xref: /openbmc/linux/fs/f2fs/segment.c (revision 5e012745)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * fs/f2fs/segment.c
4  *
5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6  *             http://www.samsung.com/
7  */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/bio.h>
11 #include <linux/blkdev.h>
12 #include <linux/prefetch.h>
13 #include <linux/kthread.h>
14 #include <linux/swap.h>
15 #include <linux/timer.h>
16 #include <linux/freezer.h>
17 #include <linux/sched/signal.h>
18 
19 #include "f2fs.h"
20 #include "segment.h"
21 #include "node.h"
22 #include "gc.h"
23 #include "trace.h"
24 #include <trace/events/f2fs.h>
25 
26 #define __reverse_ffz(x) __reverse_ffs(~(x))
27 
28 static struct kmem_cache *discard_entry_slab;
29 static struct kmem_cache *discard_cmd_slab;
30 static struct kmem_cache *sit_entry_set_slab;
31 static struct kmem_cache *inmem_entry_slab;
32 
33 static unsigned long __reverse_ulong(unsigned char *str)
34 {
35 	unsigned long tmp = 0;
36 	int shift = 24, idx = 0;
37 
38 #if BITS_PER_LONG == 64
39 	shift = 56;
40 #endif
41 	while (shift >= 0) {
42 		tmp |= (unsigned long)str[idx++] << shift;
43 		shift -= BITS_PER_BYTE;
44 	}
45 	return tmp;
46 }
47 
48 /*
49  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
50  * MSB and LSB are reversed in a byte by f2fs_set_bit.
51  */
52 static inline unsigned long __reverse_ffs(unsigned long word)
53 {
54 	int num = 0;
55 
56 #if BITS_PER_LONG == 64
57 	if ((word & 0xffffffff00000000UL) == 0)
58 		num += 32;
59 	else
60 		word >>= 32;
61 #endif
62 	if ((word & 0xffff0000) == 0)
63 		num += 16;
64 	else
65 		word >>= 16;
66 
67 	if ((word & 0xff00) == 0)
68 		num += 8;
69 	else
70 		word >>= 8;
71 
72 	if ((word & 0xf0) == 0)
73 		num += 4;
74 	else
75 		word >>= 4;
76 
77 	if ((word & 0xc) == 0)
78 		num += 2;
79 	else
80 		word >>= 2;
81 
82 	if ((word & 0x2) == 0)
83 		num += 1;
84 	return num;
85 }
86 
87 /*
88  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
89  * f2fs_set_bit makes MSB and LSB reversed in a byte.
90  * @size must be integral times of unsigned long.
91  * Example:
92  *                             MSB <--> LSB
93  *   f2fs_set_bit(0, bitmap) => 1000 0000
94  *   f2fs_set_bit(7, bitmap) => 0000 0001
95  */
96 static unsigned long __find_rev_next_bit(const unsigned long *addr,
97 			unsigned long size, unsigned long offset)
98 {
99 	const unsigned long *p = addr + BIT_WORD(offset);
100 	unsigned long result = size;
101 	unsigned long tmp;
102 
103 	if (offset >= size)
104 		return size;
105 
106 	size -= (offset & ~(BITS_PER_LONG - 1));
107 	offset %= BITS_PER_LONG;
108 
109 	while (1) {
110 		if (*p == 0)
111 			goto pass;
112 
113 		tmp = __reverse_ulong((unsigned char *)p);
114 
115 		tmp &= ~0UL >> offset;
116 		if (size < BITS_PER_LONG)
117 			tmp &= (~0UL << (BITS_PER_LONG - size));
118 		if (tmp)
119 			goto found;
120 pass:
121 		if (size <= BITS_PER_LONG)
122 			break;
123 		size -= BITS_PER_LONG;
124 		offset = 0;
125 		p++;
126 	}
127 	return result;
128 found:
129 	return result - size + __reverse_ffs(tmp);
130 }
131 
132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
133 			unsigned long size, unsigned long offset)
134 {
135 	const unsigned long *p = addr + BIT_WORD(offset);
136 	unsigned long result = size;
137 	unsigned long tmp;
138 
139 	if (offset >= size)
140 		return size;
141 
142 	size -= (offset & ~(BITS_PER_LONG - 1));
143 	offset %= BITS_PER_LONG;
144 
145 	while (1) {
146 		if (*p == ~0UL)
147 			goto pass;
148 
149 		tmp = __reverse_ulong((unsigned char *)p);
150 
151 		if (offset)
152 			tmp |= ~0UL << (BITS_PER_LONG - offset);
153 		if (size < BITS_PER_LONG)
154 			tmp |= ~0UL >> size;
155 		if (tmp != ~0UL)
156 			goto found;
157 pass:
158 		if (size <= BITS_PER_LONG)
159 			break;
160 		size -= BITS_PER_LONG;
161 		offset = 0;
162 		p++;
163 	}
164 	return result;
165 found:
166 	return result - size + __reverse_ffz(tmp);
167 }
168 
169 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
170 {
171 	int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
172 	int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
173 	int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
174 
175 	if (test_opt(sbi, LFS))
176 		return false;
177 	if (sbi->gc_mode == GC_URGENT)
178 		return true;
179 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
180 		return true;
181 
182 	return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
183 			SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
184 }
185 
186 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
187 {
188 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
189 	struct f2fs_inode_info *fi = F2FS_I(inode);
190 	struct inmem_pages *new;
191 
192 	f2fs_trace_pid(page);
193 
194 	f2fs_set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
195 
196 	new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
197 
198 	/* add atomic page indices to the list */
199 	new->page = page;
200 	INIT_LIST_HEAD(&new->list);
201 
202 	/* increase reference count with clean state */
203 	mutex_lock(&fi->inmem_lock);
204 	get_page(page);
205 	list_add_tail(&new->list, &fi->inmem_pages);
206 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
207 	if (list_empty(&fi->inmem_ilist))
208 		list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
209 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
210 	inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
211 	mutex_unlock(&fi->inmem_lock);
212 
213 	trace_f2fs_register_inmem_page(page, INMEM);
214 }
215 
216 static int __revoke_inmem_pages(struct inode *inode,
217 				struct list_head *head, bool drop, bool recover,
218 				bool trylock)
219 {
220 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
221 	struct inmem_pages *cur, *tmp;
222 	int err = 0;
223 
224 	list_for_each_entry_safe(cur, tmp, head, list) {
225 		struct page *page = cur->page;
226 
227 		if (drop)
228 			trace_f2fs_commit_inmem_page(page, INMEM_DROP);
229 
230 		if (trylock) {
231 			/*
232 			 * to avoid deadlock in between page lock and
233 			 * inmem_lock.
234 			 */
235 			if (!trylock_page(page))
236 				continue;
237 		} else {
238 			lock_page(page);
239 		}
240 
241 		f2fs_wait_on_page_writeback(page, DATA, true, true);
242 
243 		if (recover) {
244 			struct dnode_of_data dn;
245 			struct node_info ni;
246 
247 			trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
248 retry:
249 			set_new_dnode(&dn, inode, NULL, NULL, 0);
250 			err = f2fs_get_dnode_of_data(&dn, page->index,
251 								LOOKUP_NODE);
252 			if (err) {
253 				if (err == -ENOMEM) {
254 					congestion_wait(BLK_RW_ASYNC, HZ/50);
255 					cond_resched();
256 					goto retry;
257 				}
258 				err = -EAGAIN;
259 				goto next;
260 			}
261 
262 			err = f2fs_get_node_info(sbi, dn.nid, &ni);
263 			if (err) {
264 				f2fs_put_dnode(&dn);
265 				return err;
266 			}
267 
268 			if (cur->old_addr == NEW_ADDR) {
269 				f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
270 				f2fs_update_data_blkaddr(&dn, NEW_ADDR);
271 			} else
272 				f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
273 					cur->old_addr, ni.version, true, true);
274 			f2fs_put_dnode(&dn);
275 		}
276 next:
277 		/* we don't need to invalidate this in the sccessful status */
278 		if (drop || recover) {
279 			ClearPageUptodate(page);
280 			clear_cold_data(page);
281 		}
282 		f2fs_clear_page_private(page);
283 		f2fs_put_page(page, 1);
284 
285 		list_del(&cur->list);
286 		kmem_cache_free(inmem_entry_slab, cur);
287 		dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
288 	}
289 	return err;
290 }
291 
292 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
293 {
294 	struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
295 	struct inode *inode;
296 	struct f2fs_inode_info *fi;
297 next:
298 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
299 	if (list_empty(head)) {
300 		spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
301 		return;
302 	}
303 	fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
304 	inode = igrab(&fi->vfs_inode);
305 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
306 
307 	if (inode) {
308 		if (gc_failure) {
309 			if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
310 				goto drop;
311 			goto skip;
312 		}
313 drop:
314 		set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
315 		f2fs_drop_inmem_pages(inode);
316 		iput(inode);
317 	}
318 skip:
319 	congestion_wait(BLK_RW_ASYNC, HZ/50);
320 	cond_resched();
321 	goto next;
322 }
323 
324 void f2fs_drop_inmem_pages(struct inode *inode)
325 {
326 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
327 	struct f2fs_inode_info *fi = F2FS_I(inode);
328 
329 	while (!list_empty(&fi->inmem_pages)) {
330 		mutex_lock(&fi->inmem_lock);
331 		__revoke_inmem_pages(inode, &fi->inmem_pages,
332 						true, false, true);
333 
334 		if (list_empty(&fi->inmem_pages)) {
335 			spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
336 			if (!list_empty(&fi->inmem_ilist))
337 				list_del_init(&fi->inmem_ilist);
338 			spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
339 		}
340 		mutex_unlock(&fi->inmem_lock);
341 	}
342 
343 	clear_inode_flag(inode, FI_ATOMIC_FILE);
344 	fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
345 	stat_dec_atomic_write(inode);
346 }
347 
348 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
349 {
350 	struct f2fs_inode_info *fi = F2FS_I(inode);
351 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
352 	struct list_head *head = &fi->inmem_pages;
353 	struct inmem_pages *cur = NULL;
354 
355 	f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
356 
357 	mutex_lock(&fi->inmem_lock);
358 	list_for_each_entry(cur, head, list) {
359 		if (cur->page == page)
360 			break;
361 	}
362 
363 	f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
364 	list_del(&cur->list);
365 	mutex_unlock(&fi->inmem_lock);
366 
367 	dec_page_count(sbi, F2FS_INMEM_PAGES);
368 	kmem_cache_free(inmem_entry_slab, cur);
369 
370 	ClearPageUptodate(page);
371 	f2fs_clear_page_private(page);
372 	f2fs_put_page(page, 0);
373 
374 	trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
375 }
376 
377 static int __f2fs_commit_inmem_pages(struct inode *inode)
378 {
379 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
380 	struct f2fs_inode_info *fi = F2FS_I(inode);
381 	struct inmem_pages *cur, *tmp;
382 	struct f2fs_io_info fio = {
383 		.sbi = sbi,
384 		.ino = inode->i_ino,
385 		.type = DATA,
386 		.op = REQ_OP_WRITE,
387 		.op_flags = REQ_SYNC | REQ_PRIO,
388 		.io_type = FS_DATA_IO,
389 	};
390 	struct list_head revoke_list;
391 	bool submit_bio = false;
392 	int err = 0;
393 
394 	INIT_LIST_HEAD(&revoke_list);
395 
396 	list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
397 		struct page *page = cur->page;
398 
399 		lock_page(page);
400 		if (page->mapping == inode->i_mapping) {
401 			trace_f2fs_commit_inmem_page(page, INMEM);
402 
403 			f2fs_wait_on_page_writeback(page, DATA, true, true);
404 
405 			set_page_dirty(page);
406 			if (clear_page_dirty_for_io(page)) {
407 				inode_dec_dirty_pages(inode);
408 				f2fs_remove_dirty_inode(inode);
409 			}
410 retry:
411 			fio.page = page;
412 			fio.old_blkaddr = NULL_ADDR;
413 			fio.encrypted_page = NULL;
414 			fio.need_lock = LOCK_DONE;
415 			err = f2fs_do_write_data_page(&fio);
416 			if (err) {
417 				if (err == -ENOMEM) {
418 					congestion_wait(BLK_RW_ASYNC, HZ/50);
419 					cond_resched();
420 					goto retry;
421 				}
422 				unlock_page(page);
423 				break;
424 			}
425 			/* record old blkaddr for revoking */
426 			cur->old_addr = fio.old_blkaddr;
427 			submit_bio = true;
428 		}
429 		unlock_page(page);
430 		list_move_tail(&cur->list, &revoke_list);
431 	}
432 
433 	if (submit_bio)
434 		f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
435 
436 	if (err) {
437 		/*
438 		 * try to revoke all committed pages, but still we could fail
439 		 * due to no memory or other reason, if that happened, EAGAIN
440 		 * will be returned, which means in such case, transaction is
441 		 * already not integrity, caller should use journal to do the
442 		 * recovery or rewrite & commit last transaction. For other
443 		 * error number, revoking was done by filesystem itself.
444 		 */
445 		err = __revoke_inmem_pages(inode, &revoke_list,
446 						false, true, false);
447 
448 		/* drop all uncommitted pages */
449 		__revoke_inmem_pages(inode, &fi->inmem_pages,
450 						true, false, false);
451 	} else {
452 		__revoke_inmem_pages(inode, &revoke_list,
453 						false, false, false);
454 	}
455 
456 	return err;
457 }
458 
459 int f2fs_commit_inmem_pages(struct inode *inode)
460 {
461 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
462 	struct f2fs_inode_info *fi = F2FS_I(inode);
463 	int err;
464 
465 	f2fs_balance_fs(sbi, true);
466 
467 	down_write(&fi->i_gc_rwsem[WRITE]);
468 
469 	f2fs_lock_op(sbi);
470 	set_inode_flag(inode, FI_ATOMIC_COMMIT);
471 
472 	mutex_lock(&fi->inmem_lock);
473 	err = __f2fs_commit_inmem_pages(inode);
474 
475 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
476 	if (!list_empty(&fi->inmem_ilist))
477 		list_del_init(&fi->inmem_ilist);
478 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
479 	mutex_unlock(&fi->inmem_lock);
480 
481 	clear_inode_flag(inode, FI_ATOMIC_COMMIT);
482 
483 	f2fs_unlock_op(sbi);
484 	up_write(&fi->i_gc_rwsem[WRITE]);
485 
486 	return err;
487 }
488 
489 /*
490  * This function balances dirty node and dentry pages.
491  * In addition, it controls garbage collection.
492  */
493 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
494 {
495 	if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
496 		f2fs_show_injection_info(FAULT_CHECKPOINT);
497 		f2fs_stop_checkpoint(sbi, false);
498 	}
499 
500 	/* balance_fs_bg is able to be pending */
501 	if (need && excess_cached_nats(sbi))
502 		f2fs_balance_fs_bg(sbi);
503 
504 	if (f2fs_is_checkpoint_ready(sbi))
505 		return;
506 
507 	/*
508 	 * We should do GC or end up with checkpoint, if there are so many dirty
509 	 * dir/node pages without enough free segments.
