xref: /linux/fs/f2fs/segment.c (revision b68fc09be48edbc47de1a0f3d42ef8adf6c0ac55)
1 /*
2  * fs/f2fs/segment.c
3  *
4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5  *             http://www.samsung.com/
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/bio.h>
14 #include <linux/blkdev.h>
15 #include <linux/prefetch.h>
16 #include <linux/kthread.h>
17 #include <linux/swap.h>
18 #include <linux/timer.h>
19 #include <linux/freezer.h>
20 #include <linux/sched/signal.h>
21 
22 #include "f2fs.h"
23 #include "segment.h"
24 #include "node.h"
25 #include "gc.h"
26 #include "trace.h"
27 #include <trace/events/f2fs.h>
28 
29 #define __reverse_ffz(x) __reverse_ffs(~(x))
30 
31 static struct kmem_cache *discard_entry_slab;
32 static struct kmem_cache *discard_cmd_slab;
33 static struct kmem_cache *sit_entry_set_slab;
34 static struct kmem_cache *inmem_entry_slab;
35 
36 static unsigned long __reverse_ulong(unsigned char *str)
37 {
38 	unsigned long tmp = 0;
39 	int shift = 24, idx = 0;
40 
41 #if BITS_PER_LONG == 64
42 	shift = 56;
43 #endif
44 	while (shift >= 0) {
45 		tmp |= (unsigned long)str[idx++] << shift;
46 		shift -= BITS_PER_BYTE;
47 	}
48 	return tmp;
49 }
50 
51 /*
52  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
53  * MSB and LSB are reversed in a byte by f2fs_set_bit.
54  */
55 static inline unsigned long __reverse_ffs(unsigned long word)
56 {
57 	int num = 0;
58 
59 #if BITS_PER_LONG == 64
60 	if ((word & 0xffffffff00000000UL) == 0)
61 		num += 32;
62 	else
63 		word >>= 32;
64 #endif
65 	if ((word & 0xffff0000) == 0)
66 		num += 16;
67 	else
68 		word >>= 16;
69 
70 	if ((word & 0xff00) == 0)
71 		num += 8;
72 	else
73 		word >>= 8;
74 
75 	if ((word & 0xf0) == 0)
76 		num += 4;
77 	else
78 		word >>= 4;
79 
80 	if ((word & 0xc) == 0)
81 		num += 2;
82 	else
83 		word >>= 2;
84 
85 	if ((word & 0x2) == 0)
86 		num += 1;
87 	return num;
88 }
89 
90 /*
91  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
92  * f2fs_set_bit makes MSB and LSB reversed in a byte.
93  * @size must be integral times of unsigned long.
94  * Example:
95  *                             MSB <--> LSB
96  *   f2fs_set_bit(0, bitmap) => 1000 0000
97  *   f2fs_set_bit(7, bitmap) => 0000 0001
98  */
99 static unsigned long __find_rev_next_bit(const unsigned long *addr,
100 			unsigned long size, unsigned long offset)
101 {
102 	const unsigned long *p = addr + BIT_WORD(offset);
103 	unsigned long result = size;
104 	unsigned long tmp;
105 
106 	if (offset >= size)
107 		return size;
108 
109 	size -= (offset & ~(BITS_PER_LONG - 1));
110 	offset %= BITS_PER_LONG;
111 
112 	while (1) {
113 		if (*p == 0)
114 			goto pass;
115 
116 		tmp = __reverse_ulong((unsigned char *)p);
117 
118 		tmp &= ~0UL >> offset;
119 		if (size < BITS_PER_LONG)
120 			tmp &= (~0UL << (BITS_PER_LONG - size));
121 		if (tmp)
122 			goto found;
123 pass:
124 		if (size <= BITS_PER_LONG)
125 			break;
126 		size -= BITS_PER_LONG;
127 		offset = 0;
128 		p++;
129 	}
130 	return result;
131 found:
132 	return result - size + __reverse_ffs(tmp);
133 }
134 
135 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
136 			unsigned long size, unsigned long offset)
137 {
138 	const unsigned long *p = addr + BIT_WORD(offset);
139 	unsigned long result = size;
140 	unsigned long tmp;
141 
142 	if (offset >= size)
143 		return size;
144 
145 	size -= (offset & ~(BITS_PER_LONG - 1));
146 	offset %= BITS_PER_LONG;
147 
148 	while (1) {
149 		if (*p == ~0UL)
150 			goto pass;
151 
152 		tmp = __reverse_ulong((unsigned char *)p);
153 
154 		if (offset)
155 			tmp |= ~0UL << (BITS_PER_LONG - offset);
156 		if (size < BITS_PER_LONG)
157 			tmp |= ~0UL >> size;
158 		if (tmp != ~0UL)
159 			goto found;
160 pass:
161 		if (size <= BITS_PER_LONG)
162 			break;
163 		size -= BITS_PER_LONG;
164 		offset = 0;
165 		p++;
166 	}
167 	return result;
168 found:
169 	return result - size + __reverse_ffz(tmp);
170 }
171 
172 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
173 {
174 	int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
175 	int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
176 	int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
177 
178 	if (test_opt(sbi, LFS))
179 		return false;
180 	if (sbi->gc_mode == GC_URGENT)
181 		return true;
182 
183 	return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
184 			SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
185 }
186 
187 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
188 {
189 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
190 	struct f2fs_inode_info *fi = F2FS_I(inode);
191 	struct inmem_pages *new;
192 
193 	f2fs_trace_pid(page);
194 
195 	set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
196 	SetPagePrivate(page);
197 
198 	new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
199 
200 	/* add atomic page indices to the list */
201 	new->page = page;
202 	INIT_LIST_HEAD(&new->list);
203 
204 	/* increase reference count with clean state */
205 	mutex_lock(&fi->inmem_lock);
206 	get_page(page);
207 	list_add_tail(&new->list, &fi->inmem_pages);
208 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
209 	if (list_empty(&fi->inmem_ilist))
210 		list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
211 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
212 	inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
213 	mutex_unlock(&fi->inmem_lock);
214 
215 	trace_f2fs_register_inmem_page(page, INMEM);
216 }
217 
218 static int __revoke_inmem_pages(struct inode *inode,
219 				struct list_head *head, bool drop, bool recover)
220 {
221 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
222 	struct inmem_pages *cur, *tmp;
223 	int err = 0;
224 
225 	list_for_each_entry_safe(cur, tmp, head, list) {
226 		struct page *page = cur->page;
227 
228 		if (drop)
229 			trace_f2fs_commit_inmem_page(page, INMEM_DROP);
230 
231 		lock_page(page);
232 
233 		f2fs_wait_on_page_writeback(page, DATA, true);
234 
235 		if (recover) {
236 			struct dnode_of_data dn;
237 			struct node_info ni;
238 
239 			trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
240 retry:
241 			set_new_dnode(&dn, inode, NULL, NULL, 0);
242 			err = f2fs_get_dnode_of_data(&dn, page->index,
243 								LOOKUP_NODE);
244 			if (err) {
245 				if (err == -ENOMEM) {
246 					congestion_wait(BLK_RW_ASYNC, HZ/50);
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 		set_page_private(page, 0);
273 		ClearPagePrivate(page);
274 		f2fs_put_page(page, 1);
275 
276 		list_del(&cur->list);
277 		kmem_cache_free(inmem_entry_slab, cur);
278 		dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
279 	}
280 	return err;
281 }
282 
283 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
284 {
285 	struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
286 	struct inode *inode;
287 	struct f2fs_inode_info *fi;
288 next:
289 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
290 	if (list_empty(head)) {
291 		spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
292 		return;
293 	}
294 	fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
295 	inode = igrab(&fi->vfs_inode);
296 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
297 
298 	if (inode) {
299 		if (gc_failure) {
300 			if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
301 				goto drop;
302 			goto skip;
303 		}
304 drop:
305 		set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
306 		f2fs_drop_inmem_pages(inode);
307 		iput(inode);
308 	}
309 skip:
310 	congestion_wait(BLK_RW_ASYNC, HZ/50);
311 	cond_resched();
312 	goto next;
313 }
314 
315 void f2fs_drop_inmem_pages(struct inode *inode)
316 {
317 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
318 	struct f2fs_inode_info *fi = F2FS_I(inode);
319 
320 	mutex_lock(&fi->inmem_lock);
321 	__revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
322 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
323 	if (!list_empty(&fi->inmem_ilist))
324 		list_del_init(&fi->inmem_ilist);
325 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
326 	mutex_unlock(&fi->inmem_lock);
327 
328 	clear_inode_flag(inode, FI_ATOMIC_FILE);
329 	fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
330 	stat_dec_atomic_write(inode);
331 }
332 
333 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
334 {
335 	struct f2fs_inode_info *fi = F2FS_I(inode);
336 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
337 	struct list_head *head = &fi->inmem_pages;
338 	struct inmem_pages *cur = NULL;
339 
340 	f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
341 
342 	mutex_lock(&fi->inmem_lock);
343 	list_for_each_entry(cur, head, list) {
344 		if (cur->page == page)
345 			break;
346 	}
347 
348 	f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
349 	list_del(&cur->list);
350 	mutex_unlock(&fi->inmem_lock);
351 
352 	dec_page_count(sbi, F2FS_INMEM_PAGES);
353 	kmem_cache_free(inmem_entry_slab, cur);
354 
355 	ClearPageUptodate(page);
356 	set_page_private(page, 0);
357 	ClearPagePrivate(page);
358 	f2fs_put_page(page, 0);
359 
360 	trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
361 }
362 
363 static int __f2fs_commit_inmem_pages(struct inode *inode)
364 {
365 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
366 	struct f2fs_inode_info *fi = F2FS_I(inode);
367 	struct inmem_pages *cur, *tmp;
368 	struct f2fs_io_info fio = {
369 		.sbi = sbi,
370 		.ino = inode->i_ino,
371 		.type = DATA,
372 		.op = REQ_OP_WRITE,
373 		.op_flags = REQ_SYNC | REQ_PRIO,
374 		.io_type = FS_DATA_IO,
375 	};
376 	struct list_head revoke_list;
377 	pgoff_t last_idx = ULONG_MAX;
378 	int err = 0;
379 
380 	INIT_LIST_HEAD(&revoke_list);
381 
382 	list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
383 		struct page *page = cur->page;
384 
385 		lock_page(page);
386 		if (page->mapping == inode->i_mapping) {
387 			trace_f2fs_commit_inmem_page(page, INMEM);
388 
389 			set_page_dirty(page);
390 			f2fs_wait_on_page_writeback(page, DATA, true);
391 			if (clear_page_dirty_for_io(page)) {
392 				inode_dec_dirty_pages(inode);
393 				f2fs_remove_dirty_inode(inode);
394 			}
395 retry:
396 			fio.page = page;
397 			fio.old_blkaddr = NULL_ADDR;
398 			fio.encrypted_page = NULL;
399 			fio.need_lock = LOCK_DONE;
400 			err = f2fs_do_write_data_page(&fio);
401 			if (err) {
402 				if (err == -ENOMEM) {
403 					congestion_wait(BLK_RW_ASYNC, HZ/50);
404 					cond_resched();
405 					goto retry;
406 				}
407 				unlock_page(page);
408 				break;
409 			}
410 			/* record old blkaddr for revoking */
411 			cur->old_addr = fio.old_blkaddr;
412 			last_idx = page->index;
413 		}
414 		unlock_page(page);
415 		list_move_tail(&cur->list, &revoke_list);
416 	}
417 
418 	if (last_idx != ULONG_MAX)
419 		f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
420 
421 	if (err) {
422 		/*
423 		 * try to revoke all committed pages, but still we could fail
424 		 * due to no memory or other reason, if that happened, EAGAIN
425 		 * will be returned, which means in such case, transaction is
426 		 * already not integrity, caller should use journal to do the
427 		 * recovery or rewrite & commit last transaction. For other
428 		 * error number, revoking was done by filesystem itself.
429 		 */
430 		err = __revoke_inmem_pages(inode, &revoke_list, false, true);
431 
432 		/* drop all uncommitted pages */
433 		__revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
434 	} else {
435 		__revoke_inmem_pages(inode, &revoke_list, false, false);
436 	}
437 
438 	return err;
439 }
440 
441 int f2fs_commit_inmem_pages(struct inode *inode)
442 {
443 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
444 	struct f2fs_inode_info *fi = F2FS_I(inode);
445 	int err;
446 
447 	f2fs_balance_fs(sbi, true);
448 
449 	down_write(&fi->i_gc_rwsem[WRITE]);
450 
451 	f2fs_lock_op(sbi);
452 	set_inode_flag(inode, FI_ATOMIC_COMMIT);
453 
454 	mutex_lock(&fi->inmem_lock);
455 	err = __f2fs_commit_inmem_pages(inode);
456 
457 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
458 	if (!list_empty(&fi->inmem_ilist))
459 		list_del_init(&fi->inmem_ilist);
460 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
461 	mutex_unlock(&fi->inmem_lock);
462 
463 	clear_inode_flag(inode, FI_ATOMIC_COMMIT);
464 
465 	f2fs_unlock_op(sbi);
466 	up_write(&fi->i_gc_rwsem[WRITE]);
467 
468 	return err;
469 }
470 
471 /*
472  * This function balances dirty node and dentry pages.
473  * In addition, it controls garbage collection.
474  */
475 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
476 {
477 	if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
478 		f2fs_show_injection_info(FAULT_CHECKPOINT);
479 		f2fs_stop_checkpoint(sbi, false);
480 	}
481 
482 	/* balance_fs_bg is able to be pending */
483 	if (need && excess_cached_nats(sbi))
484 		f2fs_balance_fs_bg(sbi);
485 
486 	/*
487 	 * We should do GC or end up with checkpoint, if there are so many dirty
488 	 * dir/node pages without enough free segments.
