xref: /linux/fs/f2fs/segment.c (revision a36e9f5cfe9eb3a1dce8769c7058251c42705357)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * fs/f2fs/segment.c
4  *
5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6  *             http://www.samsung.com/
7  */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/bio.h>
11 #include <linux/blkdev.h>
12 #include <linux/sched/mm.h>
13 #include <linux/prefetch.h>
14 #include <linux/kthread.h>
15 #include <linux/swap.h>
16 #include <linux/timer.h>
17 #include <linux/freezer.h>
18 #include <linux/sched/signal.h>
19 #include <linux/random.h>
20 
21 #include "f2fs.h"
22 #include "segment.h"
23 #include "node.h"
24 #include "gc.h"
25 #include "iostat.h"
26 #include <trace/events/f2fs.h>
27 
28 #define __reverse_ffz(x) __reverse_ffs(~(x))
29 
30 static struct kmem_cache *discard_entry_slab;
31 static struct kmem_cache *discard_cmd_slab;
32 static struct kmem_cache *sit_entry_set_slab;
33 static struct kmem_cache *revoke_entry_slab;
34 
35 static unsigned long __reverse_ulong(unsigned char *str)
36 {
37 	unsigned long tmp = 0;
38 	int shift = 24, idx = 0;
39 
40 #if BITS_PER_LONG == 64
41 	shift = 56;
42 #endif
43 	while (shift >= 0) {
44 		tmp |= (unsigned long)str[idx++] << shift;
45 		shift -= BITS_PER_BYTE;
46 	}
47 	return tmp;
48 }
49 
50 /*
51  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
52  * MSB and LSB are reversed in a byte by f2fs_set_bit.
53  */
54 static inline unsigned long __reverse_ffs(unsigned long word)
55 {
56 	int num = 0;
57 
58 #if BITS_PER_LONG == 64
59 	if ((word & 0xffffffff00000000UL) == 0)
60 		num += 32;
61 	else
62 		word >>= 32;
63 #endif
64 	if ((word & 0xffff0000) == 0)
65 		num += 16;
66 	else
67 		word >>= 16;
68 
69 	if ((word & 0xff00) == 0)
70 		num += 8;
71 	else
72 		word >>= 8;
73 
74 	if ((word & 0xf0) == 0)
75 		num += 4;
76 	else
77 		word >>= 4;
78 
79 	if ((word & 0xc) == 0)
80 		num += 2;
81 	else
82 		word >>= 2;
83 
84 	if ((word & 0x2) == 0)
85 		num += 1;
86 	return num;
87 }
88 
89 /*
90  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
91  * f2fs_set_bit makes MSB and LSB reversed in a byte.
92  * @size must be integral times of unsigned long.
93  * Example:
94  *                             MSB <--> LSB
95  *   f2fs_set_bit(0, bitmap) => 1000 0000
96  *   f2fs_set_bit(7, bitmap) => 0000 0001
97  */
98 static unsigned long __find_rev_next_bit(const unsigned long *addr,
99 			unsigned long size, unsigned long offset)
100 {
101 	const unsigned long *p = addr + BIT_WORD(offset);
102 	unsigned long result = size;
103 	unsigned long tmp;
104 
105 	if (offset >= size)
106 		return size;
107 
108 	size -= (offset & ~(BITS_PER_LONG - 1));
109 	offset %= BITS_PER_LONG;
110 
111 	while (1) {
112 		if (*p == 0)
113 			goto pass;
114 
115 		tmp = __reverse_ulong((unsigned char *)p);
116 
117 		tmp &= ~0UL >> offset;
118 		if (size < BITS_PER_LONG)
119 			tmp &= (~0UL << (BITS_PER_LONG - size));
120 		if (tmp)
121 			goto found;
122 pass:
123 		if (size <= BITS_PER_LONG)
124 			break;
125 		size -= BITS_PER_LONG;
126 		offset = 0;
127 		p++;
128 	}
129 	return result;
130 found:
131 	return result - size + __reverse_ffs(tmp);
132 }
133 
134 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
135 			unsigned long size, unsigned long offset)
136 {
137 	const unsigned long *p = addr + BIT_WORD(offset);
138 	unsigned long result = size;
139 	unsigned long tmp;
140 
141 	if (offset >= size)
142 		return size;
143 
144 	size -= (offset & ~(BITS_PER_LONG - 1));
145 	offset %= BITS_PER_LONG;
146 
147 	while (1) {
148 		if (*p == ~0UL)
149 			goto pass;
150 
151 		tmp = __reverse_ulong((unsigned char *)p);
152 
153 		if (offset)
154 			tmp |= ~0UL << (BITS_PER_LONG - offset);
155 		if (size < BITS_PER_LONG)
156 			tmp |= ~0UL >> size;
157 		if (tmp != ~0UL)
158 			goto found;
159 pass:
160 		if (size <= BITS_PER_LONG)
161 			break;
162 		size -= BITS_PER_LONG;
163 		offset = 0;
164 		p++;
165 	}
166 	return result;
167 found:
168 	return result - size + __reverse_ffz(tmp);
169 }
170 
171 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
172 {
173 	int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
174 	int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
175 	int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
176 
177 	if (f2fs_lfs_mode(sbi))
178 		return false;
179 	if (sbi->gc_mode == GC_URGENT_HIGH)
180 		return true;
181 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
182 		return true;
183 
184 	return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
185 			SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
186 }
187 
188 void f2fs_abort_atomic_write(struct inode *inode, bool clean)
189 {
190 	struct f2fs_inode_info *fi = F2FS_I(inode);
191 
192 	if (!f2fs_is_atomic_file(inode))
193 		return;
194 
195 	if (clean)
196 		truncate_inode_pages_final(inode->i_mapping);
197 
198 	release_atomic_write_cnt(inode);
199 	clear_inode_flag(inode, FI_ATOMIC_COMMITTED);
200 	clear_inode_flag(inode, FI_ATOMIC_REPLACE);
201 	clear_inode_flag(inode, FI_ATOMIC_FILE);
202 	stat_dec_atomic_inode(inode);
203 
204 	F2FS_I(inode)->atomic_write_task = NULL;
205 
206 	if (clean) {
207 		f2fs_i_size_write(inode, fi->original_i_size);
208 		fi->original_i_size = 0;
209 	}
210 	/* avoid stale dirty inode during eviction */
211 	sync_inode_metadata(inode, 0);
212 }
213 
214 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index,
215 			block_t new_addr, block_t *old_addr, bool recover)
216 {
217 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
218 	struct dnode_of_data dn;
219 	struct node_info ni;
220 	int err;
221 
222 retry:
223 	set_new_dnode(&dn, inode, NULL, NULL, 0);
224 	err = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE);
225 	if (err) {
226 		if (err == -ENOMEM) {
227 			f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
228 			goto retry;
229 		}
230 		return err;
231 	}
232 
233 	err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
234 	if (err) {
235 		f2fs_put_dnode(&dn);
236 		return err;
237 	}
238 
239 	if (recover) {
240 		/* dn.data_blkaddr is always valid */
241 		if (!__is_valid_data_blkaddr(new_addr)) {
242 			if (new_addr == NULL_ADDR)
243 				dec_valid_block_count(sbi, inode, 1);
244 			f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
245 			f2fs_update_data_blkaddr(&dn, new_addr);
246 		} else {
247 			f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
248 				new_addr, ni.version, true, true);
249 		}
250 	} else {
251 		blkcnt_t count = 1;
252 
253 		err = inc_valid_block_count(sbi, inode, &count, true);
254 		if (err) {
255 			f2fs_put_dnode(&dn);
256 			return err;
257 		}
258 
259 		*old_addr = dn.data_blkaddr;
260 		f2fs_truncate_data_blocks_range(&dn, 1);
261 		dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count);
262 
263 		f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr,
264 					ni.version, true, false);
265 	}
266 
267 	f2fs_put_dnode(&dn);
268 
269 	trace_f2fs_replace_atomic_write_block(inode, F2FS_I(inode)->cow_inode,
270 			index, old_addr ? *old_addr : 0, new_addr, recover);
271 	return 0;
272 }
273 
274 static void __complete_revoke_list(struct inode *inode, struct list_head *head,
275 					bool revoke)
276 {
277 	struct revoke_entry *cur, *tmp;
278 	pgoff_t start_index = 0;
279 	bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE);
280 
281 	list_for_each_entry_safe(cur, tmp, head, list) {
282 		if (revoke) {
283 			__replace_atomic_write_block(inode, cur->index,
284 						cur->old_addr, NULL, true);
285 		} else if (truncate) {
286 			f2fs_truncate_hole(inode, start_index, cur->index);
287 			start_index = cur->index + 1;
288 		}
289 
290 		list_del(&cur->list);
291 		kmem_cache_free(revoke_entry_slab, cur);
292 	}
293 
294 	if (!revoke && truncate)
295 		f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false);
296 }
297 
298 static int __f2fs_commit_atomic_write(struct inode *inode)
299 {
300 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
301 	struct f2fs_inode_info *fi = F2FS_I(inode);
302 	struct inode *cow_inode = fi->cow_inode;
303 	struct revoke_entry *new;
304 	struct list_head revoke_list;
305 	block_t blkaddr;
306 	struct dnode_of_data dn;
307 	pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
308 	pgoff_t off = 0, blen, index;
309 	int ret = 0, i;
310 
311 	INIT_LIST_HEAD(&revoke_list);
312 
313 	while (len) {
314 		blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len);
315 
316 		set_new_dnode(&dn, cow_inode, NULL, NULL, 0);
317 		ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA);
318 		if (ret && ret != -ENOENT) {
319 			goto out;
320 		} else if (ret == -ENOENT) {
321 			ret = 0;
322 			if (dn.max_level == 0)
323 				goto out;
324 			goto next;
325 		}
326 
327 		blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode),
328 				len);
329 		index = off;
330 		for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) {
331 			blkaddr = f2fs_data_blkaddr(&dn);
332 
333 			if (!__is_valid_data_blkaddr(blkaddr)) {
334 				continue;
335 			} else if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
336 					DATA_GENERIC_ENHANCE)) {
337 				f2fs_put_dnode(&dn);
338 				ret = -EFSCORRUPTED;
339 				goto out;
340 			}
341 
342 			new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS,
343 							true, NULL);
344 
345 			ret = __replace_atomic_write_block(inode, index, blkaddr,
346 							&new->old_addr, false);
347 			if (ret) {
348 				f2fs_put_dnode(&dn);
349 				kmem_cache_free(revoke_entry_slab, new);
350 				goto out;
351 			}
352 
353 			f2fs_update_data_blkaddr(&dn, NULL_ADDR);
354 			new->index = index;
355 			list_add_tail(&new->list, &revoke_list);
356 		}
357 		f2fs_put_dnode(&dn);
358 next:
359 		off += blen;
360 		len -= blen;
361 	}
362 
363 out:
364 	if (ret) {
365 		sbi->revoked_atomic_block += fi->atomic_write_cnt;
366 	} else {
367 		sbi->committed_atomic_block += fi->atomic_write_cnt;
368 		set_inode_flag(inode, FI_ATOMIC_COMMITTED);
369 	}
370 
371 	__complete_revoke_list(inode, &revoke_list, ret ? true : false);
372 
373 	return ret;
374 }
375 
376 int f2fs_commit_atomic_write(struct inode *inode)
377 {
378 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
379 	struct f2fs_inode_info *fi = F2FS_I(inode);
380 	int err;
381 
382 	err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
383 	if (err)
384 		return err;
385 
386 	f2fs_down_write(&fi->i_gc_rwsem[WRITE]);
387 	f2fs_lock_op(sbi);
388 
389 	err = __f2fs_commit_atomic_write(inode);
390 
391 	f2fs_unlock_op(sbi);
392 	f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
393 
394 	return err;
395 }
396 
397 /*
398  * This function balances dirty node and dentry pages.
399  * In addition, it controls garbage collection.
400  */
401 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
402 {
403 	if (f2fs_cp_error(sbi))
404 		return;
405 
406 	if (time_to_inject(sbi, FAULT_CHECKPOINT))
407 		f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT);
408 
409 	/* balance_fs_bg is able to be pending */
410 	if (need && excess_cached_nats(sbi))
411 		f2fs_balance_fs_bg(sbi, false);
412 
413 	if (!f2fs_is_checkpoint_ready(sbi))
414 		return;
415 
416 	/*
417 	 * We should do GC or end up with checkpoint, if there are so many dirty
418 	 * dir/node pages without enough free segments.
419 	 */
420 	if (has_enough_free_secs(sbi, 0, 0))
421 		return;
422 
423 	if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
424 				sbi->gc_thread->f2fs_gc_task) {
425 		DEFINE_WAIT(wait);
426 
427 		prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
428 					TASK_UNINTERRUPTIBLE);
429 		wake_up(&sbi->gc_thread->gc_wait_queue_head);
430 		io_schedule();
431 		finish_wait(&sbi->gc_thread->fggc_wq, &wait);
432 	} else {
433 		struct f2fs_gc_control gc_control = {
434 			.victim_segno = NULL_SEGNO,
435 			.init_gc_type = BG_GC,
436 			.no_bg_gc = true,
437 			.should_migrate_blocks = false,
438 			.err_gc_skipped = false,
439 			.nr_free_secs = 1 };
440 		f2fs_down_write(&sbi->gc_lock);
441 		stat_inc_gc_call_count(sbi, FOREGROUND);
442 		f2fs_gc(sbi, &gc_control);
443 	}
444 }
445 
446 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi)
447 {
448 	int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2;
449 	unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS);
450 	unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA);
451 	unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES);
452 	unsigned int meta = get_pages(sbi, F2FS_DIRTY_META);
453 	unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA);
454 	unsigned int threshold =
455 		SEGS_TO_BLKS(sbi, (factor * DEFAULT_DIRTY_THRESHOLD));
456 	unsigned int global_threshold = threshold * 3 / 2;
457 
458 	if (dents >= threshold || qdata >= threshold ||
459 		nodes >= threshold || meta >= threshold ||
460 		imeta >= threshold)
461 		return true;
462 	return dents + qdata + nodes + meta + imeta >  global_threshold;
463 }
464 
465 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
466 {
467 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
468 		return;
469 
470 	/* try to shrink extent cache when there is no enough memory */
471 	if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE))
472 		f2fs_shrink_read_extent_tree(sbi,
473 				READ_EXTENT_CACHE_SHRINK_NUMBER);
474 
475 	/* try to shrink age extent cache when there is no enough memory */
476 	if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE))
477 		f2fs_shrink_age_extent_tree(sbi,
478 				AGE_EXTENT_CACHE_SHRINK_NUMBER);
479 
480 	/* check the # of cached NAT entries */
481 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
482 		f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
483 
484 	if (!f2fs_available_free_memory(sbi, FREE_NIDS))
485 		f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
486 	else
487 		f2fs_build_free_nids(sbi, false, false);
488 
489 	if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) ||
490 		excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi))
491 		goto do_sync;
492 
493 	/* there is background inflight IO or foreground operation recently */
494 	if (is_inflight_io(sbi, REQ_TIME) ||
495 		(!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem)))
496 		return;
497 
498 	/* exceed periodical checkpoint timeout threshold */
499 	if (f2fs_time_over(sbi, CP_TIME))
500 		goto do_sync;
501 
502 	/* checkpoint is the only way to shrink partial cached entries */
503 	if (f2fs_available_free_memory(sbi, NAT_ENTRIES) &&
504 		f2fs_available_free_memory(sbi, INO_ENTRIES))
505 		return;
506 
507 do_sync:
508 	if (test_opt(sbi, DATA_FLUSH) && from_bg) {
509 		struct blk_plug plug;
510 
511 		mutex_lock(&sbi->flush_lock);
512 
513 		blk_start_plug(&plug);
514 		f2fs_sync_dirty_inodes(sbi, FILE_INODE, false);
515 		blk_finish_plug(&plug);
516 
517 		mutex_unlock(&sbi->flush_lock);
518 	}
519 	stat_inc_cp_call_count(sbi, BACKGROUND);
520 	f2fs_sync_fs(sbi->sb, 1);
521 }
522 
523 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
524 				struct block_device *bdev)
525 {
526 	int ret = blkdev_issue_flush(bdev);
527 
528 	trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
529 				test_opt(sbi, FLUSH_MERGE), ret);
530 	if (!ret)
531 		f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0);
532 	return ret;
533 }
534 
535 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
536 {
537 	int ret = 0;
538 	int i;
539 
540 	if (!f2fs_is_multi_device(sbi))
541 		return __submit_flush_wait(sbi, sbi->sb->s_bdev);
542 
543 	for (i = 0; i < sbi->s_ndevs; i++) {
544 		if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
545 			continue;
546 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
547 		if (ret)
548 			break;
549 	}
550 	return ret;
551 }
552 
553 static int issue_flush_thread(void *data)
554 {
555 	struct f2fs_sb_info *sbi = data;
556 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
557 	wait_queue_head_t *q = &fcc->flush_wait_queue;
558 repeat:
559 	if (kthread_should_stop())
560 		return 0;
561 
562 	if (!llist_empty(&fcc->issue_list)) {
563 		struct flush_cmd *cmd, *next;
564 		int ret;
565 
566 		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
567 		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
568 
569 		cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
570 
571 		ret = submit_flush_wait(sbi, cmd->ino);
572 		atomic_inc(&fcc->issued_flush);
573 
574 		llist_for_each_entry_safe(cmd, next,
575 					  fcc->dispatch_list, llnode) {
576 			cmd->ret = ret;
577 			complete(&cmd->wait);
578 		}
579 		fcc->dispatch_list = NULL;
580 	}
581 
582 	wait_event_interruptible(*q,
583 		kthread_should_stop() || !llist_empty(&fcc->issue_list));
584 	goto repeat;
585 }
586 
587 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
588 {
589 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
590 	struct flush_cmd cmd;
591 	int ret;
592 
593 	if (test_opt(sbi, NOBARRIER))
594 		return 0;
595 
596 	if (!test_opt(sbi, FLUSH_MERGE)) {
597 		atomic_inc(&fcc->queued_flush);
598 		ret = submit_flush_wait(sbi, ino);
599 		atomic_dec(&fcc->queued_flush);
600 		atomic_inc(&fcc->issued_flush);
601 		return ret;
602 	}
603 
604 	if (atomic_inc_return(&fcc->queued_flush) == 1 ||
605 	    f2fs_is_multi_device(sbi)) {
606 		ret = submit_flush_wait(sbi, ino);
607 		atomic_dec(&fcc->queued_flush);
608 
609 		atomic_inc(&fcc->issued_flush);
610 		return ret;
611 	}
612 
613 	cmd.ino = ino;
614 	init_completion(&cmd.wait);
615 
616 	llist_add(&cmd.llnode, &fcc->issue_list);
617 
618 	/*
619 	 * update issue_list before we wake up issue_flush thread, this
620 	 * smp_mb() pairs with another barrier in ___wait_event(), see
621 	 * more details in comments of waitqueue_active().
622 	 */
623 	smp_mb();
624 
625 	if (waitqueue_active(&fcc->flush_wait_queue))
626 		wake_up(&fcc->flush_wait_queue);
627 
628 	if (fcc->f2fs_issue_flush) {
629 		wait_for_completion(&cmd.wait);
630 		atomic_dec(&fcc->queued_flush);
631 	} else {
632 		struct llist_node *list;
633 
634 		list = llist_del_all(&fcc->issue_list);
635 		if (!list) {
636 			wait_for_completion(&cmd.wait);
637 			atomic_dec(&fcc->queued_flush);
638 		} else {
639 			struct flush_cmd *tmp, *next;
640 
641 			ret = submit_flush_wait(sbi, ino);
642 
643 			llist_for_each_entry_safe(tmp, next, list, llnode) {
644 				if (tmp == &cmd) {
645 					cmd.ret = ret;
646 					atomic_dec(&fcc->queued_flush);
647 					continue;
648 				}
649 				tmp->ret = ret;
650 				complete(&tmp->wait);
651 			}
652 		}
653 	}
654 
655 	return cmd.ret;
656 }
657 
658 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
659 {
660 	dev_t dev = sbi->sb->s_bdev->bd_dev;
661 	struct flush_cmd_control *fcc;
662 
663 	if (SM_I(sbi)->fcc_info) {
664 		fcc = SM_I(sbi)->fcc_info;
665 		if (fcc->f2fs_issue_flush)
666 			return 0;
667 		goto init_thread;
668 	}
669 
670 	fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
671 	if (!fcc)
672 		return -ENOMEM;
673 	atomic_set(&fcc->issued_flush, 0);
674 	atomic_set(&fcc->queued_flush, 0);
675 	init_waitqueue_head(&fcc->flush_wait_queue);
676 	init_llist_head(&fcc->issue_list);
677 	SM_I(sbi)->fcc_info = fcc;
678 	if (!test_opt(sbi, FLUSH_MERGE))
679 		return 0;
680 
681 init_thread:
682 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
683 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
684 	if (IS_ERR(fcc->f2fs_issue_flush)) {
685 		int err = PTR_ERR(fcc->f2fs_issue_flush);
686 
687 		fcc->f2fs_issue_flush = NULL;
688 		return err;
689 	}
690 
691 	return 0;
692 }
693 
694 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
695 {
696 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
697 
698 	if (fcc && fcc->f2fs_issue_flush) {
699 		struct task_struct *flush_thread = fcc->f2fs_issue_flush;
700 
701 		fcc->f2fs_issue_flush = NULL;
702 		kthread_stop(flush_thread);
703 	}
704 	if (free) {
705 		kfree(fcc);
706 		SM_I(sbi)->fcc_info = NULL;
707 	}
708 }
709 
710 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
711 {
712 	int ret = 0, i;
713 
714 	if (!f2fs_is_multi_device(sbi))
715 		return 0;
716 
717 	if (test_opt(sbi, NOBARRIER))
718 		return 0;
719 
720 	for (i = 1; i < sbi->s_ndevs; i++) {
721 		int count = DEFAULT_RETRY_IO_COUNT;
722 
723 		if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
724 			continue;
725 
726 		do {
727 			ret = __submit_flush_wait(sbi, FDEV(i).bdev);
728 			if (ret)
729 				f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
730 		} while (ret && --count);
731 
732 		if (ret) {
733 			f2fs_stop_checkpoint(sbi, false,
734 					STOP_CP_REASON_FLUSH_FAIL);
735 			break;
736 		}
737 
738 		spin_lock(&sbi->dev_lock);
739 		f2fs_clear_bit(i, (char *)&sbi->dirty_device);
740 		spin_unlock(&sbi->dev_lock);
741 	}
742 
743 	return ret;
744 }
745 
746 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
747 		enum dirty_type dirty_type)
748 {
749 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
750 
751 	/* need not be added */
752 	if (IS_CURSEG(sbi, segno))
753 		return;
754 
755 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
756 		dirty_i->nr_dirty[dirty_type]++;
757 
758 	if (dirty_type == DIRTY) {
759 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
760 		enum dirty_type t = sentry->type;
761 
762 		if (unlikely(t >= DIRTY)) {
763 			f2fs_bug_on(sbi, 1);
764 			return;
765 		}
766 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
767 			dirty_i->nr_dirty[t]++;
768 
769 		if (__is_large_section(sbi)) {
770 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
771 			block_t valid_blocks =
772 				get_valid_blocks(sbi, segno, true);
773 
774 			f2fs_bug_on(sbi,
775 				(!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
776 				!valid_blocks) ||
777 				valid_blocks == CAP_BLKS_PER_SEC(sbi));
778 
779 			if (!IS_CURSEC(sbi, secno))
780 				set_bit(secno, dirty_i->dirty_secmap);
781 		}
782 	}
783 }
784 
785 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
786 		enum dirty_type dirty_type)
787 {
788 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
789 	block_t valid_blocks;
790 
791 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
792 		dirty_i->nr_dirty[dirty_type]--;
793 
794 	if (dirty_type == DIRTY) {
795 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
796 		enum dirty_type t = sentry->type;
797 
798 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
799 			dirty_i->nr_dirty[t]--;
800 
801 		valid_blocks = get_valid_blocks(sbi, segno, true);
802 		if (valid_blocks == 0) {
803 			clear_bit(GET_SEC_FROM_SEG(sbi, segno),
804 						dirty_i->victim_secmap);
805 #ifdef CONFIG_F2FS_CHECK_FS
806 			clear_bit(segno, SIT_I(sbi)->invalid_segmap);
807 #endif
808 		}
809 		if (__is_large_section(sbi)) {
810 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
811 
812 			if (!valid_blocks ||
813 					valid_blocks == CAP_BLKS_PER_SEC(sbi)) {
814 				clear_bit(secno, dirty_i->dirty_secmap);
815 				return;
816 			}
817 
818 			if (!IS_CURSEC(sbi, secno))
819 				set_bit(secno, dirty_i->dirty_secmap);
820 		}
821 	}
822 }
823 
824 /*
825  * Should not occur error such as -ENOMEM.
