xref: /openbmc/linux/fs/f2fs/segment.c (revision c24c57a4)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * fs/f2fs/segment.c
4  *
5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6  *             http://www.samsung.com/
7  */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/bio.h>
11 #include <linux/blkdev.h>
12 #include <linux/prefetch.h>
13 #include <linux/kthread.h>
14 #include <linux/swap.h>
15 #include <linux/timer.h>
16 #include <linux/freezer.h>
17 #include <linux/sched/signal.h>
18 
19 #include "f2fs.h"
20 #include "segment.h"
21 #include "node.h"
22 #include "gc.h"
23 #include "trace.h"
24 #include <trace/events/f2fs.h>
25 
26 #define __reverse_ffz(x) __reverse_ffs(~(x))
27 
28 static struct kmem_cache *discard_entry_slab;
29 static struct kmem_cache *discard_cmd_slab;
30 static struct kmem_cache *sit_entry_set_slab;
31 static struct kmem_cache *inmem_entry_slab;
32 
33 static unsigned long __reverse_ulong(unsigned char *str)
34 {
35 	unsigned long tmp = 0;
36 	int shift = 24, idx = 0;
37 
38 #if BITS_PER_LONG == 64
39 	shift = 56;
40 #endif
41 	while (shift >= 0) {
42 		tmp |= (unsigned long)str[idx++] << shift;
43 		shift -= BITS_PER_BYTE;
44 	}
45 	return tmp;
46 }
47 
48 /*
49  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
50  * MSB and LSB are reversed in a byte by f2fs_set_bit.
51  */
52 static inline unsigned long __reverse_ffs(unsigned long word)
53 {
54 	int num = 0;
55 
56 #if BITS_PER_LONG == 64
57 	if ((word & 0xffffffff00000000UL) == 0)
58 		num += 32;
59 	else
60 		word >>= 32;
61 #endif
62 	if ((word & 0xffff0000) == 0)
63 		num += 16;
64 	else
65 		word >>= 16;
66 
67 	if ((word & 0xff00) == 0)
68 		num += 8;
69 	else
70 		word >>= 8;
71 
72 	if ((word & 0xf0) == 0)
73 		num += 4;
74 	else
75 		word >>= 4;
76 
77 	if ((word & 0xc) == 0)
78 		num += 2;
79 	else
80 		word >>= 2;
81 
82 	if ((word & 0x2) == 0)
83 		num += 1;
84 	return num;
85 }
86 
87 /*
88  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
89  * f2fs_set_bit makes MSB and LSB reversed in a byte.
90  * @size must be integral times of unsigned long.
91  * Example:
92  *                             MSB <--> LSB
93  *   f2fs_set_bit(0, bitmap) => 1000 0000
94  *   f2fs_set_bit(7, bitmap) => 0000 0001
95  */
96 static unsigned long __find_rev_next_bit(const unsigned long *addr,
97 			unsigned long size, unsigned long offset)
98 {
99 	const unsigned long *p = addr + BIT_WORD(offset);
100 	unsigned long result = size;
101 	unsigned long tmp;
102 
103 	if (offset >= size)
104 		return size;
105 
106 	size -= (offset & ~(BITS_PER_LONG - 1));
107 	offset %= BITS_PER_LONG;
108 
109 	while (1) {
110 		if (*p == 0)
111 			goto pass;
112 
113 		tmp = __reverse_ulong((unsigned char *)p);
114 
115 		tmp &= ~0UL >> offset;
116 		if (size < BITS_PER_LONG)
117 			tmp &= (~0UL << (BITS_PER_LONG - size));
118 		if (tmp)
119 			goto found;
120 pass:
121 		if (size <= BITS_PER_LONG)
122 			break;
123 		size -= BITS_PER_LONG;
124 		offset = 0;
125 		p++;
126 	}
127 	return result;
128 found:
129 	return result - size + __reverse_ffs(tmp);
130 }
131 
132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
133 			unsigned long size, unsigned long offset)
134 {
135 	const unsigned long *p = addr + BIT_WORD(offset);
136 	unsigned long result = size;
137 	unsigned long tmp;
138 
139 	if (offset >= size)
140 		return size;
141 
142 	size -= (offset & ~(BITS_PER_LONG - 1));
143 	offset %= BITS_PER_LONG;
144 
145 	while (1) {
146 		if (*p == ~0UL)
147 			goto pass;
148 
149 		tmp = __reverse_ulong((unsigned char *)p);
150 
151 		if (offset)
152 			tmp |= ~0UL << (BITS_PER_LONG - offset);
153 		if (size < BITS_PER_LONG)
154 			tmp |= ~0UL >> size;
155 		if (tmp != ~0UL)
156 			goto found;
157 pass:
158 		if (size <= BITS_PER_LONG)
159 			break;
160 		size -= BITS_PER_LONG;
161 		offset = 0;
162 		p++;
163 	}
164 	return result;
165 found:
166 	return result - size + __reverse_ffz(tmp);
167 }
168 
169 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
170 {
171 	int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
172 	int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
173 	int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
174 
175 	if (test_opt(sbi, LFS))
176 		return false;
177 	if (sbi->gc_mode == GC_URGENT)
178 		return true;
179 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
180 		return true;
181 
182 	return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
183 			SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
184 }
185 
186 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
187 {
188 	struct inmem_pages *new;
189 
190 	f2fs_trace_pid(page);
191 
192 	f2fs_set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
193 
194 	new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
195 
196 	/* add atomic page indices to the list */
197 	new->page = page;
198 	INIT_LIST_HEAD(&new->list);
199 
200 	/* increase reference count with clean state */
201 	get_page(page);
202 	mutex_lock(&F2FS_I(inode)->inmem_lock);
203 	list_add_tail(&new->list, &F2FS_I(inode)->inmem_pages);
204 	inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
205 	mutex_unlock(&F2FS_I(inode)->inmem_lock);
206 
207 	trace_f2fs_register_inmem_page(page, INMEM);
208 }
209 
210 static int __revoke_inmem_pages(struct inode *inode,
211 				struct list_head *head, bool drop, bool recover,
212 				bool trylock)
213 {
214 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
215 	struct inmem_pages *cur, *tmp;
216 	int err = 0;
217 
218 	list_for_each_entry_safe(cur, tmp, head, list) {
219 		struct page *page = cur->page;
220 
221 		if (drop)
222 			trace_f2fs_commit_inmem_page(page, INMEM_DROP);
223 
224 		if (trylock) {
225 			/*
226 			 * to avoid deadlock in between page lock and
227 			 * inmem_lock.
228 			 */
229 			if (!trylock_page(page))
230 				continue;
231 		} else {
232 			lock_page(page);
233 		}
234 
235 		f2fs_wait_on_page_writeback(page, DATA, true, true);
236 
237 		if (recover) {
238 			struct dnode_of_data dn;
239 			struct node_info ni;
240 
241 			trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
242 retry:
243 			set_new_dnode(&dn, inode, NULL, NULL, 0);
244 			err = f2fs_get_dnode_of_data(&dn, page->index,
245 								LOOKUP_NODE);
246 			if (err) {
247 				if (err == -ENOMEM) {
248 					congestion_wait(BLK_RW_ASYNC, HZ/50);
249 					cond_resched();
250 					goto retry;
251 				}
252 				err = -EAGAIN;
253 				goto next;
254 			}
255 
256 			err = f2fs_get_node_info(sbi, dn.nid, &ni);
257 			if (err) {
258 				f2fs_put_dnode(&dn);
259 				return err;
260 			}
261 
262 			if (cur->old_addr == NEW_ADDR) {
263 				f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
264 				f2fs_update_data_blkaddr(&dn, NEW_ADDR);
265 			} else
266 				f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
267 					cur->old_addr, ni.version, true, true);
268 			f2fs_put_dnode(&dn);
269 		}
270 next:
271 		/* we don't need to invalidate this in the sccessful status */
272 		if (drop || recover) {
273 			ClearPageUptodate(page);
274 			clear_cold_data(page);
275 		}
276 		f2fs_clear_page_private(page);
277 		f2fs_put_page(page, 1);
278 
279 		list_del(&cur->list);
280 		kmem_cache_free(inmem_entry_slab, cur);
281 		dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
282 	}
283 	return err;
284 }
285 
286 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
287 {
288 	struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
289 	struct inode *inode;
290 	struct f2fs_inode_info *fi;
291 next:
292 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
293 	if (list_empty(head)) {
294 		spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
295 		return;
296 	}
297 	fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
298 	inode = igrab(&fi->vfs_inode);
299 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
300 
301 	if (inode) {
302 		if (gc_failure) {
303 			if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
304 				goto drop;
305 			goto skip;
306 		}
307 drop:
308 		set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
309 		f2fs_drop_inmem_pages(inode);
310 		iput(inode);
311 	}
312 skip:
313 	congestion_wait(BLK_RW_ASYNC, HZ/50);
314 	cond_resched();
315 	goto next;
316 }
317 
318 void f2fs_drop_inmem_pages(struct inode *inode)
319 {
320 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
321 	struct f2fs_inode_info *fi = F2FS_I(inode);
322 
323 	while (!list_empty(&fi->inmem_pages)) {
324 		mutex_lock(&fi->inmem_lock);
325 		__revoke_inmem_pages(inode, &fi->inmem_pages,
326 						true, false, true);
327 		mutex_unlock(&fi->inmem_lock);
328 	}
329 
330 	clear_inode_flag(inode, FI_ATOMIC_FILE);
331 	fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
332 	stat_dec_atomic_write(inode);
333 
334 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
335 	if (!list_empty(&fi->inmem_ilist))
336 		list_del_init(&fi->inmem_ilist);
337 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
338 }
339 
340 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
341 {
342 	struct f2fs_inode_info *fi = F2FS_I(inode);
343 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
344 	struct list_head *head = &fi->inmem_pages;
345 	struct inmem_pages *cur = NULL;
346 
347 	f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
348 
349 	mutex_lock(&fi->inmem_lock);
350 	list_for_each_entry(cur, head, list) {
351 		if (cur->page == page)
352 			break;
353 	}
354 
355 	f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
356 	list_del(&cur->list);
357 	mutex_unlock(&fi->inmem_lock);
358 
359 	dec_page_count(sbi, F2FS_INMEM_PAGES);
360 	kmem_cache_free(inmem_entry_slab, cur);
361 
362 	ClearPageUptodate(page);
363 	f2fs_clear_page_private(page);
364 	f2fs_put_page(page, 0);
365 
366 	trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
367 }
368 
369 static int __f2fs_commit_inmem_pages(struct inode *inode)
370 {
371 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
372 	struct f2fs_inode_info *fi = F2FS_I(inode);
373 	struct inmem_pages *cur, *tmp;
374 	struct f2fs_io_info fio = {
375 		.sbi = sbi,
376 		.ino = inode->i_ino,
377 		.type = DATA,
378 		.op = REQ_OP_WRITE,
379 		.op_flags = REQ_SYNC | REQ_PRIO,
380 		.io_type = FS_DATA_IO,
381 	};
382 	struct list_head revoke_list;
383 	bool submit_bio = false;
384 	int err = 0;
385 
386 	INIT_LIST_HEAD(&revoke_list);
387 
388 	list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
389 		struct page *page = cur->page;
390 
391 		lock_page(page);
392 		if (page->mapping == inode->i_mapping) {
393 			trace_f2fs_commit_inmem_page(page, INMEM);
394 
395 			f2fs_wait_on_page_writeback(page, DATA, true, true);
396 
397 			set_page_dirty(page);
398 			if (clear_page_dirty_for_io(page)) {
399 				inode_dec_dirty_pages(inode);
400 				f2fs_remove_dirty_inode(inode);
401 			}
402 retry:
403 			fio.page = page;
404 			fio.old_blkaddr = NULL_ADDR;
405 			fio.encrypted_page = NULL;
406 			fio.need_lock = LOCK_DONE;
407 			err = f2fs_do_write_data_page(&fio);
408 			if (err) {
409 				if (err == -ENOMEM) {
410 					congestion_wait(BLK_RW_ASYNC, HZ/50);
411 					cond_resched();
412 					goto retry;
413 				}
414 				unlock_page(page);
415 				break;
416 			}
417 			/* record old blkaddr for revoking */
418 			cur->old_addr = fio.old_blkaddr;
419 			submit_bio = true;
420 		}
421 		unlock_page(page);
422 		list_move_tail(&cur->list, &revoke_list);
423 	}
424 
425 	if (submit_bio)
426 		f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
427 
428 	if (err) {
429 		/*
430 		 * try to revoke all committed pages, but still we could fail
431 		 * due to no memory or other reason, if that happened, EAGAIN
432 		 * will be returned, which means in such case, transaction is
433 		 * already not integrity, caller should use journal to do the
434 		 * recovery or rewrite & commit last transaction. For other
435 		 * error number, revoking was done by filesystem itself.
436 		 */
437 		err = __revoke_inmem_pages(inode, &revoke_list,
438 						false, true, false);
439 
440 		/* drop all uncommitted pages */
441 		__revoke_inmem_pages(inode, &fi->inmem_pages,
442 						true, false, false);
443 	} else {
444 		__revoke_inmem_pages(inode, &revoke_list,
445 						false, false, false);
446 	}
447 
448 	return err;
449 }
450 
451 int f2fs_commit_inmem_pages(struct inode *inode)
452 {
453 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
454 	struct f2fs_inode_info *fi = F2FS_I(inode);
455 	int err;
456 
457 	f2fs_balance_fs(sbi, true);
458 
459 	down_write(&fi->i_gc_rwsem[WRITE]);
460 
461 	f2fs_lock_op(sbi);
462 	set_inode_flag(inode, FI_ATOMIC_COMMIT);
463 
464 	mutex_lock(&fi->inmem_lock);
465 	err = __f2fs_commit_inmem_pages(inode);
466 	mutex_unlock(&fi->inmem_lock);
467 
468 	clear_inode_flag(inode, FI_ATOMIC_COMMIT);
469 
470 	f2fs_unlock_op(sbi);
471 	up_write(&fi->i_gc_rwsem[WRITE]);
472 
473 	return err;
474 }
475 
476 /*
477  * This function balances dirty node and dentry pages.
478  * In addition, it controls garbage collection.
479  */
480 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
481 {
482 	if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
483 		f2fs_show_injection_info(FAULT_CHECKPOINT);
484 		f2fs_stop_checkpoint(sbi, false);
485 	}
486 
487 	/* balance_fs_bg is able to be pending */
488 	if (need && excess_cached_nats(sbi))
489 		f2fs_balance_fs_bg(sbi);
490 
491 	if (!f2fs_is_checkpoint_ready(sbi))
492 		return;
493 
494 	/*
495 	 * We should do GC or end up with checkpoint, if there are so many dirty
496 	 * dir/node pages without enough free segments.
