xref: /openbmc/linux/fs/f2fs/segment.c (revision 49c23519)
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 				f2fs_handle_error(sbi,
340 						ERROR_INVALID_BLKADDR);
341 				goto out;
342 			}
343 
344 			new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS,
345 							true, NULL);
346 
347 			ret = __replace_atomic_write_block(inode, index, blkaddr,
348 							&new->old_addr, false);
349 			if (ret) {
350 				f2fs_put_dnode(&dn);
351 				kmem_cache_free(revoke_entry_slab, new);
352 				goto out;
353 			}
354 
355 			f2fs_update_data_blkaddr(&dn, NULL_ADDR);
356 			new->index = index;
357 			list_add_tail(&new->list, &revoke_list);
358 		}
359 		f2fs_put_dnode(&dn);
360 next:
361 		off += blen;
362 		len -= blen;
363 	}
364 
365 out:
366 	if (ret) {
367 		sbi->revoked_atomic_block += fi->atomic_write_cnt;
368 	} else {
369 		sbi->committed_atomic_block += fi->atomic_write_cnt;
370 		set_inode_flag(inode, FI_ATOMIC_COMMITTED);
371 	}
372 
373 	__complete_revoke_list(inode, &revoke_list, ret ? true : false);
374 
375 	return ret;
376 }
377 
378 int f2fs_commit_atomic_write(struct inode *inode)
379 {
380 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
381 	struct f2fs_inode_info *fi = F2FS_I(inode);
382 	int err;
383 
384 	err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
385 	if (err)
386 		return err;
387 
388 	f2fs_down_write(&fi->i_gc_rwsem[WRITE]);
389 	f2fs_lock_op(sbi);
390 
391 	err = __f2fs_commit_atomic_write(inode);
392 
393 	f2fs_unlock_op(sbi);
394 	f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
395 
396 	return err;
397 }
398 
399 /*
400  * This function balances dirty node and dentry pages.
401  * In addition, it controls garbage collection.
402  */
403 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
404 {
405 	if (time_to_inject(sbi, FAULT_CHECKPOINT))
406 		f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT);
407 
408 	/* balance_fs_bg is able to be pending */
409 	if (need && excess_cached_nats(sbi))
410 		f2fs_balance_fs_bg(sbi, false);
411 
412 	if (!f2fs_is_checkpoint_ready(sbi))
413 		return;
414 
415 	/*
416 	 * We should do GC or end up with checkpoint, if there are so many dirty
417 	 * dir/node pages without enough free segments.
418 	 */
419 	if (has_enough_free_secs(sbi, 0, 0))
420 		return;
421 
422 	if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
423 				sbi->gc_thread->f2fs_gc_task) {
424 		DEFINE_WAIT(wait);
425 
426 		prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
427 					TASK_UNINTERRUPTIBLE);
428 		wake_up(&sbi->gc_thread->gc_wait_queue_head);
429 		io_schedule();
430 		finish_wait(&sbi->gc_thread->fggc_wq, &wait);
431 	} else {
432 		struct f2fs_gc_control gc_control = {
433 			.victim_segno = NULL_SEGNO,
434 			.init_gc_type = BG_GC,
435 			.no_bg_gc = true,
436 			.should_migrate_blocks = false,
437 			.err_gc_skipped = false,
438 			.nr_free_secs = 1 };
439 		f2fs_down_write(&sbi->gc_lock);
440 		stat_inc_gc_call_count(sbi, FOREGROUND);
441 		f2fs_gc(sbi, &gc_control);
442 	}
443 }
444 
445 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi)
446 {
447 	int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2;
448 	unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS);
449 	unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA);
450 	unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES);
451 	unsigned int meta = get_pages(sbi, F2FS_DIRTY_META);
452 	unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA);
453 	unsigned int threshold = (factor * DEFAULT_DIRTY_THRESHOLD) <<
454 				sbi->log_blocks_per_seg;
455 	unsigned int global_threshold = threshold * 3 / 2;
456 
457 	if (dents >= threshold || qdata >= threshold ||
458 		nodes >= threshold || meta >= threshold ||
459 		imeta >= threshold)
460 		return true;
461 	return dents + qdata + nodes + meta + imeta >  global_threshold;
462 }
463 
464 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
465 {
466 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
467 		return;
468 
469 	/* try to shrink extent cache when there is no enough memory */
470 	if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE))
471 		f2fs_shrink_read_extent_tree(sbi,
472 				READ_EXTENT_CACHE_SHRINK_NUMBER);
473 
474 	/* try to shrink age extent cache when there is no enough memory */