826  * Adding dirty entry into seglist is not critical operation.
827  * If a given segment is one of current working segments, it won't be added.
828  */
829 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
830 {
831 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
832 	unsigned short valid_blocks, ckpt_valid_blocks;
833 	unsigned int usable_blocks;
834 
835 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
836 		return;
837 
838 	usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
839 	mutex_lock(&dirty_i->seglist_lock);
840 
841 	valid_blocks = get_valid_blocks(sbi, segno, false);
842 	ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
843 
844 	if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
845 		ckpt_valid_blocks == usable_blocks)) {
846 		__locate_dirty_segment(sbi, segno, PRE);
847 		__remove_dirty_segment(sbi, segno, DIRTY);
848 	} else if (valid_blocks < usable_blocks) {
849 		__locate_dirty_segment(sbi, segno, DIRTY);
850 	} else {
851 		/* Recovery routine with SSR needs this */
852 		__remove_dirty_segment(sbi, segno, DIRTY);
853 	}
854 
855 	mutex_unlock(&dirty_i->seglist_lock);
856 }
857 
858 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
859 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
860 {
861 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
862 	unsigned int segno;
863 
864 	mutex_lock(&dirty_i->seglist_lock);
865 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
866 		if (get_valid_blocks(sbi, segno, false))
867 			continue;
868 		if (IS_CURSEG(sbi, segno))
869 			continue;
870 		__locate_dirty_segment(sbi, segno, PRE);
871 		__remove_dirty_segment(sbi, segno, DIRTY);
872 	}
873 	mutex_unlock(&dirty_i->seglist_lock);
874 }
875 
876 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
877 {
878 	int ovp_hole_segs =
879 		(overprovision_segments(sbi) - reserved_segments(sbi));
880 	block_t ovp_holes = SEGS_TO_BLKS(sbi, ovp_hole_segs);
881 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
882 	block_t holes[2] = {0, 0};	/* DATA and NODE */
883 	block_t unusable;
884 	struct seg_entry *se;
885 	unsigned int segno;
886 
887 	mutex_lock(&dirty_i->seglist_lock);
888 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
889 		se = get_seg_entry(sbi, segno);
890 		if (IS_NODESEG(se->type))
891 			holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
892 							se->valid_blocks;
893 		else
894 			holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
895 							se->valid_blocks;
896 	}
897 	mutex_unlock(&dirty_i->seglist_lock);
898 
899 	unusable = max(holes[DATA], holes[NODE]);
900 	if (unusable > ovp_holes)
901 		return unusable - ovp_holes;
902 	return 0;
903 }
904 
905 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
906 {
907 	int ovp_hole_segs =
908 		(overprovision_segments(sbi) - reserved_segments(sbi));
909 
910 	if (F2FS_OPTION(sbi).unusable_cap_perc == 100)
911 		return 0;
912 	if (unusable > F2FS_OPTION(sbi).unusable_cap)
913 		return -EAGAIN;
914 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
915 		dirty_segments(sbi) > ovp_hole_segs)
916 		return -EAGAIN;
917 	if (has_not_enough_free_secs(sbi, 0, 0))
918 		return -EAGAIN;
919 	return 0;
920 }
921 
922 /* This is only used by SBI_CP_DISABLED */
923 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
924 {
925 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
926 	unsigned int segno = 0;
927 
928 	mutex_lock(&dirty_i->seglist_lock);
929 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
930 		if (get_valid_blocks(sbi, segno, false))
931 			continue;
932 		if (get_ckpt_valid_blocks(sbi, segno, false))
933 			continue;
934 		mutex_unlock(&dirty_i->seglist_lock);
935 		return segno;
936 	}
937 	mutex_unlock(&dirty_i->seglist_lock);
938 	return NULL_SEGNO;
939 }
940 
941 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
942 		struct block_device *bdev, block_t lstart,
943 		block_t start, block_t len)
944 {
945 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
946 	struct list_head *pend_list;
947 	struct discard_cmd *dc;
948 
949 	f2fs_bug_on(sbi, !len);
950 
951 	pend_list = &dcc->pend_list[plist_idx(len)];
952 
953 	dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL);
954 	INIT_LIST_HEAD(&dc->list);
955 	dc->bdev = bdev;
956 	dc->di.lstart = lstart;
957 	dc->di.start = start;
958 	dc->di.len = len;
959 	dc->ref = 0;
960 	dc->state = D_PREP;
961 	dc->queued = 0;
962 	dc->error = 0;
963 	init_completion(&dc->wait);
964 	list_add_tail(&dc->list, pend_list);
965 	spin_lock_init(&dc->lock);
966 	dc->bio_ref = 0;
967 	atomic_inc(&dcc->discard_cmd_cnt);
968 	dcc->undiscard_blks += len;
969 
970 	return dc;
971 }
972 
973 static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi)
974 {
975 #ifdef CONFIG_F2FS_CHECK_FS
976 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
977 	struct rb_node *cur = rb_first_cached(&dcc->root), *next;
978 	struct discard_cmd *cur_dc, *next_dc;
979 
980 	while (cur) {
981 		next = rb_next(cur);
982 		if (!next)
983 			return true;
984 
985 		cur_dc = rb_entry(cur, struct discard_cmd, rb_node);
986 		next_dc = rb_entry(next, struct discard_cmd, rb_node);
987 
988 		if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) {
989 			f2fs_info(sbi, "broken discard_rbtree, "
990 				"cur(%u, %u) next(%u, %u)",
991 				cur_dc->di.lstart, cur_dc->di.len,
992 				next_dc->di.lstart, next_dc->di.len);
993 			return false;
994 		}
995 		cur = next;
996 	}
997 #endif
998 	return true;
999 }
1000 
1001 static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi,
1002 						block_t blkaddr)
1003 {
1004 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1005 	struct rb_node *node = dcc->root.rb_root.rb_node;
1006 	struct discard_cmd *dc;
1007 
1008 	while (node) {
1009 		dc = rb_entry(node, struct discard_cmd, rb_node);
1010 
1011 		if (blkaddr < dc->di.lstart)
1012 			node = node->rb_left;
1013 		else if (blkaddr >= dc->di.lstart + dc->di.len)
1014 			node = node->rb_right;
1015 		else
1016 			return dc;
1017 	}
1018 	return NULL;
1019 }
1020 
1021 static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root,
1022 				block_t blkaddr,
1023 				struct discard_cmd **prev_entry,
1024 				struct discard_cmd **next_entry,
1025 				struct rb_node ***insert_p,
1026 				struct rb_node **insert_parent)
1027 {
1028 	struct rb_node **pnode = &root->rb_root.rb_node;
1029 	struct rb_node *parent = NULL, *tmp_node;
1030 	struct discard_cmd *dc;
1031 
1032 	*insert_p = NULL;
1033 	*insert_parent = NULL;
1034 	*prev_entry = NULL;
1035 	*next_entry = NULL;
1036 
1037 	if (RB_EMPTY_ROOT(&root->rb_root))
1038 		return NULL;
1039 
1040 	while (*pnode) {
1041 		parent = *pnode;
1042 		dc = rb_entry(*pnode, struct discard_cmd, rb_node);
1043 
1044 		if (blkaddr < dc->di.lstart)
1045 			pnode = &(*pnode)->rb_left;
1046 		else if (blkaddr >= dc->di.lstart + dc->di.len)
1047 			pnode = &(*pnode)->rb_right;
1048 		else
1049 			goto lookup_neighbors;
1050 	}
1051 
1052 	*insert_p = pnode;
1053 	*insert_parent = parent;
1054 
1055 	dc = rb_entry(parent, struct discard_cmd, rb_node);
1056 	tmp_node = parent;
1057 	if (parent && blkaddr > dc->di.lstart)
1058 		tmp_node = rb_next(parent);
1059 	*next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1060 
1061 	tmp_node = parent;
1062 	if (parent && blkaddr < dc->di.lstart)
1063 		tmp_node = rb_prev(parent);
1064 	*prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1065 	return NULL;
1066 
1067 lookup_neighbors:
1068 	/* lookup prev node for merging backward later */
1069 	tmp_node = rb_prev(&dc->rb_node);
1070 	*prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1071 
1072 	/* lookup next node for merging frontward later */
1073 	tmp_node = rb_next(&dc->rb_node);
1074 	*next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1075 	return dc;
1076 }
1077 
1078 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1079 							struct discard_cmd *dc)
1080 {
1081 	if (dc->state == D_DONE)
1082 		atomic_sub(dc->queued, &dcc->queued_discard);
1083 
1084 	list_del(&dc->list);
1085 	rb_erase_cached(&dc->rb_node, &dcc->root);
1086 	dcc->undiscard_blks -= dc->di.len;
1087 
1088 	kmem_cache_free(discard_cmd_slab, dc);
1089 
1090 	atomic_dec(&dcc->discard_cmd_cnt);
1091 }
1092 
1093 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1094 							struct discard_cmd *dc)
1095 {
1096 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1097 	unsigned long flags;
1098 
1099 	trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len);
1100 
1101 	spin_lock_irqsave(&dc->lock, flags);
1102 	if (dc->bio_ref) {
1103 		spin_unlock_irqrestore(&dc->lock, flags);
1104 		return;
1105 	}
1106 	spin_unlock_irqrestore(&dc->lock, flags);
1107 
1108 	f2fs_bug_on(sbi, dc->ref);
1109 
1110 	if (dc->error == -EOPNOTSUPP)
1111 		dc->error = 0;
1112 
1113 	if (dc->error)
1114 		f2fs_info_ratelimited(sbi,
1115 			"Issue discard(%u, %u, %u) failed, ret: %d",
1116 			dc->di.lstart, dc->di.start, dc->di.len, dc->error);
1117 	__detach_discard_cmd(dcc, dc);
1118 }
1119 
1120 static void f2fs_submit_discard_endio(struct bio *bio)
1121 {
1122 	struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1123 	unsigned long flags;
1124 
1125 	spin_lock_irqsave(&dc->lock, flags);
1126 	if (!dc->error)
1127 		dc->error = blk_status_to_errno(bio->bi_status);
1128 	dc->bio_ref--;
1129 	if (!dc->bio_ref && dc->state == D_SUBMIT) {
1130 		dc->state = D_DONE;
1131 		complete_all(&dc->wait);
1132 	}
1133 	spin_unlock_irqrestore(&dc->lock, flags);
1134 	bio_put(bio);
1135 }
1136 
1137 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1138 				block_t start, block_t end)
1139 {
1140 #ifdef CONFIG_F2FS_CHECK_FS
1141 	struct seg_entry *sentry;
1142 	unsigned int segno;
1143 	block_t blk = start;
1144 	unsigned long offset, size, *map;
1145 
1146 	while (blk < end) {
1147 		segno = GET_SEGNO(sbi, blk);
1148 		sentry = get_seg_entry(sbi, segno);
1149 		offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1150 
1151 		if (end < START_BLOCK(sbi, segno + 1))
1152 			size = GET_BLKOFF_FROM_SEG0(sbi, end);
1153 		else
1154 			size = BLKS_PER_SEG(sbi);
1155 		map = (unsigned long *)(sentry->cur_valid_map);
1156 		offset = __find_rev_next_bit(map, size, offset);
1157 		f2fs_bug_on(sbi, offset != size);
1158 		blk = START_BLOCK(sbi, segno + 1);
1159 	}
1160 #endif
1161 }
1162 
1163 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1164 				struct discard_policy *dpolicy,
1165 				int discard_type, unsigned int granularity)
1166 {
1167 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1168 
1169 	/* common policy */
1170 	dpolicy->type = discard_type;
1171 	dpolicy->sync = true;
1172 	dpolicy->ordered = false;
1173 	dpolicy->granularity = granularity;
1174 
1175 	dpolicy->max_requests = dcc->max_discard_request;
1176 	dpolicy->io_aware_gran = dcc->discard_io_aware_gran;
1177 	dpolicy->timeout = false;
1178 
1179 	if (discard_type == DPOLICY_BG) {
1180 		dpolicy->min_interval = dcc->min_discard_issue_time;
1181 		dpolicy->mid_interval = dcc->mid_discard_issue_time;
1182 		dpolicy->max_interval = dcc->max_discard_issue_time;
1183 		if (dcc->discard_io_aware == DPOLICY_IO_AWARE_ENABLE)
1184 			dpolicy->io_aware = true;
1185 		else if (dcc->discard_io_aware == DPOLICY_IO_AWARE_DISABLE)
1186 			dpolicy->io_aware = false;
1187 		dpolicy->sync = false;
1188 		dpolicy->ordered = true;
1189 		if (utilization(sbi) > dcc->discard_urgent_util) {
1190 			dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1191 			if (atomic_read(&dcc->discard_cmd_cnt))
1192 				dpolicy->max_interval =
1193 					dcc->min_discard_issue_time;
1194 		}
1195 	} else if (discard_type == DPOLICY_FORCE) {
1196 		dpolicy->min_interval = dcc->min_discard_issue_time;
1197 		dpolicy->mid_interval = dcc->mid_discard_issue_time;
1198 		dpolicy->max_interval = dcc->max_discard_issue_time;
1199 		dpolicy->io_aware = false;
1200 	} else if (discard_type == DPOLICY_FSTRIM) {
1201 		dpolicy->io_aware = false;
1202 	} else if (discard_type == DPOLICY_UMOUNT) {
1203 		dpolicy->io_aware = false;
1204 		/* we need to issue all to keep CP_TRIMMED_FLAG */
1205 		dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1206 		dpolicy->timeout = true;
1207 	}
1208 }
1209 
1210 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1211 				struct block_device *bdev, block_t lstart,
1212 				block_t start, block_t len);
1213 
1214 #ifdef CONFIG_BLK_DEV_ZONED
1215 static void __submit_zone_reset_cmd(struct f2fs_sb_info *sbi,
1216 				   struct discard_cmd *dc, blk_opf_t flag,
1217 				   struct list_head *wait_list,
1218 				   unsigned int *issued)
1219 {
1220 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1221 	struct block_device *bdev = dc->bdev;
1222 	struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS);
1223 	unsigned long flags;
1224 
1225 	trace_f2fs_issue_reset_zone(bdev, dc->di.start);
1226 
1227 	spin_lock_irqsave(&dc->lock, flags);
1228 	dc->state = D_SUBMIT;
1229 	dc->bio_ref++;
1230 	spin_unlock_irqrestore(&dc->lock, flags);
1231 
1232 	if (issued)
1233 		(*issued)++;
1234 
1235 	atomic_inc(&dcc->queued_discard);
1236 	dc->queued++;
1237 	list_move_tail(&dc->list, wait_list);
1238 
1239 	/* sanity check on discard range */
1240 	__check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len);
1241 
1242 	bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start);
1243 	bio->bi_private = dc;
1244 	bio->bi_end_io = f2fs_submit_discard_endio;
1245 	submit_bio(bio);
1246 
1247 	atomic_inc(&dcc->issued_discard);
1248 	f2fs_update_iostat(sbi, NULL, FS_ZONE_RESET_IO, dc->di.len * F2FS_BLKSIZE);
1249 }
1250 #endif
1251 
1252 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1253 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1254 				struct discard_policy *dpolicy,
1255 				struct discard_cmd *dc, int *issued)
1256 {
1257 	struct block_device *bdev = dc->bdev;
1258 	unsigned int max_discard_blocks =
1259 			SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1260 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1261 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1262 					&(dcc->fstrim_list) : &(dcc->wait_list);
1263 	blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0;
1264 	block_t lstart, start, len, total_len;
1265 	int err = 0;
1266 
1267 	if (dc->state != D_PREP)
1268 		return 0;
1269 
1270 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1271 		return 0;
1272 
1273 #ifdef CONFIG_BLK_DEV_ZONED
1274 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) {
1275 		int devi = f2fs_bdev_index(sbi, bdev);
1276 
1277 		if (devi < 0)
1278 			return -EINVAL;
1279 
1280 		if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1281 			__submit_zone_reset_cmd(sbi, dc, flag,
1282 						wait_list, issued);
1283 			return 0;
1284 		}
1285 	}
1286 #endif
1287 
1288 	trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len);
1289 
1290 	lstart = dc->di.lstart;
1291 	start = dc->di.start;
1292 	len = dc->di.len;
1293 	total_len = len;
1294 
1295 	dc->di.len = 0;
1296 
1297 	while (total_len && *issued < dpolicy->max_requests && !err) {
1298 		struct bio *bio = NULL;
1299 		unsigned long flags;
1300 		bool last = true;
1301 
1302 		if (len > max_discard_blocks) {
1303 			len = max_discard_blocks;
1304 			last = false;
1305 		}
1306 
1307 		(*issued)++;
1308 		if (*issued == dpolicy->max_requests)
1309 			last = true;
1310 
1311 		dc->di.len += len;
1312 
1313 		if (time_to_inject(sbi, FAULT_DISCARD)) {
1314 			err = -EIO;
1315 		} else {
1316 			err = __blkdev_issue_discard(bdev,
1317 					SECTOR_FROM_BLOCK(start),
1318 					SECTOR_FROM_BLOCK(len),
1319 					GFP_NOFS, &bio);
1320 		}
1321 		if (err) {
1322 			spin_lock_irqsave(&dc->lock, flags);
1323 			if (dc->state == D_PARTIAL)
1324 				dc->state = D_SUBMIT;
1325 			spin_unlock_irqrestore(&dc->lock, flags);
1326 
1327 			break;
1328 		}
1329 
1330 		f2fs_bug_on(sbi, !bio);
1331 
1332 		/*
1333 		 * should keep before submission to avoid D_DONE
1334 		 * right away
1335 		 */
1336 		spin_lock_irqsave(&dc->lock, flags);
1337 		if (last)
1338 			dc->state = D_SUBMIT;
1339 		else
1340 			dc->state = D_PARTIAL;
1341 		dc->bio_ref++;
1342 		spin_unlock_irqrestore(&dc->lock, flags);
1343 
1344 		atomic_inc(&dcc->queued_discard);
1345 		dc->queued++;
1346 		list_move_tail(&dc->list, wait_list);
1347 
1348 		/* sanity check on discard range */
1349 		__check_sit_bitmap(sbi, lstart, lstart + len);
1350 
1351 		bio->bi_private = dc;
1352 		bio->bi_end_io = f2fs_submit_discard_endio;
1353 		bio->bi_opf |= flag;
1354 		submit_bio(bio);
1355 
1356 		atomic_inc(&dcc->issued_discard);
1357 
1358 		f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE);
1359 
1360 		lstart += len;
1361 		start += len;
1362 		total_len -= len;
1363 		len = total_len;
1364 	}
1365 
1366 	if (!err && len) {
1367 		dcc->undiscard_blks -= len;
1368 		__update_discard_tree_range(sbi, bdev, lstart, start, len);
1369 	}
1370 	return err;
1371 }
1372 
1373 static void __insert_discard_cmd(struct f2fs_sb_info *sbi,
1374 				struct block_device *bdev, block_t lstart,
1375 				block_t start, block_t len)
1376 {
1377 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1378 	struct rb_node **p = &dcc->root.rb_root.rb_node;
1379 	struct rb_node *parent = NULL;
1380 	struct discard_cmd *dc;
1381 	bool leftmost = true;
1382 
1383 	/* look up rb tree to find parent node */
1384 	while (*p) {
1385 		parent = *p;
1386 		dc = rb_entry(parent, struct discard_cmd, rb_node);
1387 
1388 		if (lstart < dc->di.lstart) {
1389 			p = &(*p)->rb_left;
1390 		} else if (lstart >= dc->di.lstart + dc->di.len) {
1391 			p = &(*p)->rb_right;
1392 			leftmost = false;
1393 		} else {
1394 			/* Let's skip to add, if exists */
1395 			return;
1396 		}
1397 	}
1398 
1399 	dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1400 
1401 	rb_link_node(&dc->rb_node, parent, p);
1402 	rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1403 }
1404 
1405 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1406 						struct discard_cmd *dc)
1407 {
1408 	list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]);
1409 }
1410 
1411 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1412 				struct discard_cmd *dc, block_t blkaddr)
1413 {
1414 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1415 	struct discard_info di = dc->di;
1416 	bool modified = false;
1417 
1418 	if (dc->state == D_DONE || dc->di.len == 1) {
1419 		__remove_discard_cmd(sbi, dc);
1420 		return;
1421 	}
1422 
1423 	dcc->undiscard_blks -= di.len;
1424 
1425 	if (blkaddr > di.lstart) {
1426 		dc->di.len = blkaddr - dc->di.lstart;
1427 		dcc->undiscard_blks += dc->di.len;
1428 		__relocate_discard_cmd(dcc, dc);
1429 		modified = true;
1430 	}
1431 
1432 	if (blkaddr < di.lstart + di.len - 1) {
1433 		if (modified) {
1434 			__insert_discard_cmd(sbi, dc->bdev, blkaddr + 1,
1435 					di.start + blkaddr + 1 - di.lstart,
1436 					di.lstart + di.len - 1 - blkaddr);
1437 		} else {
1438 			dc->di.lstart++;
1439 			dc->di.len--;
1440 			dc->di.start++;
1441 			dcc->undiscard_blks += dc->di.len;
1442 			__relocate_discard_cmd(dcc, dc);
1443 		}
1444 	}
1445 }
1446 
1447 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1448 				struct block_device *bdev, block_t lstart,
1449 				block_t start, block_t len)
1450 {
1451 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1452 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1453 	struct discard_cmd *dc;
1454 	struct discard_info di = {0};
1455 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1456 	unsigned int max_discard_blocks =
1457 			SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1458 	block_t end = lstart + len;
1459 
1460 	dc = __lookup_discard_cmd_ret(&dcc->root, lstart,
1461 				&prev_dc, &next_dc, &insert_p, &insert_parent);
1462 	if (dc)
1463 		prev_dc = dc;
1464 
1465 	if (!prev_dc) {
1466 		di.lstart = lstart;
1467 		di.len = next_dc ? next_dc->di.lstart - lstart : len;
1468 		di.len = min(di.len, len);
1469 		di.start = start;
1470 	}
1471 
1472 	while (1) {