497 	 */
498 	if (has_not_enough_free_secs(sbi, 0, 0)) {
499 		mutex_lock(&sbi->gc_mutex);
500 		f2fs_gc(sbi, false, false, NULL_SEGNO);
501 	}
502 }
503 
504 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
505 {
506 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
507 		return;
508 
509 	/* try to shrink extent cache when there is no enough memory */
510 	if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
511 		f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
512 
513 	/* check the # of cached NAT entries */
514 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
515 		f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
516 
517 	if (!f2fs_available_free_memory(sbi, FREE_NIDS))
518 		f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
519 	else
520 		f2fs_build_free_nids(sbi, false, false);
521 
522 	if (!is_idle(sbi, REQ_TIME) &&
523 		(!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
524 		return;
525 
526 	/* checkpoint is the only way to shrink partial cached entries */
527 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
528 			!f2fs_available_free_memory(sbi, INO_ENTRIES) ||
529 			excess_prefree_segs(sbi) ||
530 			excess_dirty_nats(sbi) ||
531 			excess_dirty_nodes(sbi) ||
532 			f2fs_time_over(sbi, CP_TIME)) {
533 		if (test_opt(sbi, DATA_FLUSH)) {
534 			struct blk_plug plug;
535 
536 			mutex_lock(&sbi->flush_lock);
537 
538 			blk_start_plug(&plug);
539 			f2fs_sync_dirty_inodes(sbi, FILE_INODE);
540 			blk_finish_plug(&plug);
541 
542 			mutex_unlock(&sbi->flush_lock);
543 		}
544 		f2fs_sync_fs(sbi->sb, true);
545 		stat_inc_bg_cp_count(sbi->stat_info);
546 	}
547 }
548 
549 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
550 				struct block_device *bdev)
551 {
552 	struct bio *bio;
553 	int ret;
554 
555 	bio = f2fs_bio_alloc(sbi, 0, false);
556 	if (!bio)
557 		return -ENOMEM;
558 
559 	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
560 	bio_set_dev(bio, bdev);
561 	ret = submit_bio_wait(bio);
562 	bio_put(bio);
563 
564 	trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
565 				test_opt(sbi, FLUSH_MERGE), ret);
566 	return ret;
567 }
568 
569 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
570 {
571 	int ret = 0;
572 	int i;
573 
574 	if (!f2fs_is_multi_device(sbi))
575 		return __submit_flush_wait(sbi, sbi->sb->s_bdev);
576 
577 	for (i = 0; i < sbi->s_ndevs; i++) {
578 		if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
579 			continue;
580 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
581 		if (ret)
582 			break;
583 	}
584 	return ret;
585 }
586 
587 static int issue_flush_thread(void *data)
588 {
589 	struct f2fs_sb_info *sbi = data;
590 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
591 	wait_queue_head_t *q = &fcc->flush_wait_queue;
592 repeat:
593 	if (kthread_should_stop())
594 		return 0;
595 
596 	sb_start_intwrite(sbi->sb);
597 
598 	if (!llist_empty(&fcc->issue_list)) {
599 		struct flush_cmd *cmd, *next;
600 		int ret;
601 
602 		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
603 		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
604 
605 		cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
606 
607 		ret = submit_flush_wait(sbi, cmd->ino);
608 		atomic_inc(&fcc->issued_flush);
609 
610 		llist_for_each_entry_safe(cmd, next,
611 					  fcc->dispatch_list, llnode) {
612 			cmd->ret = ret;
613 			complete(&cmd->wait);
614 		}
615 		fcc->dispatch_list = NULL;
616 	}
617 
618 	sb_end_intwrite(sbi->sb);
619 
620 	wait_event_interruptible(*q,
621 		kthread_should_stop() || !llist_empty(&fcc->issue_list));
622 	goto repeat;
623 }
624 
625 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
626 {
627 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
628 	struct flush_cmd cmd;
629 	int ret;
630 
631 	if (test_opt(sbi, NOBARRIER))
632 		return 0;
633 
634 	if (!test_opt(sbi, FLUSH_MERGE)) {
635 		atomic_inc(&fcc->queued_flush);
636 		ret = submit_flush_wait(sbi, ino);
637 		atomic_dec(&fcc->queued_flush);
638 		atomic_inc(&fcc->issued_flush);
639 		return ret;
640 	}
641 
642 	if (atomic_inc_return(&fcc->queued_flush) == 1 ||
643 	    f2fs_is_multi_device(sbi)) {
644 		ret = submit_flush_wait(sbi, ino);
645 		atomic_dec(&fcc->queued_flush);
646 
647 		atomic_inc(&fcc->issued_flush);
648 		return ret;
649 	}
650 
651 	cmd.ino = ino;
652 	init_completion(&cmd.wait);
653 
654 	llist_add(&cmd.llnode, &fcc->issue_list);
655 
656 	/* update issue_list before we wake up issue_flush thread */
657 	smp_mb();
658 
659 	if (waitqueue_active(&fcc->flush_wait_queue))
660 		wake_up(&fcc->flush_wait_queue);
661 
662 	if (fcc->f2fs_issue_flush) {
663 		wait_for_completion(&cmd.wait);
664 		atomic_dec(&fcc->queued_flush);
665 	} else {
666 		struct llist_node *list;
667 
668 		list = llist_del_all(&fcc->issue_list);
669 		if (!list) {
670 			wait_for_completion(&cmd.wait);
671 			atomic_dec(&fcc->queued_flush);
672 		} else {
673 			struct flush_cmd *tmp, *next;
674 
675 			ret = submit_flush_wait(sbi, ino);
676 
677 			llist_for_each_entry_safe(tmp, next, list, llnode) {
678 				if (tmp == &cmd) {
679 					cmd.ret = ret;
680 					atomic_dec(&fcc->queued_flush);
681 					continue;
682 				}
683 				tmp->ret = ret;
684 				complete(&tmp->wait);
685 			}
686 		}
687 	}
688 
689 	return cmd.ret;
690 }
691 
692 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
693 {
694 	dev_t dev = sbi->sb->s_bdev->bd_dev;
695 	struct flush_cmd_control *fcc;
696 	int err = 0;
697 
698 	if (SM_I(sbi)->fcc_info) {
699 		fcc = SM_I(sbi)->fcc_info;
700 		if (fcc->f2fs_issue_flush)
701 			return err;
702 		goto init_thread;
703 	}
704 
705 	fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
706 	if (!fcc)
707 		return -ENOMEM;
708 	atomic_set(&fcc->issued_flush, 0);
709 	atomic_set(&fcc->queued_flush, 0);
710 	init_waitqueue_head(&fcc->flush_wait_queue);
711 	init_llist_head(&fcc->issue_list);
712 	SM_I(sbi)->fcc_info = fcc;
713 	if (!test_opt(sbi, FLUSH_MERGE))
714 		return err;
715 
716 init_thread:
717 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
718 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
719 	if (IS_ERR(fcc->f2fs_issue_flush)) {
720 		err = PTR_ERR(fcc->f2fs_issue_flush);
721 		kvfree(fcc);
722 		SM_I(sbi)->fcc_info = NULL;
723 		return err;
724 	}
725 
726 	return err;
727 }
728 
729 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
730 {
731 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
732 
733 	if (fcc && fcc->f2fs_issue_flush) {
734 		struct task_struct *flush_thread = fcc->f2fs_issue_flush;
735 
736 		fcc->f2fs_issue_flush = NULL;
737 		kthread_stop(flush_thread);
738 	}
739 	if (free) {
740 		kvfree(fcc);
741 		SM_I(sbi)->fcc_info = NULL;
742 	}
743 }
744 
745 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
746 {
747 	int ret = 0, i;
748 
749 	if (!f2fs_is_multi_device(sbi))
750 		return 0;
751 
752 	for (i = 1; i < sbi->s_ndevs; i++) {
753 		if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
754 			continue;
755 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
756 		if (ret)
757 			break;
758 
759 		spin_lock(&sbi->dev_lock);
760 		f2fs_clear_bit(i, (char *)&sbi->dirty_device);
761 		spin_unlock(&sbi->dev_lock);
762 	}
763 
764 	return ret;
765 }
766 
767 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
768 		enum dirty_type dirty_type)
769 {
770 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
771 
772 	/* need not be added */
773 	if (IS_CURSEG(sbi, segno))
774 		return;
775 
776 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
777 		dirty_i->nr_dirty[dirty_type]++;
778 
779 	if (dirty_type == DIRTY) {
780 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
781 		enum dirty_type t = sentry->type;
782 
783 		if (unlikely(t >= DIRTY)) {
784 			f2fs_bug_on(sbi, 1);
785 			return;
786 		}
787 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
788 			dirty_i->nr_dirty[t]++;
789 	}
790 }
791 
792 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
793 		enum dirty_type dirty_type)
794 {
795 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
796 
797 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
798 		dirty_i->nr_dirty[dirty_type]--;
799 
800 	if (dirty_type == DIRTY) {
801 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
802 		enum dirty_type t = sentry->type;
803 
804 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
805 			dirty_i->nr_dirty[t]--;
806 
807 		if (get_valid_blocks(sbi, segno, true) == 0) {
808 			clear_bit(GET_SEC_FROM_SEG(sbi, segno),
809 						dirty_i->victim_secmap);
810 #ifdef CONFIG_F2FS_CHECK_FS
811 			clear_bit(segno, SIT_I(sbi)->invalid_segmap);
812 #endif
813 		}
814 	}
815 }
816 
817 /*
818  * Should not occur error such as -ENOMEM.
819  * Adding dirty entry into seglist is not critical operation.
820  * If a given segment is one of current working segments, it won't be added.
821  */
822 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
823 {
824 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
825 	unsigned short valid_blocks, ckpt_valid_blocks;
826 
827 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
828 		return;
829 
830 	mutex_lock(&dirty_i->seglist_lock);
831 
832 	valid_blocks = get_valid_blocks(sbi, segno, false);
833 	ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
834 
835 	if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
836 				ckpt_valid_blocks == sbi->blocks_per_seg)) {
837 		__locate_dirty_segment(sbi, segno, PRE);
838 		__remove_dirty_segment(sbi, segno, DIRTY);
839 	} else if (valid_blocks < sbi->blocks_per_seg) {
840 		__locate_dirty_segment(sbi, segno, DIRTY);
841 	} else {
842 		/* Recovery routine with SSR needs this */
843 		__remove_dirty_segment(sbi, segno, DIRTY);
844 	}
845 
846 	mutex_unlock(&dirty_i->seglist_lock);
847 }
848 
849 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
850 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
851 {
852 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
853 	unsigned int segno;
854 
855 	mutex_lock(&dirty_i->seglist_lock);
856 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
857 		if (get_valid_blocks(sbi, segno, false))
858 			continue;
859 		if (IS_CURSEG(sbi, segno))
860 			continue;
861 		__locate_dirty_segment(sbi, segno, PRE);
862 		__remove_dirty_segment(sbi, segno, DIRTY);
863 	}
864 	mutex_unlock(&dirty_i->seglist_lock);
865 }
866 
867 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
868 {
869 	int ovp_hole_segs =
870 		(overprovision_segments(sbi) - reserved_segments(sbi));
871 	block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
872 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
873 	block_t holes[2] = {0, 0};	/* DATA and NODE */
874 	block_t unusable;
875 	struct seg_entry *se;
876 	unsigned int segno;
877 
878 	mutex_lock(&dirty_i->seglist_lock);
879 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
880 		se = get_seg_entry(sbi, segno);
881 		if (IS_NODESEG(se->type))
882 			holes[NODE] += sbi->blocks_per_seg - se->valid_blocks;
883 		else
884 			holes[DATA] += sbi->blocks_per_seg - se->valid_blocks;
885 	}
886 	mutex_unlock(&dirty_i->seglist_lock);
887 
888 	unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
889 	if (unusable > ovp_holes)
890 		return unusable - ovp_holes;
891 	return 0;
892 }
893 
894 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
895 {
896 	int ovp_hole_segs =
897 		(overprovision_segments(sbi) - reserved_segments(sbi));
898 	if (unusable > F2FS_OPTION(sbi).unusable_cap)
899 		return -EAGAIN;
900 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
901 		dirty_segments(sbi) > ovp_hole_segs)
902 		return -EAGAIN;
903 	return 0;
904 }
905 
906 /* This is only used by SBI_CP_DISABLED */
907 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
908 {
909 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
910 	unsigned int segno = 0;
911 
912 	mutex_lock(&dirty_i->seglist_lock);
913 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
914 		if (get_valid_blocks(sbi, segno, false))
915 			continue;
916 		if (get_ckpt_valid_blocks(sbi, segno))
917 			continue;
918 		mutex_unlock(&dirty_i->seglist_lock);
919 		return segno;
920 	}
921 	mutex_unlock(&dirty_i->seglist_lock);
922 	return NULL_SEGNO;
923 }
924 
925 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
926 		struct block_device *bdev, block_t lstart,
927 		block_t start, block_t len)
928 {
929 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
930 	struct list_head *pend_list;
931 	struct discard_cmd *dc;
932 
933 	f2fs_bug_on(sbi, !len);
934 
935 	pend_list = &dcc->pend_list[plist_idx(len)];
936 
937 	dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
938 	INIT_LIST_HEAD(&dc->list);
939 	dc->bdev = bdev;
940 	dc->lstart = lstart;
941 	dc->start = start;
942 	dc->len = len;
943 	dc->ref = 0;
944 	dc->state = D_PREP;
945 	dc->queued = 0;
946 	dc->error = 0;
947 	init_completion(&dc->wait);
948 	list_add_tail(&dc->list, pend_list);
949 	spin_lock_init(&dc->lock);
950 	dc->bio_ref = 0;
951 	atomic_inc(&dcc->discard_cmd_cnt);
952 	dcc->undiscard_blks += len;
953 
954 	return dc;
955 }
956 
957 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
958 				struct block_device *bdev, block_t lstart,
959 				block_t start, block_t len,
960 				struct rb_node *parent, struct rb_node **p,
961 				bool leftmost)
962 {
963 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
964 	struct discard_cmd *dc;
965 
966 	dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
967 
968 	rb_link_node(&dc->rb_node, parent, p);
969 	rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
970 
971 	return dc;
972 }
973 
974 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
975 							struct discard_cmd *dc)
976 {
977 	if (dc->state == D_DONE)
978 		atomic_sub(dc->queued, &dcc->queued_discard);
979 
980 	list_del(&dc->list);
981 	rb_erase_cached(&dc->rb_node, &dcc->root);
982 	dcc->undiscard_blks -= dc->len;
983 
984 	kmem_cache_free(discard_cmd_slab, dc);
985 
986 	atomic_dec(&dcc->discard_cmd_cnt);
987 }
988 
989 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
990 							struct discard_cmd *dc)
991 {
992 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
993 	unsigned long flags;
994 
995 	trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
996 
997 	spin_lock_irqsave(&dc->lock, flags);
998 	if (dc->bio_ref) {
999 		spin_unlock_irqrestore(&dc->lock, flags);
1000 		return;
1001 	}
1002 	spin_unlock_irqrestore(&dc->lock, flags);
1003 
1004 	f2fs_bug_on(sbi, dc->ref);
1005 
1006 	if (dc->error == -EOPNOTSUPP)
1007 		dc->error = 0;
1008 
1009 	if (dc->error)
1010 		printk_ratelimited(
1011 			"%sF2FS-fs: Issue discard(%u, %u, %u) failed, ret: %d",
1012 			KERN_INFO, dc->lstart, dc->start, dc->len, dc->error);
1013 	__detach_discard_cmd(dcc, dc);
1014 }
1015 
1016 static void f2fs_submit_discard_endio(struct bio *bio)
1017 {
1018 	struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1019 	unsigned long flags;
1020 
1021 	dc->error = blk_status_to_errno(bio->bi_status);
1022 
1023 	spin_lock_irqsave(&dc->lock, flags);
1024 	dc->bio_ref--;
1025 	if (!dc->bio_ref && dc->state == D_SUBMIT) {
1026 		dc->state = D_DONE;
1027 		complete_all(&dc->wait);
1028 	}
1029 	spin_unlock_irqrestore(&dc->lock, flags);
1030 	bio_put(bio);
1031 }
1032 
1033 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1034 				block_t start, block_t end)
1035 {
1036 #ifdef CONFIG_F2FS_CHECK_FS
1037 	struct seg_entry *sentry;
1038 	unsigned int segno;
1039 	block_t blk = start;
1040 	unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1041 	unsigned long *map;
1042 
1043 	while (blk < end) {
1044 		segno = GET_SEGNO(sbi, blk);
1045 		sentry = get_seg_entry(sbi, segno);
1046 		offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1047 
1048 		if (end < START_BLOCK(sbi, segno + 1))
1049 			size = GET_BLKOFF_FROM_SEG0(sbi, end);
1050 		else
1051 			size = max_blocks;
1052 		map = (unsigned long *)(sentry->cur_valid_map);
1053 		offset = __find_rev_next_bit(map, size, offset);
1054 		f2fs_bug_on(sbi, offset != size);
1055 		blk = START_BLOCK(sbi, segno + 1);
1056 	}
1057 #endif
1058 }
1059 
1060 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1061 				struct discard_policy *dpolicy,
1062 				int discard_type, unsigned int granularity)
1063 {
1064 	/* common policy */
1065 	dpolicy->type = discard_type;
1066 	dpolicy->sync = true;
1067 	dpolicy->ordered = false;
1068 	dpolicy->granularity = granularity;
1069 
1070 	dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1071 	dpolicy->io_aware_gran = MAX_PLIST_NUM;
1072 	dpolicy->timeout = 0;
1073 
1074 	if (discard_type == DPOLICY_BG) {