489 	 */
490 	if (has_not_enough_free_secs(sbi, 0, 0)) {
491 		mutex_lock(&sbi->gc_mutex);
492 		f2fs_gc(sbi, false, false, NULL_SEGNO);
493 	}
494 }
495 
496 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
497 {
498 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
499 		return;
500 
501 	/* try to shrink extent cache when there is no enough memory */
502 	if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
503 		f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
504 
505 	/* check the # of cached NAT entries */
506 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
507 		f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
508 
509 	if (!f2fs_available_free_memory(sbi, FREE_NIDS))
510 		f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
511 	else
512 		f2fs_build_free_nids(sbi, false, false);
513 
514 	if (!is_idle(sbi) &&
515 		(!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
516 		return;
517 
518 	/* checkpoint is the only way to shrink partial cached entries */
519 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
520 			!f2fs_available_free_memory(sbi, INO_ENTRIES) ||
521 			excess_prefree_segs(sbi) ||
522 			excess_dirty_nats(sbi) ||
523 			excess_dirty_nodes(sbi) ||
524 			f2fs_time_over(sbi, CP_TIME)) {
525 		if (test_opt(sbi, DATA_FLUSH)) {
526 			struct blk_plug plug;
527 
528 			blk_start_plug(&plug);
529 			f2fs_sync_dirty_inodes(sbi, FILE_INODE);
530 			blk_finish_plug(&plug);
531 		}
532 		f2fs_sync_fs(sbi->sb, true);
533 		stat_inc_bg_cp_count(sbi->stat_info);
534 	}
535 }
536 
537 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
538 				struct block_device *bdev)
539 {
540 	struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
541 	int ret;
542 
543 	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
544 	bio_set_dev(bio, bdev);
545 	ret = submit_bio_wait(bio);
546 	bio_put(bio);
547 
548 	trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
549 				test_opt(sbi, FLUSH_MERGE), ret);
550 	return ret;
551 }
552 
553 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
554 {
555 	int ret = 0;
556 	int i;
557 
558 	if (!sbi->s_ndevs)
559 		return __submit_flush_wait(sbi, sbi->sb->s_bdev);
560 
561 	for (i = 0; i < sbi->s_ndevs; i++) {
562 		if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
563 			continue;
564 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
565 		if (ret)
566 			break;
567 	}
568 	return ret;
569 }
570 
571 static int issue_flush_thread(void *data)
572 {
573 	struct f2fs_sb_info *sbi = data;
574 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
575 	wait_queue_head_t *q = &fcc->flush_wait_queue;
576 repeat:
577 	if (kthread_should_stop())
578 		return 0;
579 
580 	sb_start_intwrite(sbi->sb);
581 
582 	if (!llist_empty(&fcc->issue_list)) {
583 		struct flush_cmd *cmd, *next;
584 		int ret;
585 
586 		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
587 		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
588 
589 		cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
590 
591 		ret = submit_flush_wait(sbi, cmd->ino);
592 		atomic_inc(&fcc->issued_flush);
593 
594 		llist_for_each_entry_safe(cmd, next,
595 					  fcc->dispatch_list, llnode) {
596 			cmd->ret = ret;
597 			complete(&cmd->wait);
598 		}
599 		fcc->dispatch_list = NULL;
600 	}
601 
602 	sb_end_intwrite(sbi->sb);
603 
604 	wait_event_interruptible(*q,
605 		kthread_should_stop() || !llist_empty(&fcc->issue_list));
606 	goto repeat;
607 }
608 
609 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
610 {
611 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
612 	struct flush_cmd cmd;
613 	int ret;
614 
615 	if (test_opt(sbi, NOBARRIER))
616 		return 0;
617 
618 	if (!test_opt(sbi, FLUSH_MERGE)) {
619 		ret = submit_flush_wait(sbi, ino);
620 		atomic_inc(&fcc->issued_flush);
621 		return ret;
622 	}
623 
624 	if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
625 		ret = submit_flush_wait(sbi, ino);
626 		atomic_dec(&fcc->issing_flush);
627 
628 		atomic_inc(&fcc->issued_flush);
629 		return ret;
630 	}
631 
632 	cmd.ino = ino;
633 	init_completion(&cmd.wait);
634 
635 	llist_add(&cmd.llnode, &fcc->issue_list);
636 
637 	/* update issue_list before we wake up issue_flush thread */
638 	smp_mb();
639 
640 	if (waitqueue_active(&fcc->flush_wait_queue))
641 		wake_up(&fcc->flush_wait_queue);
642 
643 	if (fcc->f2fs_issue_flush) {
644 		wait_for_completion(&cmd.wait);
645 		atomic_dec(&fcc->issing_flush);
646 	} else {
647 		struct llist_node *list;
648 
649 		list = llist_del_all(&fcc->issue_list);
650 		if (!list) {
651 			wait_for_completion(&cmd.wait);
652 			atomic_dec(&fcc->issing_flush);
653 		} else {
654 			struct flush_cmd *tmp, *next;
655 
656 			ret = submit_flush_wait(sbi, ino);
657 
658 			llist_for_each_entry_safe(tmp, next, list, llnode) {
659 				if (tmp == &cmd) {
660 					cmd.ret = ret;
661 					atomic_dec(&fcc->issing_flush);
662 					continue;
663 				}
664 				tmp->ret = ret;
665 				complete(&tmp->wait);
666 			}
667 		}
668 	}
669 
670 	return cmd.ret;
671 }
672 
673 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
674 {
675 	dev_t dev = sbi->sb->s_bdev->bd_dev;
676 	struct flush_cmd_control *fcc;
677 	int err = 0;
678 
679 	if (SM_I(sbi)->fcc_info) {
680 		fcc = SM_I(sbi)->fcc_info;
681 		if (fcc->f2fs_issue_flush)
682 			return err;
683 		goto init_thread;
684 	}
685 
686 	fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
687 	if (!fcc)
688 		return -ENOMEM;
689 	atomic_set(&fcc->issued_flush, 0);
690 	atomic_set(&fcc->issing_flush, 0);
691 	init_waitqueue_head(&fcc->flush_wait_queue);
692 	init_llist_head(&fcc->issue_list);
693 	SM_I(sbi)->fcc_info = fcc;
694 	if (!test_opt(sbi, FLUSH_MERGE))
695 		return err;
696 
697 init_thread:
698 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
699 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
700 	if (IS_ERR(fcc->f2fs_issue_flush)) {
701 		err = PTR_ERR(fcc->f2fs_issue_flush);
702 		kfree(fcc);
703 		SM_I(sbi)->fcc_info = NULL;
704 		return err;
705 	}
706 
707 	return err;
708 }
709 
710 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
711 {
712 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
713 
714 	if (fcc && fcc->f2fs_issue_flush) {
715 		struct task_struct *flush_thread = fcc->f2fs_issue_flush;
716 
717 		fcc->f2fs_issue_flush = NULL;
718 		kthread_stop(flush_thread);
719 	}
720 	if (free) {
721 		kfree(fcc);
722 		SM_I(sbi)->fcc_info = NULL;
723 	}
724 }
725 
726 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
727 {
728 	int ret = 0, i;
729 
730 	if (!sbi->s_ndevs)
731 		return 0;
732 
733 	for (i = 1; i < sbi->s_ndevs; i++) {
734 		if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
735 			continue;
736 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
737 		if (ret)
738 			break;
739 
740 		spin_lock(&sbi->dev_lock);
741 		f2fs_clear_bit(i, (char *)&sbi->dirty_device);
742 		spin_unlock(&sbi->dev_lock);
743 	}
744 
745 	return ret;
746 }
747 
748 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
749 		enum dirty_type dirty_type)
750 {
751 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
752 
753 	/* need not be added */
754 	if (IS_CURSEG(sbi, segno))
755 		return;
756 
757 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
758 		dirty_i->nr_dirty[dirty_type]++;
759 
760 	if (dirty_type == DIRTY) {
761 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
762 		enum dirty_type t = sentry->type;
763 
764 		if (unlikely(t >= DIRTY)) {
765 			f2fs_bug_on(sbi, 1);
766 			return;
767 		}
768 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
769 			dirty_i->nr_dirty[t]++;
770 	}
771 }
772 
773 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
774 		enum dirty_type dirty_type)
775 {
776 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
777 
778 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
779 		dirty_i->nr_dirty[dirty_type]--;
780 
781 	if (dirty_type == DIRTY) {
782 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
783 		enum dirty_type t = sentry->type;
784 
785 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
786 			dirty_i->nr_dirty[t]--;
787 
788 		if (get_valid_blocks(sbi, segno, true) == 0)
789 			clear_bit(GET_SEC_FROM_SEG(sbi, segno),
790 						dirty_i->victim_secmap);
791 	}
792 }
793 
794 /*
795  * Should not occur error such as -ENOMEM.
796  * Adding dirty entry into seglist is not critical operation.
797  * If a given segment is one of current working segments, it won't be added.
798  */
799 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
800 {
801 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
802 	unsigned short valid_blocks;
803 
804 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
805 		return;
806 
807 	mutex_lock(&dirty_i->seglist_lock);
808 
809 	valid_blocks = get_valid_blocks(sbi, segno, false);
810 
811 	if (valid_blocks == 0) {
812 		__locate_dirty_segment(sbi, segno, PRE);
813 		__remove_dirty_segment(sbi, segno, DIRTY);
814 	} else if (valid_blocks < sbi->blocks_per_seg) {
815 		__locate_dirty_segment(sbi, segno, DIRTY);
816 	} else {
817 		/* Recovery routine with SSR needs this */
818 		__remove_dirty_segment(sbi, segno, DIRTY);
819 	}
820 
821 	mutex_unlock(&dirty_i->seglist_lock);
822 }
823 
824 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
825 		struct block_device *bdev, block_t lstart,
826 		block_t start, block_t len)
827 {
828 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
829 	struct list_head *pend_list;
830 	struct discard_cmd *dc;
831 
832 	f2fs_bug_on(sbi, !len);
833 
834 	pend_list = &dcc->pend_list[plist_idx(len)];
835 
836 	dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
837 	INIT_LIST_HEAD(&dc->list);
838 	dc->bdev = bdev;
839 	dc->lstart = lstart;
840 	dc->start = start;
841 	dc->len = len;
842 	dc->ref = 0;
843 	dc->state = D_PREP;
844 	dc->issuing = 0;
845 	dc->error = 0;
846 	init_completion(&dc->wait);
847 	list_add_tail(&dc->list, pend_list);
848 	spin_lock_init(&dc->lock);
849 	dc->bio_ref = 0;
850 	atomic_inc(&dcc->discard_cmd_cnt);
851 	dcc->undiscard_blks += len;
852 
853 	return dc;
854 }
855 
856 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
857 				struct block_device *bdev, block_t lstart,
858 				block_t start, block_t len,
859 				struct rb_node *parent, struct rb_node **p)
860 {
861 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
862 	struct discard_cmd *dc;
863 
864 	dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
865 
866 	rb_link_node(&dc->rb_node, parent, p);
867 	rb_insert_color(&dc->rb_node, &dcc->root);
868 
869 	return dc;
870 }
871 
872 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
873 							struct discard_cmd *dc)
874 {
875 	if (dc->state == D_DONE)
876 		atomic_sub(dc->issuing, &dcc->issing_discard);
877 
878 	list_del(&dc->list);
879 	rb_erase(&dc->rb_node, &dcc->root);
880 	dcc->undiscard_blks -= dc->len;
881 
882 	kmem_cache_free(discard_cmd_slab, dc);
883 
884 	atomic_dec(&dcc->discard_cmd_cnt);
885 }
886 
887 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
888 							struct discard_cmd *dc)
889 {
890 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
891 	unsigned long flags;
892 
893 	trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
894 
895 	spin_lock_irqsave(&dc->lock, flags);
896 	if (dc->bio_ref) {
897 		spin_unlock_irqrestore(&dc->lock, flags);
898 		return;
899 	}
900 	spin_unlock_irqrestore(&dc->lock, flags);
901 
902 	f2fs_bug_on(sbi, dc->ref);
903 
904 	if (dc->error == -EOPNOTSUPP)
905 		dc->error = 0;
906 
907 	if (dc->error)
908 		f2fs_msg(sbi->sb, KERN_INFO,
909 			"Issue discard(%u, %u, %u) failed, ret: %d",
910 			dc->lstart, dc->start, dc->len, dc->error);
911 	__detach_discard_cmd(dcc, dc);
912 }
913 
914 static void f2fs_submit_discard_endio(struct bio *bio)
915 {
916 	struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
917 	unsigned long flags;
918 
919 	dc->error = blk_status_to_errno(bio->bi_status);
920 
921 	spin_lock_irqsave(&dc->lock, flags);
922 	dc->bio_ref--;
923 	if (!dc->bio_ref && dc->state == D_SUBMIT) {
924 		dc->state = D_DONE;
925 		complete_all(&dc->wait);
926 	}
927 	spin_unlock_irqrestore(&dc->lock, flags);
928 	bio_put(bio);
929 }
930 
931 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
932 				block_t start, block_t end)
933 {
934 #ifdef CONFIG_F2FS_CHECK_FS
935 	struct seg_entry *sentry;
936 	unsigned int segno;
937 	block_t blk = start;
938 	unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
939 	unsigned long *map;
940 
941 	while (blk < end) {
942 		segno = GET_SEGNO(sbi, blk);
943 		sentry = get_seg_entry(sbi, segno);
944 		offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
945 
946 		if (end < START_BLOCK(sbi, segno + 1))
947 			size = GET_BLKOFF_FROM_SEG0(sbi, end);
948 		else
949 			size = max_blocks;
950 		map = (unsigned long *)(sentry->cur_valid_map);
951 		offset = __find_rev_next_bit(map, size, offset);
952 		f2fs_bug_on(sbi, offset != size);
953 		blk = START_BLOCK(sbi, segno + 1);
954 	}
955 #endif
956 }
957 
958 static void __init_discard_policy(struct f2fs_sb_info *sbi,
959 				struct discard_policy *dpolicy,
960 				int discard_type, unsigned int granularity)
961 {
962 	/* common policy */
963 	dpolicy->type = discard_type;
964 	dpolicy->sync = true;
965 	dpolicy->ordered = false;
966 	dpolicy->granularity = granularity;
967 
968 	dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
969 	dpolicy->io_aware_gran = MAX_PLIST_NUM;
970 
971 	if (discard_type == DPOLICY_BG) {
972 		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
973 		dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
974 		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
975 		dpolicy->io_aware = true;
976 		dpolicy->sync = false;
977 		dpolicy->ordered = true;
978 		if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
979 			dpolicy->granularity = 1;
980 			dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
981 		}
982 	} else if (discard_type == DPOLICY_FORCE) {
983 		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
984 		dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
985 		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
986 		dpolicy->io_aware = false;
987 	} else if (discard_type == DPOLICY_FSTRIM) {
988 		dpolicy->io_aware = false;
989 	} else if (discard_type == DPOLICY_UMOUNT) {
990 		dpolicy->max_requests = UINT_MAX;
991 		dpolicy->io_aware = false;
992 	}
993 }
994 
995 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
996 				struct block_device *bdev, block_t lstart,
997 				block_t start, block_t len);
998 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
999 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1000 						struct discard_policy *dpolicy,
1001 						struct discard_cmd *dc,
1002 						unsigned int *issued)
1003 {
1004 	struct block_device *bdev = dc->bdev;
1005 	struct request_queue *q = bdev_get_queue(bdev);
1006 	unsigned int max_discard_blocks =
1007 			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1008 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1009 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1010 					&(dcc->fstrim_list) : &(dcc->wait_list);
1011 	int flag = dpolicy->sync ? REQ_SYNC : 0;
1012 	block_t lstart, start, len, total_len;
1013 	int err = 0;
1014 
1015 	if (dc->state != D_PREP)
1016 		return 0;
1017 
1018 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1019 		return 0;
1020 
1021 	trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1022 
1023 	lstart = dc->lstart;
1024 	start = dc->start;
1025 	len = dc->len;
1026 	total_len = len;
1027 
1028 	dc->len = 0;
1029 
1030 	while (total_len && *issued < dpolicy->max_requests && !err) {
1031 		struct bio *bio = NULL;
1032 		unsigned long flags;
1033 		bool last = true;
1034 
1035 		if (len > max_discard_blocks) {
1036 			len = max_discard_blocks;
1037 			last = false;
1038 		}
1039 
1040 		(*issued)++;
1041 		if (*issued == dpolicy->max_requests)
1042 			last = true;
1043 
1044 		dc->len += len;
1045 
1046 		if (time_to_inject(sbi, FAULT_DISCARD)) {
1047 			f2fs_show_injection_info(FAULT_DISCARD);
1048 			err = -EIO;
1049 			goto submit;
1050 		}
1051 		err = __blkdev_issue_discard(bdev,
1052 					SECTOR_FROM_BLOCK(start),
1053 					SECTOR_FROM_BLOCK(len),
1054 					GFP_NOFS, 0, &bio);
1055 submit:
1056 		if (err) {
1057 			spin_lock_irqsave(&dc->lock, flags);
