xref: /openbmc/linux/fs/f2fs/segment.c (revision d9fd5a71)
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/events/f2fs.h>
24 
25 #define __reverse_ffz(x) __reverse_ffs(~(x))
26 
27 static struct kmem_cache *discard_entry_slab;
28 static struct kmem_cache *discard_cmd_slab;
29 static struct kmem_cache *sit_entry_set_slab;
30 static struct kmem_cache *inmem_entry_slab;
31 
32 static unsigned long __reverse_ulong(unsigned char *str)
33 {
34 	unsigned long tmp = 0;
35 	int shift = 24, idx = 0;
36 
37 #if BITS_PER_LONG == 64
38 	shift = 56;
39 #endif
40 	while (shift >= 0) {
41 		tmp |= (unsigned long)str[idx++] << shift;
42 		shift -= BITS_PER_BYTE;
43 	}
44 	return tmp;
45 }
46 
47 /*
48  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
49  * MSB and LSB are reversed in a byte by f2fs_set_bit.
50  */
51 static inline unsigned long __reverse_ffs(unsigned long word)
52 {
53 	int num = 0;
54 
55 #if BITS_PER_LONG == 64
56 	if ((word & 0xffffffff00000000UL) == 0)
57 		num += 32;
58 	else
59 		word >>= 32;
60 #endif
61 	if ((word & 0xffff0000) == 0)
62 		num += 16;
63 	else
64 		word >>= 16;
65 
66 	if ((word & 0xff00) == 0)
67 		num += 8;
68 	else
69 		word >>= 8;
70 
71 	if ((word & 0xf0) == 0)
72 		num += 4;
73 	else
74 		word >>= 4;
75 
76 	if ((word & 0xc) == 0)
77 		num += 2;
78 	else
79 		word >>= 2;
80 
81 	if ((word & 0x2) == 0)
82 		num += 1;
83 	return num;
84 }
85 
86 /*
87  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
88  * f2fs_set_bit makes MSB and LSB reversed in a byte.
89  * @size must be integral times of unsigned long.
90  * Example:
91  *                             MSB <--> LSB
92  *   f2fs_set_bit(0, bitmap) => 1000 0000
93  *   f2fs_set_bit(7, bitmap) => 0000 0001
94  */
95 static unsigned long __find_rev_next_bit(const unsigned long *addr,
96 			unsigned long size, unsigned long offset)
97 {
98 	const unsigned long *p = addr + BIT_WORD(offset);
99 	unsigned long result = size;
100 	unsigned long tmp;
101 
102 	if (offset >= size)
103 		return size;
104 
105 	size -= (offset & ~(BITS_PER_LONG - 1));
106 	offset %= BITS_PER_LONG;
107 
108 	while (1) {
109 		if (*p == 0)
110 			goto pass;
111 
112 		tmp = __reverse_ulong((unsigned char *)p);
113 
114 		tmp &= ~0UL >> offset;
115 		if (size < BITS_PER_LONG)
116 			tmp &= (~0UL << (BITS_PER_LONG - size));
117 		if (tmp)
118 			goto found;
119 pass:
120 		if (size <= BITS_PER_LONG)
121 			break;
122 		size -= BITS_PER_LONG;
123 		offset = 0;
124 		p++;
125 	}
126 	return result;
127 found:
128 	return result - size + __reverse_ffs(tmp);
129 }
130 
131 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
132 			unsigned long size, unsigned long offset)
133 {
134 	const unsigned long *p = addr + BIT_WORD(offset);
135 	unsigned long result = size;
136 	unsigned long tmp;
137 
138 	if (offset >= size)
139 		return size;
140 
141 	size -= (offset & ~(BITS_PER_LONG - 1));
142 	offset %= BITS_PER_LONG;
143 
144 	while (1) {
145 		if (*p == ~0UL)
146 			goto pass;
147 
148 		tmp = __reverse_ulong((unsigned char *)p);
149 
150 		if (offset)
151 			tmp |= ~0UL << (BITS_PER_LONG - offset);
152 		if (size < BITS_PER_LONG)
153 			tmp |= ~0UL >> size;
154 		if (tmp != ~0UL)
155 			goto found;
156 pass:
157 		if (size <= BITS_PER_LONG)
158 			break;
159 		size -= BITS_PER_LONG;
160 		offset = 0;
161 		p++;
162 	}
163 	return result;
164 found:
165 	return result - size + __reverse_ffz(tmp);
166 }
167 
168 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
169 {
170 	int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
171 	int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
172 	int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
173 
174 	if (f2fs_lfs_mode(sbi))
175 		return false;
176 	if (sbi->gc_mode == GC_URGENT_HIGH)
177 		return true;
178 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
179 		return true;
180 
181 	return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
182 			SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
183 }
184 
185 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
186 {
187 	struct inmem_pages *new;
188 
189 	f2fs_set_page_private(page, ATOMIC_WRITTEN_PAGE);
190 
191 	new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
192 
193 	/* add atomic page indices to the list */
194 	new->page = page;
195 	INIT_LIST_HEAD(&new->list);
196 
197 	/* increase reference count with clean state */
198 	get_page(page);
199 	mutex_lock(&F2FS_I(inode)->inmem_lock);
200 	list_add_tail(&new->list, &F2FS_I(inode)->inmem_pages);
201 	inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
202 	mutex_unlock(&F2FS_I(inode)->inmem_lock);
203 
204 	trace_f2fs_register_inmem_page(page, INMEM);
205 }
206 
207 static int __revoke_inmem_pages(struct inode *inode,
208 				struct list_head *head, bool drop, bool recover,
209 				bool trylock)
210 {
211 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
212 	struct inmem_pages *cur, *tmp;
213 	int err = 0;
214 
215 	list_for_each_entry_safe(cur, tmp, head, list) {
216 		struct page *page = cur->page;
217 
218 		if (drop)
219 			trace_f2fs_commit_inmem_page(page, INMEM_DROP);
220 
221 		if (trylock) {
222 			/*
223 			 * to avoid deadlock in between page lock and
224 			 * inmem_lock.
225 			 */
226 			if (!trylock_page(page))
227 				continue;
228 		} else {
229 			lock_page(page);
230 		}
231 
232 		f2fs_wait_on_page_writeback(page, DATA, true, true);
233 
234 		if (recover) {
235 			struct dnode_of_data dn;
236 			struct node_info ni;
237 
238 			trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
239 retry:
240 			set_new_dnode(&dn, inode, NULL, NULL, 0);
241 			err = f2fs_get_dnode_of_data(&dn, page->index,
242 								LOOKUP_NODE);
243 			if (err) {
244 				if (err == -ENOMEM) {
245 					congestion_wait(BLK_RW_ASYNC,
246 							DEFAULT_IO_TIMEOUT);
247 					cond_resched();
248 					goto retry;
249 				}
250 				err = -EAGAIN;
251 				goto next;
252 			}
253 
254 			err = f2fs_get_node_info(sbi, dn.nid, &ni);
255 			if (err) {
256 				f2fs_put_dnode(&dn);
257 				return err;
258 			}
259 
260 			if (cur->old_addr == NEW_ADDR) {
261 				f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
262 				f2fs_update_data_blkaddr(&dn, NEW_ADDR);
263 			} else
264 				f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
265 					cur->old_addr, ni.version, true, true);
266 			f2fs_put_dnode(&dn);
267 		}
268 next:
269 		/* we don't need to invalidate this in the sccessful status */
270 		if (drop || recover) {
271 			ClearPageUptodate(page);
272 			clear_cold_data(page);
273 		}
274 		f2fs_clear_page_private(page);
275 		f2fs_put_page(page, 1);
276 
277 		list_del(&cur->list);
278 		kmem_cache_free(inmem_entry_slab, cur);
279 		dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
280 	}
281 	return err;
282 }
283 
284 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
285 {
286 	struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
287 	struct inode *inode;
288 	struct f2fs_inode_info *fi;
289 	unsigned int count = sbi->atomic_files;
290 	unsigned int looped = 0;
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 	if (inode)
300 		list_move_tail(&fi->inmem_ilist, head);
301 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
302 
303 	if (inode) {
304 		if (gc_failure) {
305 			if (!fi->i_gc_failures[GC_FAILURE_ATOMIC])
306 				goto skip;
307 		}
308 		set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
309 		f2fs_drop_inmem_pages(inode);
310 skip:
311 		iput(inode);
312 	}
313 	congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
314 	cond_resched();
315 	if (gc_failure) {
316 		if (++looped >= count)
317 			return;
318 	}
319 	goto next;
320 }
321 
322 void f2fs_drop_inmem_pages(struct inode *inode)
323 {
324 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
325 	struct f2fs_inode_info *fi = F2FS_I(inode);
326 
327 	while (!list_empty(&fi->inmem_pages)) {
328 		mutex_lock(&fi->inmem_lock);
329 		__revoke_inmem_pages(inode, &fi->inmem_pages,
330 						true, false, true);
331 		mutex_unlock(&fi->inmem_lock);
332 	}
333 
334 	fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
335 
336 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
337 	if (!list_empty(&fi->inmem_ilist))
338 		list_del_init(&fi->inmem_ilist);
339 	if (f2fs_is_atomic_file(inode)) {
340 		clear_inode_flag(inode, FI_ATOMIC_FILE);
341 		sbi->atomic_files--;
342 	}
343 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
344 }
345 
346 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
347 {
348 	struct f2fs_inode_info *fi = F2FS_I(inode);
349 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
350 	struct list_head *head = &fi->inmem_pages;
351 	struct inmem_pages *cur = NULL;
352 
353 	f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
354 
355 	mutex_lock(&fi->inmem_lock);
356 	list_for_each_entry(cur, head, list) {
357 		if (cur->page == page)
358 			break;
359 	}
360 
361 	f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
362 	list_del(&cur->list);
363 	mutex_unlock(&fi->inmem_lock);
364 
365 	dec_page_count(sbi, F2FS_INMEM_PAGES);
366 	kmem_cache_free(inmem_entry_slab, cur);
367 
368 	ClearPageUptodate(page);
369 	f2fs_clear_page_private(page);
370 	f2fs_put_page(page, 0);
371 
372 	trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
373 }
374 
375 static int __f2fs_commit_inmem_pages(struct inode *inode)
376 {
377 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
378 	struct f2fs_inode_info *fi = F2FS_I(inode);
379 	struct inmem_pages *cur, *tmp;
380 	struct f2fs_io_info fio = {
381 		.sbi = sbi,
382 		.ino = inode->i_ino,
383 		.type = DATA,
384 		.op = REQ_OP_WRITE,
385 		.op_flags = REQ_SYNC | REQ_PRIO,
386 		.io_type = FS_DATA_IO,
387 	};
388 	struct list_head revoke_list;
389 	bool submit_bio = false;
390 	int err = 0;
391 
392 	INIT_LIST_HEAD(&revoke_list);
393 
394 	list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
395 		struct page *page = cur->page;
396 
397 		lock_page(page);
398 		if (page->mapping == inode->i_mapping) {
399 			trace_f2fs_commit_inmem_page(page, INMEM);
400 
401 			f2fs_wait_on_page_writeback(page, DATA, true, true);
402 
403 			set_page_dirty(page);
404 			if (clear_page_dirty_for_io(page)) {
405 				inode_dec_dirty_pages(inode);
406 				f2fs_remove_dirty_inode(inode);
407 			}
408 retry:
409 			fio.page = page;
410 			fio.old_blkaddr = NULL_ADDR;
411 			fio.encrypted_page = NULL;
412 			fio.need_lock = LOCK_DONE;
413 			err = f2fs_do_write_data_page(&fio);
414 			if (err) {
415 				if (err == -ENOMEM) {
416 					congestion_wait(BLK_RW_ASYNC,
417 							DEFAULT_IO_TIMEOUT);
418 					cond_resched();
419 					goto retry;
420 				}
421 				unlock_page(page);
422 				break;
423 			}
424 			/* record old blkaddr for revoking */
425 			cur->old_addr = fio.old_blkaddr;
426 			submit_bio = true;
427 		}
428 		unlock_page(page);
429 		list_move_tail(&cur->list, &revoke_list);
430 	}
431 
432 	if (submit_bio)
433 		f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
434 
435 	if (err) {
436 		/*
437 		 * try to revoke all committed pages, but still we could fail
438 		 * due to no memory or other reason, if that happened, EAGAIN
439 		 * will be returned, which means in such case, transaction is
440 		 * already not integrity, caller should use journal to do the
441 		 * recovery or rewrite & commit last transaction. For other
442 		 * error number, revoking was done by filesystem itself.
443 		 */
444 		err = __revoke_inmem_pages(inode, &revoke_list,
445 						false, true, false);
446 
447 		/* drop all uncommitted pages */
448 		__revoke_inmem_pages(inode, &fi->inmem_pages,
449 						true, false, false);
450 	} else {
451 		__revoke_inmem_pages(inode, &revoke_list,
452 						false, false, false);
453 	}
454 
455 	return err;
456 }
457 
458 int f2fs_commit_inmem_pages(struct inode *inode)
459 {
460 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
461 	struct f2fs_inode_info *fi = F2FS_I(inode);
462 	int err;
463 
464 	f2fs_balance_fs(sbi, true);
465 
466 	down_write(&fi->i_gc_rwsem[WRITE]);
467 
468 	f2fs_lock_op(sbi);
469 	set_inode_flag(inode, FI_ATOMIC_COMMIT);
470 
471 	mutex_lock(&fi->inmem_lock);
472 	err = __f2fs_commit_inmem_pages(inode);
473 	mutex_unlock(&fi->inmem_lock);
474 
475 	clear_inode_flag(inode, FI_ATOMIC_COMMIT);
476 
477 	f2fs_unlock_op(sbi);
478 	up_write(&fi->i_gc_rwsem[WRITE]);
479 
480 	return err;
481 }
482 
483 /*
484  * This function balances dirty node and dentry pages.
485  * In addition, it controls garbage collection.
486  */
487 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
488 {
489 	if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
490 		f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
491 		f2fs_stop_checkpoint(sbi, false);
492 	}
493 
494 	/* balance_fs_bg is able to be pending */
495 	if (need && excess_cached_nats(sbi))
496 		f2fs_balance_fs_bg(sbi, false);
497 
498 	if (!f2fs_is_checkpoint_ready(sbi))
499 		return;
500 
501 	/*
502 	 * We should do GC or end up with checkpoint, if there are so many dirty
503 	 * dir/node pages without enough free segments.
504 	 */
505 	if (has_not_enough_free_secs(sbi, 0, 0)) {
506 		down_write(&sbi->gc_lock);
507 		f2fs_gc(sbi, false, false, NULL_SEGNO);
508 	}
509 }
510 
511 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
512 {
513 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
514 		return;
515 
516 	/* try to shrink extent cache when there is no enough memory */
517 	if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
518 		f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
519 
520 	/* check the # of cached NAT entries */
521 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
522 		f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
523 
524 	if (!f2fs_available_free_memory(sbi, FREE_NIDS))
525 		f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
526 	else
527 		f2fs_build_free_nids(sbi, false, false);
528 
529 	if (excess_dirty_nats(sbi) || excess_dirty_nodes(sbi) ||
530 		excess_prefree_segs(sbi))
531 		goto do_sync;
532 
533 	/* there is background inflight IO or foreground operation recently */
534 	if (is_inflight_io(sbi, REQ_TIME) ||
535 		(!f2fs_time_over(sbi, REQ_TIME) && rwsem_is_locked(&sbi->cp_rwsem)))
536 		return;
537 
538 	/* exceed periodical checkpoint timeout threshold */
539 	if (f2fs_time_over(sbi, CP_TIME))
540 		goto do_sync;
541 
542 	/* checkpoint is the only way to shrink partial cached entries */
543 	if (f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
544 		f2fs_available_free_memory(sbi, INO_ENTRIES))
545 		return;
546 
547 do_sync:
548 	if (test_opt(sbi, DATA_FLUSH) && from_bg) {
549 		struct blk_plug plug;
550 
551 		mutex_lock(&sbi->flush_lock);
552 
553 		blk_start_plug(&plug);
554 		f2fs_sync_dirty_inodes(sbi, FILE_INODE);
555 		blk_finish_plug(&plug);
556 
557 		mutex_unlock(&sbi->flush_lock);
558 	}
559 	f2fs_sync_fs(sbi->sb, true);
560 	stat_inc_bg_cp_count(sbi->stat_info);
561 }
562 
563 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
564 				struct block_device *bdev)
565 {
566 	int ret = blkdev_issue_flush(bdev);
567 
568 	trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
569 				test_opt(sbi, FLUSH_MERGE), ret);
570 	return ret;
571 }
572 
573 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
574 {
575 	int ret = 0;
576 	int i;
577 
578 	if (!f2fs_is_multi_device(sbi))
579 		return __submit_flush_wait(sbi, sbi->sb->s_bdev);
580 
581 	for (i = 0; i < sbi->s_ndevs; i++) {
582 		if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
583 			continue;
584 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
585 		if (ret)
586 			break;
587 	}
588 	return ret;
589 }
590 
591 static int issue_flush_thread(void *data)
592 {
593 	struct f2fs_sb_info *sbi = data;
594 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
595 	wait_queue_head_t *q = &fcc->flush_wait_queue;
596 repeat:
597 	if (kthread_should_stop())
598 		return 0;
599 
600 	if (!llist_empty(&fcc->issue_list)) {
601 		struct flush_cmd *cmd, *next;
602 		int ret;
603 
604 		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
605 		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
606 
607 		cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
608 
609 		ret = submit_flush_wait(sbi, cmd->ino);
610 		atomic_inc(&fcc->issued_flush);
611 
612 		llist_for_each_entry_safe(cmd, next,
613 					  fcc->dispatch_list, llnode) {
614 			cmd->ret = ret;
615 			complete(&cmd->wait);
616 		}
617 		fcc->dispatch_list = NULL;
618 	}
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 		kfree(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 		kfree(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 	if (test_opt(sbi, NOBARRIER))
753 		return 0;
754 
755 	for (i = 1; i < sbi->s_ndevs; i++) {
756 		if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
757 			continue;
758 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
759 		if (ret)
760 			break;
761 
762 		spin_lock(&sbi->dev_lock);
763 		f2fs_clear_bit(i, (char *)&sbi->dirty_device);
764 		spin_unlock(&sbi->dev_lock);
765 	}
766 
767 	return ret;
768 }
769 
770 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
771 		enum dirty_type dirty_type)
772 {
773 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
774 
775 	/* need not be added */
776 	if (IS_CURSEG(sbi, segno))
777 		return;
778 
779 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
780 		dirty_i->nr_dirty[dirty_type]++;
781 
782 	if (dirty_type == DIRTY) {
783 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
784 		enum dirty_type t = sentry->type;
785 
786 		if (unlikely(t >= DIRTY)) {
787 			f2fs_bug_on(sbi, 1);
788 			return;
789 		}
790 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
791 			dirty_i->nr_dirty[t]++;
792 
793 		if (__is_large_section(sbi)) {
794 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
795 			block_t valid_blocks =
796 				get_valid_blocks(sbi, segno, true);
797 
798 			f2fs_bug_on(sbi, unlikely(!valid_blocks ||
799 					valid_blocks == BLKS_PER_SEC(sbi)));
800 
801 			if (!IS_CURSEC(sbi, secno))
802 				set_bit(secno, dirty_i->dirty_secmap);
803 		}
804 	}
805 }
806 
807 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
808 		enum dirty_type dirty_type)
809 {
810 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
811 	block_t valid_blocks;
812 
813 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
814 		dirty_i->nr_dirty[dirty_type]--;
815 
816 	if (dirty_type == DIRTY) {
817 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
818 		enum dirty_type t = sentry->type;
819 
820 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
821 			dirty_i->nr_dirty[t]--;
822 
823 		valid_blocks = get_valid_blocks(sbi, segno, true);
824 		if (valid_blocks == 0) {
825 			clear_bit(GET_SEC_FROM_SEG(sbi, segno),
826 						dirty_i->victim_secmap);
827 #ifdef CONFIG_F2FS_CHECK_FS
828 			clear_bit(segno, SIT_I(sbi)->invalid_segmap);
829 #endif
830 		}
831 		if (__is_large_section(sbi)) {
832 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
833 
834 			if (!valid_blocks ||
835 					valid_blocks == BLKS_PER_SEC(sbi)) {
836 				clear_bit(secno, dirty_i->dirty_secmap);
837 				return;
838 			}
839 
840 			if (!IS_CURSEC(sbi, secno))
841 				set_bit(secno, dirty_i->dirty_secmap);
842 		}
843 	}
844 }
845 
846 /*
847  * Should not occur error such as -ENOMEM.
