xref: /openbmc/linux/fs/f2fs/segment.c (revision 9b93eb47)
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 			blk_start_plug(&plug);
550 			f2fs_sync_dirty_inodes(sbi, FILE_INODE);
551 			blk_finish_plug(&plug);
552 		}
553 		f2fs_sync_fs(sbi->sb, true);
554 		stat_inc_bg_cp_count(sbi->stat_info);
555 	}
556 }
557 
558 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
559 				struct block_device *bdev)
560 {
561 	struct bio *bio;
562 	int ret;
563 
564 	bio = f2fs_bio_alloc(sbi, 0, false);
565 	if (!bio)
566 		return -ENOMEM;
567 
568 	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
569 	bio_set_dev(bio, bdev);
570 	ret = submit_bio_wait(bio);
571 	bio_put(bio);
572 
573 	trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
574 				test_opt(sbi, FLUSH_MERGE), ret);
575 	return ret;
576 }
577 
578 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
579 {
580 	int ret = 0;
581 	int i;
582 
583 	if (!f2fs_is_multi_device(sbi))
584 		return __submit_flush_wait(sbi, sbi->sb->s_bdev);
585 
586 	for (i = 0; i < sbi->s_ndevs; i++) {
587 		if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
588 			continue;
589 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
590 		if (ret)
591 			break;
592 	}
593 	return ret;
594 }
595 
596 static int issue_flush_thread(void *data)
597 {
598 	struct f2fs_sb_info *sbi = data;
599 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
600 	wait_queue_head_t *q = &fcc->flush_wait_queue;
601 repeat:
602 	if (kthread_should_stop())
603 		return 0;
604 
605 	sb_start_intwrite(sbi->sb);
606 
607 	if (!llist_empty(&fcc->issue_list)) {
608 		struct flush_cmd *cmd, *next;
609 		int ret;
610 
611 		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
612 		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
613 
614 		cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
615 
616 		ret = submit_flush_wait(sbi, cmd->ino);
617 		atomic_inc(&fcc->issued_flush);
618 
619 		llist_for_each_entry_safe(cmd, next,
620 					  fcc->dispatch_list, llnode) {
621 			cmd->ret = ret;
622 			complete(&cmd->wait);
623 		}
624 		fcc->dispatch_list = NULL;
625 	}
626 
627 	sb_end_intwrite(sbi->sb);
628 
629 	wait_event_interruptible(*q,
630 		kthread_should_stop() || !llist_empty(&fcc->issue_list));
631 	goto repeat;
632 }
633 
634 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
635 {
636 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
637 	struct flush_cmd cmd;
638 	int ret;
639 
640 	if (test_opt(sbi, NOBARRIER))
641 		return 0;
642 
643 	if (!test_opt(sbi, FLUSH_MERGE)) {
644 		atomic_inc(&fcc->queued_flush);
645 		ret = submit_flush_wait(sbi, ino);
646 		atomic_dec(&fcc->queued_flush);
647 		atomic_inc(&fcc->issued_flush);
648 		return ret;
649 	}
650 
651 	if (atomic_inc_return(&fcc->queued_flush) == 1 ||
652 	    f2fs_is_multi_device(sbi)) {
653 		ret = submit_flush_wait(sbi, ino);
654 		atomic_dec(&fcc->queued_flush);
655 
656 		atomic_inc(&fcc->issued_flush);
657 		return ret;
658 	}
659 
660 	cmd.ino = ino;
661 	init_completion(&cmd.wait);
662 
663 	llist_add(&cmd.llnode, &fcc->issue_list);
664 
665 	/* update issue_list before we wake up issue_flush thread */
666 	smp_mb();
667 
668 	if (waitqueue_active(&fcc->flush_wait_queue))
669 		wake_up(&fcc->flush_wait_queue);
670 
671 	if (fcc->f2fs_issue_flush) {
672 		wait_for_completion(&cmd.wait);
673 		atomic_dec(&fcc->queued_flush);
674 	} else {
675 		struct llist_node *list;
676 
677 		list = llist_del_all(&fcc->issue_list);
678 		if (!list) {
679 			wait_for_completion(&cmd.wait);
680 			atomic_dec(&fcc->queued_flush);
681 		} else {
682 			struct flush_cmd *tmp, *next;
683 
684 			ret = submit_flush_wait(sbi, ino);
685 
686 			llist_for_each_entry_safe(tmp, next, list, llnode) {
687 				if (tmp == &cmd) {
688 					cmd.ret = ret;
689 					atomic_dec(&fcc->queued_flush);
690 					continue;
691 				}
692 				tmp->ret = ret;
693 				complete(&tmp->wait);
694 			}
695 		}
696 	}
697 
698 	return cmd.ret;
699 }
700 
701 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
702 {
703 	dev_t dev = sbi->sb->s_bdev->bd_dev;
704 	struct flush_cmd_control *fcc;
705 	int err = 0;
706 
707 	if (SM_I(sbi)->fcc_info) {
708 		fcc = SM_I(sbi)->fcc_info;
709 		if (fcc->f2fs_issue_flush)
710 			return err;
711 		goto init_thread;
712 	}
713 
714 	fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
715 	if (!fcc)
716 		return -ENOMEM;
717 	atomic_set(&fcc->issued_flush, 0);
718 	atomic_set(&fcc->queued_flush, 0);
719 	init_waitqueue_head(&fcc->flush_wait_queue);
720 	init_llist_head(&fcc->issue_list);
721 	SM_I(sbi)->fcc_info = fcc;
722 	if (!test_opt(sbi, FLUSH_MERGE))
723 		return err;
724 
725 init_thread:
726 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
727 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
728 	if (IS_ERR(fcc->f2fs_issue_flush)) {
729 		err = PTR_ERR(fcc->f2fs_issue_flush);
730 		kvfree(fcc);
731 		SM_I(sbi)->fcc_info = NULL;
732 		return err;
733 	}
734 
735 	return err;
736 }
737 
738 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
739 {
740 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
741 
742 	if (fcc && fcc->f2fs_issue_flush) {
743 		struct task_struct *flush_thread = fcc->f2fs_issue_flush;
744 
745 		fcc->f2fs_issue_flush = NULL;
746 		kthread_stop(flush_thread);
747 	}
748 	if (free) {
749 		kvfree(fcc);
750 		SM_I(sbi)->fcc_info = NULL;
751 	}
752 }
753 
754 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
755 {
756 	int ret = 0, i;
757 
758 	if (!f2fs_is_multi_device(sbi))
759 		return 0;
760 
761 	for (i = 1; i < sbi->s_ndevs; i++) {
762 		if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
763 			continue;
764 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
765 		if (ret)
766 			break;
767 
768 		spin_lock(&sbi->dev_lock);
769 		f2fs_clear_bit(i, (char *)&sbi->dirty_device);
770 		spin_unlock(&sbi->dev_lock);
771 	}
772 
773 	return ret;
774 }
775 
776 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
777 		enum dirty_type dirty_type)
778 {
779 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
780 
781 	/* need not be added */
782 	if (IS_CURSEG(sbi, segno))
783 		return;
784 
785 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
786 		dirty_i->nr_dirty[dirty_type]++;
787 
788 	if (dirty_type == DIRTY) {
789 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
790 		enum dirty_type t = sentry->type;
791 
792 		if (unlikely(t >= DIRTY)) {
793 			f2fs_bug_on(sbi, 1);
794 			return;
795 		}
796 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
797 			dirty_i->nr_dirty[t]++;
798 	}
799 }
800 
801 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
802 		enum dirty_type dirty_type)
803 {
804 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
805 
806 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
807 		dirty_i->nr_dirty[dirty_type]--;
808 
809 	if (dirty_type == DIRTY) {
810 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
811 		enum dirty_type t = sentry->type;
812 
813 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
814 			dirty_i->nr_dirty[t]--;
815 
816 		if (get_valid_blocks(sbi, segno, true) == 0)
817 			clear_bit(GET_SEC_FROM_SEG(sbi, segno),
818 						dirty_i->victim_secmap);
819 	}
820 }
821 
822 /*
823  * Should not occur error such as -ENOMEM.
824  * Adding dirty entry into seglist is not critical operation.
825  * If a given segment is one of current working segments, it won't be added.
826  */
827 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
828 {
829 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
830 	unsigned short valid_blocks, ckpt_valid_blocks;
831 
832 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
833 		return;
834 
835 	mutex_lock(&dirty_i->seglist_lock);
836 
837 	valid_blocks = get_valid_blocks(sbi, segno, false);
838 	ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
839 
840 	if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
841 				ckpt_valid_blocks == sbi->blocks_per_seg)) {
842 		__locate_dirty_segment(sbi, segno, PRE);
843 		__remove_dirty_segment(sbi, segno, DIRTY);
844 	} else if (valid_blocks < sbi->blocks_per_seg) {
845 		__locate_dirty_segment(sbi, segno, DIRTY);
846 	} else {
847 		/* Recovery routine with SSR needs this */
848 		__remove_dirty_segment(sbi, segno, DIRTY);
849 	}
850 
851 	mutex_unlock(&dirty_i->seglist_lock);
852 }
853 
854 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
855 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
856 {
857 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
858 	unsigned int segno;
859 
860 	mutex_lock(&dirty_i->seglist_lock);
861 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
862 		if (get_valid_blocks(sbi, segno, false))
863 			continue;
864 		if (IS_CURSEG(sbi, segno))
865 			continue;
866 		__locate_dirty_segment(sbi, segno, PRE);
867 		__remove_dirty_segment(sbi, segno, DIRTY);
868 	}
869 	mutex_unlock(&dirty_i->seglist_lock);
870 }
871 
872 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi)
873 {
874 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
875 	block_t ovp = overprovision_segments(sbi) << sbi->log_blocks_per_seg;
876 	block_t holes[2] = {0, 0};	/* DATA and NODE */
877 	struct seg_entry *se;
878 	unsigned int segno;
879 
880 	mutex_lock(&dirty_i->seglist_lock);
881 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
882 		se = get_seg_entry(sbi, segno);
883 		if (IS_NODESEG(se->type))
884 			holes[NODE] += sbi->blocks_per_seg - se->valid_blocks;
885 		else
886 			holes[DATA] += sbi->blocks_per_seg - se->valid_blocks;
887 	}
888 	mutex_unlock(&dirty_i->seglist_lock);
889 
890 	if (holes[DATA] > ovp || holes[NODE] > ovp)
891 		return -EAGAIN;
892 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
893 		dirty_segments(sbi) > overprovision_segments(sbi))
894 		return -EAGAIN;
895 	return 0;
896 }
897 
898 /* This is only used by SBI_CP_DISABLED */
899 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
900 {
901 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
902 	unsigned int segno = 0;
903 
904 	mutex_lock(&dirty_i->seglist_lock);
905 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
906 		if (get_valid_blocks(sbi, segno, false))
907 			continue;
908 		if (get_ckpt_valid_blocks(sbi, segno))
909 			continue;
910 		mutex_unlock(&dirty_i->seglist_lock);
911 		return segno;
912 	}
913 	mutex_unlock(&dirty_i->seglist_lock);
914 	return NULL_SEGNO;
915 }
916 
917 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
918 		struct block_device *bdev, block_t lstart,
919 		block_t start, block_t len)
920 {
921 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
922 	struct list_head *pend_list;
923 	struct discard_cmd *dc;
924 
925 	f2fs_bug_on(sbi, !len);
926 
927 	pend_list = &dcc->pend_list[plist_idx(len)];
928 
929 	dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
930 	INIT_LIST_HEAD(&dc->list);
931 	dc->bdev = bdev;
932 	dc->lstart = lstart;
933 	dc->start = start;
934 	dc->len = len;
935 	dc->ref = 0;
936 	dc->state = D_PREP;
937 	dc->queued = 0;
938 	dc->error = 0;
939 	init_completion(&dc->wait);
940 	list_add_tail(&dc->list, pend_list);
941 	spin_lock_init(&dc->lock);
942 	dc->bio_ref = 0;
943 	atomic_inc(&dcc->discard_cmd_cnt);
944 	dcc->undiscard_blks += len;
945 
946 	return dc;
947 }
948 
949 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
950 				struct block_device *bdev, block_t lstart,
951 				block_t start, block_t len,
952 				struct rb_node *parent, struct rb_node **p,
953 				bool leftmost)
954 {
955 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
956 	struct discard_cmd *dc;
957 
958 	dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
959 
960 	rb_link_node(&dc->rb_node, parent, p);
961 	rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
962 
963 	return dc;
964 }
965 
966 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
967 							struct discard_cmd *dc)
968 {
969 	if (dc->state == D_DONE)
970 		atomic_sub(dc->queued, &dcc->queued_discard);
971 
972 	list_del(&dc->list);
973 	rb_erase_cached(&dc->rb_node, &dcc->root);
974 	dcc->undiscard_blks -= dc->len;
975 
976 	kmem_cache_free(discard_cmd_slab, dc);
977 
978 	atomic_dec(&dcc->discard_cmd_cnt);
979 }
980 
981 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
982 							struct discard_cmd *dc)
983 {
984 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
985 	unsigned long flags;
986 
987 	trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
988 
989 	spin_lock_irqsave(&dc->lock, flags);
990 	if (dc->bio_ref) {
991 		spin_unlock_irqrestore(&dc->lock, flags);
992 		return;
993 	}
994 	spin_unlock_irqrestore(&dc->lock, flags);
995 
996 	f2fs_bug_on(sbi, dc->ref);
997 
998 	if (dc->error == -EOPNOTSUPP)
999 		dc->error = 0;
1000 
1001 	if (dc->error)
1002 		printk_ratelimited(
1003 			"%sF2FS-fs: Issue discard(%u, %u, %u) failed, ret: %d",
1004 			KERN_INFO, dc->lstart, dc->start, dc->len, dc->error);
1005 	__detach_discard_cmd(dcc, dc);
1006 }
1007 
1008 static void f2fs_submit_discard_endio(struct bio *bio)
1009 {
1010 	struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1011 	unsigned long flags;
1012 
1013 	dc->error = blk_status_to_errno(bio->bi_status);
1014 
1015 	spin_lock_irqsave(&dc->lock, flags);
1016 	dc->bio_ref--;
1017 	if (!dc->bio_ref && dc->state == D_SUBMIT) {
1018 		dc->state = D_DONE;
1019 		complete_all(&dc->wait);
1020 	}
1021 	spin_unlock_irqrestore(&dc->lock, flags);
1022 	bio_put(bio);
1023 }
1024 
1025 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1026 				block_t start, block_t end)
1027 {
1028 #ifdef CONFIG_F2FS_CHECK_FS
1029 	struct seg_entry *sentry;
1030 	unsigned int segno;
1031 	block_t blk = start;
1032 	unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1033 	unsigned long *map;
1034 
1035 	while (blk < end) {
1036 		segno = GET_SEGNO(sbi, blk);
1037 		sentry = get_seg_entry(sbi, segno);
1038 		offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1039 
1040 		if (end < START_BLOCK(sbi, segno + 1))
1041 			size = GET_BLKOFF_FROM_SEG0(sbi, end);
1042 		else
1043 			size = max_blocks;
1044 		map = (unsigned long *)(sentry->cur_valid_map);
1045 		offset = __find_rev_next_bit(map, size, offset);
1046 		f2fs_bug_on(sbi, offset != size);
1047 		blk = START_BLOCK(sbi, segno + 1);
1048 	}
1049 #endif
1050 }
1051 
1052 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1053 				struct discard_policy *dpolicy,
1054 				int discard_type, unsigned int granularity)