1058 			if (dc->state == D_PARTIAL)
1059 				dc->state = D_SUBMIT;
1060 			spin_unlock_irqrestore(&dc->lock, flags);
1061 
1062 			break;
1063 		}
1064 
1065 		f2fs_bug_on(sbi, !bio);
1066 
1067 		/*
1068 		 * should keep before submission to avoid D_DONE
1069 		 * right away
1070 		 */
1071 		spin_lock_irqsave(&dc->lock, flags);
1072 		if (last)
1073 			dc->state = D_SUBMIT;
1074 		else
1075 			dc->state = D_PARTIAL;
1076 		dc->bio_ref++;
1077 		spin_unlock_irqrestore(&dc->lock, flags);
1078 
1079 		atomic_inc(&dcc->issing_discard);
1080 		dc->issuing++;
1081 		list_move_tail(&dc->list, wait_list);
1082 
1083 		/* sanity check on discard range */
1084 		__check_sit_bitmap(sbi, start, start + len);
1085 
1086 		bio->bi_private = dc;
1087 		bio->bi_end_io = f2fs_submit_discard_endio;
1088 		bio->bi_opf |= flag;
1089 		submit_bio(bio);
1090 
1091 		atomic_inc(&dcc->issued_discard);
1092 
1093 		f2fs_update_iostat(sbi, FS_DISCARD, 1);
1094 
1095 		lstart += len;
1096 		start += len;
1097 		total_len -= len;
1098 		len = total_len;
1099 	}
1100 
1101 	if (!err && len)
1102 		__update_discard_tree_range(sbi, bdev, lstart, start, len);
1103 	return err;
1104 }
1105 
1106 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1107 				struct block_device *bdev, block_t lstart,
1108 				block_t start, block_t len,
1109 				struct rb_node **insert_p,
1110 				struct rb_node *insert_parent)
1111 {
1112 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1113 	struct rb_node **p;
1114 	struct rb_node *parent = NULL;
1115 	struct discard_cmd *dc = NULL;
1116 
1117 	if (insert_p && insert_parent) {
1118 		parent = insert_parent;
1119 		p = insert_p;
1120 		goto do_insert;
1121 	}
1122 
1123 	p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
1124 do_insert:
1125 	dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1126 	if (!dc)
1127 		return NULL;
1128 
1129 	return dc;
1130 }
1131 
1132 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1133 						struct discard_cmd *dc)
1134 {
1135 	list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1136 }
1137 
1138 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1139 				struct discard_cmd *dc, block_t blkaddr)
1140 {
1141 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1142 	struct discard_info di = dc->di;
1143 	bool modified = false;
1144 
1145 	if (dc->state == D_DONE || dc->len == 1) {
1146 		__remove_discard_cmd(sbi, dc);
1147 		return;
1148 	}
1149 
1150 	dcc->undiscard_blks -= di.len;
1151 
1152 	if (blkaddr > di.lstart) {
1153 		dc->len = blkaddr - dc->lstart;
1154 		dcc->undiscard_blks += dc->len;
1155 		__relocate_discard_cmd(dcc, dc);
1156 		modified = true;
1157 	}
1158 
1159 	if (blkaddr < di.lstart + di.len - 1) {
1160 		if (modified) {
1161 			__insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1162 					di.start + blkaddr + 1 - di.lstart,
1163 					di.lstart + di.len - 1 - blkaddr,
1164 					NULL, NULL);
1165 		} else {
1166 			dc->lstart++;
1167 			dc->len--;
1168 			dc->start++;
1169 			dcc->undiscard_blks += dc->len;
1170 			__relocate_discard_cmd(dcc, dc);
1171 		}
1172 	}
1173 }
1174 
1175 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1176 				struct block_device *bdev, block_t lstart,
1177 				block_t start, block_t len)
1178 {
1179 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1180 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1181 	struct discard_cmd *dc;
1182 	struct discard_info di = {0};
1183 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1184 	struct request_queue *q = bdev_get_queue(bdev);
1185 	unsigned int max_discard_blocks =
1186 			SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1187 	block_t end = lstart + len;
1188 
1189 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1190 					NULL, lstart,
1191 					(struct rb_entry **)&prev_dc,
1192 					(struct rb_entry **)&next_dc,
1193 					&insert_p, &insert_parent, true);
1194 	if (dc)
1195 		prev_dc = dc;
1196 
1197 	if (!prev_dc) {
1198 		di.lstart = lstart;
1199 		di.len = next_dc ? next_dc->lstart - lstart : len;
1200 		di.len = min(di.len, len);
1201 		di.start = start;
1202 	}
1203 
1204 	while (1) {
1205 		struct rb_node *node;
1206 		bool merged = false;
1207 		struct discard_cmd *tdc = NULL;
1208 
1209 		if (prev_dc) {
1210 			di.lstart = prev_dc->lstart + prev_dc->len;
1211 			if (di.lstart < lstart)
1212 				di.lstart = lstart;
1213 			if (di.lstart >= end)
1214 				break;
1215 
1216 			if (!next_dc || next_dc->lstart > end)
1217 				di.len = end - di.lstart;
1218 			else
1219 				di.len = next_dc->lstart - di.lstart;
1220 			di.start = start + di.lstart - lstart;
1221 		}
1222 
1223 		if (!di.len)
1224 			goto next;
1225 
1226 		if (prev_dc && prev_dc->state == D_PREP &&
1227 			prev_dc->bdev == bdev &&
1228 			__is_discard_back_mergeable(&di, &prev_dc->di,
1229 							max_discard_blocks)) {
1230 			prev_dc->di.len += di.len;
1231 			dcc->undiscard_blks += di.len;
1232 			__relocate_discard_cmd(dcc, prev_dc);
1233 			di = prev_dc->di;
1234 			tdc = prev_dc;
1235 			merged = true;
1236 		}
1237 
1238 		if (next_dc && next_dc->state == D_PREP &&
1239 			next_dc->bdev == bdev &&
1240 			__is_discard_front_mergeable(&di, &next_dc->di,
1241 							max_discard_blocks)) {
1242 			next_dc->di.lstart = di.lstart;
1243 			next_dc->di.len += di.len;
1244 			next_dc->di.start = di.start;
1245 			dcc->undiscard_blks += di.len;
1246 			__relocate_discard_cmd(dcc, next_dc);
1247 			if (tdc)
1248 				__remove_discard_cmd(sbi, tdc);
1249 			merged = true;
1250 		}
1251 
1252 		if (!merged) {
1253 			__insert_discard_tree(sbi, bdev, di.lstart, di.start,
1254 							di.len, NULL, NULL);
1255 		}
1256  next:
1257 		prev_dc = next_dc;
1258 		if (!prev_dc)
1259 			break;
1260 
1261 		node = rb_next(&prev_dc->rb_node);
1262 		next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1263 	}
1264 }
1265 
1266 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1267 		struct block_device *bdev, block_t blkstart, block_t blklen)
1268 {
1269 	block_t lblkstart = blkstart;
1270 
1271 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
1272 
1273 	if (sbi->s_ndevs) {
1274 		int devi = f2fs_target_device_index(sbi, blkstart);
1275 
1276 		blkstart -= FDEV(devi).start_blk;
1277 	}
1278 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1279 	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1280 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1281 	return 0;
1282 }
1283 
1284 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1285 					struct discard_policy *dpolicy)
1286 {
1287 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1288 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1289 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1290 	struct discard_cmd *dc;
1291 	struct blk_plug plug;
1292 	unsigned int pos = dcc->next_pos;
1293 	unsigned int issued = 0;
1294 	bool io_interrupted = false;
1295 
1296 	mutex_lock(&dcc->cmd_lock);
1297 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1298 					NULL, pos,
1299 					(struct rb_entry **)&prev_dc,
1300 					(struct rb_entry **)&next_dc,
1301 					&insert_p, &insert_parent, true);
1302 	if (!dc)
1303 		dc = next_dc;
1304 
1305 	blk_start_plug(&plug);
1306 
1307 	while (dc) {
1308 		struct rb_node *node;
1309 		int err = 0;
1310 
1311 		if (dc->state != D_PREP)
1312 			goto next;
1313 
1314 		if (dpolicy->io_aware && !is_idle(sbi)) {
1315 			io_interrupted = true;
1316 			break;
1317 		}
1318 
1319 		dcc->next_pos = dc->lstart + dc->len;
1320 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1321 
1322 		if (issued >= dpolicy->max_requests)
1323 			break;
1324 next:
1325 		node = rb_next(&dc->rb_node);
1326 		if (err)
1327 			__remove_discard_cmd(sbi, dc);
1328 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1329 	}
1330 
1331 	blk_finish_plug(&plug);
1332 
1333 	if (!dc)
1334 		dcc->next_pos = 0;
1335 
1336 	mutex_unlock(&dcc->cmd_lock);
1337 
1338 	if (!issued && io_interrupted)
1339 		issued = -1;
1340 
1341 	return issued;
1342 }
1343 
1344 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1345 					struct discard_policy *dpolicy)
1346 {
1347 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1348 	struct list_head *pend_list;
1349 	struct discard_cmd *dc, *tmp;
1350 	struct blk_plug plug;
1351 	int i, issued = 0;
1352 	bool io_interrupted = false;
1353 
1354 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1355 		if (i + 1 < dpolicy->granularity)
1356 			break;
1357 
1358 		if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1359 			return __issue_discard_cmd_orderly(sbi, dpolicy);
1360 
1361 		pend_list = &dcc->pend_list[i];
1362 
1363 		mutex_lock(&dcc->cmd_lock);
1364 		if (list_empty(pend_list))
1365 			goto next;
1366 		if (unlikely(dcc->rbtree_check))
1367 			f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1368 								&dcc->root));
1369 		blk_start_plug(&plug);
1370 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1371 			f2fs_bug_on(sbi, dc->state != D_PREP);
1372 
1373 			if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1374 								!is_idle(sbi)) {
1375 				io_interrupted = true;
1376 				break;
1377 			}
1378 
1379 			__submit_discard_cmd(sbi, dpolicy, dc, &issued);
1380 
1381 			if (issued >= dpolicy->max_requests)
1382 				break;
1383 		}
1384 		blk_finish_plug(&plug);
1385 next:
1386 		mutex_unlock(&dcc->cmd_lock);
1387 
1388 		if (issued >= dpolicy->max_requests || io_interrupted)
1389 			break;
1390 	}
1391 
1392 	if (!issued && io_interrupted)
1393 		issued = -1;
1394 
1395 	return issued;
1396 }
1397 
1398 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1399 {
1400 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1401 	struct list_head *pend_list;
1402 	struct discard_cmd *dc, *tmp;
1403 	int i;
1404 	bool dropped = false;
1405 
1406 	mutex_lock(&dcc->cmd_lock);
1407 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1408 		pend_list = &dcc->pend_list[i];
1409 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1410 			f2fs_bug_on(sbi, dc->state != D_PREP);
1411 			__remove_discard_cmd(sbi, dc);
1412 			dropped = true;
1413 		}
1414 	}
1415 	mutex_unlock(&dcc->cmd_lock);
1416 
1417 	return dropped;
1418 }
1419 
1420 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1421 {
1422 	__drop_discard_cmd(sbi);
1423 }
1424 
1425 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1426 							struct discard_cmd *dc)
1427 {
1428 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1429 	unsigned int len = 0;
1430 
1431 	wait_for_completion_io(&dc->wait);
1432 	mutex_lock(&dcc->cmd_lock);
1433 	f2fs_bug_on(sbi, dc->state != D_DONE);
1434 	dc->ref--;
1435 	if (!dc->ref) {
1436 		if (!dc->error)
1437 			len = dc->len;
1438 		__remove_discard_cmd(sbi, dc);
1439 	}
1440 	mutex_unlock(&dcc->cmd_lock);
1441 
1442 	return len;
1443 }
1444 
1445 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1446 						struct discard_policy *dpolicy,
1447 						block_t start, block_t end)
1448 {
1449 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1450 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1451 					&(dcc->fstrim_list) : &(dcc->wait_list);
1452 	struct discard_cmd *dc, *tmp;
1453 	bool need_wait;
1454 	unsigned int trimmed = 0;
1455 
1456 next:
1457 	need_wait = false;
1458 
1459 	mutex_lock(&dcc->cmd_lock);
1460 	list_for_each_entry_safe(dc, tmp, wait_list, list) {
1461 		if (dc->lstart + dc->len <= start || end <= dc->lstart)
1462 			continue;
1463 		if (dc->len < dpolicy->granularity)
1464 			continue;
1465 		if (dc->state == D_DONE && !dc->ref) {
1466 			wait_for_completion_io(&dc->wait);
1467 			if (!dc->error)
1468 				trimmed += dc->len;
1469 			__remove_discard_cmd(sbi, dc);
1470 		} else {
1471 			dc->ref++;
1472 			need_wait = true;
1473 			break;
1474 		}
1475 	}
1476 	mutex_unlock(&dcc->cmd_lock);
1477 
1478 	if (need_wait) {
1479 		trimmed += __wait_one_discard_bio(sbi, dc);
1480 		goto next;
1481 	}
1482 
1483 	return trimmed;
1484 }
1485 
1486 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1487 						struct discard_policy *dpolicy)
1488 {
1489 	struct discard_policy dp;
1490 	unsigned int discard_blks;
1491 
1492 	if (dpolicy)
1493 		return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1494 
1495 	/* wait all */
1496 	__init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1497 	discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1498 	__init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1499 	discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1500 
1501 	return discard_blks;
1502 }
1503 
1504 /* This should be covered by global mutex, &sit_i->sentry_lock */
1505 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1506 {
1507 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1508 	struct discard_cmd *dc;
1509 	bool need_wait = false;
1510 
1511 	mutex_lock(&dcc->cmd_lock);
1512 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1513 							NULL, blkaddr);
1514 	if (dc) {
1515 		if (dc->state == D_PREP) {
1516 			__punch_discard_cmd(sbi, dc, blkaddr);
1517 		} else {
1518 			dc->ref++;
1519 			need_wait = true;
1520 		}
1521 	}
1522 	mutex_unlock(&dcc->cmd_lock);
1523 
1524 	if (need_wait)
1525 		__wait_one_discard_bio(sbi, dc);
1526 }
1527 
1528 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1529 {
1530 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1531 
1532 	if (dcc && dcc->f2fs_issue_discard) {
1533 		struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1534 
1535 		dcc->f2fs_issue_discard = NULL;
1536 		kthread_stop(discard_thread);
1537 	}
1538 }
1539 
1540 /* This comes from f2fs_put_super */
1541 bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1542 {
1543 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1544 	struct discard_policy dpolicy;
1545 	bool dropped;
1546 
1547 	__init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1548 					dcc->discard_granularity);
1549 	__issue_discard_cmd(sbi, &dpolicy);
1550 	dropped = __drop_discard_cmd(sbi);
1551 
1552 	/* just to make sure there is no pending discard commands */
1553 	__wait_all_discard_cmd(sbi, NULL);
1554 
1555 	f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1556 	return dropped;
1557 }
1558 
1559 static int issue_discard_thread(void *data)
1560 {
1561 	struct f2fs_sb_info *sbi = data;
1562 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1563 	wait_queue_head_t *q = &dcc->discard_wait_queue;
1564 	struct discard_policy dpolicy;
1565 	unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1566 	int issued;
1567 
1568 	set_freezable();
1569 
1570 	do {
1571 		__init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1572 					dcc->discard_granularity);
1573 
1574 		wait_event_interruptible_timeout(*q,
1575 				kthread_should_stop() || freezing(current) ||
1576 				dcc->discard_wake,
1577 				msecs_to_jiffies(wait_ms));
1578 
1579 		if (dcc->discard_wake)
1580 			dcc->discard_wake = 0;
1581 
1582 		if (try_to_freeze())
1583 			continue;
1584 		if (f2fs_readonly(sbi->sb))
1585 			continue;
1586 		if (kthread_should_stop())
1587 			return 0;
1588 		if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1589 			wait_ms = dpolicy.max_interval;
1590 			continue;
1591 		}
1592 
1593 		if (sbi->gc_mode == GC_URGENT)
1594 			__init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1595 
1596 		sb_start_intwrite(sbi->sb);
1597 
1598 		issued = __issue_discard_cmd(sbi, &dpolicy);
1599 		if (issued > 0) {
1600 			__wait_all_discard_cmd(sbi, &dpolicy);
1601 			wait_ms = dpolicy.min_interval;
1602 		} else if (issued == -1){
1603 			wait_ms = dpolicy.mid_interval;
1604 		} else {
1605 			wait_ms = dpolicy.max_interval;
1606 		}
1607 
1608 		sb_end_intwrite(sbi->sb);
1609 
1610 	} while (!kthread_should_stop());
1611 	return 0;
1612 }
1613 
1614 #ifdef CONFIG_BLK_DEV_ZONED
1615 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1616 		struct block_device *bdev, block_t blkstart, block_t blklen)
1617 {
1618 	sector_t sector, nr_sects;
1619 	block_t lblkstart = blkstart;
1620 	int devi = 0;
1621 
1622 	if (sbi->s_ndevs) {
1623 		devi = f2fs_target_device_index(sbi, blkstart);
1624 		blkstart -= FDEV(devi).start_blk;
1625 	}
1626 
1627 	/*
1628 	 * We need to know the type of the zone: for conventional zones,
1629 	 * use regular discard if the drive supports it. For sequential
1630 	 * zones, reset the zone write pointer.