510 	 */
511 	if (has_not_enough_free_secs(sbi, 0, 0)) {
512 		mutex_lock(&sbi->gc_mutex);
513 		f2fs_gc(sbi, false, false, NULL_SEGNO);
514 	}
515 }
516 
517 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
518 {
519 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
520 		return;
521 
522 	/* try to shrink extent cache when there is no enough memory */
523 	if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
524 		f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
525 
526 	/* check the # of cached NAT entries */
527 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
528 		f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
529 
530 	if (!f2fs_available_free_memory(sbi, FREE_NIDS))
531 		f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
532 	else
533 		f2fs_build_free_nids(sbi, false, false);
534 
535 	if (!is_idle(sbi, REQ_TIME) &&
536 		(!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
537 		return;
538 
539 	/* checkpoint is the only way to shrink partial cached entries */
540 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
541 			!f2fs_available_free_memory(sbi, INO_ENTRIES) ||
542 			excess_prefree_segs(sbi) ||
543 			excess_dirty_nats(sbi) ||
544 			excess_dirty_nodes(sbi) ||
545 			f2fs_time_over(sbi, CP_TIME)) {
546 		if (test_opt(sbi, DATA_FLUSH)) {
547 			struct blk_plug plug;
548 
549 			mutex_lock(&sbi->flush_lock);
550 
551 			blk_start_plug(&plug);
552 			f2fs_sync_dirty_inodes(sbi, FILE_INODE);
553 			blk_finish_plug(&plug);
554 
555 			mutex_unlock(&sbi->flush_lock);
556 		}
557 		f2fs_sync_fs(sbi->sb, true);
558 		stat_inc_bg_cp_count(sbi->stat_info);
559 	}
560 }
561 
562 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
563 				struct block_device *bdev)
564 {
565 	struct bio *bio;
566 	int ret;
567 
568 	bio = f2fs_bio_alloc(sbi, 0, false);
569 	if (!bio)
570 		return -ENOMEM;
571 
572 	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
573 	bio_set_dev(bio, bdev);
574 	ret = submit_bio_wait(bio);
575 	bio_put(bio);
576 
577 	trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
578 				test_opt(sbi, FLUSH_MERGE), ret);
579 	return ret;
580 }
581 
582 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
583 {
584 	int ret = 0;
585 	int i;
586 
587 	if (!f2fs_is_multi_device(sbi))
588 		return __submit_flush_wait(sbi, sbi->sb->s_bdev);
589 
590 	for (i = 0; i < sbi->s_ndevs; i++) {
591 		if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
592 			continue;
593 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
594 		if (ret)
595 			break;
596 	}
597 	return ret;
598 }
599 
600 static int issue_flush_thread(void *data)
601 {
602 	struct f2fs_sb_info *sbi = data;
603 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
604 	wait_queue_head_t *q = &fcc->flush_wait_queue;
605 repeat:
606 	if (kthread_should_stop())
607 		return 0;
608 
609 	sb_start_intwrite(sbi->sb);
610 
611 	if (!llist_empty(&fcc->issue_list)) {
612 		struct flush_cmd *cmd, *next;
613 		int ret;
614 
615 		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
616 		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
617 
618 		cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
619 
620 		ret = submit_flush_wait(sbi, cmd->ino);
621 		atomic_inc(&fcc->issued_flush);
622 
623 		llist_for_each_entry_safe(cmd, next,
624 					  fcc->dispatch_list, llnode) {
625 			cmd->ret = ret;
626 			complete(&cmd->wait);
627 		}
628 		fcc->dispatch_list = NULL;
629 	}
630 
631 	sb_end_intwrite(sbi->sb);
632 
633 	wait_event_interruptible(*q,
634 		kthread_should_stop() || !llist_empty(&fcc->issue_list));
635 	goto repeat;
636 }
637 
638 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
639 {
640 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
641 	struct flush_cmd cmd;
642 	int ret;
643 
644 	if (test_opt(sbi, NOBARRIER))
645 		return 0;
646 
647 	if (!test_opt(sbi, FLUSH_MERGE)) {
648 		atomic_inc(&fcc->queued_flush);
649 		ret = submit_flush_wait(sbi, ino);
650 		atomic_dec(&fcc->queued_flush);
651 		atomic_inc(&fcc->issued_flush);
652 		return ret;
653 	}
654 
655 	if (atomic_inc_return(&fcc->queued_flush) == 1 ||
656 	    f2fs_is_multi_device(sbi)) {
657 		ret = submit_flush_wait(sbi, ino);
658 		atomic_dec(&fcc->queued_flush);
659 
660 		atomic_inc(&fcc->issued_flush);
661 		return ret;
662 	}
663 
664 	cmd.ino = ino;
665 	init_completion(&cmd.wait);
666 
667 	llist_add(&cmd.llnode, &fcc->issue_list);
668 
669 	/* update issue_list before we wake up issue_flush thread */
670 	smp_mb();
671 
672 	if (waitqueue_active(&fcc->flush_wait_queue))
673 		wake_up(&fcc->flush_wait_queue);
674 
675 	if (fcc->f2fs_issue_flush) {
676 		wait_for_completion(&cmd.wait);
677 		atomic_dec(&fcc->queued_flush);
678 	} else {
679 		struct llist_node *list;
680 
681 		list = llist_del_all(&fcc->issue_list);
682 		if (!list) {
683 			wait_for_completion(&cmd.wait);
684 			atomic_dec(&fcc->queued_flush);
685 		} else {
686 			struct flush_cmd *tmp, *next;
687 
688 			ret = submit_flush_wait(sbi, ino);
689 
690 			llist_for_each_entry_safe(tmp, next, list, llnode) {
691 				if (tmp == &cmd) {
692 					cmd.ret = ret;
693 					atomic_dec(&fcc->queued_flush);
694 					continue;
695 				}
696 				tmp->ret = ret;
697 				complete(&tmp->wait);
698 			}
699 		}
700 	}
701 
702 	return cmd.ret;
703 }
704 
705 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
706 {
707 	dev_t dev = sbi->sb->s_bdev->bd_dev;
708 	struct flush_cmd_control *fcc;
709 	int err = 0;
710 
711 	if (SM_I(sbi)->fcc_info) {
712 		fcc = SM_I(sbi)->fcc_info;
713 		if (fcc->f2fs_issue_flush)
714 			return err;
715 		goto init_thread;
716 	}
717 
718 	fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
719 	if (!fcc)
720 		return -ENOMEM;
721 	atomic_set(&fcc->issued_flush, 0);
722 	atomic_set(&fcc->queued_flush, 0);
723 	init_waitqueue_head(&fcc->flush_wait_queue);
724 	init_llist_head(&fcc->issue_list);
725 	SM_I(sbi)->fcc_info = fcc;
726 	if (!test_opt(sbi, FLUSH_MERGE))
727 		return err;
728 
729 init_thread:
730 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
731 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
732 	if (IS_ERR(fcc->f2fs_issue_flush)) {
733 		err = PTR_ERR(fcc->f2fs_issue_flush);
734 		kvfree(fcc);
735 		SM_I(sbi)->fcc_info = NULL;
736 		return err;
737 	}
738 
739 	return err;
740 }
741 
742 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
743 {
744 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
745 
746 	if (fcc && fcc->f2fs_issue_flush) {
747 		struct task_struct *flush_thread = fcc->f2fs_issue_flush;
748 
749 		fcc->f2fs_issue_flush = NULL;
750 		kthread_stop(flush_thread);
751 	}
752 	if (free) {
753 		kvfree(fcc);
754 		SM_I(sbi)->fcc_info = NULL;
755 	}
756 }
757 
758 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
759 {
760 	int ret = 0, i;
761 
762 	if (!f2fs_is_multi_device(sbi))
763 		return 0;
764 
765 	for (i = 1; i < sbi->s_ndevs; i++) {
766 		if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
767 			continue;
768 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
769 		if (ret)
770 			break;
771 
772 		spin_lock(&sbi->dev_lock);
773 		f2fs_clear_bit(i, (char *)&sbi->dirty_device);
774 		spin_unlock(&sbi->dev_lock);
775 	}
776 
777 	return ret;
778 }
779 
780 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
781 		enum dirty_type dirty_type)
782 {
783 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
784 
785 	/* need not be added */
786 	if (IS_CURSEG(sbi, segno))
787 		return;
788 
789 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
790 		dirty_i->nr_dirty[dirty_type]++;
791 
792 	if (dirty_type == DIRTY) {
793 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
794 		enum dirty_type t = sentry->type;
795 
796 		if (unlikely(t >= DIRTY)) {
797 			f2fs_bug_on(sbi, 1);
798 			return;
799 		}
800 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
801 			dirty_i->nr_dirty[t]++;
802 	}
803 }
804 
805 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
806 		enum dirty_type dirty_type)
807 {
808 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
809 
810 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
811 		dirty_i->nr_dirty[dirty_type]--;
812 
813 	if (dirty_type == DIRTY) {
814 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
815 		enum dirty_type t = sentry->type;
816 
817 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
818 			dirty_i->nr_dirty[t]--;
819 
820 		if (get_valid_blocks(sbi, segno, true) == 0)
821 			clear_bit(GET_SEC_FROM_SEG(sbi, segno),
822 						dirty_i->victim_secmap);
823 	}
824 }
825 
826 /*
827  * Should not occur error such as -ENOMEM.
828  * Adding dirty entry into seglist is not critical operation.
829  * If a given segment is one of current working segments, it won't be added.
830  */
831 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
832 {
833 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
834 	unsigned short valid_blocks, ckpt_valid_blocks;
835 
836 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
837 		return;
838 
839 	mutex_lock(&dirty_i->seglist_lock);
840 
841 	valid_blocks = get_valid_blocks(sbi, segno, false);
842 	ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
843 
844 	if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
845 				ckpt_valid_blocks == sbi->blocks_per_seg)) {
846 		__locate_dirty_segment(sbi, segno, PRE);
847 		__remove_dirty_segment(sbi, segno, DIRTY);
848 	} else if (valid_blocks < sbi->blocks_per_seg) {
849 		__locate_dirty_segment(sbi, segno, DIRTY);
850 	} else {
851 		/* Recovery routine with SSR needs this */
852 		__remove_dirty_segment(sbi, segno, DIRTY);
853 	}
854 
855 	mutex_unlock(&dirty_i->seglist_lock);
856 }
857 
858 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
859 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
860 {
861 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
862 	unsigned int segno;
863 
864 	mutex_lock(&dirty_i->seglist_lock);
865 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
866 		if (get_valid_blocks(sbi, segno, false))
867 			continue;
868 		if (IS_CURSEG(sbi, segno))
869 			continue;
870 		__locate_dirty_segment(sbi, segno, PRE);
871 		__remove_dirty_segment(sbi, segno, DIRTY);
872 	}
873 	mutex_unlock(&dirty_i->seglist_lock);
874 }
875 
876 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
877 {
878 	int ovp_hole_segs =
879 		(overprovision_segments(sbi) - reserved_segments(sbi));
880 	block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
881 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
882 	block_t holes[2] = {0, 0};	/* DATA and NODE */
883 	block_t unusable;
884 	struct seg_entry *se;
885 	unsigned int segno;
886 
887 	mutex_lock(&dirty_i->seglist_lock);
888 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
889 		se = get_seg_entry(sbi, segno);
890 		if (IS_NODESEG(se->type))
891 			holes[NODE] += sbi->blocks_per_seg - se->valid_blocks;
892 		else
893 			holes[DATA] += sbi->blocks_per_seg - se->valid_blocks;
894 	}
895 	mutex_unlock(&dirty_i->seglist_lock);
896 
897 	unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
898 	if (unusable > ovp_holes)
899 		return unusable - ovp_holes;
900 	return 0;
901 }
902 
903 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
904 {
905 	int ovp_hole_segs =
906 		(overprovision_segments(sbi) - reserved_segments(sbi));
907 	if (unusable > F2FS_OPTION(sbi).unusable_cap)
908 		return -EAGAIN;
909 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
910 		dirty_segments(sbi) > ovp_hole_segs)
911 		return -EAGAIN;
912 	return 0;
913 }
914 
915 /* This is only used by SBI_CP_DISABLED */
916 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
917 {
918 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
919 	unsigned int segno = 0;
920 
921 	mutex_lock(&dirty_i->seglist_lock);
922 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
923 		if (get_valid_blocks(sbi, segno, false))
924 			continue;
925 		if (get_ckpt_valid_blocks(sbi, segno))
926 			continue;
927 		mutex_unlock(&dirty_i->seglist_lock);
928 		return segno;
929 	}
930 	mutex_unlock(&dirty_i->seglist_lock);
931 	return NULL_SEGNO;
932 }
933 
934 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
935 		struct block_device *bdev, block_t lstart,
936 		block_t start, block_t len)
937 {
938 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
939 	struct list_head *pend_list;
940 	struct discard_cmd *dc;
941 
942 	f2fs_bug_on(sbi, !len);
943 
944 	pend_list = &dcc->pend_list[plist_idx(len)];
945 
946 	dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
947 	INIT_LIST_HEAD(&dc->list);
948 	dc->bdev = bdev;
949 	dc->lstart = lstart;
950 	dc->start = start;
951 	dc->len = len;
952 	dc->ref = 0;
953 	dc->state = D_PREP;
954 	dc->queued = 0;
955 	dc->error = 0;
956 	init_completion(&dc->wait);
957 	list_add_tail(&dc->list, pend_list);
958 	spin_lock_init(&dc->lock);
959 	dc->bio_ref = 0;
960 	atomic_inc(&dcc->discard_cmd_cnt);
961 	dcc->undiscard_blks += len;
962 
963 	return dc;
964 }
965 
966 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
967 				struct block_device *bdev, block_t lstart,
968 				block_t start, block_t len,
969 				struct rb_node *parent, struct rb_node **p,
970 				bool leftmost)
971 {
972 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
973 	struct discard_cmd *dc;
974 
975 	dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
976 
977 	rb_link_node(&dc->rb_node, parent, p);
978 	rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
979 
980 	return dc;
981 }
982 
983 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
984 							struct discard_cmd *dc)
985 {
986 	if (dc->state == D_DONE)
987 		atomic_sub(dc->queued, &dcc->queued_discard);
988 
989 	list_del(&dc->list);
990 	rb_erase_cached(&dc->rb_node, &dcc->root);
991 	dcc->undiscard_blks -= dc->len;
992 
993 	kmem_cache_free(discard_cmd_slab, dc);
994 
995 	atomic_dec(&dcc->discard_cmd_cnt);
996 }
997 
998 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
999 							struct discard_cmd *dc)
1000 {
1001 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1002 	unsigned long flags;
1003 
1004 	trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
1005 
1006 	spin_lock_irqsave(&dc->lock, flags);
1007 	if (dc->bio_ref) {
1008 		spin_unlock_irqrestore(&dc->lock, flags);
1009 		return;
1010 	}
1011 	spin_unlock_irqrestore(&dc->lock, flags);
1012 
1013 	f2fs_bug_on(sbi, dc->ref);
1014 
1015 	if (dc->error == -EOPNOTSUPP)
1016 		dc->error = 0;
1017 
1018 	if (dc->error)
1019 		printk_ratelimited(
1020 			"%sF2FS-fs: Issue discard(%u, %u, %u) failed, ret: %d",
1021 			KERN_INFO, dc->lstart, dc->start, dc->len, dc->error);
1022 	__detach_discard_cmd(dcc, dc);
1023 }
1024 
1025 static void f2fs_submit_discard_endio(struct bio *bio)
1026 {
1027 	struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1028 	unsigned long flags;
1029 
1030 	dc->error = blk_status_to_errno(bio->bi_status);
1031 
1032 	spin_lock_irqsave(&dc->lock, flags);
1033 	dc->bio_ref--;
1034 	if (!dc->bio_ref && dc->state == D_SUBMIT) {
1035 		dc->state = D_DONE;
1036 		complete_all(&dc->wait);
1037 	}
1038 	spin_unlock_irqrestore(&dc->lock, flags);
1039 	bio_put(bio);
1040 }
1041 
1042 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1043 				block_t start, block_t end)
1044 {
1045 #ifdef CONFIG_F2FS_CHECK_FS
1046 	struct seg_entry *sentry;
1047 	unsigned int segno;
1048 	block_t blk = start;
1049 	unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1050 	unsigned long *map;
1051 
1052 	while (blk < end) {
1053 		segno = GET_SEGNO(sbi, blk);
1054 		sentry = get_seg_entry(sbi, segno);
1055 		offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1056 
1057 		if (end < START_BLOCK(sbi, segno + 1))
1058 			size = GET_BLKOFF_FROM_SEG0(sbi, end);
1059 		else
1060 			size = max_blocks;
1061 		map = (unsigned long *)(sentry->cur_valid_map);
1062 		offset = __find_rev_next_bit(map, size, offset);
1063 		f2fs_bug_on(sbi, offset != size);
1064 		blk = START_BLOCK(sbi, segno + 1);
1065 	}
1066 #endif
1067 }
1068 
1069 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1070 				struct discard_policy *dpolicy,
1071 				int discard_type, unsigned int granularity)
1072 {
1073 	/* common policy */
1074 	dpolicy->type = discard_type;
1075 	dpolicy->sync = true;
1076 	dpolicy->ordered = false;
1077 	dpolicy->granularity = granularity;
1078 
1079 	dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1080 	dpolicy->io_aware_gran = MAX_PLIST_NUM;
1081 	dpolicy->timeout = 0;
1082 
1083 	if (discard_type == DPOLICY_BG) {
