xref: /openbmc/linux/fs/f2fs/segment.c (revision 4c9ca2fd)
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 			f2fs_get_node_info(sbi, dn.nid, &ni);
254 			if (cur->old_addr == NEW_ADDR) {
255 				f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
256 				f2fs_update_data_blkaddr(&dn, NEW_ADDR);
257 			} else
258 				f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
259 					cur->old_addr, ni.version, true, true);
260 			f2fs_put_dnode(&dn);
261 		}
262 next:
263 		/* we don't need to invalidate this in the sccessful status */
264 		if (drop || recover)
265 			ClearPageUptodate(page);
266 		set_page_private(page, 0);
267 		ClearPagePrivate(page);
268 		f2fs_put_page(page, 1);
269 
270 		list_del(&cur->list);
271 		kmem_cache_free(inmem_entry_slab, cur);
272 		dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
273 	}
274 	return err;
275 }
276 
277 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
278 {
279 	struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
280 	struct inode *inode;
281 	struct f2fs_inode_info *fi;
282 next:
283 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
284 	if (list_empty(head)) {
285 		spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
286 		return;
287 	}
288 	fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
289 	inode = igrab(&fi->vfs_inode);
290 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
291 
292 	if (inode) {
293 		if (gc_failure) {
294 			if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
295 				goto drop;
296 			goto skip;
297 		}
298 drop:
299 		set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
300 		f2fs_drop_inmem_pages(inode);
301 		iput(inode);
302 	}
303 skip:
304 	congestion_wait(BLK_RW_ASYNC, HZ/50);
305 	cond_resched();
306 	goto next;
307 }
308 
309 void f2fs_drop_inmem_pages(struct inode *inode)
310 {
311 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
312 	struct f2fs_inode_info *fi = F2FS_I(inode);
313 
314 	mutex_lock(&fi->inmem_lock);
315 	__revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
316 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
317 	if (!list_empty(&fi->inmem_ilist))
318 		list_del_init(&fi->inmem_ilist);
319 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
320 	mutex_unlock(&fi->inmem_lock);
321 
322 	clear_inode_flag(inode, FI_ATOMIC_FILE);
323 	fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
324 	stat_dec_atomic_write(inode);
325 }
326 
327 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
328 {
329 	struct f2fs_inode_info *fi = F2FS_I(inode);
330 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
331 	struct list_head *head = &fi->inmem_pages;
332 	struct inmem_pages *cur = NULL;
333 
334 	f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
335 
336 	mutex_lock(&fi->inmem_lock);
337 	list_for_each_entry(cur, head, list) {
338 		if (cur->page == page)
339 			break;
340 	}
341 
342 	f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
343 	list_del(&cur->list);
344 	mutex_unlock(&fi->inmem_lock);
345 
346 	dec_page_count(sbi, F2FS_INMEM_PAGES);
347 	kmem_cache_free(inmem_entry_slab, cur);
348 
349 	ClearPageUptodate(page);
350 	set_page_private(page, 0);
351 	ClearPagePrivate(page);
352 	f2fs_put_page(page, 0);
353 
354 	trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
355 }
356 
357 static int __f2fs_commit_inmem_pages(struct inode *inode)
358 {
359 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
360 	struct f2fs_inode_info *fi = F2FS_I(inode);
361 	struct inmem_pages *cur, *tmp;
362 	struct f2fs_io_info fio = {
363 		.sbi = sbi,
364 		.ino = inode->i_ino,
365 		.type = DATA,
366 		.op = REQ_OP_WRITE,
367 		.op_flags = REQ_SYNC | REQ_PRIO,
368 		.io_type = FS_DATA_IO,
369 	};
370 	struct list_head revoke_list;
371 	pgoff_t last_idx = ULONG_MAX;
372 	int err = 0;
373 
374 	INIT_LIST_HEAD(&revoke_list);
375 
376 	list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
377 		struct page *page = cur->page;
378 
379 		lock_page(page);
380 		if (page->mapping == inode->i_mapping) {
381 			trace_f2fs_commit_inmem_page(page, INMEM);
382 
383 			set_page_dirty(page);
384 			f2fs_wait_on_page_writeback(page, DATA, true);
385 			if (clear_page_dirty_for_io(page)) {
386 				inode_dec_dirty_pages(inode);
387 				f2fs_remove_dirty_inode(inode);
388 			}
389 retry:
390 			fio.page = page;
391 			fio.old_blkaddr = NULL_ADDR;
392 			fio.encrypted_page = NULL;
393 			fio.need_lock = LOCK_DONE;
394 			err = f2fs_do_write_data_page(&fio);
395 			if (err) {
396 				if (err == -ENOMEM) {
397 					congestion_wait(BLK_RW_ASYNC, HZ/50);
398 					cond_resched();
399 					goto retry;
400 				}
401 				unlock_page(page);
402 				break;
403 			}
404 			/* record old blkaddr for revoking */
405 			cur->old_addr = fio.old_blkaddr;
406 			last_idx = page->index;
407 		}
408 		unlock_page(page);
409 		list_move_tail(&cur->list, &revoke_list);
410 	}
411 
412 	if (last_idx != ULONG_MAX)
413 		f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
414 
415 	if (err) {
416 		/*
417 		 * try to revoke all committed pages, but still we could fail
418 		 * due to no memory or other reason, if that happened, EAGAIN
419 		 * will be returned, which means in such case, transaction is
420 		 * already not integrity, caller should use journal to do the
421 		 * recovery or rewrite & commit last transaction. For other
422 		 * error number, revoking was done by filesystem itself.
423 		 */
424 		err = __revoke_inmem_pages(inode, &revoke_list, false, true);
425 
426 		/* drop all uncommitted pages */
427 		__revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
428 	} else {
429 		__revoke_inmem_pages(inode, &revoke_list, false, false);
430 	}
431 
432 	return err;
433 }
434 
435 int f2fs_commit_inmem_pages(struct inode *inode)
436 {
437 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
438 	struct f2fs_inode_info *fi = F2FS_I(inode);
439 	int err;
440 
441 	f2fs_balance_fs(sbi, true);
442 	f2fs_lock_op(sbi);
443 
444 	set_inode_flag(inode, FI_ATOMIC_COMMIT);
445 
446 	mutex_lock(&fi->inmem_lock);
447 	err = __f2fs_commit_inmem_pages(inode);
448 
449 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
450 	if (!list_empty(&fi->inmem_ilist))
451 		list_del_init(&fi->inmem_ilist);
452 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
453 	mutex_unlock(&fi->inmem_lock);
454 
455 	clear_inode_flag(inode, FI_ATOMIC_COMMIT);
456 
457 	f2fs_unlock_op(sbi);
458 	return err;
459 }
460 
461 /*
462  * This function balances dirty node and dentry pages.
463  * In addition, it controls garbage collection.
464  */
465 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
466 {
467 #ifdef CONFIG_F2FS_FAULT_INJECTION
468 	if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
469 		f2fs_show_injection_info(FAULT_CHECKPOINT);
470 		f2fs_stop_checkpoint(sbi, false);
471 	}
472 #endif
473 
474 	/* balance_fs_bg is able to be pending */
475 	if (need && excess_cached_nats(sbi))
476 		f2fs_balance_fs_bg(sbi);
477 
478 	/*
479 	 * We should do GC or end up with checkpoint, if there are so many dirty
480 	 * dir/node pages without enough free segments.
481 	 */
482 	if (has_not_enough_free_secs(sbi, 0, 0)) {
483 		mutex_lock(&sbi->gc_mutex);
484 		f2fs_gc(sbi, false, false, NULL_SEGNO);
485 	}
486 }
487 
488 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
489 {
490 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
491 		return;
492 
493 	/* try to shrink extent cache when there is no enough memory */
494 	if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
495 		f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
496 
497 	/* check the # of cached NAT entries */
498 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
499 		f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
500 
501 	if (!f2fs_available_free_memory(sbi, FREE_NIDS))
502 		f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
503 	else
504 		f2fs_build_free_nids(sbi, false, false);
505 
506 	if (!is_idle(sbi) && !excess_dirty_nats(sbi))
507 		return;
508 
509 	/* checkpoint is the only way to shrink partial cached entries */
510 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
511 			!f2fs_available_free_memory(sbi, INO_ENTRIES) ||
512 			excess_prefree_segs(sbi) ||
513 			excess_dirty_nats(sbi) ||
514 			f2fs_time_over(sbi, CP_TIME)) {
515 		if (test_opt(sbi, DATA_FLUSH)) {
516 			struct blk_plug plug;
517 
518 			blk_start_plug(&plug);
519 			f2fs_sync_dirty_inodes(sbi, FILE_INODE);
520 			blk_finish_plug(&plug);
521 		}
522 		f2fs_sync_fs(sbi->sb, true);
523 		stat_inc_bg_cp_count(sbi->stat_info);
524 	}
525 }
526 
527 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
528 				struct block_device *bdev)
529 {
530 	struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
531 	int ret;
532 
533 	bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
534 	bio_set_dev(bio, bdev);
535 	ret = submit_bio_wait(bio);
536 	bio_put(bio);
537 
538 	trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
539 				test_opt(sbi, FLUSH_MERGE), ret);
540 	return ret;
541 }
542 
543 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
544 {
545 	int ret = 0;
546 	int i;
547 
548 	if (!sbi->s_ndevs)
549 		return __submit_flush_wait(sbi, sbi->sb->s_bdev);
550 
551 	for (i = 0; i < sbi->s_ndevs; i++) {
552 		if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
553 			continue;
554 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
555 		if (ret)
556 			break;
557 	}
558 	return ret;
559 }
560 
561 static int issue_flush_thread(void *data)
562 {
563 	struct f2fs_sb_info *sbi = data;
564 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
565 	wait_queue_head_t *q = &fcc->flush_wait_queue;
566 repeat:
567 	if (kthread_should_stop())
568 		return 0;
569 
570 	sb_start_intwrite(sbi->sb);
571 
572 	if (!llist_empty(&fcc->issue_list)) {
573 		struct flush_cmd *cmd, *next;
574 		int ret;
575 
576 		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
577 		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
578 
579 		cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
580 
581 		ret = submit_flush_wait(sbi, cmd->ino);
582 		atomic_inc(&fcc->issued_flush);
583 
584 		llist_for_each_entry_safe(cmd, next,
585 					  fcc->dispatch_list, llnode) {
586 			cmd->ret = ret;
587 			complete(&cmd->wait);
588 		}
589 		fcc->dispatch_list = NULL;
590 	}
591 
592 	sb_end_intwrite(sbi->sb);
593 
594 	wait_event_interruptible(*q,
595 		kthread_should_stop() || !llist_empty(&fcc->issue_list));
596 	goto repeat;
597 }
598 
599 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
600 {
601 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
602 	struct flush_cmd cmd;
603 	int ret;
604 
605 	if (test_opt(sbi, NOBARRIER))
606 		return 0;
607 
608 	if (!test_opt(sbi, FLUSH_MERGE)) {
609 		ret = submit_flush_wait(sbi, ino);
610 		atomic_inc(&fcc->issued_flush);
611 		return ret;
612 	}
613 
614 	if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
615 		ret = submit_flush_wait(sbi, ino);
616 		atomic_dec(&fcc->issing_flush);
617 
618 		atomic_inc(&fcc->issued_flush);
619 		return ret;
620 	}
621 
622 	cmd.ino = ino;
623 	init_completion(&cmd.wait);
624 
625 	llist_add(&cmd.llnode, &fcc->issue_list);
626 
627 	/* update issue_list before we wake up issue_flush thread */
628 	smp_mb();
629 
630 	if (waitqueue_active(&fcc->flush_wait_queue))
631 		wake_up(&fcc->flush_wait_queue);
632 
633 	if (fcc->f2fs_issue_flush) {
634 		wait_for_completion(&cmd.wait);
635 		atomic_dec(&fcc->issing_flush);
636 	} else {
637 		struct llist_node *list;
638 
639 		list = llist_del_all(&fcc->issue_list);
640 		if (!list) {
641 			wait_for_completion(&cmd.wait);
642 			atomic_dec(&fcc->issing_flush);
643 		} else {
644 			struct flush_cmd *tmp, *next;
645 
646 			ret = submit_flush_wait(sbi, ino);
647 
648 			llist_for_each_entry_safe(tmp, next, list, llnode) {
649 				if (tmp == &cmd) {
650 					cmd.ret = ret;
651 					atomic_dec(&fcc->issing_flush);
652 					continue;
653 				}
654 				tmp->ret = ret;
655 				complete(&tmp->wait);
656 			}
657 		}
658 	}
659 
660 	return cmd.ret;
661 }
662 
663 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
664 {
665 	dev_t dev = sbi->sb->s_bdev->bd_dev;
666 	struct flush_cmd_control *fcc;
667 	int err = 0;
668 
669 	if (SM_I(sbi)->fcc_info) {
670 		fcc = SM_I(sbi)->fcc_info;
671 		if (fcc->f2fs_issue_flush)
672 			return err;
673 		goto init_thread;
674 	}
675 
676 	fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
677 	if (!fcc)
678 		return -ENOMEM;
679 	atomic_set(&fcc->issued_flush, 0);
680 	atomic_set(&fcc->issing_flush, 0);
681 	init_waitqueue_head(&fcc->flush_wait_queue);
682 	init_llist_head(&fcc->issue_list);
683 	SM_I(sbi)->fcc_info = fcc;
684 	if (!test_opt(sbi, FLUSH_MERGE))
685 		return err;
686 
687 init_thread:
688 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
689 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
690 	if (IS_ERR(fcc->f2fs_issue_flush)) {
691 		err = PTR_ERR(fcc->f2fs_issue_flush);
692 		kfree(fcc);
693 		SM_I(sbi)->fcc_info = NULL;
694 		return err;
695 	}
696 
697 	return err;
698 }
699 
700 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
701 {
702 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
703 
704 	if (fcc && fcc->f2fs_issue_flush) {
705 		struct task_struct *flush_thread = fcc->f2fs_issue_flush;
706 
707 		fcc->f2fs_issue_flush = NULL;
708 		kthread_stop(flush_thread);
709 	}
710 	if (free) {
711 		kfree(fcc);
712 		SM_I(sbi)->fcc_info = NULL;
713 	}
714 }
715 
716 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
717 {
718 	int ret = 0, i;
719 
720 	if (!sbi->s_ndevs)
721 		return 0;
722 
723 	for (i = 1; i < sbi->s_ndevs; i++) {
724 		if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
725 			continue;
726 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
727 		if (ret)
728 			break;
729 
730 		spin_lock(&sbi->dev_lock);
731 		f2fs_clear_bit(i, (char *)&sbi->dirty_device);
732 		spin_unlock(&sbi->dev_lock);
733 	}
734 
735 	return ret;
736 }
737 
738 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
739 		enum dirty_type dirty_type)
740 {
741 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
742 
743 	/* need not be added */
744 	if (IS_CURSEG(sbi, segno))
745 		return;
746 
747 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
748 		dirty_i->nr_dirty[dirty_type]++;
749 
750 	if (dirty_type == DIRTY) {
751 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
752 		enum dirty_type t = sentry->type;
753 
754 		if (unlikely(t >= DIRTY)) {
755 			f2fs_bug_on(sbi, 1);
756 			return;
757 		}
758 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
759 			dirty_i->nr_dirty[t]++;
760 	}
761 }
762 
763 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
764 		enum dirty_type dirty_type)
765 {
766 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
767 
768 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
769 		dirty_i->nr_dirty[dirty_type]--;
770 
771 	if (dirty_type == DIRTY) {
772 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
773 		enum dirty_type t = sentry->type;
774 
775 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
776 			dirty_i->nr_dirty[t]--;
777 
778 		if (get_valid_blocks(sbi, segno, true) == 0)
779 			clear_bit(GET_SEC_FROM_SEG(sbi, segno),
780 						dirty_i->victim_secmap);
781 	}
782 }
783 
784 /*
785  * Should not occur error such as -ENOMEM.