475 	if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE))
476 		f2fs_shrink_age_extent_tree(sbi,
477 				AGE_EXTENT_CACHE_SHRINK_NUMBER);
478 
479 	/* check the # of cached NAT entries */
480 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
481 		f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
482 
483 	if (!f2fs_available_free_memory(sbi, FREE_NIDS))
484 		f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
485 	else
486 		f2fs_build_free_nids(sbi, false, false);
487 
488 	if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) ||
489 		excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi))
490 		goto do_sync;
491 
492 	/* there is background inflight IO or foreground operation recently */
493 	if (is_inflight_io(sbi, REQ_TIME) ||
494 		(!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem)))
495 		return;
496 
497 	/* exceed periodical checkpoint timeout threshold */
498 	if (f2fs_time_over(sbi, CP_TIME))
499 		goto do_sync;
500 
501 	/* checkpoint is the only way to shrink partial cached entries */
502 	if (f2fs_available_free_memory(sbi, NAT_ENTRIES) &&
503 		f2fs_available_free_memory(sbi, INO_ENTRIES))
504 		return;
505 
506 do_sync:
507 	if (test_opt(sbi, DATA_FLUSH) && from_bg) {
508 		struct blk_plug plug;
509 
510 		mutex_lock(&sbi->flush_lock);
511 
512 		blk_start_plug(&plug);
513 		f2fs_sync_dirty_inodes(sbi, FILE_INODE, false);
514 		blk_finish_plug(&plug);
515 
516 		mutex_unlock(&sbi->flush_lock);
517 	}
518 	stat_inc_cp_call_count(sbi, BACKGROUND);
519 	f2fs_sync_fs(sbi->sb, 1);
520 }
521 
522 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
523 				struct block_device *bdev)
524 {
525 	int ret = blkdev_issue_flush(bdev);
526 
527 	trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
528 				test_opt(sbi, FLUSH_MERGE), ret);
529 	if (!ret)
530 		f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0);
531 	return ret;
532 }
533 
534 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
535 {
536 	int ret = 0;
537 	int i;
538 
539 	if (!f2fs_is_multi_device(sbi))
540 		return __submit_flush_wait(sbi, sbi->sb->s_bdev);
541 
542 	for (i = 0; i < sbi->s_ndevs; i++) {
543 		if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
544 			continue;
545 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
546 		if (ret)
547 			break;
548 	}
549 	return ret;
550 }
551 
552 static int issue_flush_thread(void *data)
553 {
554 	struct f2fs_sb_info *sbi = data;
555 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
556 	wait_queue_head_t *q = &fcc->flush_wait_queue;
557 repeat:
558 	if (kthread_should_stop())
559 		return 0;
560 
561 	if (!llist_empty(&fcc->issue_list)) {
562 		struct flush_cmd *cmd, *next;
563 		int ret;
564 
565 		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
566 		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
567 
568 		cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
569 
570 		ret = submit_flush_wait(sbi, cmd->ino);
571 		atomic_inc(&fcc->issued_flush);
572 
573 		llist_for_each_entry_safe(cmd, next,
574 					  fcc->dispatch_list, llnode) {
575 			cmd->ret = ret;
576 			complete(&cmd->wait);
577 		}
578 		fcc->dispatch_list = NULL;
579 	}
580 
581 	wait_event_interruptible(*q,
582 		kthread_should_stop() || !llist_empty(&fcc->issue_list));
583 	goto repeat;
584 }
585 
586 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
587 {
588 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
589 	struct flush_cmd cmd;
590 	int ret;
591 
592 	if (test_opt(sbi, NOBARRIER))
593 		return 0;
594 
595 	if (!test_opt(sbi, FLUSH_MERGE)) {
596 		atomic_inc(&fcc->queued_flush);
597 		ret = submit_flush_wait(sbi, ino);
598 		atomic_dec(&fcc->queued_flush);
599 		atomic_inc(&fcc->issued_flush);
600 		return ret;
601 	}
602 
603 	if (atomic_inc_return(&fcc->queued_flush) == 1 ||
604 	    f2fs_is_multi_device(sbi)) {
605 		ret = submit_flush_wait(sbi, ino);
606 		atomic_dec(&fcc->queued_flush);
607 
608 		atomic_inc(&fcc->issued_flush);
609 		return ret;
610 	}
611 
612 	cmd.ino = ino;
613 	init_completion(&cmd.wait);
614 
615 	llist_add(&cmd.llnode, &fcc->issue_list);
616 
617 	/*
618 	 * update issue_list before we wake up issue_flush thread, this
619 	 * smp_mb() pairs with another barrier in ___wait_event(), see
620 	 * more details in comments of waitqueue_active().