1473 		struct rb_node *node;
1474 		bool merged = false;
1475 		struct discard_cmd *tdc = NULL;
1476 
1477 		if (prev_dc) {
1478 			di.lstart = prev_dc->di.lstart + prev_dc->di.len;
1479 			if (di.lstart < lstart)
1480 				di.lstart = lstart;
1481 			if (di.lstart >= end)
1482 				break;
1483 
1484 			if (!next_dc || next_dc->di.lstart > end)
1485 				di.len = end - di.lstart;
1486 			else
1487 				di.len = next_dc->di.lstart - di.lstart;
1488 			di.start = start + di.lstart - lstart;
1489 		}
1490 
1491 		if (!di.len)
1492 			goto next;
1493 
1494 		if (prev_dc && prev_dc->state == D_PREP &&
1495 			prev_dc->bdev == bdev &&
1496 			__is_discard_back_mergeable(&di, &prev_dc->di,
1497 							max_discard_blocks)) {
1498 			prev_dc->di.len += di.len;
1499 			dcc->undiscard_blks += di.len;
1500 			__relocate_discard_cmd(dcc, prev_dc);
1501 			di = prev_dc->di;
1502 			tdc = prev_dc;
1503 			merged = true;
1504 		}
1505 
1506 		if (next_dc && next_dc->state == D_PREP &&
1507 			next_dc->bdev == bdev &&
1508 			__is_discard_front_mergeable(&di, &next_dc->di,
1509 							max_discard_blocks)) {
1510 			next_dc->di.lstart = di.lstart;
1511 			next_dc->di.len += di.len;
1512 			next_dc->di.start = di.start;
1513 			dcc->undiscard_blks += di.len;
1514 			__relocate_discard_cmd(dcc, next_dc);
1515 			if (tdc)
1516 				__remove_discard_cmd(sbi, tdc);
1517 			merged = true;
1518 		}
1519 
1520 		if (!merged)
1521 			__insert_discard_cmd(sbi, bdev,
1522 						di.lstart, di.start, di.len);
1523  next:
1524 		prev_dc = next_dc;
1525 		if (!prev_dc)
1526 			break;
1527 
1528 		node = rb_next(&prev_dc->rb_node);
1529 		next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1530 	}
1531 }
1532 
1533 #ifdef CONFIG_BLK_DEV_ZONED
1534 static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi,
1535 		struct block_device *bdev, block_t blkstart, block_t lblkstart,
1536 		block_t blklen)
1537 {
1538 	trace_f2fs_queue_reset_zone(bdev, blkstart);
1539 
1540 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1541 	__insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen);
1542 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1543 }
1544 #endif
1545 
1546 static void __queue_discard_cmd(struct f2fs_sb_info *sbi,
1547 		struct block_device *bdev, block_t blkstart, block_t blklen)
1548 {
1549 	block_t lblkstart = blkstart;
1550 
1551 	if (!f2fs_bdev_support_discard(bdev))
1552 		return;
1553 
1554 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
1555 
1556 	if (f2fs_is_multi_device(sbi)) {
1557 		int devi = f2fs_target_device_index(sbi, blkstart);
1558 
1559 		blkstart -= FDEV(devi).start_blk;
1560 	}
1561 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1562 	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1563 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1564 }
1565 
1566 static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1567 		struct discard_policy *dpolicy, int *issued)
1568 {
1569 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1570 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1571 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1572 	struct discard_cmd *dc;
1573 	struct blk_plug plug;
1574 	bool io_interrupted = false;
1575 
1576 	mutex_lock(&dcc->cmd_lock);
1577 	dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos,
1578 				&prev_dc, &next_dc, &insert_p, &insert_parent);
1579 	if (!dc)
1580 		dc = next_dc;
1581 
1582 	blk_start_plug(&plug);
1583 
1584 	while (dc) {
1585 		struct rb_node *node;
1586 		int err = 0;
1587 
1588 		if (dc->state != D_PREP)
1589 			goto next;
1590 
1591 		if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1592 			io_interrupted = true;
1593 			break;
1594 		}
1595 
1596 		dcc->next_pos = dc->di.lstart + dc->di.len;
1597 		err = __submit_discard_cmd(sbi, dpolicy, dc, issued);
1598 
1599 		if (*issued >= dpolicy->max_requests)
1600 			break;
1601 next:
1602 		node = rb_next(&dc->rb_node);
1603 		if (err)
1604 			__remove_discard_cmd(sbi, dc);
1605 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1606 	}
1607 
1608 	blk_finish_plug(&plug);
1609 
1610 	if (!dc)
1611 		dcc->next_pos = 0;
1612 
1613 	mutex_unlock(&dcc->cmd_lock);
1614 
1615 	if (!(*issued) && io_interrupted)
1616 		*issued = -1;
1617 }
1618 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1619 					struct discard_policy *dpolicy);
1620 
1621 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1622 					struct discard_policy *dpolicy)
1623 {
1624 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1625 	struct list_head *pend_list;
1626 	struct discard_cmd *dc, *tmp;
1627 	struct blk_plug plug;
1628 	int i, issued;
1629 	bool io_interrupted = false;
1630 
1631 	if (dpolicy->timeout)
1632 		f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1633 
1634 retry:
1635 	issued = 0;
1636 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1637 		if (dpolicy->timeout &&
1638 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1639 			break;
1640 
1641 		if (i + 1 < dpolicy->granularity)
1642 			break;
1643 
1644 		if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) {
1645 			__issue_discard_cmd_orderly(sbi, dpolicy, &issued);
1646 			return issued;
1647 		}
1648 
1649 		pend_list = &dcc->pend_list[i];
1650 
1651 		mutex_lock(&dcc->cmd_lock);
1652 		if (list_empty(pend_list))
1653 			goto next;
1654 		if (unlikely(dcc->rbtree_check))
1655 			f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
1656 		blk_start_plug(&plug);
1657 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1658 			f2fs_bug_on(sbi, dc->state != D_PREP);
1659 
1660 			if (dpolicy->timeout &&
1661 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1662 				break;
1663 
1664 			if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1665 						!is_idle(sbi, DISCARD_TIME)) {
1666 				io_interrupted = true;
1667 				break;
1668 			}
1669 
1670 			__submit_discard_cmd(sbi, dpolicy, dc, &issued);
1671 
1672 			if (issued >= dpolicy->max_requests)
1673 				break;
1674 		}
1675 		blk_finish_plug(&plug);
1676 next:
1677 		mutex_unlock(&dcc->cmd_lock);
1678 
1679 		if (issued >= dpolicy->max_requests || io_interrupted)
1680 			break;
1681 	}
1682 
1683 	if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1684 		__wait_all_discard_cmd(sbi, dpolicy);
1685 		goto retry;
1686 	}
1687 
1688 	if (!issued && io_interrupted)
1689 		issued = -1;
1690 
1691 	return issued;
1692 }
1693 
1694 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1695 {
1696 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1697 	struct list_head *pend_list;
1698 	struct discard_cmd *dc, *tmp;
1699 	int i;
1700 	bool dropped = false;
1701 
1702 	mutex_lock(&dcc->cmd_lock);
1703 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1704 		pend_list = &dcc->pend_list[i];
1705 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1706 			f2fs_bug_on(sbi, dc->state != D_PREP);
1707 			__remove_discard_cmd(sbi, dc);
1708 			dropped = true;
1709 		}
1710 	}
1711 	mutex_unlock(&dcc->cmd_lock);
1712 
1713 	return dropped;
1714 }
1715 
1716 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1717 {
1718 	__drop_discard_cmd(sbi);
1719 }
1720 
1721 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1722 							struct discard_cmd *dc)
1723 {
1724 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1725 	unsigned int len = 0;
1726 
1727 	wait_for_completion_io(&dc->wait);
1728 	mutex_lock(&dcc->cmd_lock);
1729 	f2fs_bug_on(sbi, dc->state != D_DONE);
1730 	dc->ref--;
1731 	if (!dc->ref) {
1732 		if (!dc->error)
1733 			len = dc->di.len;
1734 		__remove_discard_cmd(sbi, dc);
1735 	}
1736 	mutex_unlock(&dcc->cmd_lock);
1737 
1738 	return len;
1739 }
1740 
1741 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1742 						struct discard_policy *dpolicy,
1743 						block_t start, block_t end)
1744 {
1745 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1746 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1747 					&(dcc->fstrim_list) : &(dcc->wait_list);
1748 	struct discard_cmd *dc = NULL, *iter, *tmp;
1749 	unsigned int trimmed = 0;
1750 
1751 next:
1752 	dc = NULL;
1753 
1754 	mutex_lock(&dcc->cmd_lock);
1755 	list_for_each_entry_safe(iter, tmp, wait_list, list) {
1756 		if (iter->di.lstart + iter->di.len <= start ||
1757 					end <= iter->di.lstart)
1758 			continue;
1759 		if (iter->di.len < dpolicy->granularity)
1760 			continue;
1761 		if (iter->state == D_DONE && !iter->ref) {
1762 			wait_for_completion_io(&iter->wait);
1763 			if (!iter->error)
1764 				trimmed += iter->di.len;
1765 			__remove_discard_cmd(sbi, iter);
1766 		} else {
1767 			iter->ref++;
1768 			dc = iter;
1769 			break;
1770 		}
1771 	}
1772 	mutex_unlock(&dcc->cmd_lock);
1773 
1774 	if (dc) {
1775 		trimmed += __wait_one_discard_bio(sbi, dc);
1776 		goto next;
1777 	}
1778 
1779 	return trimmed;
1780 }
1781 
1782 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1783 						struct discard_policy *dpolicy)
1784 {
1785 	struct discard_policy dp;
1786 	unsigned int discard_blks;
1787 
1788 	if (dpolicy)
1789 		return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1790 
1791 	/* wait all */
1792 	__init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY);
1793 	discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1794 	__init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY);
1795 	discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1796 
1797 	return discard_blks;
1798 }
1799 
1800 /* This should be covered by global mutex, &sit_i->sentry_lock */
1801 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1802 {
1803 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1804 	struct discard_cmd *dc;
1805 	bool need_wait = false;
1806 
1807 	mutex_lock(&dcc->cmd_lock);
1808 	dc = __lookup_discard_cmd(sbi, blkaddr);
1809 #ifdef CONFIG_BLK_DEV_ZONED
1810 	if (dc && f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) {
1811 		int devi = f2fs_bdev_index(sbi, dc->bdev);
1812 
1813 		if (devi < 0) {
1814 			mutex_unlock(&dcc->cmd_lock);
1815 			return;
1816 		}
1817 
1818 		if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1819 			/* force submit zone reset */
1820 			if (dc->state == D_PREP)
1821 				__submit_zone_reset_cmd(sbi, dc, REQ_SYNC,
1822 							&dcc->wait_list, NULL);
1823 			dc->ref++;
1824 			mutex_unlock(&dcc->cmd_lock);
1825 			/* wait zone reset */
1826 			__wait_one_discard_bio(sbi, dc);
1827 			return;
1828 		}
1829 	}
1830 #endif
1831 	if (dc) {
1832 		if (dc->state == D_PREP) {
1833 			__punch_discard_cmd(sbi, dc, blkaddr);
1834 		} else {
1835 			dc->ref++;
1836 			need_wait = true;
1837 		}
1838 	}
1839 	mutex_unlock(&dcc->cmd_lock);
1840 
1841 	if (need_wait)
1842 		__wait_one_discard_bio(sbi, dc);
1843 }
1844 
1845 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1846 {
1847 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1848 
1849 	if (dcc && dcc->f2fs_issue_discard) {
1850 		struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1851 
1852 		dcc->f2fs_issue_discard = NULL;
1853 		kthread_stop(discard_thread);
1854 	}
1855 }
1856 
1857 /**
1858  * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT
1859  * @sbi: the f2fs_sb_info data for discard cmd to issue
1860  *
1861  * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped
1862  *
1863  * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false.
1864  */
1865 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1866 {
1867 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1868 	struct discard_policy dpolicy;
1869 	bool dropped;
1870 
1871 	if (!atomic_read(&dcc->discard_cmd_cnt))
1872 		return true;
1873 
1874 	__init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1875 					dcc->discard_granularity);
1876 	__issue_discard_cmd(sbi, &dpolicy);
1877 	dropped = __drop_discard_cmd(sbi);
1878 
1879 	/* just to make sure there is no pending discard commands */
1880 	__wait_all_discard_cmd(sbi, NULL);
1881 
1882 	f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1883 	return !dropped;
1884 }
1885 
1886 static int issue_discard_thread(void *data)
1887 {
1888 	struct f2fs_sb_info *sbi = data;
1889 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1890 	wait_queue_head_t *q = &dcc->discard_wait_queue;
1891 	struct discard_policy dpolicy;
1892 	unsigned int wait_ms = dcc->min_discard_issue_time;
1893 	int issued;
1894 
1895 	set_freezable();
1896 
1897 	do {
1898 		wait_event_freezable_timeout(*q,
1899 				kthread_should_stop() || dcc->discard_wake,
1900 				msecs_to_jiffies(wait_ms));
1901 
1902 		if (sbi->gc_mode == GC_URGENT_HIGH ||
1903 			!f2fs_available_free_memory(sbi, DISCARD_CACHE))
1904 			__init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE,
1905 						MIN_DISCARD_GRANULARITY);
1906 		else
1907 			__init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1908 						dcc->discard_granularity);
1909 
1910 		if (dcc->discard_wake)
1911 			dcc->discard_wake = false;
1912 
1913 		/* clean up pending candidates before going to sleep */
1914 		if (atomic_read(&dcc->queued_discard))
1915 			__wait_all_discard_cmd(sbi, NULL);
1916 
1917 		if (f2fs_readonly(sbi->sb))
1918 			continue;
1919 		if (kthread_should_stop())
1920 			return 0;
1921 		if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) ||
1922 			!atomic_read(&dcc->discard_cmd_cnt)) {
1923 			wait_ms = dpolicy.max_interval;
1924 			continue;
1925 		}
1926 
1927 		sb_start_intwrite(sbi->sb);
1928 
1929 		issued = __issue_discard_cmd(sbi, &dpolicy);
1930 		if (issued > 0) {
1931 			__wait_all_discard_cmd(sbi, &dpolicy);
1932 			wait_ms = dpolicy.min_interval;
1933 		} else if (issued == -1) {
1934 			wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1935 			if (!wait_ms)
1936 				wait_ms = dpolicy.mid_interval;
1937 		} else {
1938 			wait_ms = dpolicy.max_interval;
1939 		}
1940 		if (!atomic_read(&dcc->discard_cmd_cnt))
1941 			wait_ms = dpolicy.max_interval;
1942 
1943 		sb_end_intwrite(sbi->sb);
1944 
1945 	} while (!kthread_should_stop());
1946 	return 0;
1947 }
1948 
1949 #ifdef CONFIG_BLK_DEV_ZONED
1950 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1951 		struct block_device *bdev, block_t blkstart, block_t blklen)
1952 {
1953 	sector_t sector, nr_sects;
1954 	block_t lblkstart = blkstart;
1955 	int devi = 0;
1956 	u64 remainder = 0;
1957 
1958 	if (f2fs_is_multi_device(sbi)) {
1959 		devi = f2fs_target_device_index(sbi, blkstart);
1960 		if (blkstart < FDEV(devi).start_blk ||
1961 		    blkstart > FDEV(devi).end_blk) {
1962 			f2fs_err(sbi, "Invalid block %x", blkstart);
1963 			return -EIO;
1964 		}
1965 		blkstart -= FDEV(devi).start_blk;
1966 	}
1967 
1968 	/* For sequential zones, reset the zone write pointer */
1969 	if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1970 		sector = SECTOR_FROM_BLOCK(blkstart);
1971 		nr_sects = SECTOR_FROM_BLOCK(blklen);
1972 		div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder);
1973 
1974 		if (remainder || nr_sects != bdev_zone_sectors(bdev)) {
1975 			f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1976 				 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1977 				 blkstart, blklen);
1978 			return -EIO;
1979 		}
1980 
1981 		if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) {
1982 			unsigned int nofs_flags;
1983 			int ret;
1984 
1985 			trace_f2fs_issue_reset_zone(bdev, blkstart);
1986 			nofs_flags = memalloc_nofs_save();
1987 			ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1988 						sector, nr_sects);
1989 			memalloc_nofs_restore(nofs_flags);
1990 			return ret;
1991 		}
1992 
1993 		__queue_zone_reset_cmd(sbi, bdev, blkstart, lblkstart, blklen);
1994 		return 0;
1995 	}
1996 
1997 	/* For conventional zones, use regular discard if supported */
1998 	__queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1999 	return 0;
2000 }
2001 #endif
2002 
2003 static int __issue_discard_async(struct f2fs_sb_info *sbi,
2004 		struct block_device *bdev, block_t blkstart, block_t blklen)
2005 {
2006 #ifdef CONFIG_BLK_DEV_ZONED
2007 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
2008 		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
2009 #endif
2010 	__queue_discard_cmd(sbi, bdev, blkstart, blklen);
2011 	return 0;
2012 }
2013 
2014 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
2015 				block_t blkstart, block_t blklen)
2016 {
2017 	sector_t start = blkstart, len = 0;
2018 	struct block_device *bdev;
2019 	struct seg_entry *se;
2020 	unsigned int offset;
2021 	block_t i;
2022 	int err = 0;
2023 
2024 	bdev = f2fs_target_device(sbi, blkstart, NULL);
2025 
2026 	for (i = blkstart; i < blkstart + blklen; i++, len++) {
2027 		if (i != start) {
2028 			struct block_device *bdev2 =
2029 				f2fs_target_device(sbi, i, NULL);
2030 
2031 			if (bdev2 != bdev) {
2032 				err = __issue_discard_async(sbi, bdev,
2033 						start, len);
2034 				if (err)
2035 					return err;
2036 				bdev = bdev2;
2037 				start = i;
2038 				len = 0;
2039 			}
2040 		}
2041 
2042 		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
2043 		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
2044 
2045 		if (f2fs_block_unit_discard(sbi) &&
2046 				!f2fs_test_and_set_bit(offset, se->discard_map))
2047 			sbi->discard_blks--;
2048 	}
2049 
2050 	if (len)
2051 		err = __issue_discard_async(sbi, bdev, start, len);
2052 	return err;
2053 }
2054 
2055 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
2056 							bool check_only)
2057 {
2058 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2059 	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
2060 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2061 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2062 	unsigned long *discard_map = (unsigned long *)se->discard_map;
2063 	unsigned long *dmap = SIT_I(sbi)->tmp_map;
2064 	unsigned int start = 0, end = -1;
2065 	bool force = (cpc->reason & CP_DISCARD);
2066 	struct discard_entry *de = NULL;
2067 	struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
2068 	int i;
2069 
2070 	if (se->valid_blocks == BLKS_PER_SEG(sbi) ||
2071 	    !f2fs_hw_support_discard(sbi) ||
2072 	    !f2fs_block_unit_discard(sbi))
2073 		return false;
2074 
2075 	if (!force) {
2076 		if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
2077 			SM_I(sbi)->dcc_info->nr_discards >=
2078 				SM_I(sbi)->dcc_info->max_discards)
2079 			return false;
2080 	}
2081 
2082 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
2083 	for (i = 0; i < entries; i++)
2084 		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
2085 				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
2086 
2087 	while (force || SM_I(sbi)->dcc_info->nr_discards <=
2088 				SM_I(sbi)->dcc_info->max_discards) {
2089 		start = __find_rev_next_bit(dmap, BLKS_PER_SEG(sbi), end + 1);
2090 		if (start >= BLKS_PER_SEG(sbi))
2091 			break;
2092 
2093 		end = __find_rev_next_zero_bit(dmap,
2094 						BLKS_PER_SEG(sbi), start + 1);
2095 		if (force && start && end != BLKS_PER_SEG(sbi) &&
2096 		    (end - start) < cpc->trim_minlen)
2097 			continue;
2098 
2099 		if (check_only)
2100 			return true;
2101 
2102 		if (!de) {
2103 			de = f2fs_kmem_cache_alloc(discard_entry_slab,
2104 						GFP_F2FS_ZERO, true, NULL);
2105 			de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
2106 			list_add_tail(&de->list, head);
2107 		}
2108 
2109 		for (i = start; i < end; i++)
2110 			__set_bit_le(i, (void *)de->discard_map);
2111 
2112 		SM_I(sbi)->dcc_info->nr_discards += end - start;
2113 	}
2114 	return false;
2115 }
2116 
2117 static void release_discard_addr(struct discard_entry *entry)
2118 {
2119 	list_del(&entry->list);
2120 	kmem_cache_free(discard_entry_slab, entry);
2121 }
2122 
2123 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
2124 {
2125 	struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
2126 	struct discard_entry *entry, *this;
2127 
2128 	/* drop caches */
2129 	list_for_each_entry_safe(entry, this, head, list)
2130 		release_discard_addr(entry);
2131 }
2132 
2133 /*
2134  * Should call f2fs_clear_prefree_segments after checkpoint is done.