786  * Adding dirty entry into seglist is not critical operation.
787  * If a given segment is one of current working segments, it won't be added.
788  */
789 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
790 {
791 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
792 	unsigned short valid_blocks;
793 
794 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
795 		return;
796 
797 	mutex_lock(&dirty_i->seglist_lock);
798 
799 	valid_blocks = get_valid_blocks(sbi, segno, false);
800 
801 	if (valid_blocks == 0) {
802 		__locate_dirty_segment(sbi, segno, PRE);
803 		__remove_dirty_segment(sbi, segno, DIRTY);
804 	} else if (valid_blocks < sbi->blocks_per_seg) {
805 		__locate_dirty_segment(sbi, segno, DIRTY);
806 	} else {
807 		/* Recovery routine with SSR needs this */
808 		__remove_dirty_segment(sbi, segno, DIRTY);
809 	}
810 
811 	mutex_unlock(&dirty_i->seglist_lock);
812 }
813 
814 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
815 		struct block_device *bdev, block_t lstart,
816 		block_t start, block_t len)
817 {
818 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
819 	struct list_head *pend_list;
820 	struct discard_cmd *dc;
821 
822 	f2fs_bug_on(sbi, !len);
823 
824 	pend_list = &dcc->pend_list[plist_idx(len)];
825 
826 	dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
827 	INIT_LIST_HEAD(&dc->list);
828 	dc->bdev = bdev;
829 	dc->lstart = lstart;
830 	dc->start = start;
831 	dc->len = len;
832 	dc->ref = 0;
833 	dc->state = D_PREP;
834 	dc->error = 0;
835 	init_completion(&dc->wait);
836 	list_add_tail(&dc->list, pend_list);
837 	atomic_inc(&dcc->discard_cmd_cnt);
838 	dcc->undiscard_blks += len;
839 
840 	return dc;
841 }
842 
843 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
844 				struct block_device *bdev, block_t lstart,
845 				block_t start, block_t len,
846 				struct rb_node *parent, struct rb_node **p)
847 {
848 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
849 	struct discard_cmd *dc;
850 
851 	dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
852 
853 	rb_link_node(&dc->rb_node, parent, p);
854 	rb_insert_color(&dc->rb_node, &dcc->root);
855 
856 	return dc;
857 }
858 
859 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
860 							struct discard_cmd *dc)
861 {
862 	if (dc->state == D_DONE)
863 		atomic_dec(&dcc->issing_discard);
864 
865 	list_del(&dc->list);
866 	rb_erase(&dc->rb_node, &dcc->root);
867 	dcc->undiscard_blks -= dc->len;
868 
869 	kmem_cache_free(discard_cmd_slab, dc);
870 
871 	atomic_dec(&dcc->discard_cmd_cnt);
872 }
873 
874 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
875 							struct discard_cmd *dc)
876 {
877 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
878 
879 	trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
880 
881 	f2fs_bug_on(sbi, dc->ref);
882 
883 	if (dc->error == -EOPNOTSUPP)
884 		dc->error = 0;
885 
886 	if (dc->error)
887 		f2fs_msg(sbi->sb, KERN_INFO,
888 			"Issue discard(%u, %u, %u) failed, ret: %d",
889 			dc->lstart, dc->start, dc->len, dc->error);
890 	__detach_discard_cmd(dcc, dc);
891 }
892 
893 static void f2fs_submit_discard_endio(struct bio *bio)
894 {
895 	struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
896 
897 	dc->error = blk_status_to_errno(bio->bi_status);
898 	dc->state = D_DONE;
899 	complete_all(&dc->wait);
900 	bio_put(bio);
901 }
902 
903 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
904 				block_t start, block_t end)
905 {
906 #ifdef CONFIG_F2FS_CHECK_FS
907 	struct seg_entry *sentry;
908 	unsigned int segno;
909 	block_t blk = start;
910 	unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
911 	unsigned long *map;
912 
913 	while (blk < end) {
914 		segno = GET_SEGNO(sbi, blk);
915 		sentry = get_seg_entry(sbi, segno);
916 		offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
917 
918 		if (end < START_BLOCK(sbi, segno + 1))
919 			size = GET_BLKOFF_FROM_SEG0(sbi, end);
920 		else
921 			size = max_blocks;
922 		map = (unsigned long *)(sentry->cur_valid_map);
923 		offset = __find_rev_next_bit(map, size, offset);
924 		f2fs_bug_on(sbi, offset != size);
925 		blk = START_BLOCK(sbi, segno + 1);
926 	}
927 #endif
928 }
929 
930 static void __init_discard_policy(struct f2fs_sb_info *sbi,
931 				struct discard_policy *dpolicy,
932 				int discard_type, unsigned int granularity)
933 {
934 	/* common policy */
935 	dpolicy->type = discard_type;
936 	dpolicy->sync = true;
937 	dpolicy->granularity = granularity;
938 
939 	dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
940 	dpolicy->io_aware_gran = MAX_PLIST_NUM;
941 
942 	if (discard_type == DPOLICY_BG) {
943 		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
944 		dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
945 		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
946 		dpolicy->io_aware = true;
947 		dpolicy->sync = false;
948 		if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
949 			dpolicy->granularity = 1;
950 			dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
951 		}
952 	} else if (discard_type == DPOLICY_FORCE) {
953 		dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
954 		dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
955 		dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
956 		dpolicy->io_aware = false;
957 	} else if (discard_type == DPOLICY_FSTRIM) {
958 		dpolicy->io_aware = false;
959 	} else if (discard_type == DPOLICY_UMOUNT) {
960 		dpolicy->max_requests = UINT_MAX;
961 		dpolicy->io_aware = false;
962 	}
963 }
964 
965 
966 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
967 static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
968 						struct discard_policy *dpolicy,
969 						struct discard_cmd *dc)
970 {
971 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
972 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
973 					&(dcc->fstrim_list) : &(dcc->wait_list);
974 	struct bio *bio = NULL;
975 	int flag = dpolicy->sync ? REQ_SYNC : 0;
976 
977 	if (dc->state != D_PREP)
978 		return;
979 
980 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
981 		return;
982 
983 	trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
984 
985 	dc->error = __blkdev_issue_discard(dc->bdev,
986 				SECTOR_FROM_BLOCK(dc->start),
987 				SECTOR_FROM_BLOCK(dc->len),
988 				GFP_NOFS, 0, &bio);
989 	if (!dc->error) {
990 		/* should keep before submission to avoid D_DONE right away */
991 		dc->state = D_SUBMIT;
992 		atomic_inc(&dcc->issued_discard);
993 		atomic_inc(&dcc->issing_discard);
994 		if (bio) {
995 			bio->bi_private = dc;
996 			bio->bi_end_io = f2fs_submit_discard_endio;
997 			bio->bi_opf |= flag;
998 			submit_bio(bio);
999 			list_move_tail(&dc->list, wait_list);
1000 			__check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
1001 
1002 			f2fs_update_iostat(sbi, FS_DISCARD, 1);
1003 		}
1004 	} else {
1005 		__remove_discard_cmd(sbi, dc);
1006 	}
1007 }
1008 
1009 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1010 				struct block_device *bdev, block_t lstart,
1011 				block_t start, block_t len,
1012 				struct rb_node **insert_p,
1013 				struct rb_node *insert_parent)
1014 {
1015 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1016 	struct rb_node **p;
1017 	struct rb_node *parent = NULL;
1018 	struct discard_cmd *dc = NULL;
1019 
1020 	if (insert_p && insert_parent) {
1021 		parent = insert_parent;
1022 		p = insert_p;
1023 		goto do_insert;
1024 	}
1025 
1026 	p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
1027 do_insert:
1028 	dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1029 	if (!dc)
1030 		return NULL;
1031 
1032 	return dc;
1033 }
1034 
1035 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1036 						struct discard_cmd *dc)
1037 {
1038 	list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1039 }
1040 
1041 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1042 				struct discard_cmd *dc, block_t blkaddr)
1043 {
1044 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1045 	struct discard_info di = dc->di;
1046 	bool modified = false;
1047 
1048 	if (dc->state == D_DONE || dc->len == 1) {
1049 		__remove_discard_cmd(sbi, dc);
1050 		return;
1051 	}
1052 
1053 	dcc->undiscard_blks -= di.len;
1054 
1055 	if (blkaddr > di.lstart) {
1056 		dc->len = blkaddr - dc->lstart;
1057 		dcc->undiscard_blks += dc->len;
1058 		__relocate_discard_cmd(dcc, dc);
1059 		modified = true;
1060 	}
1061 
1062 	if (blkaddr < di.lstart + di.len - 1) {
1063 		if (modified) {
1064 			__insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1065 					di.start + blkaddr + 1 - di.lstart,
1066 					di.lstart + di.len - 1 - blkaddr,
1067 					NULL, NULL);
1068 		} else {
1069 			dc->lstart++;
1070 			dc->len--;
1071 			dc->start++;
1072 			dcc->undiscard_blks += dc->len;
1073 			__relocate_discard_cmd(dcc, dc);
1074 		}
1075 	}
1076 }
1077 
1078 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1079 				struct block_device *bdev, block_t lstart,
1080 				block_t start, block_t len)
1081 {
1082 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1083 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1084 	struct discard_cmd *dc;
1085 	struct discard_info di = {0};
1086 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1087 	block_t end = lstart + len;
1088 
1089 	mutex_lock(&dcc->cmd_lock);
1090 
1091 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1092 					NULL, lstart,
1093 					(struct rb_entry **)&prev_dc,
1094 					(struct rb_entry **)&next_dc,
1095 					&insert_p, &insert_parent, true);
1096 	if (dc)
1097 		prev_dc = dc;
1098 
1099 	if (!prev_dc) {
1100 		di.lstart = lstart;
1101 		di.len = next_dc ? next_dc->lstart - lstart : len;
1102 		di.len = min(di.len, len);
1103 		di.start = start;
1104 	}
1105 
1106 	while (1) {
1107 		struct rb_node *node;
1108 		bool merged = false;
1109 		struct discard_cmd *tdc = NULL;
1110 
1111 		if (prev_dc) {
1112 			di.lstart = prev_dc->lstart + prev_dc->len;
1113 			if (di.lstart < lstart)
1114 				di.lstart = lstart;
1115 			if (di.lstart >= end)
1116 				break;
1117 
1118 			if (!next_dc || next_dc->lstart > end)
1119 				di.len = end - di.lstart;
1120 			else
1121 				di.len = next_dc->lstart - di.lstart;
1122 			di.start = start + di.lstart - lstart;
1123 		}
1124 
1125 		if (!di.len)
1126 			goto next;
1127 
1128 		if (prev_dc && prev_dc->state == D_PREP &&
1129 			prev_dc->bdev == bdev &&
1130 			__is_discard_back_mergeable(&di, &prev_dc->di)) {
1131 			prev_dc->di.len += di.len;
1132 			dcc->undiscard_blks += di.len;
1133 			__relocate_discard_cmd(dcc, prev_dc);
1134 			di = prev_dc->di;
1135 			tdc = prev_dc;
1136 			merged = true;
1137 		}
1138 
1139 		if (next_dc && next_dc->state == D_PREP &&
1140 			next_dc->bdev == bdev &&
1141 			__is_discard_front_mergeable(&di, &next_dc->di)) {
1142 			next_dc->di.lstart = di.lstart;
1143 			next_dc->di.len += di.len;
1144 			next_dc->di.start = di.start;
1145 			dcc->undiscard_blks += di.len;
1146 			__relocate_discard_cmd(dcc, next_dc);
1147 			if (tdc)
1148 				__remove_discard_cmd(sbi, tdc);
1149 			merged = true;
1150 		}
1151 
1152 		if (!merged) {
1153 			__insert_discard_tree(sbi, bdev, di.lstart, di.start,
1154 							di.len, NULL, NULL);
1155 		}
1156  next:
1157 		prev_dc = next_dc;
1158 		if (!prev_dc)
1159 			break;
1160 
1161 		node = rb_next(&prev_dc->rb_node);
1162 		next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1163 	}
1164 
1165 	mutex_unlock(&dcc->cmd_lock);
1166 }
1167 
1168 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1169 		struct block_device *bdev, block_t blkstart, block_t blklen)
1170 {
1171 	block_t lblkstart = blkstart;
1172 
1173 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
1174 
1175 	if (sbi->s_ndevs) {
1176 		int devi = f2fs_target_device_index(sbi, blkstart);
1177 
1178 		blkstart -= FDEV(devi).start_blk;
1179 	}
1180 	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1181 	return 0;
1182 }
1183 
1184 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1185 					struct discard_policy *dpolicy)
1186 {
1187 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1188 	struct list_head *pend_list;
1189 	struct discard_cmd *dc, *tmp;
1190 	struct blk_plug plug;
1191 	int i, iter = 0, issued = 0;
1192 	bool io_interrupted = false;
1193 
1194 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1195 		if (i + 1 < dpolicy->granularity)
1196 			break;
1197 		pend_list = &dcc->pend_list[i];
1198 
1199 		mutex_lock(&dcc->cmd_lock);
1200 		if (list_empty(pend_list))
1201 			goto next;
1202 		f2fs_bug_on(sbi,
1203 			!f2fs_check_rb_tree_consistence(sbi, &dcc->root));
1204 		blk_start_plug(&plug);
1205 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1206 			f2fs_bug_on(sbi, dc->state != D_PREP);
1207 
1208 			if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1209 								!is_idle(sbi)) {
1210 				io_interrupted = true;
1211 				goto skip;
1212 			}
1213 
1214 			__submit_discard_cmd(sbi, dpolicy, dc);
1215 			issued++;
1216 skip:
1217 			if (++iter >= dpolicy->max_requests)
1218 				break;
1219 		}
1220 		blk_finish_plug(&plug);
1221 next:
1222 		mutex_unlock(&dcc->cmd_lock);
1223 
1224 		if (iter >= dpolicy->max_requests)
1225 			break;
1226 	}
1227 
1228 	if (!issued && io_interrupted)
1229 		issued = -1;
1230 
1231 	return issued;
1232 }
1233 
1234 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1235 {
1236 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1237 	struct list_head *pend_list;
1238 	struct discard_cmd *dc, *tmp;
1239 	int i;
1240 	bool dropped = false;
1241 
1242 	mutex_lock(&dcc->cmd_lock);
1243 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1244 		pend_list = &dcc->pend_list[i];
1245 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1246 			f2fs_bug_on(sbi, dc->state != D_PREP);
1247 			__remove_discard_cmd(sbi, dc);
1248 			dropped = true;
1249 		}
1250 	}
1251 	mutex_unlock(&dcc->cmd_lock);
1252 
1253 	return dropped;
1254 }
1255 
1256 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1257 {
1258 	__drop_discard_cmd(sbi);