1631 	 */
1632 	switch (get_blkz_type(sbi, bdev, blkstart)) {
1633 
1634 	case BLK_ZONE_TYPE_CONVENTIONAL:
1635 		if (!blk_queue_discard(bdev_get_queue(bdev)))
1636 			return 0;
1637 		return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1638 	case BLK_ZONE_TYPE_SEQWRITE_REQ:
1639 	case BLK_ZONE_TYPE_SEQWRITE_PREF:
1640 		sector = SECTOR_FROM_BLOCK(blkstart);
1641 		nr_sects = SECTOR_FROM_BLOCK(blklen);
1642 
1643 		if (sector & (bdev_zone_sectors(bdev) - 1) ||
1644 				nr_sects != bdev_zone_sectors(bdev)) {
1645 			f2fs_msg(sbi->sb, KERN_INFO,
1646 				"(%d) %s: Unaligned discard attempted (block %x + %x)",
1647 				devi, sbi->s_ndevs ? FDEV(devi).path: "",
1648 				blkstart, blklen);
1649 			return -EIO;
1650 		}
1651 		trace_f2fs_issue_reset_zone(bdev, blkstart);
1652 		return blkdev_reset_zones(bdev, sector,
1653 					  nr_sects, GFP_NOFS);
1654 	default:
1655 		/* Unknown zone type: broken device ? */
1656 		return -EIO;
1657 	}
1658 }
1659 #endif
1660 
1661 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1662 		struct block_device *bdev, block_t blkstart, block_t blklen)
1663 {
1664 #ifdef CONFIG_BLK_DEV_ZONED
1665 	if (f2fs_sb_has_blkzoned(sbi->sb) &&
1666 				bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1667 		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1668 #endif
1669 	return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1670 }
1671 
1672 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1673 				block_t blkstart, block_t blklen)
1674 {
1675 	sector_t start = blkstart, len = 0;
1676 	struct block_device *bdev;
1677 	struct seg_entry *se;
1678 	unsigned int offset;
1679 	block_t i;
1680 	int err = 0;
1681 
1682 	bdev = f2fs_target_device(sbi, blkstart, NULL);
1683 
1684 	for (i = blkstart; i < blkstart + blklen; i++, len++) {
1685 		if (i != start) {
1686 			struct block_device *bdev2 =
1687 				f2fs_target_device(sbi, i, NULL);
1688 
1689 			if (bdev2 != bdev) {
1690 				err = __issue_discard_async(sbi, bdev,
1691 						start, len);
1692 				if (err)
1693 					return err;
1694 				bdev = bdev2;
1695 				start = i;
1696 				len = 0;
1697 			}
1698 		}
1699 
1700 		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1701 		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1702 
1703 		if (!f2fs_test_and_set_bit(offset, se->discard_map))
1704 			sbi->discard_blks--;
1705 	}
1706 
1707 	if (len)
1708 		err = __issue_discard_async(sbi, bdev, start, len);
1709 	return err;
1710 }
1711 
1712 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1713 							bool check_only)
1714 {
1715 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1716 	int max_blocks = sbi->blocks_per_seg;
1717 	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1718 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1719 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1720 	unsigned long *discard_map = (unsigned long *)se->discard_map;
1721 	unsigned long *dmap = SIT_I(sbi)->tmp_map;
1722 	unsigned int start = 0, end = -1;
1723 	bool force = (cpc->reason & CP_DISCARD);
1724 	struct discard_entry *de = NULL;
1725 	struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1726 	int i;
1727 
1728 	if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1729 		return false;
1730 
1731 	if (!force) {
1732 		if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1733 			SM_I(sbi)->dcc_info->nr_discards >=
1734 				SM_I(sbi)->dcc_info->max_discards)
1735 			return false;
1736 	}
1737 
1738 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1739 	for (i = 0; i < entries; i++)
1740 		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1741 				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1742 
1743 	while (force || SM_I(sbi)->dcc_info->nr_discards <=
1744 				SM_I(sbi)->dcc_info->max_discards) {
1745 		start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1746 		if (start >= max_blocks)
1747 			break;
1748 
1749 		end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1750 		if (force && start && end != max_blocks
1751 					&& (end - start) < cpc->trim_minlen)
1752 			continue;
1753 
1754 		if (check_only)
1755 			return true;
1756 
1757 		if (!de) {
1758 			de = f2fs_kmem_cache_alloc(discard_entry_slab,
1759 								GFP_F2FS_ZERO);
1760 			de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1761 			list_add_tail(&de->list, head);
1762 		}
1763 
1764 		for (i = start; i < end; i++)
1765 			__set_bit_le(i, (void *)de->discard_map);
1766 
1767 		SM_I(sbi)->dcc_info->nr_discards += end - start;
1768 	}
1769 	return false;
1770 }
1771 
1772 static void release_discard_addr(struct discard_entry *entry)
1773 {
1774 	list_del(&entry->list);
1775 	kmem_cache_free(discard_entry_slab, entry);
1776 }
1777 
1778 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1779 {
1780 	struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1781 	struct discard_entry *entry, *this;
1782 
1783 	/* drop caches */
1784 	list_for_each_entry_safe(entry, this, head, list)
1785 		release_discard_addr(entry);
1786 }
1787 
1788 /*
1789  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1790  */
1791 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1792 {
1793 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1794 	unsigned int segno;
1795 
1796 	mutex_lock(&dirty_i->seglist_lock);
1797 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1798 		__set_test_and_free(sbi, segno);
1799 	mutex_unlock(&dirty_i->seglist_lock);
1800 }
1801 
1802 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1803 						struct cp_control *cpc)
1804 {
1805 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1806 	struct list_head *head = &dcc->entry_list;
1807 	struct discard_entry *entry, *this;
1808 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1809 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1810 	unsigned int start = 0, end = -1;
1811 	unsigned int secno, start_segno;
1812 	bool force = (cpc->reason & CP_DISCARD);
1813 	bool need_align = test_opt(sbi, LFS) && sbi->segs_per_sec > 1;
1814 
1815 	mutex_lock(&dirty_i->seglist_lock);
1816 
1817 	while (1) {
1818 		int i;
1819 
1820 		if (need_align && end != -1)
1821 			end--;
1822 		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1823 		if (start >= MAIN_SEGS(sbi))
1824 			break;
1825 		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1826 								start + 1);
1827 
1828 		if (need_align) {
1829 			start = rounddown(start, sbi->segs_per_sec);
1830 			end = roundup(end, sbi->segs_per_sec);
1831 		}
1832 
1833 		for (i = start; i < end; i++) {
1834 			if (test_and_clear_bit(i, prefree_map))
1835 				dirty_i->nr_dirty[PRE]--;
1836 		}
1837 
1838 		if (!test_opt(sbi, DISCARD))
1839 			continue;
1840 
1841 		if (force && start >= cpc->trim_start &&
1842 					(end - 1) <= cpc->trim_end)
1843 				continue;
1844 
1845 		if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1846 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1847 				(end - start) << sbi->log_blocks_per_seg);
1848 			continue;
1849 		}
1850 next:
1851 		secno = GET_SEC_FROM_SEG(sbi, start);
1852 		start_segno = GET_SEG_FROM_SEC(sbi, secno);
1853 		if (!IS_CURSEC(sbi, secno) &&
1854 			!get_valid_blocks(sbi, start, true))
1855 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1856 				sbi->segs_per_sec << sbi->log_blocks_per_seg);
1857 
1858 		start = start_segno + sbi->segs_per_sec;
1859 		if (start < end)
1860 			goto next;
1861 		else
1862 			end = start - 1;
1863 	}
1864 	mutex_unlock(&dirty_i->seglist_lock);
1865 
1866 	/* send small discards */
1867 	list_for_each_entry_safe(entry, this, head, list) {
1868 		unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1869 		bool is_valid = test_bit_le(0, entry->discard_map);
1870 
1871 find_next:
1872 		if (is_valid) {
1873 			next_pos = find_next_zero_bit_le(entry->discard_map,
1874 					sbi->blocks_per_seg, cur_pos);
1875 			len = next_pos - cur_pos;
1876 
1877 			if (f2fs_sb_has_blkzoned(sbi->sb) ||
1878 			    (force && len < cpc->trim_minlen))
1879 				goto skip;
1880 
1881 			f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1882 									len);
1883 			total_len += len;
1884 		} else {
1885 			next_pos = find_next_bit_le(entry->discard_map,
1886 					sbi->blocks_per_seg, cur_pos);
1887 		}
1888 skip:
1889 		cur_pos = next_pos;
1890 		is_valid = !is_valid;
1891 
1892 		if (cur_pos < sbi->blocks_per_seg)
1893 			goto find_next;
1894 
1895 		release_discard_addr(entry);
1896 		dcc->nr_discards -= total_len;
1897 	}
1898 
1899 	wake_up_discard_thread(sbi, false);
1900 }
1901 
1902 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1903 {
1904 	dev_t dev = sbi->sb->s_bdev->bd_dev;
1905 	struct discard_cmd_control *dcc;
1906 	int err = 0, i;
1907 
1908 	if (SM_I(sbi)->dcc_info) {
1909 		dcc = SM_I(sbi)->dcc_info;
1910 		goto init_thread;
1911 	}
1912 
1913 	dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
1914 	if (!dcc)
1915 		return -ENOMEM;
1916 
1917 	dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
1918 	INIT_LIST_HEAD(&dcc->entry_list);
1919 	for (i = 0; i < MAX_PLIST_NUM; i++)
1920 		INIT_LIST_HEAD(&dcc->pend_list[i]);
1921 	INIT_LIST_HEAD(&dcc->wait_list);
1922 	INIT_LIST_HEAD(&dcc->fstrim_list);
1923 	mutex_init(&dcc->cmd_lock);
1924 	atomic_set(&dcc->issued_discard, 0);
1925 	atomic_set(&dcc->issing_discard, 0);
1926 	atomic_set(&dcc->discard_cmd_cnt, 0);
1927 	dcc->nr_discards = 0;
1928 	dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1929 	dcc->undiscard_blks = 0;
1930 	dcc->next_pos = 0;
1931 	dcc->root = RB_ROOT;
1932 	dcc->rbtree_check = false;
1933 
1934 	init_waitqueue_head(&dcc->discard_wait_queue);
1935 	SM_I(sbi)->dcc_info = dcc;
1936 init_thread:
1937 	dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1938 				"f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1939 	if (IS_ERR(dcc->f2fs_issue_discard)) {
1940 		err = PTR_ERR(dcc->f2fs_issue_discard);
1941 		kfree(dcc);
1942 		SM_I(sbi)->dcc_info = NULL;
1943 		return err;
1944 	}
1945 
1946 	return err;
1947 }
1948 
1949 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1950 {
1951 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1952 
1953 	if (!dcc)
1954 		return;
1955 
1956 	f2fs_stop_discard_thread(sbi);
1957 
1958 	kfree(dcc);
1959 	SM_I(sbi)->dcc_info = NULL;
1960 }
1961 
1962 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1963 {
1964 	struct sit_info *sit_i = SIT_I(sbi);
1965 
1966 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1967 		sit_i->dirty_sentries++;
1968 		return false;
1969 	}
1970 
1971 	return true;
1972 }
1973 
1974 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1975 					unsigned int segno, int modified)
1976 {
1977 	struct seg_entry *se = get_seg_entry(sbi, segno);
1978 	se->type = type;
1979 	if (modified)
1980 		__mark_sit_entry_dirty(sbi, segno);
1981 }
1982 
1983 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1984 {
1985 	struct seg_entry *se;
1986 	unsigned int segno, offset;
1987 	long int new_vblocks;
1988 	bool exist;
1989 #ifdef CONFIG_F2FS_CHECK_FS
1990 	bool mir_exist;
1991 #endif
1992 
1993 	segno = GET_SEGNO(sbi, blkaddr);
1994 
1995 	se = get_seg_entry(sbi, segno);
1996 	new_vblocks = se->valid_blocks + del;
1997 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1998 
1999 	f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
2000 				(new_vblocks > sbi->blocks_per_seg)));
2001 
2002 	se->valid_blocks = new_vblocks;
2003 	se->mtime = get_mtime(sbi, false);
2004 	if (se->mtime > SIT_I(sbi)->max_mtime)
2005 		SIT_I(sbi)->max_mtime = se->mtime;
2006 
2007 	/* Update valid block bitmap */
2008 	if (del > 0) {
2009 		exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2010 #ifdef CONFIG_F2FS_CHECK_FS
2011 		mir_exist = f2fs_test_and_set_bit(offset,
2012 						se->cur_valid_map_mir);
2013 		if (unlikely(exist != mir_exist)) {
2014 			f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2015 				"when setting bitmap, blk:%u, old bit:%d",
2016 				blkaddr, exist);
2017 			f2fs_bug_on(sbi, 1);
2018 		}
2019 #endif
2020 		if (unlikely(exist)) {
2021 			f2fs_msg(sbi->sb, KERN_ERR,
2022 				"Bitmap was wrongly set, blk:%u", blkaddr);
2023 			f2fs_bug_on(sbi, 1);
2024 			se->valid_blocks--;
2025 			del = 0;
2026 		}
2027 
2028 		if (f2fs_discard_en(sbi) &&
2029 			!f2fs_test_and_set_bit(offset, se->discard_map))
2030 			sbi->discard_blks--;
2031 
2032 		/* don't overwrite by SSR to keep node chain */
2033 		if (IS_NODESEG(se->type)) {
2034 			if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2035 				se->ckpt_valid_blocks++;
2036 		}
2037 	} else {
2038 		exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2039 #ifdef CONFIG_F2FS_CHECK_FS
2040 		mir_exist = f2fs_test_and_clear_bit(offset,
2041 						se->cur_valid_map_mir);
2042 		if (unlikely(exist != mir_exist)) {
2043 			f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2044 				"when clearing bitmap, blk:%u, old bit:%d",
2045 				blkaddr, exist);
2046 			f2fs_bug_on(sbi, 1);
2047 		}
2048 #endif
2049 		if (unlikely(!exist)) {
2050 			f2fs_msg(sbi->sb, KERN_ERR,
2051 				"Bitmap was wrongly cleared, blk:%u", blkaddr);
2052 			f2fs_bug_on(sbi, 1);
2053 			se->valid_blocks++;
2054 			del = 0;
2055 		}
2056 
2057 		if (f2fs_discard_en(sbi) &&
2058 			f2fs_test_and_clear_bit(offset, se->discard_map))
2059 			sbi->discard_blks++;
2060 	}
2061 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2062 		se->ckpt_valid_blocks += del;
2063 
2064 	__mark_sit_entry_dirty(sbi, segno);
2065 
2066 	/* update total number of valid blocks to be written in ckpt area */
2067 	SIT_I(sbi)->written_valid_blocks += del;
2068 
2069 	if (sbi->segs_per_sec > 1)
2070 		get_sec_entry(sbi, segno)->valid_blocks += del;
2071 }
2072 
2073 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2074 {
2075 	unsigned int segno = GET_SEGNO(sbi, addr);
2076 	struct sit_info *sit_i = SIT_I(sbi);
2077 
2078 	f2fs_bug_on(sbi, addr == NULL_ADDR);
2079 	if (addr == NEW_ADDR)
2080 		return;
2081 
2082 	invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2083 
2084 	/* add it into sit main buffer */
2085 	down_write(&sit_i->sentry_lock);
2086 
2087 	update_sit_entry(sbi, addr, -1);
2088 
2089 	/* add it into dirty seglist */
2090 	locate_dirty_segment(sbi, segno);
2091 
2092 	up_write(&sit_i->sentry_lock);
2093 }
2094 
2095 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2096 {
2097 	struct sit_info *sit_i = SIT_I(sbi);
2098 	unsigned int segno, offset;
2099 	struct seg_entry *se;
2100 	bool is_cp = false;
2101 
2102 	if (!is_valid_data_blkaddr(sbi, blkaddr))
2103 		return true;
2104 
2105 	down_read(&sit_i->sentry_lock);
2106 
2107 	segno = GET_SEGNO(sbi, blkaddr);
2108 	se = get_seg_entry(sbi, segno);
2109 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2110 
2111 	if (f2fs_test_bit(offset, se->ckpt_valid_map))
2112 		is_cp = true;
2113 
2114 	up_read(&sit_i->sentry_lock);
2115 
2116 	return is_cp;
2117 }
2118 
2119 /*
2120  * This function should be resided under the curseg_mutex lock
2121  */
2122 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2123 					struct f2fs_summary *sum)
2124 {
2125 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2126 	void *addr = curseg->sum_blk;
2127 	addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2128 	memcpy(addr, sum, sizeof(struct f2fs_summary));
2129 }
2130 
2131 /*
2132  * Calculate the number of current summary pages for writing
2133  */
2134 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2135 {
2136 	int valid_sum_count = 0;
2137 	int i, sum_in_page;
2138 
2139 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2140 		if (sbi->ckpt->alloc_type[i] == SSR)
2141 			valid_sum_count += sbi->blocks_per_seg;
2142 		else {
2143 			if (for_ra)
2144 				valid_sum_count += le16_to_cpu(
2145 					F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2146 			else
2147 				valid_sum_count += curseg_blkoff(sbi, i);
2148 		}
2149 	}
2150 
2151 	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2152 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2153 	if (valid_sum_count <= sum_in_page)
2154 		return 1;
2155 	else if ((valid_sum_count - sum_in_page) <=
2156 		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2157 		return 2;
2158 	return 3;
2159 }
2160 
2161 /*
2162  * Caller should put this summary page
2163  */
2164 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2165 {
2166 	return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2167 }
2168 
2169 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2170 					void *src, block_t blk_addr)
2171 {
2172 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2173 
2174 	memcpy(page_address(page), src, PAGE_SIZE);
2175 	set_page_dirty(page);
2176 	f2fs_put_page(page, 1);
2177 }
2178 
2179 static void write_sum_page(struct f2fs_sb_info *sbi,
2180 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
2181 {
2182 	f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2183 }
2184 
2185 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2186 						int type, block_t blk_addr)
2187 {
2188 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2189 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2190 	struct f2fs_summary_block *src = curseg->sum_blk;
2191 	struct f2fs_summary_block *dst;
2192 
2193 	dst = (struct f2fs_summary_block *)page_address(page);
2194 	memset(dst, 0, PAGE_SIZE);
2195 
2196 	mutex_lock(&curseg->curseg_mutex);
2197 
2198 	down_read(&curseg->journal_rwsem);
2199 	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2200 	up_read(&curseg->journal_rwsem);
2201 
2202 	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2203 	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2204 
2205 	mutex_unlock(&curseg->curseg_mutex);
2206 
2207 	set_page_dirty(page);
2208 	f2fs_put_page(page, 1);
2209 }
2210 
2211 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2212 {
2213 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2214 	unsigned int segno = curseg->segno + 1;
2215 	struct free_segmap_info *free_i = FREE_I(sbi);
2216 
2217 	if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2218 		return !test_bit(segno, free_i->free_segmap);
2219 	return 0;
2220 }
2221 
2222 /*
2223  * Find a new segment from the free segments bitmap to right order
2224  * This function should be returned with success, otherwise BUG
2225  */
2226 static void get_new_segment(struct f2fs_sb_info *sbi,
2227 			unsigned int *newseg, bool new_sec, int dir)
2228 {
2229 	struct free_segmap_info *free_i = FREE_I(sbi);
2230 	unsigned int segno, secno, zoneno;
2231 	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2232 	unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2233 	unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2234 	unsigned int left_start = hint;
2235 	bool init = true;
2236 	int go_left = 0;
2237 	int i;
2238 
2239 	spin_lock(&free_i->segmap_lock);
2240 
2241 	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2242 		segno = find_next_zero_bit(free_i->free_segmap,
2243 			GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2244 		if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2245 			goto got_it;
2246 	}
2247 find_other_zone:
2248 	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2249 	if (secno >= MAIN_SECS(sbi)) {
2250 		if (dir == ALLOC_RIGHT) {
2251 			secno = find_next_zero_bit(free_i->free_secmap,
2252 							MAIN_SECS(sbi), 0);
2253 			f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2254 		} else {
2255 			go_left = 1;
2256 			left_start = hint - 1;
2257 		}
2258 	}
2259 	if (go_left == 0)
2260 		goto skip_left;
2261 
2262 	while (test_bit(left_start, free_i->free_secmap)) {
2263 		if (left_start > 0) {
2264 			left_start--;
2265 			continue;
2266 		}
2267 		left_start = find_next_zero_bit(free_i->free_secmap,
2268 							MAIN_SECS(sbi), 0);
2269 		f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2270 		break;
2271 	}
2272 	secno = left_start;
2273 skip_left:
2274 	segno = GET_SEG_FROM_SEC(sbi, secno);
2275 	zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2276 
2277 	/* give up on finding another zone */
2278 	if (!init)
2279 		goto got_it;
2280 	if (sbi->secs_per_zone == 1)
2281 		goto got_it;
2282 	if (zoneno == old_zoneno)
2283 		goto got_it;
2284 	if (dir == ALLOC_LEFT) {
2285 		if (!go_left && zoneno + 1 >= total_zones)
2286 			goto got_it;
2287 		if (go_left && zoneno == 0)
2288 			goto got_it;
2289 	}
2290 	for (i = 0; i < NR_CURSEG_TYPE; i++)
2291 		if (CURSEG_I(sbi, i)->zone == zoneno)
2292 			break;
2293 
2294 	if (i < NR_CURSEG_TYPE) {
2295 		/* zone is in user, try another */
2296 		if (go_left)
2297 			hint = zoneno * sbi->secs_per_zone - 1;
2298 		else if (zoneno + 1 >= total_zones)
2299 			hint = 0;
2300 		else
2301 			hint = (zoneno + 1) * sbi->secs_per_zone;
2302 		init = false;
2303 		goto find_other_zone;
2304 	}
2305 got_it:
2306 	/* set it as dirty segment in free segmap */
2307 	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2308 	__set_inuse(sbi, segno);
2309 	*newseg = segno;
2310 	spin_unlock(&free_i->segmap_lock);
2311 }
2312 
2313 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2314 {
2315 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2316 	struct summary_footer *sum_footer;
2317 
2318 	curseg->segno = curseg->next_segno;
2319 	curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2320 	curseg->next_blkoff = 0;
2321 	curseg->next_segno = NULL_SEGNO;
2322 
2323 	sum_footer = &(curseg->sum_blk->footer);
2324 	memset(sum_footer, 0, sizeof(struct summary_footer));
2325 	if (IS_DATASEG(type))
2326 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2327 	if (IS_NODESEG(type))
2328 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2329 	__set_sit_entry_type(sbi, type, curseg->segno, modified);
2330 }
2331 
2332 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2333 {
2334 	/* if segs_per_sec is large than 1, we need to keep original policy. */
2335 	if (sbi->segs_per_sec != 1)
2336 		return CURSEG_I(sbi, type)->segno;
2337 
2338 	if (test_opt(sbi, NOHEAP) &&
2339 		(type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2340 		return 0;
2341 
2342 	if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2343 		return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2344 
2345 	/* find segments from 0 to reuse freed segments */
2346 	if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2347 		return 0;
2348 
2349 	return CURSEG_I(sbi, type)->segno;
2350 }
2351 
2352 /*
2353  * Allocate a current working segment.
