xref: /linux/fs/f2fs/segment.c (revision f2bf88c4afc8c5ab92b40af24819933e57d0968c)
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 (f2fs_lfs_mode(sbi))
176 		return false;
177 	if (sbi->gc_mode == GC_URGENT_HIGH)
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,
249 							DEFAULT_IO_TIMEOUT);
250 					cond_resched();
251 					goto retry;
252 				}
253 				err = -EAGAIN;
254 				goto next;
255 			}
256 
257 			err = f2fs_get_node_info(sbi, dn.nid, &ni);
258 			if (err) {
259 				f2fs_put_dnode(&dn);
260 				return err;
261 			}
262 
263 			if (cur->old_addr == NEW_ADDR) {
264 				f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
265 				f2fs_update_data_blkaddr(&dn, NEW_ADDR);
266 			} else
267 				f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
268 					cur->old_addr, ni.version, true, true);
269 			f2fs_put_dnode(&dn);
270 		}
271 next:
272 		/* we don't need to invalidate this in the sccessful status */
273 		if (drop || recover) {
274 			ClearPageUptodate(page);
275 			clear_cold_data(page);
276 		}
277 		f2fs_clear_page_private(page);
278 		f2fs_put_page(page, 1);
279 
280 		list_del(&cur->list);
281 		kmem_cache_free(inmem_entry_slab, cur);
282 		dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
283 	}
284 	return err;
285 }
286 
287 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
288 {
289 	struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
290 	struct inode *inode;
291 	struct f2fs_inode_info *fi;
292 	unsigned int count = sbi->atomic_files;
293 	unsigned int looped = 0;
294 next:
295 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
296 	if (list_empty(head)) {
297 		spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
298 		return;
299 	}
300 	fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
301 	inode = igrab(&fi->vfs_inode);
302 	if (inode)
303 		list_move_tail(&fi->inmem_ilist, head);
304 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
305 
306 	if (inode) {
307 		if (gc_failure) {
308 			if (!fi->i_gc_failures[GC_FAILURE_ATOMIC])
309 				goto skip;
310 		}
311 		set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
312 		f2fs_drop_inmem_pages(inode);
313 skip:
314 		iput(inode);
315 	}
316 	congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
317 	cond_resched();
318 	if (gc_failure) {
319 		if (++looped >= count)
320 			return;
321 	}
322 	goto next;
323 }
324 
325 void f2fs_drop_inmem_pages(struct inode *inode)
326 {
327 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
328 	struct f2fs_inode_info *fi = F2FS_I(inode);
329 
330 	while (!list_empty(&fi->inmem_pages)) {
331 		mutex_lock(&fi->inmem_lock);
332 		__revoke_inmem_pages(inode, &fi->inmem_pages,
333 						true, false, true);
334 		mutex_unlock(&fi->inmem_lock);
335 	}
336 
337 	fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
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,
420 							DEFAULT_IO_TIMEOUT);
421 					cond_resched();
422 					goto retry;
423 				}
424 				unlock_page(page);
425 				break;
426 			}
427 			/* record old blkaddr for revoking */
428 			cur->old_addr = fio.old_blkaddr;
429 			submit_bio = true;
430 		}
431 		unlock_page(page);
432 		list_move_tail(&cur->list, &revoke_list);
433 	}
434 
435 	if (submit_bio)
436 		f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
437 
438 	if (err) {
439 		/*
440 		 * try to revoke all committed pages, but still we could fail
441 		 * due to no memory or other reason, if that happened, EAGAIN
442 		 * will be returned, which means in such case, transaction is
443 		 * already not integrity, caller should use journal to do the
444 		 * recovery or rewrite & commit last transaction. For other
445 		 * error number, revoking was done by filesystem itself.
446 		 */
447 		err = __revoke_inmem_pages(inode, &revoke_list,
448 						false, true, false);
449 
450 		/* drop all uncommitted pages */
451 		__revoke_inmem_pages(inode, &fi->inmem_pages,
452 						true, false, false);
453 	} else {
454 		__revoke_inmem_pages(inode, &revoke_list,
455 						false, false, false);
456 	}
457 
458 	return err;
459 }
460 
461 int f2fs_commit_inmem_pages(struct inode *inode)
462 {
463 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
464 	struct f2fs_inode_info *fi = F2FS_I(inode);
465 	int err;
466 
467 	f2fs_balance_fs(sbi, true);
468 
469 	down_write(&fi->i_gc_rwsem[WRITE]);
470 
471 	f2fs_lock_op(sbi);
472 	set_inode_flag(inode, FI_ATOMIC_COMMIT);
473 
474 	mutex_lock(&fi->inmem_lock);
475 	err = __f2fs_commit_inmem_pages(inode);
476 	mutex_unlock(&fi->inmem_lock);
477 
478 	clear_inode_flag(inode, FI_ATOMIC_COMMIT);
479 
480 	f2fs_unlock_op(sbi);
481 	up_write(&fi->i_gc_rwsem[WRITE]);
482 
483 	return err;
484 }
485 
486 /*
487  * This function balances dirty node and dentry pages.
488  * In addition, it controls garbage collection.
489  */
490 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
491 {
492 	if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
493 		f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
494 		f2fs_stop_checkpoint(sbi, false);
495 	}
496 
497 	/* balance_fs_bg is able to be pending */
498 	if (need && excess_cached_nats(sbi))
499 		f2fs_balance_fs_bg(sbi, false);
500 
501 	if (!f2fs_is_checkpoint_ready(sbi))
502 		return;
503 
504 	/*
505 	 * We should do GC or end up with checkpoint, if there are so many dirty
506 	 * dir/node pages without enough free segments.
507 	 */
508 	if (has_not_enough_free_secs(sbi, 0, 0)) {
509 		down_write(&sbi->gc_lock);
510 		f2fs_gc(sbi, false, false, NULL_SEGNO);
511 	}
512 }
513 
514 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
515 {
516 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
517 		return;
518 
519 	/* try to shrink extent cache when there is no enough memory */
520 	if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
521 		f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
522 
523 	/* check the # of cached NAT entries */
524 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
525 		f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
526 
527 	if (!f2fs_available_free_memory(sbi, FREE_NIDS))
528 		f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
529 	else
530 		f2fs_build_free_nids(sbi, false, false);
531 
532 	if (!is_idle(sbi, REQ_TIME) &&
533 		(!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
534 		return;
535 
536 	/* checkpoint is the only way to shrink partial cached entries */
537 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
538 			!f2fs_available_free_memory(sbi, INO_ENTRIES) ||
539 			excess_prefree_segs(sbi) ||
540 			excess_dirty_nats(sbi) ||
541 			excess_dirty_nodes(sbi) ||
542 			f2fs_time_over(sbi, CP_TIME)) {
543 		if (test_opt(sbi, DATA_FLUSH) && from_bg) {
544 			struct blk_plug plug;
545 
546 			mutex_lock(&sbi->flush_lock);
547 
548 			blk_start_plug(&plug);
549 			f2fs_sync_dirty_inodes(sbi, FILE_INODE);
550 			blk_finish_plug(&plug);
551 
552 			mutex_unlock(&sbi->flush_lock);
553 		}
554 		f2fs_sync_fs(sbi->sb, true);
555 		stat_inc_bg_cp_count(sbi->stat_info);
556 	}
557 }
558 
559 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
560 				struct block_device *bdev)
561 {
562 	struct bio *bio;
563 	int ret;
564 
565 	bio = f2fs_bio_alloc(sbi, 0, false);
566 	if (!bio)
567 		return -ENOMEM;
568 
569 	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
570 	bio_set_dev(bio, bdev);
571 	ret = submit_bio_wait(bio);
572 	bio_put(bio);
573 
574 	trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
575 				test_opt(sbi, FLUSH_MERGE), ret);
576 	return ret;
577 }
578 
579 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
580 {
581 	int ret = 0;
582 	int i;
583 
584 	if (!f2fs_is_multi_device(sbi))
585 		return __submit_flush_wait(sbi, sbi->sb->s_bdev);
586 
587 	for (i = 0; i < sbi->s_ndevs; i++) {
588 		if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
589 			continue;
590 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
591 		if (ret)
592 			break;
593 	}
594 	return ret;
595 }
596 
597 static int issue_flush_thread(void *data)
598 {
599 	struct f2fs_sb_info *sbi = data;
600 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
601 	wait_queue_head_t *q = &fcc->flush_wait_queue;
602 repeat:
603 	if (kthread_should_stop())
604 		return 0;
605 
606 	sb_start_intwrite(sbi->sb);
607 
608 	if (!llist_empty(&fcc->issue_list)) {
609 		struct flush_cmd *cmd, *next;
610 		int ret;
611 
612 		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
613 		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
614 
615 		cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
616 
617 		ret = submit_flush_wait(sbi, cmd->ino);
618 		atomic_inc(&fcc->issued_flush);
619 
620 		llist_for_each_entry_safe(cmd, next,
621 					  fcc->dispatch_list, llnode) {
622 			cmd->ret = ret;
623 			complete(&cmd->wait);
624 		}
625 		fcc->dispatch_list = NULL;
626 	}
627 
628 	sb_end_intwrite(sbi->sb);
629 
630 	wait_event_interruptible(*q,
631 		kthread_should_stop() || !llist_empty(&fcc->issue_list));
632 	goto repeat;
633 }
634 
635 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
636 {
637 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
638 	struct flush_cmd cmd;
639 	int ret;
640 
641 	if (test_opt(sbi, NOBARRIER))
642 		return 0;
643 
644 	if (!test_opt(sbi, FLUSH_MERGE)) {
645 		atomic_inc(&fcc->queued_flush);
646 		ret = submit_flush_wait(sbi, ino);
647 		atomic_dec(&fcc->queued_flush);
648 		atomic_inc(&fcc->issued_flush);
649 		return ret;
650 	}
651 
652 	if (atomic_inc_return(&fcc->queued_flush) == 1 ||
653 	    f2fs_is_multi_device(sbi)) {
654 		ret = submit_flush_wait(sbi, ino);
655 		atomic_dec(&fcc->queued_flush);
656 
657 		atomic_inc(&fcc->issued_flush);
658 		return ret;
659 	}
660 
661 	cmd.ino = ino;
662 	init_completion(&cmd.wait);
663 
664 	llist_add(&cmd.llnode, &fcc->issue_list);
665 
666 	/* update issue_list before we wake up issue_flush thread */
667 	smp_mb();
668 
669 	if (waitqueue_active(&fcc->flush_wait_queue))
670 		wake_up(&fcc->flush_wait_queue);
671 
672 	if (fcc->f2fs_issue_flush) {
673 		wait_for_completion(&cmd.wait);
674 		atomic_dec(&fcc->queued_flush);
675 	} else {
676 		struct llist_node *list;
677 
678 		list = llist_del_all(&fcc->issue_list);
679 		if (!list) {
680 			wait_for_completion(&cmd.wait);
681 			atomic_dec(&fcc->queued_flush);
682 		} else {
683 			struct flush_cmd *tmp, *next;
684 
685 			ret = submit_flush_wait(sbi, ino);
686 
687 			llist_for_each_entry_safe(tmp, next, list, llnode) {
688 				if (tmp == &cmd) {
689 					cmd.ret = ret;
690 					atomic_dec(&fcc->queued_flush);
691 					continue;
692 				}
693 				tmp->ret = ret;
694 				complete(&tmp->wait);
695 			}
696 		}
697 	}
698 
699 	return cmd.ret;
700 }
701 
702 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
703 {
704 	dev_t dev = sbi->sb->s_bdev->bd_dev;
705 	struct flush_cmd_control *fcc;
706 	int err = 0;
707 
708 	if (SM_I(sbi)->fcc_info) {
709 		fcc = SM_I(sbi)->fcc_info;
710 		if (fcc->f2fs_issue_flush)
711 			return err;
712 		goto init_thread;
713 	}
714 
715 	fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
716 	if (!fcc)
717 		return -ENOMEM;
718 	atomic_set(&fcc->issued_flush, 0);
719 	atomic_set(&fcc->queued_flush, 0);
720 	init_waitqueue_head(&fcc->flush_wait_queue);
721 	init_llist_head(&fcc->issue_list);
722 	SM_I(sbi)->fcc_info = fcc;
723 	if (!test_opt(sbi, FLUSH_MERGE))
724 		return err;
725 
726 init_thread:
727 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
728 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
729 	if (IS_ERR(fcc->f2fs_issue_flush)) {
730 		err = PTR_ERR(fcc->f2fs_issue_flush);
731 		kvfree(fcc);
732 		SM_I(sbi)->fcc_info = NULL;
733 		return err;
734 	}
735 
736 	return err;
737 }
738 
739 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
740 {
741 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
742 
743 	if (fcc && fcc->f2fs_issue_flush) {
744 		struct task_struct *flush_thread = fcc->f2fs_issue_flush;
745 
746 		fcc->f2fs_issue_flush = NULL;
747 		kthread_stop(flush_thread);
748 	}
749 	if (free) {
750 		kvfree(fcc);
751 		SM_I(sbi)->fcc_info = NULL;
752 	}
753 }
754 
755 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
756 {
757 	int ret = 0, i;
758 
759 	if (!f2fs_is_multi_device(sbi))
760 		return 0;
761 
762 	for (i = 1; i < sbi->s_ndevs; i++) {
763 		if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
764 			continue;
765 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
766 		if (ret)
767 			break;
768 
769 		spin_lock(&sbi->dev_lock);
770 		f2fs_clear_bit(i, (char *)&sbi->dirty_device);
771 		spin_unlock(&sbi->dev_lock);
772 	}
773 
774 	return ret;
775 }
776 
777 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
778 		enum dirty_type dirty_type)
779 {
780 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
781 
782 	/* need not be added */
783 	if (IS_CURSEG(sbi, segno))
784 		return;
785 
786 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
787 		dirty_i->nr_dirty[dirty_type]++;
788 
789 	if (dirty_type == DIRTY) {
790 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
791 		enum dirty_type t = sentry->type;
792 
793 		if (unlikely(t >= DIRTY)) {
794 			f2fs_bug_on(sbi, 1);
795 			return;
796 		}
797 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
798 			dirty_i->nr_dirty[t]++;
799 
800 		if (__is_large_section(sbi)) {
801 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
802 			block_t valid_blocks =
803 				get_valid_blocks(sbi, segno, true);
804 
805 			f2fs_bug_on(sbi, unlikely(!valid_blocks ||
806 					valid_blocks == BLKS_PER_SEC(sbi)));
807 
808 			if (!IS_CURSEC(sbi, secno))
809 				set_bit(secno, dirty_i->dirty_secmap);
810 		}
811 	}
812 }
813 
814 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
815 		enum dirty_type dirty_type)
816 {
817 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
818 	block_t valid_blocks;
819 
820 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
821 		dirty_i->nr_dirty[dirty_type]--;
822 
823 	if (dirty_type == DIRTY) {
824 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
825 		enum dirty_type t = sentry->type;
826 
827 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
828 			dirty_i->nr_dirty[t]--;
829 
830 		valid_blocks = get_valid_blocks(sbi, segno, true);
831 		if (valid_blocks == 0) {
832 			clear_bit(GET_SEC_FROM_SEG(sbi, segno),
833 						dirty_i->victim_secmap);
834 #ifdef CONFIG_F2FS_CHECK_FS
835 			clear_bit(segno, SIT_I(sbi)->invalid_segmap);
836 #endif
837 		}
838 		if (__is_large_section(sbi)) {
839 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
840 
841 			if (!valid_blocks ||
842 					valid_blocks == BLKS_PER_SEC(sbi)) {
843 				clear_bit(secno, dirty_i->dirty_secmap);
844 				return;
845 			}
846 
847 			if (!IS_CURSEC(sbi, secno))
848 				set_bit(secno, dirty_i->dirty_secmap);
849 		}
850 	}
851 }
852 
853 /*
854  * Should not occur error such as -ENOMEM.
855  * Adding dirty entry into seglist is not critical operation.
856  * If a given segment is one of current working segments, it won't be added.