2135  */
2136 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
2137 {
2138 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2139 	unsigned int segno;
2140 
2141 	mutex_lock(&dirty_i->seglist_lock);
2142 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
2143 		__set_test_and_free(sbi, segno, false);
2144 	mutex_unlock(&dirty_i->seglist_lock);
2145 }
2146 
2147 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
2148 						struct cp_control *cpc)
2149 {
2150 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2151 	struct list_head *head = &dcc->entry_list;
2152 	struct discard_entry *entry, *this;
2153 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2154 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
2155 	unsigned int start = 0, end = -1;
2156 	unsigned int secno, start_segno;
2157 	bool force = (cpc->reason & CP_DISCARD);
2158 	bool section_alignment = F2FS_OPTION(sbi).discard_unit ==
2159 						DISCARD_UNIT_SECTION;
2160 
2161 	if (f2fs_lfs_mode(sbi) && __is_large_section(sbi))
2162 		section_alignment = true;
2163 
2164 	mutex_lock(&dirty_i->seglist_lock);
2165 
2166 	while (1) {
2167 		int i;
2168 
2169 		if (section_alignment && end != -1)
2170 			end--;
2171 		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
2172 		if (start >= MAIN_SEGS(sbi))
2173 			break;
2174 		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
2175 								start + 1);
2176 
2177 		if (section_alignment) {
2178 			start = rounddown(start, SEGS_PER_SEC(sbi));
2179 			end = roundup(end, SEGS_PER_SEC(sbi));
2180 		}
2181 
2182 		for (i = start; i < end; i++) {
2183 			if (test_and_clear_bit(i, prefree_map))
2184 				dirty_i->nr_dirty[PRE]--;
2185 		}
2186 
2187 		if (!f2fs_realtime_discard_enable(sbi))
2188 			continue;
2189 
2190 		if (force && start >= cpc->trim_start &&
2191 					(end - 1) <= cpc->trim_end)
2192 			continue;
2193 
2194 		/* Should cover 2MB zoned device for zone-based reset */
2195 		if (!f2fs_sb_has_blkzoned(sbi) &&
2196 		    (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) {
2197 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2198 				SEGS_TO_BLKS(sbi, end - start));
2199 			continue;
2200 		}
2201 next:
2202 		secno = GET_SEC_FROM_SEG(sbi, start);
2203 		start_segno = GET_SEG_FROM_SEC(sbi, secno);
2204 		if (!IS_CURSEC(sbi, secno) &&
2205 			!get_valid_blocks(sbi, start, true))
2206 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2207 						BLKS_PER_SEC(sbi));
2208 
2209 		start = start_segno + SEGS_PER_SEC(sbi);
2210 		if (start < end)
2211 			goto next;
2212 		else
2213 			end = start - 1;
2214 	}
2215 	mutex_unlock(&dirty_i->seglist_lock);
2216 
2217 	if (!f2fs_block_unit_discard(sbi))
2218 		goto wakeup;
2219 
2220 	/* send small discards */
2221 	list_for_each_entry_safe(entry, this, head, list) {
2222 		unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2223 		bool is_valid = test_bit_le(0, entry->discard_map);
2224 
2225 find_next:
2226 		if (is_valid) {
2227 			next_pos = find_next_zero_bit_le(entry->discard_map,
2228 						BLKS_PER_SEG(sbi), cur_pos);
2229 			len = next_pos - cur_pos;
2230 
2231 			if (f2fs_sb_has_blkzoned(sbi) ||
2232 			    (force && len < cpc->trim_minlen))
2233 				goto skip;
2234 
2235 			f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2236 									len);
2237 			total_len += len;
2238 		} else {
2239 			next_pos = find_next_bit_le(entry->discard_map,
2240 						BLKS_PER_SEG(sbi), cur_pos);
2241 		}
2242 skip:
2243 		cur_pos = next_pos;
2244 		is_valid = !is_valid;
2245 
2246 		if (cur_pos < BLKS_PER_SEG(sbi))
2247 			goto find_next;
2248 
2249 		release_discard_addr(entry);
2250 		dcc->nr_discards -= total_len;
2251 	}
2252 
2253 wakeup:
2254 	wake_up_discard_thread(sbi, false);
2255 }
2256 
2257 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi)
2258 {
2259 	dev_t dev = sbi->sb->s_bdev->bd_dev;
2260 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2261 	int err = 0;
2262 
2263 	if (f2fs_sb_has_readonly(sbi)) {
2264 		f2fs_info(sbi,
2265 			"Skip to start discard thread for readonly image");
2266 		return 0;
2267 	}
2268 
2269 	if (!f2fs_realtime_discard_enable(sbi))
2270 		return 0;
2271 
2272 	dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2273 				"f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2274 	if (IS_ERR(dcc->f2fs_issue_discard)) {
2275 		err = PTR_ERR(dcc->f2fs_issue_discard);
2276 		dcc->f2fs_issue_discard = NULL;
2277 	}
2278 
2279 	return err;
2280 }
2281 
2282 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2283 {
2284 	struct discard_cmd_control *dcc;
2285 	int err = 0, i;
2286 
2287 	if (SM_I(sbi)->dcc_info) {
2288 		dcc = SM_I(sbi)->dcc_info;
2289 		goto init_thread;
2290 	}
2291 
2292 	dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2293 	if (!dcc)
2294 		return -ENOMEM;
2295 
2296 	dcc->discard_io_aware_gran = MAX_PLIST_NUM;
2297 	dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2298 	dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY;
2299 	dcc->discard_io_aware = DPOLICY_IO_AWARE_ENABLE;
2300 	if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT)
2301 		dcc->discard_granularity = BLKS_PER_SEG(sbi);
2302 	else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION)
2303 		dcc->discard_granularity = BLKS_PER_SEC(sbi);
2304 
2305 	INIT_LIST_HEAD(&dcc->entry_list);
2306 	for (i = 0; i < MAX_PLIST_NUM; i++)
2307 		INIT_LIST_HEAD(&dcc->pend_list[i]);
2308 	INIT_LIST_HEAD(&dcc->wait_list);
2309 	INIT_LIST_HEAD(&dcc->fstrim_list);
2310 	mutex_init(&dcc->cmd_lock);
2311 	atomic_set(&dcc->issued_discard, 0);
2312 	atomic_set(&dcc->queued_discard, 0);
2313 	atomic_set(&dcc->discard_cmd_cnt, 0);
2314 	dcc->nr_discards = 0;
2315 	dcc->max_discards = SEGS_TO_BLKS(sbi, MAIN_SEGS(sbi));
2316 	dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST;
2317 	dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME;
2318 	dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME;
2319 	dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME;
2320 	dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL;
2321 	dcc->undiscard_blks = 0;
2322 	dcc->next_pos = 0;
2323 	dcc->root = RB_ROOT_CACHED;
2324 	dcc->rbtree_check = false;
2325 
2326 	init_waitqueue_head(&dcc->discard_wait_queue);
2327 	SM_I(sbi)->dcc_info = dcc;
2328 init_thread:
2329 	err = f2fs_start_discard_thread(sbi);
2330 	if (err) {
2331 		kfree(dcc);
2332 		SM_I(sbi)->dcc_info = NULL;
2333 	}
2334 
2335 	return err;
2336 }
2337 
2338 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2339 {
2340 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2341 
2342 	if (!dcc)
2343 		return;
2344 
2345 	f2fs_stop_discard_thread(sbi);
2346 
2347 	/*
2348 	 * Recovery can cache discard commands, so in error path of
2349 	 * fill_super(), it needs to give a chance to handle them.
2350 	 */
2351 	f2fs_issue_discard_timeout(sbi);
2352 
2353 	kfree(dcc);
2354 	SM_I(sbi)->dcc_info = NULL;
2355 }
2356 
2357 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2358 {
2359 	struct sit_info *sit_i = SIT_I(sbi);
2360 
2361 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2362 		sit_i->dirty_sentries++;
2363 		return false;
2364 	}
2365 
2366 	return true;
2367 }
2368 
2369 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2370 					unsigned int segno, int modified)
2371 {
2372 	struct seg_entry *se = get_seg_entry(sbi, segno);
2373 
2374 	se->type = type;
2375 	if (modified)
2376 		__mark_sit_entry_dirty(sbi, segno);
2377 }
2378 
2379 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2380 								block_t blkaddr)
2381 {
2382 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
2383 
2384 	if (segno == NULL_SEGNO)
2385 		return 0;
2386 	return get_seg_entry(sbi, segno)->mtime;
2387 }
2388 
2389 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2390 						unsigned long long old_mtime)
2391 {
2392 	struct seg_entry *se;
2393 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
2394 	unsigned long long ctime = get_mtime(sbi, false);
2395 	unsigned long long mtime = old_mtime ? old_mtime : ctime;
2396 
2397 	if (segno == NULL_SEGNO)
2398 		return;
2399 
2400 	se = get_seg_entry(sbi, segno);
2401 
2402 	if (!se->mtime)
2403 		se->mtime = mtime;
2404 	else
2405 		se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2406 						se->valid_blocks + 1);
2407 
2408 	if (ctime > SIT_I(sbi)->max_mtime)
2409 		SIT_I(sbi)->max_mtime = ctime;
2410 }
2411 
2412 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2413 {
2414 	struct seg_entry *se;
2415 	unsigned int segno, offset;
2416 	long int new_vblocks;
2417 	bool exist;
2418 #ifdef CONFIG_F2FS_CHECK_FS
2419 	bool mir_exist;
2420 #endif
2421 
2422 	segno = GET_SEGNO(sbi, blkaddr);
2423 	if (segno == NULL_SEGNO)
2424 		return;
2425 
2426 	se = get_seg_entry(sbi, segno);
2427 	new_vblocks = se->valid_blocks + del;
2428 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2429 
2430 	f2fs_bug_on(sbi, (new_vblocks < 0 ||
2431 			(new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2432 
2433 	se->valid_blocks = new_vblocks;
2434 
2435 	/* Update valid block bitmap */
2436 	if (del > 0) {
2437 		exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2438 #ifdef CONFIG_F2FS_CHECK_FS
2439 		mir_exist = f2fs_test_and_set_bit(offset,
2440 						se->cur_valid_map_mir);
2441 		if (unlikely(exist != mir_exist)) {
2442 			f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2443 				 blkaddr, exist);
2444 			f2fs_bug_on(sbi, 1);
2445 		}
2446 #endif
2447 		if (unlikely(exist)) {
2448 			f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2449 				 blkaddr);
2450 			f2fs_bug_on(sbi, 1);
2451 			se->valid_blocks--;
2452 			del = 0;
2453 		}
2454 
2455 		if (f2fs_block_unit_discard(sbi) &&
2456 				!f2fs_test_and_set_bit(offset, se->discard_map))
2457 			sbi->discard_blks--;
2458 
2459 		/*
2460 		 * SSR should never reuse block which is checkpointed
2461 		 * or newly invalidated.
2462 		 */
2463 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2464 			if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2465 				se->ckpt_valid_blocks++;
2466 		}
2467 	} else {
2468 		exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2469 #ifdef CONFIG_F2FS_CHECK_FS
2470 		mir_exist = f2fs_test_and_clear_bit(offset,
2471 						se->cur_valid_map_mir);
2472 		if (unlikely(exist != mir_exist)) {
2473 			f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2474 				 blkaddr, exist);
2475 			f2fs_bug_on(sbi, 1);
2476 		}
2477 #endif
2478 		if (unlikely(!exist)) {
2479 			f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2480 				 blkaddr);
2481 			f2fs_bug_on(sbi, 1);
2482 			se->valid_blocks++;
2483 			del = 0;
2484 		} else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2485 			/*
2486 			 * If checkpoints are off, we must not reuse data that
2487 			 * was used in the previous checkpoint. If it was used
2488 			 * before, we must track that to know how much space we
2489 			 * really have.
2490 			 */
2491 			if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2492 				spin_lock(&sbi->stat_lock);
2493 				sbi->unusable_block_count++;
2494 				spin_unlock(&sbi->stat_lock);
2495 			}
2496 		}
2497 
2498 		if (f2fs_block_unit_discard(sbi) &&
2499 			f2fs_test_and_clear_bit(offset, se->discard_map))
2500 			sbi->discard_blks++;
2501 	}
2502 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2503 		se->ckpt_valid_blocks += del;
2504 
2505 	__mark_sit_entry_dirty(sbi, segno);
2506 
2507 	/* update total number of valid blocks to be written in ckpt area */
2508 	SIT_I(sbi)->written_valid_blocks += del;
2509 
2510 	if (__is_large_section(sbi))
2511 		get_sec_entry(sbi, segno)->valid_blocks += del;
2512 }
2513 
2514 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2515 {
2516 	unsigned int segno = GET_SEGNO(sbi, addr);
2517 	struct sit_info *sit_i = SIT_I(sbi);
2518 
2519 	f2fs_bug_on(sbi, addr == NULL_ADDR);
2520 	if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2521 		return;
2522 
2523 	f2fs_invalidate_internal_cache(sbi, addr);
2524 
2525 	/* add it into sit main buffer */
2526 	down_write(&sit_i->sentry_lock);
2527 
2528 	update_segment_mtime(sbi, addr, 0);
2529 	update_sit_entry(sbi, addr, -1);
2530 
2531 	/* add it into dirty seglist */
2532 	locate_dirty_segment(sbi, segno);
2533 
2534 	up_write(&sit_i->sentry_lock);
2535 }
2536 
2537 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2538 {
2539 	struct sit_info *sit_i = SIT_I(sbi);
2540 	unsigned int segno, offset;
2541 	struct seg_entry *se;
2542 	bool is_cp = false;
2543 
2544 	if (!__is_valid_data_blkaddr(blkaddr))
2545 		return true;
2546 
2547 	down_read(&sit_i->sentry_lock);
2548 
2549 	segno = GET_SEGNO(sbi, blkaddr);
2550 	se = get_seg_entry(sbi, segno);
2551 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2552 
2553 	if (f2fs_test_bit(offset, se->ckpt_valid_map))
2554 		is_cp = true;
2555 
2556 	up_read(&sit_i->sentry_lock);
2557 
2558 	return is_cp;
2559 }
2560 
2561 static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type)
2562 {
2563 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2564 
2565 	if (sbi->ckpt->alloc_type[type] == SSR)
2566 		return BLKS_PER_SEG(sbi);
2567 	return curseg->next_blkoff;
2568 }
2569 
2570 /*
2571  * Calculate the number of current summary pages for writing
2572  */
2573 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2574 {
2575 	int valid_sum_count = 0;
2576 	int i, sum_in_page;
2577 
2578 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2579 		if (sbi->ckpt->alloc_type[i] != SSR && for_ra)
2580 			valid_sum_count +=
2581 				le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2582 		else
2583 			valid_sum_count += f2fs_curseg_valid_blocks(sbi, i);
2584 	}
2585 
2586 	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2587 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2588 	if (valid_sum_count <= sum_in_page)
2589 		return 1;
2590 	else if ((valid_sum_count - sum_in_page) <=
2591 		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2592 		return 2;
2593 	return 3;
2594 }
2595 
2596 /*
2597  * Caller should put this summary page
2598  */
2599 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2600 {
2601 	if (unlikely(f2fs_cp_error(sbi)))
2602 		return ERR_PTR(-EIO);
2603 	return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2604 }
2605 
2606 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2607 					void *src, block_t blk_addr)
2608 {
2609 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2610 
2611 	memcpy(page_address(page), src, PAGE_SIZE);
2612 	set_page_dirty(page);
2613 	f2fs_put_page(page, 1);
2614 }
2615 
2616 static void write_sum_page(struct f2fs_sb_info *sbi,
2617 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
2618 {
2619 	f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2620 }
2621 
2622 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2623 						int type, block_t blk_addr)
2624 {
2625 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2626 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2627 	struct f2fs_summary_block *src = curseg->sum_blk;
2628 	struct f2fs_summary_block *dst;
2629 
2630 	dst = (struct f2fs_summary_block *)page_address(page);
2631 	memset(dst, 0, PAGE_SIZE);
2632 
2633 	mutex_lock(&curseg->curseg_mutex);
2634 
2635 	down_read(&curseg->journal_rwsem);
2636 	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2637 	up_read(&curseg->journal_rwsem);
2638 
2639 	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2640 	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2641 
2642 	mutex_unlock(&curseg->curseg_mutex);
2643 
2644 	set_page_dirty(page);
2645 	f2fs_put_page(page, 1);
2646 }
2647 
2648 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2649 				struct curseg_info *curseg)
2650 {
2651 	unsigned int segno = curseg->segno + 1;
2652 	struct free_segmap_info *free_i = FREE_I(sbi);
2653 
2654 	if (segno < MAIN_SEGS(sbi) && segno % SEGS_PER_SEC(sbi))
2655 		return !test_bit(segno, free_i->free_segmap);
2656 	return 0;
2657 }
2658 
2659 /*
2660  * Find a new segment from the free segments bitmap to right order
2661  * This function should be returned with success, otherwise BUG
2662  */
2663 static int get_new_segment(struct f2fs_sb_info *sbi,
2664 			unsigned int *newseg, bool new_sec, bool pinning)
2665 {
2666 	struct free_segmap_info *free_i = FREE_I(sbi);
2667 	unsigned int segno, secno, zoneno;
2668 	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2669 	unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2670 	unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2671 	bool init = true;
2672 	int i;
2673 	int ret = 0;
2674 
2675 	spin_lock(&free_i->segmap_lock);
2676 
2677 	if (time_to_inject(sbi, FAULT_NO_SEGMENT)) {
2678 		ret = -ENOSPC;
2679 		goto out_unlock;
2680 	}
2681 
2682 	if (!new_sec && ((*newseg + 1) % SEGS_PER_SEC(sbi))) {
2683 		segno = find_next_zero_bit(free_i->free_segmap,
2684 			GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2685 		if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2686 			goto got_it;
2687 	}
2688 
2689 	/*
2690 	 * If we format f2fs on zoned storage, let's try to get pinned sections
2691 	 * from beginning of the storage, which should be a conventional one.
2692 	 */
2693 	if (f2fs_sb_has_blkzoned(sbi)) {
2694 		segno = pinning ? 0 : max(first_zoned_segno(sbi), *newseg);
2695 		hint = GET_SEC_FROM_SEG(sbi, segno);
2696 	}
2697 
2698 find_other_zone:
2699 	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2700 	if (secno >= MAIN_SECS(sbi)) {
2701 		secno = find_first_zero_bit(free_i->free_secmap,
2702 							MAIN_SECS(sbi));
2703 		if (secno >= MAIN_SECS(sbi)) {
2704 			ret = -ENOSPC;
2705 			goto out_unlock;
2706 		}
2707 	}
2708 	segno = GET_SEG_FROM_SEC(sbi, secno);
2709 	zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2710 
2711 	/* give up on finding another zone */
2712 	if (!init)
2713 		goto got_it;
2714 	if (sbi->secs_per_zone == 1)
2715 		goto got_it;
2716 	if (zoneno == old_zoneno)
2717 		goto got_it;
2718 	for (i = 0; i < NR_CURSEG_TYPE; i++)
2719 		if (CURSEG_I(sbi, i)->zone == zoneno)
2720 			break;
2721 
2722 	if (i < NR_CURSEG_TYPE) {
2723 		/* zone is in user, try another */
2724 		if (zoneno + 1 >= total_zones)
2725 			hint = 0;
2726 		else
2727 			hint = (zoneno + 1) * sbi->secs_per_zone;
2728 		init = false;
2729 		goto find_other_zone;
2730 	}
2731 got_it:
2732 	/* set it as dirty segment in free segmap */
2733 	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2734 
2735 	/* no free section in conventional zone */
2736 	if (new_sec && pinning &&
2737 		!f2fs_valid_pinned_area(sbi, START_BLOCK(sbi, segno))) {
2738 		ret = -EAGAIN;
2739 		goto out_unlock;
2740 	}
2741 	__set_inuse(sbi, segno);
2742 	*newseg = segno;
2743 out_unlock:
2744 	spin_unlock(&free_i->segmap_lock);
2745 
2746 	if (ret == -ENOSPC) {
2747 		f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_NO_SEGMENT);
2748 		f2fs_bug_on(sbi, 1);
2749 	}
2750 	return ret;
2751 }
2752 
2753 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2754 {
2755 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2756 	struct summary_footer *sum_footer;
2757 	unsigned short seg_type = curseg->seg_type;
2758 
2759 	/* only happen when get_new_segment() fails */
2760 	if (curseg->next_segno == NULL_SEGNO)
2761 		return;
2762 
2763 	curseg->inited = true;
2764 	curseg->segno = curseg->next_segno;
2765 	curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2766 	curseg->next_blkoff = 0;
2767 	curseg->next_segno = NULL_SEGNO;
2768 
2769 	sum_footer = &(curseg->sum_blk->footer);
2770 	memset(sum_footer, 0, sizeof(struct summary_footer));
2771 
2772 	sanity_check_seg_type(sbi, seg_type);
2773 
2774 	if (IS_DATASEG(seg_type))
2775 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2776 	if (IS_NODESEG(seg_type))
2777 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2778 	__set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2779 }
2780 
2781 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2782 {
2783 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2784 	unsigned short seg_type = curseg->seg_type;
2785 
2786 	sanity_check_seg_type(sbi, seg_type);
2787 	if (__is_large_section(sbi)) {
2788 		if (f2fs_need_rand_seg(sbi)) {
2789 			unsigned int hint = GET_SEC_FROM_SEG(sbi, curseg->segno);
2790 
2791 			if (GET_SEC_FROM_SEG(sbi, curseg->segno + 1) != hint)
2792 				return curseg->segno;
2793 			return get_random_u32_inclusive(curseg->segno + 1,
2794 					GET_SEG_FROM_SEC(sbi, hint + 1) - 1);
2795 		}
2796 		return curseg->segno;
2797 	} else if (f2fs_need_rand_seg(sbi)) {
2798 		return get_random_u32_below(MAIN_SECS(sbi) * SEGS_PER_SEC(sbi));
2799 	}
2800 
2801 	/* inmem log may not locate on any segment after mount */
2802 	if (!curseg->inited)
2803 		return 0;
2804 
2805 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2806 		return 0;
2807 
2808 	if (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type))
2809 		return 0;
2810 
2811 	if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2812 		return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2813 
2814 	/* find segments from 0 to reuse freed segments */
2815 	if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2816 		return 0;
2817 
2818 	return curseg->segno;
2819 }
2820 
2821 /*
2822  * Allocate a current working segment.
2823  * This function always allocates a free segment in LFS manner.