1075 		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1076 		dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1077 		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1078 		dpolicy->io_aware = true;
1079 		dpolicy->sync = false;
1080 		dpolicy->ordered = true;
1081 		if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1082 			dpolicy->granularity = 1;
1083 			dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1084 		}
1085 	} else if (discard_type == DPOLICY_FORCE) {
1086 		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1087 		dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1088 		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1089 		dpolicy->io_aware = false;
1090 	} else if (discard_type == DPOLICY_FSTRIM) {
1091 		dpolicy->io_aware = false;
1092 	} else if (discard_type == DPOLICY_UMOUNT) {
1093 		dpolicy->max_requests = UINT_MAX;
1094 		dpolicy->io_aware = false;
1095 		/* we need to issue all to keep CP_TRIMMED_FLAG */
1096 		dpolicy->granularity = 1;
1097 	}
1098 }
1099 
1100 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1101 				struct block_device *bdev, block_t lstart,
1102 				block_t start, block_t len);
1103 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1104 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1105 						struct discard_policy *dpolicy,
1106 						struct discard_cmd *dc,
1107 						unsigned int *issued)
1108 {
1109 	struct block_device *bdev = dc->bdev;
1110 	struct request_queue *q = bdev_get_queue(bdev);
1111 	unsigned int max_discard_blocks =
1112 			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1113 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1114 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1115 					&(dcc->fstrim_list) : &(dcc->wait_list);
1116 	int flag = dpolicy->sync ? REQ_SYNC : 0;
1117 	block_t lstart, start, len, total_len;
1118 	int err = 0;
1119 
1120 	if (dc->state != D_PREP)
1121 		return 0;
1122 
1123 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1124 		return 0;
1125 
1126 	trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1127 
1128 	lstart = dc->lstart;
1129 	start = dc->start;
1130 	len = dc->len;
1131 	total_len = len;
1132 
1133 	dc->len = 0;
1134 
1135 	while (total_len && *issued < dpolicy->max_requests && !err) {
1136 		struct bio *bio = NULL;
1137 		unsigned long flags;
1138 		bool last = true;
1139 
1140 		if (len > max_discard_blocks) {
1141 			len = max_discard_blocks;
1142 			last = false;
1143 		}
1144 
1145 		(*issued)++;
1146 		if (*issued == dpolicy->max_requests)
1147 			last = true;
1148 
1149 		dc->len += len;
1150 
1151 		if (time_to_inject(sbi, FAULT_DISCARD)) {
1152 			f2fs_show_injection_info(FAULT_DISCARD);
1153 			err = -EIO;
1154 			goto submit;
1155 		}
1156 		err = __blkdev_issue_discard(bdev,
1157 					SECTOR_FROM_BLOCK(start),
1158 					SECTOR_FROM_BLOCK(len),
1159 					GFP_NOFS, 0, &bio);
1160 submit:
1161 		if (err) {
1162 			spin_lock_irqsave(&dc->lock, flags);
1163 			if (dc->state == D_PARTIAL)
1164 				dc->state = D_SUBMIT;
1165 			spin_unlock_irqrestore(&dc->lock, flags);
1166 
1167 			break;
1168 		}
1169 
1170 		f2fs_bug_on(sbi, !bio);
1171 
1172 		/*
1173 		 * should keep before submission to avoid D_DONE
1174 		 * right away
1175 		 */
1176 		spin_lock_irqsave(&dc->lock, flags);
1177 		if (last)
1178 			dc->state = D_SUBMIT;
1179 		else
1180 			dc->state = D_PARTIAL;
1181 		dc->bio_ref++;
1182 		spin_unlock_irqrestore(&dc->lock, flags);
1183 
1184 		atomic_inc(&dcc->queued_discard);
1185 		dc->queued++;
1186 		list_move_tail(&dc->list, wait_list);
1187 
1188 		/* sanity check on discard range */
1189 		__check_sit_bitmap(sbi, lstart, lstart + len);
1190 
1191 		bio->bi_private = dc;
1192 		bio->bi_end_io = f2fs_submit_discard_endio;
1193 		bio->bi_opf |= flag;
1194 		submit_bio(bio);
1195 
1196 		atomic_inc(&dcc->issued_discard);
1197 
1198 		f2fs_update_iostat(sbi, FS_DISCARD, 1);
1199 
1200 		lstart += len;
1201 		start += len;
1202 		total_len -= len;
1203 		len = total_len;
1204 	}
1205 
1206 	if (!err && len)
1207 		__update_discard_tree_range(sbi, bdev, lstart, start, len);
1208 	return err;
1209 }
1210 
1211 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1212 				struct block_device *bdev, block_t lstart,
1213 				block_t start, block_t len,
1214 				struct rb_node **insert_p,
1215 				struct rb_node *insert_parent)
1216 {
1217 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1218 	struct rb_node **p;
1219 	struct rb_node *parent = NULL;
1220 	struct discard_cmd *dc = NULL;
1221 	bool leftmost = true;
1222 
1223 	if (insert_p && insert_parent) {
1224 		parent = insert_parent;
1225 		p = insert_p;
1226 		goto do_insert;
1227 	}
1228 
1229 	p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1230 							lstart, &leftmost);
1231 do_insert:
1232 	dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1233 								p, leftmost);
1234 	if (!dc)
1235 		return NULL;
1236 
1237 	return dc;
1238 }
1239 
1240 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1241 						struct discard_cmd *dc)
1242 {
1243 	list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1244 }
1245 
1246 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1247 				struct discard_cmd *dc, block_t blkaddr)
1248 {
1249 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1250 	struct discard_info di = dc->di;
1251 	bool modified = false;
1252 
1253 	if (dc->state == D_DONE || dc->len == 1) {
1254 		__remove_discard_cmd(sbi, dc);
1255 		return;
1256 	}
1257 
1258 	dcc->undiscard_blks -= di.len;
1259 
1260 	if (blkaddr > di.lstart) {
1261 		dc->len = blkaddr - dc->lstart;
1262 		dcc->undiscard_blks += dc->len;
1263 		__relocate_discard_cmd(dcc, dc);
1264 		modified = true;
1265 	}
1266 
1267 	if (blkaddr < di.lstart + di.len - 1) {
1268 		if (modified) {
1269 			__insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1270 					di.start + blkaddr + 1 - di.lstart,
1271 					di.lstart + di.len - 1 - blkaddr,
1272 					NULL, NULL);
1273 		} else {
1274 			dc->lstart++;
1275 			dc->len--;
1276 			dc->start++;
1277 			dcc->undiscard_blks += dc->len;
1278 			__relocate_discard_cmd(dcc, dc);
1279 		}
1280 	}
1281 }
1282 
1283 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1284 				struct block_device *bdev, block_t lstart,
1285 				block_t start, block_t len)
1286 {
1287 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1288 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1289 	struct discard_cmd *dc;
1290 	struct discard_info di = {0};
1291 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1292 	struct request_queue *q = bdev_get_queue(bdev);
1293 	unsigned int max_discard_blocks =
1294 			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1295 	block_t end = lstart + len;
1296 
1297 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1298 					NULL, lstart,
1299 					(struct rb_entry **)&prev_dc,
1300 					(struct rb_entry **)&next_dc,
1301 					&insert_p, &insert_parent, true, NULL);
1302 	if (dc)
1303 		prev_dc = dc;
1304 
1305 	if (!prev_dc) {
1306 		di.lstart = lstart;
1307 		di.len = next_dc ? next_dc->lstart - lstart : len;
1308 		di.len = min(di.len, len);
1309 		di.start = start;
1310 	}
1311 
1312 	while (1) {
1313 		struct rb_node *node;
1314 		bool merged = false;
1315 		struct discard_cmd *tdc = NULL;
1316 
1317 		if (prev_dc) {
1318 			di.lstart = prev_dc->lstart + prev_dc->len;
1319 			if (di.lstart < lstart)
1320 				di.lstart = lstart;
1321 			if (di.lstart >= end)
1322 				break;
1323 
1324 			if (!next_dc || next_dc->lstart > end)
1325 				di.len = end - di.lstart;
1326 			else
1327 				di.len = next_dc->lstart - di.lstart;
1328 			di.start = start + di.lstart - lstart;
1329 		}
1330 
1331 		if (!di.len)
1332 			goto next;
1333 
1334 		if (prev_dc && prev_dc->state == D_PREP &&
1335 			prev_dc->bdev == bdev &&
1336 			__is_discard_back_mergeable(&di, &prev_dc->di,
1337 							max_discard_blocks)) {
1338 			prev_dc->di.len += di.len;
1339 			dcc->undiscard_blks += di.len;
1340 			__relocate_discard_cmd(dcc, prev_dc);
1341 			di = prev_dc->di;
1342 			tdc = prev_dc;
1343 			merged = true;
1344 		}
1345 
1346 		if (next_dc && next_dc->state == D_PREP &&
1347 			next_dc->bdev == bdev &&
1348 			__is_discard_front_mergeable(&di, &next_dc->di,
1349 							max_discard_blocks)) {
1350 			next_dc->di.lstart = di.lstart;
1351 			next_dc->di.len += di.len;
1352 			next_dc->di.start = di.start;
1353 			dcc->undiscard_blks += di.len;
1354 			__relocate_discard_cmd(dcc, next_dc);
1355 			if (tdc)
1356 				__remove_discard_cmd(sbi, tdc);
1357 			merged = true;
1358 		}
1359 
1360 		if (!merged) {
1361 			__insert_discard_tree(sbi, bdev, di.lstart, di.start,
1362 							di.len, NULL, NULL);
1363 		}
1364  next:
1365 		prev_dc = next_dc;
1366 		if (!prev_dc)
1367 			break;
1368 
1369 		node = rb_next(&prev_dc->rb_node);
1370 		next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1371 	}
1372 }
1373 
1374 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1375 		struct block_device *bdev, block_t blkstart, block_t blklen)
1376 {
1377 	block_t lblkstart = blkstart;
1378 
1379 	if (!f2fs_bdev_support_discard(bdev))
1380 		return 0;
1381 
1382 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
1383 
1384 	if (f2fs_is_multi_device(sbi)) {
1385 		int devi = f2fs_target_device_index(sbi, blkstart);
1386 
1387 		blkstart -= FDEV(devi).start_blk;
1388 	}
1389 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1390 	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1391 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1392 	return 0;
1393 }
1394 
1395 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1396 					struct discard_policy *dpolicy)
1397 {
1398 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1399 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1400 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1401 	struct discard_cmd *dc;
1402 	struct blk_plug plug;
1403 	unsigned int pos = dcc->next_pos;
1404 	unsigned int issued = 0;
1405 	bool io_interrupted = false;
1406 
1407 	mutex_lock(&dcc->cmd_lock);
1408 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1409 					NULL, pos,
1410 					(struct rb_entry **)&prev_dc,
1411 					(struct rb_entry **)&next_dc,
1412 					&insert_p, &insert_parent, true, NULL);
1413 	if (!dc)
1414 		dc = next_dc;
1415 
1416 	blk_start_plug(&plug);
1417 
1418 	while (dc) {
1419 		struct rb_node *node;
1420 		int err = 0;
1421 
1422 		if (dc->state != D_PREP)
1423 			goto next;
1424 
1425 		if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1426 			io_interrupted = true;
1427 			break;
1428 		}
1429 
1430 		dcc->next_pos = dc->lstart + dc->len;
1431 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1432 
1433 		if (issued >= dpolicy->max_requests)
1434 			break;
1435 next:
1436 		node = rb_next(&dc->rb_node);
1437 		if (err)
1438 			__remove_discard_cmd(sbi, dc);
1439 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1440 	}
1441 
1442 	blk_finish_plug(&plug);
1443 
1444 	if (!dc)
1445 		dcc->next_pos = 0;
1446 
1447 	mutex_unlock(&dcc->cmd_lock);
1448 
1449 	if (!issued && io_interrupted)
1450 		issued = -1;
1451 
1452 	return issued;
1453 }
1454 
1455 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1456 					struct discard_policy *dpolicy)
1457 {
1458 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1459 	struct list_head *pend_list;
1460 	struct discard_cmd *dc, *tmp;
1461 	struct blk_plug plug;
1462 	int i, issued = 0;
1463 	bool io_interrupted = false;
1464 
1465 	if (dpolicy->timeout != 0)
1466 		f2fs_update_time(sbi, dpolicy->timeout);
1467 
1468 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1469 		if (dpolicy->timeout != 0 &&
1470 				f2fs_time_over(sbi, dpolicy->timeout))
1471 			break;
1472 
1473 		if (i + 1 < dpolicy->granularity)
1474 			break;
1475 
1476 		if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1477 			return __issue_discard_cmd_orderly(sbi, dpolicy);
1478 
1479 		pend_list = &dcc->pend_list[i];
1480 
1481 		mutex_lock(&dcc->cmd_lock);
1482 		if (list_empty(pend_list))
1483 			goto next;
1484 		if (unlikely(dcc->rbtree_check))
1485 			f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1486 								&dcc->root));
1487 		blk_start_plug(&plug);
1488 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1489 			f2fs_bug_on(sbi, dc->state != D_PREP);
1490 
1491 			if (dpolicy->timeout != 0 &&
1492 				f2fs_time_over(sbi, dpolicy->timeout))
1493 				break;
1494 
1495 			if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1496 						!is_idle(sbi, DISCARD_TIME)) {
1497 				io_interrupted = true;
1498 				break;
1499 			}
1500 
1501 			__submit_discard_cmd(sbi, dpolicy, dc, &issued);
1502 
1503 			if (issued >= dpolicy->max_requests)
1504 				break;
1505 		}
1506 		blk_finish_plug(&plug);
1507 next:
1508 		mutex_unlock(&dcc->cmd_lock);
1509 
1510 		if (issued >= dpolicy->max_requests || io_interrupted)
1511 			break;
1512 	}
1513 
1514 	if (!issued && io_interrupted)
1515 		issued = -1;
1516 
1517 	return issued;
1518 }
1519 
1520 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1521 {
1522 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1523 	struct list_head *pend_list;
1524 	struct discard_cmd *dc, *tmp;
1525 	int i;
1526 	bool dropped = false;
1527 
1528 	mutex_lock(&dcc->cmd_lock);
1529 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1530 		pend_list = &dcc->pend_list[i];
1531 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1532 			f2fs_bug_on(sbi, dc->state != D_PREP);
1533 			__remove_discard_cmd(sbi, dc);
1534 			dropped = true;
1535 		}
1536 	}
1537 	mutex_unlock(&dcc->cmd_lock);
1538 
1539 	return dropped;
1540 }
1541 
1542 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1543 {
1544 	__drop_discard_cmd(sbi);
1545 }
1546 
1547 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1548 							struct discard_cmd *dc)
1549 {
1550 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1551 	unsigned int len = 0;
1552 
1553 	wait_for_completion_io(&dc->wait);
1554 	mutex_lock(&dcc->cmd_lock);
1555 	f2fs_bug_on(sbi, dc->state != D_DONE);
1556 	dc->ref--;
1557 	if (!dc->ref) {
1558 		if (!dc->error)
1559 			len = dc->len;
1560 		__remove_discard_cmd(sbi, dc);
1561 	}
1562 	mutex_unlock(&dcc->cmd_lock);
1563 
1564 	return len;
1565 }
1566 
1567 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1568 						struct discard_policy *dpolicy,
1569 						block_t start, block_t end)
1570 {
1571 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1572 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1573 					&(dcc->fstrim_list) : &(dcc->wait_list);
1574 	struct discard_cmd *dc, *tmp;
1575 	bool need_wait;
1576 	unsigned int trimmed = 0;
1577 
1578 next:
1579 	need_wait = false;
1580 
1581 	mutex_lock(&dcc->cmd_lock);
1582 	list_for_each_entry_safe(dc, tmp, wait_list, list) {
1583 		if (dc->lstart + dc->len <= start || end <= dc->lstart)
1584 			continue;
1585 		if (dc->len < dpolicy->granularity)
1586 			continue;
1587 		if (dc->state == D_DONE && !dc->ref) {
1588 			wait_for_completion_io(&dc->wait);
1589 			if (!dc->error)
1590 				trimmed += dc->len;
1591 			__remove_discard_cmd(sbi, dc);
1592 		} else {
1593 			dc->ref++;
1594 			need_wait = true;
1595 			break;
1596 		}
1597 	}
1598 	mutex_unlock(&dcc->cmd_lock);
1599 
1600 	if (need_wait) {
1601 		trimmed += __wait_one_discard_bio(sbi, dc);
1602 		goto next;
1603 	}
1604 
1605 	return trimmed;
1606 }
1607 
1608 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1609 						struct discard_policy *dpolicy)
1610 {
1611 	struct discard_policy dp;
1612 	unsigned int discard_blks;
1613 
1614 	if (dpolicy)
1615 		return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1616 
1617 	/* wait all */
1618 	__init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1619 	discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1620 	__init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1621 	discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1622 
1623 	return discard_blks;
1624 }
1625 
1626 /* This should be covered by global mutex, &sit_i->sentry_lock */
1627 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1628 {
1629 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1630 	struct discard_cmd *dc;
1631 	bool need_wait = false;
1632 
1633 	mutex_lock(&dcc->cmd_lock);
1634 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1635 							NULL, blkaddr);
1636 	if (dc) {
1637 		if (dc->state == D_PREP) {
1638 			__punch_discard_cmd(sbi, dc, blkaddr);
1639 		} else {
1640 			dc->ref++;
1641 			need_wait = true;
1642 		}
1643 	}
1644 	mutex_unlock(&dcc->cmd_lock);
1645 
1646 	if (need_wait)
1647 		__wait_one_discard_bio(sbi, dc);
1648 }
1649 
1650 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1651 {
1652 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1653 