848  * Adding dirty entry into seglist is not critical operation.
849  * If a given segment is one of current working segments, it won't be added.
850  */
851 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
852 {
853 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
854 	unsigned short valid_blocks, ckpt_valid_blocks;
855 	unsigned int usable_blocks;
856 
857 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
858 		return;
859 
860 	usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
861 	mutex_lock(&dirty_i->seglist_lock);
862 
863 	valid_blocks = get_valid_blocks(sbi, segno, false);
864 	ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
865 
866 	if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
867 		ckpt_valid_blocks == usable_blocks)) {
868 		__locate_dirty_segment(sbi, segno, PRE);
869 		__remove_dirty_segment(sbi, segno, DIRTY);
870 	} else if (valid_blocks < usable_blocks) {
871 		__locate_dirty_segment(sbi, segno, DIRTY);
872 	} else {
873 		/* Recovery routine with SSR needs this */
874 		__remove_dirty_segment(sbi, segno, DIRTY);
875 	}
876 
877 	mutex_unlock(&dirty_i->seglist_lock);
878 }
879 
880 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
881 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
882 {
883 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
884 	unsigned int segno;
885 
886 	mutex_lock(&dirty_i->seglist_lock);
887 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
888 		if (get_valid_blocks(sbi, segno, false))
889 			continue;
890 		if (IS_CURSEG(sbi, segno))
891 			continue;
892 		__locate_dirty_segment(sbi, segno, PRE);
893 		__remove_dirty_segment(sbi, segno, DIRTY);
894 	}
895 	mutex_unlock(&dirty_i->seglist_lock);
896 }
897 
898 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
899 {
900 	int ovp_hole_segs =
901 		(overprovision_segments(sbi) - reserved_segments(sbi));
902 	block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
903 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
904 	block_t holes[2] = {0, 0};	/* DATA and NODE */
905 	block_t unusable;
906 	struct seg_entry *se;
907 	unsigned int segno;
908 
909 	mutex_lock(&dirty_i->seglist_lock);
910 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
911 		se = get_seg_entry(sbi, segno);
912 		if (IS_NODESEG(se->type))
913 			holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
914 							se->valid_blocks;
915 		else
916 			holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
917 							se->valid_blocks;
918 	}
919 	mutex_unlock(&dirty_i->seglist_lock);
920 
921 	unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
922 	if (unusable > ovp_holes)
923 		return unusable - ovp_holes;
924 	return 0;
925 }
926 
927 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
928 {
929 	int ovp_hole_segs =
930 		(overprovision_segments(sbi) - reserved_segments(sbi));
931 	if (unusable > F2FS_OPTION(sbi).unusable_cap)
932 		return -EAGAIN;
933 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
934 		dirty_segments(sbi) > ovp_hole_segs)
935 		return -EAGAIN;
936 	return 0;
937 }
938 
939 /* This is only used by SBI_CP_DISABLED */
940 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
941 {
942 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
943 	unsigned int segno = 0;
944 
945 	mutex_lock(&dirty_i->seglist_lock);
946 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
947 		if (get_valid_blocks(sbi, segno, false))
948 			continue;
949 		if (get_ckpt_valid_blocks(sbi, segno))
950 			continue;
951 		mutex_unlock(&dirty_i->seglist_lock);
952 		return segno;
953 	}
954 	mutex_unlock(&dirty_i->seglist_lock);
955 	return NULL_SEGNO;
956 }
957 
958 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
959 		struct block_device *bdev, block_t lstart,
960 		block_t start, block_t len)
961 {
962 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
963 	struct list_head *pend_list;
964 	struct discard_cmd *dc;
965 
966 	f2fs_bug_on(sbi, !len);
967 
968 	pend_list = &dcc->pend_list[plist_idx(len)];
969 
970 	dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
971 	INIT_LIST_HEAD(&dc->list);
972 	dc->bdev = bdev;
973 	dc->lstart = lstart;
974 	dc->start = start;
975 	dc->len = len;
976 	dc->ref = 0;
977 	dc->state = D_PREP;
978 	dc->queued = 0;
979 	dc->error = 0;
980 	init_completion(&dc->wait);
981 	list_add_tail(&dc->list, pend_list);
982 	spin_lock_init(&dc->lock);
983 	dc->bio_ref = 0;
984 	atomic_inc(&dcc->discard_cmd_cnt);
985 	dcc->undiscard_blks += len;
986 
987 	return dc;
988 }
989 
990 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
991 				struct block_device *bdev, block_t lstart,
992 				block_t start, block_t len,
993 				struct rb_node *parent, struct rb_node **p,
994 				bool leftmost)
995 {
996 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
997 	struct discard_cmd *dc;
998 
999 	dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1000 
1001 	rb_link_node(&dc->rb_node, parent, p);
1002 	rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1003 
1004 	return dc;
1005 }
1006 
1007 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1008 							struct discard_cmd *dc)
1009 {
1010 	if (dc->state == D_DONE)
1011 		atomic_sub(dc->queued, &dcc->queued_discard);
1012 
1013 	list_del(&dc->list);
1014 	rb_erase_cached(&dc->rb_node, &dcc->root);
1015 	dcc->undiscard_blks -= dc->len;
1016 
1017 	kmem_cache_free(discard_cmd_slab, dc);
1018 
1019 	atomic_dec(&dcc->discard_cmd_cnt);
1020 }
1021 
1022 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1023 							struct discard_cmd *dc)
1024 {
1025 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1026 	unsigned long flags;
1027 
1028 	trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
1029 
1030 	spin_lock_irqsave(&dc->lock, flags);
1031 	if (dc->bio_ref) {
1032 		spin_unlock_irqrestore(&dc->lock, flags);
1033 		return;
1034 	}
1035 	spin_unlock_irqrestore(&dc->lock, flags);
1036 
1037 	f2fs_bug_on(sbi, dc->ref);
1038 
1039 	if (dc->error == -EOPNOTSUPP)
1040 		dc->error = 0;
1041 
1042 	if (dc->error)
1043 		printk_ratelimited(
1044 			"%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1045 			KERN_INFO, sbi->sb->s_id,
1046 			dc->lstart, dc->start, dc->len, dc->error);
1047 	__detach_discard_cmd(dcc, dc);
1048 }
1049 
1050 static void f2fs_submit_discard_endio(struct bio *bio)
1051 {
1052 	struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1053 	unsigned long flags;
1054 
1055 	spin_lock_irqsave(&dc->lock, flags);
1056 	if (!dc->error)
1057 		dc->error = blk_status_to_errno(bio->bi_status);
1058 	dc->bio_ref--;
1059 	if (!dc->bio_ref && dc->state == D_SUBMIT) {
1060 		dc->state = D_DONE;
1061 		complete_all(&dc->wait);
1062 	}
1063 	spin_unlock_irqrestore(&dc->lock, flags);
1064 	bio_put(bio);
1065 }
1066 
1067 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1068 				block_t start, block_t end)
1069 {
1070 #ifdef CONFIG_F2FS_CHECK_FS
1071 	struct seg_entry *sentry;
1072 	unsigned int segno;
1073 	block_t blk = start;
1074 	unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1075 	unsigned long *map;
1076 
1077 	while (blk < end) {
1078 		segno = GET_SEGNO(sbi, blk);
1079 		sentry = get_seg_entry(sbi, segno);
1080 		offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1081 
1082 		if (end < START_BLOCK(sbi, segno + 1))
1083 			size = GET_BLKOFF_FROM_SEG0(sbi, end);
1084 		else
1085 			size = max_blocks;
1086 		map = (unsigned long *)(sentry->cur_valid_map);
1087 		offset = __find_rev_next_bit(map, size, offset);
1088 		f2fs_bug_on(sbi, offset != size);
1089 		blk = START_BLOCK(sbi, segno + 1);
1090 	}
1091 #endif
1092 }
1093 
1094 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1095 				struct discard_policy *dpolicy,
1096 				int discard_type, unsigned int granularity)
1097 {
1098 	/* common policy */
1099 	dpolicy->type = discard_type;
1100 	dpolicy->sync = true;
1101 	dpolicy->ordered = false;
1102 	dpolicy->granularity = granularity;
1103 
1104 	dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1105 	dpolicy->io_aware_gran = MAX_PLIST_NUM;
1106 	dpolicy->timeout = false;
1107 
1108 	if (discard_type == DPOLICY_BG) {
1109 		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1110 		dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1111 		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1112 		dpolicy->io_aware = true;
1113 		dpolicy->sync = false;
1114 		dpolicy->ordered = true;
1115 		if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1116 			dpolicy->granularity = 1;
1117 			dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1118 		}
1119 	} else if (discard_type == DPOLICY_FORCE) {
1120 		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1121 		dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1122 		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1123 		dpolicy->io_aware = false;
1124 	} else if (discard_type == DPOLICY_FSTRIM) {
1125 		dpolicy->io_aware = false;
1126 	} else if (discard_type == DPOLICY_UMOUNT) {
1127 		dpolicy->io_aware = false;
1128 		/* we need to issue all to keep CP_TRIMMED_FLAG */
1129 		dpolicy->granularity = 1;
1130 		dpolicy->timeout = true;
1131 	}
1132 }
1133 
1134 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1135 				struct block_device *bdev, block_t lstart,
1136 				block_t start, block_t len);
1137 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1138 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1139 						struct discard_policy *dpolicy,
1140 						struct discard_cmd *dc,
1141 						unsigned int *issued)
1142 {
1143 	struct block_device *bdev = dc->bdev;
1144 	struct request_queue *q = bdev_get_queue(bdev);
1145 	unsigned int max_discard_blocks =
1146 			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1147 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1148 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1149 					&(dcc->fstrim_list) : &(dcc->wait_list);
1150 	int flag = dpolicy->sync ? REQ_SYNC : 0;
1151 	block_t lstart, start, len, total_len;
1152 	int err = 0;
1153 
1154 	if (dc->state != D_PREP)
1155 		return 0;
1156 
1157 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1158 		return 0;
1159 
1160 	trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1161 
1162 	lstart = dc->lstart;
1163 	start = dc->start;
1164 	len = dc->len;
1165 	total_len = len;
1166 
1167 	dc->len = 0;
1168 
1169 	while (total_len && *issued < dpolicy->max_requests && !err) {
1170 		struct bio *bio = NULL;
1171 		unsigned long flags;
1172 		bool last = true;
1173 
1174 		if (len > max_discard_blocks) {
1175 			len = max_discard_blocks;
1176 			last = false;
1177 		}
1178 
1179 		(*issued)++;
1180 		if (*issued == dpolicy->max_requests)
1181 			last = true;
1182 
1183 		dc->len += len;
1184 
1185 		if (time_to_inject(sbi, FAULT_DISCARD)) {
1186 			f2fs_show_injection_info(sbi, FAULT_DISCARD);
1187 			err = -EIO;
1188 			goto submit;
1189 		}
1190 		err = __blkdev_issue_discard(bdev,
1191 					SECTOR_FROM_BLOCK(start),
1192 					SECTOR_FROM_BLOCK(len),
1193 					GFP_NOFS, 0, &bio);
1194 submit:
1195 		if (err) {
1196 			spin_lock_irqsave(&dc->lock, flags);
1197 			if (dc->state == D_PARTIAL)
1198 				dc->state = D_SUBMIT;
1199 			spin_unlock_irqrestore(&dc->lock, flags);
1200 
1201 			break;
1202 		}
1203 
1204 		f2fs_bug_on(sbi, !bio);
1205 
1206 		/*
1207 		 * should keep before submission to avoid D_DONE
1208 		 * right away
1209 		 */
1210 		spin_lock_irqsave(&dc->lock, flags);
1211 		if (last)
1212 			dc->state = D_SUBMIT;
1213 		else
1214 			dc->state = D_PARTIAL;
1215 		dc->bio_ref++;
1216 		spin_unlock_irqrestore(&dc->lock, flags);
1217 
1218 		atomic_inc(&dcc->queued_discard);
1219 		dc->queued++;
1220 		list_move_tail(&dc->list, wait_list);
1221 
1222 		/* sanity check on discard range */
1223 		__check_sit_bitmap(sbi, lstart, lstart + len);
1224 
1225 		bio->bi_private = dc;
1226 		bio->bi_end_io = f2fs_submit_discard_endio;
1227 		bio->bi_opf |= flag;
1228 		submit_bio(bio);
1229 
1230 		atomic_inc(&dcc->issued_discard);
1231 
1232 		f2fs_update_iostat(sbi, FS_DISCARD, 1);
1233 
1234 		lstart += len;
1235 		start += len;
1236 		total_len -= len;
1237 		len = total_len;
1238 	}
1239 
1240 	if (!err && len) {
1241 		dcc->undiscard_blks -= len;
1242 		__update_discard_tree_range(sbi, bdev, lstart, start, len);
1243 	}
1244 	return err;
1245 }
1246 
1247 static void __insert_discard_tree(struct f2fs_sb_info *sbi,
1248 				struct block_device *bdev, block_t lstart,
1249 				block_t start, block_t len,
1250 				struct rb_node **insert_p,
1251 				struct rb_node *insert_parent)
1252 {
1253 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1254 	struct rb_node **p;
1255 	struct rb_node *parent = NULL;
1256 	bool leftmost = true;
1257 
1258 	if (insert_p && insert_parent) {
1259 		parent = insert_parent;
1260 		p = insert_p;
1261 		goto do_insert;
1262 	}
1263 
1264 	p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1265 							lstart, &leftmost);
1266 do_insert:
1267 	__attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1268 								p, leftmost);
1269 }
1270 
1271 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1272 						struct discard_cmd *dc)
1273 {
1274 	list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1275 }
1276 
1277 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1278 				struct discard_cmd *dc, block_t blkaddr)
1279 {
1280 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1281 	struct discard_info di = dc->di;
1282 	bool modified = false;
1283 
1284 	if (dc->state == D_DONE || dc->len == 1) {
1285 		__remove_discard_cmd(sbi, dc);
1286 		return;
1287 	}
1288 
1289 	dcc->undiscard_blks -= di.len;
1290 
1291 	if (blkaddr > di.lstart) {
1292 		dc->len = blkaddr - dc->lstart;
1293 		dcc->undiscard_blks += dc->len;
1294 		__relocate_discard_cmd(dcc, dc);
1295 		modified = true;
1296 	}
1297 
1298 	if (blkaddr < di.lstart + di.len - 1) {
1299 		if (modified) {
1300 			__insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1301 					di.start + blkaddr + 1 - di.lstart,
1302 					di.lstart + di.len - 1 - blkaddr,
1303 					NULL, NULL);
1304 		} else {
1305 			dc->lstart++;
1306 			dc->len--;
1307 			dc->start++;
1308 			dcc->undiscard_blks += dc->len;
1309 			__relocate_discard_cmd(dcc, dc);
1310 		}
1311 	}
1312 }
1313 
1314 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1315 				struct block_device *bdev, block_t lstart,
1316 				block_t start, block_t len)
1317 {
1318 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1319 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1320 	struct discard_cmd *dc;
1321 	struct discard_info di = {0};
1322 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1323 	struct request_queue *q = bdev_get_queue(bdev);
1324 	unsigned int max_discard_blocks =
1325 			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1326 	block_t end = lstart + len;
1327 
1328 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1329 					NULL, lstart,
1330 					(struct rb_entry **)&prev_dc,
1331 					(struct rb_entry **)&next_dc,
1332 					&insert_p, &insert_parent, true, NULL);
1333 	if (dc)
1334 		prev_dc = dc;
1335 
1336 	if (!prev_dc) {
1337 		di.lstart = lstart;
1338 		di.len = next_dc ? next_dc->lstart - lstart : len;
1339 		di.len = min(di.len, len);
1340 		di.start = start;
1341 	}
1342 
1343 	while (1) {
1344 		struct rb_node *node;
1345 		bool merged = false;
1346 		struct discard_cmd *tdc = NULL;
1347 
1348 		if (prev_dc) {
1349 			di.lstart = prev_dc->lstart + prev_dc->len;
1350 			if (di.lstart < lstart)
1351 				di.lstart = lstart;
1352 			if (di.lstart >= end)
1353 				break;
1354 
1355 			if (!next_dc || next_dc->lstart > end)
1356 				di.len = end - di.lstart;
1357 			else
1358 				di.len = next_dc->lstart - di.lstart;
1359 			di.start = start + di.lstart - lstart;
1360 		}
1361 
1362 		if (!di.len)
1363 			goto next;
1364 
1365 		if (prev_dc && prev_dc->state == D_PREP &&
1366 			prev_dc->bdev == bdev &&
1367 			__is_discard_back_mergeable(&di, &prev_dc->di,
1368 							max_discard_blocks)) {
1369 			prev_dc->di.len += di.len;
1370 			dcc->undiscard_blks += di.len;
1371 			__relocate_discard_cmd(dcc, prev_dc);
1372 			di = prev_dc->di;
1373 			tdc = prev_dc;
1374 			merged = true;
1375 		}
1376 
1377 		if (next_dc && next_dc->state == D_PREP &&
1378 			next_dc->bdev == bdev &&
1379 			__is_discard_front_mergeable(&di, &next_dc->di,
1380 							max_discard_blocks)) {
1381 			next_dc->di.lstart = di.lstart;
1382 			next_dc->di.len += di.len;
1383 			next_dc->di.start = di.start;
1384 			dcc->undiscard_blks += di.len;
1385 			__relocate_discard_cmd(dcc, next_dc);
1386 			if (tdc)
1387 				__remove_discard_cmd(sbi, tdc);
1388 			merged = true;
1389 		}
1390 
1391 		if (!merged) {
1392 			__insert_discard_tree(sbi, bdev, di.lstart, di.start,
1393 							di.len, NULL, NULL);
1394 		}
1395  next:
1396 		prev_dc = next_dc;
1397 		if (!prev_dc)
1398 			break;
1399 
1400 		node = rb_next(&prev_dc->rb_node);
1401 		next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1402 	}
1403 }
1404 
1405 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1406 		struct block_device *bdev, block_t blkstart, block_t blklen)
1407 {
1408 	block_t lblkstart = blkstart;
1409 
1410 	if (!f2fs_bdev_support_discard(bdev))
1411 		return 0;
1412 
1413 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
1414 
1415 	if (f2fs_is_multi_device(sbi)) {
1416 		int devi = f2fs_target_device_index(sbi, blkstart);
1417 
1418 		blkstart -= FDEV(devi).start_blk;
1419 	}
1420 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1421 	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1422 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1423 	return 0;
1424 }
1425 
1426 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1427 					struct discard_policy *dpolicy)
1428 {
1429 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1430 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1431 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1432 	struct discard_cmd *dc;
1433 	struct blk_plug plug;
1434 	unsigned int pos = dcc->next_pos;
1435 	unsigned int issued = 0;
1436 	bool io_interrupted = false;
1437 
1438 	mutex_lock(&dcc->cmd_lock);
1439 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1440 					NULL, pos,
1441 					(struct rb_entry **)&prev_dc,
1442 					(struct rb_entry **)&next_dc,
1443 					&insert_p, &insert_parent, true, NULL);
1444 	if (!dc)
1445 		dc = next_dc;
1446 
1447 	blk_start_plug(&plug);
1448 
1449 	while (dc) {
1450 		struct rb_node *node;
1451 		int err = 0;
1452 
1453 		if (dc->state != D_PREP)
1454 			goto next;
1455 
1456 		if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1457 			io_interrupted = true;
1458 			break;
1459 		}
1460 
1461 		dcc->next_pos = dc->lstart + dc->len;
1462 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1463 
1464 		if (issued >= dpolicy->max_requests)
1465 			break;
1466 next:
1467 		node = rb_next(&dc->rb_node);
1468 		if (err)
1469 			__remove_discard_cmd(sbi, dc);
1470 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1471 	}
1472 
1473 	blk_finish_plug(&plug);
1474 
1475 	if (!dc)
1476 		dcc->next_pos = 0;
1477 
1478 	mutex_unlock(&dcc->cmd_lock);
1479 
1480 	if (!issued && io_interrupted)
1481 		issued = -1;
1482 
1483 	return issued;
1484 }