1084 		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1085 		dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1086 		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1087 		dpolicy->io_aware = true;
1088 		dpolicy->sync = false;
1089 		dpolicy->ordered = true;
1090 		if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1091 			dpolicy->granularity = 1;
1092 			dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1093 		}
1094 	} else if (discard_type == DPOLICY_FORCE) {
1095 		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1096 		dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1097 		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1098 		dpolicy->io_aware = false;
1099 	} else if (discard_type == DPOLICY_FSTRIM) {
1100 		dpolicy->io_aware = false;
1101 	} else if (discard_type == DPOLICY_UMOUNT) {
1102 		dpolicy->max_requests = UINT_MAX;
1103 		dpolicy->io_aware = false;
1104 		/* we need to issue all to keep CP_TRIMMED_FLAG */
1105 		dpolicy->granularity = 1;
1106 	}
1107 }
1108 
1109 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1110 				struct block_device *bdev, block_t lstart,
1111 				block_t start, block_t len);
1112 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1113 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1114 						struct discard_policy *dpolicy,
1115 						struct discard_cmd *dc,
1116 						unsigned int *issued)
1117 {
1118 	struct block_device *bdev = dc->bdev;
1119 	struct request_queue *q = bdev_get_queue(bdev);
1120 	unsigned int max_discard_blocks =
1121 			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1122 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1123 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1124 					&(dcc->fstrim_list) : &(dcc->wait_list);
1125 	int flag = dpolicy->sync ? REQ_SYNC : 0;
1126 	block_t lstart, start, len, total_len;
1127 	int err = 0;
1128 
1129 	if (dc->state != D_PREP)
1130 		return 0;
1131 
1132 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1133 		return 0;
1134 
1135 	trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1136 
1137 	lstart = dc->lstart;
1138 	start = dc->start;
1139 	len = dc->len;
1140 	total_len = len;
1141 
1142 	dc->len = 0;
1143 
1144 	while (total_len && *issued < dpolicy->max_requests && !err) {
1145 		struct bio *bio = NULL;
1146 		unsigned long flags;
1147 		bool last = true;
1148 
1149 		if (len > max_discard_blocks) {
1150 			len = max_discard_blocks;
1151 			last = false;
1152 		}
1153 
1154 		(*issued)++;
1155 		if (*issued == dpolicy->max_requests)
1156 			last = true;
1157 
1158 		dc->len += len;
1159 
1160 		if (time_to_inject(sbi, FAULT_DISCARD)) {
1161 			f2fs_show_injection_info(FAULT_DISCARD);
1162 			err = -EIO;
1163 			goto submit;
1164 		}
1165 		err = __blkdev_issue_discard(bdev,
1166 					SECTOR_FROM_BLOCK(start),
1167 					SECTOR_FROM_BLOCK(len),
1168 					GFP_NOFS, 0, &bio);
1169 submit:
1170 		if (err) {
1171 			spin_lock_irqsave(&dc->lock, flags);
1172 			if (dc->state == D_PARTIAL)
1173 				dc->state = D_SUBMIT;
1174 			spin_unlock_irqrestore(&dc->lock, flags);
1175 
1176 			break;
1177 		}
1178 
1179 		f2fs_bug_on(sbi, !bio);
1180 
1181 		/*
1182 		 * should keep before submission to avoid D_DONE
1183 		 * right away
1184 		 */
1185 		spin_lock_irqsave(&dc->lock, flags);
1186 		if (last)
1187 			dc->state = D_SUBMIT;
1188 		else
1189 			dc->state = D_PARTIAL;
1190 		dc->bio_ref++;
1191 		spin_unlock_irqrestore(&dc->lock, flags);
1192 
1193 		atomic_inc(&dcc->queued_discard);
1194 		dc->queued++;
1195 		list_move_tail(&dc->list, wait_list);
1196 
1197 		/* sanity check on discard range */
1198 		__check_sit_bitmap(sbi, lstart, lstart + len);
1199 
1200 		bio->bi_private = dc;
1201 		bio->bi_end_io = f2fs_submit_discard_endio;
1202 		bio->bi_opf |= flag;
1203 		submit_bio(bio);
1204 
1205 		atomic_inc(&dcc->issued_discard);
1206 
1207 		f2fs_update_iostat(sbi, FS_DISCARD, 1);
1208 
1209 		lstart += len;
1210 		start += len;
1211 		total_len -= len;
1212 		len = total_len;
1213 	}
1214 
1215 	if (!err && len)
1216 		__update_discard_tree_range(sbi, bdev, lstart, start, len);
1217 	return err;
1218 }
1219 
1220 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1221 				struct block_device *bdev, block_t lstart,
1222 				block_t start, block_t len,
1223 				struct rb_node **insert_p,
1224 				struct rb_node *insert_parent)
1225 {
1226 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1227 	struct rb_node **p;
1228 	struct rb_node *parent = NULL;
1229 	struct discard_cmd *dc = NULL;
1230 	bool leftmost = true;
1231 
1232 	if (insert_p && insert_parent) {
1233 		parent = insert_parent;
1234 		p = insert_p;
1235 		goto do_insert;
1236 	}
1237 
1238 	p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1239 							lstart, &leftmost);
1240 do_insert:
1241 	dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1242 								p, leftmost);
1243 	if (!dc)
1244 		return NULL;
1245 
1246 	return dc;
1247 }
1248 
1249 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1250 						struct discard_cmd *dc)
1251 {
1252 	list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1253 }
1254 
1255 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1256 				struct discard_cmd *dc, block_t blkaddr)
1257 {
1258 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1259 	struct discard_info di = dc->di;
1260 	bool modified = false;
1261 
1262 	if (dc->state == D_DONE || dc->len == 1) {
1263 		__remove_discard_cmd(sbi, dc);
1264 		return;
1265 	}
1266 
1267 	dcc->undiscard_blks -= di.len;
1268 
1269 	if (blkaddr > di.lstart) {
1270 		dc->len = blkaddr - dc->lstart;
1271 		dcc->undiscard_blks += dc->len;
1272 		__relocate_discard_cmd(dcc, dc);
1273 		modified = true;
1274 	}
1275 
1276 	if (blkaddr < di.lstart + di.len - 1) {
1277 		if (modified) {
1278 			__insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1279 					di.start + blkaddr + 1 - di.lstart,
1280 					di.lstart + di.len - 1 - blkaddr,
1281 					NULL, NULL);
1282 		} else {
1283 			dc->lstart++;
1284 			dc->len--;
1285 			dc->start++;
1286 			dcc->undiscard_blks += dc->len;
1287 			__relocate_discard_cmd(dcc, dc);
1288 		}
1289 	}
1290 }
1291 
1292 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1293 				struct block_device *bdev, block_t lstart,
1294 				block_t start, block_t len)
1295 {
1296 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1297 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1298 	struct discard_cmd *dc;
1299 	struct discard_info di = {0};
1300 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1301 	struct request_queue *q = bdev_get_queue(bdev);
1302 	unsigned int max_discard_blocks =
1303 			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1304 	block_t end = lstart + len;
1305 
1306 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1307 					NULL, lstart,
1308 					(struct rb_entry **)&prev_dc,
1309 					(struct rb_entry **)&next_dc,
1310 					&insert_p, &insert_parent, true, NULL);
1311 	if (dc)
1312 		prev_dc = dc;
1313 
1314 	if (!prev_dc) {
1315 		di.lstart = lstart;
1316 		di.len = next_dc ? next_dc->lstart - lstart : len;
1317 		di.len = min(di.len, len);
1318 		di.start = start;
1319 	}
1320 
1321 	while (1) {
1322 		struct rb_node *node;
1323 		bool merged = false;
1324 		struct discard_cmd *tdc = NULL;
1325 
1326 		if (prev_dc) {
1327 			di.lstart = prev_dc->lstart + prev_dc->len;
1328 			if (di.lstart < lstart)
1329 				di.lstart = lstart;
1330 			if (di.lstart >= end)
1331 				break;
1332 
1333 			if (!next_dc || next_dc->lstart > end)
1334 				di.len = end - di.lstart;
1335 			else
1336 				di.len = next_dc->lstart - di.lstart;
1337 			di.start = start + di.lstart - lstart;
1338 		}
1339 
1340 		if (!di.len)
1341 			goto next;
1342 
1343 		if (prev_dc && prev_dc->state == D_PREP &&
1344 			prev_dc->bdev == bdev &&
1345 			__is_discard_back_mergeable(&di, &prev_dc->di,
1346 							max_discard_blocks)) {
1347 			prev_dc->di.len += di.len;
1348 			dcc->undiscard_blks += di.len;
1349 			__relocate_discard_cmd(dcc, prev_dc);
1350 			di = prev_dc->di;
1351 			tdc = prev_dc;
1352 			merged = true;
1353 		}
1354 
1355 		if (next_dc && next_dc->state == D_PREP &&
1356 			next_dc->bdev == bdev &&
1357 			__is_discard_front_mergeable(&di, &next_dc->di,
1358 							max_discard_blocks)) {
1359 			next_dc->di.lstart = di.lstart;
1360 			next_dc->di.len += di.len;
1361 			next_dc->di.start = di.start;
1362 			dcc->undiscard_blks += di.len;
1363 			__relocate_discard_cmd(dcc, next_dc);
1364 			if (tdc)
1365 				__remove_discard_cmd(sbi, tdc);
1366 			merged = true;
1367 		}
1368 
1369 		if (!merged) {
1370 			__insert_discard_tree(sbi, bdev, di.lstart, di.start,
1371 							di.len, NULL, NULL);
1372 		}
1373  next:
1374 		prev_dc = next_dc;
1375 		if (!prev_dc)
1376 			break;
1377 
1378 		node = rb_next(&prev_dc->rb_node);
1379 		next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1380 	}
1381 }
1382 
1383 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1384 		struct block_device *bdev, block_t blkstart, block_t blklen)
1385 {
1386 	block_t lblkstart = blkstart;
1387 
1388 	if (!f2fs_bdev_support_discard(bdev))
1389 		return 0;
1390 
1391 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
1392 
1393 	if (f2fs_is_multi_device(sbi)) {
1394 		int devi = f2fs_target_device_index(sbi, blkstart);
1395 
1396 		blkstart -= FDEV(devi).start_blk;
1397 	}
1398 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1399 	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1400 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1401 	return 0;
1402 }
1403 
1404 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1405 					struct discard_policy *dpolicy)
1406 {
1407 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1408 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1409 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1410 	struct discard_cmd *dc;
1411 	struct blk_plug plug;
1412 	unsigned int pos = dcc->next_pos;
1413 	unsigned int issued = 0;
1414 	bool io_interrupted = false;
1415 
1416 	mutex_lock(&dcc->cmd_lock);
1417 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1418 					NULL, pos,
1419 					(struct rb_entry **)&prev_dc,
1420 					(struct rb_entry **)&next_dc,
1421 					&insert_p, &insert_parent, true, NULL);
1422 	if (!dc)
1423 		dc = next_dc;
1424 
1425 	blk_start_plug(&plug);
1426 
1427 	while (dc) {
1428 		struct rb_node *node;
1429 		int err = 0;
1430 
1431 		if (dc->state != D_PREP)
1432 			goto next;
1433 
1434 		if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1435 			io_interrupted = true;
1436 			break;
1437 		}
1438 
1439 		dcc->next_pos = dc->lstart + dc->len;
1440 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1441 
1442 		if (issued >= dpolicy->max_requests)
1443 			break;
1444 next:
1445 		node = rb_next(&dc->rb_node);
1446 		if (err)
1447 			__remove_discard_cmd(sbi, dc);
1448 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1449 	}
1450 
1451 	blk_finish_plug(&plug);
1452 
1453 	if (!dc)
1454 		dcc->next_pos = 0;
1455 
1456 	mutex_unlock(&dcc->cmd_lock);
1457 
1458 	if (!issued && io_interrupted)
1459 		issued = -1;
1460 
1461 	return issued;
1462 }
1463 
1464 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1465 					struct discard_policy *dpolicy)
1466 {
1467 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1468 	struct list_head *pend_list;
1469 	struct discard_cmd *dc, *tmp;
1470 	struct blk_plug plug;
1471 	int i, issued = 0;
1472 	bool io_interrupted = false;
1473 
1474 	if (dpolicy->timeout != 0)
1475 		f2fs_update_time(sbi, dpolicy->timeout);
1476 
1477 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1478 		if (dpolicy->timeout != 0 &&
1479 				f2fs_time_over(sbi, dpolicy->timeout))
1480 			break;
1481 
1482 		if (i + 1 < dpolicy->granularity)
1483 			break;
1484 
1485 		if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1486 			return __issue_discard_cmd_orderly(sbi, dpolicy);
1487 
1488 		pend_list = &dcc->pend_list[i];
1489 
1490 		mutex_lock(&dcc->cmd_lock);
1491 		if (list_empty(pend_list))
1492 			goto next;
1493 		if (unlikely(dcc->rbtree_check))
1494 			f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1495 								&dcc->root));
1496 		blk_start_plug(&plug);
1497 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1498 			f2fs_bug_on(sbi, dc->state != D_PREP);
1499 
1500 			if (dpolicy->timeout != 0 &&
1501 				f2fs_time_over(sbi, dpolicy->timeout))
1502 				break;
1503 
1504 			if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1505 						!is_idle(sbi, DISCARD_TIME)) {
1506 				io_interrupted = true;
1507 				break;
1508 			}
1509 
1510 			__submit_discard_cmd(sbi, dpolicy, dc, &issued);
1511 
1512 			if (issued >= dpolicy->max_requests)
1513 				break;
1514 		}
1515 		blk_finish_plug(&plug);
1516 next:
1517 		mutex_unlock(&dcc->cmd_lock);
1518 
1519 		if (issued >= dpolicy->max_requests || io_interrupted)
1520 			break;
1521 	}
1522 
1523 	if (!issued && io_interrupted)
1524 		issued = -1;
1525 
1526 	return issued;
1527 }
1528 
1529 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1530 {
1531 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1532 	struct list_head *pend_list;
1533 	struct discard_cmd *dc, *tmp;
1534 	int i;
1535 	bool dropped = false;
1536 
1537 	mutex_lock(&dcc->cmd_lock);
1538 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1539 		pend_list = &dcc->pend_list[i];
1540 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1541 			f2fs_bug_on(sbi, dc->state != D_PREP);
1542 			__remove_discard_cmd(sbi, dc);
1543 			dropped = true;
1544 		}
1545 	}
1546 	mutex_unlock(&dcc->cmd_lock);
1547 
1548 	return dropped;
1549 }
1550 
1551 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1552 {
1553 	__drop_discard_cmd(sbi);
1554 }
1555 
1556 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1557 							struct discard_cmd *dc)
1558 {
1559 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1560 	unsigned int len = 0;
1561 
1562 	wait_for_completion_io(&dc->wait);
1563 	mutex_lock(&dcc->cmd_lock);
1564 	f2fs_bug_on(sbi, dc->state != D_DONE);
1565 	dc->ref--;
1566 	if (!dc->ref) {
1567 		if (!dc->error)
1568 			len = dc->len;
1569 		__remove_discard_cmd(sbi, dc);
1570 	}
1571 	mutex_unlock(&dcc->cmd_lock);
1572 
1573 	return len;
1574 }
1575 
1576 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1577 						struct discard_policy *dpolicy,
1578 						block_t start, block_t end)
1579 {
1580 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1581 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1582 					&(dcc->fstrim_list) : &(dcc->wait_list);
1583 	struct discard_cmd *dc, *tmp;
1584 	bool need_wait;
1585 	unsigned int trimmed = 0;
1586 
1587 next:
1588 	need_wait = false;
1589 
1590 	mutex_lock(&dcc->cmd_lock);
1591 	list_for_each_entry_safe(dc, tmp, wait_list, list) {
1592 		if (dc->lstart + dc->len <= start || end <= dc->lstart)
1593 			continue;
1594 		if (dc->len < dpolicy->granularity)
1595 			continue;
1596 		if (dc->state == D_DONE && !dc->ref) {
1597 			wait_for_completion_io(&dc->wait);
1598 			if (!dc->error)
1599 				trimmed += dc->len;
1600 			__remove_discard_cmd(sbi, dc);
1601 		} else {
1602 			dc->ref++;
1603 			need_wait = true;
1604 			break;
1605 		}
1606 	}
1607 	mutex_unlock(&dcc->cmd_lock);
1608 
1609 	if (need_wait) {
1610 		trimmed += __wait_one_discard_bio(sbi, dc);
1611 		goto next;
1612 	}
1613 
1614 	return trimmed;
1615 }
1616 
1617 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1618 						struct discard_policy *dpolicy)
1619 {
1620 	struct discard_policy dp;
1621 	unsigned int discard_blks;
1622 
1623 	if (dpolicy)
1624 		return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1625 
1626 	/* wait all */
1627 	__init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1628 	discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1629 	__init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1630 	discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1631 
1632 	return discard_blks;
1633 }
1634 
1635 /* This should be covered by global mutex, &sit_i->sentry_lock */
1636 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1637 {
1638 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1639 	struct discard_cmd *dc;
1640 	bool need_wait = false;
1641 
1642 	mutex_lock(&dcc->cmd_lock);
1643 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1644 							NULL, blkaddr);
1645 	if (dc) {
1646 		if (dc->state == D_PREP) {
1647 			__punch_discard_cmd(sbi, dc, blkaddr);
1648 		} else {
1649 			dc->ref++;
1650 			need_wait = true;
1651 		}
1652 	}
1653 	mutex_unlock(&dcc->cmd_lock);
1654 
1655 	if (need_wait)