1259 }
1260 
1261 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1262 							struct discard_cmd *dc)
1263 {
1264 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1265 	unsigned int len = 0;
1266 
1267 	wait_for_completion_io(&dc->wait);
1268 	mutex_lock(&dcc->cmd_lock);
1269 	f2fs_bug_on(sbi, dc->state != D_DONE);
1270 	dc->ref--;
1271 	if (!dc->ref) {
1272 		if (!dc->error)
1273 			len = dc->len;
1274 		__remove_discard_cmd(sbi, dc);
1275 	}
1276 	mutex_unlock(&dcc->cmd_lock);
1277 
1278 	return len;
1279 }
1280 
1281 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1282 						struct discard_policy *dpolicy,
1283 						block_t start, block_t end)
1284 {
1285 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1286 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1287 					&(dcc->fstrim_list) : &(dcc->wait_list);
1288 	struct discard_cmd *dc, *tmp;
1289 	bool need_wait;
1290 	unsigned int trimmed = 0;
1291 
1292 next:
1293 	need_wait = false;
1294 
1295 	mutex_lock(&dcc->cmd_lock);
1296 	list_for_each_entry_safe(dc, tmp, wait_list, list) {
1297 		if (dc->lstart + dc->len <= start || end <= dc->lstart)
1298 			continue;
1299 		if (dc->len < dpolicy->granularity)
1300 			continue;
1301 		if (dc->state == D_DONE && !dc->ref) {
1302 			wait_for_completion_io(&dc->wait);
1303 			if (!dc->error)
1304 				trimmed += dc->len;
1305 			__remove_discard_cmd(sbi, dc);
1306 		} else {
1307 			dc->ref++;
1308 			need_wait = true;
1309 			break;
1310 		}
1311 	}
1312 	mutex_unlock(&dcc->cmd_lock);
1313 
1314 	if (need_wait) {
1315 		trimmed += __wait_one_discard_bio(sbi, dc);
1316 		goto next;
1317 	}
1318 
1319 	return trimmed;
1320 }
1321 
1322 static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1323 						struct discard_policy *dpolicy)
1324 {
1325 	struct discard_policy dp;
1326 
1327 	if (dpolicy) {
1328 		__wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1329 		return;
1330 	}
1331 
1332 	/* wait all */
1333 	__init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1334 	__wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1335 	__init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1336 	__wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1337 }
1338 
1339 /* This should be covered by global mutex, &sit_i->sentry_lock */
1340 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1341 {
1342 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1343 	struct discard_cmd *dc;
1344 	bool need_wait = false;
1345 
1346 	mutex_lock(&dcc->cmd_lock);
1347 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1348 							NULL, blkaddr);
1349 	if (dc) {
1350 		if (dc->state == D_PREP) {
1351 			__punch_discard_cmd(sbi, dc, blkaddr);
1352 		} else {
1353 			dc->ref++;
1354 			need_wait = true;
1355 		}
1356 	}
1357 	mutex_unlock(&dcc->cmd_lock);
1358 
1359 	if (need_wait)
1360 		__wait_one_discard_bio(sbi, dc);
1361 }
1362 
1363 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1364 {
1365 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1366 
1367 	if (dcc && dcc->f2fs_issue_discard) {
1368 		struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1369 
1370 		dcc->f2fs_issue_discard = NULL;
1371 		kthread_stop(discard_thread);
1372 	}
1373 }
1374 
1375 /* This comes from f2fs_put_super */
1376 bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1377 {
1378 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1379 	struct discard_policy dpolicy;
1380 	bool dropped;
1381 
1382 	__init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1383 					dcc->discard_granularity);
1384 	__issue_discard_cmd(sbi, &dpolicy);
1385 	dropped = __drop_discard_cmd(sbi);
1386 
1387 	/* just to make sure there is no pending discard commands */
1388 	__wait_all_discard_cmd(sbi, NULL);
1389 	return dropped;
1390 }
1391 
1392 static int issue_discard_thread(void *data)
1393 {
1394 	struct f2fs_sb_info *sbi = data;
1395 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1396 	wait_queue_head_t *q = &dcc->discard_wait_queue;
1397 	struct discard_policy dpolicy;
1398 	unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1399 	int issued;
1400 
1401 	set_freezable();
1402 
1403 	do {
1404 		__init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1405 					dcc->discard_granularity);
1406 
1407 		wait_event_interruptible_timeout(*q,
1408 				kthread_should_stop() || freezing(current) ||
1409 				dcc->discard_wake,
1410 				msecs_to_jiffies(wait_ms));
1411 
1412 		if (dcc->discard_wake)
1413 			dcc->discard_wake = 0;
1414 
1415 		if (try_to_freeze())
1416 			continue;
1417 		if (f2fs_readonly(sbi->sb))
1418 			continue;
1419 		if (kthread_should_stop())
1420 			return 0;
1421 		if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1422 			wait_ms = dpolicy.max_interval;
1423 			continue;
1424 		}
1425 
1426 		if (sbi->gc_mode == GC_URGENT)
1427 			__init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1428 
1429 		sb_start_intwrite(sbi->sb);
1430 
1431 		issued = __issue_discard_cmd(sbi, &dpolicy);
1432 		if (issued > 0) {
1433 			__wait_all_discard_cmd(sbi, &dpolicy);
1434 			wait_ms = dpolicy.min_interval;
1435 		} else if (issued == -1){
1436 			wait_ms = dpolicy.mid_interval;
1437 		} else {
1438 			wait_ms = dpolicy.max_interval;
1439 		}
1440 
1441 		sb_end_intwrite(sbi->sb);
1442 
1443 	} while (!kthread_should_stop());
1444 	return 0;
1445 }
1446 
1447 #ifdef CONFIG_BLK_DEV_ZONED
1448 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1449 		struct block_device *bdev, block_t blkstart, block_t blklen)
1450 {
1451 	sector_t sector, nr_sects;
1452 	block_t lblkstart = blkstart;
1453 	int devi = 0;
1454 
1455 	if (sbi->s_ndevs) {
1456 		devi = f2fs_target_device_index(sbi, blkstart);
1457 		blkstart -= FDEV(devi).start_blk;
1458 	}
1459 
1460 	/*
1461 	 * We need to know the type of the zone: for conventional zones,
1462 	 * use regular discard if the drive supports it. For sequential
1463 	 * zones, reset the zone write pointer.
1464 	 */
1465 	switch (get_blkz_type(sbi, bdev, blkstart)) {
1466 
1467 	case BLK_ZONE_TYPE_CONVENTIONAL:
1468 		if (!blk_queue_discard(bdev_get_queue(bdev)))
1469 			return 0;
1470 		return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1471 	case BLK_ZONE_TYPE_SEQWRITE_REQ:
1472 	case BLK_ZONE_TYPE_SEQWRITE_PREF:
1473 		sector = SECTOR_FROM_BLOCK(blkstart);
1474 		nr_sects = SECTOR_FROM_BLOCK(blklen);
1475 
1476 		if (sector & (bdev_zone_sectors(bdev) - 1) ||
1477 				nr_sects != bdev_zone_sectors(bdev)) {
1478 			f2fs_msg(sbi->sb, KERN_INFO,
1479 				"(%d) %s: Unaligned discard attempted (block %x + %x)",
1480 				devi, sbi->s_ndevs ? FDEV(devi).path: "",
1481 				blkstart, blklen);
1482 			return -EIO;
1483 		}
1484 		trace_f2fs_issue_reset_zone(bdev, blkstart);
1485 		return blkdev_reset_zones(bdev, sector,
1486 					  nr_sects, GFP_NOFS);
1487 	default:
1488 		/* Unknown zone type: broken device ? */
1489 		return -EIO;
1490 	}
1491 }
1492 #endif
1493 
1494 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1495 		struct block_device *bdev, block_t blkstart, block_t blklen)
1496 {
1497 #ifdef CONFIG_BLK_DEV_ZONED
1498 	if (f2fs_sb_has_blkzoned(sbi->sb) &&
1499 				bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1500 		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1501 #endif
1502 	return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1503 }
1504 
1505 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1506 				block_t blkstart, block_t blklen)
1507 {
1508 	sector_t start = blkstart, len = 0;
1509 	struct block_device *bdev;
1510 	struct seg_entry *se;
1511 	unsigned int offset;
1512 	block_t i;
1513 	int err = 0;
1514 
1515 	bdev = f2fs_target_device(sbi, blkstart, NULL);
1516 
1517 	for (i = blkstart; i < blkstart + blklen; i++, len++) {
1518 		if (i != start) {
1519 			struct block_device *bdev2 =
1520 				f2fs_target_device(sbi, i, NULL);
1521 
1522 			if (bdev2 != bdev) {
1523 				err = __issue_discard_async(sbi, bdev,
1524 						start, len);
1525 				if (err)
1526 					return err;
1527 				bdev = bdev2;
1528 				start = i;
1529 				len = 0;
1530 			}
1531 		}
1532 
1533 		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1534 		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1535 
1536 		if (!f2fs_test_and_set_bit(offset, se->discard_map))
1537 			sbi->discard_blks--;
1538 	}
1539 
1540 	if (len)
1541 		err = __issue_discard_async(sbi, bdev, start, len);
1542 	return err;
1543 }
1544 
1545 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1546 							bool check_only)
1547 {
1548 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1549 	int max_blocks = sbi->blocks_per_seg;
1550 	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1551 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1552 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1553 	unsigned long *discard_map = (unsigned long *)se->discard_map;
1554 	unsigned long *dmap = SIT_I(sbi)->tmp_map;
1555 	unsigned int start = 0, end = -1;
1556 	bool force = (cpc->reason & CP_DISCARD);
1557 	struct discard_entry *de = NULL;
1558 	struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1559 	int i;
1560 
1561 	if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1562 		return false;
1563 
1564 	if (!force) {
1565 		if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1566 			SM_I(sbi)->dcc_info->nr_discards >=
1567 				SM_I(sbi)->dcc_info->max_discards)
1568 			return false;
1569 	}
1570 
1571 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1572 	for (i = 0; i < entries; i++)
1573 		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1574 				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1575 
1576 	while (force || SM_I(sbi)->dcc_info->nr_discards <=
1577 				SM_I(sbi)->dcc_info->max_discards) {
1578 		start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1579 		if (start >= max_blocks)
1580 			break;
1581 
1582 		end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1583 		if (force && start && end != max_blocks
1584 					&& (end - start) < cpc->trim_minlen)
1585 			continue;
1586 
1587 		if (check_only)
1588 			return true;
1589 
1590 		if (!de) {
1591 			de = f2fs_kmem_cache_alloc(discard_entry_slab,
1592 								GFP_F2FS_ZERO);
1593 			de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1594 			list_add_tail(&de->list, head);
1595 		}
1596 
1597 		for (i = start; i < end; i++)
1598 			__set_bit_le(i, (void *)de->discard_map);
1599 
1600 		SM_I(sbi)->dcc_info->nr_discards += end - start;
1601 	}
1602 	return false;
1603 }
1604 
1605 static void release_discard_addr(struct discard_entry *entry)
1606 {
1607 	list_del(&entry->list);
1608 	kmem_cache_free(discard_entry_slab, entry);
1609 }
1610 
1611 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1612 {
1613 	struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1614 	struct discard_entry *entry, *this;
1615 
1616 	/* drop caches */
1617 	list_for_each_entry_safe(entry, this, head, list)
1618 		release_discard_addr(entry);
1619 }
1620 
1621 /*
1622  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1623  */
1624 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1625 {
1626 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1627 	unsigned int segno;
1628 
1629 	mutex_lock(&dirty_i->seglist_lock);
1630 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1631 		__set_test_and_free(sbi, segno);
1632 	mutex_unlock(&dirty_i->seglist_lock);
1633 }
1634 
1635 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1636 						struct cp_control *cpc)
1637 {
1638 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1639 	struct list_head *head = &dcc->entry_list;
1640 	struct discard_entry *entry, *this;
1641 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1642 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1643 	unsigned int start = 0, end = -1;
1644 	unsigned int secno, start_segno;
1645 	bool force = (cpc->reason & CP_DISCARD);
1646 
1647 	mutex_lock(&dirty_i->seglist_lock);
1648 
1649 	while (1) {
1650 		int i;
1651 		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1652 		if (start >= MAIN_SEGS(sbi))
1653 			break;
1654 		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1655 								start + 1);
1656 
1657 		for (i = start; i < end; i++)
1658 			clear_bit(i, prefree_map);
1659 
1660 		dirty_i->nr_dirty[PRE] -= end - start;
1661 
1662 		if (!test_opt(sbi, DISCARD))
1663 			continue;
1664 
1665 		if (force && start >= cpc->trim_start &&
1666 					(end - 1) <= cpc->trim_end)
1667 				continue;
1668 
1669 		if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1670 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1671 				(end - start) << sbi->log_blocks_per_seg);
1672 			continue;
1673 		}
1674 next:
1675 		secno = GET_SEC_FROM_SEG(sbi, start);
1676 		start_segno = GET_SEG_FROM_SEC(sbi, secno);
1677 		if (!IS_CURSEC(sbi, secno) &&
1678 			!get_valid_blocks(sbi, start, true))
1679 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1680 				sbi->segs_per_sec << sbi->log_blocks_per_seg);
1681 
1682 		start = start_segno + sbi->segs_per_sec;
1683 		if (start < end)
1684 			goto next;
1685 		else
1686 			end = start - 1;
1687 	}
1688 	mutex_unlock(&dirty_i->seglist_lock);
1689 
1690 	/* send small discards */
1691 	list_for_each_entry_safe(entry, this, head, list) {
1692 		unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1693 		bool is_valid = test_bit_le(0, entry->discard_map);
1694 
1695 find_next:
1696 		if (is_valid) {
1697 			next_pos = find_next_zero_bit_le(entry->discard_map,
1698 					sbi->blocks_per_seg, cur_pos);
1699 			len = next_pos - cur_pos;
1700 
1701 			if (f2fs_sb_has_blkzoned(sbi->sb) ||
1702 			    (force && len < cpc->trim_minlen))
1703 				goto skip;
1704 
1705 			f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1706 									len);
1707 			total_len += len;
1708 		} else {
1709 			next_pos = find_next_bit_le(entry->discard_map,
1710 					sbi->blocks_per_seg, cur_pos);
1711 		}
1712 skip:
1713 		cur_pos = next_pos;
1714 		is_valid = !is_valid;
1715 
1716 		if (cur_pos < sbi->blocks_per_seg)
