdc = prev_dc;
1353 			merged = true;
1354 		}
1355 
1356 		if (next_dc && next_dc->state == D_PREP &&
1357 			next_dc->bdev == bdev &&
1358 			__is_discard_front_mergeable(&di, &next_dc->di,
1359 							max_discard_blocks)) {
1360 			next_dc->di.lstart = di.lstart;
1361 			next_dc->di.len += di.len;
1362 			next_dc->di.start = di.start;
1363 			dcc->undiscard_blks += di.len;
1364 			__relocate_discard_cmd(dcc, next_dc);
1365 			if (tdc)
1366 				__remove_discard_cmd(sbi, tdc);
1367 			merged = true;
1368 		}
1369 
1370 		if (!merged) {
1371 			__insert_discard_tree(sbi, bdev, di.lstart, di.start,
1372 							di.len, NULL, NULL);
1373 		}
1374  next:
1375 		prev_dc = next_dc;
1376 		if (!prev_dc)
1377 			break;
1378 
1379 		node = rb_next(&prev_dc->rb_node);
1380 		next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1381 	}
1382 }
1383 
1384 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1385 		struct block_device *bdev, block_t blkstart, block_t blklen)
1386 {
1387 	block_t lblkstart = blkstart;
1388 
1389 	if (!f2fs_bdev_support_discard(bdev))
1390 		return 0;
1391 
1392 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
1393 
1394 	if (f2fs_is_multi_device(sbi)) {
1395 		int devi = f2fs_target_device_index(sbi, blkstart);
1396 
1397 		blkstart -= FDEV(devi).start_blk;
1398 	}
1399 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1400 	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1401 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1402 	return 0;
1403 }
1404 
1405 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1406 					struct discard_policy *dpolicy)
1407 {
1408 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1409 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1410 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1411 	struct discard_cmd *dc;
1412 	struct blk_plug plug;
1413 	unsigned int pos = dcc->next_pos;
1414 	unsigned int issued = 0;
1415 	bool io_interrupted = false;
1416 
1417 	mutex_lock(&dcc->cmd_lock);
1418 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1419 					NULL, pos,
1420 					(struct rb_entry **)&prev_dc,
1421 					(struct rb_entry **)&next_dc,
1422 					&insert_p, &insert_parent, true, NULL);
1423 	if (!dc)
1424 		dc = next_dc;
1425 
1426 	blk_start_plug(&plug);
1427 
1428 	while (dc) {
1429 		struct rb_node *node;
1430 		int err = 0;
1431 
1432 		if (dc->state != D_PREP)
1433 			goto next;
1434 
1435 		if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1436 			io_interrupted = true;
1437 			break;
1438 		}
1439 
1440 		dcc->next_pos = dc->lstart + dc->len;
1441 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1442 
1443 		if (issued >= dpolicy->max_requests)
1444 			break;
1445 next:
1446 		node = rb_next(&dc->rb_node);
1447 		if (err)
1448 			__remove_discard_cmd(sbi, dc);
1449 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1450 	}
1451 
1452 	blk_finish_plug(&plug);
1453 
1454 	if (!dc)
1455 		dcc->next_pos = 0;
1456 
1457 	mutex_unlock(&dcc->cmd_lock);
1458 
1459 	if (!issued && io_interrupted)
1460 		issued = -1;
1461 
1462 	return issued;
1463 }
1464 
1465 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1466 					struct discard_policy *dpolicy)
1467 {
1468 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1469 	struct list_head *pend_list;
1470 	struct discard_cmd *dc, *tmp;
1471 	struct blk_plug plug;
1472 	int i, issued = 0;
1473 	bool io_interrupted = false;
1474 
1475 	if (dpolicy->timeout != 0)
1476 		f2fs_update_time(sbi, dpolicy->timeout);
1477 
1478 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1479 		if (dpolicy->timeout != 0 &&
1480 				f2fs_time_over(sbi, dpolicy->timeout))
1481 			break;
1482 
1483 		if (i + 1 < dpolicy->granularity)
1484 			break;
1485 
1486 		if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1487 			return __issue_discard_cmd_orderly(sbi, dpolicy);
1488 
1489 		pend_list = &dcc->pend_list[i];
1490 
1491 		mutex_lock(&dcc->cmd_lock);
1492 		if (list_empty(pend_list))
1493 			goto next;
1494 		if (unlikely(dcc->rbtree_check))
1495 			f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1496 								&dcc->root));
1497 		blk_start_plug(&plug);
1498 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1499 			f2fs_bug_on(sbi, dc->state != D_PREP);
1500 
1501 			if (dpolicy->timeout != 0 &&
1502 				f2fs_time_over(sbi, dpolicy->timeout))
1503 				break;
1504 
1505 			if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1506 						!is_idle(sbi, DISCARD_TIME)) {
1507 				io_interrupted = true;
1508 				break;
1509 			}
1510 
1511 			__submit_discard_cmd(sbi, dpolicy, dc, &issued);
1512 
1513 			if (issued >= dpolicy->max_requests)
1514 				break;
1515 		}
1516 		blk_finish_plug(&plug);
1517 next:
1518 		mutex_unlock(&dcc->cmd_lock);
1519 
1520 		if (issued >= dpolicy->max_requests || io_interrupted)
1521 			break;
1522 	}
1523 
1524 	if (!issued && io_interrupted)
1525 		issued = -1;
1526 
1527 	return issued;
1528 }
1529 
1530 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1531 {
1532 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1533 	struct list_head *pend_list;
1534 	struct discard_cmd *dc, *tmp;
1535 	int i;
1536 	bool dropped = false;
1537 
1538 	mutex_lock(&dcc->cmd_lock);
1539 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1540 		pend_list = &dcc->pend_list[i];
1541 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1542 			f2fs_bug_on(sbi, dc->state != D_PREP);
1543 			__remove_discard_cmd(sbi, dc);
1544 			dropped = true;
1545 		}
1546 	}
1547 	mutex_unlock(&dcc->cmd_lock);
1548 
1549 	return dropped;
1550 }
1551 
1552 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1553 {
1554 	__drop_discard_cmd(sbi);
1555 }
1556 
1557 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1558 							struct discard_cmd *dc)
1559 {
1560 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1561 	unsigned int len = 0;
1562 
1563 	wait_for_completion_io(&dc->wait);
1564 	mutex_lock(&dcc->cmd_lock);
1565 	f2fs_bug_on(sbi, dc->state != D_DONE);
1566 	dc->ref--;
1567 	if (!dc->ref) {
1568 		if (!dc->error)
1569 			len = dc->len;
1570 		__remove_discard_cmd(sbi, dc);
1571 	}
1572 	mutex_unlock(&dcc->cmd_lock);
1573 
1574 	return len;
1575 }
1576 
1577 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1578 						struct discard_policy *dpolicy,
1579 						block_t start, block_t end)
1580 {
1581 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1582 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1583 					&(dcc->fstrim_list) : &(dcc->wait_list);
1584 	struct discard_cmd *dc, *tmp;
1585 	bool need_wait;
1586 	unsigned int trimmed = 0;
1587 
1588 next:
1589 	need_wait = false;
1590 
1591 	mutex_lock(&dcc->cmd_lock);
1592 	list_for_each_entry_safe(dc, tmp, wait_list, list) {
1593 		if (dc->lstart + dc->len <= start || end <= dc->lstart)
1594 			continue;
1595 		if (dc->len < dpolicy->granularity)
1596 			continue;
1597 		if (dc->state == D_DONE && !dc->ref) {
1598 			wait_for_completion_io(&dc->wait);
1599 			if (!dc->error)
1600 				trimmed += dc->len;
1601 			__remove_discard_cmd(sbi, dc);
1602 		} else {
1603 			dc->ref++;
1604 			need_wait = true;
1605 			break;
1606 		}
1607 	}
1608 	mutex_unlock(&dcc->cmd_lock);
1609 
1610 	if (need_wait) {
1611 		trimmed += __wait_one_discard_bio(sbi, dc);
1612 		goto next;
1613 	}
1614 
1615 	return trimmed;
1616 }
1617 
1618 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1619 						struct discard_policy *dpolicy)
1620 {
1621 	struct discard_policy dp;
1622 	unsigned int discard_blks;
1623 
1624 	if (dpolicy)
1625 		return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1626 
1627 	/* wait all */
1628 	__init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1629 	discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1630 	__init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1631 	discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1632 
1633 	return discard_blks;
1634 }
1635 
1636 /* This should be covered by global mutex, &sit_i->sentry_lock */
1637 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1638 {
1639 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1640 	struct discard_cmd *dc;
1641 	bool need_wait = false;
1642 
1643 	mutex_lock(&dcc->cmd_lock);
1644 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1645 							NULL, blkaddr);
1646 	if (dc) {
1647 		if (dc->state == D_PREP) {
1648 			__punch_discard_cmd(sbi, dc, blkaddr);
1649 		} else {
1650 			dc->ref++;
1651 			need_wait = true;
1652 		}
1653 	}
1654 	mutex_unlock(&dcc->cmd_lock);
1655 
1656 	if (need_wait)
1657 		__wait_one_discard_bio(sbi, dc);
1658 }
1659 
1660 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1661 {
1662 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1663 
1664 	if (dcc && dcc->f2fs_issue_discard) {
1665 		struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1666 
1667 		dcc->f2fs_issue_discard = NULL;
1668 		kthread_stop(discard_thread);
1669 	}
1670 }
1671 
1672 /* This comes from f2fs_put_super */
1673 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1674 {
1675 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1676 	struct discard_policy dpolicy;
1677 	bool dropped;
1678 
1679 	__init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1680 					dcc->discard_granularity);
1681 	dpolicy.timeout = UMOUNT_DISCARD_TIMEOUT;
1682 	__issue_discard_cmd(sbi, &dpolicy);
1683 	dropped = __drop_discard_cmd(sbi);
1684 
1685 	/* just to make sure there is no pending discard commands */
1686 	__wait_all_discard_cmd(sbi, NULL);
1687 
1688 	f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1689 	return dropped;
1690 }
1691 
1692 static int issue_discard_thread(void *data)
1693 {
1694 	struct f2fs_sb_info *sbi = data;
1695 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1696 	wait_queue_head_t *q = &dcc->discard_wait_queue;
1697 	struct discard_policy dpolicy;
1698 	unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1699 	int issued;
1700 
1701 	set_freezable();
1702 
1703 	do {
1704 		__init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1705 					dcc->discard_granularity);
1706 
1707 		wait_event_interruptible_timeout(*q,
1708 				kthread_should_stop() || freezing(current) ||
1709 				dcc->discard_wake,
1710 				msecs_to_jiffies(wait_ms));
1711 
1712 		if (dcc->discard_wake)
1713 			dcc->discard_wake = 0;
1714 
1715 		/* clean up pending candidates before going to sleep */
1716 		if (atomic_read(&dcc->queued_discard))
1717 			__wait_all_discard_cmd(sbi, NULL);
1718 
1719 		if (try_to_freeze())
1720 			continue;
1721 		if (f2fs_readonly(sbi->sb))
1722 			continue;
1723 		if (kthread_should_stop())
1724 			return 0;
1725 		if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1726 			wait_ms = dpolicy.max_interval;
1727 			continue;
1728 		}
1729 
1730 		if (sbi->gc_mode == GC_URGENT)
1731 			__init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1732 
1733 		sb_start_intwrite(sbi->sb);
1734 
1735 		issued = __issue_discard_cmd(sbi, &dpolicy);
1736 		if (issued > 0) {
1737 			__wait_all_discard_cmd(sbi, &dpolicy);
1738 			wait_ms = dpolicy.min_interval;
1739 		} else if (issued == -1){
1740 			wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1741 			if (!wait_ms)
1742 				wait_ms = dpolicy.mid_interval;
1743 		} else {
1744 			wait_ms = dpolicy.max_interval;
1745 		}
1746 
1747 		sb_end_intwrite(sbi->sb);
1748 
1749 	} while (!kthread_should_stop());
1750 	return 0;
1751 }
1752 
1753 #ifdef CONFIG_BLK_DEV_ZONED
1754 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1755 		struct block_device *bdev, block_t blkstart, block_t blklen)
1756 {
1757 	sector_t sector, nr_sects;
1758 	block_t lblkstart = blkstart;
1759 	int devi = 0;
1760 
1761 	if (f2fs_is_multi_device(sbi)) {
1762 		devi = f2fs_target_device_index(sbi, blkstart);
1763 		if (blkstart < FDEV(devi).start_blk ||
1764 		    blkstart > FDEV(devi).end_blk) {
1765 			f2fs_err(sbi, "Invalid block %x", blkstart);
1766 			return -EIO;
1767 		}
1768 		blkstart -= FDEV(devi).start_blk;
1769 	}
1770 
1771 	/* For sequential zones, reset the zone write pointer */
1772 	if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1773 		sector = SECTOR_FROM_BLOCK(blkstart);