1055 {
1056 	/* common policy */
1057 	dpolicy->type = discard_type;
1058 	dpolicy->sync = true;
1059 	dpolicy->ordered = false;
1060 	dpolicy->granularity = granularity;
1061 
1062 	dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1063 	dpolicy->io_aware_gran = MAX_PLIST_NUM;
1064 	dpolicy->timeout = 0;
1065 
1066 	if (discard_type == DPOLICY_BG) {
1067 		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1068 		dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1069 		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1070 		dpolicy->io_aware = true;
1071 		dpolicy->sync = false;
1072 		dpolicy->ordered = true;
1073 		if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1074 			dpolicy->granularity = 1;
1075 			dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1076 		}
1077 	} else if (discard_type == DPOLICY_FORCE) {
1078 		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1079 		dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1080 		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1081 		dpolicy->io_aware = false;
1082 	} else if (discard_type == DPOLICY_FSTRIM) {
1083 		dpolicy->io_aware = false;
1084 	} else if (discard_type == DPOLICY_UMOUNT) {
1085 		dpolicy->max_requests = UINT_MAX;
1086 		dpolicy->io_aware = false;
1087 		/* we need to issue all to keep CP_TRIMMED_FLAG */
1088 		dpolicy->granularity = 1;
1089 	}
1090 }
1091 
1092 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1093 				struct block_device *bdev, block_t lstart,
1094 				block_t start, block_t len);
1095 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1096 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1097 						struct discard_policy *dpolicy,
1098 						struct discard_cmd *dc,
1099 						unsigned int *issued)
1100 {
1101 	struct block_device *bdev = dc->bdev;
1102 	struct request_queue *q = bdev_get_queue(bdev);
1103 	unsigned int max_discard_blocks =
1104 			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1105 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1106 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1107 					&(dcc->fstrim_list) : &(dcc->wait_list);
1108 	int flag = dpolicy->sync ? REQ_SYNC : 0;
1109 	block_t lstart, start, len, total_len;
1110 	int err = 0;
1111 
1112 	if (dc->state != D_PREP)
1113 		return 0;
1114 
1115 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1116 		return 0;
1117 
1118 	trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1119 
1120 	lstart = dc->lstart;
1121 	start = dc->start;
1122 	len = dc->len;
1123 	total_len = len;
1124 
1125 	dc->len = 0;
1126 
1127 	while (total_len && *issued < dpolicy->max_requests && !err) {
1128 		struct bio *bio = NULL;
1129 		unsigned long flags;
1130 		bool last = true;
1131 
1132 		if (len > max_discard_blocks) {
1133 			len = max_discard_blocks;
1134 			last = false;
1135 		}
1136 
1137 		(*issued)++;
1138 		if (*issued == dpolicy->max_requests)
1139 			last = true;
1140 
1141 		dc->len += len;
1142 
1143 		if (time_to_inject(sbi, FAULT_DISCARD)) {
1144 			f2fs_show_injection_info(FAULT_DISCARD);
1145 			err = -EIO;
1146 			goto submit;
1147 		}
1148 		err = __blkdev_issue_discard(bdev,
1149 					SECTOR_FROM_BLOCK(start),
1150 					SECTOR_FROM_BLOCK(len),
1151 					GFP_NOFS, 0, &bio);
1152 submit:
1153 		if (err) {
1154 			spin_lock_irqsave(&dc->lock, flags);
1155 			if (dc->state == D_PARTIAL)
1156 				dc->state = D_SUBMIT;
1157 			spin_unlock_irqrestore(&dc->lock, flags);
1158 
1159 			break;
1160 		}
1161 
1162 		f2fs_bug_on(sbi, !bio);
1163 
1164 		/*
1165 		 * should keep before submission to avoid D_DONE
1166 		 * right away
1167 		 */
1168 		spin_lock_irqsave(&dc->lock, flags);
1169 		if (last)
1170 			dc->state = D_SUBMIT;
1171 		else
1172 			dc->state = D_PARTIAL;
1173 		dc->bio_ref++;
1174 		spin_unlock_irqrestore(&dc->lock, flags);
1175 
1176 		atomic_inc(&dcc->queued_discard);
1177 		dc->queued++;
1178 		list_move_tail(&dc->list, wait_list);
1179 
1180 		/* sanity check on discard range */
1181 		__check_sit_bitmap(sbi, lstart, lstart + len);
1182 
1183 		bio->bi_private = dc;
1184 		bio->bi_end_io = f2fs_submit_discard_endio;
1185 		bio->bi_opf |= flag;
1186 		submit_bio(bio);
1187 
1188 		atomic_inc(&dcc->issued_discard);
1189 
1190 		f2fs_update_iostat(sbi, FS_DISCARD, 1);
1191 
1192 		lstart += len;
1193 		start += len;
1194 		total_len -= len;
1195 		len = total_len;
1196 	}
1197 
1198 	if (!err && len)
1199 		__update_discard_tree_range(sbi, bdev, lstart, start, len);
1200 	return err;
1201 }
1202 
1203 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1204 				struct block_device *bdev, block_t lstart,
1205 				block_t start, block_t len,
1206 				struct rb_node **insert_p,
1207 				struct rb_node *insert_parent)
1208 {
1209 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1210 	struct rb_node **p;
1211 	struct rb_node *parent = NULL;
1212 	struct discard_cmd *dc = NULL;
1213 	bool leftmost = true;
1214 
1215 	if (insert_p && insert_parent) {
1216 		parent = insert_parent;
1217 		p = insert_p;
1218 		goto do_insert;
1219 	}
1220 
1221 	p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1222 							lstart, &leftmost);
1223 do_insert:
1224 	dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1225 								p, leftmost);
1226 	if (!dc)
1227 		return NULL;
1228 
1229 	return dc;
1230 }
1231 
1232 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1233 						struct discard_cmd *dc)
1234 {
1235 	list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1236 }
1237 
1238 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1239 				struct discard_cmd *dc, block_t blkaddr)
1240 {
1241 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1242 	struct discard_info di = dc->di;
1243 	bool modified = false;
1244 
1245 	if (dc->state == D_DONE || dc->len == 1) {
1246 		__remove_discard_cmd(sbi, dc);
1247 		return;
1248 	}
1249 
1250 	dcc->undiscard_blks -= di.len;
1251 
1252 	if (blkaddr > di.lstart) {
1253 		dc->len = blkaddr - dc->lstart;
1254 		dcc->undiscard_blks += dc->len;
1255 		__relocate_discard_cmd(dcc, dc);
1256 		modified = true;
1257 	}
1258 
1259 	if (blkaddr < di.lstart + di.len - 1) {
1260 		if (modified) {
1261 			__insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1262 					di.start + blkaddr + 1 - di.lstart,
1263 					di.lstart + di.len - 1 - blkaddr,
1264 					NULL, NULL);
1265 		} else {
1266 			dc->lstart++;
1267 			dc->len--;
1268 			dc->start++;
1269 			dcc->undiscard_blks += dc->len;
1270 			__relocate_discard_cmd(dcc, dc);
1271 		}
1272 	}
1273 }
1274 
1275 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1276 				struct block_device *bdev, block_t lstart,
1277 				block_t start, block_t len)
1278 {
1279 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1280 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1281 	struct discard_cmd *dc;
1282 	struct discard_info di = {0};
1283 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1284 	struct request_queue *q = bdev_get_queue(bdev);
1285 	unsigned int max_discard_blocks =
1286 			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1287 	block_t end = lstart + len;
1288 
1289 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1290 					NULL, lstart,
1291 					(struct rb_entry **)&prev_dc,
1292 					(struct rb_entry **)&next_dc,
1293 					&insert_p, &insert_parent, true, NULL);
1294 	if (dc)
1295 		prev_dc = dc;
1296 
1297 	if (!prev_dc) {
1298 		di.lstart = lstart;
1299 		di.len = next_dc ? next_dc->lstart - lstart : len;
1300 		di.len = min(di.len, len);
1301 		di.start = start;
1302 	}
1303 
1304 	while (1) {
1305 		struct rb_node *node;
1306 		bool merged = false;
1307 		struct discard_cmd *tdc = NULL;
1308 
1309 		if (prev_dc) {
1310 			di.lstart = prev_dc->lstart + prev_dc->len;
1311 			if (di.lstart < lstart)
1312 				di.lstart = lstart;
1313 			if (di.lstart >= end)
1314 				break;
1315 
1316 			if (!next_dc || next_dc->lstart > end)
1317 				di.len = end - di.lstart;
1318 			else
1319 				di.len = next_dc->lstart - di.lstart;
1320 			di.start = start + di.lstart - lstart;
1321 		}
1322 
1323 		if (!di.len)
1324 			goto next;
1325 
1326 		if (prev_dc && prev_dc->state == D_PREP &&
1327 			prev_dc->bdev == bdev &&
1328 			__is_discard_back_mergeable(&di, &prev_dc->di,
1329 							max_discard_blocks)) {
1330 			prev_dc->di.len += di.len;
1331 			dcc->undiscard_blks += di.len;
1332 			__relocate_discard_cmd(dcc, prev_dc);
1333 			di = prev_dc->di;
1334 			tdc = prev_dc;
1335 			merged = true;
1336 		}
1337 
1338 		if (next_dc && next_dc->state == D_PREP &&
1339 			next_dc->bdev == bdev &&
1340 			__is_discard_front_mergeable(&di, &next_dc->di,
1341 							max_discard_blocks)) {
1342 			next_dc->di.lstart = di.lstart;
1343 			next_dc->di.len += di.len;
1344 			next_dc->di.start = di.start;
1345 			dcc->undiscard_blks += di.len;
1346 			__relocate_discard_cmd(dcc, next_dc);
1347 			if (tdc)
1348 				__remove_discard_cmd(sbi, tdc);
1349 			merged = true;
1350 		}
1351 
1352 		if (!merged) {
1353 			__insert_discard_tree(sbi, bdev, di.lstart, di.start,
1354 							di.len, NULL, NULL);
1355 		}
1356  next:
1357 		prev_dc = next_dc;
1358 		if (!prev_dc)
1359 			break;
1360 
1361 		node = rb_next(&prev_dc->rb_node);
1362 		next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1363 	}
1364 }
1365 
1366 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1367 		struct block_device *bdev, block_t blkstart, block_t blklen)
1368 {
1369 	block_t lblkstart = blkstart;
1370 
1371 	if (!f2fs_bdev_support_discard(bdev))
1372 		return 0;
1373 
1374 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
1375 
1376 	if (f2fs_is_multi_device(sbi)) {
1377 		int devi = f2fs_target_device_index(sbi, blkstart);
1378 
1379 		blkstart -= FDEV(devi).start_blk;
1380 	}
1381 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1382 	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1383 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1384 	return 0;
1385 }
1386 
1387 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1388 					struct discard_policy *dpolicy)
1389 {
1390 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1391 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1392 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1393 	struct discard_cmd *dc;
1394 	struct blk_plug plug;
1395 	unsigned int pos = dcc->next_pos;
1396 	unsigned int issued = 0;
1397 	bool io_interrupted = false;
1398 
1399 	mutex_lock(&dcc->cmd_lock);
1400 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1401 					NULL, pos,
1402 					(struct rb_entry **)&prev_dc,
1403 					(struct rb_entry **)&next_dc,
1404 					&insert_p, &insert_parent, true, NULL);
1405 	if (!dc)
1406 		dc = next_dc;
1407 
1408 	blk_start_plug(&plug);
1409 
1410 	while (dc) {
1411 		struct rb_node *node;
1412 		int err = 0;
1413 
1414 		if (dc->state != D_PREP)
1415 			goto next;
1416 
1417 		if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1418 			io_interrupted = true;
1419 			break;
1420 		}
1421 
1422 		dcc->next_pos = dc->lstart + dc->len;
1423 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1424 
1425 		if (issued >= dpolicy->max_requests)
1426 			break;
1427 next:
1428 		node = rb_next(&dc->rb_node);
1429 		if (err)
1430 			__remove_discard_cmd(sbi, dc);
1431 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1432 	}
1433 
1434 	blk_finish_plug(&plug);
1435 
1436 	if (!dc)
1437 		dcc->next_pos = 0;
1438 
1439 	mutex_unlock(&dcc->cmd_lock);
1440 
1441 	if (!issued && io_interrupted)
1442 		issued = -1;
1443 
1444 	return issued;
1445 }
1446 
1447 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1448 					struct discard_policy *dpolicy)
1449 {
1450 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1451 	struct list_head *pend_list;
1452 	struct discard_cmd *dc, *tmp;
1453 	struct blk_plug plug;
1454 	int i, issued = 0;
1455 	bool io_interrupted = false;
1456 
1457 	if (dpolicy->timeout != 0)
1458 		f2fs_update_time(sbi, dpolicy->timeout);
1459 
1460 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1461 		if (dpolicy->timeout != 0 &&
1462 				f2fs_time_over(sbi, dpolicy->timeout))
1463 			break;
1464 
1465 		if (i + 1 < dpolicy->granularity)
1466 			break;
1467 
1468 		if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1469 			return __issue_discard_cmd_orderly(sbi, dpolicy);
1470 
1471 		pend_list = &dcc->pend_list[i];
1472 
1473 		mutex_lock(&dcc->cmd_lock);
1474 		if (list_empty(pend_list))
1475 			goto next;
1476 		if (unlikely(dcc->rbtree_check))
1477 			f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1478 								&dcc->root));
1479 		blk_start_plug(&plug);
1480 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1481 			f2fs_bug_on(sbi, dc->state != D_PREP);
1482 
1483 			if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1484 						!is_idle(sbi, DISCARD_TIME)) {
1485 				io_interrupted = true;
1486 				break;
1487 			}
1488 
1489 			__submit_discard_cmd(sbi, dpolicy, dc, &issued);
1490 
1491 			if (issued >= dpolicy->max_requests)
1492 				break;
1493 		}
1494 		blk_finish_plug(&plug);
1495 next:
1496 		mutex_unlock(&dcc->cmd_lock);
1497 
1498 		if (issued >= dpolicy->max_requests || io_interrupted)
1499 			break;
1500 	}
1501 
1502 	if (!issued && io_interrupted)
1503 		issued = -1;
1504 
1505 	return issued;
1506 }
1507 
1508 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1509 {
1510 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1511 	struct list_head *pend_list;
1512 	struct discard_cmd *dc, *tmp;
1513 	int i;
1514 	bool dropped = false;
1515 
1516 	mutex_lock(&dcc->cmd_lock);
1517 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1518 		pend_list = &dcc->pend_list[i];
1519 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1520 			f2fs_bug_on(sbi, dc->state != D_PREP);
1521 			__remove_discard_cmd(sbi, dc);
1522 			dropped = true;
1523 		}
1524 	}
1525 	mutex_unlock(&dcc->cmd_lock);
1526 
1527 	return dropped;
1528 }
1529 
1530 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1531 {
1532 	__drop_discard_cmd(sbi);
1533 }
1534 
1535 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1536 							struct discard_cmd *dc)
1537 {
1538 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1539 	unsigned int len = 0;
1540 
1541 	wait_for_completion_io(&dc->wait);
1542 	mutex_lock(&dcc->cmd_lock);
1543 	f2fs_bug_on(sbi, dc->state != D_DONE);
1544 	dc->ref--;
1545 	if (!dc->ref) {
1546 		if (!dc->error)