2824  */
2825 static int new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2826 {
2827 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2828 	unsigned int segno = curseg->segno;
2829 	bool pinning = type == CURSEG_COLD_DATA_PINNED;
2830 	int ret;
2831 
2832 	if (curseg->inited)
2833 		write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, segno));
2834 
2835 	segno = __get_next_segno(sbi, type);
2836 	ret = get_new_segment(sbi, &segno, new_sec, pinning);
2837 	if (ret) {
2838 		if (ret == -ENOSPC)
2839 			curseg->segno = NULL_SEGNO;
2840 		return ret;
2841 	}
2842 
2843 	curseg->next_segno = segno;
2844 	reset_curseg(sbi, type, 1);
2845 	curseg->alloc_type = LFS;
2846 	if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
2847 		curseg->fragment_remained_chunk =
2848 				get_random_u32_inclusive(1, sbi->max_fragment_chunk);
2849 	return 0;
2850 }
2851 
2852 static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2853 					int segno, block_t start)
2854 {
2855 	struct seg_entry *se = get_seg_entry(sbi, segno);
2856 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2857 	unsigned long *target_map = SIT_I(sbi)->tmp_map;
2858 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2859 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2860 	int i;
2861 
2862 	for (i = 0; i < entries; i++)
2863 		target_map[i] = ckpt_map[i] | cur_map[i];
2864 
2865 	return __find_rev_next_zero_bit(target_map, BLKS_PER_SEG(sbi), start);
2866 }
2867 
2868 static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi,
2869 		struct curseg_info *seg)
2870 {
2871 	return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1);
2872 }
2873 
2874 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2875 {
2876 	return __next_free_blkoff(sbi, segno, 0) < BLKS_PER_SEG(sbi);
2877 }
2878 
2879 /*
2880  * This function always allocates a used segment(from dirty seglist) by SSR
2881  * manner, so it should recover the existing segment information of valid blocks
2882  */
2883 static int change_curseg(struct f2fs_sb_info *sbi, int type)
2884 {
2885 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2886 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2887 	unsigned int new_segno = curseg->next_segno;
2888 	struct f2fs_summary_block *sum_node;
2889 	struct page *sum_page;
2890 
2891 	write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, curseg->segno));
2892 
2893 	__set_test_and_inuse(sbi, new_segno);
2894 
2895 	mutex_lock(&dirty_i->seglist_lock);
2896 	__remove_dirty_segment(sbi, new_segno, PRE);
2897 	__remove_dirty_segment(sbi, new_segno, DIRTY);
2898 	mutex_unlock(&dirty_i->seglist_lock);
2899 
2900 	reset_curseg(sbi, type, 1);
2901 	curseg->alloc_type = SSR;
2902 	curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
2903 
2904 	sum_page = f2fs_get_sum_page(sbi, new_segno);
2905 	if (IS_ERR(sum_page)) {
2906 		/* GC won't be able to use stale summary pages by cp_error */
2907 		memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2908 		return PTR_ERR(sum_page);
2909 	}
2910 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2911 	memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2912 	f2fs_put_page(sum_page, 1);
2913 	return 0;
2914 }
2915 
2916 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2917 				int alloc_mode, unsigned long long age);
2918 
2919 static int get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2920 					int target_type, int alloc_mode,
2921 					unsigned long long age)
2922 {
2923 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2924 	int ret = 0;
2925 
2926 	curseg->seg_type = target_type;
2927 
2928 	if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2929 		struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2930 
2931 		curseg->seg_type = se->type;
2932 		ret = change_curseg(sbi, type);
2933 	} else {
2934 		/* allocate cold segment by default */
2935 		curseg->seg_type = CURSEG_COLD_DATA;
2936 		ret = new_curseg(sbi, type, true);
2937 	}
2938 	stat_inc_seg_type(sbi, curseg);
2939 	return ret;
2940 }
2941 
2942 static int __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi, bool force)
2943 {
2944 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2945 	int ret = 0;
2946 
2947 	if (!sbi->am.atgc_enabled && !force)
2948 		return 0;
2949 
2950 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
2951 
2952 	mutex_lock(&curseg->curseg_mutex);
2953 	down_write(&SIT_I(sbi)->sentry_lock);
2954 
2955 	ret = get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC,
2956 					CURSEG_COLD_DATA, SSR, 0);
2957 
2958 	up_write(&SIT_I(sbi)->sentry_lock);
2959 	mutex_unlock(&curseg->curseg_mutex);
2960 
2961 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
2962 	return ret;
2963 }
2964 
2965 int f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2966 {
2967 	return __f2fs_init_atgc_curseg(sbi, false);
2968 }
2969 
2970 int f2fs_reinit_atgc_curseg(struct f2fs_sb_info *sbi)
2971 {
2972 	int ret;
2973 
2974 	if (!test_opt(sbi, ATGC))
2975 		return 0;
2976 	if (sbi->am.atgc_enabled)
2977 		return 0;
2978 	if (le64_to_cpu(F2FS_CKPT(sbi)->elapsed_time) <
2979 			sbi->am.age_threshold)
2980 		return 0;
2981 
2982 	ret = __f2fs_init_atgc_curseg(sbi, true);
2983 	if (!ret) {
2984 		sbi->am.atgc_enabled = true;
2985 		f2fs_info(sbi, "reenabled age threshold GC");
2986 	}
2987 	return ret;
2988 }
2989 
2990 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2991 {
2992 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2993 
2994 	mutex_lock(&curseg->curseg_mutex);
2995 	if (!curseg->inited)
2996 		goto out;
2997 
2998 	if (get_valid_blocks(sbi, curseg->segno, false)) {
2999 		write_sum_page(sbi, curseg->sum_blk,
3000 				GET_SUM_BLOCK(sbi, curseg->segno));
3001 	} else {
3002 		mutex_lock(&DIRTY_I(sbi)->seglist_lock);
3003 		__set_test_and_free(sbi, curseg->segno, true);
3004 		mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
3005 	}
3006 out:
3007 	mutex_unlock(&curseg->curseg_mutex);
3008 }
3009 
3010 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
3011 {
3012 	__f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
3013 
3014 	if (sbi->am.atgc_enabled)
3015 		__f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
3016 }
3017 
3018 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
3019 {
3020 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3021 
3022 	mutex_lock(&curseg->curseg_mutex);
3023 	if (!curseg->inited)
3024 		goto out;
3025 	if (get_valid_blocks(sbi, curseg->segno, false))
3026 		goto out;
3027 
3028 	mutex_lock(&DIRTY_I(sbi)->seglist_lock);
3029 	__set_test_and_inuse(sbi, curseg->segno);
3030 	mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
3031 out:
3032 	mutex_unlock(&curseg->curseg_mutex);
3033 }
3034 
3035 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
3036 {
3037 	__f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
3038 
3039 	if (sbi->am.atgc_enabled)
3040 		__f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
3041 }
3042 
3043 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
3044 				int alloc_mode, unsigned long long age)
3045 {
3046 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3047 	unsigned segno = NULL_SEGNO;
3048 	unsigned short seg_type = curseg->seg_type;
3049 	int i, cnt;
3050 	bool reversed = false;
3051 
3052 	sanity_check_seg_type(sbi, seg_type);
3053 
3054 	/* f2fs_need_SSR() already forces to do this */
3055 	if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
3056 		curseg->next_segno = segno;
3057 		return 1;
3058 	}
3059 
3060 	/* For node segments, let's do SSR more intensively */
3061 	if (IS_NODESEG(seg_type)) {
3062 		if (seg_type >= CURSEG_WARM_NODE) {
3063 			reversed = true;
3064 			i = CURSEG_COLD_NODE;
3065 		} else {
3066 			i = CURSEG_HOT_NODE;
3067 		}
3068 		cnt = NR_CURSEG_NODE_TYPE;
3069 	} else {
3070 		if (seg_type >= CURSEG_WARM_DATA) {
3071 			reversed = true;
3072 			i = CURSEG_COLD_DATA;
3073 		} else {
3074 			i = CURSEG_HOT_DATA;
3075 		}
3076 		cnt = NR_CURSEG_DATA_TYPE;
3077 	}
3078 
3079 	for (; cnt-- > 0; reversed ? i-- : i++) {
3080 		if (i == seg_type)
3081 			continue;
3082 		if (!f2fs_get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
3083 			curseg->next_segno = segno;
3084 			return 1;
3085 		}
3086 	}
3087 
3088 	/* find valid_blocks=0 in dirty list */
3089 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
3090 		segno = get_free_segment(sbi);
3091 		if (segno != NULL_SEGNO) {
3092 			curseg->next_segno = segno;
3093 			return 1;
3094 		}
3095 	}
3096 	return 0;
3097 }
3098 
3099 static bool need_new_seg(struct f2fs_sb_info *sbi, int type)
3100 {
3101 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3102 
3103 	if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
3104 	    curseg->seg_type == CURSEG_WARM_NODE)
3105 		return true;
3106 	if (curseg->alloc_type == LFS && is_next_segment_free(sbi, curseg) &&
3107 	    likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
3108 		return true;
3109 	if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0))
3110 		return true;
3111 	return false;
3112 }
3113 
3114 int f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
3115 					unsigned int start, unsigned int end)
3116 {
3117 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3118 	unsigned int segno;
3119 	int ret = 0;
3120 
3121 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3122 	mutex_lock(&curseg->curseg_mutex);
3123 	down_write(&SIT_I(sbi)->sentry_lock);
3124 
3125 	segno = CURSEG_I(sbi, type)->segno;
3126 	if (segno < start || segno > end)
3127 		goto unlock;
3128 
3129 	if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
3130 		ret = change_curseg(sbi, type);
3131 	else
3132 		ret = new_curseg(sbi, type, true);
3133 
3134 	stat_inc_seg_type(sbi, curseg);
3135 
3136 	locate_dirty_segment(sbi, segno);
3137 unlock:
3138 	up_write(&SIT_I(sbi)->sentry_lock);
3139 
3140 	if (segno != curseg->segno)
3141 		f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
3142 			    type, segno, curseg->segno);
3143 
3144 	mutex_unlock(&curseg->curseg_mutex);
3145 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3146 	return ret;
3147 }
3148 
3149 static int __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
3150 						bool new_sec, bool force)
3151 {
3152 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3153 	unsigned int old_segno;
3154 	int err = 0;
3155 
3156 	if (type == CURSEG_COLD_DATA_PINNED && !curseg->inited)
3157 		goto allocate;
3158 
3159 	if (!force && curseg->inited &&
3160 	    !curseg->next_blkoff &&
3161 	    !get_valid_blocks(sbi, curseg->segno, new_sec) &&
3162 	    !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
3163 		return 0;
3164 
3165 allocate:
3166 	old_segno = curseg->segno;
3167 	err = new_curseg(sbi, type, true);
3168 	if (err)
3169 		return err;
3170 	stat_inc_seg_type(sbi, curseg);
3171 	locate_dirty_segment(sbi, old_segno);
3172 	return 0;
3173 }
3174 
3175 int f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
3176 {
3177 	int ret;
3178 
3179 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3180 	down_write(&SIT_I(sbi)->sentry_lock);
3181 	ret = __allocate_new_segment(sbi, type, true, force);
3182 	up_write(&SIT_I(sbi)->sentry_lock);
3183 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3184 
3185 	return ret;
3186 }
3187 
3188 int f2fs_allocate_pinning_section(struct f2fs_sb_info *sbi)
3189 {
3190 	int err;
3191 	bool gc_required = true;
3192 
3193 retry:
3194 	f2fs_lock_op(sbi);
3195 	err = f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3196 	f2fs_unlock_op(sbi);
3197 
3198 	if (f2fs_sb_has_blkzoned(sbi) && err == -EAGAIN && gc_required) {
3199 		f2fs_down_write(&sbi->gc_lock);
3200 		err = f2fs_gc_range(sbi, 0, GET_SEGNO(sbi, FDEV(0).end_blk), true, 1);
3201 		f2fs_up_write(&sbi->gc_lock);
3202 
3203 		gc_required = false;
3204 		if (!err)
3205 			goto retry;
3206 	}
3207 
3208 	return err;
3209 }
3210 
3211 int f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
3212 {
3213 	int i;
3214 	int err = 0;
3215 
3216 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3217 	down_write(&SIT_I(sbi)->sentry_lock);
3218 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
3219 		err += __allocate_new_segment(sbi, i, false, false);
3220 	up_write(&SIT_I(sbi)->sentry_lock);
3221 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3222 
3223 	return err;
3224 }
3225 
3226 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
3227 						struct cp_control *cpc)
3228 {
3229 	__u64 trim_start = cpc->trim_start;
3230 	bool has_candidate = false;
3231 
3232 	down_write(&SIT_I(sbi)->sentry_lock);
3233 	for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
3234 		if (add_discard_addrs(sbi, cpc, true)) {
3235 			has_candidate = true;
3236 			break;
3237 		}
3238 	}
3239 	up_write(&SIT_I(sbi)->sentry_lock);
3240 
3241 	cpc->trim_start = trim_start;
3242 	return has_candidate;
3243 }
3244 
3245 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
3246 					struct discard_policy *dpolicy,
3247 					unsigned int start, unsigned int end)
3248 {
3249 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
3250 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
3251 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
3252 	struct discard_cmd *dc;
3253 	struct blk_plug plug;
3254 	int issued;
3255 	unsigned int trimmed = 0;
3256 
3257 next:
3258 	issued = 0;
3259 
3260 	mutex_lock(&dcc->cmd_lock);
3261 	if (unlikely(dcc->rbtree_check))
3262 		f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
3263 
3264 	dc = __lookup_discard_cmd_ret(&dcc->root, start,
3265 				&prev_dc, &next_dc, &insert_p, &insert_parent);
3266 	if (!dc)
3267 		dc = next_dc;
3268 
3269 	blk_start_plug(&plug);
3270 
3271 	while (dc && dc->di.lstart <= end) {
3272 		struct rb_node *node;
3273 		int err = 0;
3274 
3275 		if (dc->di.len < dpolicy->granularity)
3276 			goto skip;
3277 
3278 		if (dc->state != D_PREP) {
3279 			list_move_tail(&dc->list, &dcc->fstrim_list);
3280 			goto skip;
3281 		}
3282 
3283 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3284 
3285 		if (issued >= dpolicy->max_requests) {
3286 			start = dc->di.lstart + dc->di.len;
3287 
3288 			if (err)
3289 				__remove_discard_cmd(sbi, dc);
3290 
3291 			blk_finish_plug(&plug);
3292 			mutex_unlock(&dcc->cmd_lock);
3293 			trimmed += __wait_all_discard_cmd(sbi, NULL);
3294 			f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
3295 			goto next;
3296 		}
3297 skip:
3298 		node = rb_next(&dc->rb_node);
3299 		if (err)
3300 			__remove_discard_cmd(sbi, dc);
3301 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3302 
3303 		if (fatal_signal_pending(current))
3304 			break;
3305 	}
3306 
3307 	blk_finish_plug(&plug);
3308 	mutex_unlock(&dcc->cmd_lock);
3309 
3310 	return trimmed;
3311 }
3312 
3313 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3314 {
3315 	__u64 start = F2FS_BYTES_TO_BLK(range->start);
3316 	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3317 	unsigned int start_segno, end_segno;
3318 	block_t start_block, end_block;
3319 	struct cp_control cpc;
3320 	struct discard_policy dpolicy;
3321 	unsigned long long trimmed = 0;
3322 	int err = 0;
3323 	bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3324 
3325 	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3326 		return -EINVAL;
3327 
3328 	if (end < MAIN_BLKADDR(sbi))
3329 		goto out;
3330 
3331 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3332 		f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3333 		return -EFSCORRUPTED;
3334 	}
3335 
3336 	/* start/end segment number in main_area */
3337 	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3338 	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3339 						GET_SEGNO(sbi, end);
3340 	if (need_align) {
3341 		start_segno = rounddown(start_segno, SEGS_PER_SEC(sbi));
3342 		end_segno = roundup(end_segno + 1, SEGS_PER_SEC(sbi)) - 1;
3343 	}
3344 
3345 	cpc.reason = CP_DISCARD;
3346 	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3347 	cpc.trim_start = start_segno;
3348 	cpc.trim_end = end_segno;
3349 
3350 	if (sbi->discard_blks == 0)
3351 		goto out;
3352 
3353 	f2fs_down_write(&sbi->gc_lock);
3354 	stat_inc_cp_call_count(sbi, TOTAL_CALL);
3355 	err = f2fs_write_checkpoint(sbi, &cpc);
3356 	f2fs_up_write(&sbi->gc_lock);
3357 	if (err)
3358 		goto out;
3359 
3360 	/*
3361 	 * We filed discard candidates, but actually we don't need to wait for
3362 	 * all of them, since they'll be issued in idle time along with runtime
3363 	 * discard option. User configuration looks like using runtime discard
3364 	 * or periodic fstrim instead of it.
3365 	 */
3366 	if (f2fs_realtime_discard_enable(sbi))
3367 		goto out;
3368 
3369 	start_block = START_BLOCK(sbi, start_segno);
3370 	end_block = START_BLOCK(sbi, end_segno + 1);
3371 
3372 	__init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3373 	trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3374 					start_block, end_block);
3375 
3376 	trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3377 					start_block, end_block);
3378 out:
3379 	if (!err)
3380 		range->len = F2FS_BLK_TO_BYTES(trimmed);
3381 	return err;
3382 }
3383 
3384 int f2fs_rw_hint_to_seg_type(struct f2fs_sb_info *sbi, enum rw_hint hint)
3385 {
3386 	if (F2FS_OPTION(sbi).active_logs == 2)
3387 		return CURSEG_HOT_DATA;
3388 	else if (F2FS_OPTION(sbi).active_logs == 4)
3389 		return CURSEG_COLD_DATA;
3390 
3391 	/* active_log == 6 */
3392 	switch (hint) {
3393 	case WRITE_LIFE_SHORT:
3394 		return CURSEG_HOT_DATA;
3395 	case WRITE_LIFE_EXTREME:
3396 		return CURSEG_COLD_DATA;
3397 	default:
3398 		return CURSEG_WARM_DATA;
3399 	}
3400 }
3401 
3402 /*
3403  * This returns write hints for each segment type. This hints will be
3404  * passed down to block layer as below by default.
3405  *
3406  * User                  F2FS                     Block
3407  * ----                  ----                     -----
3408  *                       META                     WRITE_LIFE_NONE|REQ_META
3409  *                       HOT_NODE                 WRITE_LIFE_NONE
3410  *                       WARM_NODE                WRITE_LIFE_MEDIUM
3411  *                       COLD_NODE                WRITE_LIFE_LONG
3412  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
3413  * extension list        "                        "
3414  *
3415  * -- buffered io
3416  *                       COLD_DATA                WRITE_LIFE_EXTREME
3417  *                       HOT_DATA                 WRITE_LIFE_SHORT
3418  *                       WARM_DATA                WRITE_LIFE_NOT_SET
3419  *
3420  * -- direct io
3421  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3422  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3423  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3424  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
3425  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
3426  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
3427  */
3428 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
3429 				enum page_type type, enum temp_type temp)
3430 {
3431 	switch (type) {
3432 	case DATA:
3433 		switch (temp) {
3434 		case WARM:
3435 			return WRITE_LIFE_NOT_SET;
3436 		case HOT:
3437 			return WRITE_LIFE_SHORT;
3438 		case COLD:
3439 			return WRITE_LIFE_EXTREME;
3440 		default:
3441 			return WRITE_LIFE_NONE;
3442 		}
3443 	case NODE:
3444 		switch (temp) {
3445 		case WARM:
3446 			return WRITE_LIFE_MEDIUM;
3447 		case HOT:
3448 			return WRITE_LIFE_NONE;
3449 		case COLD:
3450 			return WRITE_LIFE_LONG;
3451 		default:
3452 			return WRITE_LIFE_NONE;
3453 		}
3454 	case META:
3455 		return WRITE_LIFE_NONE;
3456 	default:
3457 		return WRITE_LIFE_NONE;
3458 	}
3459 }
3460 
3461 static int __get_segment_type_2(struct f2fs_io_info *fio)
3462 {
3463 	if (fio->type == DATA)
3464 		return CURSEG_HOT_DATA;
3465 	else
3466 		return CURSEG_HOT_NODE;
3467 }
3468 
3469 static int __get_segment_type_4(struct f2fs_io_info *fio)
3470 {
3471 	if (fio->type == DATA) {
3472 		struct inode *inode = fio->page->mapping->host;
3473 
3474 		if (S_ISDIR(inode->i_mode))
3475 			return CURSEG_HOT_DATA;
3476 		else
3477 			return CURSEG_COLD_DATA;
3478 	} else {
3479 		if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3480 			return CURSEG_WARM_NODE;
3481 		else
3482 			return CURSEG_COLD_NODE;
3483 	}
3484 }
3485 
3486 static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs)
3487 {
3488 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3489 	struct extent_info ei = {};
3490 
3491 	if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) {
3492 		if (!ei.age)
3493 			return NO_CHECK_TYPE;
3494 		if (ei.age <= sbi->hot_data_age_threshold)
3495 			return CURSEG_HOT_DATA;
3496 		if (ei.age <= sbi->warm_data_age_threshold)
3497 			return CURSEG_WARM_DATA;
3498 		return CURSEG_COLD_DATA;
3499 	}
3500 	return NO_CHECK_TYPE;
3501 }
3502 
3503 static int __get_segment_type_6(struct f2fs_io_info *fio)
3504 {
3505 	if (fio->type == DATA) {
3506 		struct inode *inode = fio->page->mapping->host;
3507 		int type;
3508 
3509 		if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
3510 			return CURSEG_COLD_DATA_PINNED;
3511 
3512 		if (page_private_gcing(fio->page)) {
3513 			if (fio->sbi->am.atgc_enabled &&
3514 				(fio->io_type == FS_DATA_IO) &&
3515 				(fio->sbi->gc_mode != GC_URGENT_HIGH) &&
3516 				__is_valid_data_blkaddr(fio->old_blkaddr) &&
3517 				!is_inode_flag_set(inode, FI_OPU_WRITE))
3518 				return CURSEG_ALL_DATA_ATGC;
3519 			else
3520 				return CURSEG_COLD_DATA;
3521 		}
3522 		if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3523 			return CURSEG_COLD_DATA;
3524 
3525 		type = __get_age_segment_type(inode, fio->page->index);
3526 		if (type != NO_CHECK_TYPE)
3527 			return type;
3528 
3529 		if (file_is_hot(inode) ||
3530 				is_inode_flag_set(inode, FI_HOT_DATA) ||
3531 				f2fs_is_cow_file(inode))
3532 			return CURSEG_HOT_DATA;
3533 		return f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode),
3534 						inode->i_write_hint);
3535 	} else {
3536 		if (IS_DNODE(fio->page))
3537 			return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3538 						CURSEG_HOT_NODE;
3539 		return CURSEG_COLD_NODE;
3540 	}
3541 }
3542 
3543 int f2fs_get_segment_temp(int seg_type)
3544 {
3545 	if (IS_HOT(seg_type))
3546 		return HOT;
3547 	else if (IS_WARM(seg_type))
3548 		return WARM;
3549 	return COLD;
3550 }
3551 
3552 static int __get_segment_type(struct f2fs_io_info *fio)
3553 {
3554 	int type = 0;
3555 
3556 	switch (F2FS_OPTION(fio->sbi).active_logs) {
3557 	case 2:
3558 		type = __get_segment_type_2(fio);
3559 		break;
3560 	case 4:
3561 		type = __get_segment_type_4(fio);
3562 		break;
3563 	case 6:
3564 		type = __get_segment_type_6(fio);
3565 		break;
3566 	default:
3567 		f2fs_bug_on(fio->sbi, true);
3568 	}
3569 
3570 	fio->temp = f2fs_get_segment_temp(type);
3571 
3572 	return type;
3573 }
3574 
3575 static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi,
3576 		struct curseg_info *seg)
3577 {
3578 	/* To allocate block chunks in different sizes, use random number */
3579 	if (--seg->fragment_remained_chunk > 0)
3580 		return;
3581 
3582 	seg->fragment_remained_chunk =
3583 		get_random_u32_inclusive(1, sbi->max_fragment_chunk);
3584 	seg->next_blkoff +=
3585 		get_random_u32_inclusive(1, sbi->max_fragment_hole);
3586 }
3587 
3588 static void reset_curseg_fields(struct curseg_info *curseg)
3589 {
3590 	curseg->inited = false;
3591 	curseg->segno = NULL_SEGNO;
3592 	curseg->next_segno = 0;
3593 }
3594 
3595 int f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3596 		block_t old_blkaddr, block_t *new_blkaddr,
3597 		struct f2fs_summary *sum, int type,
3598 		struct f2fs_io_info *fio)
3599 {
3600 	struct sit_info *sit_i = SIT_I(sbi);
3601 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3602 	unsigned long long old_mtime;
3603 	bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3604 	struct seg_entry *se = NULL;
3605 	bool segment_full = false;
3606 	int ret = 0;
3607 
3608 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3609 
3610 	mutex_lock(&curseg->curseg_mutex);
3611 	down_write(&sit_i->sentry_lock);
3612 
3613 	if (curseg->segno == NULL_SEGNO) {
3614 		ret = -ENOSPC;
3615 		goto out_err;
3616 	}
3617 
3618 	if (from_gc) {
3619 		f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3620 		se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3621 		sanity_check_seg_type(sbi, se->type);
3622 		f2fs_bug_on(sbi, IS_NODESEG(se->type));
3623 	}
3624 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3625 
3626 	f2fs_bug_on(sbi, curseg->next_blkoff >= BLKS_PER_SEG(sbi));
3627 
3628 	f2fs_wait_discard_bio(sbi, *new_blkaddr);
3629 
3630 	curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3631 	if (curseg->alloc_type == SSR) {
3632 		curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg);
3633 	} else {
3634 		curseg->next_blkoff++;
3635 		if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
3636 			f2fs_randomize_chunk(sbi, curseg);
3637 	}
3638 	if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno))
3639 		segment_full = true;
3640 	stat_inc_block_count(sbi, curseg);
3641 
3642 	if (from_gc) {
3643 		old_mtime = get_segment_mtime(sbi, old_blkaddr);
3644 	} else {
3645 		update_segment_mtime(sbi, old_blkaddr, 0);
3646 		old_mtime = 0;
3647 	}
3648 	update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3649 
3650 	/*
3651 	 * SIT information should be updated before segment allocation,
3652 	 * since SSR needs latest valid block information.