1774 		nr_sects = SECTOR_FROM_BLOCK(blklen);
1775 
1776 		if (sector & (bdev_zone_sectors(bdev) - 1) ||
1777 				nr_sects != bdev_zone_sectors(bdev)) {
1778 			f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1779 				 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1780 				 blkstart, blklen);
1781 			return -EIO;
1782 		}
1783 		trace_f2fs_issue_reset_zone(bdev, blkstart);
1784 		return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1785 					sector, nr_sects, GFP_NOFS);
1786 	}
1787 
1788 	/* For conventional zones, use regular discard if supported */
1789 	return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1790 }
1791 #endif
1792 
1793 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1794 		struct block_device *bdev, block_t blkstart, block_t blklen)
1795 {
1796 #ifdef CONFIG_BLK_DEV_ZONED
1797 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1798 		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1799 #endif
1800 	return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1801 }
1802 
1803 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1804 				block_t blkstart, block_t blklen)
1805 {
1806 	sector_t start = blkstart, len = 0;
1807 	struct block_device *bdev;
1808 	struct seg_entry *se;
1809 	unsigned int offset;
1810 	block_t i;
1811 	int err = 0;
1812 
1813 	bdev = f2fs_target_device(sbi, blkstart, NULL);
1814 
1815 	for (i = blkstart; i < blkstart + blklen; i++, len++) {
1816 		if (i != start) {
1817 			struct block_device *bdev2 =
1818 				f2fs_target_device(sbi, i, NULL);
1819 
1820 			if (bdev2 != bdev) {
1821 				err = __issue_discard_async(sbi, bdev,
1822 						start, len);
1823 				if (err)
1824 					return err;
1825 				bdev = bdev2;
1826 				start = i;
1827 				len = 0;
1828 			}
1829 		}
1830 
1831 		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1832 		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1833 
1834 		if (!f2fs_test_and_set_bit(offset, se->discard_map))
1835 			sbi->discard_blks--;
1836 	}
1837 
1838 	if (len)
1839 		err = __issue_discard_async(sbi, bdev, start, len);
1840 	return err;
1841 }
1842 
1843 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1844 							bool check_only)
1845 {
1846 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1847 	int max_blocks = sbi->blocks_per_seg;
1848 	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1849 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1850 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1851 	unsigned long *discard_map = (unsigned long *)se->discard_map;
1852 	unsigned long *dmap = SIT_I(sbi)->tmp_map;
1853 	unsigned int start = 0, end = -1;
1854 	bool force = (cpc->reason & CP_DISCARD);
1855 	struct discard_entry *de = NULL;
1856 	struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1857 	int i;
1858 
1859 	if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1860 		return false;
1861 
1862 	if (!force) {
1863 		if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1864 			SM_I(sbi)->dcc_info->nr_discards >=
1865 				SM_I(sbi)->dcc_info->max_discards)
1866 			return false;
1867 	}
1868 
1869 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1870 	for (i = 0; i < entries; i++)
1871 		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1872 				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1873 
1874 	while (force || SM_I(sbi)->dcc_info->nr_discards <=
1875 				SM_I(sbi)->dcc_info->max_discards) {
1876 		start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1877 		if (start >= max_blocks)
1878 			break;
1879 
1880 		end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1881 		if (force && start && end != max_blocks
1882 					&& (end - start) < cpc->trim_minlen)
1883 			continue;
1884 
1885 		if (check_only)
1886 			return true;
1887 
1888 		if (!de) {
1889 			de = f2fs_kmem_cache_alloc(discard_entry_slab,
1890 								GFP_F2FS_ZERO);
1891 			de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1892 			list_add_tail(&de->list, head);
1893 		}
1894 
1895 		for (i = start; i < end; i++)
1896 			__set_bit_le(i, (void *)de->discard_map);
1897 
1898 		SM_I(sbi)->dcc_info->nr_discards += end - start;
1899 	}
1900 	return false;
1901 }
1902 
1903 static void release_discard_addr(struct discard_entry *entry)
1904 {
1905 	list_del(&entry->list);
1906 	kmem_cache_free(discard_entry_slab, entry);
1907 }
1908 
1909 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1910 {
1911 	struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1912 	struct discard_entry *entry, *this;
1913 
1914 	/* drop caches */
1915 	list_for_each_entry_safe(entry, this, head, list)
1916 		release_discard_addr(entry);
1917 }
1918 
1919 /*
1920  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1921  */
1922 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1923 {
1924 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1925 	unsigned int segno;
1926 
1927 	mutex_lock(&dirty_i->seglist_lock);
1928 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1929 		__set_test_and_free(sbi, segno);
1930 	mutex_unlock(&dirty_i->seglist_lock);
1931 }
1932 
1933 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1934 						struct cp_control *cpc)
1935 {
1936 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1937 	struct list_head *head = &dcc->entry_list;
1938 	struct discard_entry *entry, *this;
1939 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1940 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1941 	unsigned int start = 0, end = -1;
1942 	unsigned int secno, start_segno;
1943 	bool force = (cpc->reason & CP_DISCARD);
1944 	bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
1945 
1946 	mutex_lock(&dirty_i->seglist_lock);
1947 
1948 	while (1) {
1949 		int i;
1950 
1951 		if (need_align && end != -1)
1952 			end--;
1953 		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1954 		if (start >= MAIN_SEGS(sbi))
1955 			break;
1956 		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1957 								start + 1);
1958 
1959 		if (need_align) {
1960 			start = rounddown(start, sbi->segs_per_sec);
1961 			end = roundup(end, sbi->segs_per_sec);
1962 		}
1963 
1964 		for (i = start; i < end; i++) {
1965 			if (test_and_clear_bit(i, prefree_map))
1966 				dirty_i->nr_dirty[PRE]--;
1967 		}
1968 
1969 		if (!f2fs_realtime_discard_enable(sbi))
1970 			continue;
1971 
1972 		if (force && start >= cpc->trim_start &&
1973 					(end - 1) <= cpc->trim_end)
1974 				continue;
1975 
1976 		if (!test_opt(sbi, LFS) || !__is_large_section(sbi)) {
1977 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1978 				(end - start) << sbi->log_blocks_per_seg);
1979 			continue;
1980 		}
1981 next:
1982 		secno = GET_SEC_FROM_SEG(sbi, start);
1983 		start_segno = GET_SEG_FROM_SEC(sbi, secno);
1984 		if (!IS_CURSEC(sbi, secno) &&
1985 			!get_valid_blocks(sbi, start, true))
1986 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1987 				sbi->segs_per_sec << sbi->log_blocks_per_seg);
1988 
1989 		start = start_segno + sbi->segs_per_sec;
1990 		if (start < end)
1991 			goto next;
1992 		else
1993 			end = start - 1;
1994 	}
1995 	mutex_unlock(&dirty_i->seglist_lock);
1996 
1997 	/* send small discards */
1998 	list_for_each_entry_safe(entry, this, head, list) {
1999 		unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2000 		bool is_valid = test_bit_le(0, entry->discard_map);
2001 
2002 find_next:
2003 		if (is_valid) {
2004 			next_pos = find_next_zero_bit_le(entry->discard_map,
2005 					sbi->blocks_per_seg, cur_pos);
2006 			len = next_pos - cur_pos;
2007 
2008 			if (f2fs_sb_has_blkzoned(sbi) ||
2009 			    (force && len < cpc->trim_minlen))
2010 				goto skip;
2011 
2012 			f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2013 									len);
2014 			total_len += len;
2015 		} else {
2016 			next_pos = find_next_bit_le(entry->discard_map,
2017 					sbi->blocks_per_seg, cur_pos);
2018 		}
2019 skip:
2020 		cur_pos = next_pos;
2021 		is_valid = !is_valid;
2022 
2023 		if (cur_pos < sbi->blocks_per_seg)
2024 			goto find_next;
2025 
2026 		release_discard_addr(entry);
2027 		dcc->nr_discards -= total_len;
2028 	}
2029 
2030 	wake_up_discard_thread(sbi, false);
2031 }
2032 
2033 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2034 {
2035 	dev_t dev = sbi->sb->s_bdev->bd_dev;
2036 	struct discard_cmd_control *dcc;
2037 	int err = 0, i;
2038 
2039 	if (SM_I(sbi)->dcc_info) {
2040 		dcc = SM_I(sbi)->dcc_info;
2041 		goto init_thread;
2042 	}
2043 
2044 	dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2045 	if (!dcc)
2046 		return -ENOMEM;
2047 
2048 	dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2049 	INIT_LIST_HEAD(&dcc->entry_list);
2050 	for (i = 0; i < MAX_PLIST_NUM; i++)
2051 		INIT_LIST_HEAD(&dcc->pend_list[i]);
2052 	INIT_LIST_HEAD(&dcc->wait_list);
2053 	INIT_LIST_HEAD(&dcc->fstrim_list);
2054 	mutex_init(&dcc->cmd_lock);
2055 	atomic_set(&dcc->issued_discard, 0);
2056 	atomic_set(&dcc->queued_discard, 0);
2057 	atomic_set(&dcc->discard_cmd_cnt, 0);
2058 	dcc->nr_discards = 0;
2059 	dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2060 	dcc->undiscard_blks = 0;
2061 	dcc->next_pos = 0;
2062 	dcc->root = RB_ROOT_CACHED;
2063 	dcc->rbtree_check = false;
2064 
2065 	init_waitqueue_head(&dcc->discard_wait_queue);
2066 	SM_I(sbi)->dcc_info = dcc;
2067 init_thread:
2068 	dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2069 				"f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2070 	if (IS_ERR(dcc->f2fs_issue_discard)) {
2071 		err = PTR_ERR(dcc->f2fs_issue_discard);
2072 		kvfree(dcc);
2073 		SM_I(sbi)->dcc_info = NULL;
2074 		return err;
2075 	}
2076 
2077 	return err;
2078 }
2079 
2080 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2081 {
2082 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2083 
2084 	if (!dcc)
2085 		return;
2086 
2087 	f2fs_stop_discard_thread(sbi);
2088 
2089 	/*
2090 	 * Recovery can cache discard commands, so in error path of
2091 	 * fill_super(), it needs to give a chance to handle them.
2092 	 */
2093 	if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2094 		f2fs_issue_discard_timeout(sbi);
2095 
2096 	kvfree(dcc);
2097 	SM_I(sbi)->dcc_info = NULL;
2098 }
2099 
2100 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2101 {
2102 	struct sit_info *sit_i = SIT_I(sbi);
2103 
2104 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2105 		sit_i->dirty_sentries++;
2106 		return false;
2107 	}
2108 
2109 	return true;
2110 }
2111 
2112 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2113 					unsigned int segno, int modified)
2114 {
2115 	struct seg_entry *se = get_seg_entry(sbi, segno);
2116 	se->type = type;
2117 	if (modified)
2118 		__mark_sit_entry_dirty(sbi, segno);
2119 }
2120 
2121 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2122 {
2123 	struct seg_entry *se;
2124 	unsigned int segno, offset;
2125 	long int new_vblocks;
2126 	bool exist;
2127 #ifdef CONFIG_F2FS_CHECK_FS
2128 	bool mir_exist;
2129 #endif
2130 
2131 	segno = GET_SEGNO(sbi, blkaddr);
2132 
2133 	se = get_seg_entry(sbi, segno);
2134 	new_vblocks = se->valid_blocks + del;
2135 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2136 
2137 	f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
2138 				(new_vblocks > sbi->blocks_per_seg)));
2139 
2140 	se->valid_blocks = new_vblocks;
2141 	se->mtime = get_mtime(sbi, false);
2142 	if (se->mtime > SIT_I(sbi)->max_mtime)
2143 		SIT_I(sbi)->max_mtime = se->mtime;
2144 
2145 	/* Update valid block bitmap */
2146 	if (del > 0) {
2147 		exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2148 #ifdef CONFIG_F2FS_CHECK_FS
2149 		mir_exist = f2fs_test_and_set_bit(offset,
2150 						se->cur_valid_map_mir);
2151 		if (unlikely(exist != mir_exist)) {
2152 			f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2153 				 blkaddr, exist);
2154 			f2fs_bug_on(sbi, 1);
2155 		}
2156 #endif
2157 		if (unlikely(exist)) {
2158 			f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2159 				 blkaddr);
2160 			f2fs_bug_on(sbi, 1);
2161 			se->valid_blocks--;
2162 			del = 0;
2163 		}
2164 
2165 		if (!f2fs_test_and_set_bit(offset, se->discard_map))
2166 			sbi->discard_blks--;
2167 
2168 		/*
2169 		 * SSR should never reuse block which is checkpointed
2170 		 * or newly invalidated.