621 	 */
622 	smp_mb();
623 
624 	if (waitqueue_active(&fcc->flush_wait_queue))
625 		wake_up(&fcc->flush_wait_queue);
626 
627 	if (fcc->f2fs_issue_flush) {
628 		wait_for_completion(&cmd.wait);
629 		atomic_dec(&fcc->queued_flush);
630 	} else {
631 		struct llist_node *list;
632 
633 		list = llist_del_all(&fcc->issue_list);
634 		if (!list) {
635 			wait_for_completion(&cmd.wait);
636 			atomic_dec(&fcc->queued_flush);
637 		} else {
638 			struct flush_cmd *tmp, *next;
639 
640 			ret = submit_flush_wait(sbi, ino);
641 
642 			llist_for_each_entry_safe(tmp, next, list, llnode) {
643 				if (tmp == &cmd) {
644 					cmd.ret = ret;
645 					atomic_dec(&fcc->queued_flush);
646 					continue;
647 				}
648 				tmp->ret = ret;
649 				complete(&tmp->wait);
650 			}
651 		}
652 	}
653 
654 	return cmd.ret;
655 }
656 
657 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
658 {
659 	dev_t dev = sbi->sb->s_bdev->bd_dev;
660 	struct flush_cmd_control *fcc;
661 
662 	if (SM_I(sbi)->fcc_info) {
663 		fcc = SM_I(sbi)->fcc_info;
664 		if (fcc->f2fs_issue_flush)
665 			return 0;
666 		goto init_thread;
667 	}
668 
669 	fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
670 	if (!fcc)
671 		return -ENOMEM;
672 	atomic_set(&fcc->issued_flush, 0);
673 	atomic_set(&fcc->queued_flush, 0);
674 	init_waitqueue_head(&fcc->flush_wait_queue);
675 	init_llist_head(&fcc->issue_list);
676 	SM_I(sbi)->fcc_info = fcc;
677 	if (!test_opt(sbi, FLUSH_MERGE))
678 		return 0;
679 
680 init_thread:
681 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
682 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
683 	if (IS_ERR(fcc->f2fs_issue_flush)) {
684 		int err = PTR_ERR(fcc->f2fs_issue_flush);
685 
686 		fcc->f2fs_issue_flush = NULL;
687 		return err;
688 	}
689 
690 	return 0;
691 }
692 
693 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
694 {
695 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
696 
697 	if (fcc && fcc->f2fs_issue_flush) {
698 		struct task_struct *flush_thread = fcc->f2fs_issue_flush;
699 
700 		fcc->f2fs_issue_flush = NULL;
701 		kthread_stop(flush_thread);
702 	}
703 	if (free) {
704 		kfree(fcc);
705 		SM_I(sbi)->fcc_info = NULL;
706 	}
707 }
708 
709 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
710 {
711 	int ret = 0, i;
712 
713 	if (!f2fs_is_multi_device(sbi))
714 		return 0;
715 
716 	if (test_opt(sbi, NOBARRIER))
717 		return 0;
718 
719 	for (i = 1; i < sbi->s_ndevs; i++) {
720 		int count = DEFAULT_RETRY_IO_COUNT;
721 
722 		if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
723 			continue;
724 
725 		do {
726 			ret = __submit_flush_wait(sbi, FDEV(i).bdev);
727 			if (ret)
728 				f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