3653 	 */
3654 	update_sit_entry(sbi, *new_blkaddr, 1);
3655 	update_sit_entry(sbi, old_blkaddr, -1);
3656 
3657 	/*
3658 	 * If the current segment is full, flush it out and replace it with a
3659 	 * new segment.
3660 	 */
3661 	if (segment_full) {
3662 		if (type == CURSEG_COLD_DATA_PINNED &&
3663 		    !((curseg->segno + 1) % sbi->segs_per_sec)) {
3664 			write_sum_page(sbi, curseg->sum_blk,
3665 					GET_SUM_BLOCK(sbi, curseg->segno));
3666 			reset_curseg_fields(curseg);
3667 			goto skip_new_segment;
3668 		}
3669 
3670 		if (from_gc) {
3671 			ret = get_atssr_segment(sbi, type, se->type,
3672 						AT_SSR, se->mtime);
3673 		} else {
3674 			if (need_new_seg(sbi, type))
3675 				ret = new_curseg(sbi, type, false);
3676 			else
3677 				ret = change_curseg(sbi, type);
3678 			stat_inc_seg_type(sbi, curseg);
3679 		}
3680 
3681 		if (ret)
3682 			goto out_err;
3683 	}
3684 
3685 skip_new_segment:
3686 	/*
3687 	 * segment dirty status should be updated after segment allocation,
3688 	 * so we just need to update status only one time after previous
3689 	 * segment being closed.
3690 	 */
3691 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3692 	locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3693 
3694 	if (IS_DATASEG(curseg->seg_type))
3695 		atomic64_inc(&sbi->allocated_data_blocks);
3696 
3697 	up_write(&sit_i->sentry_lock);
3698 
3699 	if (page && IS_NODESEG(curseg->seg_type)) {
3700 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3701 
3702 		f2fs_inode_chksum_set(sbi, page);
3703 	}
3704 
3705 	if (fio) {
3706 		struct f2fs_bio_info *io;
3707 
3708 		INIT_LIST_HEAD(&fio->list);
3709 		fio->in_list = 1;
3710 		io = sbi->write_io[fio->type] + fio->temp;
3711 		spin_lock(&io->io_lock);
3712 		list_add_tail(&fio->list, &io->io_list);
3713 		spin_unlock(&io->io_lock);
3714 	}
3715 
3716 	mutex_unlock(&curseg->curseg_mutex);
3717 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3718 	return 0;
3719 
3720 out_err:
3721 	*new_blkaddr = NULL_ADDR;
3722 	up_write(&sit_i->sentry_lock);
3723 	mutex_unlock(&curseg->curseg_mutex);
3724 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3725 	return ret;
3726 }
3727 
3728 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino,
3729 					block_t blkaddr, unsigned int blkcnt)
3730 {
3731 	if (!f2fs_is_multi_device(sbi))
3732 		return;
3733 
3734 	while (1) {
3735 		unsigned int devidx = f2fs_target_device_index(sbi, blkaddr);
3736 		unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1;
3737 
3738 		/* update device state for fsync */
3739 		f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO);
3740 
3741 		/* update device state for checkpoint */
3742 		if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3743 			spin_lock(&sbi->dev_lock);
3744 			f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3745 			spin_unlock(&sbi->dev_lock);
3746 		}
3747 
3748 		if (blkcnt <= blks)
3749 			break;
3750 		blkcnt -= blks;
3751 		blkaddr += blks;
3752 	}
3753 }
3754 
3755 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3756 {
3757 	int type = __get_segment_type(fio);
3758 	bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3759 
3760 	if (keep_order)
3761 		f2fs_down_read(&fio->sbi->io_order_lock);
3762 
3763 	if (f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3764 			&fio->new_blkaddr, sum, type, fio)) {
3765 		if (fscrypt_inode_uses_fs_layer_crypto(fio->page->mapping->host))
3766 			fscrypt_finalize_bounce_page(&fio->encrypted_page);
3767 		end_page_writeback(fio->page);
3768 		if (f2fs_in_warm_node_list(fio->sbi, fio->page))
3769 			f2fs_del_fsync_node_entry(fio->sbi, fio->page);
3770 		goto out;
3771 	}
3772 	if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3773 		f2fs_invalidate_internal_cache(fio->sbi, fio->old_blkaddr);
3774 
3775 	/* writeout dirty page into bdev */
3776 	f2fs_submit_page_write(fio);
3777 
3778 	f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1);
3779 out:
3780 	if (keep_order)
3781 		f2fs_up_read(&fio->sbi->io_order_lock);
3782 }
3783 
3784 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3785 					enum iostat_type io_type)
3786 {
3787 	struct f2fs_io_info fio = {
3788 		.sbi = sbi,
3789 		.type = META,
3790 		.temp = HOT,
3791 		.op = REQ_OP_WRITE,
3792 		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3793 		.old_blkaddr = page->index,
3794 		.new_blkaddr = page->index,
3795 		.page = page,
3796 		.encrypted_page = NULL,
3797 		.in_list = 0,
3798 	};
3799 
3800 	if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3801 		fio.op_flags &= ~REQ_META;
3802 
3803 	set_page_writeback(page);
3804 	f2fs_submit_page_write(&fio);
3805 
3806 	stat_inc_meta_count(sbi, page->index);
3807 	f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE);
3808 }
3809 
3810 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3811 {
3812 	struct f2fs_summary sum;
3813 
3814 	set_summary(&sum, nid, 0, 0);
3815 	do_write_page(&sum, fio);
3816 
3817 	f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE);
3818 }
3819 
3820 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3821 					struct f2fs_io_info *fio)
3822 {
3823 	struct f2fs_sb_info *sbi = fio->sbi;
3824 	struct f2fs_summary sum;
3825 
3826 	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3827 	if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO)
3828 		f2fs_update_age_extent_cache(dn);
3829 	set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3830 	do_write_page(&sum, fio);
3831 	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3832 
3833 	f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE);
3834 }
3835 
3836 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3837 {
3838 	int err;
3839 	struct f2fs_sb_info *sbi = fio->sbi;
3840 	unsigned int segno;
3841 
3842 	fio->new_blkaddr = fio->old_blkaddr;
3843 	/* i/o temperature is needed for passing down write hints */
3844 	__get_segment_type(fio);
3845 
3846 	segno = GET_SEGNO(sbi, fio->new_blkaddr);
3847 
3848 	if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3849 		set_sbi_flag(sbi, SBI_NEED_FSCK);
3850 		f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3851 			  __func__, segno);
3852 		err = -EFSCORRUPTED;
3853 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
3854 		goto drop_bio;
3855 	}
3856 
3857 	if (f2fs_cp_error(sbi)) {
3858 		err = -EIO;
3859 		goto drop_bio;
3860 	}
3861 
3862 	if (fio->meta_gc)
3863 		f2fs_truncate_meta_inode_pages(sbi, fio->new_blkaddr, 1);
3864 
3865 	stat_inc_inplace_blocks(fio->sbi);
3866 
3867 	if (fio->bio && !IS_F2FS_IPU_NOCACHE(sbi))
3868 		err = f2fs_merge_page_bio(fio);
3869 	else
3870 		err = f2fs_submit_page_bio(fio);
3871 	if (!err) {
3872 		f2fs_update_device_state(fio->sbi, fio->ino,
3873 						fio->new_blkaddr, 1);
3874 		f2fs_update_iostat(fio->sbi, fio->page->mapping->host,
3875 						fio->io_type, F2FS_BLKSIZE);
3876 	}
3877 
3878 	return err;
3879 drop_bio:
3880 	if (fio->bio && *(fio->bio)) {
3881 		struct bio *bio = *(fio->bio);
3882 
3883 		bio->bi_status = BLK_STS_IOERR;
3884 		bio_endio(bio);
3885 		*(fio->bio) = NULL;
3886 	}
3887 	return err;
3888 }
3889 
3890 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3891 						unsigned int segno)
3892 {
3893 	int i;
3894 
3895 	for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3896 		if (CURSEG_I(sbi, i)->segno == segno)
3897 			break;
3898 	}
3899 	return i;
3900 }
3901 
3902 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3903 				block_t old_blkaddr, block_t new_blkaddr,
3904 				bool recover_curseg, bool recover_newaddr,
3905 				bool from_gc)
3906 {
3907 	struct sit_info *sit_i = SIT_I(sbi);
3908 	struct curseg_info *curseg;
3909 	unsigned int segno, old_cursegno;
3910 	struct seg_entry *se;
3911 	int type;
3912 	unsigned short old_blkoff;
3913 	unsigned char old_alloc_type;
3914 
3915 	segno = GET_SEGNO(sbi, new_blkaddr);
3916 	se = get_seg_entry(sbi, segno);
3917 	type = se->type;
3918 
3919 	f2fs_down_write(&SM_I(sbi)->curseg_lock);
3920 
3921 	if (!recover_curseg) {
3922 		/* for recovery flow */
3923 		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3924 			if (old_blkaddr == NULL_ADDR)
3925 				type = CURSEG_COLD_DATA;
3926 			else
3927 				type = CURSEG_WARM_DATA;
3928 		}
3929 	} else {
3930 		if (IS_CURSEG(sbi, segno)) {
3931 			/* se->type is volatile as SSR allocation */
3932 			type = __f2fs_get_curseg(sbi, segno);
3933 			f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3934 		} else {
3935 			type = CURSEG_WARM_DATA;
3936 		}
3937 	}
3938 
3939 	f2fs_bug_on(sbi, !IS_DATASEG(type));
3940 	curseg = CURSEG_I(sbi, type);
3941 
3942 	mutex_lock(&curseg->curseg_mutex);
3943 	down_write(&sit_i->sentry_lock);
3944 
3945 	old_cursegno = curseg->segno;
3946 	old_blkoff = curseg->next_blkoff;
3947 	old_alloc_type = curseg->alloc_type;
3948 
3949 	/* change the current segment */
3950 	if (segno != curseg->segno) {
3951 		curseg->next_segno = segno;
3952 		if (change_curseg(sbi, type))
3953 			goto out_unlock;
3954 	}
3955 
3956 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3957 	curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3958 
3959 	if (!recover_curseg || recover_newaddr) {
3960 		if (!from_gc)
3961 			update_segment_mtime(sbi, new_blkaddr, 0);
3962 		update_sit_entry(sbi, new_blkaddr, 1);
3963 	}
3964 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3965 		f2fs_invalidate_internal_cache(sbi, old_blkaddr);
3966 		if (!from_gc)
3967 			update_segment_mtime(sbi, old_blkaddr, 0);
3968 		update_sit_entry(sbi, old_blkaddr, -1);
3969 	}
3970 
3971 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3972 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3973 
3974 	locate_dirty_segment(sbi, old_cursegno);
3975 
3976 	if (recover_curseg) {
3977 		if (old_cursegno != curseg->segno) {
3978 			curseg->next_segno = old_cursegno;
3979 			if (change_curseg(sbi, type))
3980 				goto out_unlock;
3981 		}
3982 		curseg->next_blkoff = old_blkoff;
3983 		curseg->alloc_type = old_alloc_type;
3984 	}
3985 
3986 out_unlock:
3987 	up_write(&sit_i->sentry_lock);
3988 	mutex_unlock(&curseg->curseg_mutex);
3989 	f2fs_up_write(&SM_I(sbi)->curseg_lock);
3990 }
3991 
3992 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3993 				block_t old_addr, block_t new_addr,
3994 				unsigned char version, bool recover_curseg,
3995 				bool recover_newaddr)
3996 {
3997 	struct f2fs_summary sum;
3998 
3999 	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
4000 
4001 	f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
4002 					recover_curseg, recover_newaddr, false);
4003 
4004 	f2fs_update_data_blkaddr(dn, new_addr);
4005 }
4006 
4007 void f2fs_wait_on_page_writeback(struct page *page,
4008 				enum page_type type, bool ordered, bool locked)
4009 {
4010 	if (folio_test_writeback(page_folio(page))) {
4011 		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
4012 
4013 		/* submit cached LFS IO */
4014 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
4015 		/* submit cached IPU IO */
4016 		f2fs_submit_merged_ipu_write(sbi, NULL, page);
4017 		if (ordered) {
4018 			wait_on_page_writeback(page);
4019 			f2fs_bug_on(sbi, locked &&
4020 				folio_test_writeback(page_folio(page)));
4021 		} else {
4022 			wait_for_stable_page(page);
4023 		}
4024 	}
4025 }
4026 
4027 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
4028 {
4029 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
4030 	struct page *cpage;
4031 
4032 	if (!f2fs_meta_inode_gc_required(inode))
4033 		return;
4034 
4035 	if (!__is_valid_data_blkaddr(blkaddr))
4036 		return;
4037 
4038 	cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
4039 	if (cpage) {
4040 		f2fs_wait_on_page_writeback(cpage, DATA, true, true);
4041 		f2fs_put_page(cpage, 1);
4042 	}
4043 }
4044 
4045 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
4046 								block_t len)
4047 {
4048 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
4049 	block_t i;
4050 
4051 	if (!f2fs_meta_inode_gc_required(inode))
4052 		return;
4053 
4054 	for (i = 0; i < len; i++)
4055 		f2fs_wait_on_block_writeback(inode, blkaddr + i);
4056 
4057 	f2fs_truncate_meta_inode_pages(sbi, blkaddr, len);
4058 }
4059 
4060 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
4061 {
4062 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4063 	struct curseg_info *seg_i;
4064 	unsigned char *kaddr;
4065 	struct page *page;
4066 	block_t start;
4067 	int i, j, offset;
4068 
4069 	start = start_sum_block(sbi);
4070 
4071 	page = f2fs_get_meta_page(sbi, start++);
4072 	if (IS_ERR(page))
4073 		return PTR_ERR(page);
4074 	kaddr = (unsigned char *)page_address(page);
4075 
4076 	/* Step 1: restore nat cache */
4077 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
4078 	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
4079 
4080 	/* Step 2: restore sit cache */
4081 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
4082 	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
4083 	offset = 2 * SUM_JOURNAL_SIZE;
4084 
4085 	/* Step 3: restore summary entries */
4086 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
4087 		unsigned short blk_off;
4088 		unsigned int segno;
4089 
4090 		seg_i = CURSEG_I(sbi, i);
4091 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
4092 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
4093 		seg_i->next_segno = segno;
4094 		reset_curseg(sbi, i, 0);
4095 		seg_i->alloc_type = ckpt->alloc_type[i];
4096 		seg_i->next_blkoff = blk_off;
4097 
4098 		if (seg_i->alloc_type == SSR)
4099 			blk_off = BLKS_PER_SEG(sbi);
4100 
4101 		for (j = 0; j < blk_off; j++) {
4102 			struct f2fs_summary *s;
4103 
4104 			s = (struct f2fs_summary *)(kaddr + offset);
4105 			seg_i->sum_blk->entries[j] = *s;
4106 			offset += SUMMARY_SIZE;
4107 			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
4108 						SUM_FOOTER_SIZE)
4109 				continue;
4110 
4111 			f2fs_put_page(page, 1);
4112 			page = NULL;
4113 
4114 			page = f2fs_get_meta_page(sbi, start++);
4115 			if (IS_ERR(page))
4116 				return PTR_ERR(page);
4117 			kaddr = (unsigned char *)page_address(page);
4118 			offset = 0;
4119 		}
4120 	}
4121 	f2fs_put_page(page, 1);
4122 	return 0;
4123 }
4124 
4125 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
4126 {
4127 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4128 	struct f2fs_summary_block *sum;
4129 	struct curseg_info *curseg;
4130 	struct page *new;
4131 	unsigned short blk_off;
4132 	unsigned int segno = 0;
4133 	block_t blk_addr = 0;
4134 	int err = 0;
4135 
4136 	/* get segment number and block addr */
4137 	if (IS_DATASEG(type)) {
4138 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
4139 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
4140 							CURSEG_HOT_DATA]);
4141 		if (__exist_node_summaries(sbi))
4142 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
4143 		else
4144 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
4145 	} else {
4146 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
4147 							CURSEG_HOT_NODE]);
4148 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
4149 							CURSEG_HOT_NODE]);
4150 		if (__exist_node_summaries(sbi))
4151 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
4152 							type - CURSEG_HOT_NODE);
4153 		else
4154 			blk_addr = GET_SUM_BLOCK(sbi, segno);
4155 	}
4156 
4157 	new = f2fs_get_meta_page(sbi, blk_addr);
4158 	if (IS_ERR(new))
4159 		return PTR_ERR(new);
4160 	sum = (struct f2fs_summary_block *)page_address(new);
4161 
4162 	if (IS_NODESEG(type)) {
4163 		if (__exist_node_summaries(sbi)) {
4164 			struct f2fs_summary *ns = &sum->entries[0];
4165 			int i;
4166 
4167 			for (i = 0; i < BLKS_PER_SEG(sbi); i++, ns++) {
4168 				ns->version = 0;
4169 				ns->ofs_in_node = 0;
4170 			}
4171 		} else {
4172 			err = f2fs_restore_node_summary(sbi, segno, sum);
4173 			if (err)
4174 				goto out;
4175 		}
4176 	}
4177 
4178 	/* set uncompleted segment to curseg */
4179 	curseg = CURSEG_I(sbi, type);
4180 	mutex_lock(&curseg->curseg_mutex);
4181 
4182 	/* update journal info */
4183 	down_write(&curseg->journal_rwsem);
4184 	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
4185 	up_write(&curseg->journal_rwsem);
4186 
4187 	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
4188 	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
4189 	curseg->next_segno = segno;
4190 	reset_curseg(sbi, type, 0);
4191 	curseg->alloc_type = ckpt->alloc_type[type];
4192 	curseg->next_blkoff = blk_off;
4193 	mutex_unlock(&curseg->curseg_mutex);
4194 out:
4195 	f2fs_put_page(new, 1);
4196 	return err;
4197 }
4198 
4199 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
4200 {
4201 	struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
4202 	struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
4203 	int type = CURSEG_HOT_DATA;
4204 	int err;
4205 
4206 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
4207 		int npages = f2fs_npages_for_summary_flush(sbi, true);
4208 
4209 		if (npages >= 2)
4210 			f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
4211 							META_CP, true);
4212 
4213 		/* restore for compacted data summary */
4214 		err = read_compacted_summaries(sbi);
4215 		if (err)
4216 			return err;
4217 		type = CURSEG_HOT_NODE;
4218 	}
4219 
4220 	if (__exist_node_summaries(sbi))
4221 		f2fs_ra_meta_pages(sbi,
4222 				sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
4223 				NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
4224 
4225 	for (; type <= CURSEG_COLD_NODE; type++) {
4226 		err = read_normal_summaries(sbi, type);
4227 		if (err)
4228 			return err;
4229 	}
4230 
4231 	/* sanity check for summary blocks */
4232 	if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
4233 			sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
4234 		f2fs_err(sbi, "invalid journal entries nats %u sits %u",
4235 			 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
4236 		return -EINVAL;
4237 	}
4238 
4239 	return 0;
4240 }
4241 
4242 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
4243 {
4244 	struct page *page;
4245 	unsigned char *kaddr;
4246 	struct f2fs_summary *summary;
4247 	struct curseg_info *seg_i;
4248 	int written_size = 0;
4249 	int i, j;
4250 
4251 	page = f2fs_grab_meta_page(sbi, blkaddr++);
4252 	kaddr = (unsigned char *)page_address(page);
4253 	memset(kaddr, 0, PAGE_SIZE);
4254 
4255 	/* Step 1: write nat cache */
4256 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
4257 	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
4258 	written_size += SUM_JOURNAL_SIZE;
4259 
4260 	/* Step 2: write sit cache */
4261 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
4262 	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
4263 	written_size += SUM_JOURNAL_SIZE;
4264 
4265 	/* Step 3: write summary entries */
4266 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
4267 		seg_i = CURSEG_I(sbi, i);
4268 		for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); j++) {
4269 			if (!page) {
4270 				page = f2fs_grab_meta_page(sbi, blkaddr++);
4271 				kaddr = (unsigned char *)page_address(page);
4272 				memset(kaddr, 0, PAGE_SIZE);
4273 				written_size = 0;
4274 			}
4275 			summary = (struct f2fs_summary *)(kaddr + written_size);
4276 			*summary = seg_i->sum_blk->entries[j];
4277 			written_size += SUMMARY_SIZE;
4278 
4279 			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
4280 							SUM_FOOTER_SIZE)
4281 				continue;
4282 
4283 			set_page_dirty(page);
4284 			f2fs_put_page(page, 1);
4285 			page = NULL;
4286 		}
4287 	}
4288 	if (page) {
4289 		set_page_dirty(page);
4290 		f2fs_put_page(page, 1);
4291 	}
4292 }
4293 
4294 static void write_normal_summaries(struct f2fs_sb_info *sbi,
4295 					block_t blkaddr, int type)
4296 {
4297 	int i, end;
4298 
4299 	if (IS_DATASEG(type))
4300 		end = type + NR_CURSEG_DATA_TYPE;
4301 	else
4302 		end = type + NR_CURSEG_NODE_TYPE;
4303 
4304 	for (i = type; i < end; i++)
4305 		write_current_sum_page(sbi, i, blkaddr + (i - type));
4306 }
4307 
4308 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4309 {
4310 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
4311 		write_compacted_summaries(sbi, start_blk);
4312 	else
4313 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
4314 }
4315 
4316 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4317 {
4318 	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
4319 }
4320 
4321 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
4322 					unsigned int val, int alloc)
4323 {
4324 	int i;
4325 
4326 	if (type == NAT_JOURNAL) {
4327 		for (i = 0; i < nats_in_cursum(journal); i++) {
4328 			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
4329 				return i;
4330 		}
4331 		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
4332 			return update_nats_in_cursum(journal, 1);
4333 	} else if (type == SIT_JOURNAL) {
4334 		for (i = 0; i < sits_in_cursum(journal); i++)
4335 			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
4336 				return i;
4337 		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
4338 			return update_sits_in_cursum(journal, 1);
4339 	}
4340 	return -1;
4341 }
4342 
4343 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
4344 					unsigned int segno)
4345 {
4346 	return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
4347 }
4348 
4349 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
4350 					unsigned int start)
4351 {
4352 	struct sit_info *sit_i = SIT_I(sbi);
4353 	struct page *page;
4354 	pgoff_t src_off, dst_off;
4355 
4356 	src_off = current_sit_addr(sbi, start);
4357 	dst_off = next_sit_addr(sbi, src_off);
4358 
4359 	page = f2fs_grab_meta_page(sbi, dst_off);
4360 	seg_info_to_sit_page(sbi, page, start);
4361 
4362 	set_page_dirty(page);
4363 	set_to_next_sit(sit_i, start);
4364 
4365 	return page;
4366 }
4367 
4368 static struct sit_entry_set *grab_sit_entry_set(void)
4369 {
4370 	struct sit_entry_set *ses =
4371 			f2fs_kmem_cache_alloc(sit_entry_set_slab,
4372 						GFP_NOFS, true, NULL);
4373 
4374 	ses->entry_cnt = 0;
4375 	INIT_LIST_HEAD(&ses->set_list);
4376 	return ses;
4377 }
4378 
4379 static void release_sit_entry_set(struct sit_entry_set *ses)
4380 {
4381 	list_del(&ses->set_list);
4382 	kmem_cache_free(sit_entry_set_slab, ses);
4383 }
4384 
4385 static void adjust_sit_entry_set(struct sit_entry_set *ses,
4386 						struct list_head *head)
4387 {
4388 	struct sit_entry_set *next = ses;
4389 
4390 	if (list_is_last(&ses->set_list, head))
4391 		return;
4392 
4393 	list_for_each_entry_continue(next, head, set_list)
4394 		if (ses->entry_cnt <= next->entry_cnt) {
4395 			list_move_tail(&ses->set_list, &next->set_list);
4396 			return;
4397 		}
4398 
4399 	list_move_tail(&ses->set_list, head);
4400 }
4401 
4402 static void add_sit_entry(unsigned int segno, struct list_head *head)
4403 {
4404 	struct sit_entry_set *ses;
4405 	unsigned int start_segno = START_SEGNO(segno);
4406 
4407 	list_for_each_entry(ses, head, set_list) {
4408 		if (ses->start_segno == start_segno) {
4409 			ses->entry_cnt++;
4410 			adjust_sit_entry_set(ses, head);
4411 			return;
4412 		}
4413 	}
4414 
4415 	ses = grab_sit_entry_set();
4416 
4417 	ses->start_segno = start_segno;
4418 	ses->entry_cnt++;
4419 	list_add(&ses->set_list, head);
4420 }
4421 
4422 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4423 {
4424 	struct f2fs_sm_info *sm_info = SM_I(sbi);
4425 	struct list_head *set_list = &sm_info->sit_entry_set;
4426 	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4427 	unsigned int segno;
4428 
4429 	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4430 		add_sit_entry(segno, set_list);
4431 }
4432 
4433 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4434 {
4435 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4436 	struct f2fs_journal *journal = curseg->journal;
4437 	int i;
4438 
4439 	down_write(&curseg->journal_rwsem);
4440 	for (i = 0; i < sits_in_cursum(journal); i++) {
4441 		unsigned int segno;
4442 		bool dirtied;
4443 
4444 		segno = le32_to_cpu(segno_in_journal(journal, i));
4445 		dirtied = __mark_sit_entry_dirty(sbi, segno);
4446 
4447 		if (!dirtied)
4448 			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4449 	}
4450 	update_sits_in_cursum(journal, -i);
4451 	up_write(&curseg->journal_rwsem);
4452 }
4453 
4454 /*
4455  * CP calls this function, which flushes SIT entries including sit_journal,
4456  * and moves prefree segs to free segs.