1656 		__wait_one_discard_bio(sbi, dc);
1657 }
1658 
1659 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1660 {
1661 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1662 
1663 	if (dcc && dcc->f2fs_issue_discard) {
1664 		struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1665 
1666 		dcc->f2fs_issue_discard = NULL;
1667 		kthread_stop(discard_thread);
1668 	}
1669 }
1670 
1671 /* This comes from f2fs_put_super */
1672 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1673 {
1674 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1675 	struct discard_policy dpolicy;
1676 	bool dropped;
1677 
1678 	__init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1679 					dcc->discard_granularity);
1680 	dpolicy.timeout = UMOUNT_DISCARD_TIMEOUT;
1681 	__issue_discard_cmd(sbi, &dpolicy);
1682 	dropped = __drop_discard_cmd(sbi);
1683 
1684 	/* just to make sure there is no pending discard commands */
1685 	__wait_all_discard_cmd(sbi, NULL);
1686 
1687 	f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1688 	return dropped;
1689 }
1690 
1691 static int issue_discard_thread(void *data)
1692 {
1693 	struct f2fs_sb_info *sbi = data;
1694 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1695 	wait_queue_head_t *q = &dcc->discard_wait_queue;
1696 	struct discard_policy dpolicy;
1697 	unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1698 	int issued;
1699 
1700 	set_freezable();
1701 
1702 	do {
1703 		__init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1704 					dcc->discard_granularity);
1705 
1706 		wait_event_interruptible_timeout(*q,
1707 				kthread_should_stop() || freezing(current) ||
1708 				dcc->discard_wake,
1709 				msecs_to_jiffies(wait_ms));
1710 
1711 		if (dcc->discard_wake)
1712 			dcc->discard_wake = 0;
1713 
1714 		/* clean up pending candidates before going to sleep */
1715 		if (atomic_read(&dcc->queued_discard))
1716 			__wait_all_discard_cmd(sbi, NULL);
1717 
1718 		if (try_to_freeze())
1719 			continue;
1720 		if (f2fs_readonly(sbi->sb))
1721 			continue;
1722 		if (kthread_should_stop())
1723 			return 0;
1724 		if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1725 			wait_ms = dpolicy.max_interval;
1726 			continue;
1727 		}
1728 
1729 		if (sbi->gc_mode == GC_URGENT)
1730 			__init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1731 
1732 		sb_start_intwrite(sbi->sb);
1733 
1734 		issued = __issue_discard_cmd(sbi, &dpolicy);
1735 		if (issued > 0) {
1736 			__wait_all_discard_cmd(sbi, &dpolicy);
1737 			wait_ms = dpolicy.min_interval;
1738 		} else if (issued == -1){
1739 			wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1740 			if (!wait_ms)
1741 				wait_ms = dpolicy.mid_interval;
1742 		} else {
1743 			wait_ms = dpolicy.max_interval;
1744 		}
1745 
1746 		sb_end_intwrite(sbi->sb);
1747 
1748 	} while (!kthread_should_stop());
1749 	return 0;
1750 }
1751 
1752 #ifdef CONFIG_BLK_DEV_ZONED
1753 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1754 		struct block_device *bdev, block_t blkstart, block_t blklen)
1755 {
1756 	sector_t sector, nr_sects;
1757 	block_t lblkstart = blkstart;
1758 	int devi = 0;
1759 
1760 	if (f2fs_is_multi_device(sbi)) {
1761 		devi = f2fs_target_device_index(sbi, blkstart);
1762 		if (blkstart < FDEV(devi).start_blk ||
1763 		    blkstart > FDEV(devi).end_blk) {
1764 			f2fs_err(sbi, "Invalid block %x", blkstart);
1765 			return -EIO;
1766 		}
1767 		blkstart -= FDEV(devi).start_blk;
1768 	}
1769 
1770 	/* For sequential zones, reset the zone write pointer */
1771 	if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1772 		sector = SECTOR_FROM_BLOCK(blkstart);
1773 		nr_sects = SECTOR_FROM_BLOCK(blklen);
1774 
1775 		if (sector & (bdev_zone_sectors(bdev) - 1) ||
1776 				nr_sects != bdev_zone_sectors(bdev)) {
1777 			f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1778 				 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1779 				 blkstart, blklen);
1780 			return -EIO;
1781 		}
1782 		trace_f2fs_issue_reset_zone(bdev, blkstart);
1783 		return blkdev_reset_zones(bdev, sector, nr_sects, GFP_NOFS);
1784 	}
1785 
1786 	/* For conventional zones, use regular discard if supported */
1787 	return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1788 }
1789 #endif
1790 
1791 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1792 		struct block_device *bdev, block_t blkstart, block_t blklen)
1793 {
1794 #ifdef CONFIG_BLK_DEV_ZONED
1795 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1796 		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1797 #endif
1798 	return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1799 }
1800 
1801 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1802 				block_t blkstart, block_t blklen)
1803 {
1804 	sector_t start = blkstart, len = 0;
1805 	struct block_device *bdev;
1806 	struct seg_entry *se;
1807 	unsigned int offset;
1808 	block_t i;
1809 	int err = 0;
1810 
1811 	bdev = f2fs_target_device(sbi, blkstart, NULL);
1812 
1813 	for (i = blkstart; i < blkstart + blklen; i++, len++) {
1814 		if (i != start) {
1815 			struct block_device *bdev2 =
1816 				f2fs_target_device(sbi, i, NULL);
1817 
1818 			if (bdev2 != bdev) {
1819 				err = __issue_discard_async(sbi, bdev,
1820 						start, len);
1821 				if (err)
1822 					return err;
1823 				bdev = bdev2;
1824 				start = i;
1825 				len = 0;
1826 			}
1827 		}
1828 
1829 		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1830 		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1831 
1832 		if (!f2fs_test_and_set_bit(offset, se->discard_map))
1833 			sbi->discard_blks--;
1834 	}
1835 
1836 	if (len)
1837 		err = __issue_discard_async(sbi, bdev, start, len);
1838 	return err;
1839 }
1840 
1841 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1842 							bool check_only)
1843 {
1844 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1845 	int max_blocks = sbi->blocks_per_seg;
1846 	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1847 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1848 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1849 	unsigned long *discard_map = (unsigned long *)se->discard_map;
1850 	unsigned long *dmap = SIT_I(sbi)->tmp_map;
1851 	unsigned int start = 0, end = -1;
1852 	bool force = (cpc->reason & CP_DISCARD);
1853 	struct discard_entry *de = NULL;
1854 	struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1855 	int i;
1856 
1857 	if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1858 		return false;
1859 
1860 	if (!force) {
1861 		if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1862 			SM_I(sbi)->dcc_info->nr_discards >=
1863 				SM_I(sbi)->dcc_info->max_discards)
1864 			return false;
1865 	}
1866 
1867 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1868 	for (i = 0; i < entries; i++)
1869 		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1870 				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1871 
1872 	while (force || SM_I(sbi)->dcc_info->nr_discards <=
1873 				SM_I(sbi)->dcc_info->max_discards) {
1874 		start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1875 		if (start >= max_blocks)
1876 			break;
1877 
1878 		end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1879 		if (force && start && end != max_blocks
1880 					&& (end - start) < cpc->trim_minlen)
1881 			continue;
1882 
1883 		if (check_only)
1884 			return true;
1885 
1886 		if (!de) {
1887 			de = f2fs_kmem_cache_alloc(discard_entry_slab,
1888 								GFP_F2FS_ZERO);
1889 			de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1890 			list_add_tail(&de->list, head);
1891 		}
1892 
1893 		for (i = start; i < end; i++)
1894 			__set_bit_le(i, (void *)de->discard_map);
1895 
1896 		SM_I(sbi)->dcc_info->nr_discards += end - start;
1897 	}
1898 	return false;
1899 }
1900 
1901 static void release_discard_addr(struct discard_entry *entry)
1902 {
1903 	list_del(&entry->list);
1904 	kmem_cache_free(discard_entry_slab, entry);
1905 }
1906 
1907 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1908 {
1909 	struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1910 	struct discard_entry *entry, *this;
1911 
1912 	/* drop caches */
1913 	list_for_each_entry_safe(entry, this, head, list)
1914 		release_discard_addr(entry);
1915 }
1916 
1917 /*
1918  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1919  */
1920 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1921 {
1922 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1923 	unsigned int segno;
1924 
1925 	mutex_lock(&dirty_i->seglist_lock);
1926 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1927 		__set_test_and_free(sbi, segno);
1928 	mutex_unlock(&dirty_i->seglist_lock);
1929 }
1930 
1931 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1932 						struct cp_control *cpc)
1933 {
1934 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1935 	struct list_head *head = &dcc->entry_list;
1936 	struct discard_entry *entry, *this;
1937 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1938 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1939 	unsigned int start = 0, end = -1;
1940 	unsigned int secno, start_segno;
1941 	bool force = (cpc->reason & CP_DISCARD);
1942 	bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
1943 
1944 	mutex_lock(&dirty_i->seglist_lock);
1945 
1946 	while (1) {
1947 		int i;
1948 
1949 		if (need_align && end != -1)
1950 			end--;
1951 		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1952 		if (start >= MAIN_SEGS(sbi))
1953 			break;
1954 		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1955 								start + 1);
1956 
1957 		if (need_align) {
1958 			start = rounddown(start, sbi->segs_per_sec);
1959 			end = roundup(end, sbi->segs_per_sec);
1960 		}
1961 
1962 		for (i = start; i < end; i++) {
1963 			if (test_and_clear_bit(i, prefree_map))
1964 				dirty_i->nr_dirty[PRE]--;
1965 		}
1966 
1967 		if (!f2fs_realtime_discard_enable(sbi))
1968 			continue;
1969 
1970 		if (force && start >= cpc->trim_start &&
1971 					(end - 1) <= cpc->trim_end)
1972 				continue;
1973 
1974 		if (!test_opt(sbi, LFS) || !__is_large_section(sbi)) {
1975 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1976 				(end - start) << sbi->log_blocks_per_seg);
1977 			continue;
1978 		}
1979 next:
1980 		secno = GET_SEC_FROM_SEG(sbi, start);
1981 		start_segno = GET_SEG_FROM_SEC(sbi, secno);
1982 		if (!IS_CURSEC(sbi, secno) &&
1983 			!get_valid_blocks(sbi, start, true))
1984 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1985 				sbi->segs_per_sec << sbi->log_blocks_per_seg);
1986 
1987 		start = start_segno + sbi->segs_per_sec;
1988 		if (start < end)
1989 			goto next;
1990 		else
1991 			end = start - 1;
1992 	}
1993 	mutex_unlock(&dirty_i->seglist_lock);
1994 
1995 	/* send small discards */
1996 	list_for_each_entry_safe(entry, this, head, list) {
1997 		unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1998 		bool is_valid = test_bit_le(0, entry->discard_map);
1999 
2000 find_next:
2001 		if (is_valid) {
2002 			next_pos = find_next_zero_bit_le(entry->discard_map,
2003 					sbi->blocks_per_seg, cur_pos);
2004 			len = next_pos - cur_pos;
2005 
2006 			if (f2fs_sb_has_blkzoned(sbi) ||
2007 			    (force && len < cpc->trim_minlen))
2008 				goto skip;
2009 
2010 			f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2011 									len);
2012 			total_len += len;
2013 		} else {
2014 			next_pos = find_next_bit_le(entry->discard_map,
2015 					sbi->blocks_per_seg, cur_pos);
2016 		}
2017 skip:
2018 		cur_pos = next_pos;
2019 		is_valid = !is_valid;
2020 
2021 		if (cur_pos < sbi->blocks_per_seg)
2022 			goto find_next;
2023 
2024 		release_discard_addr(entry);
2025 		dcc->nr_discards -= total_len;
2026 	}
2027 
2028 	wake_up_discard_thread(sbi, false);
2029 }
2030 
2031 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2032 {
2033 	dev_t dev = sbi->sb->s_bdev->bd_dev;
2034 	struct discard_cmd_control *dcc;
2035 	int err = 0, i;
2036 
2037 	if (SM_I(sbi)->dcc_info) {
2038 		dcc = SM_I(sbi)->dcc_info;
2039 		goto init_thread;
2040 	}
2041 
2042 	dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2043 	if (!dcc)
2044 		return -ENOMEM;
2045 
2046 	dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2047 	INIT_LIST_HEAD(&dcc->entry_list);
2048 	for (i = 0; i < MAX_PLIST_NUM; i++)
2049 		INIT_LIST_HEAD(&dcc->pend_list[i]);
2050 	INIT_LIST_HEAD(&dcc->wait_list);
2051 	INIT_LIST_HEAD(&dcc->fstrim_list);
2052 	mutex_init(&dcc->cmd_lock);
2053 	atomic_set(&dcc->issued_discard, 0);
2054 	atomic_set(&dcc->queued_discard, 0);
2055 	atomic_set(&dcc->discard_cmd_cnt, 0);
2056 	dcc->nr_discards = 0;
2057 	dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2058 	dcc->undiscard_blks = 0;
2059 	dcc->next_pos = 0;
2060 	dcc->root = RB_ROOT_CACHED;
2061 	dcc->rbtree_check = false;
2062 
2063 	init_waitqueue_head(&dcc->discard_wait_queue);
2064 	SM_I(sbi)->dcc_info = dcc;
2065 init_thread:
2066 	dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2067 				"f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2068 	if (IS_ERR(dcc->f2fs_issue_discard)) {
2069 		err = PTR_ERR(dcc->f2fs_issue_discard);
2070 		kvfree(dcc);
2071 		SM_I(sbi)->dcc_info = NULL;
2072 		return err;
2073 	}
2074 
2075 	return err;
2076 }
2077 
2078 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2079 {
2080 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2081 
2082 	if (!dcc)
2083 		return;
2084 
2085 	f2fs_stop_discard_thread(sbi);
2086 
2087 	kvfree(dcc);
2088 	SM_I(sbi)->dcc_info = NULL;
2089 }
2090 
2091 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2092 {
2093 	struct sit_info *sit_i = SIT_I(sbi);
2094 
2095 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2096 		sit_i->dirty_sentries++;
2097 		return false;
2098 	}
2099 
2100 	return true;
2101 }
2102 
2103 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2104 					unsigned int segno, int modified)
2105 {
2106 	struct seg_entry *se = get_seg_entry(sbi, segno);
2107 	se->type = type;
2108 	if (modified)
2109 		__mark_sit_entry_dirty(sbi, segno);
2110 }
2111 
2112 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2113 {
2114 	struct seg_entry *se;
2115 	unsigned int segno, offset;
2116 	long int new_vblocks;
2117 	bool exist;
2118 #ifdef CONFIG_F2FS_CHECK_FS
2119 	bool mir_exist;
2120 #endif
2121 
2122 	segno = GET_SEGNO(sbi, blkaddr);
2123 
2124 	se = get_seg_entry(sbi, segno);
2125 	new_vblocks = se->valid_blocks + del;
2126 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2127 
2128 	f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
2129 				(new_vblocks > sbi->blocks_per_seg)));
2130 
2131 	se->valid_blocks = new_vblocks;
2132 	se->mtime = get_mtime(sbi, false);
2133 	if (se->mtime > SIT_I(sbi)->max_mtime)
2134 		SIT_I(sbi)->max_mtime = se->mtime;
2135 
2136 	/* Update valid block bitmap */
2137 	if (del > 0) {
2138 		exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2139 #ifdef CONFIG_F2FS_CHECK_FS
2140 		mir_exist = f2fs_test_and_set_bit(offset,
2141 						se->cur_valid_map_mir);
2142 		if (unlikely(exist != mir_exist)) {
2143 			f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2144 				 blkaddr, exist);
2145 			f2fs_bug_on(sbi, 1);
2146 		}
2147 #endif
2148 		if (unlikely(exist)) {
2149 			f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2150 				 blkaddr);
2151 			f2fs_bug_on(sbi, 1);
2152 			se->valid_blocks--;
2153 			del = 0;
2154 		}
2155 
2156 		if (!f2fs_test_and_set_bit(offset, se->discard_map))
2157 			sbi->discard_blks--;
2158 
2159 		/* don't overwrite by SSR to keep node chain */
2160 		if (IS_NODESEG(se->type) &&
2161 				!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2162 			if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2163 				se->ckpt_valid_blocks++;
2164 		}
2165 	} else {
2166 		exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2167 #ifdef CONFIG_F2FS_CHECK_FS
2168 		mir_exist = f2fs_test_and_clear_bit(offset,
2169 						se->cur_valid_map_mir);
2170 		if (unlikely(exist != mir_exist)) {
2171 			f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2172 				 blkaddr, exist);
2173 			f2fs_bug_on(sbi, 1);
2174 		}
2175 #endif
2176 		if (unlikely(!exist)) {
2177 			f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2178 				 blkaddr);
2179 			f2fs_bug_on(sbi, 1);
2180 			se->valid_blocks++;
2181 			del = 0;
2182 		} else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2183 			/*
2184 			 * If checkpoints are off, we must not reuse data that
2185 			 * was used in the previous checkpoint. If it was used
2186 			 * before, we must track that to know how much space we
2187 			 * really have.