1717 			goto find_next;
1718 
1719 		release_discard_addr(entry);
1720 		dcc->nr_discards -= total_len;
1721 	}
1722 
1723 	wake_up_discard_thread(sbi, false);
1724 }
1725 
1726 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1727 {
1728 	dev_t dev = sbi->sb->s_bdev->bd_dev;
1729 	struct discard_cmd_control *dcc;
1730 	int err = 0, i;
1731 
1732 	if (SM_I(sbi)->dcc_info) {
1733 		dcc = SM_I(sbi)->dcc_info;
1734 		goto init_thread;
1735 	}
1736 
1737 	dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
1738 	if (!dcc)
1739 		return -ENOMEM;
1740 
1741 	dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
1742 	INIT_LIST_HEAD(&dcc->entry_list);
1743 	for (i = 0; i < MAX_PLIST_NUM; i++)
1744 		INIT_LIST_HEAD(&dcc->pend_list[i]);
1745 	INIT_LIST_HEAD(&dcc->wait_list);
1746 	INIT_LIST_HEAD(&dcc->fstrim_list);
1747 	mutex_init(&dcc->cmd_lock);
1748 	atomic_set(&dcc->issued_discard, 0);
1749 	atomic_set(&dcc->issing_discard, 0);
1750 	atomic_set(&dcc->discard_cmd_cnt, 0);
1751 	dcc->nr_discards = 0;
1752 	dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1753 	dcc->undiscard_blks = 0;
1754 	dcc->root = RB_ROOT;
1755 
1756 	init_waitqueue_head(&dcc->discard_wait_queue);
1757 	SM_I(sbi)->dcc_info = dcc;
1758 init_thread:
1759 	dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1760 				"f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1761 	if (IS_ERR(dcc->f2fs_issue_discard)) {
1762 		err = PTR_ERR(dcc->f2fs_issue_discard);
1763 		kfree(dcc);
1764 		SM_I(sbi)->dcc_info = NULL;
1765 		return err;
1766 	}
1767 
1768 	return err;
1769 }
1770 
1771 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1772 {
1773 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1774 
1775 	if (!dcc)
1776 		return;
1777 
1778 	f2fs_stop_discard_thread(sbi);
1779 
1780 	kfree(dcc);
1781 	SM_I(sbi)->dcc_info = NULL;
1782 }
1783 
1784 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1785 {
1786 	struct sit_info *sit_i = SIT_I(sbi);
1787 
1788 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1789 		sit_i->dirty_sentries++;
1790 		return false;
1791 	}
1792 
1793 	return true;
1794 }
1795 
1796 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1797 					unsigned int segno, int modified)
1798 {
1799 	struct seg_entry *se = get_seg_entry(sbi, segno);
1800 	se->type = type;
1801 	if (modified)
1802 		__mark_sit_entry_dirty(sbi, segno);
1803 }
1804 
1805 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1806 {
1807 	struct seg_entry *se;
1808 	unsigned int segno, offset;
1809 	long int new_vblocks;
1810 	bool exist;
1811 #ifdef CONFIG_F2FS_CHECK_FS
1812 	bool mir_exist;
1813 #endif
1814 
1815 	segno = GET_SEGNO(sbi, blkaddr);
1816 
1817 	se = get_seg_entry(sbi, segno);
1818 	new_vblocks = se->valid_blocks + del;
1819 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1820 
1821 	f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1822 				(new_vblocks > sbi->blocks_per_seg)));
1823 
1824 	se->valid_blocks = new_vblocks;
1825 	se->mtime = get_mtime(sbi, false);
1826 	if (se->mtime > SIT_I(sbi)->max_mtime)
1827 		SIT_I(sbi)->max_mtime = se->mtime;
1828 
1829 	/* Update valid block bitmap */
1830 	if (del > 0) {
1831 		exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
1832 #ifdef CONFIG_F2FS_CHECK_FS
1833 		mir_exist = f2fs_test_and_set_bit(offset,
1834 						se->cur_valid_map_mir);
1835 		if (unlikely(exist != mir_exist)) {
1836 			f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1837 				"when setting bitmap, blk:%u, old bit:%d",
1838 				blkaddr, exist);
1839 			f2fs_bug_on(sbi, 1);
1840 		}
1841 #endif
1842 		if (unlikely(exist)) {
1843 			f2fs_msg(sbi->sb, KERN_ERR,
1844 				"Bitmap was wrongly set, blk:%u", blkaddr);
1845 			f2fs_bug_on(sbi, 1);
1846 			se->valid_blocks--;
1847 			del = 0;
1848 		}
1849 
1850 		if (f2fs_discard_en(sbi) &&
1851 			!f2fs_test_and_set_bit(offset, se->discard_map))
1852 			sbi->discard_blks--;
1853 
1854 		/* don't overwrite by SSR to keep node chain */
1855 		if (IS_NODESEG(se->type)) {
1856 			if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1857 				se->ckpt_valid_blocks++;
1858 		}
1859 	} else {
1860 		exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
1861 #ifdef CONFIG_F2FS_CHECK_FS
1862 		mir_exist = f2fs_test_and_clear_bit(offset,
1863 						se->cur_valid_map_mir);
1864 		if (unlikely(exist != mir_exist)) {
1865 			f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1866 				"when clearing bitmap, blk:%u, old bit:%d",
1867 				blkaddr, exist);
1868 			f2fs_bug_on(sbi, 1);
1869 		}
1870 #endif
1871 		if (unlikely(!exist)) {
1872 			f2fs_msg(sbi->sb, KERN_ERR,
1873 				"Bitmap was wrongly cleared, blk:%u", blkaddr);
1874 			f2fs_bug_on(sbi, 1);
1875 			se->valid_blocks++;
1876 			del = 0;
1877 		}
1878 
1879 		if (f2fs_discard_en(sbi) &&
1880 			f2fs_test_and_clear_bit(offset, se->discard_map))
1881 			sbi->discard_blks++;
1882 	}
1883 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1884 		se->ckpt_valid_blocks += del;
1885 
1886 	__mark_sit_entry_dirty(sbi, segno);
1887 
1888 	/* update total number of valid blocks to be written in ckpt area */
1889 	SIT_I(sbi)->written_valid_blocks += del;
1890 
1891 	if (sbi->segs_per_sec > 1)
1892 		get_sec_entry(sbi, segno)->valid_blocks += del;
1893 }
1894 
1895 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1896 {
1897 	unsigned int segno = GET_SEGNO(sbi, addr);
1898 	struct sit_info *sit_i = SIT_I(sbi);
1899 
1900 	f2fs_bug_on(sbi, addr == NULL_ADDR);
1901 	if (addr == NEW_ADDR)
1902 		return;
1903 
1904 	/* add it into sit main buffer */
1905 	down_write(&sit_i->sentry_lock);
1906 
1907 	update_sit_entry(sbi, addr, -1);
1908 
1909 	/* add it into dirty seglist */
1910 	locate_dirty_segment(sbi, segno);
1911 
1912 	up_write(&sit_i->sentry_lock);
1913 }
1914 
1915 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1916 {
1917 	struct sit_info *sit_i = SIT_I(sbi);
1918 	unsigned int segno, offset;
1919 	struct seg_entry *se;
1920 	bool is_cp = false;
1921 
1922 	if (!is_valid_blkaddr(blkaddr))
1923 		return true;
1924 
1925 	down_read(&sit_i->sentry_lock);
1926 
1927 	segno = GET_SEGNO(sbi, blkaddr);
1928 	se = get_seg_entry(sbi, segno);
1929 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1930 
1931 	if (f2fs_test_bit(offset, se->ckpt_valid_map))
1932 		is_cp = true;
1933 
1934 	up_read(&sit_i->sentry_lock);
1935 
1936 	return is_cp;
1937 }
1938 
1939 /*
1940  * This function should be resided under the curseg_mutex lock
1941  */
1942 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1943 					struct f2fs_summary *sum)
1944 {
1945 	struct curseg_info *curseg = CURSEG_I(sbi, type);
1946 	void *addr = curseg->sum_blk;
1947 	addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1948 	memcpy(addr, sum, sizeof(struct f2fs_summary));
1949 }
1950 
1951 /*
1952  * Calculate the number of current summary pages for writing
1953  */
1954 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1955 {
1956 	int valid_sum_count = 0;
1957 	int i, sum_in_page;
1958 
1959 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1960 		if (sbi->ckpt->alloc_type[i] == SSR)
1961 			valid_sum_count += sbi->blocks_per_seg;
1962 		else {
1963 			if (for_ra)
1964 				valid_sum_count += le16_to_cpu(
1965 					F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1966 			else
1967 				valid_sum_count += curseg_blkoff(sbi, i);
1968 		}
1969 	}
1970 
1971 	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1972 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1973 	if (valid_sum_count <= sum_in_page)
1974 		return 1;
1975 	else if ((valid_sum_count - sum_in_page) <=
1976 		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1977 		return 2;
1978 	return 3;
1979 }
1980 
1981 /*
1982  * Caller should put this summary page
1983  */
1984 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1985 {
1986 	return f2fs_get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1987 }
1988 
1989 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
1990 					void *src, block_t blk_addr)
1991 {
1992 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1993 
1994 	memcpy(page_address(page), src, PAGE_SIZE);
1995 	set_page_dirty(page);
1996 	f2fs_put_page(page, 1);
1997 }
1998 
1999 static void write_sum_page(struct f2fs_sb_info *sbi,
2000 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
2001 {
2002 	f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2003 }
2004 
2005 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2006 						int type, block_t blk_addr)
2007 {
2008 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2009 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2010 	struct f2fs_summary_block *src = curseg->sum_blk;
2011 	struct f2fs_summary_block *dst;
2012 
2013 	dst = (struct f2fs_summary_block *)page_address(page);
2014 	memset(dst, 0, PAGE_SIZE);
2015 
2016 	mutex_lock(&curseg->curseg_mutex);
2017 
2018 	down_read(&curseg->journal_rwsem);
2019 	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2020 	up_read(&curseg->journal_rwsem);
2021 
2022 	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2023 	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2024 
2025 	mutex_unlock(&curseg->curseg_mutex);
2026 
2027 	set_page_dirty(page);
2028 	f2fs_put_page(page, 1);
2029 }
2030 
2031 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2032 {
2033 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2034 	unsigned int segno = curseg->segno + 1;
2035 	struct free_segmap_info *free_i = FREE_I(sbi);
2036 
2037 	if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2038 		return !test_bit(segno, free_i->free_segmap);
2039 	return 0;
2040 }
2041 
2042 /*
2043  * Find a new segment from the free segments bitmap to right order
2044  * This function should be returned with success, otherwise BUG
2045  */
2046 static void get_new_segment(struct f2fs_sb_info *sbi,
2047 			unsigned int *newseg, bool new_sec, int dir)
2048 {
2049 	struct free_segmap_info *free_i = FREE_I(sbi);
2050 	unsigned int segno, secno, zoneno;
2051 	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2052 	unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2053 	unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2054 	unsigned int left_start = hint;
2055 	bool init = true;
2056 	int go_left = 0;
2057 	int i;
2058 
2059 	spin_lock(&free_i->segmap_lock);
2060 
2061 	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2062 		segno = find_next_zero_bit(free_i->free_segmap,
2063 			GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2064 		if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2065 			goto got_it;
2066 	}
2067 find_other_zone:
2068 	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2069 	if (secno >= MAIN_SECS(sbi)) {
2070 		if (dir == ALLOC_RIGHT) {
2071 			secno = find_next_zero_bit(free_i->free_secmap,
2072 							MAIN_SECS(sbi), 0);
2073 			f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2074 		} else {
2075 			go_left = 1;
2076 			left_start = hint - 1;
2077 		}
2078 	}
2079 	if (go_left == 0)
2080 		goto skip_left;
2081 
2082 	while (test_bit(left_start, free_i->free_secmap)) {
2083 		if (left_start > 0) {
2084 			left_start--;
2085 			continue;
2086 		}
2087 		left_start = find_next_zero_bit(free_i->free_secmap,
2088 							MAIN_SECS(sbi), 0);
2089 		f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2090 		break;
2091 	}
2092 	secno = left_start;
2093 skip_left:
2094 	segno = GET_SEG_FROM_SEC(sbi, secno);
2095 	zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2096 
2097 	/* give up on finding another zone */
2098 	if (!init)
2099 		goto got_it;
2100 	if (sbi->secs_per_zone == 1)
2101 		goto got_it;
2102 	if (zoneno == old_zoneno)
2103 		goto got_it;
2104 	if (dir == ALLOC_LEFT) {
2105 		if (!go_left && zoneno + 1 >= total_zones)
2106 			goto got_it;
2107 		if (go_left && zoneno == 0)
2108 			goto got_it;
2109 	}
2110 	for (i = 0; i < NR_CURSEG_TYPE; i++)
2111 		if (CURSEG_I(sbi, i)->zone == zoneno)
2112 			break;
2113 
2114 	if (i < NR_CURSEG_TYPE) {
2115 		/* zone is in user, try another */
2116 		if (go_left)
2117 			hint = zoneno * sbi->secs_per_zone - 1;
2118 		else if (zoneno + 1 >= total_zones)
2119 			hint = 0;
2120 		else
2121 			hint = (zoneno + 1) * sbi->secs_per_zone;
2122 		init = false;
2123 		goto find_other_zone;
2124 	}
2125 got_it:
2126 	/* set it as dirty segment in free segmap */
2127 	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2128 	__set_inuse(sbi, segno);
2129 	*newseg = segno;
2130 	spin_unlock(&free_i->segmap_lock);
2131 }
2132 
2133 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2134 {
2135 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2136 	struct summary_footer *sum_footer;
2137 
2138 	curseg->segno = curseg->next_segno;
2139 	curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2140 	curseg->next_blkoff = 0;
2141 	curseg->next_segno = NULL_SEGNO;
2142 
2143 	sum_footer = &(curseg->sum_blk->footer);
2144 	memset(sum_footer, 0, sizeof(struct summary_footer));
2145 	if (IS_DATASEG(type))
2146 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2147 	if (IS_NODESEG(type))
2148 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2149 	__set_sit_entry_type(sbi, type, curseg->segno, modified);
2150 }
2151 
2152 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2153 {
2154 	/* if segs_per_sec is large than 1, we need to keep original policy. */
2155 	if (sbi->segs_per_sec != 1)
2156 		return CURSEG_I(sbi, type)->segno;
2157 
2158 	if (test_opt(sbi, NOHEAP) &&
2159 		(type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2160 		return 0;
2161 
2162 	if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2163 		return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2164 
2165 	/* find segments from 0 to reuse freed segments */
2166 	if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2167 		return 0;
2168 
2169 	return CURSEG_I(sbi, type)->segno;
2170 }
2171 
2172 /*
2173  * Allocate a current working segment.
2174  * This function always allocates a free segment in LFS manner.