2171 		 */
2172 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2173 			if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2174 				se->ckpt_valid_blocks++;
2175 		}
2176 	} else {
2177 		exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2178 #ifdef CONFIG_F2FS_CHECK_FS
2179 		mir_exist = f2fs_test_and_clear_bit(offset,
2180 						se->cur_valid_map_mir);
2181 		if (unlikely(exist != mir_exist)) {
2182 			f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2183 				 blkaddr, exist);
2184 			f2fs_bug_on(sbi, 1);
2185 		}
2186 #endif
2187 		if (unlikely(!exist)) {
2188 			f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2189 				 blkaddr);
2190 			f2fs_bug_on(sbi, 1);
2191 			se->valid_blocks++;
2192 			del = 0;
2193 		} else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2194 			/*
2195 			 * If checkpoints are off, we must not reuse data that
2196 			 * was used in the previous checkpoint. If it was used
2197 			 * before, we must track that to know how much space we
2198 			 * really have.
2199 			 */
2200 			if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2201 				spin_lock(&sbi->stat_lock);
2202 				sbi->unusable_block_count++;
2203 				spin_unlock(&sbi->stat_lock);
2204 			}
2205 		}
2206 
2207 		if (f2fs_test_and_clear_bit(offset, se->discard_map))
2208 			sbi->discard_blks++;
2209 	}
2210 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2211 		se->ckpt_valid_blocks += del;
2212 
2213 	__mark_sit_entry_dirty(sbi, segno);
2214 
2215 	/* update total number of valid blocks to be written in ckpt area */
2216 	SIT_I(sbi)->written_valid_blocks += del;
2217 
2218 	if (__is_large_section(sbi))
2219 		get_sec_entry(sbi, segno)->valid_blocks += del;
2220 }
2221 
2222 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2223 {
2224 	unsigned int segno = GET_SEGNO(sbi, addr);
2225 	struct sit_info *sit_i = SIT_I(sbi);
2226 
2227 	f2fs_bug_on(sbi, addr == NULL_ADDR);
2228 	if (addr == NEW_ADDR)
2229 		return;
2230 
2231 	invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2232 
2233 	/* add it into sit main buffer */
2234 	down_write(&sit_i->sentry_lock);
2235 
2236 	update_sit_entry(sbi, addr, -1);
2237 
2238 	/* add it into dirty seglist */
2239 	locate_dirty_segment(sbi, segno);
2240 
2241 	up_write(&sit_i->sentry_lock);
2242 }
2243 
2244 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2245 {
2246 	struct sit_info *sit_i = SIT_I(sbi);
2247 	unsigned int segno, offset;
2248 	struct seg_entry *se;
2249 	bool is_cp = false;
2250 
2251 	if (!__is_valid_data_blkaddr(blkaddr))
2252 		return true;
2253 
2254 	down_read(&sit_i->sentry_lock);
2255 
2256 	segno = GET_SEGNO(sbi, blkaddr);
2257 	se = get_seg_entry(sbi, segno);
2258 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2259 
2260 	if (f2fs_test_bit(offset, se->ckpt_valid_map))
2261 		is_cp = true;
2262 
2263 	up_read(&sit_i->sentry_lock);
2264 
2265 	return is_cp;
2266 }
2267 
2268 /*
2269  * This function should be resided under the curseg_mutex lock
2270  */
2271 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2272 					struct f2fs_summary *sum)
2273 {
2274 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2275 	void *addr = curseg->sum_blk;
2276 	addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2277 	memcpy(addr, sum, sizeof(struct f2fs_summary));
2278 }
2279 
2280 /*
2281  * Calculate the number of current summary pages for writing
2282  */
2283 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2284 {
2285 	int valid_sum_count = 0;
2286 	int i, sum_in_page;
2287 
2288 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2289 		if (sbi->ckpt->alloc_type[i] == SSR)
2290 			valid_sum_count += sbi->blocks_per_seg;
2291 		else {
2292 			if (for_ra)
2293 				valid_sum_count += le16_to_cpu(
2294 					F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2295 			else
2296 				valid_sum_count += curseg_blkoff(sbi, i);
2297 		}
2298 	}
2299 
2300 	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2301 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2302 	if (valid_sum_count <= sum_in_page)
2303 		return 1;
2304 	else if ((valid_sum_count - sum_in_page) <=
2305 		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2306 		return 2;
2307 	return 3;
2308 }
2309 
2310 /*
2311  * Caller should put this summary page
2312  */
2313 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2314 {
2315 	return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2316 }
2317 
2318 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2319 					void *src, block_t blk_addr)
2320 {
2321 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2322 
2323 	memcpy(page_address(page), src, PAGE_SIZE);
2324 	set_page_dirty(page);
2325 	f2fs_put_page(page, 1);
2326 }
2327 
2328 static void write_sum_page(struct f2fs_sb_info *sbi,
2329 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
2330 {
2331 	f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2332 }
2333 
2334 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2335 						int type, block_t blk_addr)
2336 {
2337 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2338 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2339 	struct f2fs_summary_block *src = curseg->sum_blk;
2340 	struct f2fs_summary_block *dst;
2341 
2342 	dst = (struct f2fs_summary_block *)page_address(page);
2343 	memset(dst, 0, PAGE_SIZE);
2344 
2345 	mutex_lock(&curseg->curseg_mutex);
2346 
2347 	down_read(&curseg->journal_rwsem);
2348 	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2349 	up_read(&curseg->journal_rwsem);
2350 
2351 	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2352 	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2353 
2354 	mutex_unlock(&curseg->curseg_mutex);
2355 
2356 	set_page_dirty(page);
2357 	f2fs_put_page(page, 1);
2358 }
2359 
2360 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2361 {
2362 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2363 	unsigned int segno = curseg->segno + 1;
2364 	struct free_segmap_info *free_i = FREE_I(sbi);
2365 
2366 	if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2367 		return !test_bit(segno, free_i->free_segmap);
2368 	return 0;
2369 }
2370 
2371 /*
2372  * Find a new segment from the free segments bitmap to right order
2373  * This function should be returned with success, otherwise BUG
2374  */
2375 static void get_new_segment(struct f2fs_sb_info *sbi,
2376 			unsigned int *newseg, bool new_sec, int dir)
2377 {
2378 	struct free_segmap_info *free_i = FREE_I(sbi);
2379 	unsigned int segno, secno, zoneno;
2380 	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2381 	unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2382 	unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2383 	unsigned int left_start = hint;
2384 	bool init = true;
2385 	int go_left = 0;
2386 	int i;
2387 
2388 	spin_lock(&free_i->segmap_lock);
2389 
2390 	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2391 		segno = find_next_zero_bit(free_i->free_segmap,
2392 			GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2393 		if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2394 			goto got_it;
2395 	}
2396 find_other_zone:
2397 	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2398 	if (secno >= MAIN_SECS(sbi)) {
2399 		if (dir == ALLOC_RIGHT) {
2400 			secno = find_next_zero_bit(free_i->free_secmap,
2401 							MAIN_SECS(sbi), 0);
2402 			f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2403 		} else {
2404 			go_left = 1;
2405 			left_start = hint - 1;
2406 		}
2407 	}
2408 	if (go_left == 0)
2409 		goto skip_left;
2410 
2411 	while (test_bit(left_start, free_i->free_secmap)) {
2412 		if (left_start > 0) {
2413 			left_start--;
2414 			continue;
2415 		}
2416 		left_start = find_next_zero_bit(free_i->free_secmap,
2417 							MAIN_SECS(sbi), 0);
2418 		f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2419 		break;
2420 	}
2421 	secno = left_start;
2422 skip_left:
2423 	segno = GET_SEG_FROM_SEC(sbi, secno);
2424 	zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2425 
2426 	/* give up on finding another zone */
2427 	if (!init)
2428 		goto got_it;
2429 	if (sbi->secs_per_zone == 1)
2430 		goto got_it;
2431 	if (zoneno == old_zoneno)
2432 		goto got_it;
2433 	if (dir == ALLOC_LEFT) {
2434 		if (!go_left && zoneno + 1 >= total_zones)
2435 			goto got_it;
2436 		if (go_left && zoneno == 0)
2437 			goto got_it;
2438 	}
2439 	for (i = 0; i < NR_CURSEG_TYPE; i++)
2440 		if (CURSEG_I(sbi, i)->zone == zoneno)
2441 			break;
2442 
2443 	if (i < NR_CURSEG_TYPE) {
2444 		/* zone is in user, try another */
2445 		if (go_left)
2446 			hint = zoneno * sbi->secs_per_zone - 1;
2447 		else if (zoneno + 1 >= total_zones)
2448 			hint = 0;
2449 		else
2450 			hint = (zoneno + 1) * sbi->secs_per_zone;
2451 		init = false;
2452 		goto find_other_zone;
2453 	}
2454 got_it:
2455 	/* set it as dirty segment in free segmap */
2456 	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2457 	__set_inuse(sbi, segno);
2458 	*newseg = segno;
2459 	spin_unlock(&free_i->segmap_lock);
2460 }
2461 
2462 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2463 {
2464 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2465 	struct summary_footer *sum_footer;
2466 
2467 	curseg->segno = curseg->next_segno;
2468 	curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2469 	curseg->next_blkoff = 0;
2470 	curseg->next_segno = NULL_SEGNO;
2471 
2472 	sum_footer = &(curseg->sum_blk->footer);
2473 	memset(sum_footer, 0, sizeof(struct summary_footer));
2474 	if (IS_DATASEG(type))
2475 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2476 	if (IS_NODESEG(type))
2477 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2478 	__set_sit_entry_type(sbi, type, curseg->segno, modified);
2479 }
2480 
2481 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2482 {
2483 	/* if segs_per_sec is large than 1, we need to keep original policy. */
2484 	if (__is_large_section(sbi))
2485 		return CURSEG_I(sbi, type)->segno;
2486 
2487 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2488 		return 0;
2489 
2490 	if (test_opt(sbi, NOHEAP) &&
2491 		(type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2492 		return 0;
2493 
2494 	if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2495 		return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2496 
2497 	/* find segments from 0 to reuse freed segments */
2498 	if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2499 		return 0;
2500 
2501 	return CURSEG_I(sbi, type)->segno;
2502 }
2503 
2504 /*
2505  * Allocate a current working segment.
2506  * This function always allocates a free segment in LFS manner.