1654 	if (dcc && dcc->f2fs_issue_discard) {
1655 		struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1656 
1657 		dcc->f2fs_issue_discard = NULL;
1658 		kthread_stop(discard_thread);
1659 	}
1660 }
1661 
1662 /* This comes from f2fs_put_super */
1663 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1664 {
1665 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1666 	struct discard_policy dpolicy;
1667 	bool dropped;
1668 
1669 	__init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1670 					dcc->discard_granularity);
1671 	dpolicy.timeout = UMOUNT_DISCARD_TIMEOUT;
1672 	__issue_discard_cmd(sbi, &dpolicy);
1673 	dropped = __drop_discard_cmd(sbi);
1674 
1675 	/* just to make sure there is no pending discard commands */
1676 	__wait_all_discard_cmd(sbi, NULL);
1677 
1678 	f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1679 	return dropped;
1680 }
1681 
1682 static int issue_discard_thread(void *data)
1683 {
1684 	struct f2fs_sb_info *sbi = data;
1685 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1686 	wait_queue_head_t *q = &dcc->discard_wait_queue;
1687 	struct discard_policy dpolicy;
1688 	unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1689 	int issued;
1690 
1691 	set_freezable();
1692 
1693 	do {
1694 		__init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1695 					dcc->discard_granularity);
1696 
1697 		wait_event_interruptible_timeout(*q,
1698 				kthread_should_stop() || freezing(current) ||
1699 				dcc->discard_wake,
1700 				msecs_to_jiffies(wait_ms));
1701 
1702 		if (dcc->discard_wake)
1703 			dcc->discard_wake = 0;
1704 
1705 		/* clean up pending candidates before going to sleep */
1706 		if (atomic_read(&dcc->queued_discard))
1707 			__wait_all_discard_cmd(sbi, NULL);
1708 
1709 		if (try_to_freeze())
1710 			continue;
1711 		if (f2fs_readonly(sbi->sb))
1712 			continue;
1713 		if (kthread_should_stop())
1714 			return 0;
1715 		if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1716 			wait_ms = dpolicy.max_interval;
1717 			continue;
1718 		}
1719 
1720 		if (sbi->gc_mode == GC_URGENT)
1721 			__init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1722 
1723 		sb_start_intwrite(sbi->sb);
1724 
1725 		issued = __issue_discard_cmd(sbi, &dpolicy);
1726 		if (issued > 0) {
1727 			__wait_all_discard_cmd(sbi, &dpolicy);
1728 			wait_ms = dpolicy.min_interval;
1729 		} else if (issued == -1){
1730 			wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1731 			if (!wait_ms)
1732 				wait_ms = dpolicy.mid_interval;
1733 		} else {
1734 			wait_ms = dpolicy.max_interval;
1735 		}
1736 
1737 		sb_end_intwrite(sbi->sb);
1738 
1739 	} while (!kthread_should_stop());
1740 	return 0;
1741 }
1742 
1743 #ifdef CONFIG_BLK_DEV_ZONED
1744 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1745 		struct block_device *bdev, block_t blkstart, block_t blklen)
1746 {
1747 	sector_t sector, nr_sects;
1748 	block_t lblkstart = blkstart;
1749 	int devi = 0;
1750 
1751 	if (f2fs_is_multi_device(sbi)) {
1752 		devi = f2fs_target_device_index(sbi, blkstart);
1753 		if (blkstart < FDEV(devi).start_blk ||
1754 		    blkstart > FDEV(devi).end_blk) {
1755 			f2fs_err(sbi, "Invalid block %x", blkstart);
1756 			return -EIO;
1757 		}
1758 		blkstart -= FDEV(devi).start_blk;
1759 	}
1760 
1761 	/* For sequential zones, reset the zone write pointer */
1762 	if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1763 		sector = SECTOR_FROM_BLOCK(blkstart);
1764 		nr_sects = SECTOR_FROM_BLOCK(blklen);
1765 
1766 		if (sector & (bdev_zone_sectors(bdev) - 1) ||
1767 				nr_sects != bdev_zone_sectors(bdev)) {
1768 			f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1769 				 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1770 				 blkstart, blklen);
1771 			return -EIO;
1772 		}
1773 		trace_f2fs_issue_reset_zone(bdev, blkstart);
1774 		return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1775 					sector, nr_sects, GFP_NOFS);
1776 	}
1777 
1778 	/* For conventional zones, use regular discard if supported */
1779 	return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1780 }
1781 #endif
1782 
1783 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1784 		struct block_device *bdev, block_t blkstart, block_t blklen)
1785 {
1786 #ifdef CONFIG_BLK_DEV_ZONED
1787 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1788 		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1789 #endif
1790 	return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1791 }
1792 
1793 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1794 				block_t blkstart, block_t blklen)
1795 {
1796 	sector_t start = blkstart, len = 0;
1797 	struct block_device *bdev;
1798 	struct seg_entry *se;
1799 	unsigned int offset;
1800 	block_t i;
1801 	int err = 0;
1802 
1803 	bdev = f2fs_target_device(sbi, blkstart, NULL);
1804 
1805 	for (i = blkstart; i < blkstart + blklen; i++, len++) {
1806 		if (i != start) {
1807 			struct block_device *bdev2 =
1808 				f2fs_target_device(sbi, i, NULL);
1809 
1810 			if (bdev2 != bdev) {
1811 				err = __issue_discard_async(sbi, bdev,
1812 						start, len);
1813 				if (err)
1814 					return err;
1815 				bdev = bdev2;
1816 				start = i;
1817 				len = 0;
1818 			}
1819 		}
1820 
1821 		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1822 		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1823 
1824 		if (!f2fs_test_and_set_bit(offset, se->discard_map))
1825 			sbi->discard_blks--;
1826 	}
1827 
1828 	if (len)
1829 		err = __issue_discard_async(sbi, bdev, start, len);
1830 	return err;
1831 }
1832 
1833 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1834 							bool check_only)
1835 {
1836 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1837 	int max_blocks = sbi->blocks_per_seg;
1838 	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1839 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1840 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1841 	unsigned long *discard_map = (unsigned long *)se->discard_map;
1842 	unsigned long *dmap = SIT_I(sbi)->tmp_map;
1843 	unsigned int start = 0, end = -1;
1844 	bool force = (cpc->reason & CP_DISCARD);
1845 	struct discard_entry *de = NULL;
1846 	struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1847 	int i;
1848 
1849 	if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1850 		return false;
1851 
1852 	if (!force) {
1853 		if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1854 			SM_I(sbi)->dcc_info->nr_discards >=
1855 				SM_I(sbi)->dcc_info->max_discards)
1856 			return false;
1857 	}
1858 
1859 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1860 	for (i = 0; i < entries; i++)
1861 		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1862 				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1863 
1864 	while (force || SM_I(sbi)->dcc_info->nr_discards <=
1865 				SM_I(sbi)->dcc_info->max_discards) {
1866 		start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1867 		if (start >= max_blocks)
1868 			break;
1869 
1870 		end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1871 		if (force && start && end != max_blocks
1872 					&& (end - start) < cpc->trim_minlen)
1873 			continue;
1874 
1875 		if (check_only)
1876 			return true;
1877 
1878 		if (!de) {
1879 			de = f2fs_kmem_cache_alloc(discard_entry_slab,
1880 								GFP_F2FS_ZERO);
1881 			de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1882 			list_add_tail(&de->list, head);
1883 		}
1884 
1885 		for (i = start; i < end; i++)
1886 			__set_bit_le(i, (void *)de->discard_map);
1887 
1888 		SM_I(sbi)->dcc_info->nr_discards += end - start;
1889 	}
1890 	return false;
1891 }
1892 
1893 static void release_discard_addr(struct discard_entry *entry)
1894 {
1895 	list_del(&entry->list);
1896 	kmem_cache_free(discard_entry_slab, entry);
1897 }
1898 
1899 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1900 {
1901 	struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1902 	struct discard_entry *entry, *this;
1903 
1904 	/* drop caches */
1905 	list_for_each_entry_safe(entry, this, head, list)
1906 		release_discard_addr(entry);
1907 }
1908 
1909 /*
1910  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1911  */
1912 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1913 {
1914 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1915 	unsigned int segno;
1916 
1917 	mutex_lock(&dirty_i->seglist_lock);
1918 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1919 		__set_test_and_free(sbi, segno);
1920 	mutex_unlock(&dirty_i->seglist_lock);
1921 }
1922 
1923 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1924 						struct cp_control *cpc)
1925 {
1926 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1927 	struct list_head *head = &dcc->entry_list;
1928 	struct discard_entry *entry, *this;
1929 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1930 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1931 	unsigned int start = 0, end = -1;
1932 	unsigned int secno, start_segno;
1933 	bool force = (cpc->reason & CP_DISCARD);
1934 	bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
1935 
1936 	mutex_lock(&dirty_i->seglist_lock);
1937 
1938 	while (1) {
1939 		int i;
1940 
1941 		if (need_align && end != -1)
1942 			end--;
1943 		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1944 		if (start >= MAIN_SEGS(sbi))
1945 			break;
1946 		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1947 								start + 1);
1948 
1949 		if (need_align) {
1950 			start = rounddown(start, sbi->segs_per_sec);
1951 			end = roundup(end, sbi->segs_per_sec);
1952 		}
1953 
1954 		for (i = start; i < end; i++) {
1955 			if (test_and_clear_bit(i, prefree_map))
1956 				dirty_i->nr_dirty[PRE]--;
1957 		}
1958 
1959 		if (!f2fs_realtime_discard_enable(sbi))
1960 			continue;
1961 
1962 		if (force && start >= cpc->trim_start &&
1963 					(end - 1) <= cpc->trim_end)
1964 				continue;
1965 
1966 		if (!test_opt(sbi, LFS) || !__is_large_section(sbi)) {
1967 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1968 				(end - start) << sbi->log_blocks_per_seg);
1969 			continue;
1970 		}
1971 next:
1972 		secno = GET_SEC_FROM_SEG(sbi, start);
1973 		start_segno = GET_SEG_FROM_SEC(sbi, secno);
1974 		if (!IS_CURSEC(sbi, secno) &&
1975 			!get_valid_blocks(sbi, start, true))
1976 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1977 				sbi->segs_per_sec << sbi->log_blocks_per_seg);
1978 
1979 		start = start_segno + sbi->segs_per_sec;
1980 		if (start < end)
1981 			goto next;
1982 		else
1983 			end = start - 1;
1984 	}
1985 	mutex_unlock(&dirty_i->seglist_lock);
1986 
1987 	/* send small discards */
1988 	list_for_each_entry_safe(entry, this, head, list) {
1989 		unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1990 		bool is_valid = test_bit_le(0, entry->discard_map);
1991 
1992 find_next:
1993 		if (is_valid) {
1994 			next_pos = find_next_zero_bit_le(entry->discard_map,
1995 					sbi->blocks_per_seg, cur_pos);
1996 			len = next_pos - cur_pos;
1997 
1998 			if (f2fs_sb_has_blkzoned(sbi) ||
1999 			    (force && len < cpc->trim_minlen))
2000 				goto skip;
2001 
2002 			f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2003 									len);
2004 			total_len += len;
2005 		} else {
2006 			next_pos = find_next_bit_le(entry->discard_map,
2007 					sbi->blocks_per_seg, cur_pos);
2008 		}
2009 skip:
2010 		cur_pos = next_pos;
2011 		is_valid = !is_valid;
2012 
2013 		if (cur_pos < sbi->blocks_per_seg)
2014 			goto find_next;
2015 
2016 		release_discard_addr(entry);
2017 		dcc->nr_discards -= total_len;
2018 	}
2019 
2020 	wake_up_discard_thread(sbi, false);
2021 }
2022 
2023 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2024 {
2025 	dev_t dev = sbi->sb->s_bdev->bd_dev;
2026 	struct discard_cmd_control *dcc;
2027 	int err = 0, i;
2028 
2029 	if (SM_I(sbi)->dcc_info) {
2030 		dcc = SM_I(sbi)->dcc_info;
2031 		goto init_thread;
2032 	}
2033 
2034 	dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2035 	if (!dcc)
2036 		return -ENOMEM;
2037 
2038 	dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2039 	INIT_LIST_HEAD(&dcc->entry_list);
2040 	for (i = 0; i < MAX_PLIST_NUM; i++)
2041 		INIT_LIST_HEAD(&dcc->pend_list[i]);
2042 	INIT_LIST_HEAD(&dcc->wait_list);
2043 	INIT_LIST_HEAD(&dcc->fstrim_list);
2044 	mutex_init(&dcc->cmd_lock);
2045 	atomic_set(&dcc->issued_discard, 0);
2046 	atomic_set(&dcc->queued_discard, 0);
2047 	atomic_set(&dcc->discard_cmd_cnt, 0);
2048 	dcc->nr_discards = 0;
2049 	dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2050 	dcc->undiscard_blks = 0;
2051 	dcc->next_pos = 0;
2052 	dcc->root = RB_ROOT_CACHED;
2053 	dcc->rbtree_check = false;
2054 
2055 	init_waitqueue_head(&dcc->discard_wait_queue);
2056 	SM_I(sbi)->dcc_info = dcc;
2057 init_thread:
2058 	dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2059 				"f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2060 	if (IS_ERR(dcc->f2fs_issue_discard)) {
2061 		err = PTR_ERR(dcc->f2fs_issue_discard);
2062 		kvfree(dcc);
2063 		SM_I(sbi)->dcc_info = NULL;
2064 		return err;
2065 	}
2066 
2067 	return err;
2068 }
2069 
2070 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2071 {
2072 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2073 
2074 	if (!dcc)
2075 		return;
2076 
2077 	f2fs_stop_discard_thread(sbi);
2078 
2079 	/*
2080 	 * Recovery can cache discard commands, so in error path of
2081 	 * fill_super(), it needs to give a chance to handle them.
2082 	 */
2083 	if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2084 		f2fs_issue_discard_timeout(sbi);
2085 
2086 	kvfree(dcc);
2087 	SM_I(sbi)->dcc_info = NULL;
2088 }
2089 
2090 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2091 {
2092 	struct sit_info *sit_i = SIT_I(sbi);
2093 
2094 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2095 		sit_i->dirty_sentries++;
2096 		return false;
2097 	}
2098 
2099 	return true;
2100 }
2101 
2102 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2103 					unsigned int segno, int modified)
2104 {
2105 	struct seg_entry *se = get_seg_entry(sbi, segno);
2106 	se->type = type;
2107 	if (modified)
2108 		__mark_sit_entry_dirty(sbi, segno);
2109 }
2110 
2111 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2112 {
2113 	struct seg_entry *se;
2114 	unsigned int segno, offset;
2115 	long int new_vblocks;
2116 	bool exist;
2117 #ifdef CONFIG_F2FS_CHECK_FS
2118 	bool mir_exist;
2119 #endif
2120 
2121 	segno = GET_SEGNO(sbi, blkaddr);
2122 
2123 	se = get_seg_entry(sbi, segno);
2124 	new_vblocks = se->valid_blocks + del;
2125 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2126 
2127 	f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
2128 				(new_vblocks > sbi->blocks_per_seg)));
2129 
2130 	se->valid_blocks = new_vblocks;
2131 	se->mtime = get_mtime(sbi, false);
2132 	if (se->mtime > SIT_I(sbi)->max_mtime)
2133 		SIT_I(sbi)->max_mtime = se->mtime;
2134 
2135 	/* Update valid block bitmap */
2136 	if (del > 0) {
2137 		exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2138 #ifdef CONFIG_F2FS_CHECK_FS
2139 		mir_exist = f2fs_test_and_set_bit(offset,
2140 						se->cur_valid_map_mir);
2141 		if (unlikely(exist != mir_exist)) {
2142 			f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2143 				 blkaddr, exist);
2144 			f2fs_bug_on(sbi, 1);
2145 		}
2146 #endif
2147 		if (unlikely(exist)) {
2148 			f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2149 				 blkaddr);
2150 			f2fs_bug_on(sbi, 1);
2151 			se->valid_blocks--;
2152 			del = 0;
2153 		}
2154 
2155 		if (!f2fs_test_and_set_bit(offset, se->discard_map))
2156 			sbi->discard_blks--;
2157 
2158 		/*
2159 		 * SSR should never reuse block which is checkpointed
2160 		 * or newly invalidated.
2161 		 */
2162 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2163 			if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2164 				se->ckpt_valid_blocks++;
2165 		}
2166 	} else {
2167 		exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2168 #ifdef CONFIG_F2FS_CHECK_FS
2169 		mir_exist = f2fs_test_and_clear_bit(offset,
2170 						se->cur_valid_map_mir);
2171 		if (unlikely(exist != mir_exist)) {
2172 			f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2173 				 blkaddr, exist);
2174 			f2fs_bug_on(sbi, 1);
2175 		}
2176 #endif
2177 		if (unlikely(!exist)) {
2178 			f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2179 				 blkaddr);
2180 			f2fs_bug_on(sbi, 1);
2181 			se->valid_blocks++;
2182 			del = 0;
2183 		} else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2184 			/*
2185 			 * If checkpoints are off, we must not reuse data that
2186 			 * was used in the previous checkpoint. If it was used
2187 			 * before, we must track that to know how much space we
2188 			 * really have.