857  */
858 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
859 {
860 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
861 	unsigned short valid_blocks, ckpt_valid_blocks;
862 
863 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
864 		return;
865 
866 	mutex_lock(&dirty_i->seglist_lock);
867 
868 	valid_blocks = get_valid_blocks(sbi, segno, false);
869 	ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
870 
871 	if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
872 				ckpt_valid_blocks == sbi->blocks_per_seg)) {
873 		__locate_dirty_segment(sbi, segno, PRE);
874 		__remove_dirty_segment(sbi, segno, DIRTY);
875 	} else if (valid_blocks < sbi->blocks_per_seg) {
876 		__locate_dirty_segment(sbi, segno, DIRTY);
877 	} else {
878 		/* Recovery routine with SSR needs this */
879 		__remove_dirty_segment(sbi, segno, DIRTY);
880 	}
881 
882 	mutex_unlock(&dirty_i->seglist_lock);
883 }
884 
885 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
886 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
887 {
888 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
889 	unsigned int segno;
890 
891 	mutex_lock(&dirty_i->seglist_lock);
892 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
893 		if (get_valid_blocks(sbi, segno, false))
894 			continue;
895 		if (IS_CURSEG(sbi, segno))
896 			continue;
897 		__locate_dirty_segment(sbi, segno, PRE);
898 		__remove_dirty_segment(sbi, segno, DIRTY);
899 	}
900 	mutex_unlock(&dirty_i->seglist_lock);
901 }
902 
903 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
904 {
905 	int ovp_hole_segs =
906 		(overprovision_segments(sbi) - reserved_segments(sbi));
907 	block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
908 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
909 	block_t holes[2] = {0, 0};	/* DATA and NODE */
910 	block_t unusable;
911 	struct seg_entry *se;
912 	unsigned int segno;
913 
914 	mutex_lock(&dirty_i->seglist_lock);
915 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
916 		se = get_seg_entry(sbi, segno);
917 		if (IS_NODESEG(se->type))
918 			holes[NODE] += sbi->blocks_per_seg - se->valid_blocks;
919 		else
920 			holes[DATA] += sbi->blocks_per_seg - se->valid_blocks;
921 	}
922 	mutex_unlock(&dirty_i->seglist_lock);
923 
924 	unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
925 	if (unusable > ovp_holes)
926 		return unusable - ovp_holes;
927 	return 0;
928 }
929 
930 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
931 {
932 	int ovp_hole_segs =
933 		(overprovision_segments(sbi) - reserved_segments(sbi));
934 	if (unusable > F2FS_OPTION(sbi).unusable_cap)
935 		return -EAGAIN;
936 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
937 		dirty_segments(sbi) > ovp_hole_segs)
938 		return -EAGAIN;
939 	return 0;
940 }
941 
942 /* This is only used by SBI_CP_DISABLED */
943 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
944 {
945 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
946 	unsigned int segno = 0;
947 
948 	mutex_lock(&dirty_i->seglist_lock);
949 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
950 		if (get_valid_blocks(sbi, segno, false))
951 			continue;
952 		if (get_ckpt_valid_blocks(sbi, segno))
953 			continue;
954 		mutex_unlock(&dirty_i->seglist_lock);
955 		return segno;
956 	}
957 	mutex_unlock(&dirty_i->seglist_lock);
958 	return NULL_SEGNO;
959 }
960 
961 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
962 		struct block_device *bdev, block_t lstart,
963 		block_t start, block_t len)
964 {
965 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
966 	struct list_head *pend_list;
967 	struct discard_cmd *dc;
968 
969 	f2fs_bug_on(sbi, !len);
970 
971 	pend_list = &dcc->pend_list[plist_idx(len)];
972 
973 	dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
974 	INIT_LIST_HEAD(&dc->list);
975 	dc->bdev = bdev;
976 	dc->lstart = lstart;
977 	dc->start = start;
978 	dc->len = len;
979 	dc->ref = 0;
980 	dc->state = D_PREP;
981 	dc->queued = 0;
982 	dc->error = 0;
983 	init_completion(&dc->wait);
984 	list_add_tail(&dc->list, pend_list);
985 	spin_lock_init(&dc->lock);
986 	dc->bio_ref = 0;
987 	atomic_inc(&dcc->discard_cmd_cnt);
988 	dcc->undiscard_blks += len;
989 
990 	return dc;
991 }
992 
993 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
994 				struct block_device *bdev, block_t lstart,
995 				block_t start, block_t len,
996 				struct rb_node *parent, struct rb_node **p,
997 				bool leftmost)
998 {
999 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1000 	struct discard_cmd *dc;
1001 
1002 	dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1003 
1004 	rb_link_node(&dc->rb_node, parent, p);
1005 	rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1006 
1007 	return dc;
1008 }
1009 
1010 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1011 							struct discard_cmd *dc)
1012 {
1013 	if (dc->state == D_DONE)
1014 		atomic_sub(dc->queued, &dcc->queued_discard);
1015 
1016 	list_del(&dc->list);
1017 	rb_erase_cached(&dc->rb_node, &dcc->root);
1018 	dcc->undiscard_blks -= dc->len;
1019 
1020 	kmem_cache_free(discard_cmd_slab, dc);
1021 
1022 	atomic_dec(&dcc->discard_cmd_cnt);
1023 }
1024 
1025 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1026 							struct discard_cmd *dc)
1027 {
1028 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1029 	unsigned long flags;
1030 
1031 	trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
1032 
1033 	spin_lock_irqsave(&dc->lock, flags);
1034 	if (dc->bio_ref) {
1035 		spin_unlock_irqrestore(&dc->lock, flags);
1036 		return;
1037 	}
1038 	spin_unlock_irqrestore(&dc->lock, flags);
1039 
1040 	f2fs_bug_on(sbi, dc->ref);
1041 
1042 	if (dc->error == -EOPNOTSUPP)
1043 		dc->error = 0;
1044 
1045 	if (dc->error)
1046 		printk_ratelimited(
1047 			"%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1048 			KERN_INFO, sbi->sb->s_id,
1049 			dc->lstart, dc->start, dc->len, dc->error);
1050 	__detach_discard_cmd(dcc, dc);
1051 }
1052 
1053 static void f2fs_submit_discard_endio(struct bio *bio)
1054 {
1055 	struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1056 	unsigned long flags;
1057 
1058 	spin_lock_irqsave(&dc->lock, flags);
1059 	if (!dc->error)
1060 		dc->error = blk_status_to_errno(bio->bi_status);
1061 	dc->bio_ref--;
1062 	if (!dc->bio_ref && dc->state == D_SUBMIT) {
1063 		dc->state = D_DONE;
1064 		complete_all(&dc->wait);
1065 	}
1066 	spin_unlock_irqrestore(&dc->lock, flags);
1067 	bio_put(bio);
1068 }
1069 
1070 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1071 				block_t start, block_t end)
1072 {
1073 #ifdef CONFIG_F2FS_CHECK_FS
1074 	struct seg_entry *sentry;
1075 	unsigned int segno;
1076 	block_t blk = start;
1077 	unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1078 	unsigned long *map;
1079 
1080 	while (blk < end) {
1081 		segno = GET_SEGNO(sbi, blk);
1082 		sentry = get_seg_entry(sbi, segno);
1083 		offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1084 
1085 		if (end < START_BLOCK(sbi, segno + 1))
1086 			size = GET_BLKOFF_FROM_SEG0(sbi, end);
1087 		else
1088 			size = max_blocks;
1089 		map = (unsigned long *)(sentry->cur_valid_map);
1090 		offset = __find_rev_next_bit(map, size, offset);
1091 		f2fs_bug_on(sbi, offset != size);
1092 		blk = START_BLOCK(sbi, segno + 1);
1093 	}
1094 #endif
1095 }
1096 
1097 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1098 				struct discard_policy *dpolicy,
1099 				int discard_type, unsigned int granularity)
1100 {
1101 	/* common policy */
1102 	dpolicy->type = discard_type;
1103 	dpolicy->sync = true;
1104 	dpolicy->ordered = false;
1105 	dpolicy->granularity = granularity;
1106 
1107 	dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1108 	dpolicy->io_aware_gran = MAX_PLIST_NUM;
1109 	dpolicy->timeout = false;
1110 
1111 	if (discard_type == DPOLICY_BG) {
1112 		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1113 		dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1114 		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1115 		dpolicy->io_aware = true;
1116 		dpolicy->sync = false;
1117 		dpolicy->ordered = true;
1118 		if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1119 			dpolicy->granularity = 1;
1120 			dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1121 		}
1122 	} else if (discard_type == DPOLICY_FORCE) {
1123 		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1124 		dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1125 		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1126 		dpolicy->io_aware = false;
1127 	} else if (discard_type == DPOLICY_FSTRIM) {
1128 		dpolicy->io_aware = false;
1129 	} else if (discard_type == DPOLICY_UMOUNT) {
1130 		dpolicy->io_aware = false;
1131 		/* we need to issue all to keep CP_TRIMMED_FLAG */
1132 		dpolicy->granularity = 1;
1133 		dpolicy->timeout = true;
1134 	}
1135 }
1136 
1137 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1138 				struct block_device *bdev, block_t lstart,
1139 				block_t start, block_t len);
1140 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1141 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1142 						struct discard_policy *dpolicy,
1143 						struct discard_cmd *dc,
1144 						unsigned int *issued)
1145 {
1146 	struct block_device *bdev = dc->bdev;
1147 	struct request_queue *q = bdev_get_queue(bdev);
1148 	unsigned int max_discard_blocks =
1149 			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1150 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1151 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1152 					&(dcc->fstrim_list) : &(dcc->wait_list);
1153 	int flag = dpolicy->sync ? REQ_SYNC : 0;
1154 	block_t lstart, start, len, total_len;
1155 	int err = 0;
1156 
1157 	if (dc->state != D_PREP)
1158 		return 0;
1159 
1160 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1161 		return 0;
1162 
1163 	trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1164 
1165 	lstart = dc->lstart;
1166 	start = dc->start;
1167 	len = dc->len;
1168 	total_len = len;
1169 
1170 	dc->len = 0;
1171 
1172 	while (total_len && *issued < dpolicy->max_requests && !err) {
1173 		struct bio *bio = NULL;
1174 		unsigned long flags;
1175 		bool last = true;
1176 
1177 		if (len > max_discard_blocks) {
1178 			len = max_discard_blocks;
1179 			last = false;
1180 		}
1181 
1182 		(*issued)++;
1183 		if (*issued == dpolicy->max_requests)
1184 			last = true;
1185 
1186 		dc->len += len;
1187 
1188 		if (time_to_inject(sbi, FAULT_DISCARD)) {
1189 			f2fs_show_injection_info(sbi, FAULT_DISCARD);
1190 			err = -EIO;
1191 			goto submit;
1192 		}
1193 		err = __blkdev_issue_discard(bdev,
1194 					SECTOR_FROM_BLOCK(start),
1195 					SECTOR_FROM_BLOCK(len),
1196 					GFP_NOFS, 0, &bio);
1197 submit:
1198 		if (err) {
1199 			spin_lock_irqsave(&dc->lock, flags);
1200 			if (dc->state == D_PARTIAL)
1201 				dc->state = D_SUBMIT;
1202 			spin_unlock_irqrestore(&dc->lock, flags);
1203 
1204 			break;
1205 		}
1206 
1207 		f2fs_bug_on(sbi, !bio);
1208 
1209 		/*
1210 		 * should keep before submission to avoid D_DONE
1211 		 * right away
1212 		 */
1213 		spin_lock_irqsave(&dc->lock, flags);
1214 		if (last)
1215 			dc->state = D_SUBMIT;
1216 		else
1217 			dc->state = D_PARTIAL;
1218 		dc->bio_ref++;
1219 		spin_unlock_irqrestore(&dc->lock, flags);
1220 
1221 		atomic_inc(&dcc->queued_discard);
1222 		dc->queued++;
1223 		list_move_tail(&dc->list, wait_list);
1224 
1225 		/* sanity check on discard range */
1226 		__check_sit_bitmap(sbi, lstart, lstart + len);
1227 
1228 		bio->bi_private = dc;
1229 		bio->bi_end_io = f2fs_submit_discard_endio;
1230 		bio->bi_opf |= flag;
1231 		submit_bio(bio);
1232 
1233 		atomic_inc(&dcc->issued_discard);
1234 
1235 		f2fs_update_iostat(sbi, FS_DISCARD, 1);
1236 
1237 		lstart += len;
1238 		start += len;
1239 		total_len -= len;
1240 		len = total_len;
1241 	}
1242 
1243 	if (!err && len) {
1244 		dcc->undiscard_blks -= len;
1245 		__update_discard_tree_range(sbi, bdev, lstart, start, len);
1246 	}
1247 	return err;
1248 }
1249 
1250 static void __insert_discard_tree(struct f2fs_sb_info *sbi,
1251 				struct block_device *bdev, block_t lstart,
1252 				block_t start, block_t len,
1253 				struct rb_node **insert_p,
1254 				struct rb_node *insert_parent)
1255 {
1256 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1257 	struct rb_node **p;
1258 	struct rb_node *parent = NULL;
1259 	bool leftmost = true;
1260 
1261 	if (insert_p && insert_parent) {
1262 		parent = insert_parent;
1263 		p = insert_p;
1264 		goto do_insert;
1265 	}
1266 
1267 	p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1268 							lstart, &leftmost);
1269 do_insert:
1270 	__attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1271 								p, leftmost);
1272 }
1273 
1274 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1275 						struct discard_cmd *dc)
1276 {
1277 	list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1278 }
1279 
1280 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1281 				struct discard_cmd *dc, block_t blkaddr)
1282 {
1283 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1284 	struct discard_info di = dc->di;
1285 	bool modified = false;
1286 
1287 	if (dc->state == D_DONE || dc->len == 1) {
1288 		__remove_discard_cmd(sbi, dc);
1289 		return;
1290 	}
1291 
1292 	dcc->undiscard_blks -= di.len;
1293 
1294 	if (blkaddr > di.lstart) {
1295 		dc->len = blkaddr - dc->lstart;
1296 		dcc->undiscard_blks += dc->len;
1297 		__relocate_discard_cmd(dcc, dc);
1298 		modified = true;
1299 	}
1300 
1301 	if (blkaddr < di.lstart + di.len - 1) {
1302 		if (modified) {
1303 			__insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1304 					di.start + blkaddr + 1 - di.lstart,
1305 					di.lstart + di.len - 1 - blkaddr,
1306 					NULL, NULL);
1307 		} else {
1308 			dc->lstart++;
1309 			dc->len--;
1310 			dc->start++;
1311 			dcc->undiscard_blks += dc->len;
1312 			__relocate_discard_cmd(dcc, dc);
1313 		}
1314 	}
1315 }
1316 
1317 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1318 				struct block_device *bdev, block_t lstart,
1319 				block_t start, block_t len)
1320 {
1321 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1322 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1323 	struct discard_cmd *dc;
1324 	struct discard_info di = {0};
1325 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1326 	struct request_queue *q = bdev_get_queue(bdev);
1327 	unsigned int max_discard_blocks =
1328 			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1329 	block_t end = lstart + len;
1330 
1331 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1332 					NULL, lstart,
1333 					(struct rb_entry **)&prev_dc,
1334 					(struct rb_entry **)&next_dc,
1335 					&insert_p, &insert_parent, true, NULL);
1336 	if (dc)
1337 		prev_dc = dc;
1338 
1339 	if (!prev_dc) {
1340 		di.lstart = lstart;
1341 		di.len = next_dc ? next_dc->lstart - lstart : len;
1342 		di.len = min(di.len, len);
1343 		di.start = start;
1344 	}
1345 
1346 	while (1) {
1347 		struct rb_node *node;
1348 		bool merged = false;
1349 		struct discard_cmd *tdc = NULL;
1350 
1351 		if (prev_dc) {
1352 			di.lstart = prev_dc->lstart + prev_dc->len;
1353 			if (di.lstart < lstart)
1354 				di.lstart = lstart;
1355 			if (di.lstart >= end)
1356 				break;
1357 
1358 			if (!next_dc || next_dc->lstart > end)
1359 				di.len = end - di.lstart;
1360 			else
1361 				di.len = next_dc->lstart - di.lstart;
1362 			di.start = start + di.lstart - lstart;
1363 		}
1364 
1365 		if (!di.len)
1366 			goto next;
1367 
1368 		if (prev_dc && prev_dc->state == D_PREP &&
1369 			prev_dc->bdev == bdev &&
1370 			__is_discard_back_mergeable(&di, &prev_dc->di,
1371 							max_discard_blocks)) {
1372 			prev_dc->di.len += di.len;
1373 			dcc->undiscard_blks += di.len;
1374 			__relocate_discard_cmd(dcc, prev_dc);
1375 			di = prev_dc->di;
1376 			tdc = prev_dc;
1377 			merged = true;
1378 		}
1379 
1380 		if (next_dc && next_dc->state == D_PREP &&
1381 			next_dc->bdev == bdev &&
1382 			__is_discard_front_mergeable(&di, &next_dc->di,
1383 							max_discard_blocks)) {
1384 			next_dc->di.lstart = di.lstart;
1385 			next_dc->di.len += di.len;
1386 			next_dc->di.start = di.start;
1387 			dcc->undiscard_blks += di.len;
1388 			__relocate_discard_cmd(dcc, next_dc);
1389 			if (tdc)
1390 				__remove_discard_cmd(sbi, tdc);
1391 			merged = true;
1392 		}
1393 
1394 		if (!merged) {
1395 			__insert_discard_tree(sbi, bdev, di.lstart, di.start,
1396 							di.len, NULL, NULL);
1397 		}
1398  next:
1399 		prev_dc = next_dc;
1400 		if (!prev_dc)
1401 			break;
1402 
1403 		node = rb_next(&prev_dc->rb_node);
1404 		next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1405 	}
1406 }
1407 
1408 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1409 		struct block_device *bdev, block_t blkstart, block_t blklen)
1410 {
1411 	block_t lblkstart = blkstart;
1412 
1413 	if (!f2fs_bdev_support_discard(bdev))
1414 		return 0;
1415 
1416 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
1417 
1418 	if (f2fs_is_multi_device(sbi)) {
1419 		int devi = f2fs_target_device_index(sbi, blkstart);
1420 
1421 		blkstart -= FDEV(devi).start_blk;
1422 	}
1423 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1424 	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1425 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1426 	return 0;
1427 }
1428 
1429 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1430 					struct discard_policy *dpolicy)
1431 {
1432 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1433 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1434 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1435 	struct discard_cmd *dc;
1436 	struct blk_plug plug;
1437 	unsigned int pos = dcc->next_pos;
1438 	unsigned int issued = 0;
1439 	bool io_interrupted = false;
1440 
1441 	mutex_lock(&dcc->cmd_lock);
1442 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1443 					NULL, pos,
1444 					(struct rb_entry **)&prev_dc,
1445 					(struct rb_entry **)&next_dc,
1446 					&insert_p, &insert_parent, true, NULL);
1447 	if (!dc)
1448 		dc = next_dc;
1449 
1450 	blk_start_plug(&plug);
1451 
1452 	while (dc) {
1453 		struct rb_node *node;
1454 		int err = 0;
1455 
1456 		if (dc->state != D_PREP)
1457 			goto next;
1458 
1459 		if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1460 			io_interrupted = true;
1461 			break;
1462 		}
1463 
1464 		dcc->next_pos = dc->lstart + dc->len;
1465 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1466 
1467 		if (issued >= dpolicy->max_requests)
1468 			break;
1469 next:
1470 		node = rb_next(&dc->rb_node);
1471 		if (err)
1472 			__remove_discard_cmd(sbi, dc);
1473 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1474 	}
1475 
1476 	blk_finish_plug(&plug);
1477 
1478 	if (!dc)
1479 		dcc->next_pos = 0;
1480 
1481 	mutex_unlock(&dcc->cmd_lock);
1482 
1483 	if (!issued && io_interrupted)
1484 		issued = -1;
1485 
1486 	return issued;
1487 }
1488 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1489 					struct discard_policy *dpolicy);
1490 
1491 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1492 					struct discard_policy *dpolicy)