2507  */
2508 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2509 {
2510 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2511 	unsigned int segno = curseg->segno;
2512 	int dir = ALLOC_LEFT;
2513 
2514 	write_sum_page(sbi, curseg->sum_blk,
2515 				GET_SUM_BLOCK(sbi, segno));
2516 	if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2517 		dir = ALLOC_RIGHT;
2518 
2519 	if (test_opt(sbi, NOHEAP))
2520 		dir = ALLOC_RIGHT;
2521 
2522 	segno = __get_next_segno(sbi, type);
2523 	get_new_segment(sbi, &segno, new_sec, dir);
2524 	curseg->next_segno = segno;
2525 	reset_curseg(sbi, type, 1);
2526 	curseg->alloc_type = LFS;
2527 }
2528 
2529 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2530 			struct curseg_info *seg, block_t start)
2531 {
2532 	struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2533 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2534 	unsigned long *target_map = SIT_I(sbi)->tmp_map;
2535 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2536 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2537 	int i, pos;
2538 
2539 	for (i = 0; i < entries; i++)
2540 		target_map[i] = ckpt_map[i] | cur_map[i];
2541 
2542 	pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2543 
2544 	seg->next_blkoff = pos;
2545 }
2546 
2547 /*
2548  * If a segment is written by LFS manner, next block offset is just obtained
2549  * by increasing the current block offset. However, if a segment is written by
2550  * SSR manner, next block offset obtained by calling __next_free_blkoff
2551  */
2552 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2553 				struct curseg_info *seg)
2554 {
2555 	if (seg->alloc_type == SSR)
2556 		__next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2557 	else
2558 		seg->next_blkoff++;
2559 }
2560 
2561 /*
2562  * This function always allocates a used segment(from dirty seglist) by SSR
2563  * manner, so it should recover the existing segment information of valid blocks
2564  */
2565 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2566 {
2567 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2568 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2569 	unsigned int new_segno = curseg->next_segno;
2570 	struct f2fs_summary_block *sum_node;
2571 	struct page *sum_page;
2572 
2573 	write_sum_page(sbi, curseg->sum_blk,
2574 				GET_SUM_BLOCK(sbi, curseg->segno));
2575 	__set_test_and_inuse(sbi, new_segno);
2576 
2577 	mutex_lock(&dirty_i->seglist_lock);
2578 	__remove_dirty_segment(sbi, new_segno, PRE);
2579 	__remove_dirty_segment(sbi, new_segno, DIRTY);
2580 	mutex_unlock(&dirty_i->seglist_lock);
2581 
2582 	reset_curseg(sbi, type, 1);
2583 	curseg->alloc_type = SSR;
2584 	__next_free_blkoff(sbi, curseg, 0);
2585 
2586 	sum_page = f2fs_get_sum_page(sbi, new_segno);
2587 	f2fs_bug_on(sbi, IS_ERR(sum_page));
2588 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2589 	memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2590 	f2fs_put_page(sum_page, 1);
2591 }
2592 
2593 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2594 {
2595 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2596 	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2597 	unsigned segno = NULL_SEGNO;
2598 	int i, cnt;
2599 	bool reversed = false;
2600 
2601 	/* f2fs_need_SSR() already forces to do this */
2602 	if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2603 		curseg->next_segno = segno;
2604 		return 1;
2605 	}
2606 
2607 	/* For node segments, let's do SSR more intensively */
2608 	if (IS_NODESEG(type)) {
2609 		if (type >= CURSEG_WARM_NODE) {
2610 			reversed = true;
2611 			i = CURSEG_COLD_NODE;
2612 		} else {
2613 			i = CURSEG_HOT_NODE;
2614 		}
2615 		cnt = NR_CURSEG_NODE_TYPE;
2616 	} else {
2617 		if (type >= CURSEG_WARM_DATA) {
2618 			reversed = true;
2619 			i = CURSEG_COLD_DATA;
2620 		} else {
2621 			i = CURSEG_HOT_DATA;
2622 		}
2623 		cnt = NR_CURSEG_DATA_TYPE;
2624 	}
2625 
2626 	for (; cnt-- > 0; reversed ? i-- : i++) {
2627 		if (i == type)
2628 			continue;
2629 		if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2630 			curseg->next_segno = segno;
2631 			return 1;
2632 		}
2633 	}
2634 
2635 	/* find valid_blocks=0 in dirty list */
2636 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2637 		segno = get_free_segment(sbi);
2638 		if (segno != NULL_SEGNO) {
2639 			curseg->next_segno = segno;
2640 			return 1;
2641 		}
2642 	}
2643 	return 0;
2644 }
2645 
2646 /*
2647  * flush out current segment and replace it with new segment
2648  * This function should be returned with success, otherwise BUG
2649  */
2650 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2651 						int type, bool force)
2652 {
2653 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2654 
2655 	if (force)
2656 		new_curseg(sbi, type, true);
2657 	else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2658 					type == CURSEG_WARM_NODE)
2659 		new_curseg(sbi, type, false);
2660 	else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2661 			likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2662 		new_curseg(sbi, type, false);
2663 	else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2664 		change_curseg(sbi, type);
2665 	else
2666 		new_curseg(sbi, type, false);
2667 
2668 	stat_inc_seg_type(sbi, curseg);
2669 }
2670 
2671 void allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2672 					unsigned int start, unsigned int end)
2673 {
2674 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2675 	unsigned int segno;
2676 
2677 	down_read(&SM_I(sbi)->curseg_lock);
2678 	mutex_lock(&curseg->curseg_mutex);
2679 	down_write(&SIT_I(sbi)->sentry_lock);
2680 
2681 	segno = CURSEG_I(sbi, type)->segno;
2682 	if (segno < start || segno > end)
2683 		goto unlock;
2684 
2685 	if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2686 		change_curseg(sbi, type);
2687 	else
2688 		new_curseg(sbi, type, true);
2689 
2690 	stat_inc_seg_type(sbi, curseg);
2691 
2692 	locate_dirty_segment(sbi, segno);
2693 unlock:
2694 	up_write(&SIT_I(sbi)->sentry_lock);
2695 
2696 	if (segno != curseg->segno)
2697 		f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2698 			    type, segno, curseg->segno);
2699 
2700 	mutex_unlock(&curseg->curseg_mutex);
2701 	up_read(&SM_I(sbi)->curseg_lock);
2702 }
2703 
2704 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi, int type)
2705 {
2706 	struct curseg_info *curseg;
2707 	unsigned int old_segno;
2708 	int i;
2709 
2710 	down_write(&SIT_I(sbi)->sentry_lock);
2711 
2712 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2713 		if (type != NO_CHECK_TYPE && i != type)
2714 			continue;
2715 
2716 		curseg = CURSEG_I(sbi, i);
2717 		if (type == NO_CHECK_TYPE || curseg->next_blkoff ||
2718 				get_valid_blocks(sbi, curseg->segno, false) ||
2719 				get_ckpt_valid_blocks(sbi, curseg->segno)) {
2720 			old_segno = curseg->segno;
2721 			SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2722 			locate_dirty_segment(sbi, old_segno);
2723 		}
2724 	}
2725 
2726 	up_write(&SIT_I(sbi)->sentry_lock);
2727 }
2728 
2729 static const struct segment_allocation default_salloc_ops = {
2730 	.allocate_segment = allocate_segment_by_default,
2731 };
2732 
2733 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2734 						struct cp_control *cpc)
2735 {
2736 	__u64 trim_start = cpc->trim_start;
2737 	bool has_candidate = false;
2738 
2739 	down_write(&SIT_I(sbi)->sentry_lock);
2740 	for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2741 		if (add_discard_addrs(sbi, cpc, true)) {
2742 			has_candidate = true;
2743 			break;
2744 		}
2745 	}
2746 	up_write(&SIT_I(sbi)->sentry_lock);
2747 
2748 	cpc->trim_start = trim_start;
2749 	return has_candidate;
2750 }
2751 
2752 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2753 					struct discard_policy *dpolicy,
2754 					unsigned int start, unsigned int end)
2755 {
2756 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2757 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2758 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
2759 	struct discard_cmd *dc;
2760 	struct blk_plug plug;
2761 	int issued;
2762 	unsigned int trimmed = 0;
2763 
2764 next:
2765 	issued = 0;
2766 
2767 	mutex_lock(&dcc->cmd_lock);
2768 	if (unlikely(dcc->rbtree_check))
2769 		f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2770 								&dcc->root));
2771 
2772 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2773 					NULL, start,
2774 					(struct rb_entry **)&prev_dc,
2775 					(struct rb_entry **)&next_dc,
2776 					&insert_p, &insert_parent, true, NULL);
2777 	if (!dc)
2778 		dc = next_dc;
2779 
2780 	blk_start_plug(&plug);
2781 
2782 	while (dc && dc->lstart <= end) {
2783 		struct rb_node *node;
2784 		int err = 0;
2785 
2786 		if (dc->len < dpolicy->granularity)
2787 			goto skip;
2788 
2789 		if (dc->state != D_PREP) {
2790 			list_move_tail(&dc->list, &dcc->fstrim_list);
2791 			goto skip;
2792 		}
2793 
2794 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2795 
2796 		if (issued >= dpolicy->max_requests) {
2797 			start = dc->lstart + dc->len;
2798 
2799 			if (err)
2800 				__remove_discard_cmd(sbi, dc);
2801 
2802 			blk_finish_plug(&plug);
2803 			mutex_unlock(&dcc->cmd_lock);
2804 			trimmed += __wait_all_discard_cmd(sbi, NULL);
2805 			congestion_wait(BLK_RW_ASYNC, HZ/50);
2806 			goto next;
2807 		}
2808 skip:
2809 		node = rb_next(&dc->rb_node);
2810 		if (err)
2811 			__remove_discard_cmd(sbi, dc);
2812 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2813 
2814 		if (fatal_signal_pending(current))
2815 			break;
2816 	}
2817 
2818 	blk_finish_plug(&plug);
2819 	mutex_unlock(&dcc->cmd_lock);
2820 
2821 	return trimmed;
2822 }
2823 
2824 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2825 {
2826 	__u64 start = F2FS_BYTES_TO_BLK(range->start);
2827 	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2828 	unsigned int start_segno, end_segno;
2829 	block_t start_block, end_block;
2830 	struct cp_control cpc;
2831 	struct discard_policy dpolicy;
2832 	unsigned long long trimmed = 0;
2833 	int err = 0;
2834 	bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
2835 
2836 	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2837 		return -EINVAL;
2838 
2839 	if (end < MAIN_BLKADDR(sbi))
2840 		goto out;
2841 
2842 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2843 		f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
2844 		return -EFSCORRUPTED;
2845 	}
2846 
2847 	/* start/end segment number in main_area */
2848 	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2849 	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2850 						GET_SEGNO(sbi, end);
2851 	if (need_align) {
2852 		start_segno = rounddown(start_segno, sbi->segs_per_sec);
2853 		end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2854 	}
2855 
2856 	cpc.reason = CP_DISCARD;
2857 	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2858 	cpc.trim_start = start_segno;
2859 	cpc.trim_end = end_segno;
2860 
2861 	if (sbi->discard_blks == 0)
2862 		goto out;
2863 
2864 	mutex_lock(&sbi->gc_mutex);
2865 	err = f2fs_write_checkpoint(sbi, &cpc);
2866 	mutex_unlock(&sbi->gc_mutex);
2867 	if (err)
2868 		goto out;
2869 
2870 	/*
2871 	 * We filed discard candidates, but actually we don't need to wait for
2872 	 * all of them, since they'll be issued in idle time along with runtime
2873 	 * discard option. User configuration looks like using runtime discard
2874 	 * or periodic fstrim instead of it.
2875 	 */
2876 	if (f2fs_realtime_discard_enable(sbi))
2877 		goto out;
2878 
2879 	start_block = START_BLOCK(sbi, start_segno);
2880 	end_block = START_BLOCK(sbi, end_segno + 1);
2881 
2882 	__init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2883 	trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2884 					start_block, end_block);
2885 
2886 	trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2887 					start_block, end_block);
2888 out:
2889 	if (!err)
2890 		range->len = F2FS_BLK_TO_BYTES(trimmed);
2891 	return err;
2892 }
2893 
2894 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2895 {
2896 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2897 	if (curseg->next_blkoff < sbi->blocks_per_seg)
2898 		return true;
2899 	return false;
2900 }
2901 
2902 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2903 {
2904 	switch (hint) {
2905 	case WRITE_LIFE_SHORT:
2906 		return CURSEG_HOT_DATA;
2907 	case WRITE_LIFE_EXTREME:
2908 		return CURSEG_COLD_DATA;
2909 	default:
2910 		return CURSEG_WARM_DATA;
2911 	}
2912 }
2913 
2914 /* This returns write hints for each segment type. This hints will be
2915  * passed down to block layer. There are mapping tables which depend on
2916  * the mount option 'whint_mode'.