2189 			 */
2190 			if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2191 				spin_lock(&sbi->stat_lock);
2192 				sbi->unusable_block_count++;
2193 				spin_unlock(&sbi->stat_lock);
2194 			}
2195 		}
2196 
2197 		if (f2fs_test_and_clear_bit(offset, se->discard_map))
2198 			sbi->discard_blks++;
2199 	}
2200 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2201 		se->ckpt_valid_blocks += del;
2202 
2203 	__mark_sit_entry_dirty(sbi, segno);
2204 
2205 	/* update total number of valid blocks to be written in ckpt area */
2206 	SIT_I(sbi)->written_valid_blocks += del;
2207 
2208 	if (__is_large_section(sbi))
2209 		get_sec_entry(sbi, segno)->valid_blocks += del;
2210 }
2211 
2212 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2213 {
2214 	unsigned int segno = GET_SEGNO(sbi, addr);
2215 	struct sit_info *sit_i = SIT_I(sbi);
2216 
2217 	f2fs_bug_on(sbi, addr == NULL_ADDR);
2218 	if (addr == NEW_ADDR)
2219 		return;
2220 
2221 	invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2222 
2223 	/* add it into sit main buffer */
2224 	down_write(&sit_i->sentry_lock);
2225 
2226 	update_sit_entry(sbi, addr, -1);
2227 
2228 	/* add it into dirty seglist */
2229 	locate_dirty_segment(sbi, segno);
2230 
2231 	up_write(&sit_i->sentry_lock);
2232 }
2233 
2234 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2235 {
2236 	struct sit_info *sit_i = SIT_I(sbi);
2237 	unsigned int segno, offset;
2238 	struct seg_entry *se;
2239 	bool is_cp = false;
2240 
2241 	if (!__is_valid_data_blkaddr(blkaddr))
2242 		return true;
2243 
2244 	down_read(&sit_i->sentry_lock);
2245 
2246 	segno = GET_SEGNO(sbi, blkaddr);
2247 	se = get_seg_entry(sbi, segno);
2248 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2249 
2250 	if (f2fs_test_bit(offset, se->ckpt_valid_map))
2251 		is_cp = true;
2252 
2253 	up_read(&sit_i->sentry_lock);
2254 
2255 	return is_cp;
2256 }
2257 
2258 /*
2259  * This function should be resided under the curseg_mutex lock
2260  */
2261 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2262 					struct f2fs_summary *sum)
2263 {
2264 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2265 	void *addr = curseg->sum_blk;
2266 	addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2267 	memcpy(addr, sum, sizeof(struct f2fs_summary));
2268 }
2269 
2270 /*
2271  * Calculate the number of current summary pages for writing
2272  */
2273 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2274 {
2275 	int valid_sum_count = 0;
2276 	int i, sum_in_page;
2277 
2278 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2279 		if (sbi->ckpt->alloc_type[i] == SSR)
2280 			valid_sum_count += sbi->blocks_per_seg;
2281 		else {
2282 			if (for_ra)
2283 				valid_sum_count += le16_to_cpu(
2284 					F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2285 			else
2286 				valid_sum_count += curseg_blkoff(sbi, i);
2287 		}
2288 	}
2289 
2290 	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2291 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2292 	if (valid_sum_count <= sum_in_page)
2293 		return 1;
2294 	else if ((valid_sum_count - sum_in_page) <=
2295 		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2296 		return 2;
2297 	return 3;
2298 }
2299 
2300 /*
2301  * Caller should put this summary page
2302  */
2303 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2304 {
2305 	return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2306 }
2307 
2308 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2309 					void *src, block_t blk_addr)
2310 {
2311 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2312 
2313 	memcpy(page_address(page), src, PAGE_SIZE);
2314 	set_page_dirty(page);
2315 	f2fs_put_page(page, 1);
2316 }
2317 
2318 static void write_sum_page(struct f2fs_sb_info *sbi,
2319 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
2320 {
2321 	f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2322 }
2323 
2324 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2325 						int type, block_t blk_addr)
2326 {
2327 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2328 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2329 	struct f2fs_summary_block *src = curseg->sum_blk;
2330 	struct f2fs_summary_block *dst;
2331 
2332 	dst = (struct f2fs_summary_block *)page_address(page);
2333 	memset(dst, 0, PAGE_SIZE);
2334 
2335 	mutex_lock(&curseg->curseg_mutex);
2336 
2337 	down_read(&curseg->journal_rwsem);
2338 	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2339 	up_read(&curseg->journal_rwsem);
2340 
2341 	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2342 	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2343 
2344 	mutex_unlock(&curseg->curseg_mutex);
2345 
2346 	set_page_dirty(page);
2347 	f2fs_put_page(page, 1);
2348 }
2349 
2350 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2351 {
2352 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2353 	unsigned int segno = curseg->segno + 1;
2354 	struct free_segmap_info *free_i = FREE_I(sbi);
2355 
2356 	if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2357 		return !test_bit(segno, free_i->free_segmap);
2358 	return 0;
2359 }
2360 
2361 /*
2362  * Find a new segment from the free segments bitmap to right order
2363  * This function should be returned with success, otherwise BUG
2364  */
2365 static void get_new_segment(struct f2fs_sb_info *sbi,
2366 			unsigned int *newseg, bool new_sec, int dir)
2367 {
2368 	struct free_segmap_info *free_i = FREE_I(sbi);
2369 	unsigned int segno, secno, zoneno;
2370 	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2371 	unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2372 	unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2373 	unsigned int left_start = hint;
2374 	bool init = true;
2375 	int go_left = 0;
2376 	int i;
2377 
2378 	spin_lock(&free_i->segmap_lock);
2379 
2380 	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2381 		segno = find_next_zero_bit(free_i->free_segmap,
2382 			GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2383 		if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2384 			goto got_it;
2385 	}
2386 find_other_zone:
2387 	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2388 	if (secno >= MAIN_SECS(sbi)) {
2389 		if (dir == ALLOC_RIGHT) {
2390 			secno = find_next_zero_bit(free_i->free_secmap,
2391 							MAIN_SECS(sbi), 0);
2392 			f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2393 		} else {
2394 			go_left = 1;
2395 			left_start = hint - 1;
2396 		}
2397 	}
2398 	if (go_left == 0)
2399 		goto skip_left;
2400 
2401 	while (test_bit(left_start, free_i->free_secmap)) {
2402 		if (left_start > 0) {
2403 			left_start--;
2404 			continue;
2405 		}
2406 		left_start = find_next_zero_bit(free_i->free_secmap,
2407 							MAIN_SECS(sbi), 0);
2408 		f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2409 		break;
2410 	}
2411 	secno = left_start;
2412 skip_left:
2413 	segno = GET_SEG_FROM_SEC(sbi, secno);
2414 	zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2415 
2416 	/* give up on finding another zone */
2417 	if (!init)
2418 		goto got_it;
2419 	if (sbi->secs_per_zone == 1)
2420 		goto got_it;
2421 	if (zoneno == old_zoneno)
2422 		goto got_it;
2423 	if (dir == ALLOC_LEFT) {
2424 		if (!go_left && zoneno + 1 >= total_zones)
2425 			goto got_it;
2426 		if (go_left && zoneno == 0)
2427 			goto got_it;
2428 	}
2429 	for (i = 0; i < NR_CURSEG_TYPE; i++)
2430 		if (CURSEG_I(sbi, i)->zone == zoneno)
2431 			break;
2432 
2433 	if (i < NR_CURSEG_TYPE) {
2434 		/* zone is in user, try another */
2435 		if (go_left)
2436 			hint = zoneno * sbi->secs_per_zone - 1;
2437 		else if (zoneno + 1 >= total_zones)
2438 			hint = 0;
2439 		else
2440 			hint = (zoneno + 1) * sbi->secs_per_zone;
2441 		init = false;
2442 		goto find_other_zone;
2443 	}
2444 got_it:
2445 	/* set it as dirty segment in free segmap */
2446 	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2447 	__set_inuse(sbi, segno);
2448 	*newseg = segno;
2449 	spin_unlock(&free_i->segmap_lock);
2450 }
2451 
2452 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2453 {
2454 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2455 	struct summary_footer *sum_footer;
2456 
2457 	curseg->segno = curseg->next_segno;
2458 	curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2459 	curseg->next_blkoff = 0;
2460 	curseg->next_segno = NULL_SEGNO;
2461 
2462 	sum_footer = &(curseg->sum_blk->footer);
2463 	memset(sum_footer, 0, sizeof(struct summary_footer));
2464 	if (IS_DATASEG(type))
2465 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2466 	if (IS_NODESEG(type))
2467 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2468 	__set_sit_entry_type(sbi, type, curseg->segno, modified);
2469 }
2470 
2471 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2472 {
2473 	/* if segs_per_sec is large than 1, we need to keep original policy. */
2474 	if (__is_large_section(sbi))
2475 		return CURSEG_I(sbi, type)->segno;
2476 
2477 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2478 		return 0;
2479 
2480 	if (test_opt(sbi, NOHEAP) &&
2481 		(type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2482 		return 0;
2483 
2484 	if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2485 		return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2486 
2487 	/* find segments from 0 to reuse freed segments */
2488 	if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2489 		return 0;
2490 
2491 	return CURSEG_I(sbi, type)->segno;
2492 }
2493 
2494 /*
2495  * Allocate a current working segment.
2496  * This function always allocates a free segment in LFS manner.
2497  */
2498 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2499 {
2500 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2501 	unsigned int segno = curseg->segno;
2502 	int dir = ALLOC_LEFT;
2503 
2504 	write_sum_page(sbi, curseg->sum_blk,
2505 				GET_SUM_BLOCK(sbi, segno));
2506 	if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2507 		dir = ALLOC_RIGHT;
2508 
2509 	if (test_opt(sbi, NOHEAP))
2510 		dir = ALLOC_RIGHT;
2511 
2512 	segno = __get_next_segno(sbi, type);
2513 	get_new_segment(sbi, &segno, new_sec, dir);
2514 	curseg->next_segno = segno;
2515 	reset_curseg(sbi, type, 1);
2516 	curseg->alloc_type = LFS;
2517 }
2518 
2519 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2520 			struct curseg_info *seg, block_t start)
2521 {
2522 	struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2523 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2524 	unsigned long *target_map = SIT_I(sbi)->tmp_map;
2525 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2526 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2527 	int i, pos;
2528 
2529 	for (i = 0; i < entries; i++)
2530 		target_map[i] = ckpt_map[i] | cur_map[i];
2531 
2532 	pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2533 
2534 	seg->next_blkoff = pos;
2535 }
2536 
2537 /*
2538  * If a segment is written by LFS manner, next block offset is just obtained
2539  * by increasing the current block offset. However, if a segment is written by
2540  * SSR manner, next block offset obtained by calling __next_free_blkoff
2541  */
2542 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2543 				struct curseg_info *seg)
2544 {
2545 	if (seg->alloc_type == SSR)
2546 		__next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2547 	else
2548 		seg->next_blkoff++;
2549 }
2550 
2551 /*
2552  * This function always allocates a used segment(from dirty seglist) by SSR
2553  * manner, so it should recover the existing segment information of valid blocks
2554  */
2555 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2556 {
2557 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2558 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2559 	unsigned int new_segno = curseg->next_segno;
2560 	struct f2fs_summary_block *sum_node;
2561 	struct page *sum_page;
2562 
2563 	write_sum_page(sbi, curseg->sum_blk,
2564 				GET_SUM_BLOCK(sbi, curseg->segno));
2565 	__set_test_and_inuse(sbi, new_segno);
2566 
2567 	mutex_lock(&dirty_i->seglist_lock);
2568 	__remove_dirty_segment(sbi, new_segno, PRE);
2569 	__remove_dirty_segment(sbi, new_segno, DIRTY);
2570 	mutex_unlock(&dirty_i->seglist_lock);
2571 
2572 	reset_curseg(sbi, type, 1);
2573 	curseg->alloc_type = SSR;
2574 	__next_free_blkoff(sbi, curseg, 0);
2575 
2576 	sum_page = f2fs_get_sum_page(sbi, new_segno);
2577 	f2fs_bug_on(sbi, IS_ERR(sum_page));
2578 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2579 	memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2580 	f2fs_put_page(sum_page, 1);
2581 }
2582 
2583 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2584 {
2585 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2586 	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2587 	unsigned segno = NULL_SEGNO;
2588 	int i, cnt;
2589 	bool reversed = false;
2590 
2591 	/* f2fs_need_SSR() already forces to do this */
2592 	if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2593 		curseg->next_segno = segno;
2594 		return 1;
2595 	}
2596 
2597 	/* For node segments, let's do SSR more intensively */
2598 	if (IS_NODESEG(type)) {
2599 		if (type >= CURSEG_WARM_NODE) {
2600 			reversed = true;
2601 			i = CURSEG_COLD_NODE;
2602 		} else {
2603 			i = CURSEG_HOT_NODE;
2604 		}
2605 		cnt = NR_CURSEG_NODE_TYPE;
2606 	} else {
2607 		if (type >= CURSEG_WARM_DATA) {
2608 			reversed = true;
2609 			i = CURSEG_COLD_DATA;
2610 		} else {
2611 			i = CURSEG_HOT_DATA;
2612 		}
2613 		cnt = NR_CURSEG_DATA_TYPE;
2614 	}
2615 
2616 	for (; cnt-- > 0; reversed ? i-- : i++) {
2617 		if (i == type)
2618 			continue;
2619 		if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2620 			curseg->next_segno = segno;
2621 			return 1;
2622 		}
2623 	}
2624 
2625 	/* find valid_blocks=0 in dirty list */
2626 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2627 		segno = get_free_segment(sbi);
2628 		if (segno != NULL_SEGNO) {
2629 			curseg->next_segno = segno;
2630 			return 1;
2631 		}
2632 	}
2633 	return 0;
2634 }
2635 
2636 /*
2637  * flush out current segment and replace it with new segment
2638  * This function should be returned with success, otherwise BUG
2639  */
2640 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2641 						int type, bool force)
2642 {
2643 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2644 
2645 	if (force)
2646 		new_curseg(sbi, type, true);
2647 	else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2648 					type == CURSEG_WARM_NODE)
2649 		new_curseg(sbi, type, false);
2650 	else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2651 			likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2652 		new_curseg(sbi, type, false);
2653 	else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2654 		change_curseg(sbi, type);
2655 	else
2656 		new_curseg(sbi, type, false);
2657 
2658 	stat_inc_seg_type(sbi, curseg);
2659 }
2660 
2661 void allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2662 					unsigned int start, unsigned int end)
2663 {
2664 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2665 	unsigned int segno;
2666 
2667 	down_read(&SM_I(sbi)->curseg_lock);
2668 	mutex_lock(&curseg->curseg_mutex);
2669 	down_write(&SIT_I(sbi)->sentry_lock);
2670 
2671 	segno = CURSEG_I(sbi, type)->segno;
2672 	if (segno < start || segno > end)
2673 		goto unlock;
2674 
2675 	if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2676 		change_curseg(sbi, type);
2677 	else
2678 		new_curseg(sbi, type, true);
2679 
2680 	stat_inc_seg_type(sbi, curseg);
2681 
2682 	locate_dirty_segment(sbi, segno);
2683 unlock:
2684 	up_write(&SIT_I(sbi)->sentry_lock);
2685 
2686 	if (segno != curseg->segno)
2687 		f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2688 			    type, segno, curseg->segno);
2689 
2690 	mutex_unlock(&curseg->curseg_mutex);
2691 	up_read(&SM_I(sbi)->curseg_lock);
2692 }
2693 
2694 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2695 {
2696 	struct curseg_info *curseg;
2697 	unsigned int old_segno;
2698 	int i;
2699 
2700 	down_write(&SIT_I(sbi)->sentry_lock);
2701 
2702 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2703 		curseg = CURSEG_I(sbi, i);
2704 		old_segno = curseg->segno;
2705 		SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2706 		locate_dirty_segment(sbi, old_segno);
2707 	}
2708 
2709 	up_write(&SIT_I(sbi)->sentry_lock);
2710 }
2711 
2712 static const struct segment_allocation default_salloc_ops = {
2713 	.allocate_segment = allocate_segment_by_default,
2714 };
2715 
2716 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2717 						struct cp_control *cpc)
2718 {
2719 	__u64 trim_start = cpc->trim_start;
2720 	bool has_candidate = false;
2721 
2722 	down_write(&SIT_I(sbi)->sentry_lock);
2723 	for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2724 		if (add_discard_addrs(sbi, cpc, true)) {
2725 			has_candidate = true;
2726 			break;
2727 		}
2728 	}
2729 	up_write(&SIT_I(sbi)->sentry_lock);
2730 
2731 	cpc->trim_start = trim_start;
2732 	return has_candidate;
2733 }
2734 
2735 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2736 					struct discard_policy *dpolicy,
2737 					unsigned int start, unsigned int end)
2738 {
2739 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2740 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2741 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
2742 	struct discard_cmd *dc;
2743 	struct blk_plug plug;
2744 	int issued;
2745 	unsigned int trimmed = 0;
2746 
2747 next:
2748 	issued = 0;
2749 
2750 	mutex_lock(&dcc->cmd_lock);
2751 	if (unlikely(dcc->rbtree_check))
2752 		f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2753 								&dcc->root));
2754 
2755 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2756 					NULL, start,
2757 					(struct rb_entry **)&prev_dc,
2758 					(struct rb_entry **)&next_dc,
2759 					&insert_p, &insert_parent, true, NULL);
2760 	if (!dc)
2761 		dc = next_dc;
2762 
2763 	blk_start_plug(&plug);
2764 
2765 	while (dc && dc->lstart <= end) {
2766 		struct rb_node *node;
2767 		int err = 0;
2768 
2769 		if (dc->len < dpolicy->granularity)
2770 			goto skip;
2771 
2772 		if (dc->state != D_PREP) {
2773 			list_move_tail(&dc->list, &dcc->fstrim_list);
2774 			goto skip;
2775 		}
2776 
2777 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2778 
2779 		if (issued >= dpolicy->max_requests) {
2780 			start = dc->lstart + dc->len;
2781 
2782 			if (err)
2783 				__remove_discard_cmd(sbi, dc);
2784 
2785 			blk_finish_plug(&plug);
2786 			mutex_unlock(&dcc->cmd_lock);
2787 			trimmed += __wait_all_discard_cmd(sbi, NULL);
2788 			congestion_wait(BLK_RW_ASYNC, HZ/50);
2789 			goto next;
2790 		}
2791 skip:
2792 		node = rb_next(&dc->rb_node);
2793 		if (err)
2794 			__remove_discard_cmd(sbi, dc);
2795 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2796 
2797 		if (fatal_signal_pending(current))
2798 			break;
2799 	}
2800 
2801 	blk_finish_plug(&plug);
2802 	mutex_unlock(&dcc->cmd_lock);
2803 
2804 	return trimmed;
2805 }
2806 
2807 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2808 {
2809 	__u64 start = F2FS_BYTES_TO_BLK(range->start);
2810 	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2811 	unsigned int start_segno, end_segno;
2812 	block_t start_block, end_block;
2813 	struct cp_control cpc;
2814 	struct discard_policy dpolicy;
2815 	unsigned long long trimmed = 0;
2816 	int err = 0;
2817 	bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
2818 
2819 	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2820 		return -EINVAL;
2821 
2822 	if (end < MAIN_BLKADDR(sbi))
2823 		goto out;
2824 
2825 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2826 		f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
2827 		return -EFSCORRUPTED;
2828 	}
2829 
2830 	/* start/end segment number in main_area */
2831 	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2832 	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2833 						GET_SEGNO(sbi, end);
2834 	if (need_align) {
2835 		start_segno = rounddown(start_segno, sbi->segs_per_sec);
2836 		end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2837 	}
2838 
2839 	cpc.reason = CP_DISCARD;
2840 	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2841 	cpc.trim_start = start_segno;
2842 	cpc.trim_end = end_segno;
2843 
2844 	if (sbi->discard_blks == 0)
2845 		goto out;
2846 
2847 	mutex_lock(&sbi->gc_mutex);
2848 	err = f2fs_write_checkpoint(sbi, &cpc);
2849 	mutex_unlock(&sbi->gc_mutex);
2850 	if (err)
2851 		goto out;
2852 
2853 	/*
2854 	 * We filed discard candidates, but actually we don't need to wait for
2855 	 * all of them, since they'll be issued in idle time along with runtime
2856 	 * discard option. User configuration looks like using runtime discard
2857 	 * or periodic fstrim instead of it.
2858 	 */
2859 	if (f2fs_realtime_discard_enable(sbi))
2860 		goto out;
2861 
2862 	start_block = START_BLOCK(sbi, start_segno);
2863 	end_block = START_BLOCK(sbi, end_segno + 1);
2864 
2865 	__init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2866 	trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2867 					start_block, end_block);
2868 
2869 	trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2870 					start_block, end_block);
2871 out:
2872 	if (!err)
2873 		range->len = F2FS_BLK_TO_BYTES(trimmed);
2874 	return err;
2875 }
2876 
2877 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2878 {
2879 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2880 	if (curseg->next_blkoff < sbi->blocks_per_seg)
2881 		return true;
2882 	return false;
2883 }
2884 
2885 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2886 {
2887 	switch (hint) {
2888 	case WRITE_LIFE_SHORT:
2889 		return CURSEG_HOT_DATA;
2890 	case WRITE_LIFE_EXTREME:
2891 		return CURSEG_COLD_DATA;
2892 	default:
2893 		return CURSEG_WARM_DATA;
2894 	}
2895 }
2896 
2897 /* This returns write hints for each segment type. This hints will be
2898  * passed down to block layer. There are mapping tables which depend on
2899  * the mount option 'whint_mode'.