2188 			 */
2189 			if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2190 				spin_lock(&sbi->stat_lock);
2191 				sbi->unusable_block_count++;
2192 				spin_unlock(&sbi->stat_lock);
2193 			}
2194 		}
2195 
2196 		if (f2fs_test_and_clear_bit(offset, se->discard_map))
2197 			sbi->discard_blks++;
2198 	}
2199 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2200 		se->ckpt_valid_blocks += del;
2201 
2202 	__mark_sit_entry_dirty(sbi, segno);
2203 
2204 	/* update total number of valid blocks to be written in ckpt area */
2205 	SIT_I(sbi)->written_valid_blocks += del;
2206 
2207 	if (__is_large_section(sbi))
2208 		get_sec_entry(sbi, segno)->valid_blocks += del;
2209 }
2210 
2211 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2212 {
2213 	unsigned int segno = GET_SEGNO(sbi, addr);
2214 	struct sit_info *sit_i = SIT_I(sbi);
2215 
2216 	f2fs_bug_on(sbi, addr == NULL_ADDR);
2217 	if (addr == NEW_ADDR)
2218 		return;
2219 
2220 	invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2221 
2222 	/* add it into sit main buffer */
2223 	down_write(&sit_i->sentry_lock);
2224 
2225 	update_sit_entry(sbi, addr, -1);
2226 
2227 	/* add it into dirty seglist */
2228 	locate_dirty_segment(sbi, segno);
2229 
2230 	up_write(&sit_i->sentry_lock);
2231 }
2232 
2233 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2234 {
2235 	struct sit_info *sit_i = SIT_I(sbi);
2236 	unsigned int segno, offset;
2237 	struct seg_entry *se;
2238 	bool is_cp = false;
2239 
2240 	if (!__is_valid_data_blkaddr(blkaddr))
2241 		return true;
2242 
2243 	down_read(&sit_i->sentry_lock);
2244 
2245 	segno = GET_SEGNO(sbi, blkaddr);
2246 	se = get_seg_entry(sbi, segno);
2247 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2248 
2249 	if (f2fs_test_bit(offset, se->ckpt_valid_map))
2250 		is_cp = true;
2251 
2252 	up_read(&sit_i->sentry_lock);
2253 
2254 	return is_cp;
2255 }
2256 
2257 /*
2258  * This function should be resided under the curseg_mutex lock
2259  */
2260 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2261 					struct f2fs_summary *sum)
2262 {
2263 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2264 	void *addr = curseg->sum_blk;
2265 	addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2266 	memcpy(addr, sum, sizeof(struct f2fs_summary));
2267 }
2268 
2269 /*
2270  * Calculate the number of current summary pages for writing
2271  */
2272 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2273 {
2274 	int valid_sum_count = 0;
2275 	int i, sum_in_page;
2276 
2277 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2278 		if (sbi->ckpt->alloc_type[i] == SSR)
2279 			valid_sum_count += sbi->blocks_per_seg;
2280 		else {
2281 			if (for_ra)
2282 				valid_sum_count += le16_to_cpu(
2283 					F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2284 			else
2285 				valid_sum_count += curseg_blkoff(sbi, i);
2286 		}
2287 	}
2288 
2289 	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2290 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2291 	if (valid_sum_count <= sum_in_page)
2292 		return 1;
2293 	else if ((valid_sum_count - sum_in_page) <=
2294 		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2295 		return 2;
2296 	return 3;
2297 }
2298 
2299 /*
2300  * Caller should put this summary page
2301  */
2302 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2303 {
2304 	return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2305 }
2306 
2307 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2308 					void *src, block_t blk_addr)
2309 {
2310 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2311 
2312 	memcpy(page_address(page), src, PAGE_SIZE);
2313 	set_page_dirty(page);
2314 	f2fs_put_page(page, 1);
2315 }
2316 
2317 static void write_sum_page(struct f2fs_sb_info *sbi,
2318 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
2319 {
2320 	f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2321 }
2322 
2323 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2324 						int type, block_t blk_addr)
2325 {
2326 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2327 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2328 	struct f2fs_summary_block *src = curseg->sum_blk;
2329 	struct f2fs_summary_block *dst;
2330 
2331 	dst = (struct f2fs_summary_block *)page_address(page);
2332 	memset(dst, 0, PAGE_SIZE);
2333 
2334 	mutex_lock(&curseg->curseg_mutex);
2335 
2336 	down_read(&curseg->journal_rwsem);
2337 	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2338 	up_read(&curseg->journal_rwsem);
2339 
2340 	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2341 	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2342 
2343 	mutex_unlock(&curseg->curseg_mutex);
2344 
2345 	set_page_dirty(page);
2346 	f2fs_put_page(page, 1);
2347 }
2348 
2349 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2350 {
2351 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2352 	unsigned int segno = curseg->segno + 1;
2353 	struct free_segmap_info *free_i = FREE_I(sbi);
2354 
2355 	if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2356 		return !test_bit(segno, free_i->free_segmap);
2357 	return 0;
2358 }
2359 
2360 /*
2361  * Find a new segment from the free segments bitmap to right order
2362  * This function should be returned with success, otherwise BUG
2363  */
2364 static void get_new_segment(struct f2fs_sb_info *sbi,
2365 			unsigned int *newseg, bool new_sec, int dir)
2366 {
2367 	struct free_segmap_info *free_i = FREE_I(sbi);
2368 	unsigned int segno, secno, zoneno;
2369 	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2370 	unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2371 	unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2372 	unsigned int left_start = hint;
2373 	bool init = true;
2374 	int go_left = 0;
2375 	int i;
2376 
2377 	spin_lock(&free_i->segmap_lock);
2378 
2379 	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2380 		segno = find_next_zero_bit(free_i->free_segmap,
2381 			GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2382 		if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2383 			goto got_it;
2384 	}
2385 find_other_zone:
2386 	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2387 	if (secno >= MAIN_SECS(sbi)) {
2388 		if (dir == ALLOC_RIGHT) {
2389 			secno = find_next_zero_bit(free_i->free_secmap,
2390 							MAIN_SECS(sbi), 0);
2391 			f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2392 		} else {
2393 			go_left = 1;
2394 			left_start = hint - 1;
2395 		}
2396 	}
2397 	if (go_left == 0)
2398 		goto skip_left;
2399 
2400 	while (test_bit(left_start, free_i->free_secmap)) {
2401 		if (left_start > 0) {
2402 			left_start--;
2403 			continue;
2404 		}
2405 		left_start = find_next_zero_bit(free_i->free_secmap,
2406 							MAIN_SECS(sbi), 0);
2407 		f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2408 		break;
2409 	}
2410 	secno = left_start;
2411 skip_left:
2412 	segno = GET_SEG_FROM_SEC(sbi, secno);
2413 	zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2414 
2415 	/* give up on finding another zone */
2416 	if (!init)
2417 		goto got_it;
2418 	if (sbi->secs_per_zone == 1)
2419 		goto got_it;
2420 	if (zoneno == old_zoneno)
2421 		goto got_it;
2422 	if (dir == ALLOC_LEFT) {
2423 		if (!go_left && zoneno + 1 >= total_zones)
2424 			goto got_it;
2425 		if (go_left && zoneno == 0)
2426 			goto got_it;
2427 	}
2428 	for (i = 0; i < NR_CURSEG_TYPE; i++)
2429 		if (CURSEG_I(sbi, i)->zone == zoneno)
2430 			break;
2431 
2432 	if (i < NR_CURSEG_TYPE) {
2433 		/* zone is in user, try another */
2434 		if (go_left)
2435 			hint = zoneno * sbi->secs_per_zone - 1;
2436 		else if (zoneno + 1 >= total_zones)
2437 			hint = 0;
2438 		else
2439 			hint = (zoneno + 1) * sbi->secs_per_zone;
2440 		init = false;
2441 		goto find_other_zone;
2442 	}
2443 got_it:
2444 	/* set it as dirty segment in free segmap */
2445 	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2446 	__set_inuse(sbi, segno);
2447 	*newseg = segno;
2448 	spin_unlock(&free_i->segmap_lock);
2449 }
2450 
2451 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2452 {
2453 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2454 	struct summary_footer *sum_footer;
2455 
2456 	curseg->segno = curseg->next_segno;
2457 	curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2458 	curseg->next_blkoff = 0;
2459 	curseg->next_segno = NULL_SEGNO;
2460 
2461 	sum_footer = &(curseg->sum_blk->footer);
2462 	memset(sum_footer, 0, sizeof(struct summary_footer));
2463 	if (IS_DATASEG(type))
2464 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2465 	if (IS_NODESEG(type))
2466 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2467 	__set_sit_entry_type(sbi, type, curseg->segno, modified);
2468 }
2469 
2470 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2471 {
2472 	/* if segs_per_sec is large than 1, we need to keep original policy. */
2473 	if (__is_large_section(sbi))
2474 		return CURSEG_I(sbi, type)->segno;
2475 
2476 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2477 		return 0;
2478 
2479 	if (test_opt(sbi, NOHEAP) &&
2480 		(type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2481 		return 0;
2482 
2483 	if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2484 		return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2485 
2486 	/* find segments from 0 to reuse freed segments */
2487 	if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2488 		return 0;
2489 
2490 	return CURSEG_I(sbi, type)->segno;
2491 }
2492 
2493 /*
2494  * Allocate a current working segment.
2495  * This function always allocates a free segment in LFS manner.
2496  */
2497 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2498 {
2499 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2500 	unsigned int segno = curseg->segno;
2501 	int dir = ALLOC_LEFT;
2502 
2503 	write_sum_page(sbi, curseg->sum_blk,
2504 				GET_SUM_BLOCK(sbi, segno));
2505 	if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2506 		dir = ALLOC_RIGHT;
2507 
2508 	if (test_opt(sbi, NOHEAP))
2509 		dir = ALLOC_RIGHT;
2510 
2511 	segno = __get_next_segno(sbi, type);
2512 	get_new_segment(sbi, &segno, new_sec, dir);
2513 	curseg->next_segno = segno;
2514 	reset_curseg(sbi, type, 1);
2515 	curseg->alloc_type = LFS;
2516 }
2517 
2518 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2519 			struct curseg_info *seg, block_t start)
2520 {
2521 	struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2522 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2523 	unsigned long *target_map = SIT_I(sbi)->tmp_map;
2524 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2525 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2526 	int i, pos;
2527 
2528 	for (i = 0; i < entries; i++)
2529 		target_map[i] = ckpt_map[i] | cur_map[i];
2530 
2531 	pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2532 
2533 	seg->next_blkoff = pos;
2534 }
2535 
2536 /*
2537  * If a segment is written by LFS manner, next block offset is just obtained
2538  * by increasing the current block offset. However, if a segment is written by
2539  * SSR manner, next block offset obtained by calling __next_free_blkoff
2540  */
2541 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2542 				struct curseg_info *seg)
2543 {
2544 	if (seg->alloc_type == SSR)
2545 		__next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2546 	else
2547 		seg->next_blkoff++;
2548 }
2549 
2550 /*
2551  * This function always allocates a used segment(from dirty seglist) by SSR
2552  * manner, so it should recover the existing segment information of valid blocks
2553  */
2554 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2555 {
2556 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2557 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2558 	unsigned int new_segno = curseg->next_segno;
2559 	struct f2fs_summary_block *sum_node;
2560 	struct page *sum_page;
2561 
2562 	write_sum_page(sbi, curseg->sum_blk,
2563 				GET_SUM_BLOCK(sbi, curseg->segno));
2564 	__set_test_and_inuse(sbi, new_segno);
2565 
2566 	mutex_lock(&dirty_i->seglist_lock);
2567 	__remove_dirty_segment(sbi, new_segno, PRE);
2568 	__remove_dirty_segment(sbi, new_segno, DIRTY);
2569 	mutex_unlock(&dirty_i->seglist_lock);
2570 
2571 	reset_curseg(sbi, type, 1);
2572 	curseg->alloc_type = SSR;
2573 	__next_free_blkoff(sbi, curseg, 0);
2574 
2575 	sum_page = f2fs_get_sum_page(sbi, new_segno);
2576 	f2fs_bug_on(sbi, IS_ERR(sum_page));
2577 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2578 	memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2579 	f2fs_put_page(sum_page, 1);
2580 }
2581 
2582 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2583 {
2584 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2585 	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2586 	unsigned segno = NULL_SEGNO;
2587 	int i, cnt;
2588 	bool reversed = false;
2589 
2590 	/* f2fs_need_SSR() already forces to do this */
2591 	if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2592 		curseg->next_segno = segno;
2593 		return 1;
2594 	}
2595 
2596 	/* For node segments, let's do SSR more intensively */
2597 	if (IS_NODESEG(type)) {
2598 		if (type >= CURSEG_WARM_NODE) {
2599 			reversed = true;
2600 			i = CURSEG_COLD_NODE;
2601 		} else {
2602 			i = CURSEG_HOT_NODE;
2603 		}
2604 		cnt = NR_CURSEG_NODE_TYPE;
2605 	} else {
2606 		if (type >= CURSEG_WARM_DATA) {
2607 			reversed = true;
2608 			i = CURSEG_COLD_DATA;
2609 		} else {
2610 			i = CURSEG_HOT_DATA;
2611 		}
2612 		cnt = NR_CURSEG_DATA_TYPE;
2613 	}
2614 
2615 	for (; cnt-- > 0; reversed ? i-- : i++) {
2616 		if (i == type)
2617 			continue;
2618 		if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2619 			curseg->next_segno = segno;
2620 			return 1;
2621 		}
2622 	}
2623 
2624 	/* find valid_blocks=0 in dirty list */
2625 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2626 		segno = get_free_segment(sbi);
2627 		if (segno != NULL_SEGNO) {
2628 			curseg->next_segno = segno;
2629 			return 1;
2630 		}
2631 	}
2632 	return 0;
2633 }
2634 
2635 /*
2636  * flush out current segment and replace it with new segment
2637  * This function should be returned with success, otherwise BUG
2638  */
2639 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2640 						int type, bool force)
2641 {
2642 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2643 
2644 	if (force)
2645 		new_curseg(sbi, type, true);
2646 	else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2647 					type == CURSEG_WARM_NODE)
2648 		new_curseg(sbi, type, false);
2649 	else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2650 			likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2651 		new_curseg(sbi, type, false);
2652 	else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2653 		change_curseg(sbi, type);
2654 	else
2655 		new_curseg(sbi, type, false);
2656 
2657 	stat_inc_seg_type(sbi, curseg);
2658 }
2659 
2660 void allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2661 					unsigned int start, unsigned int end)
2662 {
2663 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2664 	unsigned int segno;
2665 
2666 	down_read(&SM_I(sbi)->curseg_lock);
2667 	mutex_lock(&curseg->curseg_mutex);
2668 	down_write(&SIT_I(sbi)->sentry_lock);
2669 
2670 	segno = CURSEG_I(sbi, type)->segno;
2671 	if (segno < start || segno > end)
2672 		goto unlock;
2673 
2674 	if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2675 		change_curseg(sbi, type);
2676 	else
2677 		new_curseg(sbi, type, true);
2678 
2679 	stat_inc_seg_type(sbi, curseg);
2680 
2681 	locate_dirty_segment(sbi, segno);
2682 unlock:
2683 	up_write(&SIT_I(sbi)->sentry_lock);
2684 
2685 	if (segno != curseg->segno)
2686 		f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2687 			    type, segno, curseg->segno);
2688 
2689 	mutex_unlock(&curseg->curseg_mutex);
2690 	up_read(&SM_I(sbi)->curseg_lock);
2691 }
2692 
2693 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2694 {
2695 	struct curseg_info *curseg;
2696 	unsigned int old_segno;
2697 	int i;
2698 
2699 	down_write(&SIT_I(sbi)->sentry_lock);
2700 
2701 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2702 		curseg = CURSEG_I(sbi, i);
2703 		old_segno = curseg->segno;
2704 		SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2705 		locate_dirty_segment(sbi, old_segno);
2706 	}
2707 
2708 	up_write(&SIT_I(sbi)->sentry_lock);
2709 }
2710 
2711 static const struct segment_allocation default_salloc_ops = {
2712 	.allocate_segment = allocate_segment_by_default,
2713 };
2714 
2715 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2716 						struct cp_control *cpc)
2717 {
2718 	__u64 trim_start = cpc->trim_start;
2719 	bool has_candidate = false;
2720 
2721 	down_write(&SIT_I(sbi)->sentry_lock);
2722 	for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2723 		if (add_discard_addrs(sbi, cpc, true)) {
2724 			has_candidate = true;
2725 			break;
2726 		}
2727 	}
2728 	up_write(&SIT_I(sbi)->sentry_lock);
2729 
2730 	cpc->trim_start = trim_start;
2731 	return has_candidate;
2732 }
2733 
2734 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2735 					struct discard_policy *dpolicy,
2736 					unsigned int start, unsigned int end)
2737 {
2738 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2739 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2740 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
2741 	struct discard_cmd *dc;
2742 	struct blk_plug plug;
2743 	int issued;
2744 	unsigned int trimmed = 0;
2745 
2746 next:
2747 	issued = 0;
2748 
2749 	mutex_lock(&dcc->cmd_lock);
2750 	if (unlikely(dcc->rbtree_check))
2751 		f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2752 								&dcc->root));
2753 
2754 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2755 					NULL, start,
2756 					(struct rb_entry **)&prev_dc,
2757 					(struct rb_entry **)&next_dc,
2758 					&insert_p, &insert_parent, true, NULL);
2759 	if (!dc)
2760 		dc = next_dc;
2761 
2762 	blk_start_plug(&plug);
2763 
2764 	while (dc && dc->lstart <= end) {
2765 		struct rb_node *node;
2766 		int err = 0;
2767 
2768 		if (dc->len < dpolicy->granularity)
2769 			goto skip;
2770 
2771 		if (dc->state != D_PREP) {
2772 			list_move_tail(&dc->list, &dcc->fstrim_list);
2773 			goto skip;
2774 		}
2775 
2776 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2777 
2778 		if (issued >= dpolicy->max_requests) {
2779 			start = dc->lstart + dc->len;
2780 
2781 			if (err)
2782 				__remove_discard_cmd(sbi, dc);
2783 
2784 			blk_finish_plug(&plug);
2785 			mutex_unlock(&dcc->cmd_lock);
2786 			trimmed += __wait_all_discard_cmd(sbi, NULL);
2787 			congestion_wait(BLK_RW_ASYNC, HZ/50);
2788 			goto next;
2789 		}
2790 skip:
2791 		node = rb_next(&dc->rb_node);
2792 		if (err)
2793 			__remove_discard_cmd(sbi, dc);
2794 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2795 
2796 		if (fatal_signal_pending(current))
2797 			break;
2798 	}
2799 
2800 	blk_finish_plug(&plug);
2801 	mutex_unlock(&dcc->cmd_lock);
2802 
2803 	return trimmed;
2804 }
2805 
2806 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2807 {
2808 	__u64 start = F2FS_BYTES_TO_BLK(range->start);
2809 	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2810 	unsigned int start_segno, end_segno;
2811 	block_t start_block, end_block;
2812 	struct cp_control cpc;
2813 	struct discard_policy dpolicy;
2814 	unsigned long long trimmed = 0;
2815 	int err = 0;
2816 	bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
2817 
2818 	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2819 		return -EINVAL;
2820 
2821 	if (end < MAIN_BLKADDR(sbi))
2822 		goto out;
2823 
2824 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2825 		f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
2826 		return -EFSCORRUPTED;
2827 	}
2828 
2829 	/* start/end segment number in main_area */
2830 	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2831 	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2832 						GET_SEGNO(sbi, end);
2833 	if (need_align) {
2834 		start_segno = rounddown(start_segno, sbi->segs_per_sec);
2835 		end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2836 	}
2837 
2838 	cpc.reason = CP_DISCARD;
2839 	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2840 	cpc.trim_start = start_segno;
2841 	cpc.trim_end = end_segno;
2842 
2843 	if (sbi->discard_blks == 0)
2844 		goto out;
2845 
2846 	mutex_lock(&sbi->gc_mutex);
2847 	err = f2fs_write_checkpoint(sbi, &cpc);
2848 	mutex_unlock(&sbi->gc_mutex);
2849 	if (err)
2850 		goto out;
2851 
2852 	/*
2853 	 * We filed discard candidates, but actually we don't need to wait for
2854 	 * all of them, since they'll be issued in idle time along with runtime
2855 	 * discard option. User configuration looks like using runtime discard
2856 	 * or periodic fstrim instead of it.
2857 	 */
2858 	if (f2fs_realtime_discard_enable(sbi))
2859 		goto out;
2860 
2861 	start_block = START_BLOCK(sbi, start_segno);
2862 	end_block = START_BLOCK(sbi, end_segno + 1);
2863 
2864 	__init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2865 	trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2866 					start_block, end_block);
2867 
2868 	trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2869 					start_block, end_block);
2870 out:
2871 	if (!err)
2872 		range->len = F2FS_BLK_TO_BYTES(trimmed);
2873 	return err;
2874 }
2875 
2876 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2877 {
2878 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2879 	if (curseg->next_blkoff < sbi->blocks_per_seg)
2880 		return true;
2881 	return false;
2882 }
2883 
2884 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2885 {
2886 	switch (hint) {
2887 	case WRITE_LIFE_SHORT:
2888 		return CURSEG_HOT_DATA;
2889 	case WRITE_LIFE_EXTREME:
2890 		return CURSEG_COLD_DATA;
2891 	default:
2892 		return CURSEG_WARM_DATA;
2893 	}
2894 }
2895 
2896 /* This returns write hints for each segment type. This hints will be
2897  * passed down to block layer. There are mapping tables which depend on
2898  * the mount option 'whint_mode'.