729 		} while (ret && --count);
730 
731 		if (ret) {
732 			f2fs_stop_checkpoint(sbi, false,
733 					STOP_CP_REASON_FLUSH_FAIL);
734 			break;
735 		}
736 
737 		spin_lock(&sbi->dev_lock);
738 		f2fs_clear_bit(i, (char *)&sbi->dirty_device);
739 		spin_unlock(&sbi->dev_lock);
740 	}
741 
742 	return ret;
743 }
744 
745 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
746 		enum dirty_type dirty_type)
747 {
748 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
749 
750 	/* need not be added */
751 	if (IS_CURSEG(sbi, segno))
752 		return;
753 
754 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
755 		dirty_i->nr_dirty[dirty_type]++;
756 
757 	if (dirty_type == DIRTY) {
758 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
759 		enum dirty_type t = sentry->type;
760 
761 		if (unlikely(t >= DIRTY)) {
762 			f2fs_bug_on(sbi, 1);
763 			return;
764 		}
765 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
766 			dirty_i->nr_dirty[t]++;
767 
768 		if (__is_large_section(sbi)) {
769 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
770 			block_t valid_blocks =
771 				get_valid_blocks(sbi, segno, true);
772 
773 			f2fs_bug_on(sbi, unlikely(!valid_blocks ||
774 					valid_blocks == CAP_BLKS_PER_SEC(sbi)));
775 
776 			if (!IS_CURSEC(sbi, secno))
777 				set_bit(secno, dirty_i->dirty_secmap);
778 		}
779 	}
780 }
781 
782 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
783 		enum dirty_type dirty_type)
784 {
785 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
786 	block_t valid_blocks;
787 
788 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
789 		dirty_i->nr_dirty[dirty_type]--;
790 
791 	if (dirty_type == DIRTY) {
792 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
793 		enum dirty_type t = sentry->type;
794 
795 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
796 			dirty_i->nr_dirty[t]--;
797 
798 		valid_blocks = get_valid_blocks(sbi, segno, true);
799 		if (valid_blocks == 0) {
800 			clear_bit(GET_SEC_FROM_SEG(sbi, segno),
801 						dirty_i->victim_secmap);
802 #ifdef CONFIG_F2FS_CHECK_FS
803 			clear_bit(segno, SIT_I(sbi)->invalid_segmap);
804 #endif
805 		}
806 		if (__is_large_section(sbi)) {
807 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
808 
809 			if (!valid_blocks ||
810 					valid_blocks == CAP_BLKS_PER_SEC(sbi)) {
811 				clear_bit(secno, dirty_i->dirty_secmap);
812 				return;
813 			}
814 
815 			if (!IS_CURSEC(sbi, secno))
816 				set_bit(secno, dirty_i->dirty_secmap);
817 		}
818 	}
819 }
820 
821 /*
822  * Should not occur error such as -ENOMEM.
823  * Adding dirty entry into seglist is not critical operation.
824  * If a given segment is one of current working segments, it won't be added.