2900  *
2901  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2902  *
2903  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2904  *
2905  * User                  F2FS                     Block
2906  * ----                  ----                     -----
2907  *                       META                     WRITE_LIFE_NOT_SET
2908  *                       HOT_NODE                 "
2909  *                       WARM_NODE                "
2910  *                       COLD_NODE                "
2911  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2912  * extension list        "                        "
2913  *
2914  * -- buffered io
2915  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2916  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2917  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2918  * WRITE_LIFE_NONE       "                        "
2919  * WRITE_LIFE_MEDIUM     "                        "
2920  * WRITE_LIFE_LONG       "                        "
2921  *
2922  * -- direct io
2923  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2924  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2925  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2926  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2927  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2928  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2929  *
2930  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2931  *
2932  * User                  F2FS                     Block
2933  * ----                  ----                     -----
2934  *                       META                     WRITE_LIFE_MEDIUM;
2935  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2936  *                       WARM_NODE                "
2937  *                       COLD_NODE                WRITE_LIFE_NONE
2938  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2939  * extension list        "                        "
2940  *
2941  * -- buffered io
2942  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2943  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2944  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2945  * WRITE_LIFE_NONE       "                        "
2946  * WRITE_LIFE_MEDIUM     "                        "
2947  * WRITE_LIFE_LONG       "                        "
2948  *
2949  * -- direct io
2950  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2951  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2952  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2953  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2954  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2955  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2956  */
2957 
2958 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2959 				enum page_type type, enum temp_type temp)
2960 {
2961 	if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2962 		if (type == DATA) {
2963 			if (temp == WARM)
2964 				return WRITE_LIFE_NOT_SET;
2965 			else if (temp == HOT)
2966 				return WRITE_LIFE_SHORT;
2967 			else if (temp == COLD)
2968 				return WRITE_LIFE_EXTREME;
2969 		} else {
2970 			return WRITE_LIFE_NOT_SET;
2971 		}
2972 	} else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2973 		if (type == DATA) {
2974 			if (temp == WARM)
2975 				return WRITE_LIFE_LONG;
2976 			else if (temp == HOT)
2977 				return WRITE_LIFE_SHORT;
2978 			else if (temp == COLD)
2979 				return WRITE_LIFE_EXTREME;
2980 		} else if (type == NODE) {
2981 			if (temp == WARM || temp == HOT)
2982 				return WRITE_LIFE_NOT_SET;
2983 			else if (temp == COLD)
2984 				return WRITE_LIFE_NONE;
2985 		} else if (type == META) {
2986 			return WRITE_LIFE_MEDIUM;
2987 		}
2988 	}
2989 	return WRITE_LIFE_NOT_SET;
2990 }
2991 
2992 static int __get_segment_type_2(struct f2fs_io_info *fio)
2993 {
2994 	if (fio->type == DATA)
2995 		return CURSEG_HOT_DATA;
2996 	else
2997 		return CURSEG_HOT_NODE;
2998 }
2999 
3000 static int __get_segment_type_4(struct f2fs_io_info *fio)
3001 {
3002 	if (fio->type == DATA) {
3003 		struct inode *inode = fio->page->mapping->host;
3004 
3005 		if (S_ISDIR(inode->i_mode))
3006 			return CURSEG_HOT_DATA;
3007 		else
3008 			return CURSEG_COLD_DATA;
3009 	} else {
3010 		if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3011 			return CURSEG_WARM_NODE;
3012 		else
3013 			return CURSEG_COLD_NODE;
3014 	}
3015 }
3016 
3017 static int __get_segment_type_6(struct f2fs_io_info *fio)
3018 {
3019 	if (fio->type == DATA) {
3020 		struct inode *inode = fio->page->mapping->host;
3021 
3022 		if (is_cold_data(fio->page) || file_is_cold(inode))
3023 			return CURSEG_COLD_DATA;
3024 		if (file_is_hot(inode) ||
3025 				is_inode_flag_set(inode, FI_HOT_DATA) ||
3026 				f2fs_is_atomic_file(inode) ||
3027 				f2fs_is_volatile_file(inode))
3028 			return CURSEG_HOT_DATA;
3029 		return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3030 	} else {
3031 		if (IS_DNODE(fio->page))
3032 			return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3033 						CURSEG_HOT_NODE;
3034 		return CURSEG_COLD_NODE;
3035 	}
3036 }
3037 
3038 static int __get_segment_type(struct f2fs_io_info *fio)
3039 {
3040 	int type = 0;
3041 
3042 	switch (F2FS_OPTION(fio->sbi).active_logs) {
3043 	case 2:
3044 		type = __get_segment_type_2(fio);
3045 		break;
3046 	case 4:
3047 		type = __get_segment_type_4(fio);
3048 		break;
3049 	case 6:
3050 		type = __get_segment_type_6(fio);
3051 		break;
3052 	default:
3053 		f2fs_bug_on(fio->sbi, true);
3054 	}
3055 
3056 	if (IS_HOT(type))
3057 		fio->temp = HOT;
3058 	else if (IS_WARM(type))
3059 		fio->temp = WARM;
3060 	else
3061 		fio->temp = COLD;
3062 	return type;
3063 }
3064 
3065 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3066 		block_t old_blkaddr, block_t *new_blkaddr,
3067 		struct f2fs_summary *sum, int type,
3068 		struct f2fs_io_info *fio, bool add_list)
3069 {
3070 	struct sit_info *sit_i = SIT_I(sbi);
3071 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3072 
3073 	down_read(&SM_I(sbi)->curseg_lock);
3074 
3075 	mutex_lock(&curseg->curseg_mutex);
3076 	down_write(&sit_i->sentry_lock);
3077 
3078 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3079 
3080 	f2fs_wait_discard_bio(sbi, *new_blkaddr);
3081 
3082 	/*
3083 	 * __add_sum_entry should be resided under the curseg_mutex
3084 	 * because, this function updates a summary entry in the
3085 	 * current summary block.
3086 	 */
3087 	__add_sum_entry(sbi, type, sum);
3088 
3089 	__refresh_next_blkoff(sbi, curseg);
3090 
3091 	stat_inc_block_count(sbi, curseg);
3092 
3093 	/*
3094 	 * SIT information should be updated before segment allocation,
3095 	 * since SSR needs latest valid block information.
3096 	 */
3097 	update_sit_entry(sbi, *new_blkaddr, 1);
3098 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3099 		update_sit_entry(sbi, old_blkaddr, -1);
3100 
3101 	if (!__has_curseg_space(sbi, type))
3102 		sit_i->s_ops->allocate_segment(sbi, type, false);
3103 
3104 	/*
3105 	 * segment dirty status should be updated after segment allocation,
3106 	 * so we just need to update status only one time after previous
3107 	 * segment being closed.
3108 	 */
3109 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3110 	locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3111 
3112 	up_write(&sit_i->sentry_lock);
3113 
3114 	if (page && IS_NODESEG(type)) {
3115 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3116 
3117 		f2fs_inode_chksum_set(sbi, page);
3118 	}
3119 
3120 	if (F2FS_IO_ALIGNED(sbi))
3121 		fio->retry = false;
3122 
3123 	if (add_list) {
3124 		struct f2fs_bio_info *io;
3125 
3126 		INIT_LIST_HEAD(&fio->list);
3127 		fio->in_list = true;
3128 		io = sbi->write_io[fio->type] + fio->temp;
3129 		spin_lock(&io->io_lock);
3130 		list_add_tail(&fio->list, &io->io_list);
3131 		spin_unlock(&io->io_lock);
3132 	}
3133 
3134 	mutex_unlock(&curseg->curseg_mutex);
3135 
3136 	up_read(&SM_I(sbi)->curseg_lock);
3137 }
3138 
3139 static void update_device_state(struct f2fs_io_info *fio)
3140 {
3141 	struct f2fs_sb_info *sbi = fio->sbi;
3142 	unsigned int devidx;
3143 
3144 	if (!f2fs_is_multi_device(sbi))
3145 		return;
3146 
3147 	devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3148 
3149 	/* update device state for fsync */
3150 	f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3151 
3152 	/* update device state for checkpoint */
3153 	if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3154 		spin_lock(&sbi->dev_lock);
3155 		f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3156 		spin_unlock(&sbi->dev_lock);
3157 	}
3158 }
3159 
3160 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3161 {
3162 	int type = __get_segment_type(fio);
3163 	bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
3164 
3165 	if (keep_order)
3166 		down_read(&fio->sbi->io_order_lock);
3167 reallocate:
3168 	f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3169 			&fio->new_blkaddr, sum, type, fio, true);
3170 	if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3171 		invalidate_mapping_pages(META_MAPPING(fio->sbi),
3172 					fio->old_blkaddr, fio->old_blkaddr);
3173 
3174 	/* writeout dirty page into bdev */
3175 	f2fs_submit_page_write(fio);
3176 	if (fio->retry) {
3177 		fio->old_blkaddr = fio->new_blkaddr;
3178 		goto reallocate;
3179 	}
3180 
3181 	update_device_state(fio);
3182 
3183 	if (keep_order)
3184 		up_read(&fio->sbi->io_order_lock);
3185 }
3186 
3187 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3188 					enum iostat_type io_type)
3189 {
3190 	struct f2fs_io_info fio = {
3191 		.sbi = sbi,
3192 		.type = META,
3193 		.temp = HOT,
3194 		.op = REQ_OP_WRITE,
3195 		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3196 		.old_blkaddr = page->index,
3197 		.new_blkaddr = page->index,
3198 		.page = page,
3199 		.encrypted_page = NULL,
3200 		.in_list = false,
3201 	};
3202 
3203 	if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3204 		fio.op_flags &= ~REQ_META;
3205 
3206 	set_page_writeback(page);
3207 	ClearPageError(page);
3208 	f2fs_submit_page_write(&fio);
3209 
3210 	stat_inc_meta_count(sbi, page->index);
3211 	f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3212 }
3213 
3214 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3215 {
3216 	struct f2fs_summary sum;
3217 
3218 	set_summary(&sum, nid, 0, 0);
3219 	do_write_page(&sum, fio);
3220 
3221 	f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3222 }
3223 
3224 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3225 					struct f2fs_io_info *fio)
3226 {
3227 	struct f2fs_sb_info *sbi = fio->sbi;
3228 	struct f2fs_summary sum;
3229 
3230 	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3231 	set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3232 	do_write_page(&sum, fio);
3233 	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3234 
3235 	f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3236 }
3237 
3238 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3239 {
3240 	int err;
3241 	struct f2fs_sb_info *sbi = fio->sbi;
3242 	unsigned int segno;
3243 
3244 	fio->new_blkaddr = fio->old_blkaddr;
3245 	/* i/o temperature is needed for passing down write hints */
3246 	__get_segment_type(fio);
3247 
3248 	segno = GET_SEGNO(sbi, fio->new_blkaddr);
3249 
3250 	if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3251 		set_sbi_flag(sbi, SBI_NEED_FSCK);
3252 		f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3253 			  __func__, segno);
3254 		return -EFSCORRUPTED;
3255 	}
3256 
3257 	stat_inc_inplace_blocks(fio->sbi);
3258 
3259 	if (fio->bio)
3260 		err = f2fs_merge_page_bio(fio);
3261 	else
3262 		err = f2fs_submit_page_bio(fio);
3263 	if (!err) {
3264 		update_device_state(fio);
3265 		f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3266 	}
3267 
3268 	return err;
3269 }
3270 
3271 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3272 						unsigned int segno)
3273 {
3274 	int i;
3275 
3276 	for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3277 		if (CURSEG_I(sbi, i)->segno == segno)
3278 			break;
3279 	}
3280 	return i;
3281 }
3282 
3283 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3284 				block_t old_blkaddr, block_t new_blkaddr,
3285 				bool recover_curseg, bool recover_newaddr)
3286 {
3287 	struct sit_info *sit_i = SIT_I(sbi);
3288 	struct curseg_info *curseg;
3289 	unsigned int segno, old_cursegno;
3290 	struct seg_entry *se;
3291 	int type;
3292 	unsigned short old_blkoff;
3293 
3294 	segno = GET_SEGNO(sbi, new_blkaddr);
3295 	se = get_seg_entry(sbi, segno);
3296 	type = se->type;
3297 
3298 	down_write(&SM_I(sbi)->curseg_lock);
3299 
3300 	if (!recover_curseg) {
3301 		/* for recovery flow */
3302 		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3303 			if (old_blkaddr == NULL_ADDR)
3304 				type = CURSEG_COLD_DATA;
3305 			else
3306 				type = CURSEG_WARM_DATA;
3307 		}
3308 	} else {
3309 		if (IS_CURSEG(sbi, segno)) {
3310 			/* se->type is volatile as SSR allocation */
3311 			type = __f2fs_get_curseg(sbi, segno);
3312 			f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3313 		} else {
3314 			type = CURSEG_WARM_DATA;
3315 		}
3316 	}
3317 
3318 	f2fs_bug_on(sbi, !IS_DATASEG(type));
3319 	curseg = CURSEG_I(sbi, type);
3320 
3321 	mutex_lock(&curseg->curseg_mutex);
3322 	down_write(&sit_i->sentry_lock);
3323 
3324 	old_cursegno = curseg->segno;
3325 	old_blkoff = curseg->next_blkoff;
3326 
3327 	/* change the current segment */
3328 	if (segno != curseg->segno) {
3329 		curseg->next_segno = segno;
3330 		change_curseg(sbi, type);
3331 	}
3332 
3333 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3334 	__add_sum_entry(sbi, type, sum);
3335 
3336 	if (!recover_curseg || recover_newaddr)
3337 		update_sit_entry(sbi, new_blkaddr, 1);
3338 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3339 		invalidate_mapping_pages(META_MAPPING(sbi),
3340 					old_blkaddr, old_blkaddr);
3341 		update_sit_entry(sbi, old_blkaddr, -1);
3342 	}
3343 
3344 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3345 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3346 
3347 	locate_dirty_segment(sbi, old_cursegno);
3348 
3349 	if (recover_curseg) {
3350 		if (old_cursegno != curseg->segno) {
3351 			curseg->next_segno = old_cursegno;
3352 			change_curseg(sbi, type);
3353 		}
3354 		curseg->next_blkoff = old_blkoff;
3355 	}
3356 
3357 	up_write(&sit_i->sentry_lock);
3358 	mutex_unlock(&curseg->curseg_mutex);
3359 	up_write(&SM_I(sbi)->curseg_lock);
3360 }
3361 
3362 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3363 				block_t old_addr, block_t new_addr,
3364 				unsigned char version, bool recover_curseg,
3365 				bool recover_newaddr)
3366 {
3367 	struct f2fs_summary sum;
3368 
3369 	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3370 
3371 	f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3372 					recover_curseg, recover_newaddr);
3373 
3374 	f2fs_update_data_blkaddr(dn, new_addr);
3375 }
3376 
3377 void f2fs_wait_on_page_writeback(struct page *page,
3378 				enum page_type type, bool ordered, bool locked)
3379 {
3380 	if (PageWriteback(page)) {
3381 		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3382 
3383 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3384 		if (ordered) {
3385 			wait_on_page_writeback(page);
3386 			f2fs_bug_on(sbi, locked && PageWriteback(page));
3387 		} else {
3388 			wait_for_stable_page(page);
3389 		}
3390 	}
3391 }
3392 
3393 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3394 {
3395 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3396 	struct page *cpage;
3397 
3398 	if (!f2fs_post_read_required(inode))
3399 		return;
3400 
3401 	if (!__is_valid_data_blkaddr(blkaddr))
3402 		return;
3403 
3404 	cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3405 	if (cpage) {
3406 		f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3407 		f2fs_put_page(cpage, 1);
3408 	}
3409 }
3410 
3411 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3412 								block_t len)
3413 {
3414 	block_t i;
3415 
3416 	for (i = 0; i < len; i++)
3417 		f2fs_wait_on_block_writeback(inode, blkaddr + i);
3418 }
3419 
3420 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3421 {
3422 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3423 	struct curseg_info *seg_i;
3424 	unsigned char *kaddr;
3425 	struct page *page;
3426 	block_t start;
3427 	int i, j, offset;
3428 
3429 	start = start_sum_block(sbi);
3430 
3431 	page = f2fs_get_meta_page(sbi, start++);
3432 	if (IS_ERR(page))
3433 		return PTR_ERR(page);
3434 	kaddr = (unsigned char *)page_address(page);
3435 
3436 	/* Step 1: restore nat cache */
3437 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3438 	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3439 
3440 	/* Step 2: restore sit cache */
3441 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3442 	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3443 	offset = 2 * SUM_JOURNAL_SIZE;
3444 
3445 	/* Step 3: restore summary entries */
3446 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3447 		unsigned short blk_off;
3448 		unsigned int segno;
3449 
3450 		seg_i = CURSEG_I(sbi, i);
3451 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3452 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3453 		seg_i->next_segno = segno;