1493 {
1494 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1495 	struct list_head *pend_list;
1496 	struct discard_cmd *dc, *tmp;
1497 	struct blk_plug plug;
1498 	int i, issued;
1499 	bool io_interrupted = false;
1500 
1501 	if (dpolicy->timeout)
1502 		f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1503 
1504 retry:
1505 	issued = 0;
1506 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1507 		if (dpolicy->timeout &&
1508 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1509 			break;
1510 
1511 		if (i + 1 < dpolicy->granularity)
1512 			break;
1513 
1514 		if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1515 			return __issue_discard_cmd_orderly(sbi, dpolicy);
1516 
1517 		pend_list = &dcc->pend_list[i];
1518 
1519 		mutex_lock(&dcc->cmd_lock);
1520 		if (list_empty(pend_list))
1521 			goto next;
1522 		if (unlikely(dcc->rbtree_check))
1523 			f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1524 								&dcc->root));
1525 		blk_start_plug(&plug);
1526 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1527 			f2fs_bug_on(sbi, dc->state != D_PREP);
1528 
1529 			if (dpolicy->timeout &&
1530 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1531 				break;
1532 
1533 			if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1534 						!is_idle(sbi, DISCARD_TIME)) {
1535 				io_interrupted = true;
1536 				break;
1537 			}
1538 
1539 			__submit_discard_cmd(sbi, dpolicy, dc, &issued);
1540 
1541 			if (issued >= dpolicy->max_requests)
1542 				break;
1543 		}
1544 		blk_finish_plug(&plug);
1545 next:
1546 		mutex_unlock(&dcc->cmd_lock);
1547 
1548 		if (issued >= dpolicy->max_requests || io_interrupted)
1549 			break;
1550 	}
1551 
1552 	if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1553 		__wait_all_discard_cmd(sbi, dpolicy);
1554 		goto retry;
1555 	}
1556 
1557 	if (!issued && io_interrupted)
1558 		issued = -1;
1559 
1560 	return issued;
1561 }
1562 
1563 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1564 {
1565 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1566 	struct list_head *pend_list;
1567 	struct discard_cmd *dc, *tmp;
1568 	int i;
1569 	bool dropped = false;
1570 
1571 	mutex_lock(&dcc->cmd_lock);
1572 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1573 		pend_list = &dcc->pend_list[i];
1574 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1575 			f2fs_bug_on(sbi, dc->state != D_PREP);
1576 			__remove_discard_cmd(sbi, dc);
1577 			dropped = true;
1578 		}
1579 	}
1580 	mutex_unlock(&dcc->cmd_lock);
1581 
1582 	return dropped;
1583 }
1584 
1585 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1586 {
1587 	__drop_discard_cmd(sbi);
1588 }
1589 
1590 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1591 							struct discard_cmd *dc)
1592 {
1593 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1594 	unsigned int len = 0;
1595 
1596 	wait_for_completion_io(&dc->wait);
1597 	mutex_lock(&dcc->cmd_lock);
1598 	f2fs_bug_on(sbi, dc->state != D_DONE);
1599 	dc->ref--;
1600 	if (!dc->ref) {
1601 		if (!dc->error)
1602 			len = dc->len;
1603 		__remove_discard_cmd(sbi, dc);
1604 	}
1605 	mutex_unlock(&dcc->cmd_lock);
1606 
1607 	return len;
1608 }
1609 
1610 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1611 						struct discard_policy *dpolicy,
1612 						block_t start, block_t end)
1613 {
1614 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1615 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1616 					&(dcc->fstrim_list) : &(dcc->wait_list);
1617 	struct discard_cmd *dc, *tmp;
1618 	bool need_wait;
1619 	unsigned int trimmed = 0;
1620 
1621 next:
1622 	need_wait = false;
1623 
1624 	mutex_lock(&dcc->cmd_lock);
1625 	list_for_each_entry_safe(dc, tmp, wait_list, list) {
1626 		if (dc->lstart + dc->len <= start || end <= dc->lstart)
1627 			continue;
1628 		if (dc->len < dpolicy->granularity)
1629 			continue;
1630 		if (dc->state == D_DONE && !dc->ref) {
1631 			wait_for_completion_io(&dc->wait);
1632 			if (!dc->error)
1633 				trimmed += dc->len;
1634 			__remove_discard_cmd(sbi, dc);
1635 		} else {
1636 			dc->ref++;
1637 			need_wait = true;
1638 			break;
1639 		}
1640 	}
1641 	mutex_unlock(&dcc->cmd_lock);
1642 
1643 	if (need_wait) {
1644 		trimmed += __wait_one_discard_bio(sbi, dc);
1645 		goto next;
1646 	}
1647 
1648 	return trimmed;
1649 }
1650 
1651 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1652 						struct discard_policy *dpolicy)
1653 {
1654 	struct discard_policy dp;
1655 	unsigned int discard_blks;
1656 
1657 	if (dpolicy)
1658 		return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1659 
1660 	/* wait all */
1661 	__init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1662 	discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1663 	__init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1664 	discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1665 
1666 	return discard_blks;
1667 }
1668 
1669 /* This should be covered by global mutex, &sit_i->sentry_lock */
1670 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1671 {
1672 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1673 	struct discard_cmd *dc;
1674 	bool need_wait = false;
1675 
1676 	mutex_lock(&dcc->cmd_lock);
1677 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1678 							NULL, blkaddr);
1679 	if (dc) {
1680 		if (dc->state == D_PREP) {
1681 			__punch_discard_cmd(sbi, dc, blkaddr);
1682 		} else {
1683 			dc->ref++;
1684 			need_wait = true;
1685 		}
1686 	}
1687 	mutex_unlock(&dcc->cmd_lock);
1688 
1689 	if (need_wait)
1690 		__wait_one_discard_bio(sbi, dc);
1691 }
1692 
1693 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1694 {
1695 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1696 
1697 	if (dcc && dcc->f2fs_issue_discard) {
1698 		struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1699 
1700 		dcc->f2fs_issue_discard = NULL;
1701 		kthread_stop(discard_thread);
1702 	}
1703 }
1704 
1705 /* This comes from f2fs_put_super */
1706 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1707 {
1708 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1709 	struct discard_policy dpolicy;
1710 	bool dropped;
1711 
1712 	__init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1713 					dcc->discard_granularity);
1714 	__issue_discard_cmd(sbi, &dpolicy);
1715 	dropped = __drop_discard_cmd(sbi);
1716 
1717 	/* just to make sure there is no pending discard commands */
1718 	__wait_all_discard_cmd(sbi, NULL);
1719 
1720 	f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1721 	return dropped;
1722 }
1723 
1724 static int issue_discard_thread(void *data)
1725 {
1726 	struct f2fs_sb_info *sbi = data;
1727 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1728 	wait_queue_head_t *q = &dcc->discard_wait_queue;
1729 	struct discard_policy dpolicy;
1730 	unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1731 	int issued;
1732 
1733 	set_freezable();
1734 
1735 	do {
1736 		__init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1737 					dcc->discard_granularity);
1738 
1739 		wait_event_interruptible_timeout(*q,
1740 				kthread_should_stop() || freezing(current) ||
1741 				dcc->discard_wake,
1742 				msecs_to_jiffies(wait_ms));
1743 
1744 		if (dcc->discard_wake)
1745 			dcc->discard_wake = 0;
1746 
1747 		/* clean up pending candidates before going to sleep */
1748 		if (atomic_read(&dcc->queued_discard))
1749 			__wait_all_discard_cmd(sbi, NULL);
1750 
1751 		if (try_to_freeze())
1752 			continue;
1753 		if (f2fs_readonly(sbi->sb))
1754 			continue;
1755 		if (kthread_should_stop())
1756 			return 0;
1757 		if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1758 			wait_ms = dpolicy.max_interval;
1759 			continue;
1760 		}
1761 
1762 		if (sbi->gc_mode == GC_URGENT_HIGH)
1763 			__init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1764 
1765 		sb_start_intwrite(sbi->sb);
1766 
1767 		issued = __issue_discard_cmd(sbi, &dpolicy);
1768 		if (issued > 0) {
1769 			__wait_all_discard_cmd(sbi, &dpolicy);
1770 			wait_ms = dpolicy.min_interval;
1771 		} else if (issued == -1){
1772 			wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1773 			if (!wait_ms)
1774 				wait_ms = dpolicy.mid_interval;
1775 		} else {
1776 			wait_ms = dpolicy.max_interval;
1777 		}
1778 
1779 		sb_end_intwrite(sbi->sb);
1780 
1781 	} while (!kthread_should_stop());
1782 	return 0;
1783 }
1784 
1785 #ifdef CONFIG_BLK_DEV_ZONED
1786 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1787 		struct block_device *bdev, block_t blkstart, block_t blklen)
1788 {
1789 	sector_t sector, nr_sects;
1790 	block_t lblkstart = blkstart;
1791 	int devi = 0;
1792 
1793 	if (f2fs_is_multi_device(sbi)) {
1794 		devi = f2fs_target_device_index(sbi, blkstart);
1795 		if (blkstart < FDEV(devi).start_blk ||
1796 		    blkstart > FDEV(devi).end_blk) {
1797 			f2fs_err(sbi, "Invalid block %x", blkstart);
1798 			return -EIO;
1799 		}
1800 		blkstart -= FDEV(devi).start_blk;
1801 	}
1802 
1803 	/* For sequential zones, reset the zone write pointer */
1804 	if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1805 		sector = SECTOR_FROM_BLOCK(blkstart);
1806 		nr_sects = SECTOR_FROM_BLOCK(blklen);
1807 
1808 		if (sector & (bdev_zone_sectors(bdev) - 1) ||
1809 				nr_sects != bdev_zone_sectors(bdev)) {
1810 			f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1811 				 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1812 				 blkstart, blklen);
1813 			return -EIO;
1814 		}
1815 		trace_f2fs_issue_reset_zone(bdev, blkstart);
1816 		return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1817 					sector, nr_sects, GFP_NOFS);
1818 	}
1819 
1820 	/* For conventional zones, use regular discard if supported */
1821 	return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1822 }
1823 #endif
1824 
1825 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1826 		struct block_device *bdev, block_t blkstart, block_t blklen)
1827 {
1828 #ifdef CONFIG_BLK_DEV_ZONED
1829 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1830 		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1831 #endif
1832 	return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1833 }
1834 
1835 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1836 				block_t blkstart, block_t blklen)
1837 {
1838 	sector_t start = blkstart, len = 0;
1839 	struct block_device *bdev;
1840 	struct seg_entry *se;
1841 	unsigned int offset;
1842 	block_t i;
1843 	int err = 0;
1844 
1845 	bdev = f2fs_target_device(sbi, blkstart, NULL);
1846 
1847 	for (i = blkstart; i < blkstart + blklen; i++, len++) {
1848 		if (i != start) {
1849 			struct block_device *bdev2 =
1850 				f2fs_target_device(sbi, i, NULL);
1851 
1852 			if (bdev2 != bdev) {
1853 				err = __issue_discard_async(sbi, bdev,
1854 						start, len);
1855 				if (err)
1856 					return err;
1857 				bdev = bdev2;
1858 				start = i;
1859 				len = 0;
1860 			}
1861 		}
1862 
1863 		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1864 		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1865 
1866 		if (!f2fs_test_and_set_bit(offset, se->discard_map))
1867 			sbi->discard_blks--;
1868 	}
1869 
1870 	if (len)
1871 		err = __issue_discard_async(sbi, bdev, start, len);
1872 	return err;
1873 }
1874 
1875 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1876 							bool check_only)
1877 {
1878 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1879 	int max_blocks = sbi->blocks_per_seg;
1880 	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1881 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1882 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1883 	unsigned long *discard_map = (unsigned long *)se->discard_map;
1884 	unsigned long *dmap = SIT_I(sbi)->tmp_map;
1885 	unsigned int start = 0, end = -1;
1886 	bool force = (cpc->reason & CP_DISCARD);
1887 	struct discard_entry *de = NULL;
1888 	struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1889 	int i;
1890 
1891 	if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1892 		return false;
1893 
1894 	if (!force) {
1895 		if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1896 			SM_I(sbi)->dcc_info->nr_discards >=
1897 				SM_I(sbi)->dcc_info->max_discards)
1898 			return false;
1899 	}
1900 
1901 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1902 	for (i = 0; i < entries; i++)
1903 		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1904 				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1905 
1906 	while (force || SM_I(sbi)->dcc_info->nr_discards <=
1907 				SM_I(sbi)->dcc_info->max_discards) {
1908 		start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1909 		if (start >= max_blocks)
1910 			break;
1911 
1912 		end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1913 		if (force && start && end != max_blocks
1914 					&& (end - start) < cpc->trim_minlen)
1915 			continue;
1916 
1917 		if (check_only)
1918 			return true;
1919 
1920 		if (!de) {
1921 			de = f2fs_kmem_cache_alloc(discard_entry_slab,
1922 								GFP_F2FS_ZERO);
1923 			de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1924 			list_add_tail(&de->list, head);
1925 		}
1926 
1927 		for (i = start; i < end; i++)
1928 			__set_bit_le(i, (void *)de->discard_map);
1929 
1930 		SM_I(sbi)->dcc_info->nr_discards += end - start;
1931 	}
1932 	return false;
1933 }
1934 
1935 static void release_discard_addr(struct discard_entry *entry)
1936 {
1937 	list_del(&entry->list);
1938 	kmem_cache_free(discard_entry_slab, entry);
1939 }
1940 
1941 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1942 {
1943 	struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1944 	struct discard_entry *entry, *this;
1945 
1946 	/* drop caches */
1947 	list_for_each_entry_safe(entry, this, head, list)
1948 		release_discard_addr(entry);
1949 }
1950 
1951 /*
1952  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1953  */
1954 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1955 {
1956 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1957 	unsigned int segno;
1958 
1959 	mutex_lock(&dirty_i->seglist_lock);
1960 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1961 		__set_test_and_free(sbi, segno);
1962 	mutex_unlock(&dirty_i->seglist_lock);
1963 }
1964 
1965 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1966 						struct cp_control *cpc)
1967 {
1968 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1969 	struct list_head *head = &dcc->entry_list;
1970 	struct discard_entry *entry, *this;
1971 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1972 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1973 	unsigned int start = 0, end = -1;
1974 	unsigned int secno, start_segno;
1975 	bool force = (cpc->reason & CP_DISCARD);
1976 	bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
1977 
1978 	mutex_lock(&dirty_i->seglist_lock);
1979 
1980 	while (1) {
1981 		int i;
1982 
1983 		if (need_align && end != -1)
1984 			end--;
1985 		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1986 		if (start >= MAIN_SEGS(sbi))
1987 			break;
1988 		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1989 								start + 1);
1990 
1991 		if (need_align) {
1992 			start = rounddown(start, sbi->segs_per_sec);
1993 			end = roundup(end, sbi->segs_per_sec);
1994 		}
1995 
1996 		for (i = start; i < end; i++) {
1997 			if (test_and_clear_bit(i, prefree_map))
1998 				dirty_i->nr_dirty[PRE]--;
1999 		}
2000 
2001 		if (!f2fs_realtime_discard_enable(sbi))
2002 			continue;
2003 
2004 		if (force && start >= cpc->trim_start &&
2005 					(end - 1) <= cpc->trim_end)
2006 				continue;
2007 
2008 		if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) {
2009 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2010 				(end - start) << sbi->log_blocks_per_seg);
2011 			continue;
2012 		}
2013 next:
2014 		secno = GET_SEC_FROM_SEG(sbi, start);
2015 		start_segno = GET_SEG_FROM_SEC(sbi, secno);
2016 		if (!IS_CURSEC(sbi, secno) &&
2017 			!get_valid_blocks(sbi, start, true))
2018 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2019 				sbi->segs_per_sec << sbi->log_blocks_per_seg);
2020 
2021 		start = start_segno + sbi->segs_per_sec;
2022 		if (start < end)
2023 			goto next;
2024 		else
2025 			end = start - 1;
2026 	}
2027 	mutex_unlock(&dirty_i->seglist_lock);
2028 
2029 	/* send small discards */
2030 	list_for_each_entry_safe(entry, this, head, list) {
2031 		unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2032 		bool is_valid = test_bit_le(0, entry->discard_map);
2033 
2034 find_next:
2035 		if (is_valid) {
2036 			next_pos = find_next_zero_bit_le(entry->discard_map,
2037 					sbi->blocks_per_seg, cur_pos);
2038 			len = next_pos - cur_pos;
2039 
2040 			if (f2fs_sb_has_blkzoned(sbi) ||
2041 			    (force && len < cpc->trim_minlen))
2042 				goto skip;
2043 
2044 			f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2045 									len);
2046 			total_len += len;
2047 		} else {
2048 			next_pos = find_next_bit_le(entry->discard_map,
2049 					sbi->blocks_per_seg, cur_pos);
2050 		}
2051 skip:
2052 		cur_pos = next_pos;
2053 		is_valid = !is_valid;
2054 
2055 		if (cur_pos < sbi->blocks_per_seg)
2056 			goto find_next;
2057 
2058 		release_discard_addr(entry);
2059 		dcc->nr_discards -= total_len;
2060 	}
2061 
2062 	wake_up_discard_thread(sbi, false);
2063 }
2064 
2065 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2066 {
2067 	dev_t dev = sbi->sb->s_bdev->bd_dev;
2068 	struct discard_cmd_control *dcc;
2069 	int err = 0, i;
2070 
2071 	if (SM_I(sbi)->dcc_info) {
2072 		dcc = SM_I(sbi)->dcc_info;
2073 		goto init_thread;
2074 	}
2075 
2076 	dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2077 	if (!dcc)
2078 		return -ENOMEM;
2079 
2080 	dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2081 	INIT_LIST_HEAD(&dcc->entry_list);
2082 	for (i = 0; i < MAX_PLIST_NUM; i++)
2083 		INIT_LIST_HEAD(&dcc->pend_list[i]);
2084 	INIT_LIST_HEAD(&dcc->wait_list);
2085 	INIT_LIST_HEAD(&dcc->fstrim_list);
2086 	mutex_init(&dcc->cmd_lock);
2087 	atomic_set(&dcc->issued_discard, 0);
2088 	atomic_set(&dcc->queued_discard, 0);
2089 	atomic_set(&dcc->discard_cmd_cnt, 0);
2090 	dcc->nr_discards = 0;
2091 	dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2092 	dcc->undiscard_blks = 0;
2093 	dcc->next_pos = 0;
2094 	dcc->root = RB_ROOT_CACHED;
2095 	dcc->rbtree_check = false;
2096 
2097 	init_waitqueue_head(&dcc->discard_wait_queue);
2098 	SM_I(sbi)->dcc_info = dcc;
2099 init_thread:
2100 	dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2101 				"f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2102 	if (IS_ERR(dcc->f2fs_issue_discard)) {
2103 		err = PTR_ERR(dcc->f2fs_issue_discard);
2104 		kvfree(dcc);
2105 		SM_I(sbi)->dcc_info = NULL;
2106 		return err;
2107 	}
2108 
2109 	return err;
2110 }
2111 
2112 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2113 {
2114 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2115 
2116 	if (!dcc)
2117 		return;
2118 
2119 	f2fs_stop_discard_thread(sbi);
2120 
2121 	/*
2122 	 * Recovery can cache discard commands, so in error path of
2123 	 * fill_super(), it needs to give a chance to handle them.