825  */
826 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
827 {
828 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
829 	unsigned short valid_blocks, ckpt_valid_blocks;
830 	unsigned int usable_blocks;
831 
832 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
833 		return;
834 
835 	usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
836 	mutex_lock(&dirty_i->seglist_lock);
837 
838 	valid_blocks = get_valid_blocks(sbi, segno, false);
839 	ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
840 
841 	if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
842 		ckpt_valid_blocks == usable_blocks)) {
843 		__locate_dirty_segment(sbi, segno, PRE);
844 		__remove_dirty_segment(sbi, segno, DIRTY);
845 	} else if (valid_blocks < usable_blocks) {
846 		__locate_dirty_segment(sbi, segno, DIRTY);
847 	} else {
848 		/* Recovery routine with SSR needs this */
849 		__remove_dirty_segment(sbi, segno, DIRTY);
850 	}
851 
852 	mutex_unlock(&dirty_i->seglist_lock);
853 }
854 
855 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
856 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
857 {
858 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
859 	unsigned int segno;
860 
861 	mutex_lock(&dirty_i->seglist_lock);
862 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
863 		if (get_valid_blocks(sbi, segno, false))
864 			continue;
865 		if (IS_CURSEG(sbi, segno))
866 			continue;
867 		__locate_dirty_segment(sbi, segno, PRE);
868 		__remove_dirty_segment(sbi, segno, DIRTY);
869 	}
870 	mutex_unlock(&dirty_i->seglist_lock);
871 }
872 
873 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
874 {
875 	int ovp_hole_segs =
876 		(overprovision_segments(sbi) - reserved_segments(sbi));
877 	block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
878 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
879 	block_t holes[2] = {0, 0};	/* DATA and NODE */
880 	block_t unusable;
881 	struct seg_entry *se;
882 	unsigned int segno;
883 
884 	mutex_lock(&dirty_i->seglist_lock);
885 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
886 		se = get_seg_entry(sbi, segno);
887 		if (IS_NODESEG(se->type))
888 			holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
889 							se->valid_blocks;
890 		else
891 			holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
892 							se->valid_blocks;
893 	}
894 	mutex_unlock(&dirty_i->seglist_lock);
895 
896 	unusable = max(holes[DATA], holes[NODE]);
897 	if (unusable > ovp_holes)
898 		return unusable - ovp_holes;
899 	return 0;
900 }
901 
902 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
903 {
904 	int ovp_hole_segs =
905 		(overprovision_segments(sbi) - reserved_segments(sbi));
906 	if (unusable > F2FS_OPTION(sbi).unusable_cap)
907 		return -EAGAIN;
908 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
909 		dirty_segments(sbi) > ovp_hole_segs)
910 		return -EAGAIN;
911 	return 0;
912 }
913 
914 /* This is only used by SBI_CP_DISABLED */
915 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
916 {
917 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
918 	unsigned int segno = 0;
919 
920 	mutex_lock(&dirty_i->seglist_lock);
921 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
922 		if (get_valid_blocks(sbi, segno, false))
923 			continue;
924 		if (get_ckpt_valid_blocks(sbi, segno, false))
925 			continue;
926 		mutex_unlock(&dirty_i->seglist_lock);
927 		return segno;
928 	}
929 	mutex_unlock(&dirty_i->seglist_lock);
930 	return NULL_SEGNO;
931 }
932 
933 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
934 		struct block_device *bdev, block_t lstart,
935 		block_t start, block_t len)
936 {
937 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
938 	struct list_head *pend_list;
939 	struct discard_cmd *dc;
940 
941 	f2fs_bug_on(sbi, !len);