3454 		reset_curseg(sbi, i, 0);
3455 		seg_i->alloc_type = ckpt->alloc_type[i];
3456 		seg_i->next_blkoff = blk_off;
3457 
3458 		if (seg_i->alloc_type == SSR)
3459 			blk_off = sbi->blocks_per_seg;
3460 
3461 		for (j = 0; j < blk_off; j++) {
3462 			struct f2fs_summary *s;
3463 			s = (struct f2fs_summary *)(kaddr + offset);
3464 			seg_i->sum_blk->entries[j] = *s;
3465 			offset += SUMMARY_SIZE;
3466 			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3467 						SUM_FOOTER_SIZE)
3468 				continue;
3469 
3470 			f2fs_put_page(page, 1);
3471 			page = NULL;
3472 
3473 			page = f2fs_get_meta_page(sbi, start++);
3474 			if (IS_ERR(page))
3475 				return PTR_ERR(page);
3476 			kaddr = (unsigned char *)page_address(page);
3477 			offset = 0;
3478 		}
3479 	}
3480 	f2fs_put_page(page, 1);
3481 	return 0;
3482 }
3483 
3484 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3485 {
3486 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3487 	struct f2fs_summary_block *sum;
3488 	struct curseg_info *curseg;
3489 	struct page *new;
3490 	unsigned short blk_off;
3491 	unsigned int segno = 0;
3492 	block_t blk_addr = 0;
3493 	int err = 0;
3494 
3495 	/* get segment number and block addr */
3496 	if (IS_DATASEG(type)) {
3497 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3498 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3499 							CURSEG_HOT_DATA]);
3500 		if (__exist_node_summaries(sbi))
3501 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3502 		else
3503 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3504 	} else {
3505 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
3506 							CURSEG_HOT_NODE]);
3507 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3508 							CURSEG_HOT_NODE]);
3509 		if (__exist_node_summaries(sbi))
3510 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3511 							type - CURSEG_HOT_NODE);
3512 		else
3513 			blk_addr = GET_SUM_BLOCK(sbi, segno);
3514 	}
3515 
3516 	new = f2fs_get_meta_page(sbi, blk_addr);
3517 	if (IS_ERR(new))
3518 		return PTR_ERR(new);
3519 	sum = (struct f2fs_summary_block *)page_address(new);
3520 
3521 	if (IS_NODESEG(type)) {
3522 		if (__exist_node_summaries(sbi)) {
3523 			struct f2fs_summary *ns = &sum->entries[0];
3524 			int i;
3525 			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3526 				ns->version = 0;
3527 				ns->ofs_in_node = 0;
3528 			}
3529 		} else {
3530 			err = f2fs_restore_node_summary(sbi, segno, sum);
3531 			if (err)
3532 				goto out;
3533 		}
3534 	}
3535 
3536 	/* set uncompleted segment to curseg */
3537 	curseg = CURSEG_I(sbi, type);
3538 	mutex_lock(&curseg->curseg_mutex);
3539 
3540 	/* update journal info */
3541 	down_write(&curseg->journal_rwsem);
3542 	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3543 	up_write(&curseg->journal_rwsem);
3544 
3545 	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3546 	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3547 	curseg->next_segno = segno;
3548 	reset_curseg(sbi, type, 0);
3549 	curseg->alloc_type = ckpt->alloc_type[type];
3550 	curseg->next_blkoff = blk_off;
3551 	mutex_unlock(&curseg->curseg_mutex);
3552 out:
3553 	f2fs_put_page(new, 1);
3554 	return err;
3555 }
3556 
3557 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3558 {
3559 	struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3560 	struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3561 	int type = CURSEG_HOT_DATA;
3562 	int err;
3563 
3564 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3565 		int npages = f2fs_npages_for_summary_flush(sbi, true);
3566 
3567 		if (npages >= 2)
3568 			f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3569 							META_CP, true);
3570 
3571 		/* restore for compacted data summary */
3572 		err = read_compacted_summaries(sbi);
3573 		if (err)
3574 			return err;
3575 		type = CURSEG_HOT_NODE;
3576 	}
3577 
3578 	if (__exist_node_summaries(sbi))
3579 		f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3580 					NR_CURSEG_TYPE - type, META_CP, true);
3581 
3582 	for (; type <= CURSEG_COLD_NODE; type++) {
3583 		err = read_normal_summaries(sbi, type);
3584 		if (err)
3585 			return err;
3586 	}
3587 
3588 	/* sanity check for summary blocks */
3589 	if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3590 			sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3591 		f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3592 			 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3593 		return -EINVAL;
3594 	}
3595 
3596 	return 0;
3597 }
3598 
3599 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3600 {
3601 	struct page *page;
3602 	unsigned char *kaddr;
3603 	struct f2fs_summary *summary;
3604 	struct curseg_info *seg_i;
3605 	int written_size = 0;
3606 	int i, j;
3607 
3608 	page = f2fs_grab_meta_page(sbi, blkaddr++);
3609 	kaddr = (unsigned char *)page_address(page);
3610 	memset(kaddr, 0, PAGE_SIZE);
3611 
3612 	/* Step 1: write nat cache */
3613 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3614 	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3615 	written_size += SUM_JOURNAL_SIZE;
3616 
3617 	/* Step 2: write sit cache */
3618 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3619 	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3620 	written_size += SUM_JOURNAL_SIZE;
3621 
3622 	/* Step 3: write summary entries */
3623 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3624 		unsigned short blkoff;
3625 		seg_i = CURSEG_I(sbi, i);
3626 		if (sbi->ckpt->alloc_type[i] == SSR)
3627 			blkoff = sbi->blocks_per_seg;
3628 		else
3629 			blkoff = curseg_blkoff(sbi, i);
3630 
3631 		for (j = 0; j < blkoff; j++) {
3632 			if (!page) {
3633 				page = f2fs_grab_meta_page(sbi, blkaddr++);
3634 				kaddr = (unsigned char *)page_address(page);
3635 				memset(kaddr, 0, PAGE_SIZE);
3636 				written_size = 0;
3637 			}
3638 			summary = (struct f2fs_summary *)(kaddr + written_size);
3639 			*summary = seg_i->sum_blk->entries[j];
3640 			written_size += SUMMARY_SIZE;
3641 
3642 			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3643 							SUM_FOOTER_SIZE)
3644 				continue;
3645 
3646 			set_page_dirty(page);
3647 			f2fs_put_page(page, 1);
3648 			page = NULL;
3649 		}
3650 	}
3651 	if (page) {
3652 		set_page_dirty(page);
3653 		f2fs_put_page(page, 1);
3654 	}
3655 }
3656 
3657 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3658 					block_t blkaddr, int type)
3659 {
3660 	int i, end;
3661 	if (IS_DATASEG(type))
3662 		end = type + NR_CURSEG_DATA_TYPE;
3663 	else
3664 		end = type + NR_CURSEG_NODE_TYPE;
3665 
3666 	for (i = type; i < end; i++)
3667 		write_current_sum_page(sbi, i, blkaddr + (i - type));
3668 }
3669 
3670 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3671 {
3672 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3673 		write_compacted_summaries(sbi, start_blk);
3674 	else
3675 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3676 }
3677 
3678 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3679 {
3680 	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3681 }
3682 
3683 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3684 					unsigned int val, int alloc)
3685 {
3686 	int i;
3687 
3688 	if (type == NAT_JOURNAL) {
3689 		for (i = 0; i < nats_in_cursum(journal); i++) {
3690 			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3691 				return i;
3692 		}
3693 		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3694 			return update_nats_in_cursum(journal, 1);
3695 	} else if (type == SIT_JOURNAL) {
3696 		for (i = 0; i < sits_in_cursum(journal); i++)
3697 			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3698 				return i;
3699 		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3700 			return update_sits_in_cursum(journal, 1);
3701 	}
3702 	return -1;
3703 }
3704 
3705 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3706 					unsigned int segno)
3707 {
3708 	return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3709 }
3710 
3711 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3712 					unsigned int start)
3713 {
3714 	struct sit_info *sit_i = SIT_I(sbi);
3715 	struct page *page;
3716 	pgoff_t src_off, dst_off;
3717 
3718 	src_off = current_sit_addr(sbi, start);
3719 	dst_off = next_sit_addr(sbi, src_off);
3720 
3721 	page = f2fs_grab_meta_page(sbi, dst_off);
3722 	seg_info_to_sit_page(sbi, page, start);
3723 
3724 	set_page_dirty(page);
3725 	set_to_next_sit(sit_i, start);
3726 
3727 	return page;
3728 }
3729 
3730 static struct sit_entry_set *grab_sit_entry_set(void)
3731 {
3732 	struct sit_entry_set *ses =
3733 			f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3734 
3735 	ses->entry_cnt = 0;
3736 	INIT_LIST_HEAD(&ses->set_list);
3737 	return ses;
3738 }
3739 
3740 static void release_sit_entry_set(struct sit_entry_set *ses)
3741 {
3742 	list_del(&ses->set_list);
3743 	kmem_cache_free(sit_entry_set_slab, ses);
3744 }
3745 
3746 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3747 						struct list_head *head)
3748 {
3749 	struct sit_entry_set *next = ses;
3750 
3751 	if (list_is_last(&ses->set_list, head))
3752 		return;
3753 
3754 	list_for_each_entry_continue(next, head, set_list)
3755 		if (ses->entry_cnt <= next->entry_cnt)
3756 			break;
3757 
3758 	list_move_tail(&ses->set_list, &next->set_list);
3759 }
3760 
3761 static void add_sit_entry(unsigned int segno, struct list_head *head)
3762 {
3763 	struct sit_entry_set *ses;
3764 	unsigned int start_segno = START_SEGNO(segno);
3765 
3766 	list_for_each_entry(ses, head, set_list) {
3767 		if (ses->start_segno == start_segno) {
3768 			ses->entry_cnt++;
3769 			adjust_sit_entry_set(ses, head);
3770 			return;
3771 		}
3772 	}
3773 
3774 	ses = grab_sit_entry_set();
3775 
3776 	ses->start_segno = start_segno;
3777 	ses->entry_cnt++;
3778 	list_add(&ses->set_list, head);
3779 }
3780 
3781 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3782 {
3783 	struct f2fs_sm_info *sm_info = SM_I(sbi);
3784 	struct list_head *set_list = &sm_info->sit_entry_set;
3785 	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3786 	unsigned int segno;
3787 
3788 	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3789 		add_sit_entry(segno, set_list);
3790 }
3791 
3792 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3793 {
3794 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3795 	struct f2fs_journal *journal = curseg->journal;
3796 	int i;
3797 
3798 	down_write(&curseg->journal_rwsem);
3799 	for (i = 0; i < sits_in_cursum(journal); i++) {
3800 		unsigned int segno;
3801 		bool dirtied;
3802 
3803 		segno = le32_to_cpu(segno_in_journal(journal, i));
3804 		dirtied = __mark_sit_entry_dirty(sbi, segno);
3805 
3806 		if (!dirtied)
3807 			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3808 	}
3809 	update_sits_in_cursum(journal, -i);
3810 	up_write(&curseg->journal_rwsem);
3811 }
3812 
3813 /*
3814  * CP calls this function, which flushes SIT entries including sit_journal,
3815  * and moves prefree segs to free segs.
3816  */
3817 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3818 {
3819 	struct sit_info *sit_i = SIT_I(sbi);
3820 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3821 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3822 	struct f2fs_journal *journal = curseg->journal;
3823 	struct sit_entry_set *ses, *tmp;
3824 	struct list_head *head = &SM_I(sbi)->sit_entry_set;
3825 	bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
3826 	struct seg_entry *se;
3827 
3828 	down_write(&sit_i->sentry_lock);
3829 
3830 	if (!sit_i->dirty_sentries)
3831 		goto out;
3832 
3833 	/*
3834 	 * add and account sit entries of dirty bitmap in sit entry
3835 	 * set temporarily
3836 	 */
3837 	add_sits_in_set(sbi);
3838 
3839 	/*
3840 	 * if there are no enough space in journal to store dirty sit
3841 	 * entries, remove all entries from journal and add and account
3842 	 * them in sit entry set.
3843 	 */
3844 	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
3845 								!to_journal)
3846 		remove_sits_in_journal(sbi);
3847 
3848 	/*
3849 	 * there are two steps to flush sit entries:
3850 	 * #1, flush sit entries to journal in current cold data summary block.
3851 	 * #2, flush sit entries to sit page.
3852 	 */
3853 	list_for_each_entry_safe(ses, tmp, head, set_list) {
3854 		struct page *page = NULL;
3855 		struct f2fs_sit_block *raw_sit = NULL;
3856 		unsigned int start_segno = ses->start_segno;
3857 		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3858 						(unsigned long)MAIN_SEGS(sbi));
3859 		unsigned int segno = start_segno;
3860 
3861 		if (to_journal &&
3862 			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3863 			to_journal = false;
3864 
3865 		if (to_journal) {
3866 			down_write(&curseg->journal_rwsem);
3867 		} else {
3868 			page = get_next_sit_page(sbi, start_segno);
3869 			raw_sit = page_address(page);
3870 		}
3871 
3872 		/* flush dirty sit entries in region of current sit set */
3873 		for_each_set_bit_from(segno, bitmap, end) {
3874 			int offset, sit_offset;
3875 
3876 			se = get_seg_entry(sbi, segno);
3877 #ifdef CONFIG_F2FS_CHECK_FS
3878 			if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3879 						SIT_VBLOCK_MAP_SIZE))
3880 				f2fs_bug_on(sbi, 1);
3881 #endif
3882 
3883 			/* add discard candidates */
3884 			if (!(cpc->reason & CP_DISCARD)) {
3885 				cpc->trim_start = segno;
3886 				add_discard_addrs(sbi, cpc, false);
3887 			}
3888 
3889 			if (to_journal) {
3890 				offset = f2fs_lookup_journal_in_cursum(journal,
3891 							SIT_JOURNAL, segno, 1);
3892 				f2fs_bug_on(sbi, offset < 0);
3893 				segno_in_journal(journal, offset) =
3894 							cpu_to_le32(segno);
3895 				seg_info_to_raw_sit(se,
3896 					&sit_in_journal(journal, offset));
3897 				check_block_count(sbi, segno,
3898 					&sit_in_journal(journal, offset));
3899 			} else {
3900 				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3901 				seg_info_to_raw_sit(se,
3902 						&raw_sit->entries[sit_offset]);
3903 				check_block_count(sbi, segno,
3904 						&raw_sit->entries[sit_offset]);
3905 			}
3906 
3907 			__clear_bit(segno, bitmap);
3908 			sit_i->dirty_sentries--;
3909 			ses->entry_cnt--;
3910 		}
3911 
3912 		if (to_journal)
3913 			up_write(&curseg->journal_rwsem);
3914 		else
3915 			f2fs_put_page(page, 1);
3916 
3917 		f2fs_bug_on(sbi, ses->entry_cnt);
3918 		release_sit_entry_set(ses);
3919 	}
3920 
3921 	f2fs_bug_on(sbi, !list_empty(head));
3922 	f2fs_bug_on(sbi, sit_i->dirty_sentries);
3923 out:
3924 	if (cpc->reason & CP_DISCARD) {
3925 		__u64 trim_start = cpc->trim_start;
3926 
3927 		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3928 			add_discard_addrs(sbi, cpc, false);
3929 
3930 		cpc->trim_start = trim_start;
3931 	}
3932 	up_write(&sit_i->sentry_lock);
3933 
3934 	set_prefree_as_free_segments(sbi);
3935 }
3936 
3937 static int build_sit_info(struct f2fs_sb_info *sbi)
3938 {
3939 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3940 	struct sit_info *sit_i;
3941 	unsigned int sit_segs, start;
3942 	char *src_bitmap, *bitmap;
3943 	unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
3944 
3945 	/* allocate memory for SIT information */
3946 	sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3947 	if (!sit_i)
3948 		return -ENOMEM;
3949 
3950 	SM_I(sbi)->sit_info = sit_i;
3951 
3952 	sit_i->sentries =
3953 		f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3954 					      MAIN_SEGS(sbi)),
3955 			      GFP_KERNEL);
3956 	if (!sit_i->sentries)
3957 		return -ENOMEM;
3958 
3959 	main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3960 	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
3961 								GFP_KERNEL);
3962 	if (!sit_i->dirty_sentries_bitmap)
3963 		return -ENOMEM;
3964 
3965 #ifdef CONFIG_F2FS_CHECK_FS
3966 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 4;
3967 #else
3968 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 3;
3969 #endif
3970 	sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
3971 	if (!sit_i->bitmap)
3972 		return -ENOMEM;
3973 
3974 	bitmap = sit_i->bitmap;
3975 
3976 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
3977 		sit_i->sentries[start].cur_valid_map = bitmap;
3978 		bitmap += SIT_VBLOCK_MAP_SIZE;
3979 
3980 		sit_i->sentries[start].ckpt_valid_map = bitmap;
3981 		bitmap += SIT_VBLOCK_MAP_SIZE;
3982 
3983 #ifdef CONFIG_F2FS_CHECK_FS
3984 		sit_i->sentries[start].cur_valid_map_mir = bitmap;
3985 		bitmap += SIT_VBLOCK_MAP_SIZE;
3986 #endif
3987 
3988 		sit_i->sentries[start].discard_map = bitmap;
3989 		bitmap += SIT_VBLOCK_MAP_SIZE;
3990 	}
3991 
3992 	sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3993 	if (!sit_i->tmp_map)
3994 		return -ENOMEM;
3995 
3996 	if (__is_large_section(sbi)) {
3997 		sit_i->sec_entries =
3998 			f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3999 						      MAIN_SECS(sbi)),
4000 				      GFP_KERNEL);
4001 		if (!sit_i->sec_entries)
4002 			return -ENOMEM;
4003 	}
4004 
4005 	/* get information related with SIT */
4006 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4007 
4008 	/* setup SIT bitmap from ckeckpoint pack */
4009 	sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4010 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4011 