1485 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1486 					struct discard_policy *dpolicy);
1487 
1488 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1489 					struct discard_policy *dpolicy)
1490 {
1491 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1492 	struct list_head *pend_list;
1493 	struct discard_cmd *dc, *tmp;
1494 	struct blk_plug plug;
1495 	int i, issued;
1496 	bool io_interrupted = false;
1497 
1498 	if (dpolicy->timeout)
1499 		f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1500 
1501 retry:
1502 	issued = 0;
1503 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1504 		if (dpolicy->timeout &&
1505 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1506 			break;
1507 
1508 		if (i + 1 < dpolicy->granularity)
1509 			break;
1510 
1511 		if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1512 			return __issue_discard_cmd_orderly(sbi, dpolicy);
1513 
1514 		pend_list = &dcc->pend_list[i];
1515 
1516 		mutex_lock(&dcc->cmd_lock);
1517 		if (list_empty(pend_list))
1518 			goto next;
1519 		if (unlikely(dcc->rbtree_check))
1520 			f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1521 							&dcc->root, false));
1522 		blk_start_plug(&plug);
1523 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1524 			f2fs_bug_on(sbi, dc->state != D_PREP);
1525 
1526 			if (dpolicy->timeout &&
1527 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1528 				break;
1529 
1530 			if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1531 						!is_idle(sbi, DISCARD_TIME)) {
1532 				io_interrupted = true;
1533 				break;
1534 			}
1535 
1536 			__submit_discard_cmd(sbi, dpolicy, dc, &issued);
1537 
1538 			if (issued >= dpolicy->max_requests)
1539 				break;
1540 		}
1541 		blk_finish_plug(&plug);
1542 next:
1543 		mutex_unlock(&dcc->cmd_lock);
1544 
1545 		if (issued >= dpolicy->max_requests || io_interrupted)
1546 			break;
1547 	}
1548 
1549 	if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1550 		__wait_all_discard_cmd(sbi, dpolicy);
1551 		goto retry;
1552 	}
1553 
1554 	if (!issued && io_interrupted)
1555 		issued = -1;
1556 
1557 	return issued;
1558 }
1559 
1560 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1561 {
1562 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1563 	struct list_head *pend_list;
1564 	struct discard_cmd *dc, *tmp;
1565 	int i;
1566 	bool dropped = false;
1567 
1568 	mutex_lock(&dcc->cmd_lock);
1569 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1570 		pend_list = &dcc->pend_list[i];
1571 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1572 			f2fs_bug_on(sbi, dc->state != D_PREP);
1573 			__remove_discard_cmd(sbi, dc);
1574 			dropped = true;
1575 		}
1576 	}
1577 	mutex_unlock(&dcc->cmd_lock);
1578 
1579 	return dropped;
1580 }
1581 
1582 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1583 {
1584 	__drop_discard_cmd(sbi);
1585 }
1586 
1587 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1588 							struct discard_cmd *dc)
1589 {
1590 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1591 	unsigned int len = 0;
1592 
1593 	wait_for_completion_io(&dc->wait);
1594 	mutex_lock(&dcc->cmd_lock);
1595 	f2fs_bug_on(sbi, dc->state != D_DONE);
1596 	dc->ref--;
1597 	if (!dc->ref) {
1598 		if (!dc->error)
1599 			len = dc->len;
1600 		__remove_discard_cmd(sbi, dc);
1601 	}
1602 	mutex_unlock(&dcc->cmd_lock);
1603 
1604 	return len;
1605 }
1606 
1607 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1608 						struct discard_policy *dpolicy,
1609 						block_t start, block_t end)
1610 {
1611 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1612 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1613 					&(dcc->fstrim_list) : &(dcc->wait_list);
1614 	struct discard_cmd *dc, *tmp;
1615 	bool need_wait;
1616 	unsigned int trimmed = 0;
1617 
1618 next:
1619 	need_wait = false;
1620 
1621 	mutex_lock(&dcc->cmd_lock);
1622 	list_for_each_entry_safe(dc, tmp, wait_list, list) {
1623 		if (dc->lstart + dc->len <= start || end <= dc->lstart)
1624 			continue;
1625 		if (dc->len < dpolicy->granularity)
1626 			continue;
1627 		if (dc->state == D_DONE && !dc->ref) {
1628 			wait_for_completion_io(&dc->wait);
1629 			if (!dc->error)
1630 				trimmed += dc->len;
1631 			__remove_discard_cmd(sbi, dc);
1632 		} else {
1633 			dc->ref++;
1634 			need_wait = true;
1635 			break;
1636 		}
1637 	}
1638 	mutex_unlock(&dcc->cmd_lock);
1639 
1640 	if (need_wait) {
1641 		trimmed += __wait_one_discard_bio(sbi, dc);
1642 		goto next;
1643 	}
1644 
1645 	return trimmed;
1646 }
1647 
1648 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1649 						struct discard_policy *dpolicy)
1650 {
1651 	struct discard_policy dp;
1652 	unsigned int discard_blks;
1653 
1654 	if (dpolicy)
1655 		return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1656 
1657 	/* wait all */
1658 	__init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1659 	discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1660 	__init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1661 	discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1662 
1663 	return discard_blks;
1664 }
1665 
1666 /* This should be covered by global mutex, &sit_i->sentry_lock */
1667 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1668 {
1669 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1670 	struct discard_cmd *dc;
1671 	bool need_wait = false;
1672 
1673 	mutex_lock(&dcc->cmd_lock);
1674 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1675 							NULL, blkaddr);
1676 	if (dc) {
1677 		if (dc->state == D_PREP) {
1678 			__punch_discard_cmd(sbi, dc, blkaddr);
1679 		} else {
1680 			dc->ref++;
1681 			need_wait = true;
1682 		}
1683 	}
1684 	mutex_unlock(&dcc->cmd_lock);
1685 
1686 	if (need_wait)
1687 		__wait_one_discard_bio(sbi, dc);
1688 }
1689 
1690 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1691 {
1692 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1693 
1694 	if (dcc && dcc->f2fs_issue_discard) {
1695 		struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1696 
1697 		dcc->f2fs_issue_discard = NULL;
1698 		kthread_stop(discard_thread);
1699 	}
1700 }
1701 
1702 /* This comes from f2fs_put_super */
1703 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1704 {
1705 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1706 	struct discard_policy dpolicy;
1707 	bool dropped;
1708 
1709 	__init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1710 					dcc->discard_granularity);
1711 	__issue_discard_cmd(sbi, &dpolicy);
1712 	dropped = __drop_discard_cmd(sbi);
1713 
1714 	/* just to make sure there is no pending discard commands */
1715 	__wait_all_discard_cmd(sbi, NULL);
1716 
1717 	f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1718 	return dropped;
1719 }
1720 
1721 static int issue_discard_thread(void *data)
1722 {
1723 	struct f2fs_sb_info *sbi = data;
1724 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1725 	wait_queue_head_t *q = &dcc->discard_wait_queue;
1726 	struct discard_policy dpolicy;
1727 	unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1728 	int issued;
1729 
1730 	set_freezable();
1731 
1732 	do {
1733 		__init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1734 					dcc->discard_granularity);
1735 
1736 		wait_event_interruptible_timeout(*q,
1737 				kthread_should_stop() || freezing(current) ||
1738 				dcc->discard_wake,
1739 				msecs_to_jiffies(wait_ms));
1740 
1741 		if (dcc->discard_wake)
1742 			dcc->discard_wake = 0;
1743 
1744 		/* clean up pending candidates before going to sleep */
1745 		if (atomic_read(&dcc->queued_discard))
1746 			__wait_all_discard_cmd(sbi, NULL);
1747 
1748 		if (try_to_freeze())
1749 			continue;
1750 		if (f2fs_readonly(sbi->sb))
1751 			continue;
1752 		if (kthread_should_stop())
1753 			return 0;
1754 		if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1755 			wait_ms = dpolicy.max_interval;
1756 			continue;
1757 		}
1758 
1759 		if (sbi->gc_mode == GC_URGENT_HIGH)
1760 			__init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1761 
1762 		sb_start_intwrite(sbi->sb);
1763 
1764 		issued = __issue_discard_cmd(sbi, &dpolicy);
1765 		if (issued > 0) {
1766 			__wait_all_discard_cmd(sbi, &dpolicy);
1767 			wait_ms = dpolicy.min_interval;
1768 		} else if (issued == -1){
1769 			wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1770 			if (!wait_ms)
1771 				wait_ms = dpolicy.mid_interval;
1772 		} else {
1773 			wait_ms = dpolicy.max_interval;
1774 		}
1775 
1776 		sb_end_intwrite(sbi->sb);
1777 
1778 	} while (!kthread_should_stop());
1779 	return 0;
1780 }
1781 
1782 #ifdef CONFIG_BLK_DEV_ZONED
1783 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1784 		struct block_device *bdev, block_t blkstart, block_t blklen)
1785 {
1786 	sector_t sector, nr_sects;
1787 	block_t lblkstart = blkstart;
1788 	int devi = 0;
1789 
1790 	if (f2fs_is_multi_device(sbi)) {
1791 		devi = f2fs_target_device_index(sbi, blkstart);
1792 		if (blkstart < FDEV(devi).start_blk ||
1793 		    blkstart > FDEV(devi).end_blk) {
1794 			f2fs_err(sbi, "Invalid block %x", blkstart);
1795 			return -EIO;
1796 		}
1797 		blkstart -= FDEV(devi).start_blk;
1798 	}
1799 
1800 	/* For sequential zones, reset the zone write pointer */
1801 	if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1802 		sector = SECTOR_FROM_BLOCK(blkstart);
1803 		nr_sects = SECTOR_FROM_BLOCK(blklen);
1804 
1805 		if (sector & (bdev_zone_sectors(bdev) - 1) ||
1806 				nr_sects != bdev_zone_sectors(bdev)) {
1807 			f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1808 				 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1809 				 blkstart, blklen);
1810 			return -EIO;
1811 		}
1812 		trace_f2fs_issue_reset_zone(bdev, blkstart);
1813 		return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1814 					sector, nr_sects, GFP_NOFS);
1815 	}
1816 
1817 	/* For conventional zones, use regular discard if supported */
1818 	return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1819 }
1820 #endif
1821 
1822 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1823 		struct block_device *bdev, block_t blkstart, block_t blklen)
1824 {
1825 #ifdef CONFIG_BLK_DEV_ZONED
1826 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1827 		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1828 #endif
1829 	return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1830 }
1831 
1832 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1833 				block_t blkstart, block_t blklen)
1834 {
1835 	sector_t start = blkstart, len = 0;
1836 	struct block_device *bdev;
1837 	struct seg_entry *se;
1838 	unsigned int offset;
1839 	block_t i;
1840 	int err = 0;
1841 
1842 	bdev = f2fs_target_device(sbi, blkstart, NULL);
1843 
1844 	for (i = blkstart; i < blkstart + blklen; i++, len++) {
1845 		if (i != start) {
1846 			struct block_device *bdev2 =
1847 				f2fs_target_device(sbi, i, NULL);
1848 
1849 			if (bdev2 != bdev) {
1850 				err = __issue_discard_async(sbi, bdev,
1851 						start, len);
1852 				if (err)
1853 					return err;
1854 				bdev = bdev2;
1855 				start = i;
1856 				len = 0;
1857 			}
1858 		}
1859 
1860 		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1861 		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1862 
1863 		if (!f2fs_test_and_set_bit(offset, se->discard_map))
1864 			sbi->discard_blks--;
1865 	}
1866 
1867 	if (len)
1868 		err = __issue_discard_async(sbi, bdev, start, len);
1869 	return err;
1870 }
1871 
1872 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1873 							bool check_only)
1874 {
1875 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1876 	int max_blocks = sbi->blocks_per_seg;
1877 	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1878 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1879 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1880 	unsigned long *discard_map = (unsigned long *)se->discard_map;
1881 	unsigned long *dmap = SIT_I(sbi)->tmp_map;
1882 	unsigned int start = 0, end = -1;
1883 	bool force = (cpc->reason & CP_DISCARD);
1884 	struct discard_entry *de = NULL;
1885 	struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1886 	int i;
1887 
1888 	if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1889 		return false;
1890 
1891 	if (!force) {
1892 		if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1893 			SM_I(sbi)->dcc_info->nr_discards >=
1894 				SM_I(sbi)->dcc_info->max_discards)
1895 			return false;
1896 	}
1897 
1898 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1899 	for (i = 0; i < entries; i++)
1900 		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1901 				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1902 
1903 	while (force || SM_I(sbi)->dcc_info->nr_discards <=
1904 				SM_I(sbi)->dcc_info->max_discards) {
1905 		start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1906 		if (start >= max_blocks)
1907 			break;
1908 
1909 		end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1910 		if (force && start && end != max_blocks
1911 					&& (end - start) < cpc->trim_minlen)
1912 			continue;
1913 
1914 		if (check_only)
1915 			return true;
1916 
1917 		if (!de) {
1918 			de = f2fs_kmem_cache_alloc(discard_entry_slab,
1919 								GFP_F2FS_ZERO);
1920 			de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1921 			list_add_tail(&de->list, head);
1922 		}
1923 
1924 		for (i = start; i < end; i++)
1925 			__set_bit_le(i, (void *)de->discard_map);
1926 
1927 		SM_I(sbi)->dcc_info->nr_discards += end - start;
1928 	}
1929 	return false;
1930 }
1931 
1932 static void release_discard_addr(struct discard_entry *entry)
1933 {
1934 	list_del(&entry->list);
1935 	kmem_cache_free(discard_entry_slab, entry);
1936 }
1937 
1938 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1939 {
1940 	struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1941 	struct discard_entry *entry, *this;
1942 
1943 	/* drop caches */
1944 	list_for_each_entry_safe(entry, this, head, list)
1945 		release_discard_addr(entry);
1946 }
1947 
1948 /*
1949  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1950  */
1951 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1952 {
1953 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1954 	unsigned int segno;
1955 
1956 	mutex_lock(&dirty_i->seglist_lock);
1957 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1958 		__set_test_and_free(sbi, segno, false);
1959 	mutex_unlock(&dirty_i->seglist_lock);
1960 }
1961 
1962 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1963 						struct cp_control *cpc)
1964 {
1965 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1966 	struct list_head *head = &dcc->entry_list;
1967 	struct discard_entry *entry, *this;
1968 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1969 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1970 	unsigned int start = 0, end = -1;
1971 	unsigned int secno, start_segno;
1972 	bool force = (cpc->reason & CP_DISCARD);
1973 	bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
1974 
1975 	mutex_lock(&dirty_i->seglist_lock);
1976 
1977 	while (1) {
1978 		int i;
1979 
1980 		if (need_align && end != -1)
1981 			end--;
1982 		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1983 		if (start >= MAIN_SEGS(sbi))
1984 			break;
1985 		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1986 								start + 1);
1987 
1988 		if (need_align) {
1989 			start = rounddown(start, sbi->segs_per_sec);
1990 			end = roundup(end, sbi->segs_per_sec);
1991 		}
1992 
1993 		for (i = start; i < end; i++) {
1994 			if (test_and_clear_bit(i, prefree_map))
1995 				dirty_i->nr_dirty[PRE]--;
1996 		}
1997 
1998 		if (!f2fs_realtime_discard_enable(sbi))
1999 			continue;
2000 
2001 		if (force && start >= cpc->trim_start &&
2002 					(end - 1) <= cpc->trim_end)
2003 				continue;
2004 
2005 		if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) {
2006 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2007 				(end - start) << sbi->log_blocks_per_seg);
2008 			continue;
2009 		}
2010 next:
2011 		secno = GET_SEC_FROM_SEG(sbi, start);
2012 		start_segno = GET_SEG_FROM_SEC(sbi, secno);
2013 		if (!IS_CURSEC(sbi, secno) &&
2014 			!get_valid_blocks(sbi, start, true))
2015 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2016 				sbi->segs_per_sec << sbi->log_blocks_per_seg);
2017 
2018 		start = start_segno + sbi->segs_per_sec;
2019 		if (start < end)
2020 			goto next;
2021 		else
2022 			end = start - 1;
2023 	}
2024 	mutex_unlock(&dirty_i->seglist_lock);
2025 
2026 	/* send small discards */
2027 	list_for_each_entry_safe(entry, this, head, list) {
2028 		unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2029 		bool is_valid = test_bit_le(0, entry->discard_map);
2030 
2031 find_next:
2032 		if (is_valid) {
2033 			next_pos = find_next_zero_bit_le(entry->discard_map,
2034 					sbi->blocks_per_seg, cur_pos);
2035 			len = next_pos - cur_pos;
2036 
2037 			if (f2fs_sb_has_blkzoned(sbi) ||
2038 			    (force && len < cpc->trim_minlen))
2039 				goto skip;
2040 
2041 			f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2042 									len);
2043 			total_len += len;
2044 		} else {
2045 			next_pos = find_next_bit_le(entry->discard_map,
2046 					sbi->blocks_per_seg, cur_pos);
2047 		}
2048 skip:
2049 		cur_pos = next_pos;
2050 		is_valid = !is_valid;
2051 
2052 		if (cur_pos < sbi->blocks_per_seg)
2053 			goto find_next;
2054 
2055 		release_discard_addr(entry);
2056 		dcc->nr_discards -= total_len;
2057 	}
2058 
2059 	wake_up_discard_thread(sbi, false);
2060 }
2061 
2062 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2063 {
2064 	dev_t dev = sbi->sb->s_bdev->bd_dev;
2065 	struct discard_cmd_control *dcc;
2066 	int err = 0, i;
2067 
2068 	if (SM_I(sbi)->dcc_info) {
2069 		dcc = SM_I(sbi)->dcc_info;
2070 		goto init_thread;
2071 	}
2072 
2073 	dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2074 	if (!dcc)
2075 		return -ENOMEM;
2076 
2077 	dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2078 	INIT_LIST_HEAD(&dcc->entry_list);
2079 	for (i = 0; i < MAX_PLIST_NUM; i++)
2080 		INIT_LIST_HEAD(&dcc->pend_list[i]);
2081 	INIT_LIST_HEAD(&dcc->wait_list);
2082 	INIT_LIST_HEAD(&dcc->fstrim_list);
2083 	mutex_init(&dcc->cmd_lock);
2084 	atomic_set(&dcc->issued_discard, 0);
2085 	atomic_set(&dcc->queued_discard, 0);
2086 	atomic_set(&dcc->discard_cmd_cnt, 0);
2087 	dcc->nr_discards = 0;
2088 	dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2089 	dcc->undiscard_blks = 0;
2090 	dcc->next_pos = 0;
2091 	dcc->root = RB_ROOT_CACHED;
2092 	dcc->rbtree_check = false;
2093 
2094 	init_waitqueue_head(&dcc->discard_wait_queue);
2095 	SM_I(sbi)->dcc_info = dcc;
2096 init_thread:
2097 	dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2098 				"f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2099 	if (IS_ERR(dcc->f2fs_issue_discard)) {
2100 		err = PTR_ERR(dcc->f2fs_issue_discard);
2101 		kfree(dcc);
2102 		SM_I(sbi)->dcc_info = NULL;
2103 		return err;
2104 	}
2105 
2106 	return err;
2107 }
2108 
2109 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2110 {
2111 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2112 
2113 	if (!dcc)
2114 		return;
2115 
2116 	f2fs_stop_discard_thread(sbi);
2117 
2118 	/*
2119 	 * Recovery can cache discard commands, so in error path of
2120 	 * fill_super(), it needs to give a chance to handle them.