2354  * This function always allocates a free segment in LFS manner.
2355  */
2356 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2357 {
2358 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2359 	unsigned int segno = curseg->segno;
2360 	int dir = ALLOC_LEFT;
2361 
2362 	write_sum_page(sbi, curseg->sum_blk,
2363 				GET_SUM_BLOCK(sbi, segno));
2364 	if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2365 		dir = ALLOC_RIGHT;
2366 
2367 	if (test_opt(sbi, NOHEAP))
2368 		dir = ALLOC_RIGHT;
2369 
2370 	segno = __get_next_segno(sbi, type);
2371 	get_new_segment(sbi, &segno, new_sec, dir);
2372 	curseg->next_segno = segno;
2373 	reset_curseg(sbi, type, 1);
2374 	curseg->alloc_type = LFS;
2375 }
2376 
2377 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2378 			struct curseg_info *seg, block_t start)
2379 {
2380 	struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2381 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2382 	unsigned long *target_map = SIT_I(sbi)->tmp_map;
2383 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2384 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2385 	int i, pos;
2386 
2387 	for (i = 0; i < entries; i++)
2388 		target_map[i] = ckpt_map[i] | cur_map[i];
2389 
2390 	pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2391 
2392 	seg->next_blkoff = pos;
2393 }
2394 
2395 /*
2396  * If a segment is written by LFS manner, next block offset is just obtained
2397  * by increasing the current block offset. However, if a segment is written by
2398  * SSR manner, next block offset obtained by calling __next_free_blkoff
2399  */
2400 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2401 				struct curseg_info *seg)
2402 {
2403 	if (seg->alloc_type == SSR)
2404 		__next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2405 	else
2406 		seg->next_blkoff++;
2407 }
2408 
2409 /*
2410  * This function always allocates a used segment(from dirty seglist) by SSR
2411  * manner, so it should recover the existing segment information of valid blocks
2412  */
2413 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2414 {
2415 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2416 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2417 	unsigned int new_segno = curseg->next_segno;
2418 	struct f2fs_summary_block *sum_node;
2419 	struct page *sum_page;
2420 
2421 	write_sum_page(sbi, curseg->sum_blk,
2422 				GET_SUM_BLOCK(sbi, curseg->segno));
2423 	__set_test_and_inuse(sbi, new_segno);
2424 
2425 	mutex_lock(&dirty_i->seglist_lock);
2426 	__remove_dirty_segment(sbi, new_segno, PRE);
2427 	__remove_dirty_segment(sbi, new_segno, DIRTY);
2428 	mutex_unlock(&dirty_i->seglist_lock);
2429 
2430 	reset_curseg(sbi, type, 1);
2431 	curseg->alloc_type = SSR;
2432 	__next_free_blkoff(sbi, curseg, 0);
2433 
2434 	sum_page = f2fs_get_sum_page(sbi, new_segno);
2435 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2436 	memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2437 	f2fs_put_page(sum_page, 1);
2438 }
2439 
2440 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2441 {
2442 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2443 	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2444 	unsigned segno = NULL_SEGNO;
2445 	int i, cnt;
2446 	bool reversed = false;
2447 
2448 	/* f2fs_need_SSR() already forces to do this */
2449 	if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2450 		curseg->next_segno = segno;
2451 		return 1;
2452 	}
2453 
2454 	/* For node segments, let's do SSR more intensively */
2455 	if (IS_NODESEG(type)) {
2456 		if (type >= CURSEG_WARM_NODE) {
2457 			reversed = true;
2458 			i = CURSEG_COLD_NODE;
2459 		} else {
2460 			i = CURSEG_HOT_NODE;
2461 		}
2462 		cnt = NR_CURSEG_NODE_TYPE;
2463 	} else {
2464 		if (type >= CURSEG_WARM_DATA) {
2465 			reversed = true;
2466 			i = CURSEG_COLD_DATA;
2467 		} else {
2468 			i = CURSEG_HOT_DATA;
2469 		}
2470 		cnt = NR_CURSEG_DATA_TYPE;
2471 	}
2472 
2473 	for (; cnt-- > 0; reversed ? i-- : i++) {
2474 		if (i == type)
2475 			continue;
2476 		if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2477 			curseg->next_segno = segno;
2478 			return 1;
2479 		}
2480 	}
2481 	return 0;
2482 }
2483 
2484 /*
2485  * flush out current segment and replace it with new segment
2486  * This function should be returned with success, otherwise BUG
2487  */
2488 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2489 						int type, bool force)
2490 {
2491 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2492 
2493 	if (force)
2494 		new_curseg(sbi, type, true);
2495 	else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2496 					type == CURSEG_WARM_NODE)
2497 		new_curseg(sbi, type, false);
2498 	else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2499 		new_curseg(sbi, type, false);
2500 	else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2501 		change_curseg(sbi, type);
2502 	else
2503 		new_curseg(sbi, type, false);
2504 
2505 	stat_inc_seg_type(sbi, curseg);
2506 }
2507 
2508 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2509 {
2510 	struct curseg_info *curseg;
2511 	unsigned int old_segno;
2512 	int i;
2513 
2514 	down_write(&SIT_I(sbi)->sentry_lock);
2515 
2516 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2517 		curseg = CURSEG_I(sbi, i);
2518 		old_segno = curseg->segno;
2519 		SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2520 		locate_dirty_segment(sbi, old_segno);
2521 	}
2522 
2523 	up_write(&SIT_I(sbi)->sentry_lock);
2524 }
2525 
2526 static const struct segment_allocation default_salloc_ops = {
2527 	.allocate_segment = allocate_segment_by_default,
2528 };
2529 
2530 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2531 						struct cp_control *cpc)
2532 {
2533 	__u64 trim_start = cpc->trim_start;
2534 	bool has_candidate = false;
2535 
2536 	down_write(&SIT_I(sbi)->sentry_lock);
2537 	for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2538 		if (add_discard_addrs(sbi, cpc, true)) {
2539 			has_candidate = true;
2540 			break;
2541 		}
2542 	}
2543 	up_write(&SIT_I(sbi)->sentry_lock);
2544 
2545 	cpc->trim_start = trim_start;
2546 	return has_candidate;
2547 }
2548 
2549 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2550 					struct discard_policy *dpolicy,
2551 					unsigned int start, unsigned int end)
2552 {
2553 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2554 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2555 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
2556 	struct discard_cmd *dc;
2557 	struct blk_plug plug;
2558 	int issued;
2559 	unsigned int trimmed = 0;
2560 
2561 next:
2562 	issued = 0;
2563 
2564 	mutex_lock(&dcc->cmd_lock);
2565 	if (unlikely(dcc->rbtree_check))
2566 		f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2567 								&dcc->root));
2568 
2569 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2570 					NULL, start,
2571 					(struct rb_entry **)&prev_dc,
2572 					(struct rb_entry **)&next_dc,
2573 					&insert_p, &insert_parent, true);
2574 	if (!dc)
2575 		dc = next_dc;
2576 
2577 	blk_start_plug(&plug);
2578 
2579 	while (dc && dc->lstart <= end) {
2580 		struct rb_node *node;
2581 		int err = 0;
2582 
2583 		if (dc->len < dpolicy->granularity)
2584 			goto skip;
2585 
2586 		if (dc->state != D_PREP) {
2587 			list_move_tail(&dc->list, &dcc->fstrim_list);
2588 			goto skip;
2589 		}
2590 
2591 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2592 
2593 		if (issued >= dpolicy->max_requests) {
2594 			start = dc->lstart + dc->len;
2595 
2596 			if (err)
2597 				__remove_discard_cmd(sbi, dc);
2598 
2599 			blk_finish_plug(&plug);
2600 			mutex_unlock(&dcc->cmd_lock);
2601 			trimmed += __wait_all_discard_cmd(sbi, NULL);
2602 			congestion_wait(BLK_RW_ASYNC, HZ/50);
2603 			goto next;
2604 		}
2605 skip:
2606 		node = rb_next(&dc->rb_node);
2607 		if (err)
2608 			__remove_discard_cmd(sbi, dc);
2609 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2610 
2611 		if (fatal_signal_pending(current))
2612 			break;
2613 	}
2614 
2615 	blk_finish_plug(&plug);
2616 	mutex_unlock(&dcc->cmd_lock);
2617 
2618 	return trimmed;
2619 }
2620 
2621 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2622 {
2623 	__u64 start = F2FS_BYTES_TO_BLK(range->start);
2624 	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2625 	unsigned int start_segno, end_segno;
2626 	block_t start_block, end_block;
2627 	struct cp_control cpc;
2628 	struct discard_policy dpolicy;
2629 	unsigned long long trimmed = 0;
2630 	int err = 0;
2631 	bool need_align = test_opt(sbi, LFS) && sbi->segs_per_sec > 1;
2632 
2633 	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2634 		return -EINVAL;
2635 
2636 	if (end < MAIN_BLKADDR(sbi))
2637 		goto out;
2638 
2639 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2640 		f2fs_msg(sbi->sb, KERN_WARNING,
2641 			"Found FS corruption, run fsck to fix.");
2642 		return -EIO;
2643 	}
2644 
2645 	/* start/end segment number in main_area */
2646 	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2647 	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2648 						GET_SEGNO(sbi, end);
2649 	if (need_align) {
2650 		start_segno = rounddown(start_segno, sbi->segs_per_sec);
2651 		end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2652 	}
2653 
2654 	cpc.reason = CP_DISCARD;
2655 	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2656 	cpc.trim_start = start_segno;
2657 	cpc.trim_end = end_segno;
2658 
2659 	if (sbi->discard_blks == 0)
2660 		goto out;
2661 
2662 	mutex_lock(&sbi->gc_mutex);
2663 	err = f2fs_write_checkpoint(sbi, &cpc);
2664 	mutex_unlock(&sbi->gc_mutex);
2665 	if (err)
2666 		goto out;
2667 
2668 	/*
2669 	 * We filed discard candidates, but actually we don't need to wait for
2670 	 * all of them, since they'll be issued in idle time along with runtime
2671 	 * discard option. User configuration looks like using runtime discard
2672 	 * or periodic fstrim instead of it.
2673 	 */
2674 	if (test_opt(sbi, DISCARD))
2675 		goto out;
2676 
2677 	start_block = START_BLOCK(sbi, start_segno);
2678 	end_block = START_BLOCK(sbi, end_segno + 1);
2679 
2680 	__init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2681 	trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2682 					start_block, end_block);
2683 
2684 	trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2685 					start_block, end_block);
2686 out:
2687 	if (!err)
2688 		range->len = F2FS_BLK_TO_BYTES(trimmed);
2689 	return err;
2690 }
2691 
2692 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2693 {
2694 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2695 	if (curseg->next_blkoff < sbi->blocks_per_seg)
2696 		return true;
2697 	return false;
2698 }
2699 
2700 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2701 {
2702 	switch (hint) {
2703 	case WRITE_LIFE_SHORT:
2704 		return CURSEG_HOT_DATA;
2705 	case WRITE_LIFE_EXTREME:
2706 		return CURSEG_COLD_DATA;
2707 	default:
2708 		return CURSEG_WARM_DATA;
2709 	}
2710 }
2711 
2712 /* This returns write hints for each segment type. This hints will be
2713  * passed down to block layer. There are mapping tables which depend on
2714  * the mount option 'whint_mode'.