2124 	 */
2125 	if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2126 		f2fs_issue_discard_timeout(sbi);
2127 
2128 	kvfree(dcc);
2129 	SM_I(sbi)->dcc_info = NULL;
2130 }
2131 
2132 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2133 {
2134 	struct sit_info *sit_i = SIT_I(sbi);
2135 
2136 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2137 		sit_i->dirty_sentries++;
2138 		return false;
2139 	}
2140 
2141 	return true;
2142 }
2143 
2144 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2145 					unsigned int segno, int modified)
2146 {
2147 	struct seg_entry *se = get_seg_entry(sbi, segno);
2148 	se->type = type;
2149 	if (modified)
2150 		__mark_sit_entry_dirty(sbi, segno);
2151 }
2152 
2153 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2154 {
2155 	struct seg_entry *se;
2156 	unsigned int segno, offset;
2157 	long int new_vblocks;
2158 	bool exist;
2159 #ifdef CONFIG_F2FS_CHECK_FS
2160 	bool mir_exist;
2161 #endif
2162 
2163 	segno = GET_SEGNO(sbi, blkaddr);
2164 
2165 	se = get_seg_entry(sbi, segno);
2166 	new_vblocks = se->valid_blocks + del;
2167 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2168 
2169 	f2fs_bug_on(sbi, (new_vblocks < 0 ||
2170 				(new_vblocks > sbi->blocks_per_seg)));
2171 
2172 	se->valid_blocks = new_vblocks;
2173 	se->mtime = get_mtime(sbi, false);
2174 	if (se->mtime > SIT_I(sbi)->max_mtime)
2175 		SIT_I(sbi)->max_mtime = se->mtime;
2176 
2177 	/* Update valid block bitmap */
2178 	if (del > 0) {
2179 		exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2180 #ifdef CONFIG_F2FS_CHECK_FS
2181 		mir_exist = f2fs_test_and_set_bit(offset,
2182 						se->cur_valid_map_mir);
2183 		if (unlikely(exist != mir_exist)) {
2184 			f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2185 				 blkaddr, exist);
2186 			f2fs_bug_on(sbi, 1);
2187 		}
2188 #endif
2189 		if (unlikely(exist)) {
2190 			f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2191 				 blkaddr);
2192 			f2fs_bug_on(sbi, 1);
2193 			se->valid_blocks--;
2194 			del = 0;
2195 		}
2196 
2197 		if (!f2fs_test_and_set_bit(offset, se->discard_map))
2198 			sbi->discard_blks--;
2199 
2200 		/*
2201 		 * SSR should never reuse block which is checkpointed
2202 		 * or newly invalidated.
2203 		 */
2204 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2205 			if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2206 				se->ckpt_valid_blocks++;
2207 		}
2208 	} else {
2209 		exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2210 #ifdef CONFIG_F2FS_CHECK_FS
2211 		mir_exist = f2fs_test_and_clear_bit(offset,
2212 						se->cur_valid_map_mir);
2213 		if (unlikely(exist != mir_exist)) {
2214 			f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2215 				 blkaddr, exist);
2216 			f2fs_bug_on(sbi, 1);
2217 		}
2218 #endif
2219 		if (unlikely(!exist)) {
2220 			f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2221 				 blkaddr);
2222 			f2fs_bug_on(sbi, 1);
2223 			se->valid_blocks++;
2224 			del = 0;
2225 		} else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2226 			/*
2227 			 * If checkpoints are off, we must not reuse data that
2228 			 * was used in the previous checkpoint. If it was used
2229 			 * before, we must track that to know how much space we
2230 			 * really have.
2231 			 */
2232 			if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2233 				spin_lock(&sbi->stat_lock);
2234 				sbi->unusable_block_count++;
2235 				spin_unlock(&sbi->stat_lock);
2236 			}
2237 		}
2238 
2239 		if (f2fs_test_and_clear_bit(offset, se->discard_map))
2240 			sbi->discard_blks++;
2241 	}
2242 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2243 		se->ckpt_valid_blocks += del;
2244 
2245 	__mark_sit_entry_dirty(sbi, segno);
2246 
2247 	/* update total number of valid blocks to be written in ckpt area */
2248 	SIT_I(sbi)->written_valid_blocks += del;
2249 
2250 	if (__is_large_section(sbi))
2251 		get_sec_entry(sbi, segno)->valid_blocks += del;
2252 }
2253 
2254 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2255 {
2256 	unsigned int segno = GET_SEGNO(sbi, addr);
2257 	struct sit_info *sit_i = SIT_I(sbi);
2258 
2259 	f2fs_bug_on(sbi, addr == NULL_ADDR);
2260 	if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2261 		return;
2262 
2263 	invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2264 
2265 	/* add it into sit main buffer */
2266 	down_write(&sit_i->sentry_lock);
2267 
2268 	update_sit_entry(sbi, addr, -1);
2269 
2270 	/* add it into dirty seglist */
2271 	locate_dirty_segment(sbi, segno);
2272 
2273 	up_write(&sit_i->sentry_lock);
2274 }
2275 
2276 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2277 {
2278 	struct sit_info *sit_i = SIT_I(sbi);
2279 	unsigned int segno, offset;
2280 	struct seg_entry *se;
2281 	bool is_cp = false;
2282 
2283 	if (!__is_valid_data_blkaddr(blkaddr))
2284 		return true;
2285 
2286 	down_read(&sit_i->sentry_lock);
2287 
2288 	segno = GET_SEGNO(sbi, blkaddr);
2289 	se = get_seg_entry(sbi, segno);
2290 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2291 
2292 	if (f2fs_test_bit(offset, se->ckpt_valid_map))
2293 		is_cp = true;
2294 
2295 	up_read(&sit_i->sentry_lock);
2296 
2297 	return is_cp;
2298 }
2299 
2300 /*
2301  * This function should be resided under the curseg_mutex lock
2302  */
2303 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2304 					struct f2fs_summary *sum)
2305 {
2306 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2307 	void *addr = curseg->sum_blk;
2308 	addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2309 	memcpy(addr, sum, sizeof(struct f2fs_summary));
2310 }
2311 
2312 /*
2313  * Calculate the number of current summary pages for writing
2314  */
2315 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2316 {
2317 	int valid_sum_count = 0;
2318 	int i, sum_in_page;
2319 
2320 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2321 		if (sbi->ckpt->alloc_type[i] == SSR)
2322 			valid_sum_count += sbi->blocks_per_seg;
2323 		else {
2324 			if (for_ra)
2325 				valid_sum_count += le16_to_cpu(
2326 					F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2327 			else
2328 				valid_sum_count += curseg_blkoff(sbi, i);
2329 		}
2330 	}
2331 
2332 	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2333 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2334 	if (valid_sum_count <= sum_in_page)
2335 		return 1;
2336 	else if ((valid_sum_count - sum_in_page) <=
2337 		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2338 		return 2;
2339 	return 3;
2340 }
2341 
2342 /*
2343  * Caller should put this summary page
2344  */
2345 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2346 {
2347 	return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2348 }
2349 
2350 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2351 					void *src, block_t blk_addr)
2352 {
2353 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2354 
2355 	memcpy(page_address(page), src, PAGE_SIZE);
2356 	set_page_dirty(page);
2357 	f2fs_put_page(page, 1);
2358 }
2359 
2360 static void write_sum_page(struct f2fs_sb_info *sbi,
2361 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
2362 {
2363 	f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2364 }
2365 
2366 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2367 						int type, block_t blk_addr)
2368 {
2369 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2370 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2371 	struct f2fs_summary_block *src = curseg->sum_blk;
2372 	struct f2fs_summary_block *dst;
2373 
2374 	dst = (struct f2fs_summary_block *)page_address(page);
2375 	memset(dst, 0, PAGE_SIZE);
2376 
2377 	mutex_lock(&curseg->curseg_mutex);
2378 
2379 	down_read(&curseg->journal_rwsem);
2380 	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2381 	up_read(&curseg->journal_rwsem);
2382 
2383 	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2384 	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2385 
2386 	mutex_unlock(&curseg->curseg_mutex);
2387 
2388 	set_page_dirty(page);
2389 	f2fs_put_page(page, 1);
2390 }
2391 
2392 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2393 {
2394 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2395 	unsigned int segno = curseg->segno + 1;
2396 	struct free_segmap_info *free_i = FREE_I(sbi);
2397 
2398 	if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2399 		return !test_bit(segno, free_i->free_segmap);
2400 	return 0;
2401 }
2402 
2403 /*
2404  * Find a new segment from the free segments bitmap to right order
2405  * This function should be returned with success, otherwise BUG
2406  */
2407 static void get_new_segment(struct f2fs_sb_info *sbi,
2408 			unsigned int *newseg, bool new_sec, int dir)
2409 {
2410 	struct free_segmap_info *free_i = FREE_I(sbi);
2411 	unsigned int segno, secno, zoneno;
2412 	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2413 	unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2414 	unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2415 	unsigned int left_start = hint;
2416 	bool init = true;
2417 	int go_left = 0;
2418 	int i;
2419 
2420 	spin_lock(&free_i->segmap_lock);
2421 
2422 	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2423 		segno = find_next_zero_bit(free_i->free_segmap,
2424 			GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2425 		if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2426 			goto got_it;
2427 	}
2428 find_other_zone:
2429 	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2430 	if (secno >= MAIN_SECS(sbi)) {
2431 		if (dir == ALLOC_RIGHT) {
2432 			secno = find_next_zero_bit(free_i->free_secmap,
2433 							MAIN_SECS(sbi), 0);
2434 			f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2435 		} else {
2436 			go_left = 1;
2437 			left_start = hint - 1;
2438 		}
2439 	}
2440 	if (go_left == 0)
2441 		goto skip_left;
2442 
2443 	while (test_bit(left_start, free_i->free_secmap)) {
2444 		if (left_start > 0) {
2445 			left_start--;
2446 			continue;
2447 		}
2448 		left_start = find_next_zero_bit(free_i->free_secmap,
2449 							MAIN_SECS(sbi), 0);
2450 		f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2451 		break;
2452 	}
2453 	secno = left_start;
2454 skip_left:
2455 	segno = GET_SEG_FROM_SEC(sbi, secno);
2456 	zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2457 
2458 	/* give up on finding another zone */
2459 	if (!init)
2460 		goto got_it;
2461 	if (sbi->secs_per_zone == 1)
2462 		goto got_it;
2463 	if (zoneno == old_zoneno)
2464 		goto got_it;
2465 	if (dir == ALLOC_LEFT) {
2466 		if (!go_left && zoneno + 1 >= total_zones)
2467 			goto got_it;
2468 		if (go_left && zoneno == 0)
2469 			goto got_it;
2470 	}
2471 	for (i = 0; i < NR_CURSEG_TYPE; i++)
2472 		if (CURSEG_I(sbi, i)->zone == zoneno)
2473 			break;
2474 
2475 	if (i < NR_CURSEG_TYPE) {
2476 		/* zone is in user, try another */
2477 		if (go_left)
2478 			hint = zoneno * sbi->secs_per_zone - 1;
2479 		else if (zoneno + 1 >= total_zones)
2480 			hint = 0;
2481 		else
2482 			hint = (zoneno + 1) * sbi->secs_per_zone;
2483 		init = false;
2484 		goto find_other_zone;
2485 	}
2486 got_it:
2487 	/* set it as dirty segment in free segmap */
2488 	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2489 	__set_inuse(sbi, segno);
2490 	*newseg = segno;
2491 	spin_unlock(&free_i->segmap_lock);
2492 }
2493 
2494 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2495 {
2496 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2497 	struct summary_footer *sum_footer;
2498 
2499 	curseg->segno = curseg->next_segno;
2500 	curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2501 	curseg->next_blkoff = 0;
2502 	curseg->next_segno = NULL_SEGNO;
2503 
2504 	sum_footer = &(curseg->sum_blk->footer);
2505 	memset(sum_footer, 0, sizeof(struct summary_footer));
2506 	if (IS_DATASEG(type))
2507 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2508 	if (IS_NODESEG(type))
2509 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2510 	__set_sit_entry_type(sbi, type, curseg->segno, modified);
2511 }
2512 
2513 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2514 {
2515 	/* if segs_per_sec is large than 1, we need to keep original policy. */
2516 	if (__is_large_section(sbi))
2517 		return CURSEG_I(sbi, type)->segno;
2518 
2519 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2520 		return 0;
2521 
2522 	if (test_opt(sbi, NOHEAP) &&
2523 		(type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2524 		return 0;
2525 
2526 	if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2527 		return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2528 
2529 	/* find segments from 0 to reuse freed segments */
2530 	if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2531 		return 0;
2532 
2533 	return CURSEG_I(sbi, type)->segno;
2534 }
2535 
2536 /*
2537  * Allocate a current working segment.
2538  * This function always allocates a free segment in LFS manner.