2121 	 */
2122 	if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2123 		f2fs_issue_discard_timeout(sbi);
2124 
2125 	kfree(dcc);
2126 	SM_I(sbi)->dcc_info = NULL;
2127 }
2128 
2129 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2130 {
2131 	struct sit_info *sit_i = SIT_I(sbi);
2132 
2133 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2134 		sit_i->dirty_sentries++;
2135 		return false;
2136 	}
2137 
2138 	return true;
2139 }
2140 
2141 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2142 					unsigned int segno, int modified)
2143 {
2144 	struct seg_entry *se = get_seg_entry(sbi, segno);
2145 	se->type = type;
2146 	if (modified)
2147 		__mark_sit_entry_dirty(sbi, segno);
2148 }
2149 
2150 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2151 								block_t blkaddr)
2152 {
2153 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
2154 
2155 	if (segno == NULL_SEGNO)
2156 		return 0;
2157 	return get_seg_entry(sbi, segno)->mtime;
2158 }
2159 
2160 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2161 						unsigned long long old_mtime)
2162 {
2163 	struct seg_entry *se;
2164 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
2165 	unsigned long long ctime = get_mtime(sbi, false);
2166 	unsigned long long mtime = old_mtime ? old_mtime : ctime;
2167 
2168 	if (segno == NULL_SEGNO)
2169 		return;
2170 
2171 	se = get_seg_entry(sbi, segno);
2172 
2173 	if (!se->mtime)
2174 		se->mtime = mtime;
2175 	else
2176 		se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2177 						se->valid_blocks + 1);
2178 
2179 	if (ctime > SIT_I(sbi)->max_mtime)
2180 		SIT_I(sbi)->max_mtime = ctime;
2181 }
2182 
2183 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2184 {
2185 	struct seg_entry *se;
2186 	unsigned int segno, offset;
2187 	long int new_vblocks;
2188 	bool exist;
2189 #ifdef CONFIG_F2FS_CHECK_FS
2190 	bool mir_exist;
2191 #endif
2192 
2193 	segno = GET_SEGNO(sbi, blkaddr);
2194 
2195 	se = get_seg_entry(sbi, segno);
2196 	new_vblocks = se->valid_blocks + del;
2197 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2198 
2199 	f2fs_bug_on(sbi, (new_vblocks < 0 ||
2200 			(new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2201 
2202 	se->valid_blocks = new_vblocks;
2203 
2204 	/* Update valid block bitmap */
2205 	if (del > 0) {
2206 		exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2207 #ifdef CONFIG_F2FS_CHECK_FS
2208 		mir_exist = f2fs_test_and_set_bit(offset,
2209 						se->cur_valid_map_mir);
2210 		if (unlikely(exist != mir_exist)) {
2211 			f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2212 				 blkaddr, exist);
2213 			f2fs_bug_on(sbi, 1);
2214 		}
2215 #endif
2216 		if (unlikely(exist)) {
2217 			f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2218 				 blkaddr);
2219 			f2fs_bug_on(sbi, 1);
2220 			se->valid_blocks--;
2221 			del = 0;
2222 		}
2223 
2224 		if (!f2fs_test_and_set_bit(offset, se->discard_map))
2225 			sbi->discard_blks--;
2226 
2227 		/*
2228 		 * SSR should never reuse block which is checkpointed
2229 		 * or newly invalidated.
2230 		 */
2231 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2232 			if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2233 				se->ckpt_valid_blocks++;
2234 		}
2235 	} else {
2236 		exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2237 #ifdef CONFIG_F2FS_CHECK_FS
2238 		mir_exist = f2fs_test_and_clear_bit(offset,
2239 						se->cur_valid_map_mir);
2240 		if (unlikely(exist != mir_exist)) {
2241 			f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2242 				 blkaddr, exist);
2243 			f2fs_bug_on(sbi, 1);
2244 		}
2245 #endif
2246 		if (unlikely(!exist)) {
2247 			f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2248 				 blkaddr);
2249 			f2fs_bug_on(sbi, 1);
2250 			se->valid_blocks++;
2251 			del = 0;
2252 		} else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2253 			/*
2254 			 * If checkpoints are off, we must not reuse data that
2255 			 * was used in the previous checkpoint. If it was used
2256 			 * before, we must track that to know how much space we
2257 			 * really have.
2258 			 */
2259 			if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2260 				spin_lock(&sbi->stat_lock);
2261 				sbi->unusable_block_count++;
2262 				spin_unlock(&sbi->stat_lock);
2263 			}
2264 		}
2265 
2266 		if (f2fs_test_and_clear_bit(offset, se->discard_map))
2267 			sbi->discard_blks++;
2268 	}
2269 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2270 		se->ckpt_valid_blocks += del;
2271 
2272 	__mark_sit_entry_dirty(sbi, segno);
2273 
2274 	/* update total number of valid blocks to be written in ckpt area */
2275 	SIT_I(sbi)->written_valid_blocks += del;
2276 
2277 	if (__is_large_section(sbi))
2278 		get_sec_entry(sbi, segno)->valid_blocks += del;
2279 }
2280 
2281 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2282 {
2283 	unsigned int segno = GET_SEGNO(sbi, addr);
2284 	struct sit_info *sit_i = SIT_I(sbi);
2285 
2286 	f2fs_bug_on(sbi, addr == NULL_ADDR);
2287 	if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2288 		return;
2289 
2290 	invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2291 
2292 	/* add it into sit main buffer */
2293 	down_write(&sit_i->sentry_lock);
2294 
2295 	update_segment_mtime(sbi, addr, 0);
2296 	update_sit_entry(sbi, addr, -1);
2297 
2298 	/* add it into dirty seglist */
2299 	locate_dirty_segment(sbi, segno);
2300 
2301 	up_write(&sit_i->sentry_lock);
2302 }
2303 
2304 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2305 {
2306 	struct sit_info *sit_i = SIT_I(sbi);
2307 	unsigned int segno, offset;
2308 	struct seg_entry *se;
2309 	bool is_cp = false;
2310 
2311 	if (!__is_valid_data_blkaddr(blkaddr))
2312 		return true;
2313 
2314 	down_read(&sit_i->sentry_lock);
2315 
2316 	segno = GET_SEGNO(sbi, blkaddr);
2317 	se = get_seg_entry(sbi, segno);
2318 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2319 
2320 	if (f2fs_test_bit(offset, se->ckpt_valid_map))
2321 		is_cp = true;
2322 
2323 	up_read(&sit_i->sentry_lock);
2324 
2325 	return is_cp;
2326 }
2327 
2328 /*
2329  * This function should be resided under the curseg_mutex lock
2330  */
2331 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2332 					struct f2fs_summary *sum)
2333 {
2334 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2335 	void *addr = curseg->sum_blk;
2336 	addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2337 	memcpy(addr, sum, sizeof(struct f2fs_summary));
2338 }
2339 
2340 /*
2341  * Calculate the number of current summary pages for writing
2342  */
2343 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2344 {
2345 	int valid_sum_count = 0;
2346 	int i, sum_in_page;
2347 
2348 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2349 		if (sbi->ckpt->alloc_type[i] == SSR)
2350 			valid_sum_count += sbi->blocks_per_seg;
2351 		else {
2352 			if (for_ra)
2353 				valid_sum_count += le16_to_cpu(
2354 					F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2355 			else
2356 				valid_sum_count += curseg_blkoff(sbi, i);
2357 		}
2358 	}
2359 
2360 	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2361 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2362 	if (valid_sum_count <= sum_in_page)
2363 		return 1;
2364 	else if ((valid_sum_count - sum_in_page) <=
2365 		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2366 		return 2;
2367 	return 3;
2368 }
2369 
2370 /*
2371  * Caller should put this summary page
2372  */
2373 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2374 {
2375 	if (unlikely(f2fs_cp_error(sbi)))
2376 		return ERR_PTR(-EIO);
2377 	return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2378 }
2379 
2380 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2381 					void *src, block_t blk_addr)
2382 {
2383 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2384 
2385 	memcpy(page_address(page), src, PAGE_SIZE);
2386 	set_page_dirty(page);
2387 	f2fs_put_page(page, 1);
2388 }
2389 
2390 static void write_sum_page(struct f2fs_sb_info *sbi,
2391 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
2392 {
2393 	f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2394 }
2395 
2396 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2397 						int type, block_t blk_addr)
2398 {
2399 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2400 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2401 	struct f2fs_summary_block *src = curseg->sum_blk;
2402 	struct f2fs_summary_block *dst;
2403 
2404 	dst = (struct f2fs_summary_block *)page_address(page);
2405 	memset(dst, 0, PAGE_SIZE);
2406 
2407 	mutex_lock(&curseg->curseg_mutex);
2408 
2409 	down_read(&curseg->journal_rwsem);
2410 	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2411 	up_read(&curseg->journal_rwsem);
2412 
2413 	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2414 	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2415 
2416 	mutex_unlock(&curseg->curseg_mutex);
2417 
2418 	set_page_dirty(page);
2419 	f2fs_put_page(page, 1);
2420 }
2421 
2422 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2423 				struct curseg_info *curseg, int type)
2424 {
2425 	unsigned int segno = curseg->segno + 1;
2426 	struct free_segmap_info *free_i = FREE_I(sbi);
2427 
2428 	if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2429 		return !test_bit(segno, free_i->free_segmap);
2430 	return 0;
2431 }
2432 
2433 /*
2434  * Find a new segment from the free segments bitmap to right order
2435  * This function should be returned with success, otherwise BUG
2436  */
2437 static void get_new_segment(struct f2fs_sb_info *sbi,
2438 			unsigned int *newseg, bool new_sec, int dir)
2439 {
2440 	struct free_segmap_info *free_i = FREE_I(sbi);
2441 	unsigned int segno, secno, zoneno;
2442 	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2443 	unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2444 	unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2445 	unsigned int left_start = hint;
2446 	bool init = true;
2447 	int go_left = 0;
2448 	int i;
2449 
2450 	spin_lock(&free_i->segmap_lock);
2451 
2452 	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2453 		segno = find_next_zero_bit(free_i->free_segmap,
2454 			GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2455 		if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2456 			goto got_it;
2457 	}
2458 find_other_zone:
2459 	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2460 	if (secno >= MAIN_SECS(sbi)) {
2461 		if (dir == ALLOC_RIGHT) {
2462 			secno = find_next_zero_bit(free_i->free_secmap,
2463 							MAIN_SECS(sbi), 0);
2464 			f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2465 		} else {
2466 			go_left = 1;
2467 			left_start = hint - 1;
2468 		}
2469 	}
2470 	if (go_left == 0)
2471 		goto skip_left;
2472 
2473 	while (test_bit(left_start, free_i->free_secmap)) {
2474 		if (left_start > 0) {
2475 			left_start--;
2476 			continue;
2477 		}
2478 		left_start = find_next_zero_bit(free_i->free_secmap,
2479 							MAIN_SECS(sbi), 0);
2480 		f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2481 		break;
2482 	}
2483 	secno = left_start;
2484 skip_left:
2485 	segno = GET_SEG_FROM_SEC(sbi, secno);
2486 	zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2487 
2488 	/* give up on finding another zone */
2489 	if (!init)
2490 		goto got_it;
2491 	if (sbi->secs_per_zone == 1)
2492 		goto got_it;
2493 	if (zoneno == old_zoneno)
2494 		goto got_it;
2495 	if (dir == ALLOC_LEFT) {
2496 		if (!go_left && zoneno + 1 >= total_zones)
2497 			goto got_it;
2498 		if (go_left && zoneno == 0)
2499 			goto got_it;
2500 	}
2501 	for (i = 0; i < NR_CURSEG_TYPE; i++)
2502 		if (CURSEG_I(sbi, i)->zone == zoneno)
2503 			break;
2504 
2505 	if (i < NR_CURSEG_TYPE) {
2506 		/* zone is in user, try another */
2507 		if (go_left)
2508 			hint = zoneno * sbi->secs_per_zone - 1;
2509 		else if (zoneno + 1 >= total_zones)
2510 			hint = 0;
2511 		else
2512 			hint = (zoneno + 1) * sbi->secs_per_zone;
2513 		init = false;
2514 		goto find_other_zone;
2515 	}
2516 got_it:
2517 	/* set it as dirty segment in free segmap */
2518 	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2519 	__set_inuse(sbi, segno);
2520 	*newseg = segno;
2521 	spin_unlock(&free_i->segmap_lock);
2522 }
2523 
2524 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2525 {
2526 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2527 	struct summary_footer *sum_footer;
2528 	unsigned short seg_type = curseg->seg_type;
2529 
2530 	curseg->inited = true;
2531 	curseg->segno = curseg->next_segno;
2532 	curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2533 	curseg->next_blkoff = 0;
2534 	curseg->next_segno = NULL_SEGNO;
2535 
2536 	sum_footer = &(curseg->sum_blk->footer);
2537 	memset(sum_footer, 0, sizeof(struct summary_footer));
2538 
2539 	sanity_check_seg_type(sbi, seg_type);
2540 
2541 	if (IS_DATASEG(seg_type))
2542 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2543 	if (IS_NODESEG(seg_type))
2544 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2545 	__set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2546 }
2547 
2548 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2549 {
2550 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2551 	unsigned short seg_type = curseg->seg_type;
2552 
2553 	sanity_check_seg_type(sbi, seg_type);
2554 
2555 	/* if segs_per_sec is large than 1, we need to keep original policy. */
2556 	if (__is_large_section(sbi))
2557 		return curseg->segno;
2558 
2559 	/* inmem log may not locate on any segment after mount */
2560 	if (!curseg->inited)
2561 		return 0;
2562 
2563 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2564 		return 0;
2565 
2566 	if (test_opt(sbi, NOHEAP) &&
2567 		(seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)))
2568 		return 0;
2569 
2570 	if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2571 		return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2572 
2573 	/* find segments from 0 to reuse freed segments */
2574 	if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2575 		return 0;
2576 
2577 	return curseg->segno;
2578 }
2579 
2580 /*
2581  * Allocate a current working segment.
2582  * This function always allocates a free segment in LFS manner.