4457  */
4458 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4459 {
4460 	struct sit_info *sit_i = SIT_I(sbi);
4461 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4462 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4463 	struct f2fs_journal *journal = curseg->journal;
4464 	struct sit_entry_set *ses, *tmp;
4465 	struct list_head *head = &SM_I(sbi)->sit_entry_set;
4466 	bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4467 	struct seg_entry *se;
4468 
4469 	down_write(&sit_i->sentry_lock);
4470 
4471 	if (!sit_i->dirty_sentries)
4472 		goto out;
4473 
4474 	/*
4475 	 * add and account sit entries of dirty bitmap in sit entry
4476 	 * set temporarily
4477 	 */
4478 	add_sits_in_set(sbi);
4479 
4480 	/*
4481 	 * if there are no enough space in journal to store dirty sit
4482 	 * entries, remove all entries from journal and add and account
4483 	 * them in sit entry set.
4484 	 */
4485 	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4486 								!to_journal)
4487 		remove_sits_in_journal(sbi);
4488 
4489 	/*
4490 	 * there are two steps to flush sit entries:
4491 	 * #1, flush sit entries to journal in current cold data summary block.
4492 	 * #2, flush sit entries to sit page.
4493 	 */
4494 	list_for_each_entry_safe(ses, tmp, head, set_list) {
4495 		struct page *page = NULL;
4496 		struct f2fs_sit_block *raw_sit = NULL;
4497 		unsigned int start_segno = ses->start_segno;
4498 		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4499 						(unsigned long)MAIN_SEGS(sbi));
4500 		unsigned int segno = start_segno;
4501 
4502 		if (to_journal &&
4503 			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4504 			to_journal = false;
4505 
4506 		if (to_journal) {
4507 			down_write(&curseg->journal_rwsem);
4508 		} else {
4509 			page = get_next_sit_page(sbi, start_segno);
4510 			raw_sit = page_address(page);
4511 		}
4512 
4513 		/* flush dirty sit entries in region of current sit set */
4514 		for_each_set_bit_from(segno, bitmap, end) {
4515 			int offset, sit_offset;
4516 
4517 			se = get_seg_entry(sbi, segno);
4518 #ifdef CONFIG_F2FS_CHECK_FS
4519 			if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4520 						SIT_VBLOCK_MAP_SIZE))
4521 				f2fs_bug_on(sbi, 1);
4522 #endif
4523 
4524 			/* add discard candidates */
4525 			if (!(cpc->reason & CP_DISCARD)) {
4526 				cpc->trim_start = segno;
4527 				add_discard_addrs(sbi, cpc, false);
4528 			}
4529 
4530 			if (to_journal) {
4531 				offset = f2fs_lookup_journal_in_cursum(journal,
4532 							SIT_JOURNAL, segno, 1);
4533 				f2fs_bug_on(sbi, offset < 0);
4534 				segno_in_journal(journal, offset) =
4535 							cpu_to_le32(segno);
4536 				seg_info_to_raw_sit(se,
4537 					&sit_in_journal(journal, offset));
4538 				check_block_count(sbi, segno,
4539 					&sit_in_journal(journal, offset));
4540 			} else {
4541 				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4542 				seg_info_to_raw_sit(se,
4543 						&raw_sit->entries[sit_offset]);
4544 				check_block_count(sbi, segno,
4545 						&raw_sit->entries[sit_offset]);
4546 			}
4547 
4548 			__clear_bit(segno, bitmap);
4549 			sit_i->dirty_sentries--;
4550 			ses->entry_cnt--;
4551 		}
4552 
4553 		if (to_journal)
4554 			up_write(&curseg->journal_rwsem);
4555 		else
4556 			f2fs_put_page(page, 1);
4557 
4558 		f2fs_bug_on(sbi, ses->entry_cnt);
4559 		release_sit_entry_set(ses);
4560 	}
4561 
4562 	f2fs_bug_on(sbi, !list_empty(head));
4563 	f2fs_bug_on(sbi, sit_i->dirty_sentries);
4564 out:
4565 	if (cpc->reason & CP_DISCARD) {
4566 		__u64 trim_start = cpc->trim_start;
4567 
4568 		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4569 			add_discard_addrs(sbi, cpc, false);
4570 
4571 		cpc->trim_start = trim_start;
4572 	}
4573 	up_write(&sit_i->sentry_lock);
4574 
4575 	set_prefree_as_free_segments(sbi);
4576 }
4577 
4578 static int build_sit_info(struct f2fs_sb_info *sbi)
4579 {
4580 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4581 	struct sit_info *sit_i;
4582 	unsigned int sit_segs, start;
4583 	char *src_bitmap, *bitmap;
4584 	unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4585 	unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0;
4586 
4587 	/* allocate memory for SIT information */
4588 	sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4589 	if (!sit_i)
4590 		return -ENOMEM;
4591 
4592 	SM_I(sbi)->sit_info = sit_i;
4593 
4594 	sit_i->sentries =
4595 		f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4596 					      MAIN_SEGS(sbi)),
4597 			      GFP_KERNEL);
4598 	if (!sit_i->sentries)
4599 		return -ENOMEM;
4600 
4601 	main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4602 	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4603 								GFP_KERNEL);
4604 	if (!sit_i->dirty_sentries_bitmap)
4605 		return -ENOMEM;
4606 
4607 #ifdef CONFIG_F2FS_CHECK_FS
4608 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map);
4609 #else
4610 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map);
4611 #endif
4612 	sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4613 	if (!sit_i->bitmap)
4614 		return -ENOMEM;
4615 
4616 	bitmap = sit_i->bitmap;
4617 
4618 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4619 		sit_i->sentries[start].cur_valid_map = bitmap;
4620 		bitmap += SIT_VBLOCK_MAP_SIZE;
4621 
4622 		sit_i->sentries[start].ckpt_valid_map = bitmap;
4623 		bitmap += SIT_VBLOCK_MAP_SIZE;
4624 
4625 #ifdef CONFIG_F2FS_CHECK_FS
4626 		sit_i->sentries[start].cur_valid_map_mir = bitmap;
4627 		bitmap += SIT_VBLOCK_MAP_SIZE;
4628 #endif
4629 
4630 		if (discard_map) {
4631 			sit_i->sentries[start].discard_map = bitmap;
4632 			bitmap += SIT_VBLOCK_MAP_SIZE;
4633 		}
4634 	}
4635 
4636 	sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4637 	if (!sit_i->tmp_map)
4638 		return -ENOMEM;
4639 
4640 	if (__is_large_section(sbi)) {
4641 		sit_i->sec_entries =
4642 			f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4643 						      MAIN_SECS(sbi)),
4644 				      GFP_KERNEL);
4645 		if (!sit_i->sec_entries)
4646 			return -ENOMEM;
4647 	}
4648 
4649 	/* get information related with SIT */
4650 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4651 
4652 	/* setup SIT bitmap from ckeckpoint pack */
4653 	sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4654 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4655 
4656 	sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4657 	if (!sit_i->sit_bitmap)
4658 		return -ENOMEM;
4659 
4660 #ifdef CONFIG_F2FS_CHECK_FS
4661 	sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4662 					sit_bitmap_size, GFP_KERNEL);
4663 	if (!sit_i->sit_bitmap_mir)
4664 		return -ENOMEM;
4665 
4666 	sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4667 					main_bitmap_size, GFP_KERNEL);
4668 	if (!sit_i->invalid_segmap)
4669 		return -ENOMEM;
4670 #endif
4671 
4672 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4673 	sit_i->sit_blocks = SEGS_TO_BLKS(sbi, sit_segs);
4674 	sit_i->written_valid_blocks = 0;
4675 	sit_i->bitmap_size = sit_bitmap_size;
4676 	sit_i->dirty_sentries = 0;
4677 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4678 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4679 	sit_i->mounted_time = ktime_get_boottime_seconds();
4680 	init_rwsem(&sit_i->sentry_lock);
4681 	return 0;
4682 }
4683 
4684 static int build_free_segmap(struct f2fs_sb_info *sbi)
4685 {
4686 	struct free_segmap_info *free_i;
4687 	unsigned int bitmap_size, sec_bitmap_size;
4688 
4689 	/* allocate memory for free segmap information */
4690 	free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4691 	if (!free_i)
4692 		return -ENOMEM;
4693 
4694 	SM_I(sbi)->free_info = free_i;
4695 
4696 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4697 	free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4698 	if (!free_i->free_segmap)
4699 		return -ENOMEM;
4700 
4701 	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4702 	free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4703 	if (!free_i->free_secmap)
4704 		return -ENOMEM;
4705 
4706 	/* set all segments as dirty temporarily */
4707 	memset(free_i->free_segmap, 0xff, bitmap_size);
4708 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4709 
4710 	/* init free segmap information */
4711 	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4712 	free_i->free_segments = 0;
4713 	free_i->free_sections = 0;
4714 	spin_lock_init(&free_i->segmap_lock);
4715 	return 0;
4716 }
4717 
4718 static int build_curseg(struct f2fs_sb_info *sbi)
4719 {
4720 	struct curseg_info *array;
4721 	int i;
4722 
4723 	array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4724 					sizeof(*array)), GFP_KERNEL);
4725 	if (!array)
4726 		return -ENOMEM;
4727 
4728 	SM_I(sbi)->curseg_array = array;
4729 
4730 	for (i = 0; i < NO_CHECK_TYPE; i++) {
4731 		mutex_init(&array[i].curseg_mutex);
4732 		array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4733 		if (!array[i].sum_blk)
4734 			return -ENOMEM;
4735 		init_rwsem(&array[i].journal_rwsem);
4736 		array[i].journal = f2fs_kzalloc(sbi,
4737 				sizeof(struct f2fs_journal), GFP_KERNEL);
4738 		if (!array[i].journal)
4739 			return -ENOMEM;
4740 		if (i < NR_PERSISTENT_LOG)
4741 			array[i].seg_type = CURSEG_HOT_DATA + i;
4742 		else if (i == CURSEG_COLD_DATA_PINNED)
4743 			array[i].seg_type = CURSEG_COLD_DATA;
4744 		else if (i == CURSEG_ALL_DATA_ATGC)
4745 			array[i].seg_type = CURSEG_COLD_DATA;
4746 		reset_curseg_fields(&array[i]);
4747 	}
4748 	return restore_curseg_summaries(sbi);
4749 }
4750 
4751 static int build_sit_entries(struct f2fs_sb_info *sbi)
4752 {
4753 	struct sit_info *sit_i = SIT_I(sbi);
4754 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4755 	struct f2fs_journal *journal = curseg->journal;
4756 	struct seg_entry *se;
4757 	struct f2fs_sit_entry sit;
4758 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
4759 	unsigned int i, start, end;
4760 	unsigned int readed, start_blk = 0;
4761 	int err = 0;
4762 	block_t sit_valid_blocks[2] = {0, 0};
4763 
4764 	do {
4765 		readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4766 							META_SIT, true);
4767 
4768 		start = start_blk * sit_i->sents_per_block;
4769 		end = (start_blk + readed) * sit_i->sents_per_block;
4770 
4771 		for (; start < end && start < MAIN_SEGS(sbi); start++) {
4772 			struct f2fs_sit_block *sit_blk;
4773 			struct page *page;
4774 
4775 			se = &sit_i->sentries[start];
4776 			page = get_current_sit_page(sbi, start);
4777 			if (IS_ERR(page))
4778 				return PTR_ERR(page);
4779 			sit_blk = (struct f2fs_sit_block *)page_address(page);
4780 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4781 			f2fs_put_page(page, 1);
4782 
4783 			err = check_block_count(sbi, start, &sit);
4784 			if (err)
4785 				return err;
4786 			seg_info_from_raw_sit(se, &sit);
4787 
4788 			if (se->type >= NR_PERSISTENT_LOG) {
4789 				f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4790 							se->type, start);
4791 				f2fs_handle_error(sbi,
4792 						ERROR_INCONSISTENT_SUM_TYPE);
4793 				return -EFSCORRUPTED;
4794 			}
4795 
4796 			sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4797 
4798 			if (!f2fs_block_unit_discard(sbi))
4799 				goto init_discard_map_done;
4800 
4801 			/* build discard map only one time */
4802 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4803 				memset(se->discard_map, 0xff,
4804 						SIT_VBLOCK_MAP_SIZE);
4805 				goto init_discard_map_done;
4806 			}
4807 			memcpy(se->discard_map, se->cur_valid_map,
4808 						SIT_VBLOCK_MAP_SIZE);
4809 			sbi->discard_blks += BLKS_PER_SEG(sbi) -
4810 						se->valid_blocks;
4811 init_discard_map_done:
4812 			if (__is_large_section(sbi))
4813 				get_sec_entry(sbi, start)->valid_blocks +=
4814 							se->valid_blocks;
4815 		}
4816 		start_blk += readed;
4817 	} while (start_blk < sit_blk_cnt);
4818 
4819 	down_read(&curseg->journal_rwsem);
4820 	for (i = 0; i < sits_in_cursum(journal); i++) {
4821 		unsigned int old_valid_blocks;
4822 
4823 		start = le32_to_cpu(segno_in_journal(journal, i));
4824 		if (start >= MAIN_SEGS(sbi)) {
4825 			f2fs_err(sbi, "Wrong journal entry on segno %u",
4826 				 start);
4827 			err = -EFSCORRUPTED;
4828 			f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL);
4829 			break;
4830 		}
4831 
4832 		se = &sit_i->sentries[start];
4833 		sit = sit_in_journal(journal, i);
4834 
4835 		old_valid_blocks = se->valid_blocks;
4836 
4837 		sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks;
4838 
4839 		err = check_block_count(sbi, start, &sit);
4840 		if (err)
4841 			break;
4842 		seg_info_from_raw_sit(se, &sit);
4843 
4844 		if (se->type >= NR_PERSISTENT_LOG) {
4845 			f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4846 							se->type, start);
4847 			err = -EFSCORRUPTED;
4848 			f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
4849 			break;
4850 		}
4851 
4852 		sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4853 
4854 		if (f2fs_block_unit_discard(sbi)) {
4855 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4856 				memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4857 			} else {
4858 				memcpy(se->discard_map, se->cur_valid_map,
4859 							SIT_VBLOCK_MAP_SIZE);
4860 				sbi->discard_blks += old_valid_blocks;
4861 				sbi->discard_blks -= se->valid_blocks;
4862 			}
4863 		}
4864 
4865 		if (__is_large_section(sbi)) {
4866 			get_sec_entry(sbi, start)->valid_blocks +=
4867 							se->valid_blocks;
4868 			get_sec_entry(sbi, start)->valid_blocks -=
4869 							old_valid_blocks;
4870 		}
4871 	}
4872 	up_read(&curseg->journal_rwsem);
4873 
4874 	if (err)
4875 		return err;
4876 
4877 	if (sit_valid_blocks[NODE] != valid_node_count(sbi)) {
4878 		f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4879 			 sit_valid_blocks[NODE], valid_node_count(sbi));
4880 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT);
4881 		return -EFSCORRUPTED;
4882 	}
4883 
4884 	if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] >
4885 				valid_user_blocks(sbi)) {
4886 		f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u",
4887 			 sit_valid_blocks[DATA], sit_valid_blocks[NODE],
4888 			 valid_user_blocks(sbi));
4889 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT);
4890 		return -EFSCORRUPTED;
4891 	}
4892 
4893 	return 0;
4894 }
4895 
4896 static void init_free_segmap(struct f2fs_sb_info *sbi)
4897 {
4898 	unsigned int start;
4899 	int type;
4900 	struct seg_entry *sentry;
4901 
4902 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4903 		if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4904 			continue;
4905 		sentry = get_seg_entry(sbi, start);
4906 		if (!sentry->valid_blocks)
4907 			__set_free(sbi, start);
4908 		else
4909 			SIT_I(sbi)->written_valid_blocks +=
4910 						sentry->valid_blocks;
4911 	}
4912 
4913 	/* set use the current segments */
4914 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4915 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4916 
4917 		__set_test_and_inuse(sbi, curseg_t->segno);
4918 	}
4919 }
4920 
4921 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4922 {
4923 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4924 	struct free_segmap_info *free_i = FREE_I(sbi);
4925 	unsigned int segno = 0, offset = 0, secno;
4926 	block_t valid_blocks, usable_blks_in_seg;
4927 
4928 	while (1) {
4929 		/* find dirty segment based on free segmap */
4930 		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4931 		if (segno >= MAIN_SEGS(sbi))
4932 			break;
4933 		offset = segno + 1;
4934 		valid_blocks = get_valid_blocks(sbi, segno, false);
4935 		usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4936 		if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4937 			continue;
4938 		if (valid_blocks > usable_blks_in_seg) {
4939 			f2fs_bug_on(sbi, 1);
4940 			continue;
4941 		}
4942 		mutex_lock(&dirty_i->seglist_lock);
4943 		__locate_dirty_segment(sbi, segno, DIRTY);
4944 		mutex_unlock(&dirty_i->seglist_lock);
4945 	}
4946 
4947 	if (!__is_large_section(sbi))
4948 		return;
4949 
4950 	mutex_lock(&dirty_i->seglist_lock);
4951 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) {
4952 		valid_blocks = get_valid_blocks(sbi, segno, true);
4953 		secno = GET_SEC_FROM_SEG(sbi, segno);
4954 
4955 		if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi))
4956 			continue;
4957 		if (IS_CURSEC(sbi, secno))
4958 			continue;
4959 		set_bit(secno, dirty_i->dirty_secmap);
4960 	}
4961 	mutex_unlock(&dirty_i->seglist_lock);
4962 }
4963 
4964 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4965 {
4966 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4967 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4968 
4969 	dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4970 	if (!dirty_i->victim_secmap)
4971 		return -ENOMEM;
4972 
4973 	dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4974 	if (!dirty_i->pinned_secmap)
4975 		return -ENOMEM;
4976 
4977 	dirty_i->pinned_secmap_cnt = 0;
4978 	dirty_i->enable_pin_section = true;
4979 	return 0;
4980 }
4981 
4982 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4983 {
4984 	struct dirty_seglist_info *dirty_i;
4985 	unsigned int bitmap_size, i;
4986 
4987 	/* allocate memory for dirty segments list information */
4988 	dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4989 								GFP_KERNEL);
4990 	if (!dirty_i)
4991 		return -ENOMEM;
4992 
4993 	SM_I(sbi)->dirty_info = dirty_i;
4994 	mutex_init(&dirty_i->seglist_lock);
4995 
4996 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4997 
4998 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
4999 		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
5000 								GFP_KERNEL);
5001 		if (!dirty_i->dirty_segmap[i])
5002 			return -ENOMEM;
5003 	}
5004 
5005 	if (__is_large_section(sbi)) {
5006 		bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
5007 		dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
5008 						bitmap_size, GFP_KERNEL);
5009 		if (!dirty_i->dirty_secmap)
5010 			return -ENOMEM;
5011 	}
5012 
5013 	init_dirty_segmap(sbi);
5014 	return init_victim_secmap(sbi);
5015 }
5016 
5017 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
5018 {
5019 	int i;
5020 
5021 	/*
5022 	 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
5023 	 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
5024 	 */
5025 	for (i = 0; i < NR_PERSISTENT_LOG; i++) {
5026 		struct curseg_info *curseg = CURSEG_I(sbi, i);
5027 		struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
5028 		unsigned int blkofs = curseg->next_blkoff;
5029 
5030 		if (f2fs_sb_has_readonly(sbi) &&
5031 			i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE)
5032 			continue;
5033 
5034 		sanity_check_seg_type(sbi, curseg->seg_type);
5035 
5036 		if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) {
5037 			f2fs_err(sbi,
5038 				 "Current segment has invalid alloc_type:%d",
5039 				 curseg->alloc_type);
5040 			f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
5041 			return -EFSCORRUPTED;
5042 		}
5043 
5044 		if (f2fs_test_bit(blkofs, se->cur_valid_map))
5045 			goto out;
5046 
5047 		if (curseg->alloc_type == SSR)
5048 			continue;
5049 
5050 		for (blkofs += 1; blkofs < BLKS_PER_SEG(sbi); blkofs++) {
5051 			if (!f2fs_test_bit(blkofs, se->cur_valid_map))
5052 				continue;
5053 out:
5054 			f2fs_err(sbi,
5055 				 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
5056 				 i, curseg->segno, curseg->alloc_type,
5057 				 curseg->next_blkoff, blkofs);
5058 			f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
5059 			return -EFSCORRUPTED;
5060 		}
5061 	}
5062 	return 0;
5063 }
5064 
5065 #ifdef CONFIG_BLK_DEV_ZONED
5066 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
5067 				    struct f2fs_dev_info *fdev,
5068 				    struct blk_zone *zone)
5069 {
5070 	unsigned int zone_segno;
5071 	block_t zone_block, valid_block_cnt;
5072 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
5073 	int ret;
5074 	unsigned int nofs_flags;
5075 
5076 	if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5077 		return 0;
5078 
5079 	zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
5080 	zone_segno = GET_SEGNO(sbi, zone_block);
5081 
5082 	/*
5083 	 * Skip check of zones cursegs point to, since
5084 	 * fix_curseg_write_pointer() checks them.