2539  */
2540 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2541 {
2542 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2543 	unsigned int segno = curseg->segno;
2544 	int dir = ALLOC_LEFT;
2545 
2546 	write_sum_page(sbi, curseg->sum_blk,
2547 				GET_SUM_BLOCK(sbi, segno));
2548 	if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2549 		dir = ALLOC_RIGHT;
2550 
2551 	if (test_opt(sbi, NOHEAP))
2552 		dir = ALLOC_RIGHT;
2553 
2554 	segno = __get_next_segno(sbi, type);
2555 	get_new_segment(sbi, &segno, new_sec, dir);
2556 	curseg->next_segno = segno;
2557 	reset_curseg(sbi, type, 1);
2558 	curseg->alloc_type = LFS;
2559 }
2560 
2561 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2562 			struct curseg_info *seg, block_t start)
2563 {
2564 	struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2565 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2566 	unsigned long *target_map = SIT_I(sbi)->tmp_map;
2567 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2568 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2569 	int i, pos;
2570 
2571 	for (i = 0; i < entries; i++)
2572 		target_map[i] = ckpt_map[i] | cur_map[i];
2573 
2574 	pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2575 
2576 	seg->next_blkoff = pos;
2577 }
2578 
2579 /*
2580  * If a segment is written by LFS manner, next block offset is just obtained
2581  * by increasing the current block offset. However, if a segment is written by
2582  * SSR manner, next block offset obtained by calling __next_free_blkoff
2583  */
2584 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2585 				struct curseg_info *seg)
2586 {
2587 	if (seg->alloc_type == SSR)
2588 		__next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2589 	else
2590 		seg->next_blkoff++;
2591 }
2592 
2593 /*
2594  * This function always allocates a used segment(from dirty seglist) by SSR
2595  * manner, so it should recover the existing segment information of valid blocks
2596  */
2597 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2598 {
2599 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2600 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2601 	unsigned int new_segno = curseg->next_segno;
2602 	struct f2fs_summary_block *sum_node;
2603 	struct page *sum_page;
2604 
2605 	write_sum_page(sbi, curseg->sum_blk,
2606 				GET_SUM_BLOCK(sbi, curseg->segno));
2607 	__set_test_and_inuse(sbi, new_segno);
2608 
2609 	mutex_lock(&dirty_i->seglist_lock);
2610 	__remove_dirty_segment(sbi, new_segno, PRE);
2611 	__remove_dirty_segment(sbi, new_segno, DIRTY);
2612 	mutex_unlock(&dirty_i->seglist_lock);
2613 
2614 	reset_curseg(sbi, type, 1);
2615 	curseg->alloc_type = SSR;
2616 	__next_free_blkoff(sbi, curseg, 0);
2617 
2618 	sum_page = f2fs_get_sum_page(sbi, new_segno);
2619 	f2fs_bug_on(sbi, IS_ERR(sum_page));
2620 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2621 	memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2622 	f2fs_put_page(sum_page, 1);
2623 }
2624 
2625 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2626 {
2627 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2628 	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2629 	unsigned segno = NULL_SEGNO;
2630 	int i, cnt;
2631 	bool reversed = false;
2632 
2633 	/* f2fs_need_SSR() already forces to do this */
2634 	if (!v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2635 		curseg->next_segno = segno;
2636 		return 1;
2637 	}
2638 
2639 	/* For node segments, let's do SSR more intensively */
2640 	if (IS_NODESEG(type)) {
2641 		if (type >= CURSEG_WARM_NODE) {
2642 			reversed = true;
2643 			i = CURSEG_COLD_NODE;
2644 		} else {
2645 			i = CURSEG_HOT_NODE;
2646 		}
2647 		cnt = NR_CURSEG_NODE_TYPE;
2648 	} else {
2649 		if (type >= CURSEG_WARM_DATA) {
2650 			reversed = true;
2651 			i = CURSEG_COLD_DATA;
2652 		} else {
2653 			i = CURSEG_HOT_DATA;
2654 		}
2655 		cnt = NR_CURSEG_DATA_TYPE;
2656 	}
2657 
2658 	for (; cnt-- > 0; reversed ? i-- : i++) {
2659 		if (i == type)
2660 			continue;
2661 		if (!v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2662 			curseg->next_segno = segno;
2663 			return 1;
2664 		}
2665 	}
2666 
2667 	/* find valid_blocks=0 in dirty list */
2668 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2669 		segno = get_free_segment(sbi);
2670 		if (segno != NULL_SEGNO) {
2671 			curseg->next_segno = segno;
2672 			return 1;
2673 		}
2674 	}
2675 	return 0;
2676 }
2677 
2678 /*
2679  * flush out current segment and replace it with new segment
2680  * This function should be returned with success, otherwise BUG
2681  */
2682 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2683 						int type, bool force)
2684 {
2685 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2686 
2687 	if (force)
2688 		new_curseg(sbi, type, true);
2689 	else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2690 					type == CURSEG_WARM_NODE)
2691 		new_curseg(sbi, type, false);
2692 	else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2693 			likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2694 		new_curseg(sbi, type, false);
2695 	else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2696 		change_curseg(sbi, type);
2697 	else
2698 		new_curseg(sbi, type, false);
2699 
2700 	stat_inc_seg_type(sbi, curseg);
2701 }
2702 
2703 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2704 					unsigned int start, unsigned int end)
2705 {
2706 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2707 	unsigned int segno;
2708 
2709 	down_read(&SM_I(sbi)->curseg_lock);
2710 	mutex_lock(&curseg->curseg_mutex);
2711 	down_write(&SIT_I(sbi)->sentry_lock);
2712 
2713 	segno = CURSEG_I(sbi, type)->segno;
2714 	if (segno < start || segno > end)
2715 		goto unlock;
2716 
2717 	if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2718 		change_curseg(sbi, type);
2719 	else
2720 		new_curseg(sbi, type, true);
2721 
2722 	stat_inc_seg_type(sbi, curseg);
2723 
2724 	locate_dirty_segment(sbi, segno);
2725 unlock:
2726 	up_write(&SIT_I(sbi)->sentry_lock);
2727 
2728 	if (segno != curseg->segno)
2729 		f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2730 			    type, segno, curseg->segno);
2731 
2732 	mutex_unlock(&curseg->curseg_mutex);
2733 	up_read(&SM_I(sbi)->curseg_lock);
2734 }
2735 
2736 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type)
2737 {
2738 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2739 	unsigned int old_segno;
2740 
2741 	if (!curseg->next_blkoff &&
2742 		!get_valid_blocks(sbi, curseg->segno, false) &&
2743 		!get_ckpt_valid_blocks(sbi, curseg->segno))
2744 		return;
2745 
2746 	old_segno = curseg->segno;
2747 	SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
2748 	locate_dirty_segment(sbi, old_segno);
2749 }
2750 
2751 void f2fs_allocate_new_segment(struct f2fs_sb_info *sbi, int type)
2752 {
2753 	down_write(&SIT_I(sbi)->sentry_lock);
2754 	__allocate_new_segment(sbi, type);
2755 	up_write(&SIT_I(sbi)->sentry_lock);
2756 }
2757 
2758 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2759 {
2760 	int i;
2761 
2762 	down_write(&SIT_I(sbi)->sentry_lock);
2763 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
2764 		__allocate_new_segment(sbi, i);
2765 	up_write(&SIT_I(sbi)->sentry_lock);
2766 }
2767 
2768 static const struct segment_allocation default_salloc_ops = {
2769 	.allocate_segment = allocate_segment_by_default,
2770 };
2771 
2772 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2773 						struct cp_control *cpc)
2774 {
2775 	__u64 trim_start = cpc->trim_start;
2776 	bool has_candidate = false;
2777 
2778 	down_write(&SIT_I(sbi)->sentry_lock);
2779 	for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2780 		if (add_discard_addrs(sbi, cpc, true)) {
2781 			has_candidate = true;
2782 			break;
2783 		}
2784 	}
2785 	up_write(&SIT_I(sbi)->sentry_lock);
2786 
2787 	cpc->trim_start = trim_start;
2788 	return has_candidate;
2789 }
2790 
2791 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2792 					struct discard_policy *dpolicy,
2793 					unsigned int start, unsigned int end)
2794 {
2795 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2796 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2797 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
2798 	struct discard_cmd *dc;
2799 	struct blk_plug plug;
2800 	int issued;
2801 	unsigned int trimmed = 0;
2802 
2803 next:
2804 	issued = 0;
2805 
2806 	mutex_lock(&dcc->cmd_lock);
2807 	if (unlikely(dcc->rbtree_check))
2808 		f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2809 								&dcc->root));
2810 
2811 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2812 					NULL, start,
2813 					(struct rb_entry **)&prev_dc,
2814 					(struct rb_entry **)&next_dc,
2815 					&insert_p, &insert_parent, true, NULL);
2816 	if (!dc)
2817 		dc = next_dc;
2818 
2819 	blk_start_plug(&plug);
2820 
2821 	while (dc && dc->lstart <= end) {
2822 		struct rb_node *node;
2823 		int err = 0;
2824 
2825 		if (dc->len < dpolicy->granularity)
2826 			goto skip;
2827 
2828 		if (dc->state != D_PREP) {
2829 			list_move_tail(&dc->list, &dcc->fstrim_list);
2830 			goto skip;
2831 		}
2832 
2833 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2834 
2835 		if (issued >= dpolicy->max_requests) {
2836 			start = dc->lstart + dc->len;
2837 
2838 			if (err)
2839 				__remove_discard_cmd(sbi, dc);
2840 
2841 			blk_finish_plug(&plug);
2842 			mutex_unlock(&dcc->cmd_lock);
2843 			trimmed += __wait_all_discard_cmd(sbi, NULL);
2844 			congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2845 			goto next;
2846 		}
2847 skip:
2848 		node = rb_next(&dc->rb_node);
2849 		if (err)
2850 			__remove_discard_cmd(sbi, dc);
2851 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2852 
2853 		if (fatal_signal_pending(current))
2854 			break;
2855 	}
2856 
2857 	blk_finish_plug(&plug);
2858 	mutex_unlock(&dcc->cmd_lock);
2859 
2860 	return trimmed;
2861 }
2862 
2863 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2864 {
2865 	__u64 start = F2FS_BYTES_TO_BLK(range->start);
2866 	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2867 	unsigned int start_segno, end_segno;
2868 	block_t start_block, end_block;
2869 	struct cp_control cpc;
2870 	struct discard_policy dpolicy;
2871 	unsigned long long trimmed = 0;
2872 	int err = 0;
2873 	bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
2874 
2875 	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2876 		return -EINVAL;
2877 
2878 	if (end < MAIN_BLKADDR(sbi))
2879 		goto out;
2880 
2881 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2882 		f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
2883 		return -EFSCORRUPTED;
2884 	}
2885 
2886 	/* start/end segment number in main_area */
2887 	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2888 	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2889 						GET_SEGNO(sbi, end);
2890 	if (need_align) {
2891 		start_segno = rounddown(start_segno, sbi->segs_per_sec);
2892 		end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2893 	}
2894 
2895 	cpc.reason = CP_DISCARD;
2896 	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2897 	cpc.trim_start = start_segno;
2898 	cpc.trim_end = end_segno;
2899 
2900 	if (sbi->discard_blks == 0)
2901 		goto out;
2902 
2903 	down_write(&sbi->gc_lock);
2904 	err = f2fs_write_checkpoint(sbi, &cpc);
2905 	up_write(&sbi->gc_lock);
2906 	if (err)
2907 		goto out;
2908 
2909 	/*
2910 	 * We filed discard candidates, but actually we don't need to wait for
2911 	 * all of them, since they'll be issued in idle time along with runtime
2912 	 * discard option. User configuration looks like using runtime discard
2913 	 * or periodic fstrim instead of it.
2914 	 */
2915 	if (f2fs_realtime_discard_enable(sbi))
2916 		goto out;
2917 
2918 	start_block = START_BLOCK(sbi, start_segno);
2919 	end_block = START_BLOCK(sbi, end_segno + 1);
2920 
2921 	__init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2922 	trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2923 					start_block, end_block);
2924 
2925 	trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2926 					start_block, end_block);
2927 out:
2928 	if (!err)
2929 		range->len = F2FS_BLK_TO_BYTES(trimmed);
2930 	return err;
2931 }
2932 
2933 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2934 {
2935 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2936 	if (curseg->next_blkoff < sbi->blocks_per_seg)
2937 		return true;
2938 	return false;
2939 }
2940 
2941 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2942 {
2943 	switch (hint) {
2944 	case WRITE_LIFE_SHORT:
2945 		return CURSEG_HOT_DATA;
2946 	case WRITE_LIFE_EXTREME:
2947 		return CURSEG_COLD_DATA;
2948 	default:
2949 		return CURSEG_WARM_DATA;
2950 	}
2951 }
2952 
2953 /* This returns write hints for each segment type. This hints will be
2954  * passed down to block layer. There are mapping tables which depend on
2955  * the mount option 'whint_mode'.
2956  *
2957  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2958  *
2959  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2960  *
2961  * User                  F2FS                     Block
2962  * ----                  ----                     -----
2963  *                       META                     WRITE_LIFE_NOT_SET
2964  *                       HOT_NODE                 "
2965  *                       WARM_NODE                "
2966  *                       COLD_NODE                "
2967  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2968  * extension list        "                        "
2969  *
2970  * -- buffered io
2971  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2972  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2973  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2974  * WRITE_LIFE_NONE       "                        "
2975  * WRITE_LIFE_MEDIUM     "                        "
2976  * WRITE_LIFE_LONG       "                        "
2977  *
2978  * -- direct io
2979  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2980  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2981  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2982  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2983  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2984  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2985  *
2986  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2987  *
2988  * User                  F2FS                     Block
2989  * ----                  ----                     -----
2990  *                       META                     WRITE_LIFE_MEDIUM;
2991  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2992  *                       WARM_NODE                "
2993  *                       COLD_NODE                WRITE_LIFE_NONE
2994  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2995  * extension list        "                        "
2996  *
2997  * -- buffered io
2998  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2999  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3000  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
3001  * WRITE_LIFE_NONE       "                        "
3002  * WRITE_LIFE_MEDIUM     "                        "
3003  * WRITE_LIFE_LONG       "                        "
3004  *
3005  * -- direct io
3006  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3007  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3008  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3009  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
3010  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
3011  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
3012  */
3013 
3014 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
3015 				enum page_type type, enum temp_type temp)
3016 {
3017 	if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
3018 		if (type == DATA) {
3019 			if (temp == WARM)
3020 				return WRITE_LIFE_NOT_SET;
3021 			else if (temp == HOT)
3022 				return WRITE_LIFE_SHORT;
3023 			else if (temp == COLD)
3024 				return WRITE_LIFE_EXTREME;
3025 		} else {
3026 			return WRITE_LIFE_NOT_SET;
3027 		}
3028 	} else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
3029 		if (type == DATA) {
3030 			if (temp == WARM)
3031 				return WRITE_LIFE_LONG;
3032 			else if (temp == HOT)
3033 				return WRITE_LIFE_SHORT;
3034 			else if (temp == COLD)
3035 				return WRITE_LIFE_EXTREME;
3036 		} else if (type == NODE) {
3037 			if (temp == WARM || temp == HOT)
3038 				return WRITE_LIFE_NOT_SET;
3039 			else if (temp == COLD)
3040 				return WRITE_LIFE_NONE;
3041 		} else if (type == META) {
3042 			return WRITE_LIFE_MEDIUM;
3043 		}
3044 	}
3045 	return WRITE_LIFE_NOT_SET;
3046 }
3047 
3048 static int __get_segment_type_2(struct f2fs_io_info *fio)
3049 {
3050 	if (fio->type == DATA)
3051 		return CURSEG_HOT_DATA;
3052 	else
3053 		return CURSEG_HOT_NODE;
3054 }
3055 
3056 static int __get_segment_type_4(struct f2fs_io_info *fio)
3057 {
3058 	if (fio->type == DATA) {
3059 		struct inode *inode = fio->page->mapping->host;
3060 
3061 		if (S_ISDIR(inode->i_mode))
3062 			return CURSEG_HOT_DATA;
3063 		else
3064 			return CURSEG_COLD_DATA;
3065 	} else {
3066 		if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3067 			return CURSEG_WARM_NODE;
3068 		else
3069 			return CURSEG_COLD_NODE;
3070 	}
3071 }
3072 
3073 static int __get_segment_type_6(struct f2fs_io_info *fio)
3074 {
3075 	if (fio->type == DATA) {
3076 		struct inode *inode = fio->page->mapping->host;
3077 
3078 		if (is_cold_data(fio->page) || file_is_cold(inode) ||
3079 				f2fs_compressed_file(inode))
3080 			return CURSEG_COLD_DATA;
3081 		if (file_is_hot(inode) ||
3082 				is_inode_flag_set(inode, FI_HOT_DATA) ||
3083 				f2fs_is_atomic_file(inode) ||
3084 				f2fs_is_volatile_file(inode))
3085 			return CURSEG_HOT_DATA;
3086 		return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3087 	} else {
3088 		if (IS_DNODE(fio->page))
3089 			return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3090 						CURSEG_HOT_NODE;
3091 		return CURSEG_COLD_NODE;
3092 	}
3093 }
3094 
3095 static int __get_segment_type(struct f2fs_io_info *fio)
3096 {
3097 	int type = 0;
3098 
3099 	switch (F2FS_OPTION(fio->sbi).active_logs) {
3100 	case 2:
3101 		type = __get_segment_type_2(fio);
3102 		break;
3103 	case 4:
3104 		type = __get_segment_type_4(fio);
3105 		break;
3106 	case 6:
3107 		type = __get_segment_type_6(fio);
3108 		break;
3109 	default:
3110 		f2fs_bug_on(fio->sbi, true);
3111 	}
3112 
3113 	if (IS_HOT(type))
3114 		fio->temp = HOT;
3115 	else if (IS_WARM(type))
3116 		fio->temp = WARM;
3117 	else
3118 		fio->temp = COLD;
3119 	return type;
3120 }
3121 
3122 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3123 		block_t old_blkaddr, block_t *new_blkaddr,
3124 		struct f2fs_summary *sum, int type,
3125 		struct f2fs_io_info *fio)
3126 {
3127 	struct sit_info *sit_i = SIT_I(sbi);
3128 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3129 	bool put_pin_sem = false;
3130 
3131 	if (type == CURSEG_COLD_DATA) {
3132 		/* GC during CURSEG_COLD_DATA_PINNED allocation */
3133 		if (down_read_trylock(&sbi->pin_sem)) {
3134 			put_pin_sem = true;
3135 		} else {
3136 			type = CURSEG_WARM_DATA;
3137 			curseg = CURSEG_I(sbi, type);
3138 		}
3139 	} else if (type == CURSEG_COLD_DATA_PINNED) {
3140 		type = CURSEG_COLD_DATA;
3141 	}
3142 
3143 	down_read(&SM_I(sbi)->curseg_lock);
3144 
3145 	mutex_lock(&curseg->curseg_mutex);
3146 	down_write(&sit_i->sentry_lock);
3147 
3148 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3149 
3150 	f2fs_wait_discard_bio(sbi, *new_blkaddr);
3151 
3152 	/*
3153 	 * __add_sum_entry should be resided under the curseg_mutex
3154 	 * because, this function updates a summary entry in the
3155 	 * current summary block.
3156 	 */
3157 	__add_sum_entry(sbi, type, sum);
3158 
3159 	__refresh_next_blkoff(sbi, curseg);
3160 
3161 	stat_inc_block_count(sbi, curseg);
3162 
3163 	/*
3164 	 * SIT information should be updated before segment allocation,
3165 	 * since SSR needs latest valid block information.
3166 	 */
3167 	update_sit_entry(sbi, *new_blkaddr, 1);
3168 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3169 		update_sit_entry(sbi, old_blkaddr, -1);
3170 
3171 	if (!__has_curseg_space(sbi, type))
3172 		sit_i->s_ops->allocate_segment(sbi, type, false);
3173 
3174 	/*
3175 	 * segment dirty status should be updated after segment allocation,
3176 	 * so we just need to update status only one time after previous
3177 	 * segment being closed.