2583  */
2584 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2585 {
2586 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2587 	unsigned short seg_type = curseg->seg_type;
2588 	unsigned int segno = curseg->segno;
2589 	int dir = ALLOC_LEFT;
2590 
2591 	if (curseg->inited)
2592 		write_sum_page(sbi, curseg->sum_blk,
2593 				GET_SUM_BLOCK(sbi, segno));
2594 	if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA)
2595 		dir = ALLOC_RIGHT;
2596 
2597 	if (test_opt(sbi, NOHEAP))
2598 		dir = ALLOC_RIGHT;
2599 
2600 	segno = __get_next_segno(sbi, type);
2601 	get_new_segment(sbi, &segno, new_sec, dir);
2602 	curseg->next_segno = segno;
2603 	reset_curseg(sbi, type, 1);
2604 	curseg->alloc_type = LFS;
2605 }
2606 
2607 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2608 			struct curseg_info *seg, block_t start)
2609 {
2610 	struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2611 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2612 	unsigned long *target_map = SIT_I(sbi)->tmp_map;
2613 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2614 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2615 	int i, pos;
2616 
2617 	for (i = 0; i < entries; i++)
2618 		target_map[i] = ckpt_map[i] | cur_map[i];
2619 
2620 	pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2621 
2622 	seg->next_blkoff = pos;
2623 }
2624 
2625 /*
2626  * If a segment is written by LFS manner, next block offset is just obtained
2627  * by increasing the current block offset. However, if a segment is written by
2628  * SSR manner, next block offset obtained by calling __next_free_blkoff
2629  */
2630 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2631 				struct curseg_info *seg)
2632 {
2633 	if (seg->alloc_type == SSR)
2634 		__next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2635 	else
2636 		seg->next_blkoff++;
2637 }
2638 
2639 /*
2640  * This function always allocates a used segment(from dirty seglist) by SSR
2641  * manner, so it should recover the existing segment information of valid blocks
2642  */
2643 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool flush)
2644 {
2645 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2646 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2647 	unsigned int new_segno = curseg->next_segno;
2648 	struct f2fs_summary_block *sum_node;
2649 	struct page *sum_page;
2650 
2651 	if (flush)
2652 		write_sum_page(sbi, curseg->sum_blk,
2653 					GET_SUM_BLOCK(sbi, curseg->segno));
2654 
2655 	__set_test_and_inuse(sbi, new_segno);
2656 
2657 	mutex_lock(&dirty_i->seglist_lock);
2658 	__remove_dirty_segment(sbi, new_segno, PRE);
2659 	__remove_dirty_segment(sbi, new_segno, DIRTY);
2660 	mutex_unlock(&dirty_i->seglist_lock);
2661 
2662 	reset_curseg(sbi, type, 1);
2663 	curseg->alloc_type = SSR;
2664 	__next_free_blkoff(sbi, curseg, 0);
2665 
2666 	sum_page = f2fs_get_sum_page(sbi, new_segno);
2667 	if (IS_ERR(sum_page)) {
2668 		/* GC won't be able to use stale summary pages by cp_error */
2669 		memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2670 		return;
2671 	}
2672 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2673 	memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2674 	f2fs_put_page(sum_page, 1);
2675 }
2676 
2677 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2678 				int alloc_mode, unsigned long long age);
2679 
2680 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2681 					int target_type, int alloc_mode,
2682 					unsigned long long age)
2683 {
2684 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2685 
2686 	curseg->seg_type = target_type;
2687 
2688 	if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2689 		struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2690 
2691 		curseg->seg_type = se->type;
2692 		change_curseg(sbi, type, true);
2693 	} else {
2694 		/* allocate cold segment by default */
2695 		curseg->seg_type = CURSEG_COLD_DATA;
2696 		new_curseg(sbi, type, true);
2697 	}
2698 	stat_inc_seg_type(sbi, curseg);
2699 }
2700 
2701 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
2702 {
2703 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2704 
2705 	if (!sbi->am.atgc_enabled)
2706 		return;
2707 
2708 	down_read(&SM_I(sbi)->curseg_lock);
2709 
2710 	mutex_lock(&curseg->curseg_mutex);
2711 	down_write(&SIT_I(sbi)->sentry_lock);
2712 
2713 	get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0);
2714 
2715 	up_write(&SIT_I(sbi)->sentry_lock);
2716 	mutex_unlock(&curseg->curseg_mutex);
2717 
2718 	up_read(&SM_I(sbi)->curseg_lock);
2719 
2720 }
2721 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2722 {
2723 	__f2fs_init_atgc_curseg(sbi);
2724 }
2725 
2726 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2727 {
2728 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2729 
2730 	mutex_lock(&curseg->curseg_mutex);
2731 	if (!curseg->inited)
2732 		goto out;
2733 
2734 	if (get_valid_blocks(sbi, curseg->segno, false)) {
2735 		write_sum_page(sbi, curseg->sum_blk,
2736 				GET_SUM_BLOCK(sbi, curseg->segno));
2737 	} else {
2738 		mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2739 		__set_test_and_free(sbi, curseg->segno, true);
2740 		mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2741 	}
2742 out:
2743 	mutex_unlock(&curseg->curseg_mutex);
2744 }
2745 
2746 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
2747 {
2748 	__f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2749 
2750 	if (sbi->am.atgc_enabled)
2751 		__f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2752 }
2753 
2754 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2755 {
2756 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2757 
2758 	mutex_lock(&curseg->curseg_mutex);
2759 	if (!curseg->inited)
2760 		goto out;
2761 	if (get_valid_blocks(sbi, curseg->segno, false))
2762 		goto out;
2763 
2764 	mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2765 	__set_test_and_inuse(sbi, curseg->segno);
2766 	mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2767 out:
2768 	mutex_unlock(&curseg->curseg_mutex);
2769 }
2770 
2771 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
2772 {
2773 	__f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2774 
2775 	if (sbi->am.atgc_enabled)
2776 		__f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2777 }
2778 
2779 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2780 				int alloc_mode, unsigned long long age)
2781 {
2782 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2783 	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2784 	unsigned segno = NULL_SEGNO;
2785 	unsigned short seg_type = curseg->seg_type;
2786 	int i, cnt;
2787 	bool reversed = false;
2788 
2789 	sanity_check_seg_type(sbi, seg_type);
2790 
2791 	/* f2fs_need_SSR() already forces to do this */
2792 	if (!v_ops->get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
2793 		curseg->next_segno = segno;
2794 		return 1;
2795 	}
2796 
2797 	/* For node segments, let's do SSR more intensively */
2798 	if (IS_NODESEG(seg_type)) {
2799 		if (seg_type >= CURSEG_WARM_NODE) {
2800 			reversed = true;
2801 			i = CURSEG_COLD_NODE;
2802 		} else {
2803 			i = CURSEG_HOT_NODE;
2804 		}
2805 		cnt = NR_CURSEG_NODE_TYPE;
2806 	} else {
2807 		if (seg_type >= CURSEG_WARM_DATA) {
2808 			reversed = true;
2809 			i = CURSEG_COLD_DATA;
2810 		} else {
2811 			i = CURSEG_HOT_DATA;
2812 		}
2813 		cnt = NR_CURSEG_DATA_TYPE;
2814 	}
2815 
2816 	for (; cnt-- > 0; reversed ? i-- : i++) {
2817 		if (i == seg_type)
2818 			continue;
2819 		if (!v_ops->get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
2820 			curseg->next_segno = segno;
2821 			return 1;
2822 		}
2823 	}
2824 
2825 	/* find valid_blocks=0 in dirty list */
2826 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2827 		segno = get_free_segment(sbi);
2828 		if (segno != NULL_SEGNO) {
2829 			curseg->next_segno = segno;
2830 			return 1;
2831 		}
2832 	}
2833 	return 0;
2834 }
2835 
2836 /*
2837  * flush out current segment and replace it with new segment
2838  * This function should be returned with success, otherwise BUG
2839  */
2840 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2841 						int type, bool force)
2842 {
2843 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2844 
2845 	if (force)
2846 		new_curseg(sbi, type, true);
2847 	else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2848 					curseg->seg_type == CURSEG_WARM_NODE)
2849 		new_curseg(sbi, type, false);
2850 	else if (curseg->alloc_type == LFS &&
2851 			is_next_segment_free(sbi, curseg, type) &&
2852 			likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2853 		new_curseg(sbi, type, false);
2854 	else if (f2fs_need_SSR(sbi) &&
2855 			get_ssr_segment(sbi, type, SSR, 0))
2856 		change_curseg(sbi, type, true);
2857 	else
2858 		new_curseg(sbi, type, false);
2859 
2860 	stat_inc_seg_type(sbi, curseg);
2861 }
2862 
2863 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2864 					unsigned int start, unsigned int end)
2865 {
2866 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2867 	unsigned int segno;
2868 
2869 	down_read(&SM_I(sbi)->curseg_lock);
2870 	mutex_lock(&curseg->curseg_mutex);
2871 	down_write(&SIT_I(sbi)->sentry_lock);
2872 
2873 	segno = CURSEG_I(sbi, type)->segno;
2874 	if (segno < start || segno > end)
2875 		goto unlock;
2876 
2877 	if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
2878 		change_curseg(sbi, type, true);
2879 	else
2880 		new_curseg(sbi, type, true);
2881 
2882 	stat_inc_seg_type(sbi, curseg);
2883 
2884 	locate_dirty_segment(sbi, segno);
2885 unlock:
2886 	up_write(&SIT_I(sbi)->sentry_lock);
2887 
2888 	if (segno != curseg->segno)
2889 		f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2890 			    type, segno, curseg->segno);
2891 
2892 	mutex_unlock(&curseg->curseg_mutex);
2893 	up_read(&SM_I(sbi)->curseg_lock);
2894 }
2895 
2896 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type)
2897 {
2898 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2899 	unsigned int old_segno;
2900 
2901 	if (!curseg->inited)
2902 		goto alloc;
2903 
2904 	if (!curseg->next_blkoff &&
2905 		!get_valid_blocks(sbi, curseg->segno, false) &&
2906 		!get_ckpt_valid_blocks(sbi, curseg->segno))
2907 		return;
2908 
2909 alloc:
2910 	old_segno = curseg->segno;
2911 	SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
2912 	locate_dirty_segment(sbi, old_segno);
2913 }
2914 
2915 void f2fs_allocate_new_segment(struct f2fs_sb_info *sbi, int type)
2916 {
2917 	down_write(&SIT_I(sbi)->sentry_lock);
2918 	__allocate_new_segment(sbi, type);
2919 	up_write(&SIT_I(sbi)->sentry_lock);
2920 }
2921 
2922 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2923 {
2924 	int i;
2925 
2926 	down_write(&SIT_I(sbi)->sentry_lock);
2927 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
2928 		__allocate_new_segment(sbi, i);
2929 	up_write(&SIT_I(sbi)->sentry_lock);
2930 }
2931 
2932 static const struct segment_allocation default_salloc_ops = {
2933 	.allocate_segment = allocate_segment_by_default,
2934 };
2935 
2936 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2937 						struct cp_control *cpc)
2938 {
2939 	__u64 trim_start = cpc->trim_start;
2940 	bool has_candidate = false;
2941 
2942 	down_write(&SIT_I(sbi)->sentry_lock);
2943 	for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2944 		if (add_discard_addrs(sbi, cpc, true)) {
2945 			has_candidate = true;
2946 			break;
2947 		}
2948 	}
2949 	up_write(&SIT_I(sbi)->sentry_lock);
2950 
2951 	cpc->trim_start = trim_start;
2952 	return has_candidate;
2953 }
2954 
2955 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2956 					struct discard_policy *dpolicy,
2957 					unsigned int start, unsigned int end)
2958 {
2959 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2960 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2961 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
2962 	struct discard_cmd *dc;
2963 	struct blk_plug plug;
2964 	int issued;
2965 	unsigned int trimmed = 0;
2966 
2967 next:
2968 	issued = 0;
2969 
2970 	mutex_lock(&dcc->cmd_lock);
2971 	if (unlikely(dcc->rbtree_check))
2972 		f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2973 							&dcc->root, false));
2974 
2975 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2976 					NULL, start,
2977 					(struct rb_entry **)&prev_dc,
2978 					(struct rb_entry **)&next_dc,
2979 					&insert_p, &insert_parent, true, NULL);
2980 	if (!dc)
2981 		dc = next_dc;
2982 
2983 	blk_start_plug(&plug);
2984 
2985 	while (dc && dc->lstart <= end) {
2986 		struct rb_node *node;
2987 		int err = 0;
2988 
2989 		if (dc->len < dpolicy->granularity)
2990 			goto skip;
2991 
2992 		if (dc->state != D_PREP) {
2993 			list_move_tail(&dc->list, &dcc->fstrim_list);
2994 			goto skip;
2995 		}
2996 
2997 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2998 
2999 		if (issued >= dpolicy->max_requests) {
3000 			start = dc->lstart + dc->len;
3001 
3002 			if (err)
3003 				__remove_discard_cmd(sbi, dc);
3004 
3005 			blk_finish_plug(&plug);
3006 			mutex_unlock(&dcc->cmd_lock);
3007 			trimmed += __wait_all_discard_cmd(sbi, NULL);
3008 			congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
3009 			goto next;
3010 		}
3011 skip:
3012 		node = rb_next(&dc->rb_node);
3013 		if (err)
3014 			__remove_discard_cmd(sbi, dc);
3015 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3016 
3017 		if (fatal_signal_pending(current))
3018 			break;
3019 	}
3020 
3021 	blk_finish_plug(&plug);
3022 	mutex_unlock(&dcc->cmd_lock);
3023 
3024 	return trimmed;
3025 }
3026 
3027 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3028 {
3029 	__u64 start = F2FS_BYTES_TO_BLK(range->start);
3030 	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3031 	unsigned int start_segno, end_segno;
3032 	block_t start_block, end_block;
3033 	struct cp_control cpc;
3034 	struct discard_policy dpolicy;
3035 	unsigned long long trimmed = 0;
3036 	int err = 0;
3037 	bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3038 
3039 	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3040 		return -EINVAL;
3041 
3042 	if (end < MAIN_BLKADDR(sbi))
3043 		goto out;
3044 
3045 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3046 		f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3047 		return -EFSCORRUPTED;
3048 	}
3049 
3050 	/* start/end segment number in main_area */
3051 	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3052 	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3053 						GET_SEGNO(sbi, end);
3054 	if (need_align) {
3055 		start_segno = rounddown(start_segno, sbi->segs_per_sec);
3056 		end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
3057 	}
3058 
3059 	cpc.reason = CP_DISCARD;
3060 	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3061 	cpc.trim_start = start_segno;
3062 	cpc.trim_end = end_segno;
3063 
3064 	if (sbi->discard_blks == 0)
3065 		goto out;
3066 
3067 	down_write(&sbi->gc_lock);
3068 	err = f2fs_write_checkpoint(sbi, &cpc);
3069 	up_write(&sbi->gc_lock);
3070 	if (err)
3071 		goto out;
3072 
3073 	/*
3074 	 * We filed discard candidates, but actually we don't need to wait for
3075 	 * all of them, since they'll be issued in idle time along with runtime
3076 	 * discard option. User configuration looks like using runtime discard
3077 	 * or periodic fstrim instead of it.
3078 	 */
3079 	if (f2fs_realtime_discard_enable(sbi))
3080 		goto out;
3081 
3082 	start_block = START_BLOCK(sbi, start_segno);
3083 	end_block = START_BLOCK(sbi, end_segno + 1);
3084 
3085 	__init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3086 	trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3087 					start_block, end_block);
3088 
3089 	trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3090 					start_block, end_block);
3091 out:
3092 	if (!err)
3093 		range->len = F2FS_BLK_TO_BYTES(trimmed);
3094 	return err;
3095 }
3096 
3097 static bool __has_curseg_space(struct f2fs_sb_info *sbi,
3098 					struct curseg_info *curseg)
3099 {
3100 	return curseg->next_blkoff < f2fs_usable_blks_in_seg(sbi,
3101 							curseg->segno);
3102 }
3103 
3104 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3105 {
3106 	switch (hint) {
3107 	case WRITE_LIFE_SHORT:
3108 		return CURSEG_HOT_DATA;
3109 	case WRITE_LIFE_EXTREME:
3110 		return CURSEG_COLD_DATA;
3111 	default:
3112 		return CURSEG_WARM_DATA;
3113 	}
3114 }
3115 
3116 /* This returns write hints for each segment type. This hints will be
3117  * passed down to block layer. There are mapping tables which depend on
3118  * the mount option 'whint_mode'.
3119  *
3120  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
3121  *
3122  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
3123  *
3124  * User                  F2FS                     Block
3125  * ----                  ----                     -----
3126  *                       META                     WRITE_LIFE_NOT_SET
3127  *                       HOT_NODE                 "
3128  *                       WARM_NODE                "
3129  *                       COLD_NODE                "
3130  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
3131  * extension list        "                        "
3132  *
3133  * -- buffered io
3134  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3135  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3136  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3137  * WRITE_LIFE_NONE       "                        "
3138  * WRITE_LIFE_MEDIUM     "                        "
3139  * WRITE_LIFE_LONG       "                        "
3140  *
3141  * -- direct io
3142  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3143  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3144  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3145  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
3146  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
3147  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
3148  *
3149  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
3150  *
3151  * User                  F2FS                     Block
3152  * ----                  ----                     -----
3153  *                       META                     WRITE_LIFE_MEDIUM;
3154  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
3155  *                       WARM_NODE                "
3156  *                       COLD_NODE                WRITE_LIFE_NONE
3157  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
3158  * extension list        "                        "
3159  *
3160  * -- buffered io
3161  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3162  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3163  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
3164  * WRITE_LIFE_NONE       "                        "
3165  * WRITE_LIFE_MEDIUM     "                        "
3166  * WRITE_LIFE_LONG       "                        "
3167  *
3168  * -- direct io
3169  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3170  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3171  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3172  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
3173  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
3174  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
3175  */
3176 
3177 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
3178 				enum page_type type, enum temp_type temp)
3179 {
3180 	if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
3181 		if (type == DATA) {
3182 			if (temp == WARM)
3183 				return WRITE_LIFE_NOT_SET;
3184 			else if (temp == HOT)
3185 				return WRITE_LIFE_SHORT;
3186 			else if (temp == COLD)
3187 				return WRITE_LIFE_EXTREME;
3188 		} else {
3189 			return WRITE_LIFE_NOT_SET;
3190 		}
3191 	} else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
3192 		if (type == DATA) {
3193 			if (temp == WARM)
3194 				return WRITE_LIFE_LONG;
3195 			else if (temp == HOT)
3196 				return WRITE_LIFE_SHORT;
3197 			else if (temp == COLD)
3198 				return WRITE_LIFE_EXTREME;
3199 		} else if (type == NODE) {
3200 			if (temp == WARM || temp == HOT)
3201 				return WRITE_LIFE_NOT_SET;
3202 			else if (temp == COLD)
3203 				return WRITE_LIFE_NONE;
3204 		} else if (type == META) {
3205 			return WRITE_LIFE_MEDIUM;
3206 		}
3207 	}
3208 	return WRITE_LIFE_NOT_SET;
3209 }
3210 
3211 static int __get_segment_type_2(struct f2fs_io_info *fio)
3212 {
3213 	if (fio->type == DATA)
3214 		return CURSEG_HOT_DATA;
3215 	else
3216 		return CURSEG_HOT_NODE;
3217 }
3218 
3219 static int __get_segment_type_4(struct f2fs_io_info *fio)
3220 {
3221 	if (fio->type == DATA) {
3222 		struct inode *inode = fio->page->mapping->host;
3223 
3224 		if (S_ISDIR(inode->i_mode))
3225 			return CURSEG_HOT_DATA;
3226 		else
3227 			return CURSEG_COLD_DATA;
3228 	} else {
3229 		if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3230 			return CURSEG_WARM_NODE;
3231 		else
3232 			return CURSEG_COLD_NODE;
3233 	}
3234 }
3235 
3236 static int __get_segment_type_6(struct f2fs_io_info *fio)
3237 {
3238 	if (fio->type == DATA) {
3239 		struct inode *inode = fio->page->mapping->host;
3240 
3241 		if (is_cold_data(fio->page)) {
3242 			if (fio->sbi->am.atgc_enabled)
3243 				return CURSEG_ALL_DATA_ATGC;
3244 			else
3245 				return CURSEG_COLD_DATA;
3246 		}
3247 		if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3248 			return CURSEG_COLD_DATA;
3249 		if (file_is_hot(inode) ||
3250 				is_inode_flag_set(inode, FI_HOT_DATA) ||
3251 				f2fs_is_atomic_file(inode) ||
3252 				f2fs_is_volatile_file(inode))
3253 			return CURSEG_HOT_DATA;
3254 		return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3255 	} else {
3256 		if (IS_DNODE(fio->page))
3257 			return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3258 						CURSEG_HOT_NODE;
3259 		return CURSEG_COLD_NODE;
3260 	}
3261 }
3262 
3263 static int __get_segment_type(struct f2fs_io_info *fio)
3264 {
3265 	int type = 0;
3266 
3267 	switch (F2FS_OPTION(fio->sbi).active_logs) {
3268 	case 2:
3269 		type = __get_segment_type_2(fio);
3270 		break;
3271 	case 4:
3272 		type = __get_segment_type_4(fio);
3273 		break;
3274 	case 6:
3275 		type = __get_segment_type_6(fio);
3276 		break;
3277 	default:
3278 		f2fs_bug_on(fio->sbi, true);
3279 	}
3280 
3281 	if (IS_HOT(type))
3282 		fio->temp = HOT;
3283 	else if (IS_WARM(type))
3284 		fio->temp = WARM;
3285 	else
3286 		fio->temp = COLD;
3287 	return type;
3288 }
3289 
3290 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3291 		block_t old_blkaddr, block_t *new_blkaddr,
3292 		struct f2fs_summary *sum, int type,
3293 		struct f2fs_io_info *fio)
3294 {
3295 	struct sit_info *sit_i = SIT_I(sbi);
3296 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3297 	unsigned long long old_mtime;
3298 	bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3299 	struct seg_entry *se = NULL;
3300 
3301 	down_read(&SM_I(sbi)->curseg_lock);
3302 
3303 	mutex_lock(&curseg->curseg_mutex);
3304 	down_write(&sit_i->sentry_lock);
3305 
3306 	if (from_gc) {
3307 		f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3308 		se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3309 		sanity_check_seg_type(sbi, se->type);
3310 		f2fs_bug_on(sbi, IS_NODESEG(se->type));
3311 	}
3312 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3313 
3314 	f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg);
3315 
3316 	f2fs_wait_discard_bio(sbi, *new_blkaddr);
3317 
3318 	/*
3319 	 * __add_sum_entry should be resided under the curseg_mutex
3320 	 * because, this function updates a summary entry in the
3321 	 * current summary block.
3322 	 */
3323 	__add_sum_entry(sbi, type, sum);
3324 
3325 	__refresh_next_blkoff(sbi, curseg);
3326 
3327 	stat_inc_block_count(sbi, curseg);
3328 
3329 	if (from_gc) {
3330 		old_mtime = get_segment_mtime(sbi, old_blkaddr);
3331 	} else {
3332 		update_segment_mtime(sbi, old_blkaddr, 0);
3333 		old_mtime = 0;
3334 	}
3335 	update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3336 
3337 	/*
3338 	 * SIT information should be updated before segment allocation,
3339 	 * since SSR needs latest valid block information.
3340 	 */
3341 	update_sit_entry(sbi, *new_blkaddr, 1);
3342 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3343 		update_sit_entry(sbi, old_blkaddr, -1);
3344 
3345 	if (!__has_curseg_space(sbi, curseg)) {
3346 		if (from_gc)
3347 			get_atssr_segment(sbi, type, se->type,
3348 						AT_SSR, se->mtime);
3349 		else
3350 			sit_i->s_ops->allocate_segment(sbi, type, false);
3351 	}
3352 	/*
3353 	 * segment dirty status should be updated after segment allocation,
3354 	 * so we just need to update status only one time after previous
3355 	 * segment being closed.