2917  *
2918  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2919  *
2920  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2921  *
2922  * User                  F2FS                     Block
2923  * ----                  ----                     -----
2924  *                       META                     WRITE_LIFE_NOT_SET
2925  *                       HOT_NODE                 "
2926  *                       WARM_NODE                "
2927  *                       COLD_NODE                "
2928  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2929  * extension list        "                        "
2930  *
2931  * -- buffered io
2932  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2933  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2934  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2935  * WRITE_LIFE_NONE       "                        "
2936  * WRITE_LIFE_MEDIUM     "                        "
2937  * WRITE_LIFE_LONG       "                        "
2938  *
2939  * -- direct io
2940  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2941  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2942  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2943  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2944  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2945  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2946  *
2947  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2948  *
2949  * User                  F2FS                     Block
2950  * ----                  ----                     -----
2951  *                       META                     WRITE_LIFE_MEDIUM;
2952  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2953  *                       WARM_NODE                "
2954  *                       COLD_NODE                WRITE_LIFE_NONE
2955  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2956  * extension list        "                        "
2957  *
2958  * -- buffered io
2959  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2960  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2961  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2962  * WRITE_LIFE_NONE       "                        "
2963  * WRITE_LIFE_MEDIUM     "                        "
2964  * WRITE_LIFE_LONG       "                        "
2965  *
2966  * -- direct io
2967  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2968  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2969  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2970  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2971  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2972  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2973  */
2974 
2975 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2976 				enum page_type type, enum temp_type temp)
2977 {
2978 	if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2979 		if (type == DATA) {
2980 			if (temp == WARM)
2981 				return WRITE_LIFE_NOT_SET;
2982 			else if (temp == HOT)
2983 				return WRITE_LIFE_SHORT;
2984 			else if (temp == COLD)
2985 				return WRITE_LIFE_EXTREME;
2986 		} else {
2987 			return WRITE_LIFE_NOT_SET;
2988 		}
2989 	} else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2990 		if (type == DATA) {
2991 			if (temp == WARM)
2992 				return WRITE_LIFE_LONG;
2993 			else if (temp == HOT)
2994 				return WRITE_LIFE_SHORT;
2995 			else if (temp == COLD)
2996 				return WRITE_LIFE_EXTREME;
2997 		} else if (type == NODE) {
2998 			if (temp == WARM || temp == HOT)
2999 				return WRITE_LIFE_NOT_SET;
3000 			else if (temp == COLD)
3001 				return WRITE_LIFE_NONE;
3002 		} else if (type == META) {
3003 			return WRITE_LIFE_MEDIUM;
3004 		}
3005 	}
3006 	return WRITE_LIFE_NOT_SET;
3007 }
3008 
3009 static int __get_segment_type_2(struct f2fs_io_info *fio)
3010 {
3011 	if (fio->type == DATA)
3012 		return CURSEG_HOT_DATA;
3013 	else
3014 		return CURSEG_HOT_NODE;
3015 }
3016 
3017 static int __get_segment_type_4(struct f2fs_io_info *fio)
3018 {
3019 	if (fio->type == DATA) {
3020 		struct inode *inode = fio->page->mapping->host;
3021 
3022 		if (S_ISDIR(inode->i_mode))
3023 			return CURSEG_HOT_DATA;
3024 		else
3025 			return CURSEG_COLD_DATA;
3026 	} else {
3027 		if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3028 			return CURSEG_WARM_NODE;
3029 		else
3030 			return CURSEG_COLD_NODE;
3031 	}
3032 }
3033 
3034 static int __get_segment_type_6(struct f2fs_io_info *fio)
3035 {
3036 	if (fio->type == DATA) {
3037 		struct inode *inode = fio->page->mapping->host;
3038 
3039 		if (is_cold_data(fio->page) || file_is_cold(inode))
3040 			return CURSEG_COLD_DATA;
3041 		if (file_is_hot(inode) ||
3042 				is_inode_flag_set(inode, FI_HOT_DATA) ||
3043 				f2fs_is_atomic_file(inode) ||
3044 				f2fs_is_volatile_file(inode))
3045 			return CURSEG_HOT_DATA;
3046 		return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3047 	} else {
3048 		if (IS_DNODE(fio->page))
3049 			return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3050 						CURSEG_HOT_NODE;
3051 		return CURSEG_COLD_NODE;
3052 	}
3053 }
3054 
3055 static int __get_segment_type(struct f2fs_io_info *fio)
3056 {
3057 	int type = 0;
3058 
3059 	switch (F2FS_OPTION(fio->sbi).active_logs) {
3060 	case 2:
3061 		type = __get_segment_type_2(fio);
3062 		break;
3063 	case 4:
3064 		type = __get_segment_type_4(fio);
3065 		break;
3066 	case 6:
3067 		type = __get_segment_type_6(fio);
3068 		break;
3069 	default:
3070 		f2fs_bug_on(fio->sbi, true);
3071 	}
3072 
3073 	if (IS_HOT(type))
3074 		fio->temp = HOT;
3075 	else if (IS_WARM(type))
3076 		fio->temp = WARM;
3077 	else
3078 		fio->temp = COLD;
3079 	return type;
3080 }
3081 
3082 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3083 		block_t old_blkaddr, block_t *new_blkaddr,
3084 		struct f2fs_summary *sum, int type,
3085 		struct f2fs_io_info *fio, bool add_list)
3086 {
3087 	struct sit_info *sit_i = SIT_I(sbi);
3088 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3089 	bool put_pin_sem = false;
3090 
3091 	if (type == CURSEG_COLD_DATA) {
3092 		/* GC during CURSEG_COLD_DATA_PINNED allocation */
3093 		if (down_read_trylock(&sbi->pin_sem)) {
3094 			put_pin_sem = true;
3095 		} else {
3096 			type = CURSEG_WARM_DATA;
3097 			curseg = CURSEG_I(sbi, type);
3098 		}
3099 	} else if (type == CURSEG_COLD_DATA_PINNED) {
3100 		type = CURSEG_COLD_DATA;
3101 	}
3102 
3103 	down_read(&SM_I(sbi)->curseg_lock);
3104 
3105 	mutex_lock(&curseg->curseg_mutex);
3106 	down_write(&sit_i->sentry_lock);
3107 
3108 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3109 
3110 	f2fs_wait_discard_bio(sbi, *new_blkaddr);
3111 
3112 	/*
3113 	 * __add_sum_entry should be resided under the curseg_mutex
3114 	 * because, this function updates a summary entry in the
3115 	 * current summary block.
3116 	 */
3117 	__add_sum_entry(sbi, type, sum);
3118 
3119 	__refresh_next_blkoff(sbi, curseg);
3120 
3121 	stat_inc_block_count(sbi, curseg);
3122 
3123 	/*
3124 	 * SIT information should be updated before segment allocation,
3125 	 * since SSR needs latest valid block information.
3126 	 */
3127 	update_sit_entry(sbi, *new_blkaddr, 1);
3128 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3129 		update_sit_entry(sbi, old_blkaddr, -1);
3130 
3131 	if (!__has_curseg_space(sbi, type))
3132 		sit_i->s_ops->allocate_segment(sbi, type, false);
3133 
3134 	/*
3135 	 * segment dirty status should be updated after segment allocation,
3136 	 * so we just need to update status only one time after previous
3137 	 * segment being closed.
3138 	 */
3139 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3140 	locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3141 
3142 	up_write(&sit_i->sentry_lock);
3143 
3144 	if (page && IS_NODESEG(type)) {
3145 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3146 
3147 		f2fs_inode_chksum_set(sbi, page);
3148 	}
3149 
3150 	if (F2FS_IO_ALIGNED(sbi))
3151 		fio->retry = false;
3152 
3153 	if (add_list) {
3154 		struct f2fs_bio_info *io;
3155 
3156 		INIT_LIST_HEAD(&fio->list);
3157 		fio->in_list = true;
3158 		io = sbi->write_io[fio->type] + fio->temp;
3159 		spin_lock(&io->io_lock);
3160 		list_add_tail(&fio->list, &io->io_list);
3161 		spin_unlock(&io->io_lock);
3162 	}
3163 
3164 	mutex_unlock(&curseg->curseg_mutex);
3165 
3166 	up_read(&SM_I(sbi)->curseg_lock);
3167 
3168 	if (put_pin_sem)
3169 		up_read(&sbi->pin_sem);
3170 }
3171 
3172 static void update_device_state(struct f2fs_io_info *fio)
3173 {
3174 	struct f2fs_sb_info *sbi = fio->sbi;
3175 	unsigned int devidx;
3176 
3177 	if (!f2fs_is_multi_device(sbi))
3178 		return;
3179 
3180 	devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3181 
3182 	/* update device state for fsync */
3183 	f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3184 
3185 	/* update device state for checkpoint */
3186 	if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3187 		spin_lock(&sbi->dev_lock);
3188 		f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3189 		spin_unlock(&sbi->dev_lock);
3190 	}
3191 }
3192 
3193 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3194 {
3195 	int type = __get_segment_type(fio);
3196 	bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
3197 
3198 	if (keep_order)
3199 		down_read(&fio->sbi->io_order_lock);
3200 reallocate:
3201 	f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3202 			&fio->new_blkaddr, sum, type, fio, true);
3203 	if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3204 		invalidate_mapping_pages(META_MAPPING(fio->sbi),
3205 					fio->old_blkaddr, fio->old_blkaddr);
3206 
3207 	/* writeout dirty page into bdev */
3208 	f2fs_submit_page_write(fio);
3209 	if (fio->retry) {
3210 		fio->old_blkaddr = fio->new_blkaddr;
3211 		goto reallocate;
3212 	}
3213 
3214 	update_device_state(fio);
3215 
3216 	if (keep_order)
3217 		up_read(&fio->sbi->io_order_lock);
3218 }
3219 
3220 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3221 					enum iostat_type io_type)
3222 {
3223 	struct f2fs_io_info fio = {
3224 		.sbi = sbi,
3225 		.type = META,
3226 		.temp = HOT,
3227 		.op = REQ_OP_WRITE,
3228 		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3229 		.old_blkaddr = page->index,
3230 		.new_blkaddr = page->index,
3231 		.page = page,
3232 		.encrypted_page = NULL,
3233 		.in_list = false,
3234 	};
3235 
3236 	if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3237 		fio.op_flags &= ~REQ_META;
3238 
3239 	set_page_writeback(page);
3240 	ClearPageError(page);
3241 	f2fs_submit_page_write(&fio);
3242 
3243 	stat_inc_meta_count(sbi, page->index);
3244 	f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3245 }
3246 
3247 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3248 {
3249 	struct f2fs_summary sum;
3250 
3251 	set_summary(&sum, nid, 0, 0);
3252 	do_write_page(&sum, fio);
3253 
3254 	f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3255 }
3256 
3257 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3258 					struct f2fs_io_info *fio)
3259 {
3260 	struct f2fs_sb_info *sbi = fio->sbi;
3261 	struct f2fs_summary sum;
3262 
3263 	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3264 	set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3265 	do_write_page(&sum, fio);
3266 	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3267 
3268 	f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3269 }
3270 
3271 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3272 {
3273 	int err;
3274 	struct f2fs_sb_info *sbi = fio->sbi;
3275 	unsigned int segno;
3276 
3277 	fio->new_blkaddr = fio->old_blkaddr;
3278 	/* i/o temperature is needed for passing down write hints */
3279 	__get_segment_type(fio);
3280 
3281 	segno = GET_SEGNO(sbi, fio->new_blkaddr);
3282 
3283 	if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3284 		set_sbi_flag(sbi, SBI_NEED_FSCK);
3285 		f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3286 			  __func__, segno);
3287 		return -EFSCORRUPTED;
3288 	}
3289 
3290 	stat_inc_inplace_blocks(fio->sbi);
3291 
3292 	if (fio->bio)
3293 		err = f2fs_merge_page_bio(fio);
3294 	else
3295 		err = f2fs_submit_page_bio(fio);
3296 	if (!err) {
3297 		update_device_state(fio);
3298 		f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3299 	}
3300 
3301 	return err;
3302 }
3303 
3304 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3305 						unsigned int segno)
3306 {
3307 	int i;
3308 
3309 	for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3310 		if (CURSEG_I(sbi, i)->segno == segno)
3311 			break;
3312 	}
3313 	return i;
3314 }
3315 
3316 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3317 				block_t old_blkaddr, block_t new_blkaddr,
3318 				bool recover_curseg, bool recover_newaddr)
3319 {
3320 	struct sit_info *sit_i = SIT_I(sbi);
3321 	struct curseg_info *curseg;
3322 	unsigned int segno, old_cursegno;
3323 	struct seg_entry *se;
3324 	int type;
3325 	unsigned short old_blkoff;
3326 
3327 	segno = GET_SEGNO(sbi, new_blkaddr);
3328 	se = get_seg_entry(sbi, segno);
3329 	type = se->type;
3330 
3331 	down_write(&SM_I(sbi)->curseg_lock);
3332 
3333 	if (!recover_curseg) {
3334 		/* for recovery flow */
3335 		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3336 			if (old_blkaddr == NULL_ADDR)
3337 				type = CURSEG_COLD_DATA;
3338 			else
3339 				type = CURSEG_WARM_DATA;
3340 		}
3341 	} else {
3342 		if (IS_CURSEG(sbi, segno)) {
3343 			/* se->type is volatile as SSR allocation */
3344 			type = __f2fs_get_curseg(sbi, segno);
3345 			f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3346 		} else {
3347 			type = CURSEG_WARM_DATA;
3348 		}
3349 	}
3350 
3351 	f2fs_bug_on(sbi, !IS_DATASEG(type));
3352 	curseg = CURSEG_I(sbi, type);
3353 
3354 	mutex_lock(&curseg->curseg_mutex);
3355 	down_write(&sit_i->sentry_lock);
3356 
3357 	old_cursegno = curseg->segno;
3358 	old_blkoff = curseg->next_blkoff;
3359 
3360 	/* change the current segment */
3361 	if (segno != curseg->segno) {
3362 		curseg->next_segno = segno;
3363 		change_curseg(sbi, type);
3364 	}
3365 
3366 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3367 	__add_sum_entry(sbi, type, sum);
3368 
3369 	if (!recover_curseg || recover_newaddr)
3370 		update_sit_entry(sbi, new_blkaddr, 1);
3371 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3372 		invalidate_mapping_pages(META_MAPPING(sbi),
3373 					old_blkaddr, old_blkaddr);
3374 		update_sit_entry(sbi, old_blkaddr, -1);