3178 	 */
3179 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3180 	locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3181 
3182 	up_write(&sit_i->sentry_lock);
3183 
3184 	if (page && IS_NODESEG(type)) {
3185 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3186 
3187 		f2fs_inode_chksum_set(sbi, page);
3188 	}
3189 
3190 	if (F2FS_IO_ALIGNED(sbi))
3191 		fio->retry = false;
3192 
3193 	if (fio) {
3194 		struct f2fs_bio_info *io;
3195 
3196 		INIT_LIST_HEAD(&fio->list);
3197 		fio->in_list = true;
3198 		io = sbi->write_io[fio->type] + fio->temp;
3199 		spin_lock(&io->io_lock);
3200 		list_add_tail(&fio->list, &io->io_list);
3201 		spin_unlock(&io->io_lock);
3202 	}
3203 
3204 	mutex_unlock(&curseg->curseg_mutex);
3205 
3206 	up_read(&SM_I(sbi)->curseg_lock);
3207 
3208 	if (put_pin_sem)
3209 		up_read(&sbi->pin_sem);
3210 }
3211 
3212 static void update_device_state(struct f2fs_io_info *fio)
3213 {
3214 	struct f2fs_sb_info *sbi = fio->sbi;
3215 	unsigned int devidx;
3216 
3217 	if (!f2fs_is_multi_device(sbi))
3218 		return;
3219 
3220 	devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3221 
3222 	/* update device state for fsync */
3223 	f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3224 
3225 	/* update device state for checkpoint */
3226 	if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3227 		spin_lock(&sbi->dev_lock);
3228 		f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3229 		spin_unlock(&sbi->dev_lock);
3230 	}
3231 }
3232 
3233 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3234 {
3235 	int type = __get_segment_type(fio);
3236 	bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3237 
3238 	if (keep_order)
3239 		down_read(&fio->sbi->io_order_lock);
3240 reallocate:
3241 	f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3242 			&fio->new_blkaddr, sum, type, fio);
3243 	if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3244 		invalidate_mapping_pages(META_MAPPING(fio->sbi),
3245 					fio->old_blkaddr, fio->old_blkaddr);
3246 
3247 	/* writeout dirty page into bdev */
3248 	f2fs_submit_page_write(fio);
3249 	if (fio->retry) {
3250 		fio->old_blkaddr = fio->new_blkaddr;
3251 		goto reallocate;
3252 	}
3253 
3254 	update_device_state(fio);
3255 
3256 	if (keep_order)
3257 		up_read(&fio->sbi->io_order_lock);
3258 }
3259 
3260 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3261 					enum iostat_type io_type)
3262 {
3263 	struct f2fs_io_info fio = {
3264 		.sbi = sbi,
3265 		.type = META,
3266 		.temp = HOT,
3267 		.op = REQ_OP_WRITE,
3268 		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3269 		.old_blkaddr = page->index,
3270 		.new_blkaddr = page->index,
3271 		.page = page,
3272 		.encrypted_page = NULL,
3273 		.in_list = false,
3274 	};
3275 
3276 	if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3277 		fio.op_flags &= ~REQ_META;
3278 
3279 	set_page_writeback(page);
3280 	ClearPageError(page);
3281 	f2fs_submit_page_write(&fio);
3282 
3283 	stat_inc_meta_count(sbi, page->index);
3284 	f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3285 }
3286 
3287 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3288 {
3289 	struct f2fs_summary sum;
3290 
3291 	set_summary(&sum, nid, 0, 0);
3292 	do_write_page(&sum, fio);
3293 
3294 	f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3295 }
3296 
3297 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3298 					struct f2fs_io_info *fio)
3299 {
3300 	struct f2fs_sb_info *sbi = fio->sbi;
3301 	struct f2fs_summary sum;
3302 
3303 	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3304 	set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3305 	do_write_page(&sum, fio);
3306 	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3307 
3308 	f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3309 }
3310 
3311 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3312 {
3313 	int err;
3314 	struct f2fs_sb_info *sbi = fio->sbi;
3315 	unsigned int segno;
3316 
3317 	fio->new_blkaddr = fio->old_blkaddr;
3318 	/* i/o temperature is needed for passing down write hints */
3319 	__get_segment_type(fio);
3320 
3321 	segno = GET_SEGNO(sbi, fio->new_blkaddr);
3322 
3323 	if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3324 		set_sbi_flag(sbi, SBI_NEED_FSCK);
3325 		f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3326 			  __func__, segno);
3327 		return -EFSCORRUPTED;
3328 	}
3329 
3330 	stat_inc_inplace_blocks(fio->sbi);
3331 
3332 	if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE)))
3333 		err = f2fs_merge_page_bio(fio);
3334 	else
3335 		err = f2fs_submit_page_bio(fio);
3336 	if (!err) {
3337 		update_device_state(fio);
3338 		f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3339 	}
3340 
3341 	return err;
3342 }
3343 
3344 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3345 						unsigned int segno)
3346 {
3347 	int i;
3348 
3349 	for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3350 		if (CURSEG_I(sbi, i)->segno == segno)
3351 			break;
3352 	}
3353 	return i;
3354 }
3355 
3356 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3357 				block_t old_blkaddr, block_t new_blkaddr,
3358 				bool recover_curseg, bool recover_newaddr)
3359 {
3360 	struct sit_info *sit_i = SIT_I(sbi);
3361 	struct curseg_info *curseg;
3362 	unsigned int segno, old_cursegno;
3363 	struct seg_entry *se;
3364 	int type;
3365 	unsigned short old_blkoff;
3366 
3367 	segno = GET_SEGNO(sbi, new_blkaddr);
3368 	se = get_seg_entry(sbi, segno);
3369 	type = se->type;
3370 
3371 	down_write(&SM_I(sbi)->curseg_lock);
3372 
3373 	if (!recover_curseg) {
3374 		/* for recovery flow */
3375 		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3376 			if (old_blkaddr == NULL_ADDR)
3377 				type = CURSEG_COLD_DATA;
3378 			else
3379 				type = CURSEG_WARM_DATA;
3380 		}
3381 	} else {
3382 		if (IS_CURSEG(sbi, segno)) {
3383 			/* se->type is volatile as SSR allocation */
3384 			type = __f2fs_get_curseg(sbi, segno);
3385 			f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3386 		} else {
3387 			type = CURSEG_WARM_DATA;
3388 		}
3389 	}
3390 
3391 	f2fs_bug_on(sbi, !IS_DATASEG(type));
3392 	curseg = CURSEG_I(sbi, type);
3393 
3394 	mutex_lock(&curseg->curseg_mutex);
3395 	down_write(&sit_i->sentry_lock);
3396 
3397 	old_cursegno = curseg->segno;
3398 	old_blkoff = curseg->next_blkoff;
3399 
3400 	/* change the current segment */
3401 	if (segno != curseg->segno) {
3402 		curseg->next_segno = segno;
3403 		change_curseg(sbi, type);
3404 	}
3405 
3406 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3407 	__add_sum_entry(sbi, type, sum);
3408 
3409 	if (!recover_curseg || recover_newaddr)
3410 		update_sit_entry(sbi, new_blkaddr, 1);
3411 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3412 		invalidate_mapping_pages(META_MAPPING(sbi),
3413 					old_blkaddr, old_blkaddr);
3414 		update_sit_entry(sbi, old_blkaddr, -1);
3415 	}
3416 
3417 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3418 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3419 
3420 	locate_dirty_segment(sbi, old_cursegno);
3421 
3422 	if (recover_curseg) {
3423 		if (old_cursegno != curseg->segno) {
3424 			curseg->next_segno = old_cursegno;
3425 			change_curseg(sbi, type);
3426 		}
3427 		curseg->next_blkoff = old_blkoff;
3428 	}
3429 
3430 	up_write(&sit_i->sentry_lock);
3431 	mutex_unlock(&curseg->curseg_mutex);
3432 	up_write(&SM_I(sbi)->curseg_lock);
3433 }
3434 
3435 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3436 				block_t old_addr, block_t new_addr,
3437 				unsigned char version, bool recover_curseg,
3438 				bool recover_newaddr)
3439 {
3440 	struct f2fs_summary sum;
3441 
3442 	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3443 
3444 	f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3445 					recover_curseg, recover_newaddr);
3446 
3447 	f2fs_update_data_blkaddr(dn, new_addr);
3448 }
3449 
3450 void f2fs_wait_on_page_writeback(struct page *page,
3451 				enum page_type type, bool ordered, bool locked)
3452 {
3453 	if (PageWriteback(page)) {
3454 		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3455 
3456 		/* submit cached LFS IO */
3457 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3458 		/* sbumit cached IPU IO */
3459 		f2fs_submit_merged_ipu_write(sbi, NULL, page);
3460 		if (ordered) {
3461 			wait_on_page_writeback(page);
3462 			f2fs_bug_on(sbi, locked && PageWriteback(page));
3463 		} else {
3464 			wait_for_stable_page(page);
3465 		}
3466 	}
3467 }
3468 
3469 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3470 {
3471 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3472 	struct page *cpage;
3473 
3474 	if (!f2fs_post_read_required(inode))
3475 		return;
3476 
3477 	if (!__is_valid_data_blkaddr(blkaddr))
3478 		return;
3479 
3480 	cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3481 	if (cpage) {
3482 		f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3483 		f2fs_put_page(cpage, 1);
3484 	}
3485 }
3486 
3487 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3488 								block_t len)
3489 {
3490 	block_t i;
3491 
3492 	for (i = 0; i < len; i++)
3493 		f2fs_wait_on_block_writeback(inode, blkaddr + i);
3494 }
3495 
3496 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3497 {
3498 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3499 	struct curseg_info *seg_i;
3500 	unsigned char *kaddr;
3501 	struct page *page;
3502 	block_t start;
3503 	int i, j, offset;
3504 
3505 	start = start_sum_block(sbi);
3506 
3507 	page = f2fs_get_meta_page(sbi, start++);
3508 	if (IS_ERR(page))
3509 		return PTR_ERR(page);
3510 	kaddr = (unsigned char *)page_address(page);
3511 
3512 	/* Step 1: restore nat cache */
3513 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3514 	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3515 
3516 	/* Step 2: restore sit cache */
3517 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3518 	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3519 	offset = 2 * SUM_JOURNAL_SIZE;
3520 
3521 	/* Step 3: restore summary entries */
3522 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3523 		unsigned short blk_off;
3524 		unsigned int segno;
3525 
3526 		seg_i = CURSEG_I(sbi, i);
3527 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3528 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3529 		seg_i->next_segno = segno;
3530 		reset_curseg(sbi, i, 0);
3531 		seg_i->alloc_type = ckpt->alloc_type[i];
3532 		seg_i->next_blkoff = blk_off;
3533 
3534 		if (seg_i->alloc_type == SSR)
3535 			blk_off = sbi->blocks_per_seg;
3536 
3537 		for (j = 0; j < blk_off; j++) {
3538 			struct f2fs_summary *s;
3539 			s = (struct f2fs_summary *)(kaddr + offset);
3540 			seg_i->sum_blk->entries[j] = *s;
3541 			offset += SUMMARY_SIZE;
3542 			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3543 						SUM_FOOTER_SIZE)
3544 				continue;
3545 
3546 			f2fs_put_page(page, 1);
3547 			page = NULL;
3548 
3549 			page = f2fs_get_meta_page(sbi, start++);
3550 			if (IS_ERR(page))
3551 				return PTR_ERR(page);
3552 			kaddr = (unsigned char *)page_address(page);
3553 			offset = 0;
3554 		}
3555 	}
3556 	f2fs_put_page(page, 1);
3557 	return 0;
3558 }
3559 
3560 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3561 {
3562 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3563 	struct f2fs_summary_block *sum;
3564 	struct curseg_info *curseg;
3565 	struct page *new;
3566 	unsigned short blk_off;
3567 	unsigned int segno = 0;
3568 	block_t blk_addr = 0;
3569 	int err = 0;
3570 
3571 	/* get segment number and block addr */
3572 	if (IS_DATASEG(type)) {
3573 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3574 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3575 							CURSEG_HOT_DATA]);
3576 		if (__exist_node_summaries(sbi))
3577 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3578 		else
3579 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3580 	} else {
3581 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
3582 							CURSEG_HOT_NODE]);
3583 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3584 							CURSEG_HOT_NODE]);
3585 		if (__exist_node_summaries(sbi))
3586 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3587 							type - CURSEG_HOT_NODE);
3588 		else
3589 			blk_addr = GET_SUM_BLOCK(sbi, segno);
3590 	}
3591 
3592 	new = f2fs_get_meta_page(sbi, blk_addr);
3593 	if (IS_ERR(new))
3594 		return PTR_ERR(new);
3595 	sum = (struct f2fs_summary_block *)page_address(new);
3596 
3597 	if (IS_NODESEG(type)) {
3598 		if (__exist_node_summaries(sbi)) {
3599 			struct f2fs_summary *ns = &sum->entries[0];
3600 			int i;
3601 			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3602 				ns->version = 0;
3603 				ns->ofs_in_node = 0;
3604 			}
3605 		} else {
3606 			err = f2fs_restore_node_summary(sbi, segno, sum);
3607 			if (err)
3608 				goto out;
3609 		}
3610 	}
3611 
3612 	/* set uncompleted segment to curseg */
3613 	curseg = CURSEG_I(sbi, type);
3614 	mutex_lock(&curseg->curseg_mutex);
3615 
3616 	/* update journal info */
3617 	down_write(&curseg->journal_rwsem);
3618 	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3619 	up_write(&curseg->journal_rwsem);
3620 
3621 	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3622 	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3623 	curseg->next_segno = segno;
3624 	reset_curseg(sbi, type, 0);
3625 	curseg->alloc_type = ckpt->alloc_type[type];
3626 	curseg->next_blkoff = blk_off;
3627 	mutex_unlock(&curseg->curseg_mutex);
3628 out:
3629 	f2fs_put_page(new, 1);
3630 	return err;
3631 }
3632 
3633 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3634 {
3635 	struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3636 	struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3637 	int type = CURSEG_HOT_DATA;
3638 	int err;
3639 
3640 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3641 		int npages = f2fs_npages_for_summary_flush(sbi, true);
3642 
3643 		if (npages >= 2)
3644 			f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3645 							META_CP, true);
3646 
3647 		/* restore for compacted data summary */
3648 		err = read_compacted_summaries(sbi);
3649 		if (err)
3650 			return err;
3651 		type = CURSEG_HOT_NODE;
3652 	}
3653 
3654 	if (__exist_node_summaries(sbi))
3655 		f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3656 					NR_CURSEG_TYPE - type, META_CP, true);
3657 
3658 	for (; type <= CURSEG_COLD_NODE; type++) {
3659 		err = read_normal_summaries(sbi, type);
3660 		if (err)
3661 			return err;
3662 	}
3663 
3664 	/* sanity check for summary blocks */
3665 	if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3666 			sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3667 		f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3668 			 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3669 		return -EINVAL;
3670 	}
3671 
3672 	return 0;
3673 }
3674 
3675 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3676 {
3677 	struct page *page;
3678 	unsigned char *kaddr;
3679 	struct f2fs_summary *summary;
3680 	struct curseg_info *seg_i;
3681 	int written_size = 0;
3682 	int i, j;
3683 
3684 	page = f2fs_grab_meta_page(sbi, blkaddr++);
3685 	kaddr = (unsigned char *)page_address(page);
3686 	memset(kaddr, 0, PAGE_SIZE);
3687 
3688 	/* Step 1: write nat cache */
3689 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3690 	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3691 	written_size += SUM_JOURNAL_SIZE;
3692 
3693 	/* Step 2: write sit cache */
3694 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3695 	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3696 	written_size += SUM_JOURNAL_SIZE;
3697 
3698 	/* Step 3: write summary entries */
3699 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3700 		unsigned short blkoff;
3701 		seg_i = CURSEG_I(sbi, i);
3702 		if (sbi->ckpt->alloc_type[i] == SSR)
3703 			blkoff = sbi->blocks_per_seg;
3704 		else
3705 			blkoff = curseg_blkoff(sbi, i);
3706 
3707 		for (j = 0; j < blkoff; j++) {
3708 			if (!page) {
3709 				page = f2fs_grab_meta_page(sbi, blkaddr++);
3710 				kaddr = (unsigned char *)page_address(page);
3711 				memset(kaddr, 0, PAGE_SIZE);
3712 				written_size = 0;
3713 			}
3714 			summary = (struct f2fs_summary *)(kaddr + written_size);
3715 			*summary = seg_i->sum_blk->entries[j];
3716 			written_size += SUMMARY_SIZE;
3717 
3718 			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3719 							SUM_FOOTER_SIZE)
3720 				continue;
3721 
3722 			set_page_dirty(page);
3723 			f2fs_put_page(page, 1);
3724 			page = NULL;
3725 		}
3726 	}
3727 	if (page) {
3728 		set_page_dirty(page);
3729 		f2fs_put_page(page, 1);
3730 	}
3731 }
3732 
3733 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3734 					block_t blkaddr, int type)
3735 {
3736 	int i, end;
3737 	if (IS_DATASEG(type))
3738 		end = type + NR_CURSEG_DATA_TYPE;
3739 	else
3740 		end = type + NR_CURSEG_NODE_TYPE;
3741 
3742 	for (i = type; i < end; i++)
3743 		write_current_sum_page(sbi, i, blkaddr + (i - type));
3744 }
3745 
3746 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3747 {
3748 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3749 		write_compacted_summaries(sbi, start_blk);
3750 	else
3751 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3752 }
3753 
3754 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3755 {
3756 	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3757 }
3758 
3759 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3760 					unsigned int val, int alloc)
3761 {
3762 	int i;
3763 
3764 	if (type == NAT_JOURNAL) {
3765 		for (i = 0; i < nats_in_cursum(journal); i++) {
3766 			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3767 				return i;
3768 		}
3769 		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3770 			return update_nats_in_cursum(journal, 1);
3771 	} else if (type == SIT_JOURNAL) {
3772 		for (i = 0; i < sits_in_cursum(journal); i++)
3773 			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3774 				return i;
3775 		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3776 			return update_sits_in_cursum(journal, 1);
3777 	}
3778 	return -1;
3779 }
3780 
3781 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3782 					unsigned int segno)
3783 {
3784 	return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3785 }
3786 
3787 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3788 					unsigned int start)
3789 {
3790 	struct sit_info *sit_i = SIT_I(sbi);
3791 	struct page *page;
3792 	pgoff_t src_off, dst_off;
3793 
3794 	src_off = current_sit_addr(sbi, start);
3795 	dst_off = next_sit_addr(sbi, src_off);
3796 
3797 	page = f2fs_grab_meta_page(sbi, dst_off);
3798 	seg_info_to_sit_page(sbi, page, start);
3799 
3800 	set_page_dirty(page);
3801 	set_to_next_sit(sit_i, start);
3802 
3803 	return page;
3804 }
3805 
3806 static struct sit_entry_set *grab_sit_entry_set(void)
3807 {
3808 	struct sit_entry_set *ses =
3809 			f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3810 
3811 	ses->entry_cnt = 0;
3812 	INIT_LIST_HEAD(&ses->set_list);
3813 	return ses;
3814 }
3815 
3816 static void release_sit_entry_set(struct sit_entry_set *ses)
3817 {
3818 	list_del(&ses->set_list);
3819 	kmem_cache_free(sit_entry_set_slab, ses);
3820 }
3821 
3822 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3823 						struct list_head *head)
3824 {
3825 	struct sit_entry_set *next = ses;
3826 
3827 	if (list_is_last(&ses->set_list, head))
3828 		return;
3829 
3830 	list_for_each_entry_continue(next, head, set_list)
3831 		if (ses->entry_cnt <= next->entry_cnt)
3832 			break;
3833 
3834 	list_move_tail(&ses->set_list, &next->set_list);
3835 }
3836 
3837 static void add_sit_entry(unsigned int segno, struct list_head *head)
3838 {
3839 	struct sit_entry_set *ses;
3840 	unsigned int start_segno = START_SEGNO(segno);
3841 
3842 	list_for_each_entry(ses, head, set_list) {
3843 		if (ses->start_segno == start_segno) {
3844 			ses->entry_cnt++;
3845 			adjust_sit_entry_set(ses, head);
3846 			return;
3847 		}
3848 	}
3849 
3850 	ses = grab_sit_entry_set();
3851 
3852 	ses->start_segno = start_segno;
3853 	ses->entry_cnt++;
3854 	list_add(&ses->set_list, head);
3855 }
3856 
3857 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3858 {
3859 	struct f2fs_sm_info *sm_info = SM_I(sbi);
3860 	struct list_head *set_list = &sm_info->sit_entry_set;
3861 	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3862 	unsigned int segno;
3863 
3864 	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3865 		add_sit_entry(segno, set_list);
3866 }
3867 
3868 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3869 {
3870 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3871 	struct f2fs_journal *journal = curseg->journal;
3872 	int i;
3873 
3874 	down_write(&curseg->journal_rwsem);
3875 	for (i = 0; i < sits_in_cursum(journal); i++) {
3876 		unsigned int segno;
3877 		bool dirtied;
3878 
3879 		segno = le32_to_cpu(segno_in_journal(journal, i));
3880 		dirtied = __mark_sit_entry_dirty(sbi, segno);
3881 
3882 		if (!dirtied)
3883 			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3884 	}
3885 	update_sits_in_cursum(journal, -i);
3886 	up_write(&curseg->journal_rwsem);
3887 }
3888 
3889 /*
3890  * CP calls this function, which flushes SIT entries including sit_journal,
3891  * and moves prefree segs to free segs.