1547 			len = dc->len;
1548 		__remove_discard_cmd(sbi, dc);
1549 	}
1550 	mutex_unlock(&dcc->cmd_lock);
1551 
1552 	return len;
1553 }
1554 
1555 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1556 						struct discard_policy *dpolicy,
1557 						block_t start, block_t end)
1558 {
1559 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1560 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1561 					&(dcc->fstrim_list) : &(dcc->wait_list);
1562 	struct discard_cmd *dc, *tmp;
1563 	bool need_wait;
1564 	unsigned int trimmed = 0;
1565 
1566 next:
1567 	need_wait = false;
1568 
1569 	mutex_lock(&dcc->cmd_lock);
1570 	list_for_each_entry_safe(dc, tmp, wait_list, list) {
1571 		if (dc->lstart + dc->len <= start || end <= dc->lstart)
1572 			continue;
1573 		if (dc->len < dpolicy->granularity)
1574 			continue;
1575 		if (dc->state == D_DONE && !dc->ref) {
1576 			wait_for_completion_io(&dc->wait);
1577 			if (!dc->error)
1578 				trimmed += dc->len;
1579 			__remove_discard_cmd(sbi, dc);
1580 		} else {
1581 			dc->ref++;
1582 			need_wait = true;
1583 			break;
1584 		}
1585 	}
1586 	mutex_unlock(&dcc->cmd_lock);
1587 
1588 	if (need_wait) {
1589 		trimmed += __wait_one_discard_bio(sbi, dc);
1590 		goto next;
1591 	}
1592 
1593 	return trimmed;
1594 }
1595 
1596 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1597 						struct discard_policy *dpolicy)
1598 {
1599 	struct discard_policy dp;
1600 	unsigned int discard_blks;
1601 
1602 	if (dpolicy)
1603 		return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1604 
1605 	/* wait all */
1606 	__init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1607 	discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1608 	__init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1609 	discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1610 
1611 	return discard_blks;
1612 }
1613 
1614 /* This should be covered by global mutex, &sit_i->sentry_lock */
1615 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1616 {
1617 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1618 	struct discard_cmd *dc;
1619 	bool need_wait = false;
1620 
1621 	mutex_lock(&dcc->cmd_lock);
1622 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1623 							NULL, blkaddr);
1624 	if (dc) {
1625 		if (dc->state == D_PREP) {
1626 			__punch_discard_cmd(sbi, dc, blkaddr);
1627 		} else {
1628 			dc->ref++;
1629 			need_wait = true;
1630 		}
1631 	}
1632 	mutex_unlock(&dcc->cmd_lock);
1633 
1634 	if (need_wait)
1635 		__wait_one_discard_bio(sbi, dc);
1636 }
1637 
1638 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1639 {
1640 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1641 
1642 	if (dcc && dcc->f2fs_issue_discard) {
1643 		struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1644 
1645 		dcc->f2fs_issue_discard = NULL;
1646 		kthread_stop(discard_thread);
1647 	}
1648 }
1649 
1650 /* This comes from f2fs_put_super */
1651 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1652 {
1653 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1654 	struct discard_policy dpolicy;
1655 	bool dropped;
1656 
1657 	__init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1658 					dcc->discard_granularity);
1659 	dpolicy.timeout = UMOUNT_DISCARD_TIMEOUT;
1660 	__issue_discard_cmd(sbi, &dpolicy);
1661 	dropped = __drop_discard_cmd(sbi);
1662 
1663 	/* just to make sure there is no pending discard commands */
1664 	__wait_all_discard_cmd(sbi, NULL);
1665 
1666 	f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1667 	return dropped;
1668 }
1669 
1670 static int issue_discard_thread(void *data)
1671 {
1672 	struct f2fs_sb_info *sbi = data;
1673 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1674 	wait_queue_head_t *q = &dcc->discard_wait_queue;
1675 	struct discard_policy dpolicy;
1676 	unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1677 	int issued;
1678 
1679 	set_freezable();
1680 
1681 	do {
1682 		__init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1683 					dcc->discard_granularity);
1684 
1685 		wait_event_interruptible_timeout(*q,
1686 				kthread_should_stop() || freezing(current) ||
1687 				dcc->discard_wake,
1688 				msecs_to_jiffies(wait_ms));
1689 
1690 		if (dcc->discard_wake)
1691 			dcc->discard_wake = 0;
1692 
1693 		/* clean up pending candidates before going to sleep */
1694 		if (atomic_read(&dcc->queued_discard))
1695 			__wait_all_discard_cmd(sbi, NULL);
1696 
1697 		if (try_to_freeze())
1698 			continue;
1699 		if (f2fs_readonly(sbi->sb))
1700 			continue;
1701 		if (kthread_should_stop())
1702 			return 0;
1703 		if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1704 			wait_ms = dpolicy.max_interval;
1705 			continue;
1706 		}
1707 
1708 		if (sbi->gc_mode == GC_URGENT)
1709 			__init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1710 
1711 		sb_start_intwrite(sbi->sb);
1712 
1713 		issued = __issue_discard_cmd(sbi, &dpolicy);
1714 		if (issued > 0) {
1715 			__wait_all_discard_cmd(sbi, &dpolicy);
1716 			wait_ms = dpolicy.min_interval;
1717 		} else if (issued == -1){
1718 			wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1719 			if (!wait_ms)
1720 				wait_ms = dpolicy.mid_interval;
1721 		} else {
1722 			wait_ms = dpolicy.max_interval;
1723 		}
1724 
1725 		sb_end_intwrite(sbi->sb);
1726 
1727 	} while (!kthread_should_stop());
1728 	return 0;
1729 }
1730 
1731 #ifdef CONFIG_BLK_DEV_ZONED
1732 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1733 		struct block_device *bdev, block_t blkstart, block_t blklen)
1734 {
1735 	sector_t sector, nr_sects;
1736 	block_t lblkstart = blkstart;
1737 	int devi = 0;
1738 
1739 	if (f2fs_is_multi_device(sbi)) {
1740 		devi = f2fs_target_device_index(sbi, blkstart);
1741 		if (blkstart < FDEV(devi).start_blk ||
1742 		    blkstart > FDEV(devi).end_blk) {
1743 			f2fs_msg(sbi->sb, KERN_ERR, "Invalid block %x",
1744 				 blkstart);
1745 			return -EIO;
1746 		}
1747 		blkstart -= FDEV(devi).start_blk;
1748 	}
1749 
1750 	/* For sequential zones, reset the zone write pointer */
1751 	if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1752 		sector = SECTOR_FROM_BLOCK(blkstart);
1753 		nr_sects = SECTOR_FROM_BLOCK(blklen);
1754 
1755 		if (sector & (bdev_zone_sectors(bdev) - 1) ||
1756 				nr_sects != bdev_zone_sectors(bdev)) {
1757 			f2fs_msg(sbi->sb, KERN_ERR,
1758 				"(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1759 				devi, sbi->s_ndevs ? FDEV(devi).path: "",
1760 				blkstart, blklen);
1761 			return -EIO;
1762 		}
1763 		trace_f2fs_issue_reset_zone(bdev, blkstart);
1764 		return blkdev_reset_zones(bdev, sector, nr_sects, GFP_NOFS);
1765 	}
1766 
1767 	/* For conventional zones, use regular discard if supported */
1768 	return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1769 }
1770 #endif
1771 
1772 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1773 		struct block_device *bdev, block_t blkstart, block_t blklen)
1774 {
1775 #ifdef CONFIG_BLK_DEV_ZONED
1776 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1777 		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1778 #endif
1779 	return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1780 }
1781 
1782 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1783 				block_t blkstart, block_t blklen)
1784 {
1785 	sector_t start = blkstart, len = 0;
1786 	struct block_device *bdev;
1787 	struct seg_entry *se;
1788 	unsigned int offset;
1789 	block_t i;
1790 	int err = 0;
1791 
1792 	bdev = f2fs_target_device(sbi, blkstart, NULL);
1793 
1794 	for (i = blkstart; i < blkstart + blklen; i++, len++) {
1795 		if (i != start) {
1796 			struct block_device *bdev2 =
1797 				f2fs_target_device(sbi, i, NULL);
1798 
1799 			if (bdev2 != bdev) {
1800 				err = __issue_discard_async(sbi, bdev,
1801 						start, len);
1802 				if (err)
1803 					return err;
1804 				bdev = bdev2;
1805 				start = i;
1806 				len = 0;
1807 			}
1808 		}
1809 
1810 		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1811 		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1812 
1813 		if (!f2fs_test_and_set_bit(offset, se->discard_map))
1814 			sbi->discard_blks--;
1815 	}
1816 
1817 	if (len)
1818 		err = __issue_discard_async(sbi, bdev, start, len);
1819 	return err;
1820 }
1821 
1822 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1823 							bool check_only)
1824 {
1825 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1826 	int max_blocks = sbi->blocks_per_seg;
1827 	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1828 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1829 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1830 	unsigned long *discard_map = (unsigned long *)se->discard_map;
1831 	unsigned long *dmap = SIT_I(sbi)->tmp_map;
1832 	unsigned int start = 0, end = -1;
1833 	bool force = (cpc->reason & CP_DISCARD);
1834 	struct discard_entry *de = NULL;
1835 	struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1836 	int i;
1837 
1838 	if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1839 		return false;
1840 
1841 	if (!force) {
1842 		if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1843 			SM_I(sbi)->dcc_info->nr_discards >=
1844 				SM_I(sbi)->dcc_info->max_discards)
1845 			return false;
1846 	}
1847 
1848 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1849 	for (i = 0; i < entries; i++)
1850 		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1851 				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1852 
1853 	while (force || SM_I(sbi)->dcc_info->nr_discards <=
1854 				SM_I(sbi)->dcc_info->max_discards) {
1855 		start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1856 		if (start >= max_blocks)
1857 			break;
1858 
1859 		end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1860 		if (force && start && end != max_blocks
1861 					&& (end - start) < cpc->trim_minlen)
1862 			continue;
1863 
1864 		if (check_only)
1865 			return true;
1866 
1867 		if (!de) {
1868 			de = f2fs_kmem_cache_alloc(discard_entry_slab,
1869 								GFP_F2FS_ZERO);
1870 			de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1871 			list_add_tail(&de->list, head);
1872 		}
1873 
1874 		for (i = start; i < end; i++)
1875 			__set_bit_le(i, (void *)de->discard_map);
1876 
1877 		SM_I(sbi)->dcc_info->nr_discards += end - start;
1878 	}
1879 	return false;
1880 }
1881 
1882 static void release_discard_addr(struct discard_entry *entry)
1883 {
1884 	list_del(&entry->list);
1885 	kmem_cache_free(discard_entry_slab, entry);
1886 }
1887 
1888 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1889 {
1890 	struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1891 	struct discard_entry *entry, *this;
1892 
1893 	/* drop caches */
1894 	list_for_each_entry_safe(entry, this, head, list)
1895 		release_discard_addr(entry);
1896 }
1897 
1898 /*
1899  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1900  */
1901 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1902 {
1903 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1904 	unsigned int segno;
1905 
1906 	mutex_lock(&dirty_i->seglist_lock);
1907 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1908 		__set_test_and_free(sbi, segno);
1909 	mutex_unlock(&dirty_i->seglist_lock);
1910 }
1911 
1912 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1913 						struct cp_control *cpc)
1914 {
1915 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1916 	struct list_head *head = &dcc->entry_list;
1917 	struct discard_entry *entry, *this;
1918 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1919 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1920 	unsigned int start = 0, end = -1;
1921 	unsigned int secno, start_segno;
1922 	bool force = (cpc->reason & CP_DISCARD);
1923 	bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
1924 
1925 	mutex_lock(&dirty_i->seglist_lock);
1926 
1927 	while (1) {
1928 		int i;
1929 
1930 		if (need_align && end != -1)
1931 			end--;
1932 		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1933 		if (start >= MAIN_SEGS(sbi))
1934 			break;
1935 		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1936 								start + 1);
1937 
1938 		if (need_align) {
1939 			start = rounddown(start, sbi->segs_per_sec);
1940 			end = roundup(end, sbi->segs_per_sec);
1941 		}
1942 
1943 		for (i = start; i < end; i++) {
1944 			if (test_and_clear_bit(i, prefree_map))
1945 				dirty_i->nr_dirty[PRE]--;
1946 		}
1947 
1948 		if (!f2fs_realtime_discard_enable(sbi))
1949 			continue;
1950 
1951 		if (force && start >= cpc->trim_start &&
1952 					(end - 1) <= cpc->trim_end)
1953 				continue;
1954 
1955 		if (!test_opt(sbi, LFS) || !__is_large_section(sbi)) {
1956 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1957 				(end - start) << sbi->log_blocks_per_seg);
1958 			continue;
1959 		}
1960 next:
1961 		secno = GET_SEC_FROM_SEG(sbi, start);
1962 		start_segno = GET_SEG_FROM_SEC(sbi, secno);
1963 		if (!IS_CURSEC(sbi, secno) &&
1964 			!get_valid_blocks(sbi, start, true))
1965 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1966 				sbi->segs_per_sec << sbi->log_blocks_per_seg);
1967 
1968 		start = start_segno + sbi->segs_per_sec;
1969 		if (start < end)
1970 			goto next;
1971 		else
1972 			end = start - 1;
1973 	}
1974 	mutex_unlock(&dirty_i->seglist_lock);
1975 
1976 	/* send small discards */
1977 	list_for_each_entry_safe(entry, this, head, list) {
1978 		unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1979 		bool is_valid = test_bit_le(0, entry->discard_map);
1980 
1981 find_next:
1982 		if (is_valid) {
1983 			next_pos = find_next_zero_bit_le(entry->discard_map,
1984 					sbi->blocks_per_seg, cur_pos);
1985 			len = next_pos - cur_pos;
1986 
1987 			if (f2fs_sb_has_blkzoned(sbi) ||
1988 			    (force && len < cpc->trim_minlen))
1989 				goto skip;
1990 
1991 			f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1992 									len);
1993 			total_len += len;
1994 		} else {
1995 			next_pos = find_next_bit_le(entry->discard_map,
1996 					sbi->blocks_per_seg, cur_pos);
1997 		}
1998 skip:
1999 		cur_pos = next_pos;
2000 		is_valid = !is_valid;
2001 
2002 		if (cur_pos < sbi->blocks_per_seg)
2003 			goto find_next;
2004 
2005 		release_discard_addr(entry);
2006 		dcc->nr_discards -= total_len;
2007 	}
2008 
2009 	wake_up_discard_thread(sbi, false);
2010 }
2011 
2012 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2013 {
2014 	dev_t dev = sbi->sb->s_bdev->bd_dev;
2015 	struct discard_cmd_control *dcc;