3356 	 */
3357 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3358 	locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3359 
3360 	up_write(&sit_i->sentry_lock);
3361 
3362 	if (page && IS_NODESEG(type)) {
3363 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3364 
3365 		f2fs_inode_chksum_set(sbi, page);
3366 	}
3367 
3368 	if (F2FS_IO_ALIGNED(sbi))
3369 		fio->retry = false;
3370 
3371 	if (fio) {
3372 		struct f2fs_bio_info *io;
3373 
3374 		INIT_LIST_HEAD(&fio->list);
3375 		fio->in_list = true;
3376 		io = sbi->write_io[fio->type] + fio->temp;
3377 		spin_lock(&io->io_lock);
3378 		list_add_tail(&fio->list, &io->io_list);
3379 		spin_unlock(&io->io_lock);
3380 	}
3381 
3382 	mutex_unlock(&curseg->curseg_mutex);
3383 
3384 	up_read(&SM_I(sbi)->curseg_lock);
3385 }
3386 
3387 static void update_device_state(struct f2fs_io_info *fio)
3388 {
3389 	struct f2fs_sb_info *sbi = fio->sbi;
3390 	unsigned int devidx;
3391 
3392 	if (!f2fs_is_multi_device(sbi))
3393 		return;
3394 
3395 	devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3396 
3397 	/* update device state for fsync */
3398 	f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3399 
3400 	/* update device state for checkpoint */
3401 	if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3402 		spin_lock(&sbi->dev_lock);
3403 		f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3404 		spin_unlock(&sbi->dev_lock);
3405 	}
3406 }
3407 
3408 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3409 {
3410 	int type = __get_segment_type(fio);
3411 	bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3412 
3413 	if (keep_order)
3414 		down_read(&fio->sbi->io_order_lock);
3415 reallocate:
3416 	f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3417 			&fio->new_blkaddr, sum, type, fio);
3418 	if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3419 		invalidate_mapping_pages(META_MAPPING(fio->sbi),
3420 					fio->old_blkaddr, fio->old_blkaddr);
3421 
3422 	/* writeout dirty page into bdev */
3423 	f2fs_submit_page_write(fio);
3424 	if (fio->retry) {
3425 		fio->old_blkaddr = fio->new_blkaddr;
3426 		goto reallocate;
3427 	}
3428 
3429 	update_device_state(fio);
3430 
3431 	if (keep_order)
3432 		up_read(&fio->sbi->io_order_lock);
3433 }
3434 
3435 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3436 					enum iostat_type io_type)
3437 {
3438 	struct f2fs_io_info fio = {
3439 		.sbi = sbi,
3440 		.type = META,
3441 		.temp = HOT,
3442 		.op = REQ_OP_WRITE,
3443 		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3444 		.old_blkaddr = page->index,
3445 		.new_blkaddr = page->index,
3446 		.page = page,
3447 		.encrypted_page = NULL,
3448 		.in_list = false,
3449 	};
3450 
3451 	if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3452 		fio.op_flags &= ~REQ_META;
3453 
3454 	set_page_writeback(page);
3455 	ClearPageError(page);
3456 	f2fs_submit_page_write(&fio);
3457 
3458 	stat_inc_meta_count(sbi, page->index);
3459 	f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3460 }
3461 
3462 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3463 {
3464 	struct f2fs_summary sum;
3465 
3466 	set_summary(&sum, nid, 0, 0);
3467 	do_write_page(&sum, fio);
3468 
3469 	f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3470 }
3471 
3472 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3473 					struct f2fs_io_info *fio)
3474 {
3475 	struct f2fs_sb_info *sbi = fio->sbi;
3476 	struct f2fs_summary sum;
3477 
3478 	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3479 	set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3480 	do_write_page(&sum, fio);
3481 	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3482 
3483 	f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3484 }
3485 
3486 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3487 {
3488 	int err;
3489 	struct f2fs_sb_info *sbi = fio->sbi;
3490 	unsigned int segno;
3491 
3492 	fio->new_blkaddr = fio->old_blkaddr;
3493 	/* i/o temperature is needed for passing down write hints */
3494 	__get_segment_type(fio);
3495 
3496 	segno = GET_SEGNO(sbi, fio->new_blkaddr);
3497 
3498 	if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3499 		set_sbi_flag(sbi, SBI_NEED_FSCK);
3500 		f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3501 			  __func__, segno);
3502 		return -EFSCORRUPTED;
3503 	}
3504 
3505 	stat_inc_inplace_blocks(fio->sbi);
3506 
3507 	if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE)))
3508 		err = f2fs_merge_page_bio(fio);
3509 	else
3510 		err = f2fs_submit_page_bio(fio);
3511 	if (!err) {
3512 		update_device_state(fio);
3513 		f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3514 	}
3515 
3516 	return err;
3517 }
3518 
3519 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3520 						unsigned int segno)
3521 {
3522 	int i;
3523 
3524 	for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3525 		if (CURSEG_I(sbi, i)->segno == segno)
3526 			break;
3527 	}
3528 	return i;
3529 }
3530 
3531 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3532 				block_t old_blkaddr, block_t new_blkaddr,
3533 				bool recover_curseg, bool recover_newaddr,
3534 				bool from_gc)
3535 {
3536 	struct sit_info *sit_i = SIT_I(sbi);
3537 	struct curseg_info *curseg;
3538 	unsigned int segno, old_cursegno;
3539 	struct seg_entry *se;
3540 	int type;
3541 	unsigned short old_blkoff;
3542 
3543 	segno = GET_SEGNO(sbi, new_blkaddr);
3544 	se = get_seg_entry(sbi, segno);
3545 	type = se->type;
3546 
3547 	down_write(&SM_I(sbi)->curseg_lock);
3548 
3549 	if (!recover_curseg) {
3550 		/* for recovery flow */
3551 		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3552 			if (old_blkaddr == NULL_ADDR)
3553 				type = CURSEG_COLD_DATA;
3554 			else
3555 				type = CURSEG_WARM_DATA;
3556 		}
3557 	} else {
3558 		if (IS_CURSEG(sbi, segno)) {
3559 			/* se->type is volatile as SSR allocation */
3560 			type = __f2fs_get_curseg(sbi, segno);
3561 			f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3562 		} else {
3563 			type = CURSEG_WARM_DATA;
3564 		}
3565 	}
3566 
3567 	f2fs_bug_on(sbi, !IS_DATASEG(type));
3568 	curseg = CURSEG_I(sbi, type);
3569 
3570 	mutex_lock(&curseg->curseg_mutex);
3571 	down_write(&sit_i->sentry_lock);
3572 
3573 	old_cursegno = curseg->segno;
3574 	old_blkoff = curseg->next_blkoff;
3575 
3576 	/* change the current segment */
3577 	if (segno != curseg->segno) {
3578 		curseg->next_segno = segno;
3579 		change_curseg(sbi, type, true);
3580 	}
3581 
3582 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3583 	__add_sum_entry(sbi, type, sum);
3584 
3585 	if (!recover_curseg || recover_newaddr) {
3586 		if (!from_gc)
3587 			update_segment_mtime(sbi, new_blkaddr, 0);
3588 		update_sit_entry(sbi, new_blkaddr, 1);
3589 	}
3590 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3591 		invalidate_mapping_pages(META_MAPPING(sbi),
3592 					old_blkaddr, old_blkaddr);
3593 		if (!from_gc)
3594 			update_segment_mtime(sbi, old_blkaddr, 0);
3595 		update_sit_entry(sbi, old_blkaddr, -1);
3596 	}
3597 
3598 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3599 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3600 
3601 	locate_dirty_segment(sbi, old_cursegno);
3602 
3603 	if (recover_curseg) {
3604 		if (old_cursegno != curseg->segno) {
3605 			curseg->next_segno = old_cursegno;
3606 			change_curseg(sbi, type, true);
3607 		}
3608 		curseg->next_blkoff = old_blkoff;
3609 	}
3610 
3611 	up_write(&sit_i->sentry_lock);
3612 	mutex_unlock(&curseg->curseg_mutex);
3613 	up_write(&SM_I(sbi)->curseg_lock);
3614 }
3615 
3616 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3617 				block_t old_addr, block_t new_addr,
3618 				unsigned char version, bool recover_curseg,
3619 				bool recover_newaddr)
3620 {
3621 	struct f2fs_summary sum;
3622 
3623 	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3624 
3625 	f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3626 					recover_curseg, recover_newaddr, false);
3627 
3628 	f2fs_update_data_blkaddr(dn, new_addr);
3629 }
3630 
3631 void f2fs_wait_on_page_writeback(struct page *page,
3632 				enum page_type type, bool ordered, bool locked)
3633 {
3634 	if (PageWriteback(page)) {
3635 		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3636 
3637 		/* submit cached LFS IO */
3638 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3639 		/* sbumit cached IPU IO */
3640 		f2fs_submit_merged_ipu_write(sbi, NULL, page);
3641 		if (ordered) {
3642 			wait_on_page_writeback(page);
3643 			f2fs_bug_on(sbi, locked && PageWriteback(page));
3644 		} else {
3645 			wait_for_stable_page(page);
3646 		}
3647 	}
3648 }
3649 
3650 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3651 {
3652 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3653 	struct page *cpage;
3654 
3655 	if (!f2fs_post_read_required(inode))
3656 		return;
3657 
3658 	if (!__is_valid_data_blkaddr(blkaddr))
3659 		return;
3660 
3661 	cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3662 	if (cpage) {
3663 		f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3664 		f2fs_put_page(cpage, 1);
3665 	}
3666 }
3667 
3668 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3669 								block_t len)
3670 {
3671 	block_t i;
3672 
3673 	for (i = 0; i < len; i++)
3674 		f2fs_wait_on_block_writeback(inode, blkaddr + i);
3675 }
3676 
3677 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3678 {
3679 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3680 	struct curseg_info *seg_i;
3681 	unsigned char *kaddr;
3682 	struct page *page;
3683 	block_t start;
3684 	int i, j, offset;
3685 
3686 	start = start_sum_block(sbi);
3687 
3688 	page = f2fs_get_meta_page(sbi, start++);
3689 	if (IS_ERR(page))
3690 		return PTR_ERR(page);
3691 	kaddr = (unsigned char *)page_address(page);
3692 
3693 	/* Step 1: restore nat cache */
3694 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3695 	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3696 
3697 	/* Step 2: restore sit cache */
3698 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3699 	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3700 	offset = 2 * SUM_JOURNAL_SIZE;
3701 
3702 	/* Step 3: restore summary entries */
3703 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3704 		unsigned short blk_off;
3705 		unsigned int segno;
3706 
3707 		seg_i = CURSEG_I(sbi, i);
3708 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3709 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3710 		seg_i->next_segno = segno;
3711 		reset_curseg(sbi, i, 0);
3712 		seg_i->alloc_type = ckpt->alloc_type[i];
3713 		seg_i->next_blkoff = blk_off;
3714 
3715 		if (seg_i->alloc_type == SSR)
3716 			blk_off = sbi->blocks_per_seg;
3717 
3718 		for (j = 0; j < blk_off; j++) {
3719 			struct f2fs_summary *s;
3720 			s = (struct f2fs_summary *)(kaddr + offset);
3721 			seg_i->sum_blk->entries[j] = *s;
3722 			offset += SUMMARY_SIZE;
3723 			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3724 						SUM_FOOTER_SIZE)
3725 				continue;
3726 
3727 			f2fs_put_page(page, 1);
3728 			page = NULL;
3729 
3730 			page = f2fs_get_meta_page(sbi, start++);
3731 			if (IS_ERR(page))
3732 				return PTR_ERR(page);
3733 			kaddr = (unsigned char *)page_address(page);
3734 			offset = 0;
3735 		}
3736 	}
3737 	f2fs_put_page(page, 1);
3738 	return 0;
3739 }
3740 
3741 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3742 {
3743 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3744 	struct f2fs_summary_block *sum;
3745 	struct curseg_info *curseg;
3746 	struct page *new;
3747 	unsigned short blk_off;
3748 	unsigned int segno = 0;
3749 	block_t blk_addr = 0;
3750 	int err = 0;
3751 
3752 	/* get segment number and block addr */
3753 	if (IS_DATASEG(type)) {
3754 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3755 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3756 							CURSEG_HOT_DATA]);
3757 		if (__exist_node_summaries(sbi))
3758 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
3759 		else
3760 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3761 	} else {
3762 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
3763 							CURSEG_HOT_NODE]);
3764 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3765 							CURSEG_HOT_NODE]);
3766 		if (__exist_node_summaries(sbi))
3767 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3768 							type - CURSEG_HOT_NODE);
3769 		else
3770 			blk_addr = GET_SUM_BLOCK(sbi, segno);
3771 	}
3772 
3773 	new = f2fs_get_meta_page(sbi, blk_addr);
3774 	if (IS_ERR(new))
3775 		return PTR_ERR(new);
3776 	sum = (struct f2fs_summary_block *)page_address(new);
3777 
3778 	if (IS_NODESEG(type)) {
3779 		if (__exist_node_summaries(sbi)) {
3780 			struct f2fs_summary *ns = &sum->entries[0];
3781 			int i;
3782 			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3783 				ns->version = 0;
3784 				ns->ofs_in_node = 0;
3785 			}
3786 		} else {
3787 			err = f2fs_restore_node_summary(sbi, segno, sum);
3788 			if (err)
3789 				goto out;
3790 		}
3791 	}
3792 
3793 	/* set uncompleted segment to curseg */
3794 	curseg = CURSEG_I(sbi, type);
3795 	mutex_lock(&curseg->curseg_mutex);
3796 
3797 	/* update journal info */
3798 	down_write(&curseg->journal_rwsem);
3799 	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3800 	up_write(&curseg->journal_rwsem);
3801 
3802 	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3803 	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3804 	curseg->next_segno = segno;
3805 	reset_curseg(sbi, type, 0);
3806 	curseg->alloc_type = ckpt->alloc_type[type];
3807 	curseg->next_blkoff = blk_off;
3808 	mutex_unlock(&curseg->curseg_mutex);
3809 out:
3810 	f2fs_put_page(new, 1);
3811 	return err;
3812 }
3813 
3814 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3815 {
3816 	struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3817 	struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3818 	int type = CURSEG_HOT_DATA;
3819 	int err;
3820 
3821 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3822 		int npages = f2fs_npages_for_summary_flush(sbi, true);
3823 
3824 		if (npages >= 2)
3825 			f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3826 							META_CP, true);
3827 
3828 		/* restore for compacted data summary */
3829 		err = read_compacted_summaries(sbi);
3830 		if (err)
3831 			return err;
3832 		type = CURSEG_HOT_NODE;
3833 	}
3834 
3835 	if (__exist_node_summaries(sbi))
3836 		f2fs_ra_meta_pages(sbi,
3837 				sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
3838 				NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
3839 
3840 	for (; type <= CURSEG_COLD_NODE; type++) {
3841 		err = read_normal_summaries(sbi, type);
3842 		if (err)
3843 			return err;
3844 	}
3845 
3846 	/* sanity check for summary blocks */
3847 	if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3848 			sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3849 		f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3850 			 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3851 		return -EINVAL;
3852 	}
3853 
3854 	return 0;
3855 }
3856 
3857 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3858 {
3859 	struct page *page;
3860 	unsigned char *kaddr;
3861 	struct f2fs_summary *summary;
3862 	struct curseg_info *seg_i;
3863 	int written_size = 0;
3864 	int i, j;
3865 
3866 	page = f2fs_grab_meta_page(sbi, blkaddr++);
3867 	kaddr = (unsigned char *)page_address(page);
3868 	memset(kaddr, 0, PAGE_SIZE);
3869 
3870 	/* Step 1: write nat cache */
3871 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3872 	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3873 	written_size += SUM_JOURNAL_SIZE;
3874 
3875 	/* Step 2: write sit cache */
3876 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3877 	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3878 	written_size += SUM_JOURNAL_SIZE;
3879 
3880 	/* Step 3: write summary entries */
3881 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3882 		unsigned short blkoff;
3883 		seg_i = CURSEG_I(sbi, i);
3884 		if (sbi->ckpt->alloc_type[i] == SSR)
3885 			blkoff = sbi->blocks_per_seg;
3886 		else
3887 			blkoff = curseg_blkoff(sbi, i);
3888 
3889 		for (j = 0; j < blkoff; j++) {
3890 			if (!page) {
3891 				page = f2fs_grab_meta_page(sbi, blkaddr++);
3892 				kaddr = (unsigned char *)page_address(page);
3893 				memset(kaddr, 0, PAGE_SIZE);
3894 				written_size = 0;
3895 			}
3896 			summary = (struct f2fs_summary *)(kaddr + written_size);
3897 			*summary = seg_i->sum_blk->entries[j];
3898 			written_size += SUMMARY_SIZE;
3899 
3900 			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3901 							SUM_FOOTER_SIZE)
3902 				continue;
3903 
3904 			set_page_dirty(page);
3905 			f2fs_put_page(page, 1);
3906 			page = NULL;
3907 		}
3908 	}
3909 	if (page) {
3910 		set_page_dirty(page);
3911 		f2fs_put_page(page, 1);
3912 	}
3913 }
3914 
3915 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3916 					block_t blkaddr, int type)
3917 {
3918 	int i, end;
3919 	if (IS_DATASEG(type))
3920 		end = type + NR_CURSEG_DATA_TYPE;
3921 	else
3922 		end = type + NR_CURSEG_NODE_TYPE;
3923 
3924 	for (i = type; i < end; i++)
3925 		write_current_sum_page(sbi, i, blkaddr + (i - type));
3926 }
3927 
3928 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3929 {
3930 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3931 		write_compacted_summaries(sbi, start_blk);
3932 	else
3933 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3934 }
3935 
3936 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3937 {
3938 	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3939 }
3940 
3941 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3942 					unsigned int val, int alloc)
3943 {
3944 	int i;
3945 
3946 	if (type == NAT_JOURNAL) {
3947 		for (i = 0; i < nats_in_cursum(journal); i++) {
3948 			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3949 				return i;
3950 		}
3951 		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3952 			return update_nats_in_cursum(journal, 1);
3953 	} else if (type == SIT_JOURNAL) {
3954 		for (i = 0; i < sits_in_cursum(journal); i++)
3955 			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3956 				return i;
3957 		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3958 			return update_sits_in_cursum(journal, 1);
3959 	}
3960 	return -1;
3961 }
3962 
3963 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3964 					unsigned int segno)
3965 {
3966 	return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
3967 }
3968 
3969 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3970 					unsigned int start)
3971 {
3972 	struct sit_info *sit_i = SIT_I(sbi);
3973 	struct page *page;
3974 	pgoff_t src_off, dst_off;
3975 
3976 	src_off = current_sit_addr(sbi, start);
3977 	dst_off = next_sit_addr(sbi, src_off);
3978 
3979 	page = f2fs_grab_meta_page(sbi, dst_off);
3980 	seg_info_to_sit_page(sbi, page, start);
3981 
3982 	set_page_dirty(page);
3983 	set_to_next_sit(sit_i, start);
3984 
3985 	return page;
3986 }
3987 
3988 static struct sit_entry_set *grab_sit_entry_set(void)
3989 {
3990 	struct sit_entry_set *ses =
3991 			f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3992 
3993 	ses->entry_cnt = 0;
3994 	INIT_LIST_HEAD(&ses->set_list);
3995 	return ses;
3996 }
3997 
3998 static void release_sit_entry_set(struct sit_entry_set *ses)
3999 {
4000 	list_del(&ses->set_list);
4001 	kmem_cache_free(sit_entry_set_slab, ses);
4002 }
4003 
4004 static void adjust_sit_entry_set(struct sit_entry_set *ses,
4005 						struct list_head *head)
4006 {
4007 	struct sit_entry_set *next = ses;
4008 
4009 	if (list_is_last(&ses->set_list, head))
4010 		return;
4011 
4012 	list_for_each_entry_continue(next, head, set_list)
4013 		if (ses->entry_cnt <= next->entry_cnt)
4014 			break;
4015 
4016 	list_move_tail(&ses->set_list, &next->set_list);
4017 }
4018 
4019 static void add_sit_entry(unsigned int segno, struct list_head *head)
4020 {
4021 	struct sit_entry_set *ses;
4022 	unsigned int start_segno = START_SEGNO(segno);
4023 
4024 	list_for_each_entry(ses, head, set_list) {
4025 		if (ses->start_segno == start_segno) {
4026 			ses->entry_cnt++;
4027 			adjust_sit_entry_set(ses, head);
4028 			return;
4029 		}
4030 	}
4031 
4032 	ses = grab_sit_entry_set();
4033 
4034 	ses->start_segno = start_segno;
4035 	ses->entry_cnt++;
4036 	list_add(&ses->set_list, head);
4037 }
4038 
4039 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4040 {
4041 	struct f2fs_sm_info *sm_info = SM_I(sbi);
4042 	struct list_head *set_list = &sm_info->sit_entry_set;
4043 	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4044 	unsigned int segno;
4045 
4046 	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4047 		add_sit_entry(segno, set_list);
4048 }
4049 
4050 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4051 {
4052 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4053 	struct f2fs_journal *journal = curseg->journal;
4054 	int i;
4055 
4056 	down_write(&curseg->journal_rwsem);
4057 	for (i = 0; i < sits_in_cursum(journal); i++) {
4058 		unsigned int segno;
4059 		bool dirtied;
4060 
4061 		segno = le32_to_cpu(segno_in_journal(journal, i));
4062 		dirtied = __mark_sit_entry_dirty(sbi, segno);
4063 
4064 		if (!dirtied)
4065 			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4066 	}
4067 	update_sits_in_cursum(journal, -i);
4068 	up_write(&curseg->journal_rwsem);
4069 }
4070 
4071 /*
4072  * CP calls this function, which flushes SIT entries including sit_journal,
4073  * and moves prefree segs to free segs.