2899  *
2900  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2901  *
2902  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2903  *
2904  * User                  F2FS                     Block
2905  * ----                  ----                     -----
2906  *                       META                     WRITE_LIFE_NOT_SET
2907  *                       HOT_NODE                 "
2908  *                       WARM_NODE                "
2909  *                       COLD_NODE                "
2910  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2911  * extension list        "                        "
2912  *
2913  * -- buffered io
2914  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2915  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2916  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2917  * WRITE_LIFE_NONE       "                        "
2918  * WRITE_LIFE_MEDIUM     "                        "
2919  * WRITE_LIFE_LONG       "                        "
2920  *
2921  * -- direct io
2922  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2923  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2924  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2925  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2926  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2927  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2928  *
2929  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2930  *
2931  * User                  F2FS                     Block
2932  * ----                  ----                     -----
2933  *                       META                     WRITE_LIFE_MEDIUM;
2934  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2935  *                       WARM_NODE                "
2936  *                       COLD_NODE                WRITE_LIFE_NONE
2937  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2938  * extension list        "                        "
2939  *
2940  * -- buffered io
2941  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2942  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2943  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2944  * WRITE_LIFE_NONE       "                        "
2945  * WRITE_LIFE_MEDIUM     "                        "
2946  * WRITE_LIFE_LONG       "                        "
2947  *
2948  * -- direct io
2949  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2950  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2951  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2952  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2953  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2954  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2955  */
2956 
2957 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2958 				enum page_type type, enum temp_type temp)
2959 {
2960 	if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2961 		if (type == DATA) {
2962 			if (temp == WARM)
2963 				return WRITE_LIFE_NOT_SET;
2964 			else if (temp == HOT)
2965 				return WRITE_LIFE_SHORT;
2966 			else if (temp == COLD)
2967 				return WRITE_LIFE_EXTREME;
2968 		} else {
2969 			return WRITE_LIFE_NOT_SET;
2970 		}
2971 	} else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2972 		if (type == DATA) {
2973 			if (temp == WARM)
2974 				return WRITE_LIFE_LONG;
2975 			else if (temp == HOT)
2976 				return WRITE_LIFE_SHORT;
2977 			else if (temp == COLD)
2978 				return WRITE_LIFE_EXTREME;
2979 		} else if (type == NODE) {
2980 			if (temp == WARM || temp == HOT)
2981 				return WRITE_LIFE_NOT_SET;
2982 			else if (temp == COLD)
2983 				return WRITE_LIFE_NONE;
2984 		} else if (type == META) {
2985 			return WRITE_LIFE_MEDIUM;
2986 		}
2987 	}
2988 	return WRITE_LIFE_NOT_SET;
2989 }
2990 
2991 static int __get_segment_type_2(struct f2fs_io_info *fio)
2992 {
2993 	if (fio->type == DATA)
2994 		return CURSEG_HOT_DATA;
2995 	else
2996 		return CURSEG_HOT_NODE;
2997 }
2998 
2999 static int __get_segment_type_4(struct f2fs_io_info *fio)
3000 {
3001 	if (fio->type == DATA) {
3002 		struct inode *inode = fio->page->mapping->host;
3003 
3004 		if (S_ISDIR(inode->i_mode))
3005 			return CURSEG_HOT_DATA;
3006 		else
3007 			return CURSEG_COLD_DATA;
3008 	} else {
3009 		if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3010 			return CURSEG_WARM_NODE;
3011 		else
3012 			return CURSEG_COLD_NODE;
3013 	}
3014 }
3015 
3016 static int __get_segment_type_6(struct f2fs_io_info *fio)
3017 {
3018 	if (fio->type == DATA) {
3019 		struct inode *inode = fio->page->mapping->host;
3020 
3021 		if (is_cold_data(fio->page) || file_is_cold(inode))
3022 			return CURSEG_COLD_DATA;
3023 		if (file_is_hot(inode) ||
3024 				is_inode_flag_set(inode, FI_HOT_DATA) ||
3025 				f2fs_is_atomic_file(inode) ||
3026 				f2fs_is_volatile_file(inode))
3027 			return CURSEG_HOT_DATA;
3028 		return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3029 	} else {
3030 		if (IS_DNODE(fio->page))
3031 			return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3032 						CURSEG_HOT_NODE;
3033 		return CURSEG_COLD_NODE;
3034 	}
3035 }
3036 
3037 static int __get_segment_type(struct f2fs_io_info *fio)
3038 {
3039 	int type = 0;
3040 
3041 	switch (F2FS_OPTION(fio->sbi).active_logs) {
3042 	case 2:
3043 		type = __get_segment_type_2(fio);
3044 		break;
3045 	case 4:
3046 		type = __get_segment_type_4(fio);
3047 		break;
3048 	case 6:
3049 		type = __get_segment_type_6(fio);
3050 		break;
3051 	default:
3052 		f2fs_bug_on(fio->sbi, true);
3053 	}
3054 
3055 	if (IS_HOT(type))
3056 		fio->temp = HOT;
3057 	else if (IS_WARM(type))
3058 		fio->temp = WARM;
3059 	else
3060 		fio->temp = COLD;
3061 	return type;
3062 }
3063 
3064 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3065 		block_t old_blkaddr, block_t *new_blkaddr,
3066 		struct f2fs_summary *sum, int type,
3067 		struct f2fs_io_info *fio, bool add_list)
3068 {
3069 	struct sit_info *sit_i = SIT_I(sbi);
3070 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3071 
3072 	down_read(&SM_I(sbi)->curseg_lock);
3073 
3074 	mutex_lock(&curseg->curseg_mutex);
3075 	down_write(&sit_i->sentry_lock);
3076 
3077 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3078 
3079 	f2fs_wait_discard_bio(sbi, *new_blkaddr);
3080 
3081 	/*
3082 	 * __add_sum_entry should be resided under the curseg_mutex
3083 	 * because, this function updates a summary entry in the
3084 	 * current summary block.
3085 	 */
3086 	__add_sum_entry(sbi, type, sum);
3087 
3088 	__refresh_next_blkoff(sbi, curseg);
3089 
3090 	stat_inc_block_count(sbi, curseg);
3091 
3092 	/*
3093 	 * SIT information should be updated before segment allocation,
3094 	 * since SSR needs latest valid block information.
3095 	 */
3096 	update_sit_entry(sbi, *new_blkaddr, 1);
3097 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3098 		update_sit_entry(sbi, old_blkaddr, -1);
3099 
3100 	if (!__has_curseg_space(sbi, type))
3101 		sit_i->s_ops->allocate_segment(sbi, type, false);
3102 
3103 	/*
3104 	 * segment dirty status should be updated after segment allocation,
3105 	 * so we just need to update status only one time after previous
3106 	 * segment being closed.
3107 	 */
3108 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3109 	locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3110 
3111 	up_write(&sit_i->sentry_lock);
3112 
3113 	if (page && IS_NODESEG(type)) {
3114 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3115 
3116 		f2fs_inode_chksum_set(sbi, page);
3117 	}
3118 
3119 	if (add_list) {
3120 		struct f2fs_bio_info *io;
3121 
3122 		INIT_LIST_HEAD(&fio->list);
3123 		fio->in_list = true;
3124 		fio->retry = false;
3125 		io = sbi->write_io[fio->type] + fio->temp;
3126 		spin_lock(&io->io_lock);
3127 		list_add_tail(&fio->list, &io->io_list);
3128 		spin_unlock(&io->io_lock);
3129 	}
3130 
3131 	mutex_unlock(&curseg->curseg_mutex);
3132 
3133 	up_read(&SM_I(sbi)->curseg_lock);
3134 }
3135 
3136 static void update_device_state(struct f2fs_io_info *fio)
3137 {
3138 	struct f2fs_sb_info *sbi = fio->sbi;
3139 	unsigned int devidx;
3140 
3141 	if (!f2fs_is_multi_device(sbi))
3142 		return;
3143 
3144 	devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3145 
3146 	/* update device state for fsync */
3147 	f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3148 
3149 	/* update device state for checkpoint */
3150 	if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3151 		spin_lock(&sbi->dev_lock);
3152 		f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3153 		spin_unlock(&sbi->dev_lock);
3154 	}
3155 }
3156 
3157 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3158 {
3159 	int type = __get_segment_type(fio);
3160 	bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
3161 
3162 	if (keep_order)
3163 		down_read(&fio->sbi->io_order_lock);
3164 reallocate:
3165 	f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3166 			&fio->new_blkaddr, sum, type, fio, true);
3167 	if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3168 		invalidate_mapping_pages(META_MAPPING(fio->sbi),
3169 					fio->old_blkaddr, fio->old_blkaddr);
3170 
3171 	/* writeout dirty page into bdev */
3172 	f2fs_submit_page_write(fio);
3173 	if (fio->retry) {
3174 		fio->old_blkaddr = fio->new_blkaddr;
3175 		goto reallocate;
3176 	}
3177 
3178 	update_device_state(fio);
3179 
3180 	if (keep_order)
3181 		up_read(&fio->sbi->io_order_lock);
3182 }
3183 
3184 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3185 					enum iostat_type io_type)
3186 {
3187 	struct f2fs_io_info fio = {
3188 		.sbi = sbi,
3189 		.type = META,
3190 		.temp = HOT,
3191 		.op = REQ_OP_WRITE,
3192 		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3193 		.old_blkaddr = page->index,
3194 		.new_blkaddr = page->index,
3195 		.page = page,
3196 		.encrypted_page = NULL,
3197 		.in_list = false,
3198 	};
3199 
3200 	if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3201 		fio.op_flags &= ~REQ_META;
3202 
3203 	set_page_writeback(page);
3204 	ClearPageError(page);
3205 	f2fs_submit_page_write(&fio);
3206 
3207 	stat_inc_meta_count(sbi, page->index);
3208 	f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3209 }
3210 
3211 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3212 {
3213 	struct f2fs_summary sum;
3214 
3215 	set_summary(&sum, nid, 0, 0);
3216 	do_write_page(&sum, fio);
3217 
3218 	f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3219 }
3220 
3221 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3222 					struct f2fs_io_info *fio)
3223 {
3224 	struct f2fs_sb_info *sbi = fio->sbi;
3225 	struct f2fs_summary sum;
3226 
3227 	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3228 	set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3229 	do_write_page(&sum, fio);
3230 	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3231 
3232 	f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3233 }
3234 
3235 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3236 {
3237 	int err;
3238 	struct f2fs_sb_info *sbi = fio->sbi;
3239 	unsigned int segno;
3240 
3241 	fio->new_blkaddr = fio->old_blkaddr;
3242 	/* i/o temperature is needed for passing down write hints */
3243 	__get_segment_type(fio);
3244 
3245 	segno = GET_SEGNO(sbi, fio->new_blkaddr);
3246 
3247 	if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3248 		set_sbi_flag(sbi, SBI_NEED_FSCK);
3249 		f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3250 			  __func__, segno);
3251 		return -EFSCORRUPTED;
3252 	}
3253 
3254 	stat_inc_inplace_blocks(fio->sbi);
3255 
3256 	if (fio->bio)
3257 		err = f2fs_merge_page_bio(fio);
3258 	else
3259 		err = f2fs_submit_page_bio(fio);
3260 	if (!err) {
3261 		update_device_state(fio);
3262 		f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3263 	}
3264 
3265 	return err;
3266 }
3267 
3268 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3269 						unsigned int segno)
3270 {
3271 	int i;
3272 
3273 	for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3274 		if (CURSEG_I(sbi, i)->segno == segno)
3275 			break;
3276 	}
3277 	return i;
3278 }
3279 
3280 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3281 				block_t old_blkaddr, block_t new_blkaddr,
3282 				bool recover_curseg, bool recover_newaddr)
3283 {
3284 	struct sit_info *sit_i = SIT_I(sbi);
3285 	struct curseg_info *curseg;
3286 	unsigned int segno, old_cursegno;
3287 	struct seg_entry *se;
3288 	int type;
3289 	unsigned short old_blkoff;
3290 
3291 	segno = GET_SEGNO(sbi, new_blkaddr);
3292 	se = get_seg_entry(sbi, segno);
3293 	type = se->type;
3294 
3295 	down_write(&SM_I(sbi)->curseg_lock);
3296 
3297 	if (!recover_curseg) {
3298 		/* for recovery flow */
3299 		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3300 			if (old_blkaddr == NULL_ADDR)
3301 				type = CURSEG_COLD_DATA;
3302 			else
3303 				type = CURSEG_WARM_DATA;
3304 		}
3305 	} else {
3306 		if (IS_CURSEG(sbi, segno)) {
3307 			/* se->type is volatile as SSR allocation */
3308 			type = __f2fs_get_curseg(sbi, segno);
3309 			f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3310 		} else {
3311 			type = CURSEG_WARM_DATA;
3312 		}
3313 	}
3314 
3315 	f2fs_bug_on(sbi, !IS_DATASEG(type));
3316 	curseg = CURSEG_I(sbi, type);
3317 
3318 	mutex_lock(&curseg->curseg_mutex);
3319 	down_write(&sit_i->sentry_lock);
3320 
3321 	old_cursegno = curseg->segno;
3322 	old_blkoff = curseg->next_blkoff;
3323 
3324 	/* change the current segment */
3325 	if (segno != curseg->segno) {
3326 		curseg->next_segno = segno;
3327 		change_curseg(sbi, type);
3328 	}
3329 
3330 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3331 	__add_sum_entry(sbi, type, sum);
3332 
3333 	if (!recover_curseg || recover_newaddr)
3334 		update_sit_entry(sbi, new_blkaddr, 1);
3335 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3336 		invalidate_mapping_pages(META_MAPPING(sbi),
3337 					old_blkaddr, old_blkaddr);
3338 		update_sit_entry(sbi, old_blkaddr, -1);
3339 	}
3340 
3341 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3342 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3343 
3344 	locate_dirty_segment(sbi, old_cursegno);
3345 
3346 	if (recover_curseg) {
3347 		if (old_cursegno != curseg->segno) {
3348 			curseg->next_segno = old_cursegno;
3349 			change_curseg(sbi, type);
3350 		}
3351 		curseg->next_blkoff = old_blkoff;
3352 	}
3353 
3354 	up_write(&sit_i->sentry_lock);
3355 	mutex_unlock(&curseg->curseg_mutex);
3356 	up_write(&SM_I(sbi)->curseg_lock);
3357 }
3358 
3359 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3360 				block_t old_addr, block_t new_addr,
3361 				unsigned char version, bool recover_curseg,
3362 				bool recover_newaddr)
3363 {
3364 	struct f2fs_summary sum;
3365 
3366 	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3367 
3368 	f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3369 					recover_curseg, recover_newaddr);
3370 
3371 	f2fs_update_data_blkaddr(dn, new_addr);
3372 }
3373 
3374 void f2fs_wait_on_page_writeback(struct page *page,
3375 				enum page_type type, bool ordered, bool locked)
3376 {
3377 	if (PageWriteback(page)) {
3378 		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3379 
3380 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3381 		if (ordered) {
3382 			wait_on_page_writeback(page);
3383 			f2fs_bug_on(sbi, locked && PageWriteback(page));
3384 		} else {
3385 			wait_for_stable_page(page);
3386 		}
3387 	}
3388 }
3389 
3390 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3391 {
3392 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3393 	struct page *cpage;
3394 
3395 	if (!f2fs_post_read_required(inode))
3396 		return;
3397 
3398 	if (!__is_valid_data_blkaddr(blkaddr))
3399 		return;
3400 
3401 	cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3402 	if (cpage) {
3403 		f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3404 		f2fs_put_page(cpage, 1);
3405 	}
3406 }
3407 
3408 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3409 								block_t len)
3410 {
3411 	block_t i;
3412 
3413 	for (i = 0; i < len; i++)
3414 		f2fs_wait_on_block_writeback(inode, blkaddr + i);
3415 }
3416 
3417 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3418 {
3419 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3420 	struct curseg_info *seg_i;
3421 	unsigned char *kaddr;
3422 	struct page *page;
3423 	block_t start;
3424 	int i, j, offset;
3425 
3426 	start = start_sum_block(sbi);
3427 
3428 	page = f2fs_get_meta_page(sbi, start++);
3429 	if (IS_ERR(page))
3430 		return PTR_ERR(page);
3431 	kaddr = (unsigned char *)page_address(page);
3432 
3433 	/* Step 1: restore nat cache */
3434 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3435 	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3436 
3437 	/* Step 2: restore sit cache */
3438 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3439 	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3440 	offset = 2 * SUM_JOURNAL_SIZE;
3441 
3442 	/* Step 3: restore summary entries */
3443 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3444 		unsigned short blk_off;
3445 		unsigned int segno;
3446 
3447 		seg_i = CURSEG_I(sbi, i);
3448 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3449 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3450 		if (blk_off > ENTRIES_IN_SUM) {
3451 			f2fs_bug_on(sbi, 1);
3452 			f2fs_put_page(page, 1);