2016 	int err = 0, i;
2017 
2018 	if (SM_I(sbi)->dcc_info) {
2019 		dcc = SM_I(sbi)->dcc_info;
2020 		goto init_thread;
2021 	}
2022 
2023 	dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2024 	if (!dcc)
2025 		return -ENOMEM;
2026 
2027 	dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2028 	INIT_LIST_HEAD(&dcc->entry_list);
2029 	for (i = 0; i < MAX_PLIST_NUM; i++)
2030 		INIT_LIST_HEAD(&dcc->pend_list[i]);
2031 	INIT_LIST_HEAD(&dcc->wait_list);
2032 	INIT_LIST_HEAD(&dcc->fstrim_list);
2033 	mutex_init(&dcc->cmd_lock);
2034 	atomic_set(&dcc->issued_discard, 0);
2035 	atomic_set(&dcc->queued_discard, 0);
2036 	atomic_set(&dcc->discard_cmd_cnt, 0);
2037 	dcc->nr_discards = 0;
2038 	dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2039 	dcc->undiscard_blks = 0;
2040 	dcc->next_pos = 0;
2041 	dcc->root = RB_ROOT_CACHED;
2042 	dcc->rbtree_check = false;
2043 
2044 	init_waitqueue_head(&dcc->discard_wait_queue);
2045 	SM_I(sbi)->dcc_info = dcc;
2046 init_thread:
2047 	dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2048 				"f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2049 	if (IS_ERR(dcc->f2fs_issue_discard)) {
2050 		err = PTR_ERR(dcc->f2fs_issue_discard);
2051 		kvfree(dcc);
2052 		SM_I(sbi)->dcc_info = NULL;
2053 		return err;
2054 	}
2055 
2056 	return err;
2057 }
2058 
2059 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2060 {
2061 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2062 
2063 	if (!dcc)
2064 		return;
2065 
2066 	f2fs_stop_discard_thread(sbi);
2067 
2068 	kvfree(dcc);
2069 	SM_I(sbi)->dcc_info = NULL;
2070 }
2071 
2072 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2073 {
2074 	struct sit_info *sit_i = SIT_I(sbi);
2075 
2076 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2077 		sit_i->dirty_sentries++;
2078 		return false;
2079 	}
2080 
2081 	return true;
2082 }
2083 
2084 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2085 					unsigned int segno, int modified)
2086 {
2087 	struct seg_entry *se = get_seg_entry(sbi, segno);
2088 	se->type = type;
2089 	if (modified)
2090 		__mark_sit_entry_dirty(sbi, segno);
2091 }
2092 
2093 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2094 {
2095 	struct seg_entry *se;
2096 	unsigned int segno, offset;
2097 	long int new_vblocks;
2098 	bool exist;
2099 #ifdef CONFIG_F2FS_CHECK_FS
2100 	bool mir_exist;
2101 #endif
2102 
2103 	segno = GET_SEGNO(sbi, blkaddr);
2104 
2105 	se = get_seg_entry(sbi, segno);
2106 	new_vblocks = se->valid_blocks + del;
2107 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2108 
2109 	f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
2110 				(new_vblocks > sbi->blocks_per_seg)));
2111 
2112 	se->valid_blocks = new_vblocks;
2113 	se->mtime = get_mtime(sbi, false);
2114 	if (se->mtime > SIT_I(sbi)->max_mtime)
2115 		SIT_I(sbi)->max_mtime = se->mtime;
2116 
2117 	/* Update valid block bitmap */
2118 	if (del > 0) {
2119 		exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2120 #ifdef CONFIG_F2FS_CHECK_FS
2121 		mir_exist = f2fs_test_and_set_bit(offset,
2122 						se->cur_valid_map_mir);
2123 		if (unlikely(exist != mir_exist)) {
2124 			f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2125 				"when setting bitmap, blk:%u, old bit:%d",
2126 				blkaddr, exist);
2127 			f2fs_bug_on(sbi, 1);
2128 		}
2129 #endif
2130 		if (unlikely(exist)) {
2131 			f2fs_msg(sbi->sb, KERN_ERR,
2132 				"Bitmap was wrongly set, blk:%u", blkaddr);
2133 			f2fs_bug_on(sbi, 1);
2134 			se->valid_blocks--;
2135 			del = 0;
2136 		}
2137 
2138 		if (!f2fs_test_and_set_bit(offset, se->discard_map))
2139 			sbi->discard_blks--;
2140 
2141 		/* don't overwrite by SSR to keep node chain */
2142 		if (IS_NODESEG(se->type) &&
2143 				!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2144 			if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2145 				se->ckpt_valid_blocks++;
2146 		}
2147 	} else {
2148 		exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2149 #ifdef CONFIG_F2FS_CHECK_FS
2150 		mir_exist = f2fs_test_and_clear_bit(offset,
2151 						se->cur_valid_map_mir);
2152 		if (unlikely(exist != mir_exist)) {
2153 			f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2154 				"when clearing bitmap, blk:%u, old bit:%d",
2155 				blkaddr, exist);
2156 			f2fs_bug_on(sbi, 1);
2157 		}
2158 #endif
2159 		if (unlikely(!exist)) {
2160 			f2fs_msg(sbi->sb, KERN_ERR,
2161 				"Bitmap was wrongly cleared, blk:%u", blkaddr);
2162 			f2fs_bug_on(sbi, 1);
2163 			se->valid_blocks++;
2164 			del = 0;
2165 		} else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2166 			/*
2167 			 * If checkpoints are off, we must not reuse data that
2168 			 * was used in the previous checkpoint. If it was used
2169 			 * before, we must track that to know how much space we
2170 			 * really have.
2171 			 */
2172 			if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2173 				spin_lock(&sbi->stat_lock);
2174 				sbi->unusable_block_count++;
2175 				spin_unlock(&sbi->stat_lock);
2176 			}
2177 		}
2178 
2179 		if (f2fs_test_and_clear_bit(offset, se->discard_map))
2180 			sbi->discard_blks++;
2181 	}
2182 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2183 		se->ckpt_valid_blocks += del;
2184 
2185 	__mark_sit_entry_dirty(sbi, segno);
2186 
2187 	/* update total number of valid blocks to be written in ckpt area */
2188 	SIT_I(sbi)->written_valid_blocks += del;
2189 
2190 	if (__is_large_section(sbi))
2191 		get_sec_entry(sbi, segno)->valid_blocks += del;
2192 }
2193 
2194 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2195 {
2196 	unsigned int segno = GET_SEGNO(sbi, addr);
2197 	struct sit_info *sit_i = SIT_I(sbi);
2198 
2199 	f2fs_bug_on(sbi, addr == NULL_ADDR);
2200 	if (addr == NEW_ADDR)
2201 		return;
2202 
2203 	invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2204 
2205 	/* add it into sit main buffer */
2206 	down_write(&sit_i->sentry_lock);
2207 
2208 	update_sit_entry(sbi, addr, -1);
2209 
2210 	/* add it into dirty seglist */
2211 	locate_dirty_segment(sbi, segno);
2212 
2213 	up_write(&sit_i->sentry_lock);
2214 }
2215 
2216 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2217 {
2218 	struct sit_info *sit_i = SIT_I(sbi);
2219 	unsigned int segno, offset;
2220 	struct seg_entry *se;
2221 	bool is_cp = false;
2222 
2223 	if (!__is_valid_data_blkaddr(blkaddr))
2224 		return true;
2225 
2226 	down_read(&sit_i->sentry_lock);
2227 
2228 	segno = GET_SEGNO(sbi, blkaddr);
2229 	se = get_seg_entry(sbi, segno);
2230 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2231 
2232 	if (f2fs_test_bit(offset, se->ckpt_valid_map))
2233 		is_cp = true;
2234 
2235 	up_read(&sit_i->sentry_lock);
2236 
2237 	return is_cp;
2238 }
2239 
2240 /*
2241  * This function should be resided under the curseg_mutex lock
2242  */
2243 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2244 					struct f2fs_summary *sum)
2245 {
2246 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2247 	void *addr = curseg->sum_blk;
2248 	addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2249 	memcpy(addr, sum, sizeof(struct f2fs_summary));
2250 }
2251 
2252 /*
2253  * Calculate the number of current summary pages for writing
2254  */
2255 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2256 {
2257 	int valid_sum_count = 0;
2258 	int i, sum_in_page;
2259 
2260 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2261 		if (sbi->ckpt->alloc_type[i] == SSR)
2262 			valid_sum_count += sbi->blocks_per_seg;
2263 		else {
2264 			if (for_ra)
2265 				valid_sum_count += le16_to_cpu(
2266 					F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2267 			else
2268 				valid_sum_count += curseg_blkoff(sbi, i);
2269 		}
2270 	}
2271 
2272 	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2273 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2274 	if (valid_sum_count <= sum_in_page)
2275 		return 1;
2276 	else if ((valid_sum_count - sum_in_page) <=
2277 		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2278 		return 2;
2279 	return 3;
2280 }
2281 
2282 /*
2283  * Caller should put this summary page
2284  */
2285 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2286 {
2287 	return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2288 }
2289 
2290 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2291 					void *src, block_t blk_addr)
2292 {
2293 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2294 
2295 	memcpy(page_address(page), src, PAGE_SIZE);
2296 	set_page_dirty(page);
2297 	f2fs_put_page(page, 1);
2298 }
2299 
2300 static void write_sum_page(struct f2fs_sb_info *sbi,
2301 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
2302 {
2303 	f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2304 }
2305 
2306 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2307 						int type, block_t blk_addr)
2308 {
2309 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2310 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2311 	struct f2fs_summary_block *src = curseg->sum_blk;
2312 	struct f2fs_summary_block *dst;
2313 
2314 	dst = (struct f2fs_summary_block *)page_address(page);
2315 	memset(dst, 0, PAGE_SIZE);
2316 
2317 	mutex_lock(&curseg->curseg_mutex);
2318 
2319 	down_read(&curseg->journal_rwsem);
2320 	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2321 	up_read(&curseg->journal_rwsem);
2322 
2323 	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2324 	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2325 
2326 	mutex_unlock(&curseg->curseg_mutex);
2327 
2328 	set_page_dirty(page);
2329 	f2fs_put_page(page, 1);
2330 }
2331 
2332 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2333 {
2334 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2335 	unsigned int segno = curseg->segno + 1;
2336 	struct free_segmap_info *free_i = FREE_I(sbi);
2337 
2338 	if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2339 		return !test_bit(segno, free_i->free_segmap);
2340 	return 0;
2341 }
2342 
2343 /*
2344  * Find a new segment from the free segments bitmap to right order
2345  * This function should be returned with success, otherwise BUG
2346  */
2347 static void get_new_segment(struct f2fs_sb_info *sbi,
2348 			unsigned int *newseg, bool new_sec, int dir)
2349 {
2350 	struct free_segmap_info *free_i = FREE_I(sbi);
2351 	unsigned int segno, secno, zoneno;
2352 	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2353 	unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2354 	unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2355 	unsigned int left_start = hint;
2356 	bool init = true;
2357 	int go_left = 0;
2358 	int i;
2359 
2360 	spin_lock(&free_i->segmap_lock);
2361 
2362 	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2363 		segno = find_next_zero_bit(free_i->free_segmap,
2364 			GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2365 		if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2366 			goto got_it;
2367 	}
2368 find_other_zone:
2369 	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2370 	if (secno >= MAIN_SECS(sbi)) {
2371 		if (dir == ALLOC_RIGHT) {
2372 			secno = find_next_zero_bit(free_i->free_secmap,
2373 							MAIN_SECS(sbi), 0);
2374 			f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2375 		} else {
2376 			go_left = 1;
2377 			left_start = hint - 1;
2378 		}
2379 	}
2380 	if (go_left == 0)
2381 		goto skip_left;
2382 
2383 	while (test_bit(left_start, free_i->free_secmap)) {
2384 		if (left_start > 0) {
2385 			left_start--;
2386 			continue;
2387 		}
2388 		left_start = find_next_zero_bit(free_i->free_secmap,
2389 							MAIN_SECS(sbi), 0);
2390 		f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2391 		break;
2392 	}
2393 	secno = left_start;
2394 skip_left:
2395 	segno = GET_SEG_FROM_SEC(sbi, secno);
2396 	zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2397 
2398 	/* give up on finding another zone */
2399 	if (!init)
2400 		goto got_it;
2401 	if (sbi->secs_per_zone == 1)
2402 		goto got_it;
2403 	if (zoneno == old_zoneno)
2404 		goto got_it;
2405 	if (dir == ALLOC_LEFT) {
2406 		if (!go_left && zoneno + 1 >= total_zones)
2407 			goto got_it;
2408 		if (go_left && zoneno == 0)
2409 			goto got_it;
2410 	}
2411 	for (i = 0; i < NR_CURSEG_TYPE; i++)
2412 		if (CURSEG_I(sbi, i)->zone == zoneno)
2413 			break;
2414 
2415 	if (i < NR_CURSEG_TYPE) {
2416 		/* zone is in user, try another */
2417 		if (go_left)
2418 			hint = zoneno * sbi->secs_per_zone - 1;
2419 		else if (zoneno + 1 >= total_zones)
2420 			hint = 0;
2421 		else
2422 			hint = (zoneno + 1) * sbi->secs_per_zone;
2423 		init = false;
2424 		goto find_other_zone;
2425 	}
2426 got_it:
2427 	/* set it as dirty segment in free segmap */
2428 	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2429 	__set_inuse(sbi, segno);
2430 	*newseg = segno;
2431 	spin_unlock(&free_i->segmap_lock);
2432 }
2433 
2434 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2435 {
2436 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2437 	struct summary_footer *sum_footer;
2438 
2439 	curseg->segno = curseg->next_segno;
2440 	curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2441 	curseg->next_blkoff = 0;
2442 	curseg->next_segno = NULL_SEGNO;
2443 
2444 	sum_footer = &(curseg->sum_blk->footer);
2445 	memset(sum_footer, 0, sizeof(struct summary_footer));
2446 	if (IS_DATASEG(type))
2447 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2448 	if (IS_NODESEG(type))
2449 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2450 	__set_sit_entry_type(sbi, type, curseg->segno, modified);
2451 }
2452 
2453 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2454 {
2455 	/* if segs_per_sec is large than 1, we need to keep original policy. */
2456 	if (__is_large_section(sbi))
2457 		return CURSEG_I(sbi, type)->segno;
2458 
2459 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2460 		return 0;
2461 
2462 	if (test_opt(sbi, NOHEAP) &&
2463 		(type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2464 		return 0;
2465 
2466 	if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2467 		return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2468 
2469 	/* find segments from 0 to reuse freed segments */
2470 	if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2471 		return 0;
2472 
2473 	return CURSEG_I(sbi, type)->segno;
2474 }
2475 
2476 /*
2477  * Allocate a current working segment.
2478  * This function always allocates a free segment in LFS manner.