4074  */
4075 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4076 {
4077 	struct sit_info *sit_i = SIT_I(sbi);
4078 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4079 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4080 	struct f2fs_journal *journal = curseg->journal;
4081 	struct sit_entry_set *ses, *tmp;
4082 	struct list_head *head = &SM_I(sbi)->sit_entry_set;
4083 	bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4084 	struct seg_entry *se;
4085 
4086 	down_write(&sit_i->sentry_lock);
4087 
4088 	if (!sit_i->dirty_sentries)
4089 		goto out;
4090 
4091 	/*
4092 	 * add and account sit entries of dirty bitmap in sit entry
4093 	 * set temporarily
4094 	 */
4095 	add_sits_in_set(sbi);
4096 
4097 	/*
4098 	 * if there are no enough space in journal to store dirty sit
4099 	 * entries, remove all entries from journal and add and account
4100 	 * them in sit entry set.
4101 	 */
4102 	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4103 								!to_journal)
4104 		remove_sits_in_journal(sbi);
4105 
4106 	/*
4107 	 * there are two steps to flush sit entries:
4108 	 * #1, flush sit entries to journal in current cold data summary block.
4109 	 * #2, flush sit entries to sit page.
4110 	 */
4111 	list_for_each_entry_safe(ses, tmp, head, set_list) {
4112 		struct page *page = NULL;
4113 		struct f2fs_sit_block *raw_sit = NULL;
4114 		unsigned int start_segno = ses->start_segno;
4115 		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4116 						(unsigned long)MAIN_SEGS(sbi));
4117 		unsigned int segno = start_segno;
4118 
4119 		if (to_journal &&
4120 			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4121 			to_journal = false;
4122 
4123 		if (to_journal) {
4124 			down_write(&curseg->journal_rwsem);
4125 		} else {
4126 			page = get_next_sit_page(sbi, start_segno);
4127 			raw_sit = page_address(page);
4128 		}
4129 
4130 		/* flush dirty sit entries in region of current sit set */
4131 		for_each_set_bit_from(segno, bitmap, end) {
4132 			int offset, sit_offset;
4133 
4134 			se = get_seg_entry(sbi, segno);
4135 #ifdef CONFIG_F2FS_CHECK_FS
4136 			if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4137 						SIT_VBLOCK_MAP_SIZE))
4138 				f2fs_bug_on(sbi, 1);
4139 #endif
4140 
4141 			/* add discard candidates */
4142 			if (!(cpc->reason & CP_DISCARD)) {
4143 				cpc->trim_start = segno;
4144 				add_discard_addrs(sbi, cpc, false);
4145 			}
4146 
4147 			if (to_journal) {
4148 				offset = f2fs_lookup_journal_in_cursum(journal,
4149 							SIT_JOURNAL, segno, 1);
4150 				f2fs_bug_on(sbi, offset < 0);
4151 				segno_in_journal(journal, offset) =
4152 							cpu_to_le32(segno);
4153 				seg_info_to_raw_sit(se,
4154 					&sit_in_journal(journal, offset));
4155 				check_block_count(sbi, segno,
4156 					&sit_in_journal(journal, offset));
4157 			} else {
4158 				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4159 				seg_info_to_raw_sit(se,
4160 						&raw_sit->entries[sit_offset]);
4161 				check_block_count(sbi, segno,
4162 						&raw_sit->entries[sit_offset]);
4163 			}
4164 
4165 			__clear_bit(segno, bitmap);
4166 			sit_i->dirty_sentries--;
4167 			ses->entry_cnt--;
4168 		}
4169 
4170 		if (to_journal)
4171 			up_write(&curseg->journal_rwsem);
4172 		else
4173 			f2fs_put_page(page, 1);
4174 
4175 		f2fs_bug_on(sbi, ses->entry_cnt);
4176 		release_sit_entry_set(ses);
4177 	}
4178 
4179 	f2fs_bug_on(sbi, !list_empty(head));
4180 	f2fs_bug_on(sbi, sit_i->dirty_sentries);
4181 out:
4182 	if (cpc->reason & CP_DISCARD) {
4183 		__u64 trim_start = cpc->trim_start;
4184 
4185 		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4186 			add_discard_addrs(sbi, cpc, false);
4187 
4188 		cpc->trim_start = trim_start;
4189 	}
4190 	up_write(&sit_i->sentry_lock);
4191 
4192 	set_prefree_as_free_segments(sbi);
4193 }
4194 
4195 static int build_sit_info(struct f2fs_sb_info *sbi)
4196 {
4197 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4198 	struct sit_info *sit_i;
4199 	unsigned int sit_segs, start;
4200 	char *src_bitmap, *bitmap;
4201 	unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4202 
4203 	/* allocate memory for SIT information */
4204 	sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4205 	if (!sit_i)
4206 		return -ENOMEM;
4207 
4208 	SM_I(sbi)->sit_info = sit_i;
4209 
4210 	sit_i->sentries =
4211 		f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4212 					      MAIN_SEGS(sbi)),
4213 			      GFP_KERNEL);
4214 	if (!sit_i->sentries)
4215 		return -ENOMEM;
4216 
4217 	main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4218 	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4219 								GFP_KERNEL);
4220 	if (!sit_i->dirty_sentries_bitmap)
4221 		return -ENOMEM;
4222 
4223 #ifdef CONFIG_F2FS_CHECK_FS
4224 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 4;
4225 #else
4226 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 3;
4227 #endif
4228 	sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4229 	if (!sit_i->bitmap)
4230 		return -ENOMEM;
4231 
4232 	bitmap = sit_i->bitmap;
4233 
4234 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4235 		sit_i->sentries[start].cur_valid_map = bitmap;
4236 		bitmap += SIT_VBLOCK_MAP_SIZE;
4237 
4238 		sit_i->sentries[start].ckpt_valid_map = bitmap;
4239 		bitmap += SIT_VBLOCK_MAP_SIZE;
4240 
4241 #ifdef CONFIG_F2FS_CHECK_FS
4242 		sit_i->sentries[start].cur_valid_map_mir = bitmap;
4243 		bitmap += SIT_VBLOCK_MAP_SIZE;
4244 #endif
4245 
4246 		sit_i->sentries[start].discard_map = bitmap;
4247 		bitmap += SIT_VBLOCK_MAP_SIZE;
4248 	}
4249 
4250 	sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4251 	if (!sit_i->tmp_map)
4252 		return -ENOMEM;
4253 
4254 	if (__is_large_section(sbi)) {
4255 		sit_i->sec_entries =
4256 			f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4257 						      MAIN_SECS(sbi)),
4258 				      GFP_KERNEL);
4259 		if (!sit_i->sec_entries)
4260 			return -ENOMEM;
4261 	}
4262 
4263 	/* get information related with SIT */
4264 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4265 
4266 	/* setup SIT bitmap from ckeckpoint pack */
4267 	sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4268 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4269 
4270 	sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4271 	if (!sit_i->sit_bitmap)
4272 		return -ENOMEM;
4273 
4274 #ifdef CONFIG_F2FS_CHECK_FS
4275 	sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4276 					sit_bitmap_size, GFP_KERNEL);
4277 	if (!sit_i->sit_bitmap_mir)
4278 		return -ENOMEM;
4279 
4280 	sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4281 					main_bitmap_size, GFP_KERNEL);
4282 	if (!sit_i->invalid_segmap)
4283 		return -ENOMEM;
4284 #endif
4285 
4286 	/* init SIT information */
4287 	sit_i->s_ops = &default_salloc_ops;
4288 
4289 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4290 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4291 	sit_i->written_valid_blocks = 0;
4292 	sit_i->bitmap_size = sit_bitmap_size;
4293 	sit_i->dirty_sentries = 0;
4294 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4295 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4296 	sit_i->mounted_time = ktime_get_boottime_seconds();
4297 	init_rwsem(&sit_i->sentry_lock);
4298 	return 0;
4299 }
4300 
4301 static int build_free_segmap(struct f2fs_sb_info *sbi)
4302 {
4303 	struct free_segmap_info *free_i;
4304 	unsigned int bitmap_size, sec_bitmap_size;
4305 
4306 	/* allocate memory for free segmap information */
4307 	free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4308 	if (!free_i)
4309 		return -ENOMEM;
4310 
4311 	SM_I(sbi)->free_info = free_i;
4312 
4313 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4314 	free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4315 	if (!free_i->free_segmap)
4316 		return -ENOMEM;
4317 
4318 	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4319 	free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4320 	if (!free_i->free_secmap)
4321 		return -ENOMEM;
4322 
4323 	/* set all segments as dirty temporarily */
4324 	memset(free_i->free_segmap, 0xff, bitmap_size);
4325 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4326 
4327 	/* init free segmap information */
4328 	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4329 	free_i->free_segments = 0;
4330 	free_i->free_sections = 0;
4331 	spin_lock_init(&free_i->segmap_lock);
4332 	return 0;
4333 }
4334 
4335 static int build_curseg(struct f2fs_sb_info *sbi)
4336 {
4337 	struct curseg_info *array;
4338 	int i;
4339 
4340 	array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4341 					sizeof(*array)), GFP_KERNEL);
4342 	if (!array)
4343 		return -ENOMEM;
4344 
4345 	SM_I(sbi)->curseg_array = array;
4346 
4347 	for (i = 0; i < NO_CHECK_TYPE; i++) {
4348 		mutex_init(&array[i].curseg_mutex);
4349 		array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4350 		if (!array[i].sum_blk)
4351 			return -ENOMEM;
4352 		init_rwsem(&array[i].journal_rwsem);
4353 		array[i].journal = f2fs_kzalloc(sbi,
4354 				sizeof(struct f2fs_journal), GFP_KERNEL);
4355 		if (!array[i].journal)
4356 			return -ENOMEM;
4357 		if (i < NR_PERSISTENT_LOG)
4358 			array[i].seg_type = CURSEG_HOT_DATA + i;
4359 		else if (i == CURSEG_COLD_DATA_PINNED)
4360 			array[i].seg_type = CURSEG_COLD_DATA;
4361 		else if (i == CURSEG_ALL_DATA_ATGC)
4362 			array[i].seg_type = CURSEG_COLD_DATA;
4363 		array[i].segno = NULL_SEGNO;
4364 		array[i].next_blkoff = 0;
4365 		array[i].inited = false;
4366 	}
4367 	return restore_curseg_summaries(sbi);
4368 }
4369 
4370 static int build_sit_entries(struct f2fs_sb_info *sbi)
4371 {
4372 	struct sit_info *sit_i = SIT_I(sbi);
4373 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4374 	struct f2fs_journal *journal = curseg->journal;
4375 	struct seg_entry *se;
4376 	struct f2fs_sit_entry sit;
4377 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
4378 	unsigned int i, start, end;
4379 	unsigned int readed, start_blk = 0;
4380 	int err = 0;
4381 	block_t total_node_blocks = 0;
4382 
4383 	do {
4384 		readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4385 							META_SIT, true);
4386 
4387 		start = start_blk * sit_i->sents_per_block;
4388 		end = (start_blk + readed) * sit_i->sents_per_block;
4389 
4390 		for (; start < end && start < MAIN_SEGS(sbi); start++) {
4391 			struct f2fs_sit_block *sit_blk;
4392 			struct page *page;
4393 
4394 			se = &sit_i->sentries[start];
4395 			page = get_current_sit_page(sbi, start);
4396 			if (IS_ERR(page))
4397 				return PTR_ERR(page);
4398 			sit_blk = (struct f2fs_sit_block *)page_address(page);
4399 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4400 			f2fs_put_page(page, 1);
4401 
4402 			err = check_block_count(sbi, start, &sit);
4403 			if (err)
4404 				return err;
4405 			seg_info_from_raw_sit(se, &sit);
4406 			if (IS_NODESEG(se->type))
4407 				total_node_blocks += se->valid_blocks;
4408 
4409 			/* build discard map only one time */
4410 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4411 				memset(se->discard_map, 0xff,
4412 					SIT_VBLOCK_MAP_SIZE);
4413 			} else {
4414 				memcpy(se->discard_map,
4415 					se->cur_valid_map,
4416 					SIT_VBLOCK_MAP_SIZE);
4417 				sbi->discard_blks +=
4418 					sbi->blocks_per_seg -
4419 					se->valid_blocks;
4420 			}
4421 
4422 			if (__is_large_section(sbi))
4423 				get_sec_entry(sbi, start)->valid_blocks +=
4424 							se->valid_blocks;
4425 		}
4426 		start_blk += readed;
4427 	} while (start_blk < sit_blk_cnt);
4428 
4429 	down_read(&curseg->journal_rwsem);
4430 	for (i = 0; i < sits_in_cursum(journal); i++) {
4431 		unsigned int old_valid_blocks;
4432 
4433 		start = le32_to_cpu(segno_in_journal(journal, i));
4434 		if (start >= MAIN_SEGS(sbi)) {
4435 			f2fs_err(sbi, "Wrong journal entry on segno %u",
4436 				 start);
4437 			err = -EFSCORRUPTED;
4438 			break;
4439 		}
4440 
4441 		se = &sit_i->sentries[start];
4442 		sit = sit_in_journal(journal, i);
4443 
4444 		old_valid_blocks = se->valid_blocks;
4445 		if (IS_NODESEG(se->type))
4446 			total_node_blocks -= old_valid_blocks;
4447 
4448 		err = check_block_count(sbi, start, &sit);
4449 		if (err)
4450 			break;
4451 		seg_info_from_raw_sit(se, &sit);
4452 		if (IS_NODESEG(se->type))
4453 			total_node_blocks += se->valid_blocks;
4454 
4455 		if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4456 			memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4457 		} else {
4458 			memcpy(se->discard_map, se->cur_valid_map,
4459 						SIT_VBLOCK_MAP_SIZE);
4460 			sbi->discard_blks += old_valid_blocks;
4461 			sbi->discard_blks -= se->valid_blocks;
4462 		}
4463 
4464 		if (__is_large_section(sbi)) {
4465 			get_sec_entry(sbi, start)->valid_blocks +=
4466 							se->valid_blocks;
4467 			get_sec_entry(sbi, start)->valid_blocks -=
4468 							old_valid_blocks;
4469 		}
4470 	}
4471 	up_read(&curseg->journal_rwsem);
4472 
4473 	if (!err && total_node_blocks != valid_node_count(sbi)) {
4474 		f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4475 			 total_node_blocks, valid_node_count(sbi));
4476 		err = -EFSCORRUPTED;
4477 	}
4478 
4479 	return err;
4480 }
4481 
4482 static void init_free_segmap(struct f2fs_sb_info *sbi)
4483 {
4484 	unsigned int start;
4485 	int type;
4486 	struct seg_entry *sentry;
4487 
4488 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4489 		if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4490 			continue;
4491 		sentry = get_seg_entry(sbi, start);
4492 		if (!sentry->valid_blocks)
4493 			__set_free(sbi, start);
4494 		else
4495 			SIT_I(sbi)->written_valid_blocks +=
4496 						sentry->valid_blocks;
4497 	}
4498 
4499 	/* set use the current segments */
4500 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4501 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4502 		__set_test_and_inuse(sbi, curseg_t->segno);
4503 	}
4504 }
4505 
4506 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4507 {
4508 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4509 	struct free_segmap_info *free_i = FREE_I(sbi);
4510 	unsigned int segno = 0, offset = 0, secno;
4511 	block_t valid_blocks, usable_blks_in_seg;
4512 	block_t blks_per_sec = BLKS_PER_SEC(sbi);
4513 
4514 	while (1) {
4515 		/* find dirty segment based on free segmap */
4516 		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4517 		if (segno >= MAIN_SEGS(sbi))
4518 			break;
4519 		offset = segno + 1;
4520 		valid_blocks = get_valid_blocks(sbi, segno, false);
4521 		usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4522 		if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4523 			continue;
4524 		if (valid_blocks > usable_blks_in_seg) {
4525 			f2fs_bug_on(sbi, 1);
4526 			continue;
4527 		}
4528 		mutex_lock(&dirty_i->seglist_lock);
4529 		__locate_dirty_segment(sbi, segno, DIRTY);
4530 		mutex_unlock(&dirty_i->seglist_lock);
4531 	}
4532 
4533 	if (!__is_large_section(sbi))
4534 		return;
4535 
4536 	mutex_lock(&dirty_i->seglist_lock);
4537 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4538 		valid_blocks = get_valid_blocks(sbi, segno, true);
4539 		secno = GET_SEC_FROM_SEG(sbi, segno);
4540 
4541 		if (!valid_blocks || valid_blocks == blks_per_sec)
4542 			continue;
4543 		if (IS_CURSEC(sbi, secno))
4544 			continue;
4545 		set_bit(secno, dirty_i->dirty_secmap);
4546 	}
4547 	mutex_unlock(&dirty_i->seglist_lock);
4548 }
4549 
4550 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4551 {
4552 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4553 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4554 
4555 	dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4556 	if (!dirty_i->victim_secmap)
4557 		return -ENOMEM;
4558 	return 0;
4559 }
4560 
4561 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4562 {
4563 	struct dirty_seglist_info *dirty_i;
4564 	unsigned int bitmap_size, i;
4565 
4566 	/* allocate memory for dirty segments list information */
4567 	dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4568 								GFP_KERNEL);
4569 	if (!dirty_i)
4570 		return -ENOMEM;
4571 
4572 	SM_I(sbi)->dirty_info = dirty_i;
4573 	mutex_init(&dirty_i->seglist_lock);
4574 
4575 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4576 
4577 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
4578 		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4579 								GFP_KERNEL);
4580 		if (!dirty_i->dirty_segmap[i])
4581 			return -ENOMEM;
4582 	}
4583 
4584 	if (__is_large_section(sbi)) {
4585 		bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4586 		dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4587 						bitmap_size, GFP_KERNEL);
4588 		if (!dirty_i->dirty_secmap)
4589 			return -ENOMEM;
4590 	}
4591 
4592 	init_dirty_segmap(sbi);
4593 	return init_victim_secmap(sbi);
4594 }
4595 
4596 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4597 {
4598 	int i;
4599 
4600 	/*
4601 	 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4602 	 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4603 	 */
4604 	for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4605 		struct curseg_info *curseg = CURSEG_I(sbi, i);
4606 		struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4607 		unsigned int blkofs = curseg->next_blkoff;
4608 
4609 		sanity_check_seg_type(sbi, curseg->seg_type);
4610 
4611 		if (f2fs_test_bit(blkofs, se->cur_valid_map))
4612 			goto out;
4613 
4614 		if (curseg->alloc_type == SSR)
4615 			continue;
4616 
4617 		for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4618 			if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4619 				continue;
4620 out:
4621 			f2fs_err(sbi,
4622 				 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4623 				 i, curseg->segno, curseg->alloc_type,
4624 				 curseg->next_blkoff, blkofs);
4625 			return -EFSCORRUPTED;
4626 		}
4627 	}
4628 	return 0;
4629 }
4630 
4631 #ifdef CONFIG_BLK_DEV_ZONED
4632 
4633 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4634 				    struct f2fs_dev_info *fdev,
4635 				    struct blk_zone *zone)
4636 {
4637 	unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4638 	block_t zone_block, wp_block, last_valid_block;
4639 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4640 	int i, s, b, ret;
4641 	struct seg_entry *se;
4642 
4643 	if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4644 		return 0;
4645 
4646 	wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4647 	wp_segno = GET_SEGNO(sbi, wp_block);
4648 	wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4649 	zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4650 	zone_segno = GET_SEGNO(sbi, zone_block);
4651 	zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4652 
4653 	if (zone_segno >= MAIN_SEGS(sbi))
4654 		return 0;
4655 
4656 	/*
4657 	 * Skip check of zones cursegs point to, since
4658 	 * fix_curseg_write_pointer() checks them.