2715  *
2716  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2717  *
2718  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2719  *
2720  * User                  F2FS                     Block
2721  * ----                  ----                     -----
2722  *                       META                     WRITE_LIFE_NOT_SET
2723  *                       HOT_NODE                 "
2724  *                       WARM_NODE                "
2725  *                       COLD_NODE                "
2726  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2727  * extension list        "                        "
2728  *
2729  * -- buffered io
2730  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2731  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2732  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2733  * WRITE_LIFE_NONE       "                        "
2734  * WRITE_LIFE_MEDIUM     "                        "
2735  * WRITE_LIFE_LONG       "                        "
2736  *
2737  * -- direct io
2738  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2739  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2740  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2741  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2742  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2743  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2744  *
2745  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2746  *
2747  * User                  F2FS                     Block
2748  * ----                  ----                     -----
2749  *                       META                     WRITE_LIFE_MEDIUM;
2750  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2751  *                       WARM_NODE                "
2752  *                       COLD_NODE                WRITE_LIFE_NONE
2753  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2754  * extension list        "                        "
2755  *
2756  * -- buffered io
2757  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2758  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2759  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2760  * WRITE_LIFE_NONE       "                        "
2761  * WRITE_LIFE_MEDIUM     "                        "
2762  * WRITE_LIFE_LONG       "                        "
2763  *
2764  * -- direct io
2765  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2766  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2767  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2768  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2769  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2770  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2771  */
2772 
2773 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2774 				enum page_type type, enum temp_type temp)
2775 {
2776 	if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2777 		if (type == DATA) {
2778 			if (temp == WARM)
2779 				return WRITE_LIFE_NOT_SET;
2780 			else if (temp == HOT)
2781 				return WRITE_LIFE_SHORT;
2782 			else if (temp == COLD)
2783 				return WRITE_LIFE_EXTREME;
2784 		} else {
2785 			return WRITE_LIFE_NOT_SET;
2786 		}
2787 	} else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2788 		if (type == DATA) {
2789 			if (temp == WARM)
2790 				return WRITE_LIFE_LONG;
2791 			else if (temp == HOT)
2792 				return WRITE_LIFE_SHORT;
2793 			else if (temp == COLD)
2794 				return WRITE_LIFE_EXTREME;
2795 		} else if (type == NODE) {
2796 			if (temp == WARM || temp == HOT)
2797 				return WRITE_LIFE_NOT_SET;
2798 			else if (temp == COLD)
2799 				return WRITE_LIFE_NONE;
2800 		} else if (type == META) {
2801 			return WRITE_LIFE_MEDIUM;
2802 		}
2803 	}
2804 	return WRITE_LIFE_NOT_SET;
2805 }
2806 
2807 static int __get_segment_type_2(struct f2fs_io_info *fio)
2808 {
2809 	if (fio->type == DATA)
2810 		return CURSEG_HOT_DATA;
2811 	else
2812 		return CURSEG_HOT_NODE;
2813 }
2814 
2815 static int __get_segment_type_4(struct f2fs_io_info *fio)
2816 {
2817 	if (fio->type == DATA) {
2818 		struct inode *inode = fio->page->mapping->host;
2819 
2820 		if (S_ISDIR(inode->i_mode))
2821 			return CURSEG_HOT_DATA;
2822 		else
2823 			return CURSEG_COLD_DATA;
2824 	} else {
2825 		if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2826 			return CURSEG_WARM_NODE;
2827 		else
2828 			return CURSEG_COLD_NODE;
2829 	}
2830 }
2831 
2832 static int __get_segment_type_6(struct f2fs_io_info *fio)
2833 {
2834 	if (fio->type == DATA) {
2835 		struct inode *inode = fio->page->mapping->host;
2836 
2837 		if (is_cold_data(fio->page) || file_is_cold(inode))
2838 			return CURSEG_COLD_DATA;
2839 		if (file_is_hot(inode) ||
2840 				is_inode_flag_set(inode, FI_HOT_DATA) ||
2841 				f2fs_is_atomic_file(inode) ||
2842 				f2fs_is_volatile_file(inode))
2843 			return CURSEG_HOT_DATA;
2844 		return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
2845 	} else {
2846 		if (IS_DNODE(fio->page))
2847 			return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2848 						CURSEG_HOT_NODE;
2849 		return CURSEG_COLD_NODE;
2850 	}
2851 }
2852 
2853 static int __get_segment_type(struct f2fs_io_info *fio)
2854 {
2855 	int type = 0;
2856 
2857 	switch (F2FS_OPTION(fio->sbi).active_logs) {
2858 	case 2:
2859 		type = __get_segment_type_2(fio);
2860 		break;
2861 	case 4:
2862 		type = __get_segment_type_4(fio);
2863 		break;
2864 	case 6:
2865 		type = __get_segment_type_6(fio);
2866 		break;
2867 	default:
2868 		f2fs_bug_on(fio->sbi, true);
2869 	}
2870 
2871 	if (IS_HOT(type))
2872 		fio->temp = HOT;
2873 	else if (IS_WARM(type))
2874 		fio->temp = WARM;
2875 	else
2876 		fio->temp = COLD;
2877 	return type;
2878 }
2879 
2880 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2881 		block_t old_blkaddr, block_t *new_blkaddr,
2882 		struct f2fs_summary *sum, int type,
2883 		struct f2fs_io_info *fio, bool add_list)
2884 {
2885 	struct sit_info *sit_i = SIT_I(sbi);
2886 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2887 
2888 	down_read(&SM_I(sbi)->curseg_lock);
2889 
2890 	mutex_lock(&curseg->curseg_mutex);
2891 	down_write(&sit_i->sentry_lock);
2892 
2893 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2894 
2895 	f2fs_wait_discard_bio(sbi, *new_blkaddr);
2896 
2897 	/*
2898 	 * __add_sum_entry should be resided under the curseg_mutex
2899 	 * because, this function updates a summary entry in the
2900 	 * current summary block.
2901 	 */
2902 	__add_sum_entry(sbi, type, sum);
2903 
2904 	__refresh_next_blkoff(sbi, curseg);
2905 
2906 	stat_inc_block_count(sbi, curseg);
2907 
2908 	/*
2909 	 * SIT information should be updated before segment allocation,
2910 	 * since SSR needs latest valid block information.
2911 	 */
2912 	update_sit_entry(sbi, *new_blkaddr, 1);
2913 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2914 		update_sit_entry(sbi, old_blkaddr, -1);
2915 
2916 	if (!__has_curseg_space(sbi, type))
2917 		sit_i->s_ops->allocate_segment(sbi, type, false);
2918 
2919 	/*
2920 	 * segment dirty status should be updated after segment allocation,
2921 	 * so we just need to update status only one time after previous
2922 	 * segment being closed.
2923 	 */
2924 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2925 	locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
2926 
2927 	up_write(&sit_i->sentry_lock);
2928 
2929 	if (page && IS_NODESEG(type)) {
2930 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2931 
2932 		f2fs_inode_chksum_set(sbi, page);
2933 	}
2934 
2935 	if (add_list) {
2936 		struct f2fs_bio_info *io;
2937 
2938 		INIT_LIST_HEAD(&fio->list);
2939 		fio->in_list = true;
2940 		fio->retry = false;
2941 		io = sbi->write_io[fio->type] + fio->temp;
2942 		spin_lock(&io->io_lock);
2943 		list_add_tail(&fio->list, &io->io_list);
2944 		spin_unlock(&io->io_lock);
2945 	}
2946 
2947 	mutex_unlock(&curseg->curseg_mutex);
2948 
2949 	up_read(&SM_I(sbi)->curseg_lock);
2950 }
2951 
2952 static void update_device_state(struct f2fs_io_info *fio)
2953 {
2954 	struct f2fs_sb_info *sbi = fio->sbi;
2955 	unsigned int devidx;
2956 
2957 	if (!sbi->s_ndevs)
2958 		return;
2959 
2960 	devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2961 
2962 	/* update device state for fsync */
2963 	f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
2964 
2965 	/* update device state for checkpoint */
2966 	if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2967 		spin_lock(&sbi->dev_lock);
2968 		f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2969 		spin_unlock(&sbi->dev_lock);
2970 	}
2971 }
2972 
2973 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2974 {
2975 	int type = __get_segment_type(fio);
2976 	bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
2977 
2978 	if (keep_order)
2979 		down_read(&fio->sbi->io_order_lock);
2980 reallocate:
2981 	f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2982 			&fio->new_blkaddr, sum, type, fio, true);
2983 	if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
2984 		invalidate_mapping_pages(META_MAPPING(fio->sbi),
2985 					fio->old_blkaddr, fio->old_blkaddr);
2986 
2987 	/* writeout dirty page into bdev */
2988 	f2fs_submit_page_write(fio);
2989 	if (fio->retry) {
2990 		fio->old_blkaddr = fio->new_blkaddr;
2991 		goto reallocate;
2992 	}
2993 
2994 	update_device_state(fio);
2995 
2996 	if (keep_order)
2997 		up_read(&fio->sbi->io_order_lock);
2998 }
2999 
3000 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3001 					enum iostat_type io_type)
3002 {
3003 	struct f2fs_io_info fio = {
3004 		.sbi = sbi,
3005 		.type = META,
3006 		.temp = HOT,
3007 		.op = REQ_OP_WRITE,
3008 		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3009 		.old_blkaddr = page->index,
3010 		.new_blkaddr = page->index,
3011 		.page = page,
3012 		.encrypted_page = NULL,
3013 		.in_list = false,
3014 	};
3015 
3016 	if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3017 		fio.op_flags &= ~REQ_META;
3018 
3019 	set_page_writeback(page);
3020 	ClearPageError(page);
3021 	f2fs_submit_page_write(&fio);
3022 
3023 	f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3024 }
3025 
3026 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3027 {
3028 	struct f2fs_summary sum;
3029 
3030 	set_summary(&sum, nid, 0, 0);
3031 	do_write_page(&sum, fio);
3032 
3033 	f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3034 }
3035 
3036 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3037 					struct f2fs_io_info *fio)
3038 {
3039 	struct f2fs_sb_info *sbi = fio->sbi;
3040 	struct f2fs_summary sum;
3041 
3042 	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3043 	set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3044 	do_write_page(&sum, fio);
3045 	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3046 
3047 	f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3048 }
3049 
3050 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3051 {
3052 	int err;
3053 	struct f2fs_sb_info *sbi = fio->sbi;
3054 
3055 	fio->new_blkaddr = fio->old_blkaddr;
3056 	/* i/o temperature is needed for passing down write hints */
3057 	__get_segment_type(fio);
3058 
3059 	f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
3060 			GET_SEGNO(sbi, fio->new_blkaddr))->type));
3061 
3062 	stat_inc_inplace_blocks(fio->sbi);
3063 
3064 	err = f2fs_submit_page_bio(fio);
3065 	if (!err)
3066 		update_device_state(fio);
3067 
3068 	f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3069 
3070 	return err;
3071 }
3072 
3073 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3074 						unsigned int segno)
3075 {
3076 	int i;
3077 
3078 	for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3079 		if (CURSEG_I(sbi, i)->segno == segno)
3080 			break;
3081 	}
3082 	return i;
3083 }
3084 
3085 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3086 				block_t old_blkaddr, block_t new_blkaddr,
3087 				bool recover_curseg, bool recover_newaddr)
3088 {
3089 	struct sit_info *sit_i = SIT_I(sbi);
3090 	struct curseg_info *curseg;
3091 	unsigned int segno, old_cursegno;
3092 	struct seg_entry *se;
3093 	int type;
3094 	unsigned short old_blkoff;
3095 
3096 	segno = GET_SEGNO(sbi, new_blkaddr);
3097 	se = get_seg_entry(sbi, segno);
3098 	type = se->type;
3099 
3100 	down_write(&SM_I(sbi)->curseg_lock);
3101 
3102 	if (!recover_curseg) {
3103 		/* for recovery flow */
3104 		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3105 			if (old_blkaddr == NULL_ADDR)
3106 				type = CURSEG_COLD_DATA;
3107 			else
3108 				type = CURSEG_WARM_DATA;
3109 		}
3110 	} else {
3111 		if (IS_CURSEG(sbi, segno)) {
3112 			/* se->type is volatile as SSR allocation */
3113 			type = __f2fs_get_curseg(sbi, segno);
3114 			f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3115 		} else {
3116 			type = CURSEG_WARM_DATA;
3117 		}
3118 	}
3119 
3120 	f2fs_bug_on(sbi, !IS_DATASEG(type));
3121 	curseg = CURSEG_I(sbi, type);
3122 
3123 	mutex_lock(&curseg->curseg_mutex);
3124 	down_write(&sit_i->sentry_lock);
3125 
3126 	old_cursegno = curseg->segno;
3127 	old_blkoff = curseg->next_blkoff;
3128 
3129 	/* change the current segment */
3130 	if (segno != curseg->segno) {
3131 		curseg->next_segno = segno;
3132 		change_curseg(sbi, type);
3133 	}
3134 
3135 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3136 	__add_sum_entry(sbi, type, sum);
3137 
3138 	if (!recover_curseg || recover_newaddr)
3139 		update_sit_entry(sbi, new_blkaddr, 1);
3140 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3141 		invalidate_mapping_pages(META_MAPPING(sbi),
3142 					old_blkaddr, old_blkaddr);
3143 		update_sit_entry(sbi, old_blkaddr, -1);
3144 	}
3145 
3146 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3147 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3148 
3149 	locate_dirty_segment(sbi, old_cursegno);
3150 
3151 	if (recover_curseg) {
3152 		if (old_cursegno != curseg->segno) {
3153 			curseg->next_segno = old_cursegno;
3154 			change_curseg(sbi, type);
3155 		}
3156 		curseg->next_blkoff = old_blkoff;
3157 	}
3158 
3159 	up_write(&sit_i->sentry_lock);
3160 	mutex_unlock(&curseg->curseg_mutex);
3161 	up_write(&SM_I(sbi)->curseg_lock);
3162 }
3163 
3164 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3165 				block_t old_addr, block_t new_addr,
3166 				unsigned char version, bool recover_curseg,
3167 				bool recover_newaddr)
3168 {
3169 	struct f2fs_summary sum;
3170 
3171 	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3172 
3173 	f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3174 					recover_curseg, recover_newaddr);
3175 
3176 	f2fs_update_data_blkaddr(dn, new_addr);
3177 }
3178 
3179 void f2fs_wait_on_page_writeback(struct page *page,
3180 				enum page_type type, bool ordered)
3181 {
3182 	if (PageWriteback(page)) {
3183 		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3184 
3185 		f2fs_submit_merged_write_cond(sbi, page->mapping->host,
3186 						0, page->index, type);
3187 		if (ordered)
3188 			wait_on_page_writeback(page);
3189 		else
3190 			wait_for_stable_page(page);
3191 	}
3192 }
3193 
3194 void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
3195 {
3196 	struct page *cpage;
3197 
3198 	if (!is_valid_data_blkaddr(sbi, blkaddr))
3199 		return;
3200 
3201 	cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3202 	if (cpage) {
3203 		f2fs_wait_on_page_writeback(cpage, DATA, true);
3204 		f2fs_put_page(cpage, 1);
3205 	}
3206 }
3207 
3208 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3209 {
3210 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3211 	struct curseg_info *seg_i;
3212 	unsigned char *kaddr;
3213 	struct page *page;
3214 	block_t start;
3215 	int i, j, offset;
3216 
3217 	start = start_sum_block(sbi);
3218 
3219 	page = f2fs_get_meta_page(sbi, start++);
3220 	if (IS_ERR(page))
3221 		return PTR_ERR(page);
3222 	kaddr = (unsigned char *)page_address(page);
3223 
3224 	/* Step 1: restore nat cache */
3225 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3226 	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3227 
3228 	/* Step 2: restore sit cache */
3229 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3230 	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3231 	offset = 2 * SUM_JOURNAL_SIZE;
3232 
3233 	/* Step 3: restore summary entries */
3234 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3235 		unsigned short blk_off;
3236 		unsigned int segno;
3237 
3238 		seg_i = CURSEG_I(sbi, i);
3239 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3240 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3241 		seg_i->next_segno = segno;
3242 		reset_curseg(sbi, i, 0);
3243 		seg_i->alloc_type = ckpt->alloc_type[i];
3244 		seg_i->next_blkoff = blk_off;