2479  */
2480 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2481 {
2482 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2483 	unsigned int segno = curseg->segno;
2484 	int dir = ALLOC_LEFT;
2485 
2486 	write_sum_page(sbi, curseg->sum_blk,
2487 				GET_SUM_BLOCK(sbi, segno));
2488 	if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2489 		dir = ALLOC_RIGHT;
2490 
2491 	if (test_opt(sbi, NOHEAP))
2492 		dir = ALLOC_RIGHT;
2493 
2494 	segno = __get_next_segno(sbi, type);
2495 	get_new_segment(sbi, &segno, new_sec, dir);
2496 	curseg->next_segno = segno;
2497 	reset_curseg(sbi, type, 1);
2498 	curseg->alloc_type = LFS;
2499 }
2500 
2501 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2502 			struct curseg_info *seg, block_t start)
2503 {
2504 	struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2505 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2506 	unsigned long *target_map = SIT_I(sbi)->tmp_map;
2507 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2508 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2509 	int i, pos;
2510 
2511 	for (i = 0; i < entries; i++)
2512 		target_map[i] = ckpt_map[i] | cur_map[i];
2513 
2514 	pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2515 
2516 	seg->next_blkoff = pos;
2517 }
2518 
2519 /*
2520  * If a segment is written by LFS manner, next block offset is just obtained
2521  * by increasing the current block offset. However, if a segment is written by
2522  * SSR manner, next block offset obtained by calling __next_free_blkoff
2523  */
2524 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2525 				struct curseg_info *seg)
2526 {
2527 	if (seg->alloc_type == SSR)
2528 		__next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2529 	else
2530 		seg->next_blkoff++;
2531 }
2532 
2533 /*
2534  * This function always allocates a used segment(from dirty seglist) by SSR
2535  * manner, so it should recover the existing segment information of valid blocks
2536  */
2537 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2538 {
2539 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2540 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2541 	unsigned int new_segno = curseg->next_segno;
2542 	struct f2fs_summary_block *sum_node;
2543 	struct page *sum_page;
2544 
2545 	write_sum_page(sbi, curseg->sum_blk,
2546 				GET_SUM_BLOCK(sbi, curseg->segno));
2547 	__set_test_and_inuse(sbi, new_segno);
2548 
2549 	mutex_lock(&dirty_i->seglist_lock);
2550 	__remove_dirty_segment(sbi, new_segno, PRE);
2551 	__remove_dirty_segment(sbi, new_segno, DIRTY);
2552 	mutex_unlock(&dirty_i->seglist_lock);
2553 
2554 	reset_curseg(sbi, type, 1);
2555 	curseg->alloc_type = SSR;
2556 	__next_free_blkoff(sbi, curseg, 0);
2557 
2558 	sum_page = f2fs_get_sum_page(sbi, new_segno);
2559 	f2fs_bug_on(sbi, IS_ERR(sum_page));
2560 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2561 	memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2562 	f2fs_put_page(sum_page, 1);
2563 }
2564 
2565 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2566 {
2567 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2568 	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2569 	unsigned segno = NULL_SEGNO;
2570 	int i, cnt;
2571 	bool reversed = false;
2572 
2573 	/* f2fs_need_SSR() already forces to do this */
2574 	if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2575 		curseg->next_segno = segno;
2576 		return 1;
2577 	}
2578 
2579 	/* For node segments, let's do SSR more intensively */
2580 	if (IS_NODESEG(type)) {
2581 		if (type >= CURSEG_WARM_NODE) {
2582 			reversed = true;
2583 			i = CURSEG_COLD_NODE;
2584 		} else {
2585 			i = CURSEG_HOT_NODE;
2586 		}
2587 		cnt = NR_CURSEG_NODE_TYPE;
2588 	} else {
2589 		if (type >= CURSEG_WARM_DATA) {
2590 			reversed = true;
2591 			i = CURSEG_COLD_DATA;
2592 		} else {
2593 			i = CURSEG_HOT_DATA;
2594 		}
2595 		cnt = NR_CURSEG_DATA_TYPE;
2596 	}
2597 
2598 	for (; cnt-- > 0; reversed ? i-- : i++) {
2599 		if (i == type)
2600 			continue;
2601 		if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2602 			curseg->next_segno = segno;
2603 			return 1;
2604 		}
2605 	}
2606 
2607 	/* find valid_blocks=0 in dirty list */
2608 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2609 		segno = get_free_segment(sbi);
2610 		if (segno != NULL_SEGNO) {
2611 			curseg->next_segno = segno;
2612 			return 1;
2613 		}
2614 	}
2615 	return 0;
2616 }
2617 
2618 /*
2619  * flush out current segment and replace it with new segment
2620  * This function should be returned with success, otherwise BUG
2621  */
2622 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2623 						int type, bool force)
2624 {
2625 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2626 
2627 	if (force)
2628 		new_curseg(sbi, type, true);
2629 	else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2630 					type == CURSEG_WARM_NODE)
2631 		new_curseg(sbi, type, false);
2632 	else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2633 			likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2634 		new_curseg(sbi, type, false);
2635 	else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2636 		change_curseg(sbi, type);
2637 	else
2638 		new_curseg(sbi, type, false);
2639 
2640 	stat_inc_seg_type(sbi, curseg);
2641 }
2642 
2643 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2644 {
2645 	struct curseg_info *curseg;
2646 	unsigned int old_segno;
2647 	int i;
2648 
2649 	down_write(&SIT_I(sbi)->sentry_lock);
2650 
2651 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2652 		curseg = CURSEG_I(sbi, i);
2653 		old_segno = curseg->segno;
2654 		SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2655 		locate_dirty_segment(sbi, old_segno);
2656 	}
2657 
2658 	up_write(&SIT_I(sbi)->sentry_lock);
2659 }
2660 
2661 static const struct segment_allocation default_salloc_ops = {
2662 	.allocate_segment = allocate_segment_by_default,
2663 };
2664 
2665 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2666 						struct cp_control *cpc)
2667 {
2668 	__u64 trim_start = cpc->trim_start;
2669 	bool has_candidate = false;
2670 
2671 	down_write(&SIT_I(sbi)->sentry_lock);
2672 	for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2673 		if (add_discard_addrs(sbi, cpc, true)) {
2674 			has_candidate = true;
2675 			break;
2676 		}
2677 	}
2678 	up_write(&SIT_I(sbi)->sentry_lock);
2679 
2680 	cpc->trim_start = trim_start;
2681 	return has_candidate;
2682 }
2683 
2684 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2685 					struct discard_policy *dpolicy,
2686 					unsigned int start, unsigned int end)
2687 {
2688 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2689 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2690 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
2691 	struct discard_cmd *dc;
2692 	struct blk_plug plug;
2693 	int issued;
2694 	unsigned int trimmed = 0;
2695 
2696 next:
2697 	issued = 0;
2698 
2699 	mutex_lock(&dcc->cmd_lock);
2700 	if (unlikely(dcc->rbtree_check))
2701 		f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2702 								&dcc->root));
2703 
2704 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2705 					NULL, start,
2706 					(struct rb_entry **)&prev_dc,
2707 					(struct rb_entry **)&next_dc,
2708 					&insert_p, &insert_parent, true, NULL);
2709 	if (!dc)
2710 		dc = next_dc;
2711 
2712 	blk_start_plug(&plug);
2713 
2714 	while (dc && dc->lstart <= end) {
2715 		struct rb_node *node;
2716 		int err = 0;
2717 
2718 		if (dc->len < dpolicy->granularity)
2719 			goto skip;
2720 
2721 		if (dc->state != D_PREP) {
2722 			list_move_tail(&dc->list, &dcc->fstrim_list);
2723 			goto skip;
2724 		}
2725 
2726 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2727 
2728 		if (issued >= dpolicy->max_requests) {
2729 			start = dc->lstart + dc->len;
2730 
2731 			if (err)
2732 				__remove_discard_cmd(sbi, dc);
2733 
2734 			blk_finish_plug(&plug);
2735 			mutex_unlock(&dcc->cmd_lock);
2736 			trimmed += __wait_all_discard_cmd(sbi, NULL);
2737 			congestion_wait(BLK_RW_ASYNC, HZ/50);
2738 			goto next;
2739 		}
2740 skip:
2741 		node = rb_next(&dc->rb_node);
2742 		if (err)
2743 			__remove_discard_cmd(sbi, dc);
2744 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2745 
2746 		if (fatal_signal_pending(current))
2747 			break;
2748 	}
2749 
2750 	blk_finish_plug(&plug);
2751 	mutex_unlock(&dcc->cmd_lock);
2752 
2753 	return trimmed;
2754 }
2755 
2756 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2757 {
2758 	__u64 start = F2FS_BYTES_TO_BLK(range->start);
2759 	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2760 	unsigned int start_segno, end_segno;
2761 	block_t start_block, end_block;
2762 	struct cp_control cpc;
2763 	struct discard_policy dpolicy;
2764 	unsigned long long trimmed = 0;
2765 	int err = 0;
2766 	bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
2767 
2768 	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2769 		return -EINVAL;
2770 
2771 	if (end < MAIN_BLKADDR(sbi))
2772 		goto out;
2773 
2774 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2775 		f2fs_msg(sbi->sb, KERN_WARNING,
2776 			"Found FS corruption, run fsck to fix.");
2777 		return -EIO;
2778 	}
2779 
2780 	/* start/end segment number in main_area */
2781 	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2782 	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2783 						GET_SEGNO(sbi, end);
2784 	if (need_align) {
2785 		start_segno = rounddown(start_segno, sbi->segs_per_sec);
2786 		end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2787 	}
2788 
2789 	cpc.reason = CP_DISCARD;
2790 	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2791 	cpc.trim_start = start_segno;
2792 	cpc.trim_end = end_segno;
2793 
2794 	if (sbi->discard_blks == 0)
2795 		goto out;
2796 
2797 	mutex_lock(&sbi->gc_mutex);
2798 	err = f2fs_write_checkpoint(sbi, &cpc);
2799 	mutex_unlock(&sbi->gc_mutex);
2800 	if (err)
2801 		goto out;
2802 
2803 	/*
2804 	 * We filed discard candidates, but actually we don't need to wait for
2805 	 * all of them, since they'll be issued in idle time along with runtime
2806 	 * discard option. User configuration looks like using runtime discard
2807 	 * or periodic fstrim instead of it.
2808 	 */
2809 	if (f2fs_realtime_discard_enable(sbi))
2810 		goto out;
2811 
2812 	start_block = START_BLOCK(sbi, start_segno);
2813 	end_block = START_BLOCK(sbi, end_segno + 1);
2814 
2815 	__init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2816 	trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2817 					start_block, end_block);
2818 
2819 	trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2820 					start_block, end_block);
2821 out:
2822 	if (!err)
2823 		range->len = F2FS_BLK_TO_BYTES(trimmed);
2824 	return err;
2825 }
2826 
2827 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2828 {
2829 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2830 	if (curseg->next_blkoff < sbi->blocks_per_seg)
2831 		return true;
2832 	return false;
2833 }
2834 
2835 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2836 {
2837 	switch (hint) {
2838 	case WRITE_LIFE_SHORT:
2839 		return CURSEG_HOT_DATA;
2840 	case WRITE_LIFE_EXTREME:
2841 		return CURSEG_COLD_DATA;
2842 	default:
2843 		return CURSEG_WARM_DATA;
2844 	}
2845 }
2846 
2847 /* This returns write hints for each segment type. This hints will be
2848  * passed down to block layer. There are mapping tables which depend on
2849  * the mount option 'whint_mode'.
2850  *
2851  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2852  *
2853  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2854  *
2855  * User                  F2FS                     Block
2856  * ----                  ----                     -----
2857  *                       META                     WRITE_LIFE_NOT_SET
2858  *                       HOT_NODE                 "
2859  *                       WARM_NODE                "
2860  *                       COLD_NODE                "
2861  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2862  * extension list        "                        "
2863  *
2864  * -- buffered io
2865  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2866  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2867  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2868  * WRITE_LIFE_NONE       "                        "
2869  * WRITE_LIFE_MEDIUM     "                        "
2870  * WRITE_LIFE_LONG       "                        "
2871  *
2872  * -- direct io
2873  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2874  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2875  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2876  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2877  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2878  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2879  *
2880  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2881  *
2882  * User                  F2FS                     Block
2883  * ----                  ----                     -----
2884  *                       META                     WRITE_LIFE_MEDIUM;
2885  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2886  *                       WARM_NODE                "
2887  *                       COLD_NODE                WRITE_LIFE_NONE
2888  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2889  * extension list        "                        "
2890  *
2891  * -- buffered io
2892  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2893  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2894  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2895  * WRITE_LIFE_NONE       "                        "
2896  * WRITE_LIFE_MEDIUM     "                        "
2897  * WRITE_LIFE_LONG       "                        "
2898  *
2899  * -- direct io
2900  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2901  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2902  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2903  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2904  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2905  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2906  */
2907 
2908 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2909 				enum page_type type, enum temp_type temp)
2910 {
2911 	if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2912 		if (type == DATA) {
2913 			if (temp == WARM)
2914 				return WRITE_LIFE_NOT_SET;
2915 			else if (temp == HOT)
2916 				return WRITE_LIFE_SHORT;
2917 			else if (temp == COLD)
2918 				return WRITE_LIFE_EXTREME;
2919 		} else {
2920 			return WRITE_LIFE_NOT_SET;
2921 		}
2922 	} else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2923 		if (type == DATA) {
2924 			if (temp == WARM)
2925 				return WRITE_LIFE_LONG;
2926 			else if (temp == HOT)
2927 				return WRITE_LIFE_SHORT;
2928 			else if (temp == COLD)
2929 				return WRITE_LIFE_EXTREME;
2930 		} else if (type == NODE) {
2931 			if (temp == WARM || temp == HOT)
2932 				return WRITE_LIFE_NOT_SET;
2933 			else if (temp == COLD)
2934 				return WRITE_LIFE_NONE;
2935 		} else if (type == META) {
2936 			return WRITE_LIFE_MEDIUM;
2937 		}
2938 	}
2939 	return WRITE_LIFE_NOT_SET;
2940 }
2941 
2942 static int __get_segment_type_2(struct f2fs_io_info *fio)
2943 {
2944 	if (fio->type == DATA)
2945 		return CURSEG_HOT_DATA;
2946 	else
2947 		return CURSEG_HOT_NODE;
2948 }
2949 
2950 static int __get_segment_type_4(struct f2fs_io_info *fio)
2951 {
2952 	if (fio->type == DATA) {
2953 		struct inode *inode = fio->page->mapping->host;
2954 
2955 		if (S_ISDIR(inode->i_mode))
2956 			return CURSEG_HOT_DATA;
2957 		else
2958 			return CURSEG_COLD_DATA;
2959 	} else {
2960 		if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2961 			return CURSEG_WARM_NODE;
2962 		else
2963 			return CURSEG_COLD_NODE;
2964 	}
2965 }
2966 
2967 static int __get_segment_type_6(struct f2fs_io_info *fio)
2968 {
2969 	if (fio->type == DATA) {
2970 		struct inode *inode = fio->page->mapping->host;
2971 
2972 		if (is_cold_data(fio->page) || file_is_cold(inode))
2973 			return CURSEG_COLD_DATA;
2974 		if (file_is_hot(inode) ||
2975 				is_inode_flag_set(inode, FI_HOT_DATA) ||
2976 				f2fs_is_atomic_file(inode) ||
2977 				f2fs_is_volatile_file(inode))
2978 			return CURSEG_HOT_DATA;
2979 		return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
2980 	} else {
2981 		if (IS_DNODE(fio->page))
2982 			return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2983 						CURSEG_HOT_NODE;
2984 		return CURSEG_COLD_NODE;
2985 	}
2986 }
2987 
2988 static int __get_segment_type(struct f2fs_io_info *fio)
2989 {
2990 	int type = 0;
2991 
2992 	switch (F2FS_OPTION(fio->sbi).active_logs) {
2993 	case 2:
2994 		type = __get_segment_type_2(fio);
2995 		break;
2996 	case 4:
2997 		type = __get_segment_type_4(fio);
2998 		break;
2999 	case 6:
3000 		type = __get_segment_type_6(fio);
3001 		break;
3002 	default:
3003 		f2fs_bug_on(fio->sbi, true);
3004 	}
3005 
3006 	if (IS_HOT(type))
3007 		fio->temp = HOT;
3008 	else if (IS_WARM(type))
3009 		fio->temp = WARM;
3010 	else
3011 		fio->temp = COLD;
3012 	return type;
3013 }
3014 
3015 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3016 		block_t old_blkaddr, block_t *new_blkaddr,
3017 		struct f2fs_summary *sum, int type,
3018 		struct f2fs_io_info *fio, bool add_list)
3019 {
3020 	struct sit_info *sit_i = SIT_I(sbi);
3021 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3022 
3023 	down_read(&SM_I(sbi)->curseg_lock);
3024 
3025 	mutex_lock(&curseg->curseg_mutex);
3026 	down_write(&sit_i->sentry_lock);
3027 
3028 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3029 
3030 	f2fs_wait_discard_bio(sbi, *new_blkaddr);
3031 
3032 	/*
3033 	 * __add_sum_entry should be resided under the curseg_mutex
3034 	 * because, this function updates a summary entry in the
3035 	 * current summary block.