3375 	}
3376 
3377 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3378 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3379 
3380 	locate_dirty_segment(sbi, old_cursegno);
3381 
3382 	if (recover_curseg) {
3383 		if (old_cursegno != curseg->segno) {
3384 			curseg->next_segno = old_cursegno;
3385 			change_curseg(sbi, type);
3386 		}
3387 		curseg->next_blkoff = old_blkoff;
3388 	}
3389 
3390 	up_write(&sit_i->sentry_lock);
3391 	mutex_unlock(&curseg->curseg_mutex);
3392 	up_write(&SM_I(sbi)->curseg_lock);
3393 }
3394 
3395 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3396 				block_t old_addr, block_t new_addr,
3397 				unsigned char version, bool recover_curseg,
3398 				bool recover_newaddr)
3399 {
3400 	struct f2fs_summary sum;
3401 
3402 	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3403 
3404 	f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3405 					recover_curseg, recover_newaddr);
3406 
3407 	f2fs_update_data_blkaddr(dn, new_addr);
3408 }
3409 
3410 void f2fs_wait_on_page_writeback(struct page *page,
3411 				enum page_type type, bool ordered, bool locked)
3412 {
3413 	if (PageWriteback(page)) {
3414 		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3415 
3416 		/* submit cached LFS IO */
3417 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3418 		/* sbumit cached IPU IO */
3419 		f2fs_submit_merged_ipu_write(sbi, NULL, page);
3420 		if (ordered) {
3421 			wait_on_page_writeback(page);
3422 			f2fs_bug_on(sbi, locked && PageWriteback(page));
3423 		} else {
3424 			wait_for_stable_page(page);
3425 		}
3426 	}
3427 }
3428 
3429 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3430 {
3431 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3432 	struct page *cpage;
3433 
3434 	if (!f2fs_post_read_required(inode))
3435 		return;
3436 
3437 	if (!__is_valid_data_blkaddr(blkaddr))
3438 		return;
3439 
3440 	cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3441 	if (cpage) {
3442 		f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3443 		f2fs_put_page(cpage, 1);
3444 	}
3445 }
3446 
3447 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3448 								block_t len)
3449 {
3450 	block_t i;
3451 
3452 	for (i = 0; i < len; i++)
3453 		f2fs_wait_on_block_writeback(inode, blkaddr + i);
3454 }
3455 
3456 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3457 {
3458 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3459 	struct curseg_info *seg_i;
3460 	unsigned char *kaddr;
3461 	struct page *page;
3462 	block_t start;
3463 	int i, j, offset;
3464 
3465 	start = start_sum_block(sbi);
3466 
3467 	page = f2fs_get_meta_page(sbi, start++);
3468 	if (IS_ERR(page))
3469 		return PTR_ERR(page);
3470 	kaddr = (unsigned char *)page_address(page);
3471 
3472 	/* Step 1: restore nat cache */
3473 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3474 	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3475 
3476 	/* Step 2: restore sit cache */
3477 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3478 	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3479 	offset = 2 * SUM_JOURNAL_SIZE;
3480 
3481 	/* Step 3: restore summary entries */
3482 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3483 		unsigned short blk_off;
3484 		unsigned int segno;
3485 
3486 		seg_i = CURSEG_I(sbi, i);
3487 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3488 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3489 		seg_i->next_segno = segno;
3490 		reset_curseg(sbi, i, 0);
3491 		seg_i->alloc_type = ckpt->alloc_type[i];
3492 		seg_i->next_blkoff = blk_off;
3493 
3494 		if (seg_i->alloc_type == SSR)
3495 			blk_off = sbi->blocks_per_seg;
3496 
3497 		for (j = 0; j < blk_off; j++) {
3498 			struct f2fs_summary *s;
3499 			s = (struct f2fs_summary *)(kaddr + offset);
3500 			seg_i->sum_blk->entries[j] = *s;
3501 			offset += SUMMARY_SIZE;
3502 			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3503 						SUM_FOOTER_SIZE)
3504 				continue;
3505 
3506 			f2fs_put_page(page, 1);
3507 			page = NULL;
3508 
3509 			page = f2fs_get_meta_page(sbi, start++);
3510 			if (IS_ERR(page))
3511 				return PTR_ERR(page);
3512 			kaddr = (unsigned char *)page_address(page);
3513 			offset = 0;
3514 		}
3515 	}
3516 	f2fs_put_page(page, 1);
3517 	return 0;
3518 }
3519 
3520 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3521 {
3522 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3523 	struct f2fs_summary_block *sum;
3524 	struct curseg_info *curseg;
3525 	struct page *new;
3526 	unsigned short blk_off;
3527 	unsigned int segno = 0;
3528 	block_t blk_addr = 0;
3529 	int err = 0;
3530 
3531 	/* get segment number and block addr */
3532 	if (IS_DATASEG(type)) {
3533 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3534 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3535 							CURSEG_HOT_DATA]);
3536 		if (__exist_node_summaries(sbi))
3537 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3538 		else
3539 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3540 	} else {
3541 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
3542 							CURSEG_HOT_NODE]);
3543 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3544 							CURSEG_HOT_NODE]);
3545 		if (__exist_node_summaries(sbi))
3546 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3547 							type - CURSEG_HOT_NODE);
3548 		else
3549 			blk_addr = GET_SUM_BLOCK(sbi, segno);
3550 	}
3551 
3552 	new = f2fs_get_meta_page(sbi, blk_addr);
3553 	if (IS_ERR(new))
3554 		return PTR_ERR(new);
3555 	sum = (struct f2fs_summary_block *)page_address(new);
3556 
3557 	if (IS_NODESEG(type)) {
3558 		if (__exist_node_summaries(sbi)) {
3559 			struct f2fs_summary *ns = &sum->entries[0];
3560 			int i;
3561 			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3562 				ns->version = 0;
3563 				ns->ofs_in_node = 0;
3564 			}
3565 		} else {
3566 			err = f2fs_restore_node_summary(sbi, segno, sum);
3567 			if (err)
3568 				goto out;
3569 		}
3570 	}
3571 
3572 	/* set uncompleted segment to curseg */
3573 	curseg = CURSEG_I(sbi, type);
3574 	mutex_lock(&curseg->curseg_mutex);
3575 
3576 	/* update journal info */
3577 	down_write(&curseg->journal_rwsem);
3578 	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3579 	up_write(&curseg->journal_rwsem);
3580 
3581 	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3582 	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3583 	curseg->next_segno = segno;
3584 	reset_curseg(sbi, type, 0);
3585 	curseg->alloc_type = ckpt->alloc_type[type];
3586 	curseg->next_blkoff = blk_off;
3587 	mutex_unlock(&curseg->curseg_mutex);
3588 out:
3589 	f2fs_put_page(new, 1);
3590 	return err;
3591 }
3592 
3593 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3594 {
3595 	struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3596 	struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3597 	int type = CURSEG_HOT_DATA;
3598 	int err;
3599 
3600 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3601 		int npages = f2fs_npages_for_summary_flush(sbi, true);
3602 
3603 		if (npages >= 2)
3604 			f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3605 							META_CP, true);
3606 
3607 		/* restore for compacted data summary */
3608 		err = read_compacted_summaries(sbi);
3609 		if (err)
3610 			return err;
3611 		type = CURSEG_HOT_NODE;
3612 	}
3613 
3614 	if (__exist_node_summaries(sbi))
3615 		f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3616 					NR_CURSEG_TYPE - type, META_CP, true);
3617 
3618 	for (; type <= CURSEG_COLD_NODE; type++) {
3619 		err = read_normal_summaries(sbi, type);
3620 		if (err)
3621 			return err;
3622 	}
3623 
3624 	/* sanity check for summary blocks */
3625 	if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3626 			sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3627 		f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3628 			 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3629 		return -EINVAL;
3630 	}
3631 
3632 	return 0;
3633 }
3634 
3635 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3636 {
3637 	struct page *page;
3638 	unsigned char *kaddr;
3639 	struct f2fs_summary *summary;
3640 	struct curseg_info *seg_i;
3641 	int written_size = 0;
3642 	int i, j;
3643 
3644 	page = f2fs_grab_meta_page(sbi, blkaddr++);
3645 	kaddr = (unsigned char *)page_address(page);
3646 	memset(kaddr, 0, PAGE_SIZE);
3647 
3648 	/* Step 1: write nat cache */
3649 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3650 	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3651 	written_size += SUM_JOURNAL_SIZE;
3652 
3653 	/* Step 2: write sit cache */
3654 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3655 	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3656 	written_size += SUM_JOURNAL_SIZE;
3657 
3658 	/* Step 3: write summary entries */
3659 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3660 		unsigned short blkoff;
3661 		seg_i = CURSEG_I(sbi, i);
3662 		if (sbi->ckpt->alloc_type[i] == SSR)
3663 			blkoff = sbi->blocks_per_seg;
3664 		else
3665 			blkoff = curseg_blkoff(sbi, i);
3666 
3667 		for (j = 0; j < blkoff; j++) {
3668 			if (!page) {
3669 				page = f2fs_grab_meta_page(sbi, blkaddr++);
3670 				kaddr = (unsigned char *)page_address(page);
3671 				memset(kaddr, 0, PAGE_SIZE);
3672 				written_size = 0;
3673 			}
3674 			summary = (struct f2fs_summary *)(kaddr + written_size);
3675 			*summary = seg_i->sum_blk->entries[j];
3676 			written_size += SUMMARY_SIZE;
3677 
3678 			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3679 							SUM_FOOTER_SIZE)
3680 				continue;
3681 
3682 			set_page_dirty(page);
3683 			f2fs_put_page(page, 1);
3684 			page = NULL;
3685 		}
3686 	}
3687 	if (page) {
3688 		set_page_dirty(page);
3689 		f2fs_put_page(page, 1);
3690 	}
3691 }
3692 
3693 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3694 					block_t blkaddr, int type)
3695 {
3696 	int i, end;
3697 	if (IS_DATASEG(type))
3698 		end = type + NR_CURSEG_DATA_TYPE;
3699 	else
3700 		end = type + NR_CURSEG_NODE_TYPE;
3701 
3702 	for (i = type; i < end; i++)
3703 		write_current_sum_page(sbi, i, blkaddr + (i - type));
3704 }
3705 
3706 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3707 {
3708 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3709 		write_compacted_summaries(sbi, start_blk);
3710 	else
3711 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3712 }
3713 
3714 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3715 {
3716 	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3717 }
3718 
3719 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3720 					unsigned int val, int alloc)
3721 {
3722 	int i;
3723 
3724 	if (type == NAT_JOURNAL) {
3725 		for (i = 0; i < nats_in_cursum(journal); i++) {
3726 			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3727 				return i;
3728 		}
3729 		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3730 			return update_nats_in_cursum(journal, 1);
3731 	} else if (type == SIT_JOURNAL) {
3732 		for (i = 0; i < sits_in_cursum(journal); i++)
3733 			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3734 				return i;
3735 		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3736 			return update_sits_in_cursum(journal, 1);
3737 	}
3738 	return -1;
3739 }
3740 
3741 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3742 					unsigned int segno)
3743 {
3744 	return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3745 }
3746 
3747 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3748 					unsigned int start)
3749 {
3750 	struct sit_info *sit_i = SIT_I(sbi);
3751 	struct page *page;
3752 	pgoff_t src_off, dst_off;
3753 
3754 	src_off = current_sit_addr(sbi, start);
3755 	dst_off = next_sit_addr(sbi, src_off);
3756 
3757 	page = f2fs_grab_meta_page(sbi, dst_off);
3758 	seg_info_to_sit_page(sbi, page, start);
3759 
3760 	set_page_dirty(page);
3761 	set_to_next_sit(sit_i, start);
3762 
3763 	return page;
3764 }
3765 
3766 static struct sit_entry_set *grab_sit_entry_set(void)
3767 {
3768 	struct sit_entry_set *ses =
3769 			f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3770 
3771 	ses->entry_cnt = 0;
3772 	INIT_LIST_HEAD(&ses->set_list);
3773 	return ses;
3774 }
3775 
3776 static void release_sit_entry_set(struct sit_entry_set *ses)
3777 {
3778 	list_del(&ses->set_list);
3779 	kmem_cache_free(sit_entry_set_slab, ses);
3780 }
3781 
3782 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3783 						struct list_head *head)
3784 {
3785 	struct sit_entry_set *next = ses;
3786 
3787 	if (list_is_last(&ses->set_list, head))
3788 		return;
3789 
3790 	list_for_each_entry_continue(next, head, set_list)
3791 		if (ses->entry_cnt <= next->entry_cnt)
3792 			break;
3793 
3794 	list_move_tail(&ses->set_list, &next->set_list);
3795 }
3796 
3797 static void add_sit_entry(unsigned int segno, struct list_head *head)
3798 {
3799 	struct sit_entry_set *ses;
3800 	unsigned int start_segno = START_SEGNO(segno);
3801 
3802 	list_for_each_entry(ses, head, set_list) {
3803 		if (ses->start_segno == start_segno) {
3804 			ses->entry_cnt++;
3805 			adjust_sit_entry_set(ses, head);
3806 			return;
3807 		}
3808 	}
3809 
3810 	ses = grab_sit_entry_set();
3811 
3812 	ses->start_segno = start_segno;
3813 	ses->entry_cnt++;
3814 	list_add(&ses->set_list, head);
3815 }
3816 
3817 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3818 {
3819 	struct f2fs_sm_info *sm_info = SM_I(sbi);
3820 	struct list_head *set_list = &sm_info->sit_entry_set;
3821 	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3822 	unsigned int segno;
3823 
3824 	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3825 		add_sit_entry(segno, set_list);
3826 }
3827 
3828 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3829 {
3830 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3831 	struct f2fs_journal *journal = curseg->journal;
3832 	int i;
3833 
3834 	down_write(&curseg->journal_rwsem);
3835 	for (i = 0; i < sits_in_cursum(journal); i++) {
3836 		unsigned int segno;
3837 		bool dirtied;
3838 
3839 		segno = le32_to_cpu(segno_in_journal(journal, i));
3840 		dirtied = __mark_sit_entry_dirty(sbi, segno);
3841 
3842 		if (!dirtied)
3843 			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3844 	}
3845 	update_sits_in_cursum(journal, -i);
3846 	up_write(&curseg->journal_rwsem);
3847 }
3848 
3849 /*
3850  * CP calls this function, which flushes SIT entries including sit_journal,
3851  * and moves prefree segs to free segs.