2175  */
2176 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2177 {
2178 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2179 	unsigned int segno = curseg->segno;
2180 	int dir = ALLOC_LEFT;
2181 
2182 	write_sum_page(sbi, curseg->sum_blk,
2183 				GET_SUM_BLOCK(sbi, segno));
2184 	if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2185 		dir = ALLOC_RIGHT;
2186 
2187 	if (test_opt(sbi, NOHEAP))
2188 		dir = ALLOC_RIGHT;
2189 
2190 	segno = __get_next_segno(sbi, type);
2191 	get_new_segment(sbi, &segno, new_sec, dir);
2192 	curseg->next_segno = segno;
2193 	reset_curseg(sbi, type, 1);
2194 	curseg->alloc_type = LFS;
2195 }
2196 
2197 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2198 			struct curseg_info *seg, block_t start)
2199 {
2200 	struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2201 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2202 	unsigned long *target_map = SIT_I(sbi)->tmp_map;
2203 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2204 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2205 	int i, pos;
2206 
2207 	for (i = 0; i < entries; i++)
2208 		target_map[i] = ckpt_map[i] | cur_map[i];
2209 
2210 	pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2211 
2212 	seg->next_blkoff = pos;
2213 }
2214 
2215 /*
2216  * If a segment is written by LFS manner, next block offset is just obtained
2217  * by increasing the current block offset. However, if a segment is written by
2218  * SSR manner, next block offset obtained by calling __next_free_blkoff
2219  */
2220 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2221 				struct curseg_info *seg)
2222 {
2223 	if (seg->alloc_type == SSR)
2224 		__next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2225 	else
2226 		seg->next_blkoff++;
2227 }
2228 
2229 /*
2230  * This function always allocates a used segment(from dirty seglist) by SSR
2231  * manner, so it should recover the existing segment information of valid blocks
2232  */
2233 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2234 {
2235 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2236 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2237 	unsigned int new_segno = curseg->next_segno;
2238 	struct f2fs_summary_block *sum_node;
2239 	struct page *sum_page;
2240 
2241 	write_sum_page(sbi, curseg->sum_blk,
2242 				GET_SUM_BLOCK(sbi, curseg->segno));
2243 	__set_test_and_inuse(sbi, new_segno);
2244 
2245 	mutex_lock(&dirty_i->seglist_lock);
2246 	__remove_dirty_segment(sbi, new_segno, PRE);
2247 	__remove_dirty_segment(sbi, new_segno, DIRTY);
2248 	mutex_unlock(&dirty_i->seglist_lock);
2249 
2250 	reset_curseg(sbi, type, 1);
2251 	curseg->alloc_type = SSR;
2252 	__next_free_blkoff(sbi, curseg, 0);
2253 
2254 	sum_page = f2fs_get_sum_page(sbi, new_segno);
2255 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2256 	memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2257 	f2fs_put_page(sum_page, 1);
2258 }
2259 
2260 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2261 {
2262 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2263 	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2264 	unsigned segno = NULL_SEGNO;
2265 	int i, cnt;
2266 	bool reversed = false;
2267 
2268 	/* f2fs_need_SSR() already forces to do this */
2269 	if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2270 		curseg->next_segno = segno;
2271 		return 1;
2272 	}
2273 
2274 	/* For node segments, let's do SSR more intensively */
2275 	if (IS_NODESEG(type)) {
2276 		if (type >= CURSEG_WARM_NODE) {
2277 			reversed = true;
2278 			i = CURSEG_COLD_NODE;
2279 		} else {
2280 			i = CURSEG_HOT_NODE;
2281 		}
2282 		cnt = NR_CURSEG_NODE_TYPE;
2283 	} else {
2284 		if (type >= CURSEG_WARM_DATA) {
2285 			reversed = true;
2286 			i = CURSEG_COLD_DATA;
2287 		} else {
2288 			i = CURSEG_HOT_DATA;
2289 		}
2290 		cnt = NR_CURSEG_DATA_TYPE;
2291 	}
2292 
2293 	for (; cnt-- > 0; reversed ? i-- : i++) {
2294 		if (i == type)
2295 			continue;
2296 		if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2297 			curseg->next_segno = segno;
2298 			return 1;
2299 		}
2300 	}
2301 	return 0;
2302 }
2303 
2304 /*
2305  * flush out current segment and replace it with new segment
2306  * This function should be returned with success, otherwise BUG
2307  */
2308 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2309 						int type, bool force)
2310 {
2311 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2312 
2313 	if (force)
2314 		new_curseg(sbi, type, true);
2315 	else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2316 					type == CURSEG_WARM_NODE)
2317 		new_curseg(sbi, type, false);
2318 	else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2319 		new_curseg(sbi, type, false);
2320 	else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2321 		change_curseg(sbi, type);
2322 	else
2323 		new_curseg(sbi, type, false);
2324 
2325 	stat_inc_seg_type(sbi, curseg);
2326 }
2327 
2328 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2329 {
2330 	struct curseg_info *curseg;
2331 	unsigned int old_segno;
2332 	int i;
2333 
2334 	down_write(&SIT_I(sbi)->sentry_lock);
2335 
2336 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2337 		curseg = CURSEG_I(sbi, i);
2338 		old_segno = curseg->segno;
2339 		SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2340 		locate_dirty_segment(sbi, old_segno);
2341 	}
2342 
2343 	up_write(&SIT_I(sbi)->sentry_lock);
2344 }
2345 
2346 static const struct segment_allocation default_salloc_ops = {
2347 	.allocate_segment = allocate_segment_by_default,
2348 };
2349 
2350 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2351 						struct cp_control *cpc)
2352 {
2353 	__u64 trim_start = cpc->trim_start;
2354 	bool has_candidate = false;
2355 
2356 	down_write(&SIT_I(sbi)->sentry_lock);
2357 	for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2358 		if (add_discard_addrs(sbi, cpc, true)) {
2359 			has_candidate = true;
2360 			break;
2361 		}
2362 	}
2363 	up_write(&SIT_I(sbi)->sentry_lock);
2364 
2365 	cpc->trim_start = trim_start;
2366 	return has_candidate;
2367 }
2368 
2369 static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2370 					struct discard_policy *dpolicy,
2371 					unsigned int start, unsigned int end)
2372 {
2373 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2374 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2375 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
2376 	struct discard_cmd *dc;
2377 	struct blk_plug plug;
2378 	int issued;
2379 
2380 next:
2381 	issued = 0;
2382 
2383 	mutex_lock(&dcc->cmd_lock);
2384 	f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi, &dcc->root));
2385 
2386 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2387 					NULL, start,
2388 					(struct rb_entry **)&prev_dc,
2389 					(struct rb_entry **)&next_dc,
2390 					&insert_p, &insert_parent, true);
2391 	if (!dc)
2392 		dc = next_dc;
2393 
2394 	blk_start_plug(&plug);
2395 
2396 	while (dc && dc->lstart <= end) {
2397 		struct rb_node *node;
2398 
2399 		if (dc->len < dpolicy->granularity)
2400 			goto skip;
2401 
2402 		if (dc->state != D_PREP) {
2403 			list_move_tail(&dc->list, &dcc->fstrim_list);
2404 			goto skip;
2405 		}
2406 
2407 		__submit_discard_cmd(sbi, dpolicy, dc);
2408 
2409 		if (++issued >= dpolicy->max_requests) {
2410 			start = dc->lstart + dc->len;
2411 
2412 			blk_finish_plug(&plug);
2413 			mutex_unlock(&dcc->cmd_lock);
2414 			__wait_all_discard_cmd(sbi, NULL);
2415 			congestion_wait(BLK_RW_ASYNC, HZ/50);
2416 			goto next;
2417 		}
2418 skip:
2419 		node = rb_next(&dc->rb_node);
2420 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2421 
2422 		if (fatal_signal_pending(current))
2423 			break;
2424 	}
2425 
2426 	blk_finish_plug(&plug);
2427 	mutex_unlock(&dcc->cmd_lock);
2428 }
2429 
2430 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2431 {
2432 	__u64 start = F2FS_BYTES_TO_BLK(range->start);
2433 	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2434 	unsigned int start_segno, end_segno;
2435 	block_t start_block, end_block;
2436 	struct cp_control cpc;
2437 	struct discard_policy dpolicy;
2438 	unsigned long long trimmed = 0;
2439 	int err = 0;
2440 
2441 	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2442 		return -EINVAL;
2443 
2444 	if (end <= MAIN_BLKADDR(sbi))
2445 		return -EINVAL;
2446 
2447 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2448 		f2fs_msg(sbi->sb, KERN_WARNING,
2449 			"Found FS corruption, run fsck to fix.");
2450 		return -EIO;
2451 	}
2452 
2453 	/* start/end segment number in main_area */
2454 	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2455 	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2456 						GET_SEGNO(sbi, end);
2457 
2458 	cpc.reason = CP_DISCARD;
2459 	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2460 	cpc.trim_start = start_segno;
2461 	cpc.trim_end = end_segno;
2462 
2463 	if (sbi->discard_blks == 0)
2464 		goto out;
2465 
2466 	mutex_lock(&sbi->gc_mutex);
2467 	err = f2fs_write_checkpoint(sbi, &cpc);
2468 	mutex_unlock(&sbi->gc_mutex);
2469 	if (err)
2470 		goto out;
2471 
2472 	start_block = START_BLOCK(sbi, start_segno);
2473 	end_block = START_BLOCK(sbi, end_segno + 1);
2474 
2475 	__init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2476 	__issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block);
2477 
2478 	/*
2479 	 * We filed discard candidates, but actually we don't need to wait for
2480 	 * all of them, since they'll be issued in idle time along with runtime
2481 	 * discard option. User configuration looks like using runtime discard
2482 	 * or periodic fstrim instead of it.
2483 	 */
2484 	if (!test_opt(sbi, DISCARD)) {
2485 		trimmed = __wait_discard_cmd_range(sbi, &dpolicy,
2486 					start_block, end_block);
2487 		range->len = F2FS_BLK_TO_BYTES(trimmed);
2488 	}
2489 out:
2490 	return err;
2491 }
2492 
2493 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2494 {
2495 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2496 	if (curseg->next_blkoff < sbi->blocks_per_seg)
2497 		return true;
2498 	return false;
2499 }
2500 
2501 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2502 {
2503 	switch (hint) {
2504 	case WRITE_LIFE_SHORT:
2505 		return CURSEG_HOT_DATA;
2506 	case WRITE_LIFE_EXTREME:
2507 		return CURSEG_COLD_DATA;
2508 	default:
2509 		return CURSEG_WARM_DATA;
2510 	}
2511 }
2512 
2513 /* This returns write hints for each segment type. This hints will be
2514  * passed down to block layer. There are mapping tables which depend on
2515  * the mount option 'whint_mode'.
2516  *
2517  * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2518  *
2519  * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2520  *
2521  * User                  F2FS                     Block
2522  * ----                  ----                     -----
2523  *                       META                     WRITE_LIFE_NOT_SET
2524  *                       HOT_NODE                 "
2525  *                       WARM_NODE                "
2526  *                       COLD_NODE                "
2527  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2528  * extension list        "                        "
2529  *
2530  * -- buffered io
2531  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2532  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2533  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2534  * WRITE_LIFE_NONE       "                        "
2535  * WRITE_LIFE_MEDIUM     "                        "
2536  * WRITE_LIFE_LONG       "                        "
2537  *
2538  * -- direct io
2539  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2540  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2541  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2542  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2543  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2544  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2545  *
2546  * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2547  *
2548  * User                  F2FS                     Block
2549  * ----                  ----                     -----
2550  *                       META                     WRITE_LIFE_MEDIUM;
2551  *                       HOT_NODE                 WRITE_LIFE_NOT_SET
2552  *                       WARM_NODE                "
2553  *                       COLD_NODE                WRITE_LIFE_NONE
2554  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
2555  * extension list        "                        "
2556  *
2557  * -- buffered io
2558  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2559  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2560  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_LONG
2561  * WRITE_LIFE_NONE       "                        "
2562  * WRITE_LIFE_MEDIUM     "                        "
2563  * WRITE_LIFE_LONG       "                        "
2564  *
2565  * -- direct io
2566  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
2567  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
2568  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
2569  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
2570  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
2571  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
2572  */
2573 
2574 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2575 				enum page_type type, enum temp_type temp)
2576 {
2577 	if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2578 		if (type == DATA) {
2579 			if (temp == WARM)
2580 				return WRITE_LIFE_NOT_SET;
2581 			else if (temp == HOT)
2582 				return WRITE_LIFE_SHORT;
2583 			else if (temp == COLD)
2584 				return WRITE_LIFE_EXTREME;
2585 		} else {
2586 			return WRITE_LIFE_NOT_SET;
2587 		}
2588 	} else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2589 		if (type == DATA) {
2590 			if (temp == WARM)
2591 				return WRITE_LIFE_LONG;
2592 			else if (temp == HOT)
2593 				return WRITE_LIFE_SHORT;
2594 			else if (temp == COLD)
2595 				return WRITE_LIFE_EXTREME;
2596 		} else if (type == NODE) {
2597 			if (temp == WARM || temp == HOT)
2598 				return WRITE_LIFE_NOT_SET;
2599 			else if (temp == COLD)
2600 				return WRITE_LIFE_NONE;
2601 		} else if (type == META) {
2602 			return WRITE_LIFE_MEDIUM;
2603 		}
2604 	}
2605 	return WRITE_LIFE_NOT_SET;
2606 }
2607 
2608 static int __get_segment_type_2(struct f2fs_io_info *fio)
2609 {
2610 	if (fio->type == DATA)
2611 		return CURSEG_HOT_DATA;
2612 	else
2613 		return CURSEG_HOT_NODE;
2614 }
2615 
2616 static int __get_segment_type_4(struct f2fs_io_info *fio)
2617 {
2618 	if (fio->type == DATA) {
2619 		struct inode *inode = fio->page->mapping->host;
2620 
2621 		if (S_ISDIR(inode->i_mode))
2622 			return CURSEG_HOT_DATA;
2623 		else
2624 			return CURSEG_COLD_DATA;
2625 	} else {
2626 		if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2627 			return CURSEG_WARM_NODE;
2628 		else
2629 			return CURSEG_COLD_NODE;
2630 	}
2631 }
2632 
2633 static int __get_segment_type_6(struct f2fs_io_info *fio)
2634 {
2635 	if (fio->type == DATA) {
2636 		struct inode *inode = fio->page->mapping->host;
2637 
2638 		if (is_cold_data(fio->page) || file_is_cold(inode))
2639 			return CURSEG_COLD_DATA;
2640 		if (file_is_hot(inode) ||
2641 				is_inode_flag_set(inode, FI_HOT_DATA) ||
2642 				is_inode_flag_set(inode, FI_ATOMIC_FILE) ||
2643 				is_inode_flag_set(inode, FI_VOLATILE_FILE))
2644 			return CURSEG_HOT_DATA;
2645 		return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
2646 	} else {
2647 		if (IS_DNODE(fio->page))
2648 			return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2649 						CURSEG_HOT_NODE;
2650 		return CURSEG_COLD_NODE;
2651 	}
2652 }
2653 
2654 static int __get_segment_type(struct f2fs_io_info *fio)
2655 {
2656 	int type = 0;
2657 
2658 	switch (F2FS_OPTION(fio->sbi).active_logs) {
2659 	case 2:
2660 		type = __get_segment_type_2(fio);
2661 		break;
2662 	case 4:
2663 		type = __get_segment_type_4(fio);
2664 		break;
2665 	case 6:
2666 		type = __get_segment_type_6(fio);
2667 		break;
2668 	default:
2669 		f2fs_bug_on(fio->sbi, true);
2670 	}
2671 
2672 	if (IS_HOT(type))
2673 		fio->temp = HOT;
2674 	else if (IS_WARM(type))
2675 		fio->temp = WARM;
2676 	else
2677 		fio->temp = COLD;
2678 	return type;
2679 }
2680 
2681 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2682 		block_t old_blkaddr, block_t *new_blkaddr,
2683 		struct f2fs_summary *sum, int type,
2684 		struct f2fs_io_info *fio, bool add_list)
2685 {
2686 	struct sit_info *sit_i = SIT_I(sbi);
2687 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2688 
2689 	down_read(&SM_I(sbi)->curseg_lock);
2690 
2691 	mutex_lock(&curseg->curseg_mutex);
2692 	down_write(&sit_i->sentry_lock);
2693 
2694 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2695 
2696 	f2fs_wait_discard_bio(sbi, *new_blkaddr);
2697 
2698 	/*
2699 	 * __add_sum_entry should be resided under the curseg_mutex
2700 	 * because, this function updates a summary entry in the
2701 	 * current summary block.