3245 
3246 		if (seg_i->alloc_type == SSR)
3247 			blk_off = sbi->blocks_per_seg;
3248 
3249 		for (j = 0; j < blk_off; j++) {
3250 			struct f2fs_summary *s;
3251 			s = (struct f2fs_summary *)(kaddr + offset);
3252 			seg_i->sum_blk->entries[j] = *s;
3253 			offset += SUMMARY_SIZE;
3254 			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3255 						SUM_FOOTER_SIZE)
3256 				continue;
3257 
3258 			f2fs_put_page(page, 1);
3259 			page = NULL;
3260 
3261 			page = f2fs_get_meta_page(sbi, start++);
3262 			if (IS_ERR(page))
3263 				return PTR_ERR(page);
3264 			kaddr = (unsigned char *)page_address(page);
3265 			offset = 0;
3266 		}
3267 	}
3268 	f2fs_put_page(page, 1);
3269 	return 0;
3270 }
3271 
3272 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3273 {
3274 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3275 	struct f2fs_summary_block *sum;
3276 	struct curseg_info *curseg;
3277 	struct page *new;
3278 	unsigned short blk_off;
3279 	unsigned int segno = 0;
3280 	block_t blk_addr = 0;
3281 	int err = 0;
3282 
3283 	/* get segment number and block addr */
3284 	if (IS_DATASEG(type)) {
3285 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3286 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3287 							CURSEG_HOT_DATA]);
3288 		if (__exist_node_summaries(sbi))
3289 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3290 		else
3291 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3292 	} else {
3293 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
3294 							CURSEG_HOT_NODE]);
3295 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3296 							CURSEG_HOT_NODE]);
3297 		if (__exist_node_summaries(sbi))
3298 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3299 							type - CURSEG_HOT_NODE);
3300 		else
3301 			blk_addr = GET_SUM_BLOCK(sbi, segno);
3302 	}
3303 
3304 	new = f2fs_get_meta_page(sbi, blk_addr);
3305 	if (IS_ERR(new))
3306 		return PTR_ERR(new);
3307 	sum = (struct f2fs_summary_block *)page_address(new);
3308 
3309 	if (IS_NODESEG(type)) {
3310 		if (__exist_node_summaries(sbi)) {
3311 			struct f2fs_summary *ns = &sum->entries[0];
3312 			int i;
3313 			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3314 				ns->version = 0;
3315 				ns->ofs_in_node = 0;
3316 			}
3317 		} else {
3318 			err = f2fs_restore_node_summary(sbi, segno, sum);
3319 			if (err)
3320 				goto out;
3321 		}
3322 	}
3323 
3324 	/* set uncompleted segment to curseg */
3325 	curseg = CURSEG_I(sbi, type);
3326 	mutex_lock(&curseg->curseg_mutex);
3327 
3328 	/* update journal info */
3329 	down_write(&curseg->journal_rwsem);
3330 	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3331 	up_write(&curseg->journal_rwsem);
3332 
3333 	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3334 	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3335 	curseg->next_segno = segno;
3336 	reset_curseg(sbi, type, 0);
3337 	curseg->alloc_type = ckpt->alloc_type[type];
3338 	curseg->next_blkoff = blk_off;
3339 	mutex_unlock(&curseg->curseg_mutex);
3340 out:
3341 	f2fs_put_page(new, 1);
3342 	return err;
3343 }
3344 
3345 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3346 {
3347 	struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3348 	struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3349 	int type = CURSEG_HOT_DATA;
3350 	int err;
3351 
3352 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3353 		int npages = f2fs_npages_for_summary_flush(sbi, true);
3354 
3355 		if (npages >= 2)
3356 			f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3357 							META_CP, true);
3358 
3359 		/* restore for compacted data summary */
3360 		err = read_compacted_summaries(sbi);
3361 		if (err)
3362 			return err;
3363 		type = CURSEG_HOT_NODE;
3364 	}
3365 
3366 	if (__exist_node_summaries(sbi))
3367 		f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3368 					NR_CURSEG_TYPE - type, META_CP, true);
3369 
3370 	for (; type <= CURSEG_COLD_NODE; type++) {
3371 		err = read_normal_summaries(sbi, type);
3372 		if (err)
3373 			return err;
3374 	}
3375 
3376 	/* sanity check for summary blocks */
3377 	if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3378 			sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3379 		return -EINVAL;
3380 
3381 	return 0;
3382 }
3383 
3384 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3385 {
3386 	struct page *page;
3387 	unsigned char *kaddr;
3388 	struct f2fs_summary *summary;
3389 	struct curseg_info *seg_i;
3390 	int written_size = 0;
3391 	int i, j;
3392 
3393 	page = f2fs_grab_meta_page(sbi, blkaddr++);
3394 	kaddr = (unsigned char *)page_address(page);
3395 	memset(kaddr, 0, PAGE_SIZE);
3396 
3397 	/* Step 1: write nat cache */
3398 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3399 	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3400 	written_size += SUM_JOURNAL_SIZE;
3401 
3402 	/* Step 2: write sit cache */
3403 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3404 	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3405 	written_size += SUM_JOURNAL_SIZE;
3406 
3407 	/* Step 3: write summary entries */
3408 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3409 		unsigned short blkoff;
3410 		seg_i = CURSEG_I(sbi, i);
3411 		if (sbi->ckpt->alloc_type[i] == SSR)
3412 			blkoff = sbi->blocks_per_seg;
3413 		else
3414 			blkoff = curseg_blkoff(sbi, i);
3415 
3416 		for (j = 0; j < blkoff; j++) {
3417 			if (!page) {
3418 				page = f2fs_grab_meta_page(sbi, blkaddr++);
3419 				kaddr = (unsigned char *)page_address(page);
3420 				memset(kaddr, 0, PAGE_SIZE);
3421 				written_size = 0;
3422 			}
3423 			summary = (struct f2fs_summary *)(kaddr + written_size);
3424 			*summary = seg_i->sum_blk->entries[j];
3425 			written_size += SUMMARY_SIZE;
3426 
3427 			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3428 							SUM_FOOTER_SIZE)
3429 				continue;
3430 
3431 			set_page_dirty(page);
3432 			f2fs_put_page(page, 1);
3433 			page = NULL;
3434 		}
3435 	}
3436 	if (page) {
3437 		set_page_dirty(page);
3438 		f2fs_put_page(page, 1);
3439 	}
3440 }
3441 
3442 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3443 					block_t blkaddr, int type)
3444 {
3445 	int i, end;
3446 	if (IS_DATASEG(type))
3447 		end = type + NR_CURSEG_DATA_TYPE;
3448 	else
3449 		end = type + NR_CURSEG_NODE_TYPE;
3450 
3451 	for (i = type; i < end; i++)
3452 		write_current_sum_page(sbi, i, blkaddr + (i - type));
3453 }
3454 
3455 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3456 {
3457 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3458 		write_compacted_summaries(sbi, start_blk);
3459 	else
3460 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3461 }
3462 
3463 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3464 {
3465 	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3466 }
3467 
3468 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3469 					unsigned int val, int alloc)
3470 {
3471 	int i;
3472 
3473 	if (type == NAT_JOURNAL) {
3474 		for (i = 0; i < nats_in_cursum(journal); i++) {
3475 			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3476 				return i;
3477 		}
3478 		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3479 			return update_nats_in_cursum(journal, 1);
3480 	} else if (type == SIT_JOURNAL) {
3481 		for (i = 0; i < sits_in_cursum(journal); i++)
3482 			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3483 				return i;
3484 		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3485 			return update_sits_in_cursum(journal, 1);
3486 	}
3487 	return -1;
3488 }
3489 
3490 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3491 					unsigned int segno)
3492 {
3493 	return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3494 }
3495 
3496 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3497 					unsigned int start)
3498 {
3499 	struct sit_info *sit_i = SIT_I(sbi);
3500 	struct page *page;
3501 	pgoff_t src_off, dst_off;
3502 
3503 	src_off = current_sit_addr(sbi, start);
3504 	dst_off = next_sit_addr(sbi, src_off);
3505 
3506 	page = f2fs_grab_meta_page(sbi, dst_off);
3507 	seg_info_to_sit_page(sbi, page, start);
3508 
3509 	set_page_dirty(page);
3510 	set_to_next_sit(sit_i, start);
3511 
3512 	return page;
3513 }
3514 
3515 static struct sit_entry_set *grab_sit_entry_set(void)
3516 {
3517 	struct sit_entry_set *ses =
3518 			f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3519 
3520 	ses->entry_cnt = 0;
3521 	INIT_LIST_HEAD(&ses->set_list);
3522 	return ses;
3523 }
3524 
3525 static void release_sit_entry_set(struct sit_entry_set *ses)
3526 {
3527 	list_del(&ses->set_list);
3528 	kmem_cache_free(sit_entry_set_slab, ses);
3529 }
3530 
3531 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3532 						struct list_head *head)
3533 {
3534 	struct sit_entry_set *next = ses;
3535 
3536 	if (list_is_last(&ses->set_list, head))
3537 		return;
3538 
3539 	list_for_each_entry_continue(next, head, set_list)
3540 		if (ses->entry_cnt <= next->entry_cnt)
3541 			break;
3542 
3543 	list_move_tail(&ses->set_list, &next->set_list);
3544 }
3545 
3546 static void add_sit_entry(unsigned int segno, struct list_head *head)
3547 {
3548 	struct sit_entry_set *ses;
3549 	unsigned int start_segno = START_SEGNO(segno);
3550 
3551 	list_for_each_entry(ses, head, set_list) {
3552 		if (ses->start_segno == start_segno) {
3553 			ses->entry_cnt++;
3554 			adjust_sit_entry_set(ses, head);
3555 			return;
3556 		}
3557 	}
3558 
3559 	ses = grab_sit_entry_set();
3560 
3561 	ses->start_segno = start_segno;
3562 	ses->entry_cnt++;
3563 	list_add(&ses->set_list, head);
3564 }
3565 
3566 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3567 {
3568 	struct f2fs_sm_info *sm_info = SM_I(sbi);
3569 	struct list_head *set_list = &sm_info->sit_entry_set;
3570 	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3571 	unsigned int segno;
3572 
3573 	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3574 		add_sit_entry(segno, set_list);
3575 }
3576 
3577 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3578 {
3579 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3580 	struct f2fs_journal *journal = curseg->journal;
3581 	int i;
3582 
3583 	down_write(&curseg->journal_rwsem);
3584 	for (i = 0; i < sits_in_cursum(journal); i++) {
3585 		unsigned int segno;
3586 		bool dirtied;
3587 
3588 		segno = le32_to_cpu(segno_in_journal(journal, i));
3589 		dirtied = __mark_sit_entry_dirty(sbi, segno);
3590 
3591 		if (!dirtied)
3592 			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3593 	}
3594 	update_sits_in_cursum(journal, -i);
3595 	up_write(&curseg->journal_rwsem);
3596 }
3597 
3598 /*
3599  * CP calls this function, which flushes SIT entries including sit_journal,
3600  * and moves prefree segs to free segs.
3601  */
3602 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3603 {
3604 	struct sit_info *sit_i = SIT_I(sbi);
3605 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3606 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3607 	struct f2fs_journal *journal = curseg->journal;
3608 	struct sit_entry_set *ses, *tmp;
3609 	struct list_head *head = &SM_I(sbi)->sit_entry_set;
3610 	bool to_journal = true;
3611 	struct seg_entry *se;
3612 
3613 	down_write(&sit_i->sentry_lock);
3614 
3615 	if (!sit_i->dirty_sentries)
3616 		goto out;
3617 
3618 	/*
3619 	 * add and account sit entries of dirty bitmap in sit entry
3620 	 * set temporarily
3621 	 */
3622 	add_sits_in_set(sbi);
3623 
3624 	/*
3625 	 * if there are no enough space in journal to store dirty sit
3626 	 * entries, remove all entries from journal and add and account
3627 	 * them in sit entry set.
3628 	 */
3629 	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3630 		remove_sits_in_journal(sbi);
3631 
3632 	/*
3633 	 * there are two steps to flush sit entries:
3634 	 * #1, flush sit entries to journal in current cold data summary block.
3635 	 * #2, flush sit entries to sit page.
3636 	 */
3637 	list_for_each_entry_safe(ses, tmp, head, set_list) {
3638 		struct page *page = NULL;
3639 		struct f2fs_sit_block *raw_sit = NULL;
3640 		unsigned int start_segno = ses->start_segno;
3641 		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3642 						(unsigned long)MAIN_SEGS(sbi));
3643 		unsigned int segno = start_segno;
3644 
3645 		if (to_journal &&
3646 			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3647 			to_journal = false;
3648 
3649 		if (to_journal) {
3650 			down_write(&curseg->journal_rwsem);
3651 		} else {
3652 			page = get_next_sit_page(sbi, start_segno);
3653 			raw_sit = page_address(page);
3654 		}
3655 
3656 		/* flush dirty sit entries in region of current sit set */
3657 		for_each_set_bit_from(segno, bitmap, end) {
3658 			int offset, sit_offset;
3659 
3660 			se = get_seg_entry(sbi, segno);
3661 #ifdef CONFIG_F2FS_CHECK_FS
3662 			if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3663 						SIT_VBLOCK_MAP_SIZE))
3664 				f2fs_bug_on(sbi, 1);
3665 #endif
3666 
3667 			/* add discard candidates */
3668 			if (!(cpc->reason & CP_DISCARD)) {
3669 				cpc->trim_start = segno;
3670 				add_discard_addrs(sbi, cpc, false);
3671 			}
3672 
3673 			if (to_journal) {
3674 				offset = f2fs_lookup_journal_in_cursum(journal,
3675 							SIT_JOURNAL, segno, 1);
3676 				f2fs_bug_on(sbi, offset < 0);
3677 				segno_in_journal(journal, offset) =
3678 							cpu_to_le32(segno);
3679 				seg_info_to_raw_sit(se,
3680 					&sit_in_journal(journal, offset));
3681 				check_block_count(sbi, segno,
3682 					&sit_in_journal(journal, offset));
3683 			} else {
3684 				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3685 				seg_info_to_raw_sit(se,
3686 						&raw_sit->entries[sit_offset]);
3687 				check_block_count(sbi, segno,
3688 						&raw_sit->entries[sit_offset]);
3689 			}
3690 
3691 			__clear_bit(segno, bitmap);
3692 			sit_i->dirty_sentries--;
3693 			ses->entry_cnt--;
3694 		}
3695 
3696 		if (to_journal)
3697 			up_write(&curseg->journal_rwsem);
3698 		else
3699 			f2fs_put_page(page, 1);
3700 
3701 		f2fs_bug_on(sbi, ses->entry_cnt);
3702 		release_sit_entry_set(ses);
3703 	}
3704 
3705 	f2fs_bug_on(sbi, !list_empty(head));
3706 	f2fs_bug_on(sbi, sit_i->dirty_sentries);
3707 out:
3708 	if (cpc->reason & CP_DISCARD) {
3709 		__u64 trim_start = cpc->trim_start;
3710 
3711 		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3712 			add_discard_addrs(sbi, cpc, false);
3713 
3714 		cpc->trim_start = trim_start;
3715 	}
3716 	up_write(&sit_i->sentry_lock);
3717 
3718 	set_prefree_as_free_segments(sbi);
3719 }
3720 
3721 static int build_sit_info(struct f2fs_sb_info *sbi)
3722 {
3723 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3724 	struct sit_info *sit_i;
3725 	unsigned int sit_segs, start;
3726 	char *src_bitmap;
3727 	unsigned int bitmap_size;
3728 
3729 	/* allocate memory for SIT information */
3730 	sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3731 	if (!sit_i)
3732 		return -ENOMEM;
3733 
3734 	SM_I(sbi)->sit_info = sit_i;
3735 
3736 	sit_i->sentries =
3737 		f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3738 					      MAIN_SEGS(sbi)),
3739 			      GFP_KERNEL);
3740 	if (!sit_i->sentries)
3741 		return -ENOMEM;
3742 
3743 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3744 	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3745 								GFP_KERNEL);
3746 	if (!sit_i->dirty_sentries_bitmap)
3747 		return -ENOMEM;
3748 
3749 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
3750 		sit_i->sentries[start].cur_valid_map
3751 			= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3752 		sit_i->sentries[start].ckpt_valid_map
3753 			= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3754 		if (!sit_i->sentries[start].cur_valid_map ||
3755 				!sit_i->sentries[start].ckpt_valid_map)
3756 			return -ENOMEM;
3757 
3758 #ifdef CONFIG_F2FS_CHECK_FS
3759 		sit_i->sentries[start].cur_valid_map_mir
3760 			= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3761 		if (!sit_i->sentries[start].cur_valid_map_mir)
3762 			return -ENOMEM;
3763 #endif
3764 
3765 		if (f2fs_discard_en(sbi)) {
3766 			sit_i->sentries[start].discard_map
3767 				= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3768 								GFP_KERNEL);
3769 			if (!sit_i->sentries[start].discard_map)
3770 				return -ENOMEM;
3771 		}
3772 	}
3773 
3774 	sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3775 	if (!sit_i->tmp_map)
3776 		return -ENOMEM;
3777 
3778 	if (sbi->segs_per_sec > 1) {
3779 		sit_i->sec_entries =
3780 			f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3781 						      MAIN_SECS(sbi)),
3782 				      GFP_KERNEL);
3783 		if (!sit_i->sec_entries)
3784 			return -ENOMEM;
3785 	}
3786 
3787 	/* get information related with SIT */