3852  */
3853 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3854 {
3855 	struct sit_info *sit_i = SIT_I(sbi);
3856 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3857 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3858 	struct f2fs_journal *journal = curseg->journal;
3859 	struct sit_entry_set *ses, *tmp;
3860 	struct list_head *head = &SM_I(sbi)->sit_entry_set;
3861 	bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
3862 	struct seg_entry *se;
3863 
3864 	down_write(&sit_i->sentry_lock);
3865 
3866 	if (!sit_i->dirty_sentries)
3867 		goto out;
3868 
3869 	/*
3870 	 * add and account sit entries of dirty bitmap in sit entry
3871 	 * set temporarily
3872 	 */
3873 	add_sits_in_set(sbi);
3874 
3875 	/*
3876 	 * if there are no enough space in journal to store dirty sit
3877 	 * entries, remove all entries from journal and add and account
3878 	 * them in sit entry set.
3879 	 */
3880 	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
3881 								!to_journal)
3882 		remove_sits_in_journal(sbi);
3883 
3884 	/*
3885 	 * there are two steps to flush sit entries:
3886 	 * #1, flush sit entries to journal in current cold data summary block.
3887 	 * #2, flush sit entries to sit page.
3888 	 */
3889 	list_for_each_entry_safe(ses, tmp, head, set_list) {
3890 		struct page *page = NULL;
3891 		struct f2fs_sit_block *raw_sit = NULL;
3892 		unsigned int start_segno = ses->start_segno;
3893 		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3894 						(unsigned long)MAIN_SEGS(sbi));
3895 		unsigned int segno = start_segno;
3896 
3897 		if (to_journal &&
3898 			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3899 			to_journal = false;
3900 
3901 		if (to_journal) {
3902 			down_write(&curseg->journal_rwsem);
3903 		} else {
3904 			page = get_next_sit_page(sbi, start_segno);
3905 			raw_sit = page_address(page);
3906 		}
3907 
3908 		/* flush dirty sit entries in region of current sit set */
3909 		for_each_set_bit_from(segno, bitmap, end) {
3910 			int offset, sit_offset;
3911 
3912 			se = get_seg_entry(sbi, segno);
3913 #ifdef CONFIG_F2FS_CHECK_FS
3914 			if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3915 						SIT_VBLOCK_MAP_SIZE))
3916 				f2fs_bug_on(sbi, 1);
3917 #endif
3918 
3919 			/* add discard candidates */
3920 			if (!(cpc->reason & CP_DISCARD)) {
3921 				cpc->trim_start = segno;
3922 				add_discard_addrs(sbi, cpc, false);
3923 			}
3924 
3925 			if (to_journal) {
3926 				offset = f2fs_lookup_journal_in_cursum(journal,
3927 							SIT_JOURNAL, segno, 1);
3928 				f2fs_bug_on(sbi, offset < 0);
3929 				segno_in_journal(journal, offset) =
3930 							cpu_to_le32(segno);
3931 				seg_info_to_raw_sit(se,
3932 					&sit_in_journal(journal, offset));
3933 				check_block_count(sbi, segno,
3934 					&sit_in_journal(journal, offset));
3935 			} else {
3936 				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3937 				seg_info_to_raw_sit(se,
3938 						&raw_sit->entries[sit_offset]);
3939 				check_block_count(sbi, segno,
3940 						&raw_sit->entries[sit_offset]);
3941 			}
3942 
3943 			__clear_bit(segno, bitmap);
3944 			sit_i->dirty_sentries--;
3945 			ses->entry_cnt--;
3946 		}
3947 
3948 		if (to_journal)
3949 			up_write(&curseg->journal_rwsem);
3950 		else
3951 			f2fs_put_page(page, 1);
3952 
3953 		f2fs_bug_on(sbi, ses->entry_cnt);
3954 		release_sit_entry_set(ses);
3955 	}
3956 
3957 	f2fs_bug_on(sbi, !list_empty(head));
3958 	f2fs_bug_on(sbi, sit_i->dirty_sentries);
3959 out:
3960 	if (cpc->reason & CP_DISCARD) {
3961 		__u64 trim_start = cpc->trim_start;
3962 
3963 		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3964 			add_discard_addrs(sbi, cpc, false);
3965 
3966 		cpc->trim_start = trim_start;
3967 	}
3968 	up_write(&sit_i->sentry_lock);
3969 
3970 	set_prefree_as_free_segments(sbi);
3971 }
3972 
3973 static int build_sit_info(struct f2fs_sb_info *sbi)
3974 {
3975 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3976 	struct sit_info *sit_i;
3977 	unsigned int sit_segs, start;
3978 	char *src_bitmap, *bitmap;
3979 	unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
3980 
3981 	/* allocate memory for SIT information */
3982 	sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3983 	if (!sit_i)
3984 		return -ENOMEM;
3985 
3986 	SM_I(sbi)->sit_info = sit_i;
3987 
3988 	sit_i->sentries =
3989 		f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3990 					      MAIN_SEGS(sbi)),
3991 			      GFP_KERNEL);
3992 	if (!sit_i->sentries)
3993 		return -ENOMEM;
3994 
3995 	main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3996 	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
3997 								GFP_KERNEL);
3998 	if (!sit_i->dirty_sentries_bitmap)
3999 		return -ENOMEM;
4000 
4001 #ifdef CONFIG_F2FS_CHECK_FS
4002 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 4;
4003 #else
4004 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 3;
4005 #endif
4006 	sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4007 	if (!sit_i->bitmap)
4008 		return -ENOMEM;
4009 
4010 	bitmap = sit_i->bitmap;
4011 
4012 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4013 		sit_i->sentries[start].cur_valid_map = bitmap;
4014 		bitmap += SIT_VBLOCK_MAP_SIZE;
4015 
4016 		sit_i->sentries[start].ckpt_valid_map = bitmap;
4017 		bitmap += SIT_VBLOCK_MAP_SIZE;
4018 
4019 #ifdef CONFIG_F2FS_CHECK_FS
4020 		sit_i->sentries[start].cur_valid_map_mir = bitmap;
4021 		bitmap += SIT_VBLOCK_MAP_SIZE;
4022 #endif
4023 
4024 		sit_i->sentries[start].discard_map = bitmap;
4025 		bitmap += SIT_VBLOCK_MAP_SIZE;
4026 	}
4027 
4028 	sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4029 	if (!sit_i->tmp_map)
4030 		return -ENOMEM;
4031 
4032 	if (__is_large_section(sbi)) {
4033 		sit_i->sec_entries =
4034 			f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4035 						      MAIN_SECS(sbi)),
4036 				      GFP_KERNEL);
4037 		if (!sit_i->sec_entries)
4038 			return -ENOMEM;
4039 	}
4040 
4041 	/* get information related with SIT */
4042 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4043 
4044 	/* setup SIT bitmap from ckeckpoint pack */
4045 	sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4046 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4047 
4048 	sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4049 	if (!sit_i->sit_bitmap)
4050 		return -ENOMEM;
4051 
4052 #ifdef CONFIG_F2FS_CHECK_FS
4053 	sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4054 					sit_bitmap_size, GFP_KERNEL);
4055 	if (!sit_i->sit_bitmap_mir)
4056 		return -ENOMEM;
4057 
4058 	sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4059 					main_bitmap_size, GFP_KERNEL);
4060 	if (!sit_i->invalid_segmap)
4061 		return -ENOMEM;
4062 #endif
4063 
4064 	/* init SIT information */
4065 	sit_i->s_ops = &default_salloc_ops;
4066 
4067 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4068 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4069 	sit_i->written_valid_blocks = 0;
4070 	sit_i->bitmap_size = sit_bitmap_size;
4071 	sit_i->dirty_sentries = 0;
4072 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4073 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4074 	sit_i->mounted_time = ktime_get_real_seconds();
4075 	init_rwsem(&sit_i->sentry_lock);
4076 	return 0;
4077 }
4078 
4079 static int build_free_segmap(struct f2fs_sb_info *sbi)
4080 {
4081 	struct free_segmap_info *free_i;
4082 	unsigned int bitmap_size, sec_bitmap_size;
4083 
4084 	/* allocate memory for free segmap information */
4085 	free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4086 	if (!free_i)
4087 		return -ENOMEM;
4088 
4089 	SM_I(sbi)->free_info = free_i;
4090 
4091 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4092 	free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4093 	if (!free_i->free_segmap)
4094 		return -ENOMEM;
4095 
4096 	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4097 	free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4098 	if (!free_i->free_secmap)
4099 		return -ENOMEM;
4100 
4101 	/* set all segments as dirty temporarily */
4102 	memset(free_i->free_segmap, 0xff, bitmap_size);
4103 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4104 
4105 	/* init free segmap information */
4106 	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4107 	free_i->free_segments = 0;
4108 	free_i->free_sections = 0;
4109 	spin_lock_init(&free_i->segmap_lock);
4110 	return 0;
4111 }
4112 
4113 static int build_curseg(struct f2fs_sb_info *sbi)
4114 {
4115 	struct curseg_info *array;
4116 	int i;
4117 
4118 	array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
4119 			     GFP_KERNEL);
4120 	if (!array)
4121 		return -ENOMEM;
4122 
4123 	SM_I(sbi)->curseg_array = array;
4124 
4125 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
4126 		mutex_init(&array[i].curseg_mutex);
4127 		array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4128 		if (!array[i].sum_blk)
4129 			return -ENOMEM;
4130 		init_rwsem(&array[i].journal_rwsem);
4131 		array[i].journal = f2fs_kzalloc(sbi,
4132 				sizeof(struct f2fs_journal), GFP_KERNEL);
4133 		if (!array[i].journal)
4134 			return -ENOMEM;
4135 		array[i].segno = NULL_SEGNO;
4136 		array[i].next_blkoff = 0;
4137 	}
4138 	return restore_curseg_summaries(sbi);
4139 }
4140 
4141 static int build_sit_entries(struct f2fs_sb_info *sbi)
4142 {
4143 	struct sit_info *sit_i = SIT_I(sbi);
4144 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4145 	struct f2fs_journal *journal = curseg->journal;
4146 	struct seg_entry *se;
4147 	struct f2fs_sit_entry sit;
4148 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
4149 	unsigned int i, start, end;
4150 	unsigned int readed, start_blk = 0;
4151 	int err = 0;
4152 	block_t total_node_blocks = 0;
4153 
4154 	do {
4155 		readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4156 							META_SIT, true);
4157 
4158 		start = start_blk * sit_i->sents_per_block;
4159 		end = (start_blk + readed) * sit_i->sents_per_block;
4160 
4161 		for (; start < end && start < MAIN_SEGS(sbi); start++) {
4162 			struct f2fs_sit_block *sit_blk;
4163 			struct page *page;
4164 
4165 			se = &sit_i->sentries[start];
4166 			page = get_current_sit_page(sbi, start);
4167 			if (IS_ERR(page))
4168 				return PTR_ERR(page);
4169 			sit_blk = (struct f2fs_sit_block *)page_address(page);
4170 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4171 			f2fs_put_page(page, 1);
4172 
4173 			err = check_block_count(sbi, start, &sit);
4174 			if (err)
4175 				return err;
4176 			seg_info_from_raw_sit(se, &sit);
4177 			if (IS_NODESEG(se->type))
4178 				total_node_blocks += se->valid_blocks;
4179 
4180 			/* build discard map only one time */
4181 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4182 				memset(se->discard_map, 0xff,
4183 					SIT_VBLOCK_MAP_SIZE);
4184 			} else {
4185 				memcpy(se->discard_map,
4186 					se->cur_valid_map,
4187 					SIT_VBLOCK_MAP_SIZE);
4188 				sbi->discard_blks +=
4189 					sbi->blocks_per_seg -
4190 					se->valid_blocks;
4191 			}
4192 
4193 			if (__is_large_section(sbi))
4194 				get_sec_entry(sbi, start)->valid_blocks +=
4195 							se->valid_blocks;
4196 		}
4197 		start_blk += readed;
4198 	} while (start_blk < sit_blk_cnt);
4199 
4200 	down_read(&curseg->journal_rwsem);
4201 	for (i = 0; i < sits_in_cursum(journal); i++) {
4202 		unsigned int old_valid_blocks;
4203 
4204 		start = le32_to_cpu(segno_in_journal(journal, i));
4205 		if (start >= MAIN_SEGS(sbi)) {
4206 			f2fs_err(sbi, "Wrong journal entry on segno %u",
4207 				 start);
4208 			err = -EFSCORRUPTED;
4209 			break;
4210 		}
4211 
4212 		se = &sit_i->sentries[start];
4213 		sit = sit_in_journal(journal, i);
4214 
4215 		old_valid_blocks = se->valid_blocks;
4216 		if (IS_NODESEG(se->type))