3892  */
3893 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3894 {
3895 	struct sit_info *sit_i = SIT_I(sbi);
3896 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3897 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3898 	struct f2fs_journal *journal = curseg->journal;
3899 	struct sit_entry_set *ses, *tmp;
3900 	struct list_head *head = &SM_I(sbi)->sit_entry_set;
3901 	bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
3902 	struct seg_entry *se;
3903 
3904 	down_write(&sit_i->sentry_lock);
3905 
3906 	if (!sit_i->dirty_sentries)
3907 		goto out;
3908 
3909 	/*
3910 	 * add and account sit entries of dirty bitmap in sit entry
3911 	 * set temporarily
3912 	 */
3913 	add_sits_in_set(sbi);
3914 
3915 	/*
3916 	 * if there are no enough space in journal to store dirty sit
3917 	 * entries, remove all entries from journal and add and account
3918 	 * them in sit entry set.
3919 	 */
3920 	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
3921 								!to_journal)
3922 		remove_sits_in_journal(sbi);
3923 
3924 	/*
3925 	 * there are two steps to flush sit entries:
3926 	 * #1, flush sit entries to journal in current cold data summary block.
3927 	 * #2, flush sit entries to sit page.
3928 	 */
3929 	list_for_each_entry_safe(ses, tmp, head, set_list) {
3930 		struct page *page = NULL;
3931 		struct f2fs_sit_block *raw_sit = NULL;
3932 		unsigned int start_segno = ses->start_segno;
3933 		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3934 						(unsigned long)MAIN_SEGS(sbi));
3935 		unsigned int segno = start_segno;
3936 
3937 		if (to_journal &&
3938 			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3939 			to_journal = false;
3940 
3941 		if (to_journal) {
3942 			down_write(&curseg->journal_rwsem);
3943 		} else {
3944 			page = get_next_sit_page(sbi, start_segno);
3945 			raw_sit = page_address(page);
3946 		}
3947 
3948 		/* flush dirty sit entries in region of current sit set */
3949 		for_each_set_bit_from(segno, bitmap, end) {
3950 			int offset, sit_offset;
3951 
3952 			se = get_seg_entry(sbi, segno);
3953 #ifdef CONFIG_F2FS_CHECK_FS
3954 			if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3955 						SIT_VBLOCK_MAP_SIZE))
3956 				f2fs_bug_on(sbi, 1);
3957 #endif
3958 
3959 			/* add discard candidates */
3960 			if (!(cpc->reason & CP_DISCARD)) {
3961 				cpc->trim_start = segno;
3962 				add_discard_addrs(sbi, cpc, false);
3963 			}
3964 
3965 			if (to_journal) {
3966 				offset = f2fs_lookup_journal_in_cursum(journal,
3967 							SIT_JOURNAL, segno, 1);
3968 				f2fs_bug_on(sbi, offset < 0);
3969 				segno_in_journal(journal, offset) =
3970 							cpu_to_le32(segno);
3971 				seg_info_to_raw_sit(se,
3972 					&sit_in_journal(journal, offset));
3973 				check_block_count(sbi, segno,
3974 					&sit_in_journal(journal, offset));
3975 			} else {
3976 				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3977 				seg_info_to_raw_sit(se,
3978 						&raw_sit->entries[sit_offset]);
3979 				check_block_count(sbi, segno,
3980 						&raw_sit->entries[sit_offset]);
3981 			}
3982 
3983 			__clear_bit(segno, bitmap);
3984 			sit_i->dirty_sentries--;
3985 			ses->entry_cnt--;
3986 		}
3987 
3988 		if (to_journal)
3989 			up_write(&curseg->journal_rwsem);
3990 		else
3991 			f2fs_put_page(page, 1);
3992 
3993 		f2fs_bug_on(sbi, ses->entry_cnt);
3994 		release_sit_entry_set(ses);
3995 	}
3996 
3997 	f2fs_bug_on(sbi, !list_empty(head));
3998 	f2fs_bug_on(sbi, sit_i->dirty_sentries);
3999 out:
4000 	if (cpc->reason & CP_DISCARD) {
4001 		__u64 trim_start = cpc->trim_start;
4002 
4003 		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4004 			add_discard_addrs(sbi, cpc, false);
4005 
4006 		cpc->trim_start = trim_start;
4007 	}
4008 	up_write(&sit_i->sentry_lock);
4009 
4010 	set_prefree_as_free_segments(sbi);
4011 }
4012 
4013 static int build_sit_info(struct f2fs_sb_info *sbi)
4014 {
4015 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4016 	struct sit_info *sit_i;
4017 	unsigned int sit_segs, start;
4018 	char *src_bitmap, *bitmap;
4019 	unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4020 
4021 	/* allocate memory for SIT information */
4022 	sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4023 	if (!sit_i)
4024 		return -ENOMEM;
4025 
4026 	SM_I(sbi)->sit_info = sit_i;
4027 
4028 	sit_i->sentries =
4029 		f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4030 					      MAIN_SEGS(sbi)),
4031 			      GFP_KERNEL);
4032 	if (!sit_i->sentries)
4033 		return -ENOMEM;
4034 
4035 	main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4036 	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4037 								GFP_KERNEL);
4038 	if (!sit_i->dirty_sentries_bitmap)
4039 		return -ENOMEM;
4040 
4041 #ifdef CONFIG_F2FS_CHECK_FS
4042 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 4;
4043 #else
4044 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 3;
4045 #endif
4046 	sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4047 	if (!sit_i->bitmap)
4048 		return -ENOMEM;
4049 
4050 	bitmap = sit_i->bitmap;
4051 
4052 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4053 		sit_i->sentries[start].cur_valid_map = bitmap;
4054 		bitmap += SIT_VBLOCK_MAP_SIZE;
4055 
4056 		sit_i->sentries[start].ckpt_valid_map = bitmap;
4057 		bitmap += SIT_VBLOCK_MAP_SIZE;
4058 
4059 #ifdef CONFIG_F2FS_CHECK_FS
4060 		sit_i->sentries[start].cur_valid_map_mir = bitmap;
4061 		bitmap += SIT_VBLOCK_MAP_SIZE;
4062 #endif
4063 
4064 		sit_i->sentries[start].discard_map = bitmap;
4065 		bitmap += SIT_VBLOCK_MAP_SIZE;
4066 	}
4067 
4068 	sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4069 	if (!sit_i->tmp_map)
4070 		return -ENOMEM;
4071 
4072 	if (__is_large_section(sbi)) {
4073 		sit_i->sec_entries =
4074 			f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4075 						      MAIN_SECS(sbi)),
4076 				      GFP_KERNEL);
4077 		if (!sit_i->sec_entries)
4078 			return -ENOMEM;
4079 	}
4080 
4081 	/* get information related with SIT */
4082 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4083 
4084 	/* setup SIT bitmap from ckeckpoint pack */
4085 	sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4086 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4087 
4088 	sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4089 	if (!sit_i->sit_bitmap)
4090 		return -ENOMEM;
4091 
4092 #ifdef CONFIG_F2FS_CHECK_FS
4093 	sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4094 					sit_bitmap_size, GFP_KERNEL);
4095 	if (!sit_i->sit_bitmap_mir)
4096 		return -ENOMEM;
4097 
4098 	sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4099 					main_bitmap_size, GFP_KERNEL);
4100 	if (!sit_i->invalid_segmap)
4101 		return -ENOMEM;
4102 #endif
4103 
4104 	/* init SIT information */
4105 	sit_i->s_ops = &default_salloc_ops;
4106 
4107 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4108 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4109 	sit_i->written_valid_blocks = 0;
4110 	sit_i->bitmap_size = sit_bitmap_size;
4111 	sit_i->dirty_sentries = 0;
4112 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4113 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4114 	sit_i->mounted_time = ktime_get_boottime_seconds();
4115 	init_rwsem(&sit_i->sentry_lock);
4116 	return 0;
4117 }
4118 
4119 static int build_free_segmap(struct f2fs_sb_info *sbi)
4120 {
4121 	struct free_segmap_info *free_i;
4122 	unsigned int bitmap_size, sec_bitmap_size;
4123 
4124 	/* allocate memory for free segmap information */
4125 	free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4126 	if (!free_i)
4127 		return -ENOMEM;
4128 
4129 	SM_I(sbi)->free_info = free_i;
4130 
4131 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4132 	free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4133 	if (!free_i->free_segmap)
4134 		return -ENOMEM;
4135 
4136 	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4137 	free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4138 	if (!free_i->free_secmap)
4139 		return -ENOMEM;
4140 
4141 	/* set all segments as dirty temporarily */
4142 	memset(free_i->free_segmap, 0xff, bitmap_size);
4143 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4144 
4145 	/* init free segmap information */
4146 	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4147 	free_i->free_segments = 0;
4148 	free_i->free_sections = 0;
4149 	spin_lock_init(&free_i->segmap_lock);
4150 	return 0;
4151 }
4152 
4153 static int build_curseg(struct f2fs_sb_info *sbi)
4154 {
4155 	struct curseg_info *array;
4156 	int i;
4157 
4158 	array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
4159 			     GFP_KERNEL);
4160 	if (!array)
4161 		return -ENOMEM;
4162 
4163 	SM_I(sbi)->curseg_array = array;
4164 
4165 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
4166 		mutex_init(&array[i].curseg_mutex);
4167 		array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4168 		if (!array[i].sum_blk)
4169 			return -ENOMEM;
4170 		init_rwsem(&array[i].journal_rwsem);
4171 		array[i].journal = f2fs_kzalloc(sbi,
4172 				sizeof(struct f2fs_journal), GFP_KERNEL);
4173 		if (!array[i].journal)
4174 			return -ENOMEM;
4175 		array[i].segno = NULL_SEGNO;
4176 		array[i].next_blkoff = 0;
4177 	}
4178 	return restore_curseg_summaries(sbi);
4179 }
4180 
4181 static int build_sit_entries(struct f2fs_sb_info *sbi)
4182 {
4183 	struct sit_info *sit_i = SIT_I(sbi);
4184 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4185 	struct f2fs_journal *journal = curseg->journal;
4186 	struct seg_entry *se;
4187 	struct f2fs_sit_entry sit;
4188 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
4189 	unsigned int i, start, end;
4190 	unsigned int readed, start_blk = 0;
4191 	int err = 0;
4192 	block_t total_node_blocks = 0;
4193 
4194 	do {
4195 		readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4196 							META_SIT, true);
4197 
4198 		start = start_blk * sit_i->sents_per_block;
4199 		end = (start_blk + readed) * sit_i->sents_per_block;
4200 
4201 		for (; start < end && start < MAIN_SEGS(sbi); start++) {
4202 			struct f2fs_sit_block *sit_blk;
4203 			struct page *page;
4204 
4205 			se = &sit_i->sentries[start];
4206 			page = get_current_sit_page(sbi, start);
4207 			if (IS_ERR(page))
4208 				return PTR_ERR(page);
4209 			sit_blk = (struct f2fs_sit_block *)page_address(page);
4210 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4211 			f2fs_put_page(page, 1);
4212 
4213 			err = check_block_count(sbi, start, &sit);
4214 			if (err)
4215 				return err;
4216 			seg_info_from_raw_sit(se, &sit);
4217 			if (IS_NODESEG(se->type))
4218 				total_node_blocks += se->valid_blocks;
4219 
4220 			/* build discard map only one time */
4221 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4222 				memset(se->discard_map, 0xff,
4223 					SIT_VBLOCK_MAP_SIZE);
4224 			} else {
4225 				memcpy(se->discard_map,
4226 					se->cur_valid_map,
4227 					SIT_VBLOCK_MAP_SIZE);
4228 				sbi->discard_blks +=
4229 					sbi->blocks_per_seg -
4230 					se->valid_blocks;
4231 			}
4232 
4233 			if (__is_large_section(sbi))
4234 				get_sec_entry(sbi, start)->valid_blocks +=
4235 							se->valid_blocks;
4236 		}
4237 		start_blk += readed;
4238 	} while (start_blk < sit_blk_cnt);
4239 
4240 	down_read(&curseg->journal_rwsem);
4241 	for (i = 0; i < sits_in_cursum(journal); i++) {
4242 		unsigned int old_valid_blocks;
4243 
4244 		start = le32_to_cpu(segno_in_journal(journal, i));
4245 		if (start >= MAIN_SEGS(sbi)) {
4246 			f2fs_err(sbi, "Wrong journal entry on segno %u",
4247 				 start);
4248 			err = -EFSCORRUPTED;
4249 			break;
4250 		}
4251 
4252 		se = &sit_i->sentries[start];
4253 		sit = sit_in_journal(journal, i);
4254 
4255 		old_valid_blocks = se->valid_blocks;
4256 		if (IS_NODESEG(se->type))
4257 			total_node_blocks -= old_valid_blocks;
4258 
4259 		err = check_block_count(sbi, start, &sit);
4260 		if (err)
4261 			break;
4262 		seg_info_from_raw_sit(se, &sit);
4263 		if (IS_NODESEG(se->type))
4264 			total_node_blocks += se->valid_blocks;
4265 
4266 		if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4267 			memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4268 		} else {
4269 			memcpy(se->discard_map, se->cur_valid_map,
4270 						SIT_VBLOCK_MAP_SIZE);
4271 			sbi->discard_blks += old_valid_blocks;
4272 			sbi->discard_blks -= se->valid_blocks;
4273 		}
4274 
4275 		if (__is_large_section(sbi)) {
4276 			get_sec_entry(sbi, start)->valid_blocks +=
4277 							se->valid_blocks;
4278 			get_sec_entry(sbi, start)->valid_blocks -=
4279 							old_valid_blocks;
4280 		}
4281 	}
4282 	up_read(&curseg->journal_rwsem);
4283 
4284 	if (!err && total_node_blocks != valid_node_count(sbi)) {
4285 		f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4286 			 total_node_blocks, valid_node_count(sbi));
4287 		err = -EFSCORRUPTED;
4288 	}
4289 
4290 	return err;
4291 }
4292 
4293 static void init_free_segmap(struct f2fs_sb_info *sbi)
4294 {
4295 	unsigned int start;
4296 	int type;
4297 
4298 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4299 		struct seg_entry *sentry = get_seg_entry(sbi, start);
4300 		if (!sentry->valid_blocks)
4301 			__set_free(sbi, start);
4302 		else
4303 			SIT_I(sbi)->written_valid_blocks +=
4304 						sentry->valid_blocks;
4305 	}
4306 
4307 	/* set use the current segments */
4308 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4309 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4310 		__set_test_and_inuse(sbi, curseg_t->segno);
4311 	}
4312 }
4313 
4314 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4315 {
4316 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4317 	struct free_segmap_info *free_i = FREE_I(sbi);
4318 	unsigned int segno = 0, offset = 0, secno;
4319 	block_t valid_blocks;
4320 	block_t blks_per_sec = BLKS_PER_SEC(sbi);
4321 
4322 	while (1) {
4323 		/* find dirty segment based on free segmap */
4324 		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4325 		if (segno >= MAIN_SEGS(sbi))
4326 			break;
4327 		offset = segno + 1;
4328 		valid_blocks = get_valid_blocks(sbi, segno, false);
4329 		if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4330 			continue;
4331 		if (valid_blocks > sbi->blocks_per_seg) {
4332 			f2fs_bug_on(sbi, 1);
4333 			continue;
4334 		}
4335 		mutex_lock(&dirty_i->seglist_lock);
4336 		__locate_dirty_segment(sbi, segno, DIRTY);
4337 		mutex_unlock(&dirty_i->seglist_lock);
4338 	}
4339 
4340 	if (!__is_large_section(sbi))
4341 		return;
4342 
4343 	mutex_lock(&dirty_i->seglist_lock);
4344 	for (segno = 0; segno < MAIN_SECS(sbi); segno += blks_per_sec) {
4345 		valid_blocks = get_valid_blocks(sbi, segno, true);
4346 		secno = GET_SEC_FROM_SEG(sbi, segno);
4347 
4348 		if (!valid_blocks || valid_blocks == blks_per_sec)
4349 			continue;
4350 		if (IS_CURSEC(sbi, secno))
4351 			continue;
4352 		set_bit(secno, dirty_i->dirty_secmap);
4353 	}
4354 	mutex_unlock(&dirty_i->seglist_lock);
4355 }
4356 
4357 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4358 {
4359 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4360 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4361 
4362 	dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4363 	if (!dirty_i->victim_secmap)
4364 		return -ENOMEM;
4365 	return 0;
4366 }
4367 
4368 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4369 {
4370 	struct dirty_seglist_info *dirty_i;
4371 	unsigned int bitmap_size, i;
4372 
4373 	/* allocate memory for dirty segments list information */
4374 	dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4375 								GFP_KERNEL);
4376 	if (!dirty_i)
4377 		return -ENOMEM;
4378 
4379 	SM_I(sbi)->dirty_info = dirty_i;
4380 	mutex_init(&dirty_i->seglist_lock);
4381 
4382 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4383 
4384 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
4385 		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4386 								GFP_KERNEL);
4387 		if (!dirty_i->dirty_segmap[i])
4388 			return -ENOMEM;
4389 	}
4390 
4391 	if (__is_large_section(sbi)) {
4392 		bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4393 		dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4394 						bitmap_size, GFP_KERNEL);
4395 		if (!dirty_i->dirty_secmap)
4396 			return -ENOMEM;
4397 	}
4398 
4399 	init_dirty_segmap(sbi);
4400 	return init_victim_secmap(sbi);
4401 }
4402 
4403 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4404 {
4405 	int i;
4406 
4407 	/*
4408 	 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4409 	 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4410 	 */
4411 	for (i = 0; i < NO_CHECK_TYPE; i++) {
4412 		struct curseg_info *curseg = CURSEG_I(sbi, i);
4413 		struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4414 		unsigned int blkofs = curseg->next_blkoff;
4415 
4416 		if (f2fs_test_bit(blkofs, se->cur_valid_map))
4417 			goto out;
4418 
4419 		if (curseg->alloc_type == SSR)
4420 			continue;
4421 
4422 		for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4423 			if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4424 				continue;
4425 out:
4426 			f2fs_err(sbi,
4427 				 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4428 				 i, curseg->segno, curseg->alloc_type,
4429 				 curseg->next_blkoff, blkofs);
4430 			return -EFSCORRUPTED;
4431 		}
4432 	}
4433 	return 0;
4434 }
4435 
4436 #ifdef CONFIG_BLK_DEV_ZONED
4437 
4438 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4439 				    struct f2fs_dev_info *fdev,
4440 				    struct blk_zone *zone)
4441 {
4442 	unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4443 	block_t zone_block, wp_block, last_valid_block;
4444 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4445 	int i, s, b, ret;
4446 	struct seg_entry *se;
4447 
4448 	if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4449 		return 0;
4450 
4451 	wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4452 	wp_segno = GET_SEGNO(sbi, wp_block);
4453 	wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4454 	zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4455 	zone_segno = GET_SEGNO(sbi, zone_block);
4456 	zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4457 
4458 	if (zone_segno >= MAIN_SEGS(sbi))
4459 		return 0;
4460 
4461 	/*
4462 	 * Skip check of zones cursegs point to, since
4463 	 * fix_curseg_write_pointer() checks them.