5085 	 */
5086 	if (zone_segno >= MAIN_SEGS(sbi))
5087 		return 0;
5088 
5089 	/*
5090 	 * Get # of valid block of the zone.
5091 	 */
5092 	valid_block_cnt = get_valid_blocks(sbi, zone_segno, true);
5093 	if (IS_CURSEC(sbi, GET_SEC_FROM_SEG(sbi, zone_segno))) {
5094 		f2fs_notice(sbi, "Open zones: valid block[0x%x,0x%x] cond[%s]",
5095 				zone_segno, valid_block_cnt,
5096 				blk_zone_cond_str(zone->cond));
5097 		return 0;
5098 	}
5099 
5100 	if ((!valid_block_cnt && zone->cond == BLK_ZONE_COND_EMPTY) ||
5101 	    (valid_block_cnt && zone->cond == BLK_ZONE_COND_FULL))
5102 		return 0;
5103 
5104 	if (!valid_block_cnt) {
5105 		f2fs_notice(sbi, "Zone without valid block has non-zero write "
5106 			    "pointer. Reset the write pointer: cond[%s]",
5107 			    blk_zone_cond_str(zone->cond));
5108 		ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
5109 					zone->len >> log_sectors_per_block);
5110 		if (ret)
5111 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
5112 				 fdev->path, ret);
5113 		return ret;
5114 	}
5115 
5116 	/*
5117 	 * If there are valid blocks and the write pointer doesn't match
5118 	 * with them, we need to report the inconsistency and fill
5119 	 * the zone till the end to close the zone. This inconsistency
5120 	 * does not cause write error because the zone will not be
5121 	 * selected for write operation until it get discarded.
5122 	 */
5123 	f2fs_notice(sbi, "Valid blocks are not aligned with write "
5124 		    "pointer: valid block[0x%x,0x%x] cond[%s]",
5125 		    zone_segno, valid_block_cnt, blk_zone_cond_str(zone->cond));
5126 
5127 	nofs_flags = memalloc_nofs_save();
5128 	ret = blkdev_zone_mgmt(fdev->bdev, REQ_OP_ZONE_FINISH,
5129 				zone->start, zone->len);
5130 	memalloc_nofs_restore(nofs_flags);
5131 	if (ret == -EOPNOTSUPP) {
5132 		ret = blkdev_issue_zeroout(fdev->bdev, zone->wp,
5133 					zone->len - (zone->wp - zone->start),
5134 					GFP_NOFS, 0);
5135 		if (ret)
5136 			f2fs_err(sbi, "Fill up zone failed: %s (errno=%d)",
5137 					fdev->path, ret);
5138 	} else if (ret) {
5139 		f2fs_err(sbi, "Finishing zone failed: %s (errno=%d)",
5140 				fdev->path, ret);
5141 	}
5142 
5143 	return ret;
5144 }
5145 
5146 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
5147 						  block_t zone_blkaddr)
5148 {
5149 	int i;
5150 
5151 	for (i = 0; i < sbi->s_ndevs; i++) {
5152 		if (!bdev_is_zoned(FDEV(i).bdev))
5153 			continue;
5154 		if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
5155 				zone_blkaddr <= FDEV(i).end_blk))
5156 			return &FDEV(i);
5157 	}
5158 
5159 	return NULL;
5160 }
5161 
5162 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
5163 			      void *data)
5164 {
5165 	memcpy(data, zone, sizeof(struct blk_zone));
5166 	return 0;
5167 }
5168 
5169 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
5170 {
5171 	struct curseg_info *cs = CURSEG_I(sbi, type);
5172 	struct f2fs_dev_info *zbd;
5173 	struct blk_zone zone;
5174 	unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
5175 	block_t cs_zone_block, wp_block;
5176 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
5177 	sector_t zone_sector;
5178 	int err;
5179 
5180 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
5181 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
5182 
5183 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
5184 	if (!zbd)
5185 		return 0;
5186 
5187 	/* report zone for the sector the curseg points to */
5188 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5189 		<< log_sectors_per_block;
5190 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5191 				  report_one_zone_cb, &zone);
5192 	if (err != 1) {
5193 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5194 			 zbd->path, err);
5195 		return err;
5196 	}
5197 
5198 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5199 		return 0;
5200 
5201 	/*
5202 	 * When safely unmounted in the previous mount, we could use current
5203 	 * segments. Otherwise, allocate new sections.
5204 	 */
5205 	if (is_set_ckpt_flags(sbi, CP_UMOUNT_FLAG)) {
5206 		wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
5207 		wp_segno = GET_SEGNO(sbi, wp_block);
5208 		wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
5209 		wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
5210 
5211 		if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
5212 				wp_sector_off == 0)
5213 			return 0;
5214 
5215 		f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
5216 			    "curseg[0x%x,0x%x] wp[0x%x,0x%x]", type, cs->segno,
5217 			    cs->next_blkoff, wp_segno, wp_blkoff);
5218 	}
5219 
5220 	/* Allocate a new section if it's not new. */
5221 	if (cs->next_blkoff ||
5222 	    cs->segno != GET_SEG_FROM_SEC(sbi, GET_ZONE_FROM_SEC(sbi, cs_section))) {
5223 		unsigned int old_segno = cs->segno, old_blkoff = cs->next_blkoff;
5224 
5225 		f2fs_allocate_new_section(sbi, type, true);
5226 		f2fs_notice(sbi, "Assign new section to curseg[%d]: "
5227 				"[0x%x,0x%x] -> [0x%x,0x%x]",
5228 				type, old_segno, old_blkoff,
5229 				cs->segno, cs->next_blkoff);
5230 	}
5231 
5232 	/* check consistency of the zone curseg pointed to */
5233 	if (check_zone_write_pointer(sbi, zbd, &zone))
5234 		return -EIO;
5235 
5236 	/* check newly assigned zone */
5237 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
5238 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
5239 
5240 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
5241 	if (!zbd)
5242 		return 0;
5243 
5244 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5245 		<< log_sectors_per_block;
5246 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5247 				  report_one_zone_cb, &zone);
5248 	if (err != 1) {
5249 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5250 			 zbd->path, err);
5251 		return err;
5252 	}
5253 
5254 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5255 		return 0;
5256 
5257 	if (zone.wp != zone.start) {
5258 		f2fs_notice(sbi,
5259 			    "New zone for curseg[%d] is not yet discarded. "
5260 			    "Reset the zone: curseg[0x%x,0x%x]",
5261 			    type, cs->segno, cs->next_blkoff);
5262 		err = __f2fs_issue_discard_zone(sbi, zbd->bdev,	cs_zone_block,
5263 					zone.len >> log_sectors_per_block);
5264 		if (err) {
5265 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
5266 				 zbd->path, err);
5267 			return err;
5268 		}
5269 	}
5270 
5271 	return 0;
5272 }
5273 
5274 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5275 {
5276 	int i, ret;
5277 
5278 	for (i = 0; i < NR_PERSISTENT_LOG; i++) {
5279 		ret = fix_curseg_write_pointer(sbi, i);
5280 		if (ret)
5281 			return ret;
5282 	}
5283 
5284 	return 0;
5285 }
5286 
5287 struct check_zone_write_pointer_args {
5288 	struct f2fs_sb_info *sbi;
5289 	struct f2fs_dev_info *fdev;
5290 };
5291 
5292 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
5293 				      void *data)
5294 {
5295 	struct check_zone_write_pointer_args *args;
5296 
5297 	args = (struct check_zone_write_pointer_args *)data;
5298 
5299 	return check_zone_write_pointer(args->sbi, args->fdev, zone);
5300 }
5301 
5302 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5303 {
5304 	int i, ret;
5305 	struct check_zone_write_pointer_args args;
5306 
5307 	for (i = 0; i < sbi->s_ndevs; i++) {
5308 		if (!bdev_is_zoned(FDEV(i).bdev))
5309 			continue;
5310 
5311 		args.sbi = sbi;
5312 		args.fdev = &FDEV(i);
5313 		ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
5314 					  check_zone_write_pointer_cb, &args);
5315 		if (ret < 0)
5316 			return ret;
5317 	}
5318 
5319 	return 0;
5320 }
5321 
5322 /*
5323  * Return the number of usable blocks in a segment. The number of blocks
5324  * returned is always equal to the number of blocks in a segment for
5325  * segments fully contained within a sequential zone capacity or a
5326  * conventional zone. For segments partially contained in a sequential
5327  * zone capacity, the number of usable blocks up to the zone capacity
5328  * is returned. 0 is returned in all other cases.
5329  */
5330 static inline unsigned int f2fs_usable_zone_blks_in_seg(
5331 			struct f2fs_sb_info *sbi, unsigned int segno)
5332 {
5333 	block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
5334 	unsigned int secno;
5335 
5336 	if (!sbi->unusable_blocks_per_sec)
5337 		return BLKS_PER_SEG(sbi);
5338 
5339 	secno = GET_SEC_FROM_SEG(sbi, segno);
5340 	seg_start = START_BLOCK(sbi, segno);
5341 	sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5342 	sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi);
5343 
5344 	/*
5345 	 * If segment starts before zone capacity and spans beyond
5346 	 * zone capacity, then usable blocks are from seg start to
5347 	 * zone capacity. If the segment starts after the zone capacity,
5348 	 * then there are no usable blocks.
5349 	 */
5350 	if (seg_start >= sec_cap_blkaddr)
5351 		return 0;
5352 	if (seg_start + BLKS_PER_SEG(sbi) > sec_cap_blkaddr)
5353 		return sec_cap_blkaddr - seg_start;
5354 
5355 	return BLKS_PER_SEG(sbi);
5356 }
5357 #else
5358 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5359 {
5360 	return 0;
5361 }
5362 
5363 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5364 {
5365 	return 0;
5366 }
5367 
5368 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5369 							unsigned int segno)
5370 {
5371 	return 0;
5372 }
5373 
5374 #endif
5375 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5376 					unsigned int segno)
5377 {
5378 	if (f2fs_sb_has_blkzoned(sbi))
5379 		return f2fs_usable_zone_blks_in_seg(sbi, segno);
5380 
5381 	return BLKS_PER_SEG(sbi);
5382 }
5383 
5384 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5385 					unsigned int segno)
5386 {
5387 	if (f2fs_sb_has_blkzoned(sbi))
5388 		return CAP_SEGS_PER_SEC(sbi);
5389 
5390 	return SEGS_PER_SEC(sbi);
5391 }
5392 
5393 /*
5394  * Update min, max modified time for cost-benefit GC algorithm
5395  */
5396 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5397 {
5398 	struct sit_info *sit_i = SIT_I(sbi);
5399 	unsigned int segno;
5400 
5401 	down_write(&sit_i->sentry_lock);
5402 
5403 	sit_i->min_mtime = ULLONG_MAX;
5404 
5405 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) {
5406 		unsigned int i;
5407 		unsigned long long mtime = 0;
5408 
5409 		for (i = 0; i < SEGS_PER_SEC(sbi); i++)
5410 			mtime += get_seg_entry(sbi, segno + i)->mtime;
5411 
5412 		mtime = div_u64(mtime, SEGS_PER_SEC(sbi));
5413 
5414 		if (sit_i->min_mtime > mtime)
5415 			sit_i->min_mtime = mtime;
5416 	}
5417 	sit_i->max_mtime = get_mtime(sbi, false);
5418 	sit_i->dirty_max_mtime = 0;
5419 	up_write(&sit_i->sentry_lock);
5420 }
5421 
5422 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5423 {
5424 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5425 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5426 	struct f2fs_sm_info *sm_info;
5427 	int err;
5428 
5429 	sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5430 	if (!sm_info)
5431 		return -ENOMEM;
5432 
5433 	/* init sm info */
5434 	sbi->sm_info = sm_info;
5435 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5436 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5437 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5438 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5439 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5440 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5441 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5442 	sm_info->rec_prefree_segments = sm_info->main_segments *
5443 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5444 	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5445 		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5446 
5447 	if (!f2fs_lfs_mode(sbi))
5448 		sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC);
5449 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5450 	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5451 	sm_info->min_seq_blocks = BLKS_PER_SEG(sbi);
5452 	sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5453 	sm_info->min_ssr_sections = reserved_sections(sbi);
5454 
5455 	INIT_LIST_HEAD(&sm_info->sit_entry_set);
5456 
5457 	init_f2fs_rwsem(&sm_info->curseg_lock);
5458 
5459 	err = f2fs_create_flush_cmd_control(sbi);
5460 	if (err)
5461 		return err;
5462 
5463 	err = create_discard_cmd_control(sbi);
5464 	if (err)
5465 		return err;
5466 
5467 	err = build_sit_info(sbi);
5468 	if (err)
5469 		return err;
5470 	err = build_free_segmap(sbi);
5471 	if (err)
5472 		return err;
5473 	err = build_curseg(sbi);
5474 	if (err)
5475 		return err;
5476 
5477 	/* reinit free segmap based on SIT */
5478 	err = build_sit_entries(sbi);
5479 	if (err)
5480 		return err;
5481 
5482 	init_free_segmap(sbi);
5483 	err = build_dirty_segmap(sbi);
5484 	if (err)
5485 		return err;
5486 
5487 	err = sanity_check_curseg(sbi);
5488 	if (err)
5489 		return err;
5490 
5491 	init_min_max_mtime(sbi);
5492 	return 0;
5493 }
5494 
5495 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5496 		enum dirty_type dirty_type)
5497 {
5498 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5499 
5500 	mutex_lock(&dirty_i->seglist_lock);
5501 	kvfree(dirty_i->dirty_segmap[dirty_type]);
5502 	dirty_i->nr_dirty[dirty_type] = 0;
5503 	mutex_unlock(&dirty_i->seglist_lock);
5504 }
5505 
5506 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5507 {
5508 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5509 
5510 	kvfree(dirty_i->pinned_secmap);
5511 	kvfree(dirty_i->victim_secmap);
5512 }
5513 
5514 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5515 {
5516 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5517 	int i;
5518 
5519 	if (!dirty_i)
5520 		return;
5521 
5522 	/* discard pre-free/dirty segments list */
5523 	for (i = 0; i < NR_DIRTY_TYPE; i++)
5524 		discard_dirty_segmap(sbi, i);
5525 
5526 	if (__is_large_section(sbi)) {
5527 		mutex_lock(&dirty_i->seglist_lock);
5528 		kvfree(dirty_i->dirty_secmap);
5529 		mutex_unlock(&dirty_i->seglist_lock);
5530 	}
5531 
5532 	destroy_victim_secmap(sbi);
5533 	SM_I(sbi)->dirty_info = NULL;
5534 	kfree(dirty_i);
5535 }
5536 
5537 static void destroy_curseg(struct f2fs_sb_info *sbi)
5538 {
5539 	struct curseg_info *array = SM_I(sbi)->curseg_array;
5540 	int i;
5541 
5542 	if (!array)
5543 		return;
5544 	SM_I(sbi)->curseg_array = NULL;
5545 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
5546 		kfree(array[i].sum_blk);
5547 		kfree(array[i].journal);
5548 	}
5549 	kfree(array);
5550 }
5551 
5552 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5553 {
5554 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5555 
5556 	if (!free_i)
5557 		return;
5558 	SM_I(sbi)->free_info = NULL;
5559 	kvfree(free_i->free_segmap);
5560 	kvfree(free_i->free_secmap);
5561 	kfree(free_i);
5562 }
5563 
5564 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5565 {
5566 	struct sit_info *sit_i = SIT_I(sbi);
5567 
5568 	if (!sit_i)
5569 		return;
5570 
5571 	if (sit_i->sentries)
5572 		kvfree(sit_i->bitmap);
5573 	kfree(sit_i->tmp_map);
5574 
5575 	kvfree(sit_i->sentries);
5576 	kvfree(sit_i->sec_entries);
5577 	kvfree(sit_i->dirty_sentries_bitmap);
5578 
5579 	SM_I(sbi)->sit_info = NULL;
5580 	kvfree(sit_i->sit_bitmap);
5581 #ifdef CONFIG_F2FS_CHECK_FS
5582 	kvfree(sit_i->sit_bitmap_mir);
5583 	kvfree(sit_i->invalid_segmap);
5584 #endif
5585 	kfree(sit_i);
5586 }
5587 
5588 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5589 {
5590 	struct f2fs_sm_info *sm_info = SM_I(sbi);
5591 
5592 	if (!sm_info)
5593 		return;
5594 	f2fs_destroy_flush_cmd_control(sbi, true);
5595 	destroy_discard_cmd_control(sbi);
5596 	destroy_dirty_segmap(sbi);
5597 	destroy_curseg(sbi);
5598 	destroy_free_segmap(sbi);
5599 	destroy_sit_info(sbi);
5600 	sbi->sm_info = NULL;
5601 	kfree(sm_info);
5602 }
5603 
5604 int __init f2fs_create_segment_manager_caches(void)
5605 {
5606 	discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5607 			sizeof(struct discard_entry));
5608 	if (!discard_entry_slab)
5609 		goto fail;
5610 
5611 	discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5612 			sizeof(struct discard_cmd));
5613 	if (!discard_cmd_slab)
5614 		goto destroy_discard_entry;
5615 
5616 	sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5617 			sizeof(struct sit_entry_set));
5618 	if (!sit_entry_set_slab)
5619 		goto destroy_discard_cmd;
5620 
5621 	revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry",
5622 			sizeof(struct revoke_entry));
5623 	if (!revoke_entry_slab)
5624 		goto destroy_sit_entry_set;
5625 	return 0;
5626 
5627 destroy_sit_entry_set:
5628 	kmem_cache_destroy(sit_entry_set_slab);
5629 destroy_discard_cmd:
5630 	kmem_cache_destroy(discard_cmd_slab);
5631 destroy_discard_entry:
5632 	kmem_cache_destroy(discard_entry_slab);
5633 fail:
5634 	return -ENOMEM;
5635 }
5636 
5637 void f2fs_destroy_segment_manager_caches(void)
5638 {
5639 	kmem_cache_destroy(sit_entry_set_slab);
5640 	kmem_cache_destroy(discard_cmd_slab);
5641 	kmem_cache_destroy(discard_entry_slab);
5642 	kmem_cache_destroy(revoke_entry_slab);
5643 }
5644