942 
943 	pend_list = &dcc->pend_list[plist_idx(len)];
944 
945 	dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL);
946 	INIT_LIST_HEAD(&dc->list);
947 	dc->bdev = bdev;
948 	dc->di.lstart = lstart;
949 	dc->di.start = start;
950 	dc->di.len = len;
951 	dc->ref = 0;
952 	dc->state = D_PREP;
953 	dc->queued = 0;
954 	dc->error = 0;
955 	init_completion(&dc->wait);
956 	list_add_tail(&dc->list, pend_list);
957 	spin_lock_init(&dc->lock);
958 	dc->bio_ref = 0;
959 	atomic_inc(&dcc->discard_cmd_cnt);
960 	dcc->undiscard_blks += len;
961 
962 	return dc;
963 }
964 
965 static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi)
966 {
967 #ifdef CONFIG_F2FS_CHECK_FS
968 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
969 	struct rb_node *cur = rb_first_cached(&dcc->root), *next;
970 	struct discard_cmd *cur_dc, *next_dc;
971 
972 	while (cur) {
973 		next = rb_next(cur);
974 		if (!next)
975 			return true;
976 
977 		cur_dc = rb_entry(cur, struct discard_cmd, rb_node);
978 		next_dc = rb_entry(next, struct discard_cmd, rb_node);
979 
980 		if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) {
981 			f2fs_info(sbi, "broken discard_rbtree, "
982 				"cur(%u, %u) next(%u, %u)",
983 				cur_dc->di.lstart, cur_dc->di.len,
984 				next_dc->di.lstart, next_dc->di.len);
985 			return false;
986 		}
987 		cur = next;
988 	}
989 #endif
990 	return true;
991 }
992 
993 static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi,
994 						block_t blkaddr)
995 {
996 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
997 	struct rb_node *node = dcc->root.rb_root.rb_node;
998 	struct discard_cmd *dc;
999 
1000 	while (node) {
1001 		dc = rb_entry(node, struct discard_cmd, rb_node);
1002 
1003 		if (blkaddr < dc->di.lstart)
1004 			node = node->rb_left;
1005 		else if (blkaddr >= dc->di.lstart + dc->di.len)
1006 			node = node->rb_right;
1007 		else
1008 			return dc;
1009 	}
1010 	return NULL;
1011 }
1012 
1013 static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root,
1014 				block_t blkaddr,
1015 				struct discard_cmd **prev_entry,
1016 				struct discard_cmd **next_entry,
1017 				struct rb_node ***insert_p,
1018 				struct rb_node **insert_parent)
1019 {
1020 	struct rb_node **pnode = &root->rb_root.rb_node;
1021 	struct rb_node *parent = NULL, *tmp_node;
1022 	struct discard_cmd *dc;
1023 
1024 	*insert_p = NULL;
1025 	*insert_parent = NULL;
1026 	*prev_entry = NULL;
1027 	*next_entry = NULL;
1028 
1029 	if (RB_EMPTY_ROOT(&root->rb_root))
1030 		return NULL;
1031 
1032 	while (*pnode) {
1033 		parent = *pnode;
1034 		dc = rb_entry(*pnode, struct discard_cmd, rb_node);
1035 
1036 		if (blkaddr < dc->di.lstart)
1037 			pnode = &(*pnode)->rb_left;
1038 		else if (blkaddr >= dc->di.lstart + dc->di.len)
1039 			pnode = &(*pnode)->rb_right;
1040 		else
1041 			goto lookup_neighbors;
1042 	}
1043 
1044 	*insert_p = pnode;
1045 	*insert_parent = parent;
1046 
1047 	dc = rb_entry(parent, struct discard_cmd, rb_node);
1048 	tmp_node = parent;
1049 	if (parent && blkaddr > dc->di.lstart)
1050 		tmp_node = rb_next(parent);
1051 	*next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1052 
1053 	tmp_node = parent;
1054 	if (parent && blkaddr < dc->di.lstart)
1055 		tmp_node = rb_prev(parent);
1056 	*prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1057 	return NULL;
1058 
1059 lookup_neighbors:
1060 	/* lookup prev node for merging backward later */
1061 	tmp_node = rb_prev(&dc->rb_node);
1062 	*prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1063 
1064 	/* lookup next node for merging frontward later */
1065 	tmp_node = rb_next(&dc->rb_node);
1066 	*next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1067 	return dc;
1068 }
1069 
1070 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1071 							struct discard_cmd *dc)