2702 	 */
2703 	__add_sum_entry(sbi, type, sum);
2704 
2705 	__refresh_next_blkoff(sbi, curseg);
2706 
2707 	stat_inc_block_count(sbi, curseg);
2708 
2709 	/*
2710 	 * SIT information should be updated before segment allocation,
2711 	 * since SSR needs latest valid block information.
2712 	 */
2713 	update_sit_entry(sbi, *new_blkaddr, 1);
2714 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2715 		update_sit_entry(sbi, old_blkaddr, -1);
2716 
2717 	if (!__has_curseg_space(sbi, type))
2718 		sit_i->s_ops->allocate_segment(sbi, type, false);
2719 
2720 	/*
2721 	 * segment dirty status should be updated after segment allocation,
2722 	 * so we just need to update status only one time after previous
2723 	 * segment being closed.
2724 	 */
2725 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2726 	locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
2727 
2728 	up_write(&sit_i->sentry_lock);
2729 
2730 	if (page && IS_NODESEG(type)) {
2731 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2732 
2733 		f2fs_inode_chksum_set(sbi, page);
2734 	}
2735 
2736 	if (add_list) {
2737 		struct f2fs_bio_info *io;
2738 
2739 		INIT_LIST_HEAD(&fio->list);
2740 		fio->in_list = true;
2741 		fio->retry = false;
2742 		io = sbi->write_io[fio->type] + fio->temp;
2743 		spin_lock(&io->io_lock);
2744 		list_add_tail(&fio->list, &io->io_list);
2745 		spin_unlock(&io->io_lock);
2746 	}
2747 
2748 	mutex_unlock(&curseg->curseg_mutex);
2749 
2750 	up_read(&SM_I(sbi)->curseg_lock);
2751 }
2752 
2753 static void update_device_state(struct f2fs_io_info *fio)
2754 {
2755 	struct f2fs_sb_info *sbi = fio->sbi;
2756 	unsigned int devidx;
2757 
2758 	if (!sbi->s_ndevs)
2759 		return;
2760 
2761 	devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2762 
2763 	/* update device state for fsync */
2764 	f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
2765 
2766 	/* update device state for checkpoint */
2767 	if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2768 		spin_lock(&sbi->dev_lock);
2769 		f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2770 		spin_unlock(&sbi->dev_lock);
2771 	}
2772 }
2773 
2774 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2775 {
2776 	int type = __get_segment_type(fio);
2777 	bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
2778 
2779 	if (keep_order)
2780 		down_read(&fio->sbi->io_order_lock);
2781 reallocate:
2782 	f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2783 			&fio->new_blkaddr, sum, type, fio, true);
2784 
2785 	/* writeout dirty page into bdev */
2786 	f2fs_submit_page_write(fio);
2787 	if (fio->retry) {
2788 		fio->old_blkaddr = fio->new_blkaddr;
2789 		goto reallocate;
2790 	}
2791 
2792 	update_device_state(fio);
2793 
2794 	if (keep_order)
2795 		up_read(&fio->sbi->io_order_lock);
2796 }
2797 
2798 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2799 					enum iostat_type io_type)
2800 {
2801 	struct f2fs_io_info fio = {
2802 		.sbi = sbi,
2803 		.type = META,
2804 		.temp = HOT,
2805 		.op = REQ_OP_WRITE,
2806 		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
2807 		.old_blkaddr = page->index,
2808 		.new_blkaddr = page->index,
2809 		.page = page,
2810 		.encrypted_page = NULL,
2811 		.in_list = false,
2812 	};
2813 
2814 	if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
2815 		fio.op_flags &= ~REQ_META;
2816 
2817 	set_page_writeback(page);
2818 	ClearPageError(page);
2819 	f2fs_submit_page_write(&fio);
2820 
2821 	f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
2822 }
2823 
2824 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
2825 {
2826 	struct f2fs_summary sum;
2827 
2828 	set_summary(&sum, nid, 0, 0);
2829 	do_write_page(&sum, fio);
2830 
2831 	f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2832 }
2833 
2834 void f2fs_outplace_write_data(struct dnode_of_data *dn,
2835 					struct f2fs_io_info *fio)
2836 {
2837 	struct f2fs_sb_info *sbi = fio->sbi;
2838 	struct f2fs_summary sum;
2839 	struct node_info ni;
2840 
2841 	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
2842 	f2fs_get_node_info(sbi, dn->nid, &ni);
2843 	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
2844 	do_write_page(&sum, fio);
2845 	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
2846 
2847 	f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
2848 }
2849 
2850 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
2851 {
2852 	int err;
2853 	struct f2fs_sb_info *sbi = fio->sbi;
2854 
2855 	fio->new_blkaddr = fio->old_blkaddr;
2856 	/* i/o temperature is needed for passing down write hints */
2857 	__get_segment_type(fio);
2858 
2859 	f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
2860 			GET_SEGNO(sbi, fio->new_blkaddr))->type));
2861 
2862 	stat_inc_inplace_blocks(fio->sbi);
2863 
2864 	err = f2fs_submit_page_bio(fio);
2865 	if (!err)
2866 		update_device_state(fio);
2867 
2868 	f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2869 
2870 	return err;
2871 }
2872 
2873 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
2874 						unsigned int segno)
2875 {
2876 	int i;
2877 
2878 	for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
2879 		if (CURSEG_I(sbi, i)->segno == segno)
2880 			break;
2881 	}
2882 	return i;
2883 }
2884 
2885 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
2886 				block_t old_blkaddr, block_t new_blkaddr,
2887 				bool recover_curseg, bool recover_newaddr)
2888 {
2889 	struct sit_info *sit_i = SIT_I(sbi);
2890 	struct curseg_info *curseg;
2891 	unsigned int segno, old_cursegno;
2892 	struct seg_entry *se;
2893 	int type;
2894 	unsigned short old_blkoff;
2895 
2896 	segno = GET_SEGNO(sbi, new_blkaddr);
2897 	se = get_seg_entry(sbi, segno);
2898 	type = se->type;
2899 
2900 	down_write(&SM_I(sbi)->curseg_lock);
2901 
2902 	if (!recover_curseg) {
2903 		/* for recovery flow */
2904 		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2905 			if (old_blkaddr == NULL_ADDR)
2906 				type = CURSEG_COLD_DATA;
2907 			else
2908 				type = CURSEG_WARM_DATA;
2909 		}
2910 	} else {
2911 		if (IS_CURSEG(sbi, segno)) {
2912 			/* se->type is volatile as SSR allocation */
2913 			type = __f2fs_get_curseg(sbi, segno);
2914 			f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
2915 		} else {
2916 			type = CURSEG_WARM_DATA;
2917 		}
2918 	}
2919 
2920 	f2fs_bug_on(sbi, !IS_DATASEG(type));
2921 	curseg = CURSEG_I(sbi, type);
2922 
2923 	mutex_lock(&curseg->curseg_mutex);
2924 	down_write(&sit_i->sentry_lock);
2925 
2926 	old_cursegno = curseg->segno;
2927 	old_blkoff = curseg->next_blkoff;
2928 
2929 	/* change the current segment */
2930 	if (segno != curseg->segno) {
2931 		curseg->next_segno = segno;
2932 		change_curseg(sbi, type);
2933 	}
2934 
2935 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
2936 	__add_sum_entry(sbi, type, sum);
2937 
2938 	if (!recover_curseg || recover_newaddr)
2939 		update_sit_entry(sbi, new_blkaddr, 1);
2940 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2941 		update_sit_entry(sbi, old_blkaddr, -1);
2942 
2943 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2944 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2945 
2946 	locate_dirty_segment(sbi, old_cursegno);
2947 
2948 	if (recover_curseg) {
2949 		if (old_cursegno != curseg->segno) {
2950 			curseg->next_segno = old_cursegno;
2951 			change_curseg(sbi, type);
2952 		}
2953 		curseg->next_blkoff = old_blkoff;
2954 	}
2955 
2956 	up_write(&sit_i->sentry_lock);
2957 	mutex_unlock(&curseg->curseg_mutex);
2958 	up_write(&SM_I(sbi)->curseg_lock);
2959 }
2960 
2961 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2962 				block_t old_addr, block_t new_addr,
2963 				unsigned char version, bool recover_curseg,
2964 				bool recover_newaddr)
2965 {
2966 	struct f2fs_summary sum;
2967 
2968 	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2969 
2970 	f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
2971 					recover_curseg, recover_newaddr);
2972 
2973 	f2fs_update_data_blkaddr(dn, new_addr);
2974 }
2975 
2976 void f2fs_wait_on_page_writeback(struct page *page,
2977 				enum page_type type, bool ordered)
2978 {
2979 	if (PageWriteback(page)) {
2980 		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2981 
2982 		f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2983 						0, page->index, type);
2984 		if (ordered)
2985 			wait_on_page_writeback(page);
2986 		else
2987 			wait_for_stable_page(page);
2988 	}
2989 }
2990 
2991 void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
2992 {
2993 	struct page *cpage;
2994 
2995 	if (!is_valid_blkaddr(blkaddr))
2996 		return;
2997 
2998 	cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2999 	if (cpage) {
3000 		f2fs_wait_on_page_writeback(cpage, DATA, true);
3001 		f2fs_put_page(cpage, 1);
3002 	}
3003 }
3004 
3005 static void read_compacted_summaries(struct f2fs_sb_info *sbi)
3006 {
3007 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3008 	struct curseg_info *seg_i;
3009 	unsigned char *kaddr;
3010 	struct page *page;
3011 	block_t start;
3012 	int i, j, offset;
3013 
3014 	start = start_sum_block(sbi);
3015 
3016 	page = f2fs_get_meta_page(sbi, start++);
3017 	kaddr = (unsigned char *)page_address(page);
3018 
3019 	/* Step 1: restore nat cache */
3020 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3021 	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3022 
3023 	/* Step 2: restore sit cache */
3024 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3025 	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3026 	offset = 2 * SUM_JOURNAL_SIZE;
3027 
3028 	/* Step 3: restore summary entries */
3029 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3030 		unsigned short blk_off;
3031 		unsigned int segno;
3032 
3033 		seg_i = CURSEG_I(sbi, i);
3034 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3035 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3036 		seg_i->next_segno = segno;
3037 		reset_curseg(sbi, i, 0);
3038 		seg_i->alloc_type = ckpt->alloc_type[i];
3039 		seg_i->next_blkoff = blk_off;
3040 
3041 		if (seg_i->alloc_type == SSR)
3042 			blk_off = sbi->blocks_per_seg;
3043 
3044 		for (j = 0; j < blk_off; j++) {
3045 			struct f2fs_summary *s;
3046 			s = (struct f2fs_summary *)(kaddr + offset);
3047 			seg_i->sum_blk->entries[j] = *s;
3048 			offset += SUMMARY_SIZE;
3049 			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3050 						SUM_FOOTER_SIZE)
3051 				continue;
3052 
3053 			f2fs_put_page(page, 1);
3054 			page = NULL;
3055 
3056 			page = f2fs_get_meta_page(sbi, start++);
3057 			kaddr = (unsigned char *)page_address(page);
3058 			offset = 0;
3059 		}
3060 	}
3061 	f2fs_put_page(page, 1);
3062 }
3063 
3064 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3065 {
3066 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3067 	struct f2fs_summary_block *sum;
3068 	struct curseg_info *curseg;
3069 	struct page *new;
3070 	unsigned short blk_off;
3071 	unsigned int segno = 0;
3072 	block_t blk_addr = 0;
3073 
3074 	/* get segment number and block addr */
3075 	if (IS_DATASEG(type)) {
3076 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3077 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3078 							CURSEG_HOT_DATA]);
3079 		if (__exist_node_summaries(sbi))
3080 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3081 		else
3082 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3083 	} else {
3084 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
3085 							CURSEG_HOT_NODE]);
3086 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3087 							CURSEG_HOT_NODE]);
3088 		if (__exist_node_summaries(sbi))
3089 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3090 							type - CURSEG_HOT_NODE);
3091 		else
3092 			blk_addr = GET_SUM_BLOCK(sbi, segno);
3093 	}
3094 
3095 	new = f2fs_get_meta_page(sbi, blk_addr);
3096 	sum = (struct f2fs_summary_block *)page_address(new);
3097 
3098 	if (IS_NODESEG(type)) {
3099 		if (__exist_node_summaries(sbi)) {
3100 			struct f2fs_summary *ns = &sum->entries[0];
3101 			int i;
3102 			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3103 				ns->version = 0;
3104 				ns->ofs_in_node = 0;
3105 			}
3106 		} else {
3107 			f2fs_restore_node_summary(sbi, segno, sum);
3108 		}
3109 	}
3110 
3111 	/* set uncompleted segment to curseg */
3112 	curseg = CURSEG_I(sbi, type);
3113 	mutex_lock(&curseg->curseg_mutex);
3114 
3115 	/* update journal info */
3116 	down_write(&curseg->journal_rwsem);
3117 	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3118 	up_write(&curseg->journal_rwsem);
3119 
3120 	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3121 	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3122 	curseg->next_segno = segno;
3123 	reset_curseg(sbi, type, 0);
3124 	curseg->alloc_type = ckpt->alloc_type[type];
3125 	curseg->next_blkoff = blk_off;
3126 	mutex_unlock(&curseg->curseg_mutex);
3127 	f2fs_put_page(new, 1);
3128 	return 0;
3129 }
3130 
3131 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3132 {
3133 	struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3134 	struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3135 	int type = CURSEG_HOT_DATA;
3136 	int err;
3137 
3138 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3139 		int npages = f2fs_npages_for_summary_flush(sbi, true);
3140 
3141 		if (npages >= 2)
3142 			f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3143 							META_CP, true);
3144 
3145 		/* restore for compacted data summary */
3146 		read_compacted_summaries(sbi);
3147 		type = CURSEG_HOT_NODE;
3148 	}
3149 
3150 	if (__exist_node_summaries(sbi))
3151 		f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3152 					NR_CURSEG_TYPE - type, META_CP, true);
3153 
3154 	for (; type <= CURSEG_COLD_NODE; type++) {
3155 		err = read_normal_summaries(sbi, type);
3156 		if (err)
3157 			return err;
3158 	}
3159 
3160 	/* sanity check for summary blocks */
3161 	if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3162 			sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3163 		return -EINVAL;
3164 
3165 	return 0;
3166 }
3167 
3168 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3169 {
3170 	struct page *page;
3171 	unsigned char *kaddr;
3172 	struct f2fs_summary *summary;
3173 	struct curseg_info *seg_i;
3174 	int written_size = 0;
3175 	int i, j;
3176 
3177 	page = f2fs_grab_meta_page(sbi, blkaddr++);