4217 			total_node_blocks -= old_valid_blocks;
4218 
4219 		err = check_block_count(sbi, start, &sit);
4220 		if (err)
4221 			break;
4222 		seg_info_from_raw_sit(se, &sit);
4223 		if (IS_NODESEG(se->type))
4224 			total_node_blocks += se->valid_blocks;
4225 
4226 		if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4227 			memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4228 		} else {
4229 			memcpy(se->discard_map, se->cur_valid_map,
4230 						SIT_VBLOCK_MAP_SIZE);
4231 			sbi->discard_blks += old_valid_blocks;
4232 			sbi->discard_blks -= se->valid_blocks;
4233 		}
4234 
4235 		if (__is_large_section(sbi)) {
4236 			get_sec_entry(sbi, start)->valid_blocks +=
4237 							se->valid_blocks;
4238 			get_sec_entry(sbi, start)->valid_blocks -=
4239 							old_valid_blocks;
4240 		}
4241 	}
4242 	up_read(&curseg->journal_rwsem);
4243 
4244 	if (!err && total_node_blocks != valid_node_count(sbi)) {
4245 		f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4246 			 total_node_blocks, valid_node_count(sbi));
4247 		err = -EFSCORRUPTED;
4248 	}
4249 
4250 	return err;
4251 }
4252 
4253 static void init_free_segmap(struct f2fs_sb_info *sbi)
4254 {
4255 	unsigned int start;
4256 	int type;
4257 
4258 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4259 		struct seg_entry *sentry = get_seg_entry(sbi, start);
4260 		if (!sentry->valid_blocks)
4261 			__set_free(sbi, start);
4262 		else
4263 			SIT_I(sbi)->written_valid_blocks +=
4264 						sentry->valid_blocks;
4265 	}
4266 
4267 	/* set use the current segments */
4268 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4269 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4270 		__set_test_and_inuse(sbi, curseg_t->segno);
4271 	}
4272 }
4273 
4274 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4275 {
4276 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4277 	struct free_segmap_info *free_i = FREE_I(sbi);
4278 	unsigned int segno = 0, offset = 0;
4279 	unsigned short valid_blocks;
4280 
4281 	while (1) {
4282 		/* find dirty segment based on free segmap */
4283 		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4284 		if (segno >= MAIN_SEGS(sbi))
4285 			break;
4286 		offset = segno + 1;
4287 		valid_blocks = get_valid_blocks(sbi, segno, false);
4288 		if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4289 			continue;
4290 		if (valid_blocks > sbi->blocks_per_seg) {
4291 			f2fs_bug_on(sbi, 1);
4292 			continue;
4293 		}
4294 		mutex_lock(&dirty_i->seglist_lock);
4295 		__locate_dirty_segment(sbi, segno, DIRTY);
4296 		mutex_unlock(&dirty_i->seglist_lock);
4297 	}
4298 }
4299 
4300 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4301 {
4302 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4303 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4304 
4305 	dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4306 	if (!dirty_i->victim_secmap)
4307 		return -ENOMEM;
4308 	return 0;
4309 }
4310 
4311 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4312 {
4313 	struct dirty_seglist_info *dirty_i;
4314 	unsigned int bitmap_size, i;
4315 
4316 	/* allocate memory for dirty segments list information */
4317 	dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4318 								GFP_KERNEL);
4319 	if (!dirty_i)
4320 		return -ENOMEM;
4321 
4322 	SM_I(sbi)->dirty_info = dirty_i;
4323 	mutex_init(&dirty_i->seglist_lock);
4324 
4325 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4326 
4327 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
4328 		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4329 								GFP_KERNEL);
4330 		if (!dirty_i->dirty_segmap[i])
4331 			return -ENOMEM;
4332 	}
4333 
4334 	init_dirty_segmap(sbi);
4335 	return init_victim_secmap(sbi);
4336 }
4337 
4338 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4339 {
4340 	int i;
4341 
4342 	/*
4343 	 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4344 	 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4345 	 */
4346 	for (i = 0; i < NO_CHECK_TYPE; i++) {
4347 		struct curseg_info *curseg = CURSEG_I(sbi, i);
4348 		struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4349 		unsigned int blkofs = curseg->next_blkoff;
4350 
4351 		if (f2fs_test_bit(blkofs, se->cur_valid_map))
4352 			goto out;
4353 
4354 		if (curseg->alloc_type == SSR)
4355 			continue;
4356 
4357 		for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4358 			if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4359 				continue;
4360 out:
4361 			f2fs_err(sbi,
4362 				 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4363 				 i, curseg->segno, curseg->alloc_type,
4364 				 curseg->next_blkoff, blkofs);
4365 			return -EFSCORRUPTED;
4366 		}
4367 	}
4368 	return 0;
4369 }
4370 
4371 /*
4372  * Update min, max modified time for cost-benefit GC algorithm
4373  */
4374 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4375 {
4376 	struct sit_info *sit_i = SIT_I(sbi);
4377 	unsigned int segno;
4378 
4379 	down_write(&sit_i->sentry_lock);
4380 
4381 	sit_i->min_mtime = ULLONG_MAX;
4382 
4383 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4384 		unsigned int i;
4385 		unsigned long long mtime = 0;
4386 
4387 		for (i = 0; i < sbi->segs_per_sec; i++)
4388 			mtime += get_seg_entry(sbi, segno + i)->mtime;
4389 
4390 		mtime = div_u64(mtime, sbi->segs_per_sec);
4391 
4392 		if (sit_i->min_mtime > mtime)
4393 			sit_i->min_mtime = mtime;
4394 	}
4395 	sit_i->max_mtime = get_mtime(sbi, false);
4396 	up_write(&sit_i->sentry_lock);
4397 }
4398 
4399 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4400 {
4401 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4402 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4403 	struct f2fs_sm_info *sm_info;
4404 	int err;
4405 
4406 	sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4407 	if (!sm_info)
4408 		return -ENOMEM;
4409 
4410 	/* init sm info */
4411 	sbi->sm_info = sm_info;
4412 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4413 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4414 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4415 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4416 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4417 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4418 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4419 	sm_info->rec_prefree_segments = sm_info->main_segments *
4420 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4421 	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4422 		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4423 
4424 	if (!test_opt(sbi, LFS))
4425 		sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4426 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4427 	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4428 	sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4429 	sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4430 	sm_info->min_ssr_sections = reserved_sections(sbi);
4431 
4432 	INIT_LIST_HEAD(&sm_info->sit_entry_set);
4433 
4434 	init_rwsem(&sm_info->curseg_lock);
4435 
4436 	if (!f2fs_readonly(sbi->sb)) {
4437 		err = f2fs_create_flush_cmd_control(sbi);
4438 		if (err)
4439 			return err;
4440 	}
4441 
4442 	err = create_discard_cmd_control(sbi);
4443 	if (err)
4444 		return err;
4445 
4446 	err = build_sit_info(sbi);
4447 	if (err)
4448 		return err;
4449 	err = build_free_segmap(sbi);
4450 	if (err)
4451 		return err;
4452 	err = build_curseg(sbi);
4453 	if (err)
4454 		return err;
4455 
4456 	/* reinit free segmap based on SIT */
4457 	err = build_sit_entries(sbi);
4458 	if (err)
4459 		return err;
4460 
4461 	init_free_segmap(sbi);
4462 	err = build_dirty_segmap(sbi);
4463 	if (err)
4464 		return err;
4465 
4466 	err = sanity_check_curseg(sbi);
4467 	if (err)
4468 		return err;
4469 
4470 	init_min_max_mtime(sbi);
4471 	return 0;
4472 }
4473 
4474 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4475 		enum dirty_type dirty_type)
4476 {
4477 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4478 
4479 	mutex_lock(&dirty_i->seglist_lock);
4480 	kvfree(dirty_i->dirty_segmap[dirty_type]);
4481 	dirty_i->nr_dirty[dirty_type] = 0;
4482 	mutex_unlock(&dirty_i->seglist_lock);
4483 }
4484 
4485 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4486 {
4487 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4488 	kvfree(dirty_i->victim_secmap);
4489 }
4490 
4491 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4492 {
4493 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4494 	int i;
4495 
4496 	if (!dirty_i)
4497 		return;
4498 
4499 	/* discard pre-free/dirty segments list */
4500 	for (i = 0; i < NR_DIRTY_TYPE; i++)
4501 		discard_dirty_segmap(sbi, i);
4502 
4503 	destroy_victim_secmap(sbi);
4504 	SM_I(sbi)->dirty_info = NULL;
4505 	kvfree(dirty_i);
4506 }
4507 
4508 static void destroy_curseg(struct f2fs_sb_info *sbi)
4509 {
4510 	struct curseg_info *array = SM_I(sbi)->curseg_array;
4511 	int i;
4512 
4513 	if (!array)
4514 		return;
4515 	SM_I(sbi)->curseg_array = NULL;
4516 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
4517 		kvfree(array[i].sum_blk);
4518 		kvfree(array[i].journal);
4519 	}
4520 	kvfree(array);
4521 }
4522 
4523 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4524 {
4525 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4526 	if (!free_i)
4527 		return;
4528 	SM_I(sbi)->free_info = NULL;
4529 	kvfree(free_i->free_segmap);
4530 	kvfree(free_i->free_secmap);
4531 	kvfree(free_i);
4532 }
4533 
4534 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4535 {
4536 	struct sit_info *sit_i = SIT_I(sbi);
4537 
4538 	if (!sit_i)
4539 		return;
4540 
4541 	if (sit_i->sentries)
4542 		kvfree(sit_i->bitmap);
4543 	kvfree(sit_i->tmp_map);
4544 
4545 	kvfree(sit_i->sentries);
4546 	kvfree(sit_i->sec_entries);
4547 	kvfree(sit_i->dirty_sentries_bitmap);
4548 
4549 	SM_I(sbi)->sit_info = NULL;
4550 	kvfree(sit_i->sit_bitmap);
4551 #ifdef CONFIG_F2FS_CHECK_FS
4552 	kvfree(sit_i->sit_bitmap_mir);
4553 	kvfree(sit_i->invalid_segmap);
4554 #endif
4555 	kvfree(sit_i);
4556 }
4557 
4558 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4559 {
4560 	struct f2fs_sm_info *sm_info = SM_I(sbi);
4561 
4562 	if (!sm_info)
4563 		return;
4564 	f2fs_destroy_flush_cmd_control(sbi, true);
4565 	destroy_discard_cmd_control(sbi);
4566 	destroy_dirty_segmap(sbi);
4567 	destroy_curseg(sbi);
4568 	destroy_free_segmap(sbi);
4569 	destroy_sit_info(sbi);
4570 	sbi->sm_info = NULL;
4571 	kvfree(sm_info);
4572 }
4573 
4574 int __init f2fs_create_segment_manager_caches(void)
4575 {
4576 	discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4577 			sizeof(struct discard_entry));
4578 	if (!discard_entry_slab)
4579 		goto fail;
4580 
4581 	discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4582 			sizeof(struct discard_cmd));
4583 	if (!discard_cmd_slab)
4584 		goto destroy_discard_entry;
4585 
4586 	sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4587 			sizeof(struct sit_entry_set));
4588 	if (!sit_entry_set_slab)
4589 		goto destroy_discard_cmd;
4590 
4591 	inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4592 			sizeof(struct inmem_pages));
4593 	if (!inmem_entry_slab)
4594 		goto destroy_sit_entry_set;
4595 	return 0;
4596 
4597 destroy_sit_entry_set:
4598 	kmem_cache_destroy(sit_entry_set_slab);
4599 destroy_discard_cmd:
4600 	kmem_cache_destroy(discard_cmd_slab);
4601 destroy_discard_entry:
4602 	kmem_cache_destroy(discard_entry_slab);
4603 fail:
4604 	return -ENOMEM;
4605 }
4606 
4607 void f2fs_destroy_segment_manager_caches(void)
4608 {
4609 	kmem_cache_destroy(sit_entry_set_slab);
4610 	kmem_cache_destroy(discard_cmd_slab);
4611 	kmem_cache_destroy(discard_entry_slab);
4612 	kmem_cache_destroy(inmem_entry_slab);
4613 }
4614