4464 	 */
4465 	for (i = 0; i < NO_CHECK_TYPE; i++)
4466 		if (zone_secno == GET_SEC_FROM_SEG(sbi,
4467 						   CURSEG_I(sbi, i)->segno))
4468 			return 0;
4469 
4470 	/*
4471 	 * Get last valid block of the zone.
4472 	 */
4473 	last_valid_block = zone_block - 1;
4474 	for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4475 		segno = zone_segno + s;
4476 		se = get_seg_entry(sbi, segno);
4477 		for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4478 			if (f2fs_test_bit(b, se->cur_valid_map)) {
4479 				last_valid_block = START_BLOCK(sbi, segno) + b;
4480 				break;
4481 			}
4482 		if (last_valid_block >= zone_block)
4483 			break;
4484 	}
4485 
4486 	/*
4487 	 * If last valid block is beyond the write pointer, report the
4488 	 * inconsistency. This inconsistency does not cause write error
4489 	 * because the zone will not be selected for write operation until
4490 	 * it get discarded. Just report it.
4491 	 */
4492 	if (last_valid_block >= wp_block) {
4493 		f2fs_notice(sbi, "Valid block beyond write pointer: "
4494 			    "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4495 			    GET_SEGNO(sbi, last_valid_block),
4496 			    GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4497 			    wp_segno, wp_blkoff);
4498 		return 0;
4499 	}
4500 
4501 	/*
4502 	 * If there is no valid block in the zone and if write pointer is
4503 	 * not at zone start, reset the write pointer.
4504 	 */
4505 	if (last_valid_block + 1 == zone_block && zone->wp != zone->start) {
4506 		f2fs_notice(sbi,
4507 			    "Zone without valid block has non-zero write "
4508 			    "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4509 			    wp_segno, wp_blkoff);
4510 		ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4511 					zone->len >> log_sectors_per_block);
4512 		if (ret) {
4513 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4514 				 fdev->path, ret);
4515 			return ret;
4516 		}
4517 	}
4518 
4519 	return 0;
4520 }
4521 
4522 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4523 						  block_t zone_blkaddr)
4524 {
4525 	int i;
4526 
4527 	for (i = 0; i < sbi->s_ndevs; i++) {
4528 		if (!bdev_is_zoned(FDEV(i).bdev))
4529 			continue;
4530 		if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4531 				zone_blkaddr <= FDEV(i).end_blk))
4532 			return &FDEV(i);
4533 	}
4534 
4535 	return NULL;
4536 }
4537 
4538 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4539 			      void *data) {
4540 	memcpy(data, zone, sizeof(struct blk_zone));
4541 	return 0;
4542 }
4543 
4544 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4545 {
4546 	struct curseg_info *cs = CURSEG_I(sbi, type);
4547 	struct f2fs_dev_info *zbd;
4548 	struct blk_zone zone;
4549 	unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4550 	block_t cs_zone_block, wp_block;
4551 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4552 	sector_t zone_sector;
4553 	int err;
4554 
4555 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4556 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4557 
4558 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
4559 	if (!zbd)
4560 		return 0;
4561 
4562 	/* report zone for the sector the curseg points to */
4563 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4564 		<< log_sectors_per_block;
4565 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4566 				  report_one_zone_cb, &zone);
4567 	if (err != 1) {
4568 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4569 			 zbd->path, err);
4570 		return err;
4571 	}
4572 
4573 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4574 		return 0;
4575 
4576 	wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
4577 	wp_segno = GET_SEGNO(sbi, wp_block);
4578 	wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4579 	wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
4580 
4581 	if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
4582 		wp_sector_off == 0)
4583 		return 0;
4584 
4585 	f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
4586 		    "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
4587 		    type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
4588 
4589 	f2fs_notice(sbi, "Assign new section to curseg[%d]: "
4590 		    "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
4591 	allocate_segment_by_default(sbi, type, true);
4592 
4593 	/* check consistency of the zone curseg pointed to */
4594 	if (check_zone_write_pointer(sbi, zbd, &zone))
4595 		return -EIO;
4596 
4597 	/* check newly assigned zone */
4598 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4599 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4600 
4601 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
4602 	if (!zbd)
4603 		return 0;
4604 
4605 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4606 		<< log_sectors_per_block;
4607 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4608 				  report_one_zone_cb, &zone);
4609 	if (err != 1) {
4610 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4611 			 zbd->path, err);
4612 		return err;
4613 	}
4614 
4615 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4616 		return 0;
4617 
4618 	if (zone.wp != zone.start) {
4619 		f2fs_notice(sbi,
4620 			    "New zone for curseg[%d] is not yet discarded. "
4621 			    "Reset the zone: curseg[0x%x,0x%x]",
4622 			    type, cs->segno, cs->next_blkoff);
4623 		err = __f2fs_issue_discard_zone(sbi, zbd->bdev,
4624 				zone_sector >> log_sectors_per_block,
4625 				zone.len >> log_sectors_per_block);
4626 		if (err) {
4627 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4628 				 zbd->path, err);
4629 			return err;
4630 		}
4631 	}
4632 
4633 	return 0;
4634 }
4635 
4636 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4637 {
4638 	int i, ret;
4639 
4640 	for (i = 0; i < NO_CHECK_TYPE; i++) {
4641 		ret = fix_curseg_write_pointer(sbi, i);
4642 		if (ret)
4643 			return ret;
4644 	}
4645 
4646 	return 0;
4647 }
4648 
4649 struct check_zone_write_pointer_args {
4650 	struct f2fs_sb_info *sbi;
4651 	struct f2fs_dev_info *fdev;
4652 };
4653 
4654 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
4655 				      void *data) {
4656 	struct check_zone_write_pointer_args *args;
4657 	args = (struct check_zone_write_pointer_args *)data;
4658 
4659 	return check_zone_write_pointer(args->sbi, args->fdev, zone);
4660 }
4661 
4662 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4663 {
4664 	int i, ret;
4665 	struct check_zone_write_pointer_args args;
4666 
4667 	for (i = 0; i < sbi->s_ndevs; i++) {
4668 		if (!bdev_is_zoned(FDEV(i).bdev))
4669 			continue;
4670 
4671 		args.sbi = sbi;
4672 		args.fdev = &FDEV(i);
4673 		ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
4674 					  check_zone_write_pointer_cb, &args);
4675 		if (ret < 0)
4676 			return ret;
4677 	}
4678 
4679 	return 0;
4680 }
4681 #else
4682 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4683 {
4684 	return 0;
4685 }
4686 
4687 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4688 {
4689 	return 0;
4690 }
4691 #endif
4692 
4693 /*
4694  * Update min, max modified time for cost-benefit GC algorithm
4695  */
4696 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4697 {
4698 	struct sit_info *sit_i = SIT_I(sbi);
4699 	unsigned int segno;
4700 
4701 	down_write(&sit_i->sentry_lock);
4702 
4703 	sit_i->min_mtime = ULLONG_MAX;
4704 
4705 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4706 		unsigned int i;
4707 		unsigned long long mtime = 0;
4708 
4709 		for (i = 0; i < sbi->segs_per_sec; i++)
4710 			mtime += get_seg_entry(sbi, segno + i)->mtime;
4711 
4712 		mtime = div_u64(mtime, sbi->segs_per_sec);
4713 
4714 		if (sit_i->min_mtime > mtime)
4715 			sit_i->min_mtime = mtime;
4716 	}
4717 	sit_i->max_mtime = get_mtime(sbi, false);
4718 	up_write(&sit_i->sentry_lock);
4719 }
4720 
4721 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4722 {
4723 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4724 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4725 	struct f2fs_sm_info *sm_info;
4726 	int err;
4727 
4728 	sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4729 	if (!sm_info)
4730 		return -ENOMEM;
4731 
4732 	/* init sm info */
4733 	sbi->sm_info = sm_info;
4734 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4735 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4736 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4737 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4738 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4739 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4740 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4741 	sm_info->rec_prefree_segments = sm_info->main_segments *
4742 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4743 	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4744 		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4745 
4746 	if (!f2fs_lfs_mode(sbi))
4747 		sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4748 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4749 	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4750 	sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4751 	sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4752 	sm_info->min_ssr_sections = reserved_sections(sbi);
4753 
4754 	INIT_LIST_HEAD(&sm_info->sit_entry_set);
4755 
4756 	init_rwsem(&sm_info->curseg_lock);
4757 
4758 	if (!f2fs_readonly(sbi->sb)) {
4759 		err = f2fs_create_flush_cmd_control(sbi);
4760 		if (err)
4761 			return err;
4762 	}
4763 
4764 	err = create_discard_cmd_control(sbi);
4765 	if (err)
4766 		return err;
4767 
4768 	err = build_sit_info(sbi);
4769 	if (err)
4770 		return err;
4771 	err = build_free_segmap(sbi);
4772 	if (err)
4773 		return err;
4774 	err = build_curseg(sbi);
4775 	if (err)
4776 		return err;
4777 
4778 	/* reinit free segmap based on SIT */
4779 	err = build_sit_entries(sbi);
4780 	if (err)
4781 		return err;
4782 
4783 	init_free_segmap(sbi);
4784 	err = build_dirty_segmap(sbi);
4785 	if (err)
4786 		return err;
4787 
4788 	err = sanity_check_curseg(sbi);
4789 	if (err)
4790 		return err;
4791 
4792 	init_min_max_mtime(sbi);
4793 	return 0;
4794 }
4795 
4796 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4797 		enum dirty_type dirty_type)
4798 {
4799 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4800 
4801 	mutex_lock(&dirty_i->seglist_lock);
4802 	kvfree(dirty_i->dirty_segmap[dirty_type]);
4803 	dirty_i->nr_dirty[dirty_type] = 0;
4804 	mutex_unlock(&dirty_i->seglist_lock);
4805 }
4806 
4807 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4808 {
4809 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4810 	kvfree(dirty_i->victim_secmap);
4811 }
4812 
4813 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4814 {
4815 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4816 	int i;
4817 
4818 	if (!dirty_i)
4819 		return;
4820 
4821 	/* discard pre-free/dirty segments list */
4822 	for (i = 0; i < NR_DIRTY_TYPE; i++)
4823 		discard_dirty_segmap(sbi, i);
4824 
4825 	if (__is_large_section(sbi)) {
4826 		mutex_lock(&dirty_i->seglist_lock);
4827 		kvfree(dirty_i->dirty_secmap);
4828 		mutex_unlock(&dirty_i->seglist_lock);
4829 	}
4830 
4831 	destroy_victim_secmap(sbi);
4832 	SM_I(sbi)->dirty_info = NULL;
4833 	kvfree(dirty_i);
4834 }
4835 
4836 static void destroy_curseg(struct f2fs_sb_info *sbi)
4837 {
4838 	struct curseg_info *array = SM_I(sbi)->curseg_array;
4839 	int i;
4840 
4841 	if (!array)
4842 		return;
4843 	SM_I(sbi)->curseg_array = NULL;
4844 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
4845 		kvfree(array[i].sum_blk);
4846 		kvfree(array[i].journal);
4847 	}
4848 	kvfree(array);
4849 }
4850 
4851 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4852 {
4853 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4854 	if (!free_i)
4855 		return;
4856 	SM_I(sbi)->free_info = NULL;
4857 	kvfree(free_i->free_segmap);
4858 	kvfree(free_i->free_secmap);
4859 	kvfree(free_i);
4860 }
4861 
4862 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4863 {
4864 	struct sit_info *sit_i = SIT_I(sbi);
4865 
4866 	if (!sit_i)
4867 		return;
4868 
4869 	if (sit_i->sentries)
4870 		kvfree(sit_i->bitmap);
4871 	kvfree(sit_i->tmp_map);
4872 
4873 	kvfree(sit_i->sentries);
4874 	kvfree(sit_i->sec_entries);
4875 	kvfree(sit_i->dirty_sentries_bitmap);
4876 
4877 	SM_I(sbi)->sit_info = NULL;
4878 	kvfree(sit_i->sit_bitmap);
4879 #ifdef CONFIG_F2FS_CHECK_FS
4880 	kvfree(sit_i->sit_bitmap_mir);
4881 	kvfree(sit_i->invalid_segmap);
4882 #endif
4883 	kvfree(sit_i);
4884 }
4885 
4886 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4887 {
4888 	struct f2fs_sm_info *sm_info = SM_I(sbi);
4889 
4890 	if (!sm_info)
4891 		return;
4892 	f2fs_destroy_flush_cmd_control(sbi, true);
4893 	destroy_discard_cmd_control(sbi);
4894 	destroy_dirty_segmap(sbi);
4895 	destroy_curseg(sbi);
4896 	destroy_free_segmap(sbi);
4897 	destroy_sit_info(sbi);
4898 	sbi->sm_info = NULL;
4899 	kvfree(sm_info);
4900 }
4901 
4902 int __init f2fs_create_segment_manager_caches(void)
4903 {
4904 	discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
4905 			sizeof(struct discard_entry));
4906 	if (!discard_entry_slab)
4907 		goto fail;
4908 
4909 	discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
4910 			sizeof(struct discard_cmd));
4911 	if (!discard_cmd_slab)
4912 		goto destroy_discard_entry;
4913 
4914 	sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
4915 			sizeof(struct sit_entry_set));
4916 	if (!sit_entry_set_slab)
4917 		goto destroy_discard_cmd;
4918 
4919 	inmem_entry_slab = f2fs_kmem_cache_create("f2fs_inmem_page_entry",
4920 			sizeof(struct inmem_pages));
4921 	if (!inmem_entry_slab)
4922 		goto destroy_sit_entry_set;
4923 	return 0;
4924 
4925 destroy_sit_entry_set:
4926 	kmem_cache_destroy(sit_entry_set_slab);
4927 destroy_discard_cmd:
4928 	kmem_cache_destroy(discard_cmd_slab);
4929 destroy_discard_entry:
4930 	kmem_cache_destroy(discard_entry_slab);
4931 fail:
4932 	return -ENOMEM;
4933 }
4934 
4935 void f2fs_destroy_segment_manager_caches(void)
4936 {
4937 	kmem_cache_destroy(sit_entry_set_slab);
4938 	kmem_cache_destroy(discard_cmd_slab);
4939 	kmem_cache_destroy(discard_entry_slab);
4940 	kmem_cache_destroy(inmem_entry_slab);
4941 }
4942