3178 	kaddr = (unsigned char *)page_address(page);
3179 	memset(kaddr, 0, PAGE_SIZE);
3180 
3181 	/* Step 1: write nat cache */
3182 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3183 	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3184 	written_size += SUM_JOURNAL_SIZE;
3185 
3186 	/* Step 2: write sit cache */
3187 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3188 	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3189 	written_size += SUM_JOURNAL_SIZE;
3190 
3191 	/* Step 3: write summary entries */
3192 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3193 		unsigned short blkoff;
3194 		seg_i = CURSEG_I(sbi, i);
3195 		if (sbi->ckpt->alloc_type[i] == SSR)
3196 			blkoff = sbi->blocks_per_seg;
3197 		else
3198 			blkoff = curseg_blkoff(sbi, i);
3199 
3200 		for (j = 0; j < blkoff; j++) {
3201 			if (!page) {
3202 				page = f2fs_grab_meta_page(sbi, blkaddr++);
3203 				kaddr = (unsigned char *)page_address(page);
3204 				memset(kaddr, 0, PAGE_SIZE);
3205 				written_size = 0;
3206 			}
3207 			summary = (struct f2fs_summary *)(kaddr + written_size);
3208 			*summary = seg_i->sum_blk->entries[j];
3209 			written_size += SUMMARY_SIZE;
3210 
3211 			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3212 							SUM_FOOTER_SIZE)
3213 				continue;
3214 
3215 			set_page_dirty(page);
3216 			f2fs_put_page(page, 1);
3217 			page = NULL;
3218 		}
3219 	}
3220 	if (page) {
3221 		set_page_dirty(page);
3222 		f2fs_put_page(page, 1);
3223 	}
3224 }
3225 
3226 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3227 					block_t blkaddr, int type)
3228 {
3229 	int i, end;
3230 	if (IS_DATASEG(type))
3231 		end = type + NR_CURSEG_DATA_TYPE;
3232 	else
3233 		end = type + NR_CURSEG_NODE_TYPE;
3234 
3235 	for (i = type; i < end; i++)
3236 		write_current_sum_page(sbi, i, blkaddr + (i - type));
3237 }
3238 
3239 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3240 {
3241 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3242 		write_compacted_summaries(sbi, start_blk);
3243 	else
3244 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3245 }
3246 
3247 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3248 {
3249 	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3250 }
3251 
3252 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3253 					unsigned int val, int alloc)
3254 {
3255 	int i;
3256 
3257 	if (type == NAT_JOURNAL) {
3258 		for (i = 0; i < nats_in_cursum(journal); i++) {
3259 			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3260 				return i;
3261 		}
3262 		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3263 			return update_nats_in_cursum(journal, 1);
3264 	} else if (type == SIT_JOURNAL) {
3265 		for (i = 0; i < sits_in_cursum(journal); i++)
3266 			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3267 				return i;
3268 		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3269 			return update_sits_in_cursum(journal, 1);
3270 	}
3271 	return -1;
3272 }
3273 
3274 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3275 					unsigned int segno)
3276 {
3277 	return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
3278 }
3279 
3280 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3281 					unsigned int start)
3282 {
3283 	struct sit_info *sit_i = SIT_I(sbi);
3284 	struct page *page;
3285 	pgoff_t src_off, dst_off;
3286 
3287 	src_off = current_sit_addr(sbi, start);
3288 	dst_off = next_sit_addr(sbi, src_off);
3289 
3290 	page = f2fs_grab_meta_page(sbi, dst_off);
3291 	seg_info_to_sit_page(sbi, page, start);
3292 
3293 	set_page_dirty(page);
3294 	set_to_next_sit(sit_i, start);
3295 
3296 	return page;
3297 }
3298 
3299 static struct sit_entry_set *grab_sit_entry_set(void)
3300 {
3301 	struct sit_entry_set *ses =
3302 			f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3303 
3304 	ses->entry_cnt = 0;
3305 	INIT_LIST_HEAD(&ses->set_list);
3306 	return ses;
3307 }
3308 
3309 static void release_sit_entry_set(struct sit_entry_set *ses)
3310 {
3311 	list_del(&ses->set_list);
3312 	kmem_cache_free(sit_entry_set_slab, ses);
3313 }
3314 
3315 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3316 						struct list_head *head)
3317 {
3318 	struct sit_entry_set *next = ses;
3319 
3320 	if (list_is_last(&ses->set_list, head))
3321 		return;
3322 
3323 	list_for_each_entry_continue(next, head, set_list)
3324 		if (ses->entry_cnt <= next->entry_cnt)
3325 			break;
3326 
3327 	list_move_tail(&ses->set_list, &next->set_list);
3328 }
3329 
3330 static void add_sit_entry(unsigned int segno, struct list_head *head)
3331 {
3332 	struct sit_entry_set *ses;
3333 	unsigned int start_segno = START_SEGNO(segno);
3334 
3335 	list_for_each_entry(ses, head, set_list) {
3336 		if (ses->start_segno == start_segno) {
3337 			ses->entry_cnt++;
3338 			adjust_sit_entry_set(ses, head);
3339 			return;
3340 		}
3341 	}
3342 
3343 	ses = grab_sit_entry_set();
3344 
3345 	ses->start_segno = start_segno;
3346 	ses->entry_cnt++;
3347 	list_add(&ses->set_list, head);
3348 }
3349 
3350 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3351 {
3352 	struct f2fs_sm_info *sm_info = SM_I(sbi);
3353 	struct list_head *set_list = &sm_info->sit_entry_set;
3354 	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3355 	unsigned int segno;
3356 
3357 	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3358 		add_sit_entry(segno, set_list);
3359 }
3360 
3361 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3362 {
3363 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3364 	struct f2fs_journal *journal = curseg->journal;
3365 	int i;
3366 
3367 	down_write(&curseg->journal_rwsem);
3368 	for (i = 0; i < sits_in_cursum(journal); i++) {
3369 		unsigned int segno;
3370 		bool dirtied;
3371 
3372 		segno = le32_to_cpu(segno_in_journal(journal, i));
3373 		dirtied = __mark_sit_entry_dirty(sbi, segno);
3374 
3375 		if (!dirtied)
3376 			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3377 	}
3378 	update_sits_in_cursum(journal, -i);
3379 	up_write(&curseg->journal_rwsem);
3380 }
3381 
3382 /*
3383  * CP calls this function, which flushes SIT entries including sit_journal,
3384  * and moves prefree segs to free segs.
3385  */
3386 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3387 {
3388 	struct sit_info *sit_i = SIT_I(sbi);
3389 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3390 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3391 	struct f2fs_journal *journal = curseg->journal;
3392 	struct sit_entry_set *ses, *tmp;
3393 	struct list_head *head = &SM_I(sbi)->sit_entry_set;
3394 	bool to_journal = true;
3395 	struct seg_entry *se;
3396 
3397 	down_write(&sit_i->sentry_lock);
3398 
3399 	if (!sit_i->dirty_sentries)
3400 		goto out;
3401 
3402 	/*
3403 	 * add and account sit entries of dirty bitmap in sit entry
3404 	 * set temporarily
3405 	 */
3406 	add_sits_in_set(sbi);
3407 
3408 	/*
3409 	 * if there are no enough space in journal to store dirty sit
3410 	 * entries, remove all entries from journal and add and account
3411 	 * them in sit entry set.
3412 	 */
3413 	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3414 		remove_sits_in_journal(sbi);
3415 
3416 	/*
3417 	 * there are two steps to flush sit entries:
3418 	 * #1, flush sit entries to journal in current cold data summary block.
3419 	 * #2, flush sit entries to sit page.
3420 	 */
3421 	list_for_each_entry_safe(ses, tmp, head, set_list) {
3422 		struct page *page = NULL;
3423 		struct f2fs_sit_block *raw_sit = NULL;
3424 		unsigned int start_segno = ses->start_segno;
3425 		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3426 						(unsigned long)MAIN_SEGS(sbi));
3427 		unsigned int segno = start_segno;
3428 
3429 		if (to_journal &&
3430 			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3431 			to_journal = false;
3432 
3433 		if (to_journal) {
3434 			down_write(&curseg->journal_rwsem);
3435 		} else {
3436 			page = get_next_sit_page(sbi, start_segno);
3437 			raw_sit = page_address(page);
3438 		}
3439 
3440 		/* flush dirty sit entries in region of current sit set */
3441 		for_each_set_bit_from(segno, bitmap, end) {
3442 			int offset, sit_offset;
3443 
3444 			se = get_seg_entry(sbi, segno);
3445 #ifdef CONFIG_F2FS_CHECK_FS
3446 			if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3447 						SIT_VBLOCK_MAP_SIZE))
3448 				f2fs_bug_on(sbi, 1);
3449 #endif
3450 
3451 			/* add discard candidates */
3452 			if (!(cpc->reason & CP_DISCARD)) {
3453 				cpc->trim_start = segno;
3454 				add_discard_addrs(sbi, cpc, false);
3455 			}
3456 
3457 			if (to_journal) {
3458 				offset = f2fs_lookup_journal_in_cursum(journal,
3459 							SIT_JOURNAL, segno, 1);
3460 				f2fs_bug_on(sbi, offset < 0);
3461 				segno_in_journal(journal, offset) =
3462 							cpu_to_le32(segno);
3463 				seg_info_to_raw_sit(se,
3464 					&sit_in_journal(journal, offset));
3465 				check_block_count(sbi, segno,
3466 					&sit_in_journal(journal, offset));
3467 			} else {
3468 				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3469 				seg_info_to_raw_sit(se,
3470 						&raw_sit->entries[sit_offset]);
3471 				check_block_count(sbi, segno,
3472 						&raw_sit->entries[sit_offset]);
3473 			}
3474 
3475 			__clear_bit(segno, bitmap);
3476 			sit_i->dirty_sentries--;
3477 			ses->entry_cnt--;
3478 		}
3479 
3480 		if (to_journal)
3481 			up_write(&curseg->journal_rwsem);
3482 		else
3483 			f2fs_put_page(page, 1);
3484 
3485 		f2fs_bug_on(sbi, ses->entry_cnt);
3486 		release_sit_entry_set(ses);
3487 	}
3488 
3489 	f2fs_bug_on(sbi, !list_empty(head));
3490 	f2fs_bug_on(sbi, sit_i->dirty_sentries);
3491 out:
3492 	if (cpc->reason & CP_DISCARD) {
3493 		__u64 trim_start = cpc->trim_start;
3494 
3495 		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3496 			add_discard_addrs(sbi, cpc, false);
3497 
3498 		cpc->trim_start = trim_start;
3499 	}
3500 	up_write(&sit_i->sentry_lock);
3501 
3502 	set_prefree_as_free_segments(sbi);
3503 }
3504 
3505 static int build_sit_info(struct f2fs_sb_info *sbi)
3506 {
3507 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3508 	struct sit_info *sit_i;
3509 	unsigned int sit_segs, start;
3510 	char *src_bitmap;
3511 	unsigned int bitmap_size;
3512 
3513 	/* allocate memory for SIT information */
3514 	sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3515 	if (!sit_i)
3516 		return -ENOMEM;
3517 
3518 	SM_I(sbi)->sit_info = sit_i;
3519 
3520 	sit_i->sentries =
3521 		f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3522 					      MAIN_SEGS(sbi)),
3523 			      GFP_KERNEL);
3524 	if (!sit_i->sentries)
3525 		return -ENOMEM;
3526 
3527 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3528 	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3529 								GFP_KERNEL);
3530 	if (!sit_i->dirty_sentries_bitmap)
3531 		return -ENOMEM;
3532 
3533 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
3534 		sit_i->sentries[start].cur_valid_map
3535 			= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3536 		sit_i->sentries[start].ckpt_valid_map
3537 			= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3538 		if (!sit_i->sentries[start].cur_valid_map ||
3539 				!sit_i->sentries[start].ckpt_valid_map)
3540 			return -ENOMEM;
3541 
3542 #ifdef CONFIG_F2FS_CHECK_FS
3543 		sit_i->sentries[start].cur_valid_map_mir
3544 			= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3545 		if (!sit_i->sentries[start].cur_valid_map_mir)
3546 			return -ENOMEM;
3547 #endif
3548 
3549 		if (f2fs_discard_en(sbi)) {
3550 			sit_i->sentries[start].discard_map
3551 				= f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3552 								GFP_KERNEL);
3553 			if (!sit_i->sentries[start].discard_map)
3554 				return -ENOMEM;
3555 		}
3556 	}
3557 
3558 	sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3559 	if (!sit_i->tmp_map)
3560 		return -ENOMEM;
3561 
3562 	if (sbi->segs_per_sec > 1) {
3563 		sit_i->sec_entries =
3564 			f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3565 						      MAIN_SECS(sbi)),
3566 				      GFP_KERNEL);
3567 		if (!sit_i->sec_entries)
3568 			return -ENOMEM;
3569 	}
3570 
3571 	/* get information related with SIT */
3572 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3573 
3574 	/* setup SIT bitmap from ckeckpoint pack */
3575 	bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3576 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3577 
3578 	sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3579 	if (!sit_i->sit_bitmap)
3580 		return -ENOMEM;
3581 
3582 #ifdef CONFIG_F2FS_CHECK_FS
3583 	sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3584 	if (!sit_i->sit_bitmap_mir)
3585 		return -ENOMEM;
3586 #endif
3587 
3588 	/* init SIT information */
3589 	sit_i->s_ops = &default_salloc_ops;
3590 
3591 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3592 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3593 	sit_i->written_valid_blocks = 0;
3594 	sit_i->bitmap_size = bitmap_size;
3595 	sit_i->dirty_sentries = 0;
3596 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3597 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3598 	sit_i->mounted_time = ktime_get_real_seconds();
3599 	init_rwsem(&sit_i->sentry_lock);
3600 	return 0;
3601 }
3602 
3603 static int build_free_segmap(struct f2fs_sb_info *sbi)
3604 {
3605 	struct free_segmap_info *free_i;
3606 	unsigned int bitmap_size, sec_bitmap_size;
3607 
3608 	/* allocate memory for free segmap information */
3609 	free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
3610 	if (!free_i)
3611 		return -ENOMEM;
3612 
3613 	SM_I(sbi)->free_info = free_i;
3614 
3615 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3616 	free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
3617 	if (!free_i->free_segmap)
3618 		return -ENOMEM;
3619 
3620 	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3621 	free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
3622 	if (!free_i->free_secmap)
3623 		return -ENOMEM;
3624 
3625 	/* set all segments as dirty temporarily */
3626 	memset(free_i->free_segmap, 0xff, bitmap_size);
3627 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3628 
3629 	/* init free segmap information */
3630 	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3631 	free_i->free_segments = 0;
3632 	free_i->free_sections = 0;
3633 	spin_lock_init(&free_i->segmap_lock);
3634 	return 0;
3635 }
3636 
3637 static int build_curseg(struct f2fs_sb_info *sbi)