4012 	sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4013 	if (!sit_i->sit_bitmap)
4014 		return -ENOMEM;
4015 
4016 #ifdef CONFIG_F2FS_CHECK_FS
4017 	sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4018 					sit_bitmap_size, GFP_KERNEL);
4019 	if (!sit_i->sit_bitmap_mir)
4020 		return -ENOMEM;
4021 
4022 	sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4023 					main_bitmap_size, GFP_KERNEL);
4024 	if (!sit_i->invalid_segmap)
4025 		return -ENOMEM;
4026 #endif
4027 
4028 	/* init SIT information */
4029 	sit_i->s_ops = &default_salloc_ops;
4030 
4031 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4032 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4033 	sit_i->written_valid_blocks = 0;
4034 	sit_i->bitmap_size = sit_bitmap_size;
4035 	sit_i->dirty_sentries = 0;
4036 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4037 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4038 	sit_i->mounted_time = ktime_get_real_seconds();
4039 	init_rwsem(&sit_i->sentry_lock);
4040 	return 0;
4041 }
4042 
4043 static int build_free_segmap(struct f2fs_sb_info *sbi)
4044 {
4045 	struct free_segmap_info *free_i;
4046 	unsigned int bitmap_size, sec_bitmap_size;
4047 
4048 	/* allocate memory for free segmap information */
4049 	free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4050 	if (!free_i)
4051 		return -ENOMEM;
4052 
4053 	SM_I(sbi)->free_info = free_i;
4054 
4055 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4056 	free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4057 	if (!free_i->free_segmap)
4058 		return -ENOMEM;
4059 
4060 	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4061 	free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4062 	if (!free_i->free_secmap)
4063 		return -ENOMEM;
4064 
4065 	/* set all segments as dirty temporarily */
4066 	memset(free_i->free_segmap, 0xff, bitmap_size);
4067 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4068 
4069 	/* init free segmap information */
4070 	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4071 	free_i->free_segments = 0;
4072 	free_i->free_sections = 0;
4073 	spin_lock_init(&free_i->segmap_lock);
4074 	return 0;
4075 }
4076 
4077 static int build_curseg(struct f2fs_sb_info *sbi)
4078 {
4079 	struct curseg_info *array;
4080 	int i;
4081 
4082 	array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
4083 			     GFP_KERNEL);
4084 	if (!array)
4085 		return -ENOMEM;
4086 
4087 	SM_I(sbi)->curseg_array = array;
4088 
4089 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
4090 		mutex_init(&array[i].curseg_mutex);
4091 		array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4092 		if (!array[i].sum_blk)
4093 			return -ENOMEM;
4094 		init_rwsem(&array[i].journal_rwsem);
4095 		array[i].journal = f2fs_kzalloc(sbi,
4096 				sizeof(struct f2fs_journal), GFP_KERNEL);
4097 		if (!array[i].journal)
4098 			return -ENOMEM;
4099 		array[i].segno = NULL_SEGNO;
4100 		array[i].next_blkoff = 0;
4101 	}
4102 	return restore_curseg_summaries(sbi);
4103 }
4104 
4105 static int build_sit_entries(struct f2fs_sb_info *sbi)
4106 {
4107 	struct sit_info *sit_i = SIT_I(sbi);
4108 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4109 	struct f2fs_journal *journal = curseg->journal;
4110 	struct seg_entry *se;
4111 	struct f2fs_sit_entry sit;
4112 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
4113 	unsigned int i, start, end;
4114 	unsigned int readed, start_blk = 0;
4115 	int err = 0;
4116 	block_t total_node_blocks = 0;
4117 
4118 	do {
4119 		readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4120 							META_SIT, true);
4121 
4122 		start = start_blk * sit_i->sents_per_block;
4123 		end = (start_blk + readed) * sit_i->sents_per_block;
4124 
4125 		for (; start < end && start < MAIN_SEGS(sbi); start++) {
4126 			struct f2fs_sit_block *sit_blk;
4127 			struct page *page;
4128 
4129 			se = &sit_i->sentries[start];
4130 			page = get_current_sit_page(sbi, start);
4131 			if (IS_ERR(page))
4132 				return PTR_ERR(page);
4133 			sit_blk = (struct f2fs_sit_block *)page_address(page);
4134 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4135 			f2fs_put_page(page, 1);
4136 
4137 			err = check_block_count(sbi, start, &sit);
4138 			if (err)
4139 				return err;
4140 			seg_info_from_raw_sit(se, &sit);
4141 			if (IS_NODESEG(se->type))
4142 				total_node_blocks += se->valid_blocks;
4143 
4144 			/* build discard map only one time */
4145 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4146 				memset(se->discard_map, 0xff,
4147 					SIT_VBLOCK_MAP_SIZE);
4148 			} else {
4149 				memcpy(se->discard_map,
4150 					se->cur_valid_map,
4151 					SIT_VBLOCK_MAP_SIZE);
4152 				sbi->discard_blks +=
4153 					sbi->blocks_per_seg -
4154 					se->valid_blocks;
4155 			}
4156 
4157 			if (__is_large_section(sbi))
4158 				get_sec_entry(sbi, start)->valid_blocks +=
4159 							se->valid_blocks;
4160 		}
4161 		start_blk += readed;
4162 	} while (start_blk < sit_blk_cnt);
4163 
4164 	down_read(&curseg->journal_rwsem);
4165 	for (i = 0; i < sits_in_cursum(journal); i++) {
4166 		unsigned int old_valid_blocks;
4167 
4168 		start = le32_to_cpu(segno_in_journal(journal, i));
4169 		if (start >= MAIN_SEGS(sbi)) {
4170 			f2fs_err(sbi, "Wrong journal entry on segno %u",
4171 				 start);
4172 			err = -EFSCORRUPTED;
4173 			break;
4174 		}
4175 
4176 		se = &sit_i->sentries[start];
4177 		sit = sit_in_journal(journal, i);
4178 
4179 		old_valid_blocks = se->valid_blocks;
4180 		if (IS_NODESEG(se->type))
4181 			total_node_blocks -= old_valid_blocks;
4182 
4183 		err = check_block_count(sbi, start, &sit);
4184 		if (err)
4185 			break;
4186 		seg_info_from_raw_sit(se, &sit);
4187 		if (IS_NODESEG(se->type))
4188 			total_node_blocks += se->valid_blocks;
4189 
4190 		if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4191 			memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4192 		} else {
4193 			memcpy(se->discard_map, se->cur_valid_map,
4194 						SIT_VBLOCK_MAP_SIZE);
4195 			sbi->discard_blks += old_valid_blocks;
4196 			sbi->discard_blks -= se->valid_blocks;
4197 		}
4198 
4199 		if (__is_large_section(sbi)) {
4200 			get_sec_entry(sbi, start)->valid_blocks +=
4201 							se->valid_blocks;
4202 			get_sec_entry(sbi, start)->valid_blocks -=
4203 							old_valid_blocks;
4204 		}
4205 	}
4206 	up_read(&curseg->journal_rwsem);
4207 
4208 	if (!err && total_node_blocks != valid_node_count(sbi)) {
4209 		f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4210 			 total_node_blocks, valid_node_count(sbi));
4211 		err = -EFSCORRUPTED;
4212 	}
4213 
4214 	return err;
4215 }
4216 
4217 static void init_free_segmap(struct f2fs_sb_info *sbi)
4218 {
4219 	unsigned int start;
4220 	int type;
4221 
4222 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4223 		struct seg_entry *sentry = get_seg_entry(sbi, start);
4224 		if (!sentry->valid_blocks)
4225 			__set_free(sbi, start);
4226 		else
4227 			SIT_I(sbi)->written_valid_blocks +=
4228 						sentry->valid_blocks;
4229 	}
4230 
4231 	/* set use the current segments */
4232 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4233 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4234 		__set_test_and_inuse(sbi, curseg_t->segno);
4235 	}
4236 }
4237 
4238 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4239 {
4240 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4241 	struct free_segmap_info *free_i = FREE_I(sbi);
4242 	unsigned int segno = 0, offset = 0;
4243 	unsigned short valid_blocks;
4244 
4245 	while (1) {
4246 		/* find dirty segment based on free segmap */
4247 		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4248 		if (segno >= MAIN_SEGS(sbi))
4249 			break;
4250 		offset = segno + 1;
4251 		valid_blocks = get_valid_blocks(sbi, segno, false);
4252 		if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4253 			continue;
4254 		if (valid_blocks > sbi->blocks_per_seg) {
4255 			f2fs_bug_on(sbi, 1);
4256 			continue;
4257 		}
4258 		mutex_lock(&dirty_i->seglist_lock);
4259 		__locate_dirty_segment(sbi, segno, DIRTY);
4260 		mutex_unlock(&dirty_i->seglist_lock);
4261 	}
4262 }
4263 
4264 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4265 {
4266 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4267 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4268 
4269 	dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4270 	if (!dirty_i->victim_secmap)
4271 		return -ENOMEM;
4272 	return 0;
4273 }
4274 
4275 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4276 {
4277 	struct dirty_seglist_info *dirty_i;
4278 	unsigned int bitmap_size, i;
4279 
4280 	/* allocate memory for dirty segments list information */
4281 	dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4282 								GFP_KERNEL);
4283 	if (!dirty_i)
4284 		return -ENOMEM;
4285 
4286 	SM_I(sbi)->dirty_info = dirty_i;
4287 	mutex_init(&dirty_i->seglist_lock);
4288 
4289 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4290 
4291 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
4292 		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4293 								GFP_KERNEL);
4294 		if (!dirty_i->dirty_segmap[i])
4295 			return -ENOMEM;
4296 	}
4297 
4298 	init_dirty_segmap(sbi);
4299 	return init_victim_secmap(sbi);
4300 }
4301 
4302 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4303 {
4304 	int i;
4305 
4306 	/*
4307 	 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4308 	 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4309 	 */
4310 	for (i = 0; i < NO_CHECK_TYPE; i++) {
4311 		struct curseg_info *curseg = CURSEG_I(sbi, i);
4312 		struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4313 		unsigned int blkofs = curseg->next_blkoff;
4314 
4315 		if (f2fs_test_bit(blkofs, se->cur_valid_map))
4316 			goto out;
4317 
4318 		if (curseg->alloc_type == SSR)
4319 			continue;
4320 
4321 		for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4322 			if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4323 				continue;
4324 out:
4325 			f2fs_err(sbi,
4326 				 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4327 				 i, curseg->segno, curseg->alloc_type,
4328 				 curseg->next_blkoff, blkofs);
4329 			return -EFSCORRUPTED;
4330 		}
4331 	}
4332 	return 0;
4333 }
4334 
4335 /*
4336  * Update min, max modified time for cost-benefit GC algorithm
4337  */
4338 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4339 {
4340 	struct sit_info *sit_i = SIT_I(sbi);
4341 	unsigned int segno;
4342 
4343 	down_write(&sit_i->sentry_lock);
4344 
4345 	sit_i->min_mtime = ULLONG_MAX;
4346 
4347 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4348 		unsigned int i;
4349 		unsigned long long mtime = 0;
4350 
4351 		for (i = 0; i < sbi->segs_per_sec; i++)
4352 			mtime += get_seg_entry(sbi, segno + i)->mtime;
4353 
4354 		mtime = div_u64(mtime, sbi->segs_per_sec);
4355 
4356 		if (sit_i->min_mtime > mtime)
4357 			sit_i->min_mtime = mtime;
4358 	}
4359 	sit_i->max_mtime = get_mtime(sbi, false);
4360 	up_write(&sit_i->sentry_lock);
4361 }
4362 
4363 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4364 {
4365 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4366 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4367 	struct f2fs_sm_info *sm_info;
4368 	int err;
4369 
4370 	sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4371 	if (!sm_info)
4372 		return -ENOMEM;
4373 
4374 	/* init sm info */
4375 	sbi->sm_info = sm_info;
4376 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4377 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4378 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4379 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4380 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4381 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4382 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4383 	sm_info->rec_prefree_segments = sm_info->main_segments *
4384 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4385 	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4386 		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4387 
4388 	if (!test_opt(sbi, LFS))
4389 		sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4390 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4391 	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4392 	sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4393 	sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4394 	sm_info->min_ssr_sections = reserved_sections(sbi);
4395 
4396 	INIT_LIST_HEAD(&sm_info->sit_entry_set);
4397 
4398 	init_rwsem(&sm_info->curseg_lock);
4399 
4400 	if (!f2fs_readonly(sbi->sb)) {
4401 		err = f2fs_create_flush_cmd_control(sbi);
4402 		if (err)
4403 			return err;
4404 	}
4405 
4406 	err = create_discard_cmd_control(sbi);
4407 	if (err)
4408 		return err;
4409 
4410 	err = build_sit_info(sbi);
4411 	if (err)
4412 		return err;
4413 	err = build_free_segmap(sbi);
4414 	if (err)
4415 		return err;
4416 	err = build_curseg(sbi);
4417 	if (err)
4418 		return err;
4419 
4420 	/* reinit free segmap based on SIT */
4421 	err = build_sit_entries(sbi);
4422 	if (err)
4423 		return err;
4424 
4425 	init_free_segmap(sbi);
4426 	err = build_dirty_segmap(sbi);
4427 	if (err)
4428 		return err;
4429 
4430 	err = sanity_check_curseg(sbi);
4431 	if (err)
4432 		return err;
4433 
4434 	init_min_max_mtime(sbi);
4435 	return 0;
4436 }
4437 
4438 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4439 		enum dirty_type dirty_type)
4440 {
4441 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4442 
4443 	mutex_lock(&dirty_i->seglist_lock);
4444 	kvfree(dirty_i->dirty_segmap[dirty_type]);
4445 	dirty_i->nr_dirty[dirty_type] = 0;
4446 	mutex_unlock(&dirty_i->seglist_lock);
4447 }
4448 
4449 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4450 {
4451 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4452 	kvfree(dirty_i->victim_secmap);
4453 }
4454 
4455 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4456 {
4457 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4458 	int i;
4459 
4460 	if (!dirty_i)
4461 		return;
4462 
4463 	/* discard pre-free/dirty segments list */
4464 	for (i = 0; i < NR_DIRTY_TYPE; i++)
4465 		discard_dirty_segmap(sbi, i);
4466 
4467 	destroy_victim_secmap(sbi);
4468 	SM_I(sbi)->dirty_info = NULL;
4469 	kvfree(dirty_i);
4470 }
4471 
4472 static void destroy_curseg(struct f2fs_sb_info *sbi)
4473 {
4474 	struct curseg_info *array = SM_I(sbi)->curseg_array;
4475 	int i;
4476 
4477 	if (!array)
4478 		return;
4479 	SM_I(sbi)->curseg_array = NULL;
4480 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
4481 		kvfree(array[i].sum_blk);
4482 		kvfree(array[i].journal);
4483 	}
4484 	kvfree(array);
4485 }
4486 
4487 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4488 {
4489 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4490 	if (!free_i)
4491 		return;
4492 	SM_I(sbi)->free_info = NULL;
4493 	kvfree(free_i->free_segmap);
4494 	kvfree(free_i->free_secmap);
4495 	kvfree(free_i);
4496 }
4497 
4498 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4499 {
4500 	struct sit_info *sit_i = SIT_I(sbi);
4501 
4502 	if (!sit_i)
4503 		return;
4504 
4505 	if (sit_i->sentries)
4506 		kvfree(sit_i->bitmap);
4507 	kvfree(sit_i->tmp_map);
4508 
4509 	kvfree(sit_i->sentries);
4510 	kvfree(sit_i->sec_entries);
4511 	kvfree(sit_i->dirty_sentries_bitmap);
4512 
4513 	SM_I(sbi)->sit_info = NULL;
4514 	kvfree(sit_i->sit_bitmap);
4515 #ifdef CONFIG_F2FS_CHECK_FS
4516 	kvfree(sit_i->sit_bitmap_mir);
4517 	kvfree(sit_i->invalid_segmap);
4518 #endif
4519 	kvfree(sit_i);
4520 }
4521 
4522 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4523 {
4524 	struct f2fs_sm_info *sm_info = SM_I(sbi);
4525 
4526 	if (!sm_info)
4527 		return;
4528 	f2fs_destroy_flush_cmd_control(sbi, true);
4529 	destroy_discard_cmd_control(sbi);
4530 	destroy_dirty_segmap(sbi);
4531 	destroy_curseg(sbi);
4532 	destroy_free_segmap(sbi);
4533 	destroy_sit_info(sbi);
4534 	sbi->sm_info = NULL;
4535 	kvfree(sm_info);
4536 }
4537 
4538 int __init f2fs_create_segment_manager_caches(void)
4539 {
4540 	discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4541 			sizeof(struct discard_entry));
4542 	if (!discard_entry_slab)
4543 		goto fail;
4544 
4545 	discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4546 			sizeof(struct discard_cmd));
4547 	if (!discard_cmd_slab)
4548 		goto destroy_discard_entry;
4549 
4550 	sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4551 			sizeof(struct sit_entry_set));
4552 	if (!sit_entry_set_slab)
4553 		goto destroy_discard_cmd;
4554 
4555 	inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4556 			sizeof(struct inmem_pages));
4557 	if (!inmem_entry_slab)
4558 		goto destroy_sit_entry_set;
4559 	return 0;
4560 
4561 destroy_sit_entry_set:
4562 	kmem_cache_destroy(sit_entry_set_slab);
4563 destroy_discard_cmd:
4564 	kmem_cache_destroy(discard_cmd_slab);
4565 destroy_discard_entry:
4566 	kmem_cache_destroy(discard_entry_slab);
4567 fail:
4568 	return -ENOMEM;
4569 }
4570 
4571 void f2fs_destroy_segment_manager_caches(void)
4572 {
4573 	kmem_cache_destroy(sit_entry_set_slab);
4574 	kmem_cache_destroy(discard_cmd_slab);
4575 	kmem_cache_destroy(discard_entry_slab);
4576 	kmem_cache_destroy(inmem_entry_slab);
4577 }
4578