3453 			return -EFAULT;
3454 		}
3455 		seg_i->next_segno = segno;
3456 		reset_curseg(sbi, i, 0);
3457 		seg_i->alloc_type = ckpt->alloc_type[i];
3458 		seg_i->next_blkoff = blk_off;
3459 
3460 		if (seg_i->alloc_type == SSR)
3461 			blk_off = sbi->blocks_per_seg;
3462 
3463 		for (j = 0; j < blk_off; j++) {
3464 			struct f2fs_summary *s;
3465 			s = (struct f2fs_summary *)(kaddr + offset);
3466 			seg_i->sum_blk->entries[j] = *s;
3467 			offset += SUMMARY_SIZE;
3468 			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3469 						SUM_FOOTER_SIZE)
3470 				continue;
3471 
3472 			f2fs_put_page(page, 1);
3473 			page = NULL;
3474 
3475 			page = f2fs_get_meta_page(sbi, start++);
3476 			if (IS_ERR(page))
3477 				return PTR_ERR(page);
3478 			kaddr = (unsigned char *)page_address(page);
3479 			offset = 0;
3480 		}
3481 	}
3482 	f2fs_put_page(page, 1);
3483 	return 0;
3484 }
3485 
3486 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3487 {
3488 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3489 	struct f2fs_summary_block *sum;
3490 	struct curseg_info *curseg;
3491 	struct page *new;
3492 	unsigned short blk_off;
3493 	unsigned int segno = 0;
3494 	block_t blk_addr = 0;
3495 	int err = 0;
3496 
3497 	/* get segment number and block addr */
3498 	if (IS_DATASEG(type)) {
3499 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3500 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3501 							CURSEG_HOT_DATA]);
3502 		if (__exist_node_summaries(sbi))
3503 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3504 		else
3505 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3506 	} else {
3507 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
3508 							CURSEG_HOT_NODE]);
3509 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3510 							CURSEG_HOT_NODE]);
3511 		if (__exist_node_summaries(sbi))
3512 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3513 							type - CURSEG_HOT_NODE);
3514 		else
3515 			blk_addr = GET_SUM_BLOCK(sbi, segno);
3516 	}
3517 
3518 	new = f2fs_get_meta_page(sbi, blk_addr);
3519 	if (IS_ERR(new))
3520 		return PTR_ERR(new);
3521 	sum = (struct f2fs_summary_block *)page_address(new);
3522 
3523 	if (IS_NODESEG(type)) {
3524 		if (__exist_node_summaries(sbi)) {
3525 			struct f2fs_summary *ns = &sum->entries[0];
3526 			int i;
3527 			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3528 				ns->version = 0;
3529 				ns->ofs_in_node = 0;
3530 			}
3531 		} else {
3532 			err = f2fs_restore_node_summary(sbi, segno, sum);
3533 			if (err)
3534 				goto out;
3535 		}
3536 	}
3537 
3538 	/* set uncompleted segment to curseg */
3539 	curseg = CURSEG_I(sbi, type);
3540 	mutex_lock(&curseg->curseg_mutex);
3541 
3542 	/* update journal info */
3543 	down_write(&curseg->journal_rwsem);
3544 	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3545 	up_write(&curseg->journal_rwsem);
3546 
3547 	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3548 	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3549 	curseg->next_segno = segno;
3550 	reset_curseg(sbi, type, 0);
3551 	curseg->alloc_type = ckpt->alloc_type[type];
3552 	curseg->next_blkoff = blk_off;
3553 	mutex_unlock(&curseg->curseg_mutex);
3554 out:
3555 	f2fs_put_page(new, 1);
3556 	return err;
3557 }
3558 
3559 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3560 {
3561 	struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3562 	struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3563 	int type = CURSEG_HOT_DATA;
3564 	int err;
3565 
3566 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3567 		int npages = f2fs_npages_for_summary_flush(sbi, true);
3568 
3569 		if (npages >= 2)
3570 			f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3571 							META_CP, true);
3572 
3573 		/* restore for compacted data summary */
3574 		err = read_compacted_summaries(sbi);
3575 		if (err)
3576 			return err;
3577 		type = CURSEG_HOT_NODE;
3578 	}
3579 
3580 	if (__exist_node_summaries(sbi))
3581 		f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3582 					NR_CURSEG_TYPE - type, META_CP, true);
3583 
3584 	for (; type <= CURSEG_COLD_NODE; type++) {
3585 		err = read_normal_summaries(sbi, type);
3586 		if (err)
3587 			return err;
3588 	}
3589 
3590 	/* sanity check for summary blocks */
3591 	if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3592 			sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3593 		f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3594 			 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3595 		return -EINVAL;
3596 	}
3597 
3598 	return 0;
3599 }
3600 
3601 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3602 {
3603 	struct page *page;
3604 	unsigned char *kaddr;
3605 	struct f2fs_summary *summary;
3606 	struct curseg_info *seg_i;
3607 	int written_size = 0;
3608 	int i, j;
3609 
3610 	page = f2fs_grab_meta_page(sbi, blkaddr++);
3611 	kaddr = (unsigned char *)page_address(page);
3612 	memset(kaddr, 0, PAGE_SIZE);
3613 
3614 	/* Step 1: write nat cache */
3615 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3616 	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3617 	written_size += SUM_JOURNAL_SIZE;
3618 
3619 	/* Step 2: write sit cache */
3620 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3621 	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3622 	written_size += SUM_JOURNAL_SIZE;
3623 
3624 	/* Step 3: write summary entries */
3625 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3626 		unsigned short blkoff;
3627 		seg_i = CURSEG_I(sbi, i);
3628 		if (sbi->ckpt->alloc_type[i] == SSR)
3629 			blkoff = sbi->blocks_per_seg;
3630 		else
3631 			blkoff = curseg_blkoff(sbi, i);
3632 
3633 		for (j = 0; j < blkoff; j++) {
3634 			if (!page) {
3635 				page = f2fs_grab_meta_page(sbi, blkaddr++);
3636 				kaddr = (unsigned char *)page_address(page);
3637 				memset(kaddr, 0, PAGE_SIZE);
3638 				written_size = 0;
3639 			}
3640 			summary = (struct f2fs_summary *)(kaddr + written_size);
3641 			*summary = seg_i->sum_blk->entries[j];
3642 			written_size += SUMMARY_SIZE;
3643 
3644 			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3645 							SUM_FOOTER_SIZE)
3646 				continue;
3647 
3648 			set_page_dirty(page);
3649 			f2fs_put_page(page, 1);
3650 			page = NULL;
3651 		}
3652 	}
3653 	if (page) {
3654 		set_page_dirty(page);
3655 		f2fs_put_page(page, 1);
3656 	}
3657 }
3658 
3659 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3660 					block_t blkaddr, int type)
3661 {
3662 	int i, end;
3663 	if (IS_DATASEG(type))
3664 		end = type + NR_CURSEG_DATA_TYPE;
3665 	else
3666 		end = type + NR_CURSEG_NODE_TYPE;
3667 
3668 	for (i = type; i < end; i++)
3669 		write_current_sum_page(sbi, i, blkaddr + (i - type));
3670 }
3671 
3672 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3673 {
3674 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3675 		write_compacted_summaries(sbi, start_blk);
3676 	else
3677 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3678 }
3679 
3680 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3681 {
3682 	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3683 }
3684 
3685 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3686 					unsigned int val, int alloc)
3687 {
3688 	int i;
3689 
3690 	if (type == NAT_JOURNAL) {
3691 		for (i = 0; i < nats_in_cursum(journal); i++) {
3692 			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3693 				return i;
3694 		}
3695 		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3696 			return update_nats_in_cursum(journal, 1);
3697 	} else if (type == SIT_JOURNAL) {
3698 		for (i = 0; i < sits_in_cursum(journal); i++)
3699 			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3700 				return i;
3701 		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3702 			return update_sits_in_cursum(journal, 1);
3703 	}
3704 	return -1;
3705 }
3706 
3707 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3708 					unsigned int segno)
3709 {
3710 	return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3711 }
3712 
3713 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3714 					unsigned int start)
3715 {
3716 	struct sit_info *sit_i = SIT_I(sbi);
3717 	struct page *page;
3718 	pgoff_t src_off, dst_off;
3719 
3720 	src_off = current_sit_addr(sbi, start);
3721 	dst_off = next_sit_addr(sbi, src_off);
3722 
3723 	page = f2fs_grab_meta_page(sbi, dst_off);
3724 	seg_info_to_sit_page(sbi, page, start);
3725 
3726 	set_page_dirty(page);
3727 	set_to_next_sit(sit_i, start);
3728 
3729 	return page;
3730 }
3731 
3732 static struct sit_entry_set *grab_sit_entry_set(void)
3733 {
3734 	struct sit_entry_set *ses =
3735 			f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3736 
3737 	ses->entry_cnt = 0;
3738 	INIT_LIST_HEAD(&ses->set_list);
3739 	return ses;
3740 }
3741 
3742 static void release_sit_entry_set(struct sit_entry_set *ses)
3743 {
3744 	list_del(&ses->set_list);
3745 	kmem_cache_free(sit_entry_set_slab, ses);
3746 }
3747 
3748 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3749 						struct list_head *head)
3750 {
3751 	struct sit_entry_set *next = ses;
3752 
3753 	if (list_is_last(&ses->set_list, head))
3754 		return;
3755 
3756 	list_for_each_entry_continue(next, head, set_list)
3757 		if (ses->entry_cnt <= next->entry_cnt)
3758 			break;
3759 
3760 	list_move_tail(&ses->set_list, &next->set_list);
3761 }
3762 
3763 static void add_sit_entry(unsigned int segno, struct list_head *head)
3764 {
3765 	struct sit_entry_set *ses;
3766 	unsigned int start_segno = START_SEGNO(segno);
3767 
3768 	list_for_each_entry(ses, head, set_list) {
3769 		if (ses->start_segno == start_segno) {
3770 			ses->entry_cnt++;
3771 			adjust_sit_entry_set(ses, head);
3772 			return;
3773 		}
3774 	}
3775 
3776 	ses = grab_sit_entry_set();
3777 
3778 	ses->start_segno = start_segno;
3779 	ses->entry_cnt++;
3780 	list_add(&ses->set_list, head);
3781 }
3782 
3783 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3784 {
3785 	struct f2fs_sm_info *sm_info = SM_I(sbi);
3786 	struct list_head *set_list = &sm_info->sit_entry_set;
3787 	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3788 	unsigned int segno;
3789 
3790 	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3791 		add_sit_entry(segno, set_list);
3792 }
3793 
3794 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3795 {
3796 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3797 	struct f2fs_journal *journal = curseg->journal;
3798 	int i;
3799 
3800 	down_write(&curseg->journal_rwsem);
3801 	for (i = 0; i < sits_in_cursum(journal); i++) {
3802 		unsigned int segno;
3803 		bool dirtied;
3804 
3805 		segno = le32_to_cpu(segno_in_journal(journal, i));
3806 		dirtied = __mark_sit_entry_dirty(sbi, segno);
3807 
3808 		if (!dirtied)
3809 			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3810 	}
3811 	update_sits_in_cursum(journal, -i);
3812 	up_write(&curseg->journal_rwsem);
3813 }
3814 
3815 /*
3816  * CP calls this function, which flushes SIT entries including sit_journal,
3817  * and moves prefree segs to free segs.
3818  */
3819 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3820 {
3821 	struct sit_info *sit_i = SIT_I(sbi);
3822 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3823 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3824 	struct f2fs_journal *journal = curseg->journal;
3825 	struct sit_entry_set *ses, *tmp;
3826 	struct list_head *head = &SM_I(sbi)->sit_entry_set;
3827 	bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
3828 	struct seg_entry *se;
3829 
3830 	down_write(&sit_i->sentry_lock);
3831 
3832 	if (!sit_i->dirty_sentries)
3833 		goto out;
3834 
3835 	/*
3836 	 * add and account sit entries of dirty bitmap in sit entry
3837 	 * set temporarily
3838 	 */
3839 	add_sits_in_set(sbi);
3840 
3841 	/*
3842 	 * if there are no enough space in journal to store dirty sit
3843 	 * entries, remove all entries from journal and add and account
3844 	 * them in sit entry set.
3845 	 */
3846 	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
3847 								!to_journal)
3848 		remove_sits_in_journal(sbi);
3849 
3850 	/*
3851 	 * there are two steps to flush sit entries:
3852 	 * #1, flush sit entries to journal in current cold data summary block.
3853 	 * #2, flush sit entries to sit page.
3854 	 */
3855 	list_for_each_entry_safe(ses, tmp, head, set_list) {
3856 		struct page *page = NULL;
3857 		struct f2fs_sit_block *raw_sit = NULL;
3858 		unsigned int start_segno = ses->start_segno;
3859 		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3860 						(unsigned long)MAIN_SEGS(sbi));
3861 		unsigned int segno = start_segno;
3862 
3863 		if (to_journal &&
3864 			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3865 			to_journal = false;
3866 
3867 		if (to_journal) {
3868 			down_write(&curseg->journal_rwsem);
3869 		} else {
3870 			page = get_next_sit_page(sbi, start_segno);
3871 			raw_sit = page_address(page);
3872 		}
3873 
3874 		/* flush dirty sit entries in region of current sit set */
3875 		for_each_set_bit_from(segno, bitmap, end) {
3876 			int offset, sit_offset;
3877 
3878 			se = get_seg_entry(sbi, segno);
3879 #ifdef CONFIG_F2FS_CHECK_FS
3880 			if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3881 						SIT_VBLOCK_MAP_SIZE))
3882 				f2fs_bug_on(sbi, 1);
3883 #endif
3884 
3885 			/* add discard candidates */
3886 			if (!(cpc->reason & CP_DISCARD)) {
3887 				cpc->trim_start = segno;
3888 				add_discard_addrs(sbi, cpc, false);
3889 			}
3890 
3891 			if (to_journal) {
3892 				offset = f2fs_lookup_journal_in_cursum(journal,
3893 							SIT_JOURNAL, segno, 1);
3894 				f2fs_bug_on(sbi, offset < 0);
3895 				segno_in_journal(journal, offset) =
3896 							cpu_to_le32(segno);
3897 				seg_info_to_raw_sit(se,
3898 					&sit_in_journal(journal, offset));
3899 				check_block_count(sbi, segno,
3900 					&sit_in_journal(journal, offset));
3901 			} else {
3902 				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3903 				seg_info_to_raw_sit(se,
3904 						&raw_sit->entries[sit_offset]);
3905 				check_block_count(sbi, segno,
3906 						&raw_sit->entries[sit_offset]);
3907 			}
3908 
3909 			__clear_bit(segno, bitmap);
3910 			sit_i->dirty_sentries--;
3911 			ses->entry_cnt--;
3912 		}
3913 
3914 		if (to_journal)
3915 			up_write(&curseg->journal_rwsem);
3916 		else
3917 			f2fs_put_page(page, 1);
3918 
3919 		f2fs_bug_on(sbi, ses->entry_cnt);
3920 		release_sit_entry_set(ses);
3921 	}
3922 
3923 	f2fs_bug_on(sbi, !list_empty(head));
3924 	f2fs_bug_on(sbi, sit_i->dirty_sentries);
3925 out:
3926 	if (cpc->reason & CP_DISCARD) {
3927 		__u64 trim_start = cpc->trim_start;
3928 
3929 		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3930 			add_discard_addrs(sbi, cpc, false);
3931 
3932 		cpc->trim_start = trim_start;
3933 	}
3934 	up_write(&sit_i->sentry_lock);
3935 
3936 	set_prefree_as_free_segments(sbi);
3937 }
3938 
3939 static int build_sit_info(struct f2fs_sb_info *sbi)
3940 {
3941 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3942 	struct sit_info *sit_i;
3943 	unsigned int sit_segs, start;
3944 	char *src_bitmap;
3945 	unsigned int bitmap_size;
3946 
3947 	/* allocate memory for SIT information */
3948 	sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3949 	if (!sit_i)
3950 		return -ENOMEM;
3951 
3952 	SM_I(sbi)->sit_info = sit_i;
3953 
3954 	sit_i->sentries =
3955 		f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3956 					      MAIN_SEGS(sbi)),
3957 			      GFP_KERNEL);
3958 	if (!sit_i->sentries)
3959 		return -ENOMEM;
3960 
3961 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3962 	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3963 								GFP_KERNEL);
3964 	if (!sit_i->dirty_sentries_bitmap)
3965 		return -ENOMEM;
3966 
3967 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
3968 		sit_i->sentries[start].cur_valid_map
3969 			= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3970 		sit_i->sentries[start].ckpt_valid_map
3971 			= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3972 		if (!sit_i->sentries[start].cur_valid_map ||
3973 				!sit_i->sentries[start].ckpt_valid_map)
3974 			return -ENOMEM;
3975 
3976 #ifdef CONFIG_F2FS_CHECK_FS
3977 		sit_i->sentries[start].cur_valid_map_mir
3978 			= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3979 		if (!sit_i->sentries[start].cur_valid_map_mir)
3980 			return -ENOMEM;
3981 #endif
3982 
3983 		sit_i->sentries[start].discard_map
3984 			= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3985 							GFP_KERNEL);
3986 		if (!sit_i->sentries[start].discard_map)
3987 			return -ENOMEM;
3988 	}
3989 
3990 	sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3991 	if (!sit_i->tmp_map)
3992 		return -ENOMEM;
3993 
3994 	if (__is_large_section(sbi)) {
3995 		sit_i->sec_entries =
3996 			f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3997 						      MAIN_SECS(sbi)),
3998 				      GFP_KERNEL);
3999 		if (!sit_i->sec_entries)
4000 			return -ENOMEM;
4001 	}
4002 
4003 	/* get information related with SIT */
4004 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4005 
4006 	/* setup SIT bitmap from ckeckpoint pack */
4007 	bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4008 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4009 
4010 	sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
4011 	if (!sit_i->sit_bitmap)