3638 {
3639 	struct curseg_info *array;
3640 	int i;
3641 
3642 	array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
3643 			     GFP_KERNEL);
3644 	if (!array)
3645 		return -ENOMEM;
3646 
3647 	SM_I(sbi)->curseg_array = array;
3648 
3649 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
3650 		mutex_init(&array[i].curseg_mutex);
3651 		array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
3652 		if (!array[i].sum_blk)
3653 			return -ENOMEM;
3654 		init_rwsem(&array[i].journal_rwsem);
3655 		array[i].journal = f2fs_kzalloc(sbi,
3656 				sizeof(struct f2fs_journal), GFP_KERNEL);
3657 		if (!array[i].journal)
3658 			return -ENOMEM;
3659 		array[i].segno = NULL_SEGNO;
3660 		array[i].next_blkoff = 0;
3661 	}
3662 	return restore_curseg_summaries(sbi);
3663 }
3664 
3665 static int build_sit_entries(struct f2fs_sb_info *sbi)
3666 {
3667 	struct sit_info *sit_i = SIT_I(sbi);
3668 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3669 	struct f2fs_journal *journal = curseg->journal;
3670 	struct seg_entry *se;
3671 	struct f2fs_sit_entry sit;
3672 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
3673 	unsigned int i, start, end;
3674 	unsigned int readed, start_blk = 0;
3675 	int err = 0;
3676 	block_t total_node_blocks = 0;
3677 
3678 	do {
3679 		readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3680 							META_SIT, true);
3681 
3682 		start = start_blk * sit_i->sents_per_block;
3683 		end = (start_blk + readed) * sit_i->sents_per_block;
3684 
3685 		for (; start < end && start < MAIN_SEGS(sbi); start++) {
3686 			struct f2fs_sit_block *sit_blk;
3687 			struct page *page;
3688 
3689 			se = &sit_i->sentries[start];
3690 			page = get_current_sit_page(sbi, start);
3691 			sit_blk = (struct f2fs_sit_block *)page_address(page);
3692 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3693 			f2fs_put_page(page, 1);
3694 
3695 			err = check_block_count(sbi, start, &sit);
3696 			if (err)
3697 				return err;
3698 			seg_info_from_raw_sit(se, &sit);
3699 			if (IS_NODESEG(se->type))
3700 				total_node_blocks += se->valid_blocks;
3701 
3702 			/* build discard map only one time */
3703 			if (f2fs_discard_en(sbi)) {
3704 				if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3705 					memset(se->discard_map, 0xff,
3706 						SIT_VBLOCK_MAP_SIZE);
3707 				} else {
3708 					memcpy(se->discard_map,
3709 						se->cur_valid_map,
3710 						SIT_VBLOCK_MAP_SIZE);
3711 					sbi->discard_blks +=
3712 						sbi->blocks_per_seg -
3713 						se->valid_blocks;
3714 				}
3715 			}
3716 
3717 			if (sbi->segs_per_sec > 1)
3718 				get_sec_entry(sbi, start)->valid_blocks +=
3719 							se->valid_blocks;
3720 		}
3721 		start_blk += readed;
3722 	} while (start_blk < sit_blk_cnt);
3723 
3724 	down_read(&curseg->journal_rwsem);
3725 	for (i = 0; i < sits_in_cursum(journal); i++) {
3726 		unsigned int old_valid_blocks;
3727 
3728 		start = le32_to_cpu(segno_in_journal(journal, i));
3729 		if (start >= MAIN_SEGS(sbi)) {
3730 			f2fs_msg(sbi->sb, KERN_ERR,
3731 					"Wrong journal entry on segno %u",
3732 					start);
3733 			set_sbi_flag(sbi, SBI_NEED_FSCK);
3734 			err = -EINVAL;
3735 			break;
3736 		}
3737 
3738 		se = &sit_i->sentries[start];
3739 		sit = sit_in_journal(journal, i);
3740 
3741 		old_valid_blocks = se->valid_blocks;
3742 		if (IS_NODESEG(se->type))
3743 			total_node_blocks -= old_valid_blocks;
3744 
3745 		err = check_block_count(sbi, start, &sit);
3746 		if (err)
3747 			break;
3748 		seg_info_from_raw_sit(se, &sit);
3749 		if (IS_NODESEG(se->type))
3750 			total_node_blocks += se->valid_blocks;
3751 
3752 		if (f2fs_discard_en(sbi)) {
3753 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3754 				memset(se->discard_map, 0xff,
3755 							SIT_VBLOCK_MAP_SIZE);
3756 			} else {
3757 				memcpy(se->discard_map, se->cur_valid_map,
3758 							SIT_VBLOCK_MAP_SIZE);
3759 				sbi->discard_blks += old_valid_blocks;
3760 				sbi->discard_blks -= se->valid_blocks;
3761 			}
3762 		}
3763 
3764 		if (sbi->segs_per_sec > 1) {
3765 			get_sec_entry(sbi, start)->valid_blocks +=
3766 							se->valid_blocks;
3767 			get_sec_entry(sbi, start)->valid_blocks -=
3768 							old_valid_blocks;
3769 		}
3770 	}
3771 	up_read(&curseg->journal_rwsem);
3772 
3773 	if (!err && total_node_blocks != valid_node_count(sbi)) {
3774 		f2fs_msg(sbi->sb, KERN_ERR,
3775 			"SIT is corrupted node# %u vs %u",
3776 			total_node_blocks, valid_node_count(sbi));
3777 		set_sbi_flag(sbi, SBI_NEED_FSCK);
3778 		err = -EINVAL;
3779 	}
3780 
3781 	return err;
3782 }
3783 
3784 static void init_free_segmap(struct f2fs_sb_info *sbi)
3785 {
3786 	unsigned int start;
3787 	int type;
3788 
3789 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
3790 		struct seg_entry *sentry = get_seg_entry(sbi, start);
3791 		if (!sentry->valid_blocks)
3792 			__set_free(sbi, start);
3793 		else
3794 			SIT_I(sbi)->written_valid_blocks +=
3795 						sentry->valid_blocks;
3796 	}
3797 
3798 	/* set use the current segments */
3799 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3800 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3801 		__set_test_and_inuse(sbi, curseg_t->segno);
3802 	}
3803 }
3804 
3805 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3806 {
3807 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3808 	struct free_segmap_info *free_i = FREE_I(sbi);
3809 	unsigned int segno = 0, offset = 0;
3810 	unsigned short valid_blocks;
3811 
3812 	while (1) {
3813 		/* find dirty segment based on free segmap */
3814 		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3815 		if (segno >= MAIN_SEGS(sbi))
3816 			break;
3817 		offset = segno + 1;
3818 		valid_blocks = get_valid_blocks(sbi, segno, false);
3819 		if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
3820 			continue;
3821 		if (valid_blocks > sbi->blocks_per_seg) {
3822 			f2fs_bug_on(sbi, 1);
3823 			continue;
3824 		}
3825 		mutex_lock(&dirty_i->seglist_lock);
3826 		__locate_dirty_segment(sbi, segno, DIRTY);
3827 		mutex_unlock(&dirty_i->seglist_lock);
3828 	}
3829 }
3830 
3831 static int init_victim_secmap(struct f2fs_sb_info *sbi)
3832 {
3833 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3834 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3835 
3836 	dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
3837 	if (!dirty_i->victim_secmap)
3838 		return -ENOMEM;
3839 	return 0;
3840 }
3841 
3842 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3843 {
3844 	struct dirty_seglist_info *dirty_i;
3845 	unsigned int bitmap_size, i;
3846 
3847 	/* allocate memory for dirty segments list information */
3848 	dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
3849 								GFP_KERNEL);
3850 	if (!dirty_i)
3851 		return -ENOMEM;
3852 
3853 	SM_I(sbi)->dirty_info = dirty_i;
3854 	mutex_init(&dirty_i->seglist_lock);
3855 
3856 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3857 
3858 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
3859 		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
3860 								GFP_KERNEL);
3861 		if (!dirty_i->dirty_segmap[i])
3862 			return -ENOMEM;
3863 	}
3864 
3865 	init_dirty_segmap(sbi);
3866 	return init_victim_secmap(sbi);
3867 }
3868 
3869 /*
3870  * Update min, max modified time for cost-benefit GC algorithm
3871  */
3872 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3873 {
3874 	struct sit_info *sit_i = SIT_I(sbi);
3875 	unsigned int segno;
3876 
3877 	down_write(&sit_i->sentry_lock);
3878 
3879 	sit_i->min_mtime = ULLONG_MAX;
3880 
3881 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
3882 		unsigned int i;
3883 		unsigned long long mtime = 0;
3884 
3885 		for (i = 0; i < sbi->segs_per_sec; i++)
3886 			mtime += get_seg_entry(sbi, segno + i)->mtime;
3887 
3888 		mtime = div_u64(mtime, sbi->segs_per_sec);
3889 
3890 		if (sit_i->min_mtime > mtime)
3891 			sit_i->min_mtime = mtime;
3892 	}
3893 	sit_i->max_mtime = get_mtime(sbi, false);
3894 	up_write(&sit_i->sentry_lock);
3895 }
3896 
3897 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
3898 {
3899 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3900 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3901 	struct f2fs_sm_info *sm_info;
3902 	int err;
3903 
3904 	sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
3905 	if (!sm_info)
3906 		return -ENOMEM;
3907 
3908 	/* init sm info */
3909 	sbi->sm_info = sm_info;
3910 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3911 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3912 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3913 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3914 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3915 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3916 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
3917 	sm_info->rec_prefree_segments = sm_info->main_segments *
3918 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
3919 	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3920 		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3921 
3922 	if (!test_opt(sbi, LFS))
3923 		sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
3924 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
3925 	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
3926 	sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
3927 	sm_info->min_ssr_sections = reserved_sections(sbi);
3928 
3929 	INIT_LIST_HEAD(&sm_info->sit_entry_set);
3930 
3931 	init_rwsem(&sm_info->curseg_lock);
3932 
3933 	if (!f2fs_readonly(sbi->sb)) {
3934 		err = f2fs_create_flush_cmd_control(sbi);
3935 		if (err)
3936 			return err;
3937 	}
3938 
3939 	err = create_discard_cmd_control(sbi);
3940 	if (err)
3941 		return err;
3942 
3943 	err = build_sit_info(sbi);
3944 	if (err)
3945 		return err;
3946 	err = build_free_segmap(sbi);
3947 	if (err)
3948 		return err;
3949 	err = build_curseg(sbi);
3950 	if (err)
3951 		return err;
3952 
3953 	/* reinit free segmap based on SIT */
3954 	err = build_sit_entries(sbi);
3955 	if (err)
3956 		return err;
3957 
3958 	init_free_segmap(sbi);
3959 	err = build_dirty_segmap(sbi);
3960 	if (err)
3961 		return err;
3962 
3963 	init_min_max_mtime(sbi);
3964 	return 0;
3965 }
3966 
3967 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3968 		enum dirty_type dirty_type)
3969 {
3970 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3971 
3972 	mutex_lock(&dirty_i->seglist_lock);
3973 	kvfree(dirty_i->dirty_segmap[dirty_type]);
3974 	dirty_i->nr_dirty[dirty_type] = 0;
3975 	mutex_unlock(&dirty_i->seglist_lock);
3976 }
3977 
3978 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
3979 {
3980 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3981 	kvfree(dirty_i->victim_secmap);
3982 }
3983 
3984 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3985 {
3986 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3987 	int i;
3988 
3989 	if (!dirty_i)
3990 		return;
3991 
3992 	/* discard pre-free/dirty segments list */
3993 	for (i = 0; i < NR_DIRTY_TYPE; i++)
3994 		discard_dirty_segmap(sbi, i);
3995 
3996 	destroy_victim_secmap(sbi);
3997 	SM_I(sbi)->dirty_info = NULL;
3998 	kfree(dirty_i);
3999 }
4000 
4001 static void destroy_curseg(struct f2fs_sb_info *sbi)
4002 {
4003 	struct curseg_info *array = SM_I(sbi)->curseg_array;
4004 	int i;
4005 
4006 	if (!array)
4007 		return;
4008 	SM_I(sbi)->curseg_array = NULL;
4009 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
4010 		kfree(array[i].sum_blk);
4011 		kfree(array[i].journal);
4012 	}
4013 	kfree(array);
4014 }
4015 
4016 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4017 {
4018 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4019 	if (!free_i)
4020 		return;
4021 	SM_I(sbi)->free_info = NULL;
4022 	kvfree(free_i->free_segmap);
4023 	kvfree(free_i->free_secmap);
4024 	kfree(free_i);
4025 }
4026 
4027 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4028 {
4029 	struct sit_info *sit_i = SIT_I(sbi);
4030 	unsigned int start;
4031 
4032 	if (!sit_i)
4033 		return;
4034 
4035 	if (sit_i->sentries) {
4036 		for (start = 0; start < MAIN_SEGS(sbi); start++) {
4037 			kfree(sit_i->sentries[start].cur_valid_map);
4038 #ifdef CONFIG_F2FS_CHECK_FS
4039 			kfree(sit_i->sentries[start].cur_valid_map_mir);
4040 #endif
4041 			kfree(sit_i->sentries[start].ckpt_valid_map);
4042 			kfree(sit_i->sentries[start].discard_map);
4043 		}
4044 	}
4045 	kfree(sit_i->tmp_map);
4046 
4047 	kvfree(sit_i->sentries);
4048 	kvfree(sit_i->sec_entries);
4049 	kvfree(sit_i->dirty_sentries_bitmap);
4050 
4051 	SM_I(sbi)->sit_info = NULL;
4052 	kfree(sit_i->sit_bitmap);
4053 #ifdef CONFIG_F2FS_CHECK_FS
4054 	kfree(sit_i->sit_bitmap_mir);
4055 #endif
4056 	kfree(sit_i);
4057 }
4058 
4059 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4060 {
4061 	struct f2fs_sm_info *sm_info = SM_I(sbi);
4062 
4063 	if (!sm_info)
4064 		return;
4065 	f2fs_destroy_flush_cmd_control(sbi, true);
4066 	destroy_discard_cmd_control(sbi);
4067 	destroy_dirty_segmap(sbi);
4068 	destroy_curseg(sbi);
4069 	destroy_free_segmap(sbi);
4070 	destroy_sit_info(sbi);
4071 	sbi->sm_info = NULL;
4072 	kfree(sm_info);
4073 }
4074 
4075 int __init f2fs_create_segment_manager_caches(void)
4076 {
4077 	discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4078 			sizeof(struct discard_entry));
4079 	if (!discard_entry_slab)
4080 		goto fail;
4081 
4082 	discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4083 			sizeof(struct discard_cmd));
4084 	if (!discard_cmd_slab)
4085 		goto destroy_discard_entry;
4086 
4087 	sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4088 			sizeof(struct sit_entry_set));
4089 	if (!sit_entry_set_slab)
4090 		goto destroy_discard_cmd;
4091 
4092 	inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4093 			sizeof(struct inmem_pages));
4094 	if (!inmem_entry_slab)
4095 		goto destroy_sit_entry_set;
4096 	return 0;
4097 
4098 destroy_sit_entry_set:
4099 	kmem_cache_destroy(sit_entry_set_slab);
4100 destroy_discard_cmd:
4101 	kmem_cache_destroy(discard_cmd_slab);
4102 destroy_discard_entry:
4103 	kmem_cache_destroy(discard_entry_slab);
4104 fail:
4105 	return -ENOMEM;
4106 }
4107 
4108 void f2fs_destroy_segment_manager_caches(void)
4109 {
4110 	kmem_cache_destroy(sit_entry_set_slab);
4111 	kmem_cache_destroy(discard_cmd_slab);
4112 	kmem_cache_destroy(discard_entry_slab);
4113 	kmem_cache_destroy(inmem_entry_slab);
4114 }
4115