1072 {
1073 	if (dc->state == D_DONE)
1074 		atomic_sub(dc->queued, &dcc->queued_discard);
1075 
1076 	list_del(&dc->list);
1077 	rb_erase_cached(&dc->rb_node, &dcc->root);
1078 	dcc->undiscard_blks -= dc->di.len;
1079 
1080 	kmem_cache_free(discard_cmd_slab, dc);
1081 
1082 	atomic_dec(&dcc->discard_cmd_cnt);
1083 }
1084 
1085 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1086 							struct discard_cmd *dc)
1087 {
1088 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1089 	unsigned long flags;
1090 
1091 	trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len);
1092 
1093 	spin_lock_irqsave(&dc->lock, flags);
1094 	if (dc->bio_ref) {
1095 		spin_unlock_irqrestore(&dc->lock, flags);
1096 		return;
1097 	}
1098 	spin_unlock_irqrestore(&dc->lock, flags);
1099 
1100 	f2fs_bug_on(sbi, dc->ref);
1101 
1102 	if (dc->error == -EOPNOTSUPP)
1103 		dc->error = 0;
1104 
1105 	if (dc->error)
1106 		f2fs_info_ratelimited(sbi,
1107 			"Issue discard(%u, %u, %u) failed, ret: %d",
1108 			dc->di.lstart, dc->di.start, dc->di.len, dc->error);
1109 	__detach_discard_cmd(dcc, dc);
1110 }
1111 
1112 static void f2fs_submit_discard_endio(struct bio *bio)
1113 {
1114 	struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1115 	unsigned long flags;
1116 
1117 	spin_lock_irqsave(&dc->lock, flags);
1118 	if (!dc->error)
1119 		dc->error = blk_status_to_errno(bio->bi_status);
1120 	dc->bio_ref--;
1121 	if (!dc->bio_ref && dc->state == D_SUBMIT) {
1122 		dc->state = D_DONE;
1123 		complete_all(&dc->wait);
1124 	}
1125 	spin_unlock_irqrestore(&dc->lock, flags);
1126 	bio_put(bio);
1127 }
1128 
1129 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1130 				block_t start, block_t end)
1131 {
1132 #ifdef CONFIG_F2FS_CHECK_FS
1133 	struct seg_entry *sentry;
1134 	unsigned int segno;
1135 	block_t blk = start;
1136 	unsigned long offset, size, *map;
1137 
1138 	while (blk < end) {
1139 		segno = GET_SEGNO(sbi, blk);
1140 		sentry = get_seg_entry(sbi, segno);
1141 		offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1142 
1143 		if (end < START_BLOCK(sbi, segno + 1))
1144 			size = GET_BLKOFF_FROM_SEG0(sbi, end);
1145 		else
1146 			size = BLKS_PER_SEG(sbi);
1147 		map = (unsigned long *)(sentry->cur_valid_map);
1148 		offset = __find_rev_next_bit(map, size, offset);
1149 		f2fs_bug_on(sbi, offset != size);
1150 		blk = START_BLOCK(sbi, segno + 1);
1151 	}
1152 #endif
1153 }
1154 
1155 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1156 				struct discard_policy *dpolicy,
1157 				int discard_type, unsigned int granularity)
1158 {
1159 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1160 
1161 	/* common policy */
1162 	dpolicy->type = discard_type;
1163 	dpolicy->sync = true;
1164 	dpolicy->ordered = false;
1165 	dpolicy->granularity = granularity;
1166 
1167 	dpolicy->max_requests = dcc->max_discard_request;
1168 	dpolicy->io_aware_gran = dcc->discard_io_aware_gran;
1169 	dpolicy->timeout = false;
1170 
1171 	if (discard_type == DPOLICY_BG) {
1172 		dpolicy->min_interval = dcc->min_discard_issue_time;
1173 		dpolicy->mid_interval = dcc->mid_discard_issue_time;
1174 		dpolicy->max_interval = dcc->max_discard_issue_time;
1175 		dpolicy->io_aware = true;
1176 		dpolicy->sync = false;
1177 		dpolicy->ordered = true;
1178 		if (utilization(sbi) > dcc->discard_urgent_util) {
1179 			dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1180 			if (atomic_read(&dcc->discard_cmd_cnt))
1181 				dpolicy->max_interval =
1182 					dcc->min_discard_issue_time;
1183 		}
1184 	} else if (discard_type == DPOLICY_FORCE) {
1185 		dpolicy->min_interval = dcc->min_discard_issue_time;
1186 		dpolicy->mid_interval = dcc->mid_discard_issue_time;
1187 		dpolicy->max_interval = dcc->max_discard_issue_time;
1188 		dpolicy->io_aware = false;