4659 	 */
4660 	for (i = 0; i < NO_CHECK_TYPE; i++)
4661 		if (zone_secno == GET_SEC_FROM_SEG(sbi,
4662 						   CURSEG_I(sbi, i)->segno))
4663 			return 0;
4664 
4665 	/*
4666 	 * Get last valid block of the zone.
4667 	 */
4668 	last_valid_block = zone_block - 1;
4669 	for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4670 		segno = zone_segno + s;
4671 		se = get_seg_entry(sbi, segno);
4672 		for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4673 			if (f2fs_test_bit(b, se->cur_valid_map)) {
4674 				last_valid_block = START_BLOCK(sbi, segno) + b;
4675 				break;
4676 			}
4677 		if (last_valid_block >= zone_block)
4678 			break;
4679 	}
4680 
4681 	/*
4682 	 * If last valid block is beyond the write pointer, report the
4683 	 * inconsistency. This inconsistency does not cause write error
4684 	 * because the zone will not be selected for write operation until
4685 	 * it get discarded. Just report it.
4686 	 */
4687 	if (last_valid_block >= wp_block) {
4688 		f2fs_notice(sbi, "Valid block beyond write pointer: "
4689 			    "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4690 			    GET_SEGNO(sbi, last_valid_block),
4691 			    GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4692 			    wp_segno, wp_blkoff);
4693 		return 0;
4694 	}
4695 
4696 	/*
4697 	 * If there is no valid block in the zone and if write pointer is
4698 	 * not at zone start, reset the write pointer.
4699 	 */
4700 	if (last_valid_block + 1 == zone_block && zone->wp != zone->start) {
4701 		f2fs_notice(sbi,
4702 			    "Zone without valid block has non-zero write "
4703 			    "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4704 			    wp_segno, wp_blkoff);
4705 		ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4706 					zone->len >> log_sectors_per_block);
4707 		if (ret) {
4708 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4709 				 fdev->path, ret);
4710 			return ret;
4711 		}
4712 	}
4713 
4714 	return 0;
4715 }
4716 
4717 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4718 						  block_t zone_blkaddr)
4719 {
4720 	int i;
4721 
4722 	for (i = 0; i < sbi->s_ndevs; i++) {
4723 		if (!bdev_is_zoned(FDEV(i).bdev))
4724 			continue;
4725 		if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4726 				zone_blkaddr <= FDEV(i).end_blk))
4727 			return &FDEV(i);
4728 	}
4729 
4730 	return NULL;
4731 }
4732 
4733 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4734 			      void *data) {
4735 	memcpy(data, zone, sizeof(struct blk_zone));
4736 	return 0;
4737 }
4738 
4739 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4740 {
4741 	struct curseg_info *cs = CURSEG_I(sbi, type);
4742 	struct f2fs_dev_info *zbd;
4743 	struct blk_zone zone;
4744 	unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4745 	block_t cs_zone_block, wp_block;
4746 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4747 	sector_t zone_sector;
4748 	int err;
4749 
4750 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4751 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4752 
4753 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
4754 	if (!zbd)
4755 		return 0;
4756 
4757 	/* report zone for the sector the curseg points to */
4758 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4759 		<< log_sectors_per_block;
4760 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4761 				  report_one_zone_cb, &zone);
4762 	if (err != 1) {
4763 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4764 			 zbd->path, err);
4765 		return err;
4766 	}
4767 
4768 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4769 		return 0;
4770 
4771 	wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
4772 	wp_segno = GET_SEGNO(sbi, wp_block);
4773 	wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4774 	wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
4775 
4776 	if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
4777 		wp_sector_off == 0)
4778 		return 0;
4779 
4780 	f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
4781 		    "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
4782 		    type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
4783 
4784 	f2fs_notice(sbi, "Assign new section to curseg[%d]: "
4785 		    "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
4786 	allocate_segment_by_default(sbi, type, true);
4787 
4788 	/* check consistency of the zone curseg pointed to */
4789 	if (check_zone_write_pointer(sbi, zbd, &zone))
4790 		return -EIO;
4791 
4792 	/* check newly assigned zone */
4793 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4794 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4795 
4796 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
4797 	if (!zbd)
4798 		return 0;
4799 
4800 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4801 		<< log_sectors_per_block;
4802 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4803 				  report_one_zone_cb, &zone);
4804 	if (err != 1) {
4805 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4806 			 zbd->path, err);
4807 		return err;
4808 	}
4809 
4810 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4811 		return 0;
4812 
4813 	if (zone.wp != zone.start) {
4814 		f2fs_notice(sbi,
4815 			    "New zone for curseg[%d] is not yet discarded. "
4816 			    "Reset the zone: curseg[0x%x,0x%x]",
4817 			    type, cs->segno, cs->next_blkoff);
4818 		err = __f2fs_issue_discard_zone(sbi, zbd->bdev,
4819 				zone_sector >> log_sectors_per_block,
4820 				zone.len >> log_sectors_per_block);
4821 		if (err) {
4822 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4823 				 zbd->path, err);
4824 			return err;
4825 		}
4826 	}
4827 
4828 	return 0;
4829 }
4830 
4831 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4832 {
4833 	int i, ret;
4834 
4835 	for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4836 		ret = fix_curseg_write_pointer(sbi, i);
4837 		if (ret)
4838 			return ret;
4839 	}
4840 
4841 	return 0;
4842 }
4843 
4844 struct check_zone_write_pointer_args {
4845 	struct f2fs_sb_info *sbi;
4846 	struct f2fs_dev_info *fdev;
4847 };
4848 
4849 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
4850 				      void *data) {
4851 	struct check_zone_write_pointer_args *args;
4852 	args = (struct check_zone_write_pointer_args *)data;
4853 
4854 	return check_zone_write_pointer(args->sbi, args->fdev, zone);
4855 }
4856 
4857 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4858 {
4859 	int i, ret;
4860 	struct check_zone_write_pointer_args args;
4861 
4862 	for (i = 0; i < sbi->s_ndevs; i++) {
4863 		if (!bdev_is_zoned(FDEV(i).bdev))
4864 			continue;
4865 
4866 		args.sbi = sbi;
4867 		args.fdev = &FDEV(i);
4868 		ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
4869 					  check_zone_write_pointer_cb, &args);
4870 		if (ret < 0)
4871 			return ret;
4872 	}
4873 
4874 	return 0;
4875 }
4876 
4877 static bool is_conv_zone(struct f2fs_sb_info *sbi, unsigned int zone_idx,
4878 						unsigned int dev_idx)
4879 {
4880 	if (!bdev_is_zoned(FDEV(dev_idx).bdev))
4881 		return true;
4882 	return !test_bit(zone_idx, FDEV(dev_idx).blkz_seq);
4883 }
4884 
4885 /* Return the zone index in the given device */
4886 static unsigned int get_zone_idx(struct f2fs_sb_info *sbi, unsigned int secno,
4887 					int dev_idx)
4888 {
4889 	block_t sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
4890 
4891 	return (sec_start_blkaddr - FDEV(dev_idx).start_blk) >>
4892 						sbi->log_blocks_per_blkz;
4893 }
4894 
4895 /*
4896  * Return the usable segments in a section based on the zone's
4897  * corresponding zone capacity. Zone is equal to a section.
4898  */
4899 static inline unsigned int f2fs_usable_zone_segs_in_sec(
4900 		struct f2fs_sb_info *sbi, unsigned int segno)
4901 {
4902 	unsigned int dev_idx, zone_idx, unusable_segs_in_sec;
4903 
4904 	dev_idx = f2fs_target_device_index(sbi, START_BLOCK(sbi, segno));
4905 	zone_idx = get_zone_idx(sbi, GET_SEC_FROM_SEG(sbi, segno), dev_idx);
4906 
4907 	/* Conventional zone's capacity is always equal to zone size */
4908 	if (is_conv_zone(sbi, zone_idx, dev_idx))
4909 		return sbi->segs_per_sec;
4910 
4911 	/*
4912 	 * If the zone_capacity_blocks array is NULL, then zone capacity
4913 	 * is equal to the zone size for all zones
4914 	 */
4915 	if (!FDEV(dev_idx).zone_capacity_blocks)
4916 		return sbi->segs_per_sec;
4917 
4918 	/* Get the segment count beyond zone capacity block */
4919 	unusable_segs_in_sec = (sbi->blocks_per_blkz -
4920 				FDEV(dev_idx).zone_capacity_blocks[zone_idx]) >>
4921 				sbi->log_blocks_per_seg;
4922 	return sbi->segs_per_sec - unusable_segs_in_sec;
4923 }
4924 
4925 /*
4926  * Return the number of usable blocks in a segment. The number of blocks
4927  * returned is always equal to the number of blocks in a segment for
4928  * segments fully contained within a sequential zone capacity or a
4929  * conventional zone. For segments partially contained in a sequential
4930  * zone capacity, the number of usable blocks up to the zone capacity
4931  * is returned. 0 is returned in all other cases.
4932  */
4933 static inline unsigned int f2fs_usable_zone_blks_in_seg(
4934 			struct f2fs_sb_info *sbi, unsigned int segno)
4935 {
4936 	block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
4937 	unsigned int zone_idx, dev_idx, secno;
4938 
4939 	secno = GET_SEC_FROM_SEG(sbi, segno);
4940 	seg_start = START_BLOCK(sbi, segno);
4941 	dev_idx = f2fs_target_device_index(sbi, seg_start);
4942 	zone_idx = get_zone_idx(sbi, secno, dev_idx);
4943 
4944 	/*
4945 	 * Conventional zone's capacity is always equal to zone size,
4946 	 * so, blocks per segment is unchanged.
4947 	 */
4948 	if (is_conv_zone(sbi, zone_idx, dev_idx))
4949 		return sbi->blocks_per_seg;
4950 
4951 	if (!FDEV(dev_idx).zone_capacity_blocks)
4952 		return sbi->blocks_per_seg;
4953 
4954 	sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
4955 	sec_cap_blkaddr = sec_start_blkaddr +
4956 				FDEV(dev_idx).zone_capacity_blocks[zone_idx];
4957 
4958 	/*
4959 	 * If segment starts before zone capacity and spans beyond
4960 	 * zone capacity, then usable blocks are from seg start to
4961 	 * zone capacity. If the segment starts after the zone capacity,
4962 	 * then there are no usable blocks.
4963 	 */
4964 	if (seg_start >= sec_cap_blkaddr)
4965 		return 0;
4966 	if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr)
4967 		return sec_cap_blkaddr - seg_start;
4968 
4969 	return sbi->blocks_per_seg;
4970 }
4971 #else
4972 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4973 {
4974 	return 0;
4975 }
4976 
4977 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4978 {
4979 	return 0;
4980 }
4981 
4982 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
4983 							unsigned int segno)
4984 {
4985 	return 0;
4986 }
4987 
4988 static inline unsigned int f2fs_usable_zone_segs_in_sec(struct f2fs_sb_info *sbi,
4989 							unsigned int segno)
4990 {
4991 	return 0;
4992 }
4993 #endif
4994 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
4995 					unsigned int segno)
4996 {
4997 	if (f2fs_sb_has_blkzoned(sbi))
4998 		return f2fs_usable_zone_blks_in_seg(sbi, segno);
4999 
5000 	return sbi->blocks_per_seg;
5001 }
5002 
5003 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5004 					unsigned int segno)
5005 {
5006 	if (f2fs_sb_has_blkzoned(sbi))
5007 		return f2fs_usable_zone_segs_in_sec(sbi, segno);
5008 
5009 	return sbi->segs_per_sec;
5010 }
5011 
5012 /*
5013  * Update min, max modified time for cost-benefit GC algorithm
5014  */
5015 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5016 {
5017 	struct sit_info *sit_i = SIT_I(sbi);
5018 	unsigned int segno;
5019 
5020 	down_write(&sit_i->sentry_lock);
5021 
5022 	sit_i->min_mtime = ULLONG_MAX;
5023 
5024 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
5025 		unsigned int i;
5026 		unsigned long long mtime = 0;
5027 
5028 		for (i = 0; i < sbi->segs_per_sec; i++)
5029 			mtime += get_seg_entry(sbi, segno + i)->mtime;
5030 
5031 		mtime = div_u64(mtime, sbi->segs_per_sec);
5032 
5033 		if (sit_i->min_mtime > mtime)
5034 			sit_i->min_mtime = mtime;
5035 	}
5036 	sit_i->max_mtime = get_mtime(sbi, false);
5037 	sit_i->dirty_max_mtime = 0;
5038 	up_write(&sit_i->sentry_lock);
5039 }
5040 
5041 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5042 {
5043 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5044 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5045 	struct f2fs_sm_info *sm_info;
5046 	int err;
5047 
5048 	sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5049 	if (!sm_info)
5050 		return -ENOMEM;
5051 
5052 	/* init sm info */
5053 	sbi->sm_info = sm_info;
5054 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5055 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5056 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5057 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5058 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5059 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5060 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5061 	sm_info->rec_prefree_segments = sm_info->main_segments *
5062 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5063 	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5064 		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5065 
5066 	if (!f2fs_lfs_mode(sbi))
5067 		sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
5068 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5069 	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5070 	sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
5071 	sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5072 	sm_info->min_ssr_sections = reserved_sections(sbi);
5073 
5074 	INIT_LIST_HEAD(&sm_info->sit_entry_set);
5075 
5076 	init_rwsem(&sm_info->curseg_lock);
5077 
5078 	if (!f2fs_readonly(sbi->sb)) {
5079 		err = f2fs_create_flush_cmd_control(sbi);
5080 		if (err)
5081 			return err;
5082 	}
5083 
5084 	err = create_discard_cmd_control(sbi);
5085 	if (err)
5086 		return err;
5087 
5088 	err = build_sit_info(sbi);
5089 	if (err)
5090 		return err;
5091 	err = build_free_segmap(sbi);
5092 	if (err)
5093 		return err;
5094 	err = build_curseg(sbi);
5095 	if (err)
5096 		return err;
5097 
5098 	/* reinit free segmap based on SIT */
5099 	err = build_sit_entries(sbi);
5100 	if (err)
5101 		return err;
5102 
5103 	init_free_segmap(sbi);
5104 	err = build_dirty_segmap(sbi);
5105 	if (err)
5106 		return err;
5107 
5108 	err = sanity_check_curseg(sbi);
5109 	if (err)
5110 		return err;
5111 
5112 	init_min_max_mtime(sbi);
5113 	return 0;
5114 }
5115 
5116 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5117 		enum dirty_type dirty_type)
5118 {
5119 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5120 
5121 	mutex_lock(&dirty_i->seglist_lock);
5122 	kvfree(dirty_i->dirty_segmap[dirty_type]);
5123 	dirty_i->nr_dirty[dirty_type] = 0;
5124 	mutex_unlock(&dirty_i->seglist_lock);
5125 }
5126 
5127 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5128 {
5129 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5130 	kvfree(dirty_i->victim_secmap);
5131 }
5132 
5133 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5134 {
5135 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5136 	int i;
5137 
5138 	if (!dirty_i)
5139 		return;
5140 
5141 	/* discard pre-free/dirty segments list */
5142 	for (i = 0; i < NR_DIRTY_TYPE; i++)
5143 		discard_dirty_segmap(sbi, i);
5144 
5145 	if (__is_large_section(sbi)) {
5146 		mutex_lock(&dirty_i->seglist_lock);
5147 		kvfree(dirty_i->dirty_secmap);
5148 		mutex_unlock(&dirty_i->seglist_lock);
5149 	}
5150 
5151 	destroy_victim_secmap(sbi);
5152 	SM_I(sbi)->dirty_info = NULL;
5153 	kfree(dirty_i);
5154 }
5155 
5156 static void destroy_curseg(struct f2fs_sb_info *sbi)
5157 {
5158 	struct curseg_info *array = SM_I(sbi)->curseg_array;
5159 	int i;
5160 
5161 	if (!array)
5162 		return;
5163 	SM_I(sbi)->curseg_array = NULL;
5164 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
5165 		kfree(array[i].sum_blk);
5166 		kfree(array[i].journal);
5167 	}
5168 	kfree(array);
5169 }
5170 
5171 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5172 {
5173 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5174 	if (!free_i)
5175 		return;
5176 	SM_I(sbi)->free_info = NULL;
5177 	kvfree(free_i->free_segmap);
5178 	kvfree(free_i->free_secmap);
5179 	kfree(free_i);
5180 }
5181 
5182 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5183 {
5184 	struct sit_info *sit_i = SIT_I(sbi);
5185 
5186 	if (!sit_i)
5187 		return;
5188 
5189 	if (sit_i->sentries)
5190 		kvfree(sit_i->bitmap);
5191 	kfree(sit_i->tmp_map);
5192 
5193 	kvfree(sit_i->sentries);
5194 	kvfree(sit_i->sec_entries);
5195 	kvfree(sit_i->dirty_sentries_bitmap);
5196 
5197 	SM_I(sbi)->sit_info = NULL;
5198 	kvfree(sit_i->sit_bitmap);
5199 #ifdef CONFIG_F2FS_CHECK_FS
5200 	kvfree(sit_i->sit_bitmap_mir);
5201 	kvfree(sit_i->invalid_segmap);
5202 #endif
5203 	kfree(sit_i);
5204 }
5205 
5206 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5207 {
5208 	struct f2fs_sm_info *sm_info = SM_I(sbi);
5209 
5210 	if (!sm_info)
5211 		return;
5212 	f2fs_destroy_flush_cmd_control(sbi, true);
5213 	destroy_discard_cmd_control(sbi);
5214 	destroy_dirty_segmap(sbi);
5215 	destroy_curseg(sbi);
5216 	destroy_free_segmap(sbi);
5217 	destroy_sit_info(sbi);
5218 	sbi->sm_info = NULL;
5219 	kfree(sm_info);
5220 }
5221 
5222 int __init f2fs_create_segment_manager_caches(void)
5223 {
5224 	discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5225 			sizeof(struct discard_entry));
5226 	if (!discard_entry_slab)
5227 		goto fail;
5228 
5229 	discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5230 			sizeof(struct discard_cmd));
5231 	if (!discard_cmd_slab)
5232 		goto destroy_discard_entry;
5233 
5234 	sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5235 			sizeof(struct sit_entry_set));
5236 	if (!sit_entry_set_slab)
5237 		goto destroy_discard_cmd;
5238 
5239 	inmem_entry_slab = f2fs_kmem_cache_create("f2fs_inmem_page_entry",
5240 			sizeof(struct inmem_pages));
5241 	if (!inmem_entry_slab)
5242 		goto destroy_sit_entry_set;
5243 	return 0;
5244 
5245 destroy_sit_entry_set:
5246 	kmem_cache_destroy(sit_entry_set_slab);
5247 destroy_discard_cmd:
5248 	kmem_cache_destroy(discard_cmd_slab);
5249 destroy_discard_entry:
5250 	kmem_cache_destroy(discard_entry_slab);
5251 fail:
5252 	return -ENOMEM;
5253 }
5254 
5255 void f2fs_destroy_segment_manager_caches(void)
5256 {
5257 	kmem_cache_destroy(sit_entry_set_slab);
5258 	kmem_cache_destroy(discard_cmd_slab);
5259 	kmem_cache_destroy(discard_entry_slab);
5260 	kmem_cache_destroy(inmem_entry_slab);
5261 }
5262