3036 	 */
3037 	__add_sum_entry(sbi, type, sum);
3038 
3039 	__refresh_next_blkoff(sbi, curseg);
3040 
3041 	stat_inc_block_count(sbi, curseg);
3042 
3043 	/*
3044 	 * SIT information should be updated before segment allocation,
3045 	 * since SSR needs latest valid block information.
3046 	 */
3047 	update_sit_entry(sbi, *new_blkaddr, 1);
3048 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3049 		update_sit_entry(sbi, old_blkaddr, -1);
3050 
3051 	if (!__has_curseg_space(sbi, type))
3052 		sit_i->s_ops->allocate_segment(sbi, type, false);
3053 
3054 	/*
3055 	 * segment dirty status should be updated after segment allocation,
3056 	 * so we just need to update status only one time after previous
3057 	 * segment being closed.
3058 	 */
3059 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3060 	locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3061 
3062 	up_write(&sit_i->sentry_lock);
3063 
3064 	if (page && IS_NODESEG(type)) {
3065 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3066 
3067 		f2fs_inode_chksum_set(sbi, page);
3068 	}
3069 
3070 	if (add_list) {
3071 		struct f2fs_bio_info *io;
3072 
3073 		INIT_LIST_HEAD(&fio->list);
3074 		fio->in_list = true;
3075 		fio->retry = false;
3076 		io = sbi->write_io[fio->type] + fio->temp;
3077 		spin_lock(&io->io_lock);
3078 		list_add_tail(&fio->list, &io->io_list);
3079 		spin_unlock(&io->io_lock);
3080 	}
3081 
3082 	mutex_unlock(&curseg->curseg_mutex);
3083 
3084 	up_read(&SM_I(sbi)->curseg_lock);
3085 }
3086 
3087 static void update_device_state(struct f2fs_io_info *fio)
3088 {
3089 	struct f2fs_sb_info *sbi = fio->sbi;
3090 	unsigned int devidx;
3091 
3092 	if (!f2fs_is_multi_device(sbi))
3093 		return;
3094 
3095 	devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3096 
3097 	/* update device state for fsync */
3098 	f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3099 
3100 	/* update device state for checkpoint */
3101 	if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3102 		spin_lock(&sbi->dev_lock);
3103 		f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3104 		spin_unlock(&sbi->dev_lock);
3105 	}
3106 }
3107 
3108 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3109 {
3110 	int type = __get_segment_type(fio);
3111 	bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
3112 
3113 	if (keep_order)
3114 		down_read(&fio->sbi->io_order_lock);
3115 reallocate:
3116 	f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3117 			&fio->new_blkaddr, sum, type, fio, true);
3118 	if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3119 		invalidate_mapping_pages(META_MAPPING(fio->sbi),
3120 					fio->old_blkaddr, fio->old_blkaddr);
3121 
3122 	/* writeout dirty page into bdev */
3123 	f2fs_submit_page_write(fio);
3124 	if (fio->retry) {
3125 		fio->old_blkaddr = fio->new_blkaddr;
3126 		goto reallocate;
3127 	}
3128 
3129 	update_device_state(fio);
3130 
3131 	if (keep_order)
3132 		up_read(&fio->sbi->io_order_lock);
3133 }
3134 
3135 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3136 					enum iostat_type io_type)
3137 {
3138 	struct f2fs_io_info fio = {
3139 		.sbi = sbi,
3140 		.type = META,
3141 		.temp = HOT,
3142 		.op = REQ_OP_WRITE,
3143 		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3144 		.old_blkaddr = page->index,
3145 		.new_blkaddr = page->index,
3146 		.page = page,
3147 		.encrypted_page = NULL,
3148 		.in_list = false,
3149 	};
3150 
3151 	if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3152 		fio.op_flags &= ~REQ_META;
3153 
3154 	set_page_writeback(page);
3155 	ClearPageError(page);
3156 	f2fs_submit_page_write(&fio);
3157 
3158 	stat_inc_meta_count(sbi, page->index);
3159 	f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3160 }
3161 
3162 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3163 {
3164 	struct f2fs_summary sum;
3165 
3166 	set_summary(&sum, nid, 0, 0);
3167 	do_write_page(&sum, fio);
3168 
3169 	f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3170 }
3171 
3172 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3173 					struct f2fs_io_info *fio)
3174 {
3175 	struct f2fs_sb_info *sbi = fio->sbi;
3176 	struct f2fs_summary sum;
3177 
3178 	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3179 	set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3180 	do_write_page(&sum, fio);
3181 	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3182 
3183 	f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3184 }
3185 
3186 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3187 {
3188 	int err;
3189 	struct f2fs_sb_info *sbi = fio->sbi;
3190 	unsigned int segno;
3191 
3192 	fio->new_blkaddr = fio->old_blkaddr;
3193 	/* i/o temperature is needed for passing down write hints */
3194 	__get_segment_type(fio);
3195 
3196 	segno = GET_SEGNO(sbi, fio->new_blkaddr);
3197 
3198 	if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3199 		set_sbi_flag(sbi, SBI_NEED_FSCK);
3200 		return -EFAULT;
3201 	}
3202 
3203 	stat_inc_inplace_blocks(fio->sbi);
3204 
3205 	err = f2fs_submit_page_bio(fio);
3206 	if (!err) {
3207 		update_device_state(fio);
3208 		f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3209 	}
3210 
3211 	return err;
3212 }
3213 
3214 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3215 						unsigned int segno)
3216 {
3217 	int i;
3218 
3219 	for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3220 		if (CURSEG_I(sbi, i)->segno == segno)
3221 			break;
3222 	}
3223 	return i;
3224 }
3225 
3226 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3227 				block_t old_blkaddr, block_t new_blkaddr,
3228 				bool recover_curseg, bool recover_newaddr)
3229 {
3230 	struct sit_info *sit_i = SIT_I(sbi);
3231 	struct curseg_info *curseg;
3232 	unsigned int segno, old_cursegno;
3233 	struct seg_entry *se;
3234 	int type;
3235 	unsigned short old_blkoff;
3236 
3237 	segno = GET_SEGNO(sbi, new_blkaddr);
3238 	se = get_seg_entry(sbi, segno);
3239 	type = se->type;
3240 
3241 	down_write(&SM_I(sbi)->curseg_lock);
3242 
3243 	if (!recover_curseg) {
3244 		/* for recovery flow */
3245 		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3246 			if (old_blkaddr == NULL_ADDR)
3247 				type = CURSEG_COLD_DATA;
3248 			else
3249 				type = CURSEG_WARM_DATA;
3250 		}
3251 	} else {
3252 		if (IS_CURSEG(sbi, segno)) {
3253 			/* se->type is volatile as SSR allocation */
3254 			type = __f2fs_get_curseg(sbi, segno);
3255 			f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3256 		} else {
3257 			type = CURSEG_WARM_DATA;
3258 		}
3259 	}
3260 
3261 	f2fs_bug_on(sbi, !IS_DATASEG(type));
3262 	curseg = CURSEG_I(sbi, type);
3263 
3264 	mutex_lock(&curseg->curseg_mutex);
3265 	down_write(&sit_i->sentry_lock);
3266 
3267 	old_cursegno = curseg->segno;
3268 	old_blkoff = curseg->next_blkoff;
3269 
3270 	/* change the current segment */
3271 	if (segno != curseg->segno) {
3272 		curseg->next_segno = segno;
3273 		change_curseg(sbi, type);
3274 	}
3275 
3276 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3277 	__add_sum_entry(sbi, type, sum);
3278 
3279 	if (!recover_curseg || recover_newaddr)
3280 		update_sit_entry(sbi, new_blkaddr, 1);
3281 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3282 		invalidate_mapping_pages(META_MAPPING(sbi),
3283 					old_blkaddr, old_blkaddr);
3284 		update_sit_entry(sbi, old_blkaddr, -1);
3285 	}
3286 
3287 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3288 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3289 
3290 	locate_dirty_segment(sbi, old_cursegno);
3291 
3292 	if (recover_curseg) {
3293 		if (old_cursegno != curseg->segno) {
3294 			curseg->next_segno = old_cursegno;
3295 			change_curseg(sbi, type);
3296 		}
3297 		curseg->next_blkoff = old_blkoff;
3298 	}
3299 
3300 	up_write(&sit_i->sentry_lock);
3301 	mutex_unlock(&curseg->curseg_mutex);
3302 	up_write(&SM_I(sbi)->curseg_lock);
3303 }
3304 
3305 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3306 				block_t old_addr, block_t new_addr,
3307 				unsigned char version, bool recover_curseg,
3308 				bool recover_newaddr)
3309 {
3310 	struct f2fs_summary sum;
3311 
3312 	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3313 
3314 	f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3315 					recover_curseg, recover_newaddr);
3316 
3317 	f2fs_update_data_blkaddr(dn, new_addr);
3318 }
3319 
3320 void f2fs_wait_on_page_writeback(struct page *page,
3321 				enum page_type type, bool ordered, bool locked)
3322 {
3323 	if (PageWriteback(page)) {
3324 		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3325 
3326 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3327 		if (ordered) {
3328 			wait_on_page_writeback(page);
3329 			f2fs_bug_on(sbi, locked && PageWriteback(page));
3330 		} else {
3331 			wait_for_stable_page(page);
3332 		}
3333 	}
3334 }
3335 
3336 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3337 {
3338 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3339 	struct page *cpage;
3340 
3341 	if (!f2fs_post_read_required(inode))
3342 		return;
3343 
3344 	if (!__is_valid_data_blkaddr(blkaddr))
3345 		return;
3346 
3347 	cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3348 	if (cpage) {
3349 		f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3350 		f2fs_put_page(cpage, 1);
3351 	}
3352 }
3353 
3354 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3355 								block_t len)
3356 {
3357 	block_t i;
3358 
3359 	for (i = 0; i < len; i++)
3360 		f2fs_wait_on_block_writeback(inode, blkaddr + i);
3361 }
3362 
3363 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3364 {
3365 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3366 	struct curseg_info *seg_i;
3367 	unsigned char *kaddr;
3368 	struct page *page;
3369 	block_t start;
3370 	int i, j, offset;
3371 
3372 	start = start_sum_block(sbi);
3373 
3374 	page = f2fs_get_meta_page(sbi, start++);
3375 	if (IS_ERR(page))
3376 		return PTR_ERR(page);
3377 	kaddr = (unsigned char *)page_address(page);
3378 
3379 	/* Step 1: restore nat cache */
3380 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3381 	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3382 
3383 	/* Step 2: restore sit cache */
3384 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3385 	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3386 	offset = 2 * SUM_JOURNAL_SIZE;
3387 
3388 	/* Step 3: restore summary entries */
3389 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3390 		unsigned short blk_off;
3391 		unsigned int segno;
3392 
3393 		seg_i = CURSEG_I(sbi, i);
3394 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3395 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3396 		seg_i->next_segno = segno;
3397 		reset_curseg(sbi, i, 0);
3398 		seg_i->alloc_type = ckpt->alloc_type[i];
3399 		seg_i->next_blkoff = blk_off;
3400 
3401 		if (seg_i->alloc_type == SSR)
3402 			blk_off = sbi->blocks_per_seg;
3403 
3404 		for (j = 0; j < blk_off; j++) {
3405 			struct f2fs_summary *s;
3406 			s = (struct f2fs_summary *)(kaddr + offset);
3407 			seg_i->sum_blk->entries[j] = *s;
3408 			offset += SUMMARY_SIZE;
3409 			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3410 						SUM_FOOTER_SIZE)
3411 				continue;
3412 
3413 			f2fs_put_page(page, 1);
3414 			page = NULL;
3415 
3416 			page = f2fs_get_meta_page(sbi, start++);
3417 			if (IS_ERR(page))
3418 				return PTR_ERR(page);
3419 			kaddr = (unsigned char *)page_address(page);
3420 			offset = 0;
3421 		}
3422 	}
3423 	f2fs_put_page(page, 1);
3424 	return 0;
3425 }
3426 
3427 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3428 {
3429 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3430 	struct f2fs_summary_block *sum;
3431 	struct curseg_info *curseg;
3432 	struct page *new;
3433 	unsigned short blk_off;
3434 	unsigned int segno = 0;
3435 	block_t blk_addr = 0;
3436 	int err = 0;
3437 
3438 	/* get segment number and block addr */
3439 	if (IS_DATASEG(type)) {
3440 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3441 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3442 							CURSEG_HOT_DATA]);
3443 		if (__exist_node_summaries(sbi))
3444 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3445 		else
3446 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3447 	} else {
3448 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
3449 							CURSEG_HOT_NODE]);
3450 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3451 							CURSEG_HOT_NODE]);
3452 		if (__exist_node_summaries(sbi))
3453 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3454 							type - CURSEG_HOT_NODE);
3455 		else
3456 			blk_addr = GET_SUM_BLOCK(sbi, segno);
3457 	}
3458 
3459 	new = f2fs_get_meta_page(sbi, blk_addr);
3460 	if (IS_ERR(new))
3461 		return PTR_ERR(new);
3462 	sum = (struct f2fs_summary_block *)page_address(new);
3463 
3464 	if (IS_NODESEG(type)) {
3465 		if (__exist_node_summaries(sbi)) {
3466 			struct f2fs_summary *ns = &sum->entries[0];
3467 			int i;
3468 			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3469 				ns->version = 0;
3470 				ns->ofs_in_node = 0;
3471 			}
3472 		} else {
3473 			err = f2fs_restore_node_summary(sbi, segno, sum);
3474 			if (err)
3475 				goto out;
3476 		}
3477 	}
3478 
3479 	/* set uncompleted segment to curseg */
3480 	curseg = CURSEG_I(sbi, type);
3481 	mutex_lock(&curseg->curseg_mutex);
3482 
3483 	/* update journal info */
3484 	down_write(&curseg->journal_rwsem);
3485 	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3486 	up_write(&curseg->journal_rwsem);
3487 
3488 	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3489 	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3490 	curseg->next_segno = segno;
3491 	reset_curseg(sbi, type, 0);
3492 	curseg->alloc_type = ckpt->alloc_type[type];
3493 	curseg->next_blkoff = blk_off;
3494 	mutex_unlock(&curseg->curseg_mutex);
3495 out:
3496 	f2fs_put_page(new, 1);
3497 	return err;
3498 }
3499 
3500 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3501 {
3502 	struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3503 	struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3504 	int type = CURSEG_HOT_DATA;
3505 	int err;
3506 
3507 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3508 		int npages = f2fs_npages_for_summary_flush(sbi, true);
3509 
3510 		if (npages >= 2)
3511 			f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3512 							META_CP, true);
3513 
3514 		/* restore for compacted data summary */
3515 		err = read_compacted_summaries(sbi);
3516 		if (err)
3517 			return err;
3518 		type = CURSEG_HOT_NODE;
3519 	}
3520 
3521 	if (__exist_node_summaries(sbi))
3522 		f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3523 					NR_CURSEG_TYPE - type, META_CP, true);