4012 		return -ENOMEM;
4013 
4014 #ifdef CONFIG_F2FS_CHECK_FS
4015 	sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
4016 	if (!sit_i->sit_bitmap_mir)
4017 		return -ENOMEM;
4018 #endif
4019 
4020 	/* init SIT information */
4021 	sit_i->s_ops = &default_salloc_ops;
4022 
4023 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4024 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4025 	sit_i->written_valid_blocks = 0;
4026 	sit_i->bitmap_size = bitmap_size;
4027 	sit_i->dirty_sentries = 0;
4028 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4029 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4030 	sit_i->mounted_time = ktime_get_real_seconds();
4031 	init_rwsem(&sit_i->sentry_lock);
4032 	return 0;
4033 }
4034 
4035 static int build_free_segmap(struct f2fs_sb_info *sbi)
4036 {
4037 	struct free_segmap_info *free_i;
4038 	unsigned int bitmap_size, sec_bitmap_size;
4039 
4040 	/* allocate memory for free segmap information */
4041 	free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4042 	if (!free_i)
4043 		return -ENOMEM;
4044 
4045 	SM_I(sbi)->free_info = free_i;
4046 
4047 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4048 	free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4049 	if (!free_i->free_segmap)
4050 		return -ENOMEM;
4051 
4052 	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4053 	free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4054 	if (!free_i->free_secmap)
4055 		return -ENOMEM;
4056 
4057 	/* set all segments as dirty temporarily */
4058 	memset(free_i->free_segmap, 0xff, bitmap_size);
4059 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4060 
4061 	/* init free segmap information */
4062 	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4063 	free_i->free_segments = 0;
4064 	free_i->free_sections = 0;
4065 	spin_lock_init(&free_i->segmap_lock);
4066 	return 0;
4067 }
4068 
4069 static int build_curseg(struct f2fs_sb_info *sbi)
4070 {
4071 	struct curseg_info *array;
4072 	int i;
4073 
4074 	array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
4075 			     GFP_KERNEL);
4076 	if (!array)
4077 		return -ENOMEM;
4078 
4079 	SM_I(sbi)->curseg_array = array;
4080 
4081 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
4082 		mutex_init(&array[i].curseg_mutex);
4083 		array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4084 		if (!array[i].sum_blk)
4085 			return -ENOMEM;
4086 		init_rwsem(&array[i].journal_rwsem);
4087 		array[i].journal = f2fs_kzalloc(sbi,
4088 				sizeof(struct f2fs_journal), GFP_KERNEL);
4089 		if (!array[i].journal)
4090 			return -ENOMEM;
4091 		array[i].segno = NULL_SEGNO;
4092 		array[i].next_blkoff = 0;
4093 	}
4094 	return restore_curseg_summaries(sbi);
4095 }
4096 
4097 static int build_sit_entries(struct f2fs_sb_info *sbi)
4098 {
4099 	struct sit_info *sit_i = SIT_I(sbi);
4100 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4101 	struct f2fs_journal *journal = curseg->journal;
4102 	struct seg_entry *se;
4103 	struct f2fs_sit_entry sit;
4104 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
4105 	unsigned int i, start, end;
4106 	unsigned int readed, start_blk = 0;
4107 	int err = 0;
4108 	block_t total_node_blocks = 0;
4109 
4110 	do {
4111 		readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4112 							META_SIT, true);
4113 
4114 		start = start_blk * sit_i->sents_per_block;
4115 		end = (start_blk + readed) * sit_i->sents_per_block;
4116 
4117 		for (; start < end && start < MAIN_SEGS(sbi); start++) {
4118 			struct f2fs_sit_block *sit_blk;
4119 			struct page *page;
4120 
4121 			se = &sit_i->sentries[start];
4122 			page = get_current_sit_page(sbi, start);
4123 			if (IS_ERR(page))
4124 				return PTR_ERR(page);
4125 			sit_blk = (struct f2fs_sit_block *)page_address(page);
4126 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4127 			f2fs_put_page(page, 1);
4128 
4129 			err = check_block_count(sbi, start, &sit);
4130 			if (err)
4131 				return err;
4132 			seg_info_from_raw_sit(se, &sit);
4133 			if (IS_NODESEG(se->type))
4134 				total_node_blocks += se->valid_blocks;
4135 
4136 			/* build discard map only one time */
4137 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4138 				memset(se->discard_map, 0xff,
4139 					SIT_VBLOCK_MAP_SIZE);
4140 			} else {
4141 				memcpy(se->discard_map,
4142 					se->cur_valid_map,
4143 					SIT_VBLOCK_MAP_SIZE);
4144 				sbi->discard_blks +=
4145 					sbi->blocks_per_seg -
4146 					se->valid_blocks;
4147 			}
4148 
4149 			if (__is_large_section(sbi))
4150 				get_sec_entry(sbi, start)->valid_blocks +=
4151 							se->valid_blocks;
4152 		}
4153 		start_blk += readed;
4154 	} while (start_blk < sit_blk_cnt);
4155 
4156 	down_read(&curseg->journal_rwsem);
4157 	for (i = 0; i < sits_in_cursum(journal); i++) {
4158 		unsigned int old_valid_blocks;
4159 
4160 		start = le32_to_cpu(segno_in_journal(journal, i));
4161 		if (start >= MAIN_SEGS(sbi)) {
4162 			f2fs_err(sbi, "Wrong journal entry on segno %u",
4163 				 start);
4164 			set_sbi_flag(sbi, SBI_NEED_FSCK);
4165 			err = -EFSCORRUPTED;
4166 			break;
4167 		}
4168 
4169 		se = &sit_i->sentries[start];
4170 		sit = sit_in_journal(journal, i);
4171 
4172 		old_valid_blocks = se->valid_blocks;
4173 		if (IS_NODESEG(se->type))
4174 			total_node_blocks -= old_valid_blocks;
4175 
4176 		err = check_block_count(sbi, start, &sit);
4177 		if (err)
4178 			break;
4179 		seg_info_from_raw_sit(se, &sit);
4180 		if (IS_NODESEG(se->type))
4181 			total_node_blocks += se->valid_blocks;
4182 
4183 		if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4184 			memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4185 		} else {
4186 			memcpy(se->discard_map, se->cur_valid_map,
4187 						SIT_VBLOCK_MAP_SIZE);
4188 			sbi->discard_blks += old_valid_blocks;
4189 			sbi->discard_blks -= se->valid_blocks;
4190 		}
4191 
4192 		if (__is_large_section(sbi)) {
4193 			get_sec_entry(sbi, start)->valid_blocks +=
4194 							se->valid_blocks;
4195 			get_sec_entry(sbi, start)->valid_blocks -=
4196 							old_valid_blocks;
4197 		}
4198 	}
4199 	up_read(&curseg->journal_rwsem);
4200 
4201 	if (!err && total_node_blocks != valid_node_count(sbi)) {
4202 		f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4203 			 total_node_blocks, valid_node_count(sbi));
4204 		set_sbi_flag(sbi, SBI_NEED_FSCK);
4205 		err = -EFSCORRUPTED;
4206 	}
4207 
4208 	return err;
4209 }
4210 
4211 static void init_free_segmap(struct f2fs_sb_info *sbi)
4212 {
4213 	unsigned int start;
4214 	int type;
4215 
4216 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4217 		struct seg_entry *sentry = get_seg_entry(sbi, start);
4218 		if (!sentry->valid_blocks)
4219 			__set_free(sbi, start);
4220 		else
4221 			SIT_I(sbi)->written_valid_blocks +=
4222 						sentry->valid_blocks;
4223 	}
4224 
4225 	/* set use the current segments */
4226 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4227 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4228 		__set_test_and_inuse(sbi, curseg_t->segno);
4229 	}
4230 }
4231 
4232 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4233 {
4234 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4235 	struct free_segmap_info *free_i = FREE_I(sbi);
4236 	unsigned int segno = 0, offset = 0;
4237 	unsigned short valid_blocks;
4238 
4239 	while (1) {
4240 		/* find dirty segment based on free segmap */
4241 		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4242 		if (segno >= MAIN_SEGS(sbi))
4243 			break;
4244 		offset = segno + 1;
4245 		valid_blocks = get_valid_blocks(sbi, segno, false);
4246 		if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4247 			continue;
4248 		if (valid_blocks > sbi->blocks_per_seg) {
4249 			f2fs_bug_on(sbi, 1);
4250 			continue;
4251 		}
4252 		mutex_lock(&dirty_i->seglist_lock);
4253 		__locate_dirty_segment(sbi, segno, DIRTY);
4254 		mutex_unlock(&dirty_i->seglist_lock);
4255 	}
4256 }
4257 
4258 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4259 {
4260 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4261 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4262 
4263 	dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4264 	if (!dirty_i->victim_secmap)
4265 		return -ENOMEM;
4266 	return 0;
4267 }
4268 
4269 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4270 {
4271 	struct dirty_seglist_info *dirty_i;
4272 	unsigned int bitmap_size, i;
4273 
4274 	/* allocate memory for dirty segments list information */
4275 	dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4276 								GFP_KERNEL);
4277 	if (!dirty_i)
4278 		return -ENOMEM;
4279 
4280 	SM_I(sbi)->dirty_info = dirty_i;
4281 	mutex_init(&dirty_i->seglist_lock);
4282 
4283 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4284 
4285 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
4286 		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4287 								GFP_KERNEL);
4288 		if (!dirty_i->dirty_segmap[i])
4289 			return -ENOMEM;
4290 	}
4291 
4292 	init_dirty_segmap(sbi);
4293 	return init_victim_secmap(sbi);
4294 }
4295 
4296 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4297 {
4298 	int i;
4299 
4300 	/*
4301 	 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4302 	 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4303 	 */
4304 	for (i = 0; i < NO_CHECK_TYPE; i++) {
4305 		struct curseg_info *curseg = CURSEG_I(sbi, i);
4306 		struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4307 		unsigned int blkofs = curseg->next_blkoff;
4308 
4309 		if (f2fs_test_bit(blkofs, se->cur_valid_map))
4310 			goto out;
4311 
4312 		if (curseg->alloc_type == SSR)
4313 			continue;
4314 
4315 		for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4316 			if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4317 				continue;
4318 out:
4319 			f2fs_err(sbi,
4320 				 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4321 				 i, curseg->segno, curseg->alloc_type,
4322 				 curseg->next_blkoff, blkofs);
4323 			return -EFSCORRUPTED;
4324 		}
4325 	}
4326 	return 0;
4327 }
4328 
4329 /*
4330  * Update min, max modified time for cost-benefit GC algorithm
4331  */
4332 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4333 {
4334 	struct sit_info *sit_i = SIT_I(sbi);
4335 	unsigned int segno;
4336 
4337 	down_write(&sit_i->sentry_lock);
4338 
4339 	sit_i->min_mtime = ULLONG_MAX;
4340 
4341 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4342 		unsigned int i;
4343 		unsigned long long mtime = 0;
4344 
4345 		for (i = 0; i < sbi->segs_per_sec; i++)
4346 			mtime += get_seg_entry(sbi, segno + i)->mtime;
4347 
4348 		mtime = div_u64(mtime, sbi->segs_per_sec);
4349 
4350 		if (sit_i->min_mtime > mtime)
4351 			sit_i->min_mtime = mtime;
4352 	}
4353 	sit_i->max_mtime = get_mtime(sbi, false);
4354 	up_write(&sit_i->sentry_lock);
4355 }
4356 
4357 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4358 {
4359 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4360 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4361 	struct f2fs_sm_info *sm_info;
4362 	int err;
4363 
4364 	sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4365 	if (!sm_info)
4366 		return -ENOMEM;
4367 
4368 	/* init sm info */
4369 	sbi->sm_info = sm_info;
4370 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4371 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4372 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4373 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4374 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4375 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4376 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4377 	sm_info->rec_prefree_segments = sm_info->main_segments *
4378 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4379 	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4380 		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4381 
4382 	if (!test_opt(sbi, LFS))
4383 		sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4384 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4385 	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4386 	sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4387 	sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4388 	sm_info->min_ssr_sections = reserved_sections(sbi);
4389 
4390 	INIT_LIST_HEAD(&sm_info->sit_entry_set);
4391 
4392 	init_rwsem(&sm_info->curseg_lock);
4393 
4394 	if (!f2fs_readonly(sbi->sb)) {
4395 		err = f2fs_create_flush_cmd_control(sbi);
4396 		if (err)
4397 			return err;
4398 	}
4399 
4400 	err = create_discard_cmd_control(sbi);
4401 	if (err)
4402 		return err;
4403 
4404 	err = build_sit_info(sbi);
4405 	if (err)
4406 		return err;
4407 	err = build_free_segmap(sbi);
4408 	if (err)
4409 		return err;
4410 	err = build_curseg(sbi);
4411 	if (err)
4412 		return err;
4413 
4414 	/* reinit free segmap based on SIT */
4415 	err = build_sit_entries(sbi);
4416 	if (err)
4417 		return err;
4418 
4419 	init_free_segmap(sbi);
4420 	err = build_dirty_segmap(sbi);
4421 	if (err)
4422 		return err;
4423 
4424 	err = sanity_check_curseg(sbi);
4425 	if (err)
4426 		return err;
4427 
4428 	init_min_max_mtime(sbi);
4429 	return 0;
4430 }
4431 
4432 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4433 		enum dirty_type dirty_type)
4434 {
4435 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4436 
4437 	mutex_lock(&dirty_i->seglist_lock);
4438 	kvfree(dirty_i->dirty_segmap[dirty_type]);
4439 	dirty_i->nr_dirty[dirty_type] = 0;
4440 	mutex_unlock(&dirty_i->seglist_lock);
4441 }
4442 
4443 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4444 {
4445 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4446 	kvfree(dirty_i->victim_secmap);
4447 }
4448 
4449 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4450 {
4451 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4452 	int i;
4453 
4454 	if (!dirty_i)
4455 		return;
4456 
4457 	/* discard pre-free/dirty segments list */
4458 	for (i = 0; i < NR_DIRTY_TYPE; i++)
4459 		discard_dirty_segmap(sbi, i);
4460 
4461 	destroy_victim_secmap(sbi);
4462 	SM_I(sbi)->dirty_info = NULL;
4463 	kvfree(dirty_i);
4464 }
4465 
4466 static void destroy_curseg(struct f2fs_sb_info *sbi)
4467 {
4468 	struct curseg_info *array = SM_I(sbi)->curseg_array;
4469 	int i;
4470 
4471 	if (!array)
4472 		return;
4473 	SM_I(sbi)->curseg_array = NULL;
4474 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
4475 		kvfree(array[i].sum_blk);
4476 		kvfree(array[i].journal);
4477 	}
4478 	kvfree(array);
4479 }
4480 
4481 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4482 {
4483 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4484 	if (!free_i)
4485 		return;
4486 	SM_I(sbi)->free_info = NULL;
4487 	kvfree(free_i->free_segmap);
4488 	kvfree(free_i->free_secmap);
4489 	kvfree(free_i);
4490 }
4491 
4492 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4493 {
4494 	struct sit_info *sit_i = SIT_I(sbi);
4495 	unsigned int start;
4496 
4497 	if (!sit_i)
4498 		return;
4499 
4500 	if (sit_i->sentries) {
4501 		for (start = 0; start < MAIN_SEGS(sbi); start++) {
4502 			kvfree(sit_i->sentries[start].cur_valid_map);
4503 #ifdef CONFIG_F2FS_CHECK_FS
4504 			kvfree(sit_i->sentries[start].cur_valid_map_mir);
4505 #endif
4506 			kvfree(sit_i->sentries[start].ckpt_valid_map);
4507 			kvfree(sit_i->sentries[start].discard_map);
4508 		}
4509 	}
4510 	kvfree(sit_i->tmp_map);
4511 
4512 	kvfree(sit_i->sentries);
4513 	kvfree(sit_i->sec_entries);
4514 	kvfree(sit_i->dirty_sentries_bitmap);
4515 
4516 	SM_I(sbi)->sit_info = NULL;
4517 	kvfree(sit_i->sit_bitmap);
4518 #ifdef CONFIG_F2FS_CHECK_FS
4519 	kvfree(sit_i->sit_bitmap_mir);
4520 #endif
4521 	kvfree(sit_i);
4522 }
4523 
4524 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4525 {
4526 	struct f2fs_sm_info *sm_info = SM_I(sbi);
4527 
4528 	if (!sm_info)
4529 		return;
4530 	f2fs_destroy_flush_cmd_control(sbi, true);
4531 	destroy_discard_cmd_control(sbi);
4532 	destroy_dirty_segmap(sbi);
4533 	destroy_curseg(sbi);
4534 	destroy_free_segmap(sbi);
4535 	destroy_sit_info(sbi);
4536 	sbi->sm_info = NULL;
4537 	kvfree(sm_info);
4538 }
4539 
4540 int __init f2fs_create_segment_manager_caches(void)
4541 {
4542 	discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4543 			sizeof(struct discard_entry));
4544 	if (!discard_entry_slab)
4545 		goto fail;
4546 
4547 	discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4548 			sizeof(struct discard_cmd));
4549 	if (!discard_cmd_slab)
4550 		goto destroy_discard_entry;
4551 
4552 	sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4553 			sizeof(struct sit_entry_set));
4554 	if (!sit_entry_set_slab)
4555 		goto destroy_discard_cmd;
4556 
4557 	inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4558 			sizeof(struct inmem_pages));
4559 	if (!inmem_entry_slab)
4560 		goto destroy_sit_entry_set;
4561 	return 0;
4562 
4563 destroy_sit_entry_set:
4564 	kmem_cache_destroy(sit_entry_set_slab);
4565 destroy_discard_cmd:
4566 	kmem_cache_destroy(discard_cmd_slab);
4567 destroy_discard_entry:
4568 	kmem_cache_destroy(discard_entry_slab);
4569 fail:
4570 	return -ENOMEM;
4571 }
4572 
4573 void f2fs_destroy_segment_manager_caches(void)
4574 {
4575 	kmem_cache_destroy(sit_entry_set_slab);
4576 	kmem_cache_destroy(discard_cmd_slab);
4577 	kmem_cache_destroy(discard_entry_slab);
4578 	kmem_cache_destroy(inmem_entry_slab);
4579 }
4580