3788 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3789 
3790 	/* setup SIT bitmap from ckeckpoint pack */
3791 	bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3792 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3793 
3794 	sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3795 	if (!sit_i->sit_bitmap)
3796 		return -ENOMEM;
3797 
3798 #ifdef CONFIG_F2FS_CHECK_FS
3799 	sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3800 	if (!sit_i->sit_bitmap_mir)
3801 		return -ENOMEM;
3802 #endif
3803 
3804 	/* init SIT information */
3805 	sit_i->s_ops = &default_salloc_ops;
3806 
3807 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3808 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3809 	sit_i->written_valid_blocks = 0;
3810 	sit_i->bitmap_size = bitmap_size;
3811 	sit_i->dirty_sentries = 0;
3812 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3813 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3814 	sit_i->mounted_time = ktime_get_real_seconds();
3815 	init_rwsem(&sit_i->sentry_lock);
3816 	return 0;
3817 }
3818 
3819 static int build_free_segmap(struct f2fs_sb_info *sbi)
3820 {
3821 	struct free_segmap_info *free_i;
3822 	unsigned int bitmap_size, sec_bitmap_size;
3823 
3824 	/* allocate memory for free segmap information */
3825 	free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
3826 	if (!free_i)
3827 		return -ENOMEM;
3828 
3829 	SM_I(sbi)->free_info = free_i;
3830 
3831 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3832 	free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
3833 	if (!free_i->free_segmap)
3834 		return -ENOMEM;
3835 
3836 	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3837 	free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
3838 	if (!free_i->free_secmap)
3839 		return -ENOMEM;
3840 
3841 	/* set all segments as dirty temporarily */
3842 	memset(free_i->free_segmap, 0xff, bitmap_size);
3843 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3844 
3845 	/* init free segmap information */
3846 	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3847 	free_i->free_segments = 0;
3848 	free_i->free_sections = 0;
3849 	spin_lock_init(&free_i->segmap_lock);
3850 	return 0;
3851 }
3852 
3853 static int build_curseg(struct f2fs_sb_info *sbi)
3854 {
3855 	struct curseg_info *array;
3856 	int i;
3857 
3858 	array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
3859 			     GFP_KERNEL);
3860 	if (!array)
3861 		return -ENOMEM;
3862 
3863 	SM_I(sbi)->curseg_array = array;
3864 
3865 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
3866 		mutex_init(&array[i].curseg_mutex);
3867 		array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
3868 		if (!array[i].sum_blk)
3869 			return -ENOMEM;
3870 		init_rwsem(&array[i].journal_rwsem);
3871 		array[i].journal = f2fs_kzalloc(sbi,
3872 				sizeof(struct f2fs_journal), GFP_KERNEL);
3873 		if (!array[i].journal)
3874 			return -ENOMEM;
3875 		array[i].segno = NULL_SEGNO;
3876 		array[i].next_blkoff = 0;
3877 	}
3878 	return restore_curseg_summaries(sbi);
3879 }
3880 
3881 static int build_sit_entries(struct f2fs_sb_info *sbi)
3882 {
3883 	struct sit_info *sit_i = SIT_I(sbi);
3884 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3885 	struct f2fs_journal *journal = curseg->journal;
3886 	struct seg_entry *se;
3887 	struct f2fs_sit_entry sit;
3888 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
3889 	unsigned int i, start, end;
3890 	unsigned int readed, start_blk = 0;
3891 	int err = 0;
3892 	block_t total_node_blocks = 0;
3893 
3894 	do {
3895 		readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3896 							META_SIT, true);
3897 
3898 		start = start_blk * sit_i->sents_per_block;
3899 		end = (start_blk + readed) * sit_i->sents_per_block;
3900 
3901 		for (; start < end && start < MAIN_SEGS(sbi); start++) {
3902 			struct f2fs_sit_block *sit_blk;
3903 			struct page *page;
3904 
3905 			se = &sit_i->sentries[start];
3906 			page = get_current_sit_page(sbi, start);
3907 			sit_blk = (struct f2fs_sit_block *)page_address(page);
3908 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3909 			f2fs_put_page(page, 1);
3910 
3911 			err = check_block_count(sbi, start, &sit);
3912 			if (err)
3913 				return err;
3914 			seg_info_from_raw_sit(se, &sit);
3915 			if (IS_NODESEG(se->type))
3916 				total_node_blocks += se->valid_blocks;
3917 
3918 			/* build discard map only one time */
3919 			if (f2fs_discard_en(sbi)) {
3920 				if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3921 					memset(se->discard_map, 0xff,
3922 						SIT_VBLOCK_MAP_SIZE);
3923 				} else {
3924 					memcpy(se->discard_map,
3925 						se->cur_valid_map,
3926 						SIT_VBLOCK_MAP_SIZE);
3927 					sbi->discard_blks +=
3928 						sbi->blocks_per_seg -
3929 						se->valid_blocks;
3930 				}
3931 			}
3932 
3933 			if (sbi->segs_per_sec > 1)
3934 				get_sec_entry(sbi, start)->valid_blocks +=
3935 							se->valid_blocks;
3936 		}
3937 		start_blk += readed;
3938 	} while (start_blk < sit_blk_cnt);
3939 
3940 	down_read(&curseg->journal_rwsem);
3941 	for (i = 0; i < sits_in_cursum(journal); i++) {
3942 		unsigned int old_valid_blocks;
3943 
3944 		start = le32_to_cpu(segno_in_journal(journal, i));
3945 		if (start >= MAIN_SEGS(sbi)) {
3946 			f2fs_msg(sbi->sb, KERN_ERR,
3947 					"Wrong journal entry on segno %u",
3948 					start);
3949 			set_sbi_flag(sbi, SBI_NEED_FSCK);
3950 			err = -EINVAL;
3951 			break;
3952 		}
3953 
3954 		se = &sit_i->sentries[start];
3955 		sit = sit_in_journal(journal, i);
3956 
3957 		old_valid_blocks = se->valid_blocks;
3958 		if (IS_NODESEG(se->type))
3959 			total_node_blocks -= old_valid_blocks;
3960 
3961 		err = check_block_count(sbi, start, &sit);
3962 		if (err)
3963 			break;
3964 		seg_info_from_raw_sit(se, &sit);
3965 		if (IS_NODESEG(se->type))
3966 			total_node_blocks += se->valid_blocks;
3967 
3968 		if (f2fs_discard_en(sbi)) {
3969 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3970 				memset(se->discard_map, 0xff,
3971 							SIT_VBLOCK_MAP_SIZE);
3972 			} else {
3973 				memcpy(se->discard_map, se->cur_valid_map,
3974 							SIT_VBLOCK_MAP_SIZE);
3975 				sbi->discard_blks += old_valid_blocks;
3976 				sbi->discard_blks -= se->valid_blocks;
3977 			}
3978 		}
3979 
3980 		if (sbi->segs_per_sec > 1) {
3981 			get_sec_entry(sbi, start)->valid_blocks +=
3982 							se->valid_blocks;
3983 			get_sec_entry(sbi, start)->valid_blocks -=
3984 							old_valid_blocks;
3985 		}
3986 	}
3987 	up_read(&curseg->journal_rwsem);
3988 
3989 	if (!err && total_node_blocks != valid_node_count(sbi)) {
3990 		f2fs_msg(sbi->sb, KERN_ERR,
3991 			"SIT is corrupted node# %u vs %u",
3992 			total_node_blocks, valid_node_count(sbi));
3993 		set_sbi_flag(sbi, SBI_NEED_FSCK);
3994 		err = -EINVAL;
3995 	}
3996 
3997 	return err;
3998 }
3999 
4000 static void init_free_segmap(struct f2fs_sb_info *sbi)
4001 {
4002 	unsigned int start;
4003 	int type;
4004 
4005 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4006 		struct seg_entry *sentry = get_seg_entry(sbi, start);
4007 		if (!sentry->valid_blocks)
4008 			__set_free(sbi, start);
4009 		else
4010 			SIT_I(sbi)->written_valid_blocks +=
4011 						sentry->valid_blocks;
4012 	}
4013 
4014 	/* set use the current segments */
4015 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4016 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4017 		__set_test_and_inuse(sbi, curseg_t->segno);
4018 	}
4019 }
4020 
4021 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4022 {
4023 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4024 	struct free_segmap_info *free_i = FREE_I(sbi);
4025 	unsigned int segno = 0, offset = 0;
4026 	unsigned short valid_blocks;
4027 
4028 	while (1) {
4029 		/* find dirty segment based on free segmap */
4030 		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4031 		if (segno >= MAIN_SEGS(sbi))
4032 			break;
4033 		offset = segno + 1;
4034 		valid_blocks = get_valid_blocks(sbi, segno, false);
4035 		if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4036 			continue;
4037 		if (valid_blocks > sbi->blocks_per_seg) {
4038 			f2fs_bug_on(sbi, 1);
4039 			continue;
4040 		}
4041 		mutex_lock(&dirty_i->seglist_lock);
4042 		__locate_dirty_segment(sbi, segno, DIRTY);
4043 		mutex_unlock(&dirty_i->seglist_lock);
4044 	}
4045 }
4046 
4047 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4048 {
4049 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4050 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4051 
4052 	dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4053 	if (!dirty_i->victim_secmap)
4054 		return -ENOMEM;
4055 	return 0;
4056 }
4057 
4058 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4059 {
4060 	struct dirty_seglist_info *dirty_i;
4061 	unsigned int bitmap_size, i;
4062 
4063 	/* allocate memory for dirty segments list information */
4064 	dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4065 								GFP_KERNEL);
4066 	if (!dirty_i)
4067 		return -ENOMEM;
4068 
4069 	SM_I(sbi)->dirty_info = dirty_i;
4070 	mutex_init(&dirty_i->seglist_lock);
4071 
4072 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4073 
4074 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
4075 		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4076 								GFP_KERNEL);
4077 		if (!dirty_i->dirty_segmap[i])
4078 			return -ENOMEM;
4079 	}
4080 
4081 	init_dirty_segmap(sbi);
4082 	return init_victim_secmap(sbi);
4083 }
4084 
4085 /*
4086  * Update min, max modified time for cost-benefit GC algorithm
4087  */
4088 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4089 {
4090 	struct sit_info *sit_i = SIT_I(sbi);
4091 	unsigned int segno;
4092 
4093 	down_write(&sit_i->sentry_lock);
4094 
4095 	sit_i->min_mtime = ULLONG_MAX;
4096 
4097 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4098 		unsigned int i;
4099 		unsigned long long mtime = 0;
4100 
4101 		for (i = 0; i < sbi->segs_per_sec; i++)
4102 			mtime += get_seg_entry(sbi, segno + i)->mtime;
4103 
4104 		mtime = div_u64(mtime, sbi->segs_per_sec);
4105 
4106 		if (sit_i->min_mtime > mtime)
4107 			sit_i->min_mtime = mtime;
4108 	}
4109 	sit_i->max_mtime = get_mtime(sbi, false);
4110 	up_write(&sit_i->sentry_lock);
4111 }
4112 
4113 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4114 {
4115 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4116 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4117 	struct f2fs_sm_info *sm_info;
4118 	int err;
4119 
4120 	sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4121 	if (!sm_info)
4122 		return -ENOMEM;
4123 
4124 	/* init sm info */
4125 	sbi->sm_info = sm_info;
4126 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4127 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4128 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4129 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4130 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4131 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4132 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4133 	sm_info->rec_prefree_segments = sm_info->main_segments *
4134 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4135 	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4136 		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4137 
4138 	if (!test_opt(sbi, LFS))
4139 		sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4140 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4141 	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4142 	sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4143 	sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4144 	sm_info->min_ssr_sections = reserved_sections(sbi);
4145 
4146 	INIT_LIST_HEAD(&sm_info->sit_entry_set);
4147 
4148 	init_rwsem(&sm_info->curseg_lock);
4149 
4150 	if (!f2fs_readonly(sbi->sb)) {
4151 		err = f2fs_create_flush_cmd_control(sbi);
4152 		if (err)
4153 			return err;
4154 	}
4155 
4156 	err = create_discard_cmd_control(sbi);
4157 	if (err)
4158 		return err;
4159 
4160 	err = build_sit_info(sbi);
4161 	if (err)
4162 		return err;
4163 	err = build_free_segmap(sbi);
4164 	if (err)
4165 		return err;
4166 	err = build_curseg(sbi);
4167 	if (err)
4168 		return err;
4169 
4170 	/* reinit free segmap based on SIT */
4171 	err = build_sit_entries(sbi);
4172 	if (err)
4173 		return err;
4174 
4175 	init_free_segmap(sbi);
4176 	err = build_dirty_segmap(sbi);
4177 	if (err)
4178 		return err;
4179 
4180 	init_min_max_mtime(sbi);
4181 	return 0;
4182 }
4183 
4184 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4185 		enum dirty_type dirty_type)
4186 {
4187 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4188 
4189 	mutex_lock(&dirty_i->seglist_lock);
4190 	kvfree(dirty_i->dirty_segmap[dirty_type]);
4191 	dirty_i->nr_dirty[dirty_type] = 0;
4192 	mutex_unlock(&dirty_i->seglist_lock);
4193 }
4194 
4195 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4196 {
4197 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4198 	kvfree(dirty_i->victim_secmap);
4199 }
4200 
4201 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4202 {
4203 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4204 	int i;
4205 
4206 	if (!dirty_i)
4207 		return;
4208 
4209 	/* discard pre-free/dirty segments list */
4210 	for (i = 0; i < NR_DIRTY_TYPE; i++)
4211 		discard_dirty_segmap(sbi, i);
4212 
4213 	destroy_victim_secmap(sbi);
4214 	SM_I(sbi)->dirty_info = NULL;
4215 	kfree(dirty_i);
4216 }
4217 
4218 static void destroy_curseg(struct f2fs_sb_info *sbi)
4219 {
4220 	struct curseg_info *array = SM_I(sbi)->curseg_array;
4221 	int i;
4222 
4223 	if (!array)
4224 		return;
4225 	SM_I(sbi)->curseg_array = NULL;
4226 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
4227 		kfree(array[i].sum_blk);
4228 		kfree(array[i].journal);
4229 	}
4230 	kfree(array);
4231 }
4232 
4233 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4234 {
4235 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4236 	if (!free_i)
4237 		return;
4238 	SM_I(sbi)->free_info = NULL;
4239 	kvfree(free_i->free_segmap);
4240 	kvfree(free_i->free_secmap);
4241 	kfree(free_i);
4242 }
4243 
4244 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4245 {
4246 	struct sit_info *sit_i = SIT_I(sbi);
4247 	unsigned int start;
4248 
4249 	if (!sit_i)
4250 		return;
4251 
4252 	if (sit_i->sentries) {
4253 		for (start = 0; start < MAIN_SEGS(sbi); start++) {
4254 			kfree(sit_i->sentries[start].cur_valid_map);
4255 #ifdef CONFIG_F2FS_CHECK_FS
4256 			kfree(sit_i->sentries[start].cur_valid_map_mir);
4257 #endif
4258 			kfree(sit_i->sentries[start].ckpt_valid_map);
4259 			kfree(sit_i->sentries[start].discard_map);
4260 		}
4261 	}
4262 	kfree(sit_i->tmp_map);
4263 
4264 	kvfree(sit_i->sentries);
4265 	kvfree(sit_i->sec_entries);
4266 	kvfree(sit_i->dirty_sentries_bitmap);
4267 
4268 	SM_I(sbi)->sit_info = NULL;
4269 	kfree(sit_i->sit_bitmap);
4270 #ifdef CONFIG_F2FS_CHECK_FS
4271 	kfree(sit_i->sit_bitmap_mir);
4272 #endif
4273 	kfree(sit_i);
4274 }
4275 
4276 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4277 {
4278 	struct f2fs_sm_info *sm_info = SM_I(sbi);
4279 
4280 	if (!sm_info)
4281 		return;
4282 	f2fs_destroy_flush_cmd_control(sbi, true);
4283 	destroy_discard_cmd_control(sbi);
4284 	destroy_dirty_segmap(sbi);
4285 	destroy_curseg(sbi);
4286 	destroy_free_segmap(sbi);
4287 	destroy_sit_info(sbi);
4288 	sbi->sm_info = NULL;
4289 	kfree(sm_info);
4290 }
4291 
4292 int __init f2fs_create_segment_manager_caches(void)
4293 {
4294 	discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4295 			sizeof(struct discard_entry));
4296 	if (!discard_entry_slab)
4297 		goto fail;
4298 
4299 	discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4300 			sizeof(struct discard_cmd));
4301 	if (!discard_cmd_slab)
4302 		goto destroy_discard_entry;
4303 
4304 	sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4305 			sizeof(struct sit_entry_set));
4306 	if (!sit_entry_set_slab)
4307 		goto destroy_discard_cmd;
4308 
4309 	inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4310 			sizeof(struct inmem_pages));
4311 	if (!inmem_entry_slab)
4312 		goto destroy_sit_entry_set;
4313 	return 0;
4314 
4315 destroy_sit_entry_set:
4316 	kmem_cache_destroy(sit_entry_set_slab);
4317 destroy_discard_cmd:
4318 	kmem_cache_destroy(discard_cmd_slab);
4319 destroy_discard_entry:
4320 	kmem_cache_destroy(discard_entry_slab);
4321 fail:
4322 	return -ENOMEM;
4323 }
4324 
4325 void f2fs_destroy_segment_manager_caches(void)
4326 {
4327 	kmem_cache_destroy(sit_entry_set_slab);
4328 	kmem_cache_destroy(discard_cmd_slab);
4329 	kmem_cache_destroy(discard_entry_slab);
4330 	kmem_cache_destroy(inmem_entry_slab);
4331 }
4332