3524 
3525 	for (; type <= CURSEG_COLD_NODE; type++) {
3526 		err = read_normal_summaries(sbi, type);
3527 		if (err)
3528 			return err;
3529 	}
3530 
3531 	/* sanity check for summary blocks */
3532 	if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3533 			sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3534 		return -EINVAL;
3535 
3536 	return 0;
3537 }
3538 
3539 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3540 {
3541 	struct page *page;
3542 	unsigned char *kaddr;
3543 	struct f2fs_summary *summary;
3544 	struct curseg_info *seg_i;
3545 	int written_size = 0;
3546 	int i, j;
3547 
3548 	page = f2fs_grab_meta_page(sbi, blkaddr++);
3549 	kaddr = (unsigned char *)page_address(page);
3550 	memset(kaddr, 0, PAGE_SIZE);
3551 
3552 	/* Step 1: write nat cache */
3553 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3554 	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3555 	written_size += SUM_JOURNAL_SIZE;
3556 
3557 	/* Step 2: write sit cache */
3558 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3559 	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3560 	written_size += SUM_JOURNAL_SIZE;
3561 
3562 	/* Step 3: write summary entries */
3563 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3564 		unsigned short blkoff;
3565 		seg_i = CURSEG_I(sbi, i);
3566 		if (sbi->ckpt->alloc_type[i] == SSR)
3567 			blkoff = sbi->blocks_per_seg;
3568 		else
3569 			blkoff = curseg_blkoff(sbi, i);
3570 
3571 		for (j = 0; j < blkoff; j++) {
3572 			if (!page) {
3573 				page = f2fs_grab_meta_page(sbi, blkaddr++);
3574 				kaddr = (unsigned char *)page_address(page);
3575 				memset(kaddr, 0, PAGE_SIZE);
3576 				written_size = 0;
3577 			}
3578 			summary = (struct f2fs_summary *)(kaddr + written_size);
3579 			*summary = seg_i->sum_blk->entries[j];
3580 			written_size += SUMMARY_SIZE;
3581 
3582 			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3583 							SUM_FOOTER_SIZE)
3584 				continue;
3585 
3586 			set_page_dirty(page);
3587 			f2fs_put_page(page, 1);
3588 			page = NULL;
3589 		}
3590 	}
3591 	if (page) {
3592 		set_page_dirty(page);
3593 		f2fs_put_page(page, 1);
3594 	}
3595 }
3596 
3597 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3598 					block_t blkaddr, int type)
3599 {
3600 	int i, end;
3601 	if (IS_DATASEG(type))
3602 		end = type + NR_CURSEG_DATA_TYPE;
3603 	else
3604 		end = type + NR_CURSEG_NODE_TYPE;
3605 
3606 	for (i = type; i < end; i++)
3607 		write_current_sum_page(sbi, i, blkaddr + (i - type));
3608 }
3609 
3610 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3611 {
3612 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3613 		write_compacted_summaries(sbi, start_blk);
3614 	else
3615 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3616 }
3617 
3618 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3619 {
3620 	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3621 }
3622 
3623 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3624 					unsigned int val, int alloc)
3625 {
3626 	int i;
3627 
3628 	if (type == NAT_JOURNAL) {
3629 		for (i = 0; i < nats_in_cursum(journal); i++) {
3630 			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3631 				return i;
3632 		}
3633 		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3634 			return update_nats_in_cursum(journal, 1);
3635 	} else if (type == SIT_JOURNAL) {
3636 		for (i = 0; i < sits_in_cursum(journal); i++)
3637 			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3638 				return i;
3639 		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3640 			return update_sits_in_cursum(journal, 1);
3641 	}
3642 	return -1;
3643 }
3644 
3645 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3646 					unsigned int segno)
3647 {
3648 	return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3649 }
3650 
3651 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3652 					unsigned int start)
3653 {
3654 	struct sit_info *sit_i = SIT_I(sbi);
3655 	struct page *page;
3656 	pgoff_t src_off, dst_off;
3657 
3658 	src_off = current_sit_addr(sbi, start);
3659 	dst_off = next_sit_addr(sbi, src_off);
3660 
3661 	page = f2fs_grab_meta_page(sbi, dst_off);
3662 	seg_info_to_sit_page(sbi, page, start);
3663 
3664 	set_page_dirty(page);
3665 	set_to_next_sit(sit_i, start);
3666 
3667 	return page;
3668 }
3669 
3670 static struct sit_entry_set *grab_sit_entry_set(void)
3671 {
3672 	struct sit_entry_set *ses =
3673 			f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3674 
3675 	ses->entry_cnt = 0;
3676 	INIT_LIST_HEAD(&ses->set_list);
3677 	return ses;
3678 }
3679 
3680 static void release_sit_entry_set(struct sit_entry_set *ses)
3681 {
3682 	list_del(&ses->set_list);
3683 	kmem_cache_free(sit_entry_set_slab, ses);
3684 }
3685 
3686 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3687 						struct list_head *head)
3688 {
3689 	struct sit_entry_set *next = ses;
3690 
3691 	if (list_is_last(&ses->set_list, head))
3692 		return;
3693 
3694 	list_for_each_entry_continue(next, head, set_list)
3695 		if (ses->entry_cnt <= next->entry_cnt)
3696 			break;
3697 
3698 	list_move_tail(&ses->set_list, &next->set_list);
3699 }
3700 
3701 static void add_sit_entry(unsigned int segno, struct list_head *head)
3702 {
3703 	struct sit_entry_set *ses;
3704 	unsigned int start_segno = START_SEGNO(segno);
3705 
3706 	list_for_each_entry(ses, head, set_list) {
3707 		if (ses->start_segno == start_segno) {
3708 			ses->entry_cnt++;
3709 			adjust_sit_entry_set(ses, head);
3710 			return;
3711 		}
3712 	}
3713 
3714 	ses = grab_sit_entry_set();
3715 
3716 	ses->start_segno = start_segno;
3717 	ses->entry_cnt++;
3718 	list_add(&ses->set_list, head);
3719 }
3720 
3721 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3722 {
3723 	struct f2fs_sm_info *sm_info = SM_I(sbi);
3724 	struct list_head *set_list = &sm_info->sit_entry_set;
3725 	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3726 	unsigned int segno;
3727 
3728 	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3729 		add_sit_entry(segno, set_list);
3730 }
3731 
3732 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3733 {
3734 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3735 	struct f2fs_journal *journal = curseg->journal;
3736 	int i;
3737 
3738 	down_write(&curseg->journal_rwsem);
3739 	for (i = 0; i < sits_in_cursum(journal); i++) {
3740 		unsigned int segno;
3741 		bool dirtied;
3742 
3743 		segno = le32_to_cpu(segno_in_journal(journal, i));
3744 		dirtied = __mark_sit_entry_dirty(sbi, segno);
3745 
3746 		if (!dirtied)
3747 			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3748 	}
3749 	update_sits_in_cursum(journal, -i);
3750 	up_write(&curseg->journal_rwsem);
3751 }
3752 
3753 /*
3754  * CP calls this function, which flushes SIT entries including sit_journal,
3755  * and moves prefree segs to free segs.
3756  */
3757 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3758 {
3759 	struct sit_info *sit_i = SIT_I(sbi);
3760 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3761 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3762 	struct f2fs_journal *journal = curseg->journal;
3763 	struct sit_entry_set *ses, *tmp;
3764 	struct list_head *head = &SM_I(sbi)->sit_entry_set;
3765 	bool to_journal = true;
3766 	struct seg_entry *se;
3767 
3768 	down_write(&sit_i->sentry_lock);
3769 
3770 	if (!sit_i->dirty_sentries)
3771 		goto out;
3772 
3773 	/*
3774 	 * add and account sit entries of dirty bitmap in sit entry
3775 	 * set temporarily
3776 	 */
3777 	add_sits_in_set(sbi);
3778 
3779 	/*
3780 	 * if there are no enough space in journal to store dirty sit
3781 	 * entries, remove all entries from journal and add and account
3782 	 * them in sit entry set.
3783 	 */
3784 	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3785 		remove_sits_in_journal(sbi);
3786 
3787 	/*
3788 	 * there are two steps to flush sit entries:
3789 	 * #1, flush sit entries to journal in current cold data summary block.
3790 	 * #2, flush sit entries to sit page.
3791 	 */
3792 	list_for_each_entry_safe(ses, tmp, head, set_list) {
3793 		struct page *page = NULL;
3794 		struct f2fs_sit_block *raw_sit = NULL;
3795 		unsigned int start_segno = ses->start_segno;
3796 		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3797 						(unsigned long)MAIN_SEGS(sbi));
3798 		unsigned int segno = start_segno;
3799 
3800 		if (to_journal &&
3801 			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3802 			to_journal = false;
3803 
3804 		if (to_journal) {
3805 			down_write(&curseg->journal_rwsem);
3806 		} else {
3807 			page = get_next_sit_page(sbi, start_segno);
3808 			raw_sit = page_address(page);
3809 		}
3810 
3811 		/* flush dirty sit entries in region of current sit set */
3812 		for_each_set_bit_from(segno, bitmap, end) {
3813 			int offset, sit_offset;
3814 
3815 			se = get_seg_entry(sbi, segno);
3816 #ifdef CONFIG_F2FS_CHECK_FS
3817 			if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3818 						SIT_VBLOCK_MAP_SIZE))
3819 				f2fs_bug_on(sbi, 1);
3820 #endif
3821 
3822 			/* add discard candidates */
3823 			if (!(cpc->reason & CP_DISCARD)) {
3824 				cpc->trim_start = segno;
3825 				add_discard_addrs(sbi, cpc, false);
3826 			}
3827 
3828 			if (to_journal) {
3829 				offset = f2fs_lookup_journal_in_cursum(journal,
3830 							SIT_JOURNAL, segno, 1);
3831 				f2fs_bug_on(sbi, offset < 0);
3832 				segno_in_journal(journal, offset) =
3833 							cpu_to_le32(segno);
3834 				seg_info_to_raw_sit(se,
3835 					&sit_in_journal(journal, offset));
3836 				check_block_count(sbi, segno,
3837 					&sit_in_journal(journal, offset));
3838 			} else {
3839 				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3840 				seg_info_to_raw_sit(se,
3841 						&raw_sit->entries[sit_offset]);
3842 				check_block_count(sbi, segno,
3843 						&raw_sit->entries[sit_offset]);
3844 			}
3845 
3846 			__clear_bit(segno, bitmap);
3847 			sit_i->dirty_sentries--;
3848 			ses->entry_cnt--;
3849 		}
3850 
3851 		if (to_journal)
3852 			up_write(&curseg->journal_rwsem);
3853 		else
3854 			f2fs_put_page(page, 1);
3855 
3856 		f2fs_bug_on(sbi, ses->entry_cnt);
3857 		release_sit_entry_set(ses);
3858 	}
3859 
3860 	f2fs_bug_on(sbi, !list_empty(head));
3861 	f2fs_bug_on(sbi, sit_i->dirty_sentries);
3862 out:
3863 	if (cpc->reason & CP_DISCARD) {
3864 		__u64 trim_start = cpc->trim_start;
3865 
3866 		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3867 			add_discard_addrs(sbi, cpc, false);
3868 
3869 		cpc->trim_start = trim_start;
3870 	}
3871 	up_write(&sit_i->sentry_lock);
3872 
3873 	set_prefree_as_free_segments(sbi);
3874 }
3875 
3876 static int build_sit_info(struct f2fs_sb_info *sbi)
3877 {
3878 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3879 	struct sit_info *sit_i;
3880 	unsigned int sit_segs, start;
3881 	char *src_bitmap;
3882 	unsigned int bitmap_size;
3883 
3884 	/* allocate memory for SIT information */
3885 	sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3886 	if (!sit_i)
3887 		return -ENOMEM;
3888 
3889 	SM_I(sbi)->sit_info = sit_i;
3890 
3891 	sit_i->sentries =
3892 		f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3893 					      MAIN_SEGS(sbi)),
3894 			      GFP_KERNEL);
3895 	if (!sit_i->sentries)
3896 		return -ENOMEM;
3897 
3898 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3899 	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3900 								GFP_KERNEL);
3901 	if (!sit_i->dirty_sentries_bitmap)
3902 		return -ENOMEM;
3903 
3904 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
3905 		sit_i->sentries[start].cur_valid_map
3906 			= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3907 		sit_i->sentries[start].ckpt_valid_map
3908 			= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3909 		if (!sit_i->sentries[start].cur_valid_map ||
3910 				!sit_i->sentries[start].ckpt_valid_map)
3911 			return -ENOMEM;
3912 
3913 #ifdef CONFIG_F2FS_CHECK_FS
3914 		sit_i->sentries[start].cur_valid_map_mir
3915 			= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3916 		if (!sit_i->sentries[start].cur_valid_map_mir)
3917 			return -ENOMEM;
3918 #endif
3919 
3920 		sit_i->sentries[start].discard_map
3921 			= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3922 							GFP_KERNEL);
3923 		if (!sit_i->sentries[start].discard_map)
3924 			return -ENOMEM;
3925 	}
3926 
3927 	sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3928 	if (!sit_i->tmp_map)
3929 		return -ENOMEM;
3930 
3931 	if (__is_large_section(sbi)) {
3932 		sit_i->sec_entries =
3933 			f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3934 						      MAIN_SECS(sbi)),
3935 				      GFP_KERNEL);
3936 		if (!sit_i->sec_entries)
3937 			return -ENOMEM;
3938 	}
3939 
3940 	/* get information related with SIT */
3941 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3942 
3943 	/* setup SIT bitmap from ckeckpoint pack */
3944 	bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3945 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3946 
3947 	sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3948 	if (!sit_i->sit_bitmap)
3949 		return -ENOMEM;
3950 
3951 #ifdef CONFIG_F2FS_CHECK_FS
3952 	sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3953 	if (!sit_i->sit_bitmap_mir)
3954 		return -ENOMEM;
3955 #endif
3956 
3957 	/* init SIT information */
3958 	sit_i->s_ops = &default_salloc_ops;
3959 
3960 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3961 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3962 	sit_i->written_valid_blocks = 0;
3963 	sit_i->bitmap_size = bitmap_size;
3964 	sit_i->dirty_sentries = 0;
3965 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3966 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3967 	sit_i->mounted_time = ktime_get_real_seconds();
3968 	init_rwsem(&sit_i->sentry_lock);
3969 	return 0;
3970 }
3971 
3972 static int build_free_segmap(struct f2fs_sb_info *sbi)
3973 {
3974 	struct free_segmap_info *free_i;
3975 	unsigned int bitmap_size, sec_bitmap_size;
3976 
3977 	/* allocate memory for free segmap information */
3978 	free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
3979 	if (!free_i)
3980 		return -ENOMEM;
3981 
3982 	SM_I(sbi)->free_info = free_i;
3983 
3984 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3985 	free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
3986 	if (!free_i->free_segmap)
3987 		return -ENOMEM;
3988 
3989 	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3990 	free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
3991 	if (!free_i->free_secmap)
3992 		return -ENOMEM;
3993 
3994 	/* set all segments as dirty temporarily */
3995 	memset(free_i->free_segmap, 0xff, bitmap_size);
3996 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3997 
3998 	/* init free segmap information */
3999 	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4000 	free_i->free_segments = 0;
4001 	free_i->free_sections = 0;
4002 	spin_lock_init(&free_i->segmap_lock);
4003 	return 0;
4004 }
4005 
4006 static int build_curseg(struct f2fs_sb_info *sbi)
4007 {
4008 	struct curseg_info *array;
4009 	int i;
4010 
4011 	array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
4012 			     GFP_KERNEL);
4013 	if (!array)
4014 		return -ENOMEM;
4015 
4016 	SM_I(sbi)->curseg_array = array;
4017 
4018 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
4019 		mutex_init(&array[i].curseg_mutex);
4020 		array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4021 		if (!array[i].sum_blk)
4022 			return -ENOMEM;
4023 		init_rwsem(&array[i].journal_rwsem);
4024 		array[i].journal = f2fs_kzalloc(sbi,
4025 				sizeof(struct f2fs_journal), GFP_KERNEL);
4026 		if (!array[i].journal)
4027 			return -ENOMEM;
4028 		array[i].segno = NULL_SEGNO;
4029 		array[i].next_blkoff = 0;
4030 	}
4031 	return restore_curseg_summaries(sbi);
4032 }
4033 
4034 static int build_sit_entries(struct f2fs_sb_info *sbi)
4035 {
4036 	struct sit_info *sit_i = SIT_I(sbi);
4037 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4038 	struct f2fs_journal *journal = curseg->journal;
4039 	struct seg_entry *se;
4040 	struct f2fs_sit_entry sit;
4041 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
4042 	unsigned int i, start, end;
4043 	unsigned int readed, start_blk = 0;
4044 	int err = 0;
4045 	block_t total_node_blocks = 0;
4046 
4047 	do {
4048 		readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4049 							META_SIT, true);
4050 
4051 		start = start_blk * sit_i->sents_per_block;
4052 		end = (start_blk + readed) * sit_i->sents_per_block;
4053 
4054 		for (; start < end && start < MAIN_SEGS(sbi); start++) {
4055 			struct f2fs_sit_block *sit_blk;
4056 			struct page *page;
4057 
4058 			se = &sit_i->sentries[start];
4059 			page = get_current_sit_page(sbi, start);
4060 			if (IS_ERR(page))
4061 				return PTR_ERR(page);
4062 			sit_blk = (struct f2fs_sit_block *)page_address(page);
4063 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4064 			f2fs_put_page(page, 1);
4065 
4066 			err = check_block_count(sbi, start, &sit);
4067 			if (err)
4068 				return err;
4069 			seg_info_from_raw_sit(se, &sit);
4070 			if (IS_NODESEG(se->type))
4071 				total_node_blocks += se->valid_blocks;
4072 
4073 			/* build discard map only one time */
4074 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4075 				memset(se->discard_map, 0xff,
4076 					SIT_VBLOCK_MAP_SIZE);
4077 			} else {
4078 				memcpy(se->discard_map,
4079 					se->cur_valid_map,
4080 					SIT_VBLOCK_MAP_SIZE);
4081 				sbi->discard_blks +=
4082 					sbi->blocks_per_seg -
4083 					se->valid_blocks;
4084 			}
4085 
4086 			if (__is_large_section(sbi))
4087 				get_sec_entry(sbi, start)->valid_blocks +=
4088 							se->valid_blocks;
4089 		}
4090 		start_blk += readed;
4091 	} while (start_blk < sit_blk_cnt);
4092 
4093 	down_read(&curseg->journal_rwsem);
4094 	for (i = 0; i < sits_in_cursum(journal); i++) {
4095 		unsigned int old_valid_blocks;
4096 
4097 		start = le32_to_cpu(segno_in_journal(journal, i));
4098 		if (start >= MAIN_SEGS(sbi)) {
4099 			f2fs_msg(sbi->sb, KERN_ERR,
4100 					"Wrong journal entry on segno %u",
4101 					start);
4102 			set_sbi_flag(sbi, SBI_NEED_FSCK);
4103 			err = -EINVAL;
4104 			break;
4105 		}
4106 
4107 		se = &sit_i->sentries[start];
4108 		sit = sit_in_journal(journal, i);
4109 
4110 		old_valid_blocks = se->valid_blocks;
4111 		if (IS_NODESEG(se->type))
4112 			total_node_blocks -= old_valid_blocks;
4113 
4114 		err = check_block_count(sbi, start, &sit);
4115 		if (err)
4116 			break;
4117 		seg_info_from_raw_sit(se, &sit);
4118 		if (IS_NODESEG(se->type))
4119 			total_node_blocks += se->valid_blocks;
4120 
4121 		if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4122 			memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4123 		} else {
4124 			memcpy(se->discard_map, se->cur_valid_map,
4125 						SIT_VBLOCK_MAP_SIZE);
4126 			sbi->discard_blks += old_valid_blocks;
4127 			sbi->discard_blks -= se->valid_blocks;
4128 		}
4129 
4130 		if (__is_large_section(sbi)) {
4131 			get_sec_entry(sbi, start)->valid_blocks +=
4132 							se->valid_blocks;
4133 			get_sec_entry(sbi, start)->valid_blocks -=
4134 							old_valid_blocks;
4135 		}
4136 	}
4137 	up_read(&curseg->journal_rwsem);
4138 
4139 	if (!err && total_node_blocks != valid_node_count(sbi)) {
4140 		f2fs_msg(sbi->sb, KERN_ERR,
4141 			"SIT is corrupted node# %u vs %u",
4142 			total_node_blocks, valid_node_count(sbi));
4143 		set_sbi_flag(sbi, SBI_NEED_FSCK);
4144 		err = -EINVAL;
4145 	}
4146 
4147 	return err;
4148 }
4149 
4150 static void init_free_segmap(struct f2fs_sb_info *sbi)
4151 {
4152 	unsigned int start;
4153 	int type;
4154 
4155 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4156 		struct seg_entry *sentry = get_seg_entry(sbi, start);
4157 		if (!sentry->valid_blocks)
4158 			__set_free(sbi, start);
4159 		else
4160 			SIT_I(sbi)->written_valid_blocks +=
4161 						sentry->valid_blocks;
4162 	}
4163 
4164 	/* set use the current segments */
4165 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4166 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4167 		__set_test_and_inuse(sbi, curseg_t->segno);
4168 	}
4169 }
4170 
4171 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4172 {
4173 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4174 	struct free_segmap_info *free_i = FREE_I(sbi);
4175 	unsigned int segno = 0, offset = 0;
4176 	unsigned short valid_blocks;
4177 
4178 	while (1) {
4179 		/* find dirty segment based on free segmap */
4180 		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4181 		if (segno >= MAIN_SEGS(sbi))
4182 			break;
4183 		offset = segno + 1;
4184 		valid_blocks = get_valid_blocks(sbi, segno, false);
4185 		if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4186 			continue;
4187 		if (valid_blocks > sbi->blocks_per_seg) {
4188 			f2fs_bug_on(sbi, 1);
4189 			continue;
4190 		}
4191 		mutex_lock(&dirty_i->seglist_lock);
4192 		__locate_dirty_segment(sbi, segno, DIRTY);
4193 		mutex_unlock(&dirty_i->seglist_lock);
4194 	}
4195 }
4196 
4197 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4198 {
4199 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4200 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4201 
4202 	dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4203 	if (!dirty_i->victim_secmap)
4204 		return -ENOMEM;
4205 	return 0;
4206 }
4207 
4208 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4209 {
4210 	struct dirty_seglist_info *dirty_i;
4211 	unsigned int bitmap_size, i;
4212 
4213 	/* allocate memory for dirty segments list information */
4214 	dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4215 								GFP_KERNEL);
4216 	if (!dirty_i)
4217 		return -ENOMEM;
4218 
4219 	SM_I(sbi)->dirty_info = dirty_i;
4220 	mutex_init(&dirty_i->seglist_lock);
4221 
4222 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4223 
4224 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
4225 		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4226 								GFP_KERNEL);
4227 		if (!dirty_i->dirty_segmap[i])
4228 			return -ENOMEM;
4229 	}
4230 
4231 	init_dirty_segmap(sbi);
4232 	return init_victim_secmap(sbi);
4233 }
4234 
4235 /*
4236  * Update min, max modified time for cost-benefit GC algorithm
4237  */
4238 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4239 {
4240 	struct sit_info *sit_i = SIT_I(sbi);
4241 	unsigned int segno;
4242 
4243 	down_write(&sit_i->sentry_lock);
4244 
4245 	sit_i->min_mtime = ULLONG_MAX;
4246 
4247 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4248 		unsigned int i;
4249 		unsigned long long mtime = 0;
4250 
4251 		for (i = 0; i < sbi->segs_per_sec; i++)
4252 			mtime += get_seg_entry(sbi, segno + i)->mtime;
4253 
4254 		mtime = div_u64(mtime, sbi->segs_per_sec);
4255 
4256 		if (sit_i->min_mtime > mtime)
4257 			sit_i->min_mtime = mtime;
4258 	}
4259 	sit_i->max_mtime = get_mtime(sbi, false);
4260 	up_write(&sit_i->sentry_lock);
4261 }
4262 
4263 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4264 {
4265 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4266 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4267 	struct f2fs_sm_info *sm_info;
4268 	int err;
4269 
4270 	sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4271 	if (!sm_info)
4272 		return -ENOMEM;
4273 
4274 	/* init sm info */
4275 	sbi->sm_info = sm_info;
4276 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4277 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4278 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4279 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4280 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4281 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4282 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4283 	sm_info->rec_prefree_segments = sm_info->main_segments *
4284 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4285 	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4286 		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4287 
4288 	if (!test_opt(sbi, LFS))
4289 		sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4290 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4291 	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4292 	sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4293 	sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4294 	sm_info->min_ssr_sections = reserved_sections(sbi);
4295 
4296 	INIT_LIST_HEAD(&sm_info->sit_entry_set);
4297 
4298 	init_rwsem(&sm_info->curseg_lock);
4299 
4300 	if (!f2fs_readonly(sbi->sb)) {
4301 		err = f2fs_create_flush_cmd_control(sbi);
4302 		if (err)
4303 			return err;
4304 	}
4305 
4306 	err = create_discard_cmd_control(sbi);
4307 	if (err)
4308 		return err;
4309 
4310 	err = build_sit_info(sbi);
4311 	if (err)
4312 		return err;
4313 	err = build_free_segmap(sbi);
4314 	if (err)
4315 		return err;
4316 	err = build_curseg(sbi);
4317 	if (err)
4318 		return err;
4319 
4320 	/* reinit free segmap based on SIT */
4321 	err = build_sit_entries(sbi);
4322 	if (err)
4323 		return err;
4324 
4325 	init_free_segmap(sbi);
4326 	err = build_dirty_segmap(sbi);
4327 	if (err)
4328 		return err;
4329 
4330 	init_min_max_mtime(sbi);
4331 	return 0;
4332 }
4333 
4334 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4335 		enum dirty_type dirty_type)
4336 {
4337 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4338 
4339 	mutex_lock(&dirty_i->seglist_lock);
4340 	kvfree(dirty_i->dirty_segmap[dirty_type]);
4341 	dirty_i->nr_dirty[dirty_type] = 0;
4342 	mutex_unlock(&dirty_i->seglist_lock);
4343 }
4344 
4345 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4346 {
4347 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4348 	kvfree(dirty_i->victim_secmap);
4349 }
4350 
4351 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4352 {
4353 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4354 	int i;
4355 
4356 	if (!dirty_i)
4357 		return;
4358 
4359 	/* discard pre-free/dirty segments list */
4360 	for (i = 0; i < NR_DIRTY_TYPE; i++)
4361 		discard_dirty_segmap(sbi, i);
4362 
4363 	destroy_victim_secmap(sbi);
4364 	SM_I(sbi)->dirty_info = NULL;
4365 	kvfree(dirty_i);
4366 }
4367 
4368 static void destroy_curseg(struct f2fs_sb_info *sbi)
4369 {
4370 	struct curseg_info *array = SM_I(sbi)->curseg_array;
4371 	int i;
4372 
4373 	if (!array)
4374 		return;
4375 	SM_I(sbi)->curseg_array = NULL;
4376 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
4377 		kvfree(array[i].sum_blk);
4378 		kvfree(array[i].journal);
4379 	}
4380 	kvfree(array);
4381 }
4382 
4383 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4384 {
4385 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4386 	if (!free_i)
4387 		return;
4388 	SM_I(sbi)->free_info = NULL;
4389 	kvfree(free_i->free_segmap);
4390 	kvfree(free_i->free_secmap);
4391 	kvfree(free_i);
4392 }
4393 
4394 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4395 {
4396 	struct sit_info *sit_i = SIT_I(sbi);
4397 	unsigned int start;
4398 
4399 	if (!sit_i)
4400 		return;
4401 
4402 	if (sit_i->sentries) {
4403 		for (start = 0; start < MAIN_SEGS(sbi); start++) {
4404 			kvfree(sit_i->sentries[start].cur_valid_map);
4405 #ifdef CONFIG_F2FS_CHECK_FS
4406 			kvfree(sit_i->sentries[start].cur_valid_map_mir);
4407 #endif
4408 			kvfree(sit_i->sentries[start].ckpt_valid_map);
4409 			kvfree(sit_i->sentries[start].discard_map);
4410 		}
4411 	}
4412 	kvfree(sit_i->tmp_map);
4413 
4414 	kvfree(sit_i->sentries);
4415 	kvfree(sit_i->sec_entries);
4416 	kvfree(sit_i->dirty_sentries_bitmap);
4417 
4418 	SM_I(sbi)->sit_info = NULL;
4419 	kvfree(sit_i->sit_bitmap);
4420 #ifdef CONFIG_F2FS_CHECK_FS
4421 	kvfree(sit_i->sit_bitmap_mir);
4422 #endif
4423 	kvfree(sit_i);
4424 }
4425 
4426 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4427 {
4428 	struct f2fs_sm_info *sm_info = SM_I(sbi);
4429 
4430 	if (!sm_info)
4431 		return;
4432 	f2fs_destroy_flush_cmd_control(sbi, true);
4433 	destroy_discard_cmd_control(sbi);
4434 	destroy_dirty_segmap(sbi);
4435 	destroy_curseg(sbi);
4436 	destroy_free_segmap(sbi);
4437 	destroy_sit_info(sbi);
4438 	sbi->sm_info = NULL;
4439 	kvfree(sm_info);
4440 }
4441 
4442 int __init f2fs_create_segment_manager_caches(void)
4443 {
4444 	discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4445 			sizeof(struct discard_entry));
4446 	if (!discard_entry_slab)
4447 		goto fail;
4448 
4449 	discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4450 			sizeof(struct discard_cmd));
4451 	if (!discard_cmd_slab)
4452 		goto destroy_discard_entry;
4453 
4454 	sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4455 			sizeof(struct sit_entry_set));
4456 	if (!sit_entry_set_slab)
4457 		goto destroy_discard_cmd;
4458 
4459 	inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4460 			sizeof(struct inmem_pages));
4461 	if (!inmem_entry_slab)
4462 		goto destroy_sit_entry_set;
4463 	return 0;
4464 
4465 destroy_sit_entry_set:
4466 	kmem_cache_destroy(sit_entry_set_slab);
4467 destroy_discard_cmd:
4468 	kmem_cache_destroy(discard_cmd_slab);
4469 destroy_discard_entry:
4470 	kmem_cache_destroy(discard_entry_slab);
4471 fail:
4472 	return -ENOMEM;
4473 }
4474 
4475 void f2fs_destroy_segment_manager_caches(void)
4476 {
4477 	kmem_cache_destroy(sit_entry_set_slab);
4478 	kmem_cache_destroy(discard_cmd_slab);
4479 	kmem_cache_destroy(discard_entry_slab);
4480 	kmem_cache_destroy(inmem_entry_slab);
4481 }
4482