1189 	} else if (discard_type == DPOLICY_FSTRIM) {
1190 		dpolicy->io_aware = false;
1191 	} else if (discard_type == DPOLICY_UMOUNT) {
1192 		dpolicy->io_aware = false;
1193 		/* we need to issue all to keep CP_TRIMMED_FLAG */
1194 		dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1195 		dpolicy->timeout = true;
1196 	}
1197 }
1198 
1199 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1200 				struct block_device *bdev, block_t lstart,
1201 				block_t start, block_t len);
1202 
1203 #ifdef CONFIG_BLK_DEV_ZONED
1204 static void __submit_zone_reset_cmd(struct f2fs_sb_info *sbi,
1205 				   struct discard_cmd *dc, blk_opf_t flag,
1206 				   struct list_head *wait_list,
1207 				   unsigned int *issued)
1208 {
1209 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1210 	struct block_device *bdev = dc->bdev;
1211 	struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS);
1212 	unsigned long flags;
1213 
1214 	trace_f2fs_issue_reset_zone(bdev, dc->di.start);
1215 
1216 	spin_lock_irqsave(&dc->lock, flags);
1217 	dc->state = D_SUBMIT;
1218 	dc->bio_ref++;
1219 	spin_unlock_irqrestore(&dc->lock, flags);
1220 
1221 	if (issued)
1222 		(*issued)++;
1223 
1224 	atomic_inc(&dcc->queued_discard);
1225 	dc->queued++;
1226 	list_move_tail(&dc->list, wait_list);
1227 
1228 	/* sanity check on discard range */
1229 	__check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len);
1230 
1231 	bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start);
1232 	bio->bi_private = dc;
1233 	bio->bi_end_io = f2fs_submit_discard_endio;
1234 	submit_bio(bio);
1235 
1236 	atomic_inc(&dcc->issued_discard);
1237 	f2fs_update_iostat(sbi, NULL, FS_ZONE_RESET_IO, dc->di.len * F2FS_BLKSIZE);
1238 }
1239 #endif
1240 
1241 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1242 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1243 				struct discard_policy *dpolicy,
1244 				struct discard_cmd *dc, int *issued)
1245 {
1246 	struct block_device *bdev = dc->bdev;
1247 	unsigned int max_discard_blocks =
1248 			SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1249 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1250 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1251 					&(dcc->fstrim_list) : &(dcc->wait_list);
1252 	blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0;
1253 	block_t lstart, start, len, total_len;
1254 	int err = 0;
1255 
1256 	if (dc->state != D_PREP)
1257 		return 0;
1258 
1259 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1260 		return 0;
1261 
1262 #ifdef CONFIG_BLK_DEV_ZONED
1263 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) {
1264 		int devi = f2fs_bdev_index(sbi, bdev);
1265 
1266 		if (devi < 0)
1267 			return -EINVAL;
1268 
1269 		if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1270 			__submit_zone_reset_cmd(sbi, dc, flag,
1271 						wait_list, issued);
1272 			return 0;
1273 		}
1274 	}
1275 #endif
1276 
1277 	trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len);
1278 
1279 	lstart = dc->di.lstart;
1280 	start = dc->di.start;
1281 	len = dc->di.len;
1282 	total_len = len;
1283 
1284 	dc->di.len = 0;
1285 
1286 	while (total_len && *issued < dpolicy->max_requests && !err) {
1287 		struct bio *bio = NULL;
1288 		unsigned long flags;
1289 		bool last = true;
1290 
1291 		if (len > max_discard_blocks) {
1292 			len = max_discard_blocks;
1293 			last = false;
1294 		}
1295 
1296 		(*issued)++;
1297 		if (*issued == dpolicy->max_requests)
1298 			last = true;
1299 
1300 		dc->di.len += len;
1301 
1302 		if (time_to_inject(sbi, FAULT_DISCARD)) {
1303 			err = -EIO;
1304 		} else {
1305 			err = __blkdev_issue_discard(bdev,
1306 					SECTOR_FROM_BLOCK(start),
1307 					SECTOR_FROM_BLOCK(len),
1308 					GFP_NOFS, &bio);
1309 		}
1310 		if (err) {
1311 			spin_lock_irqsave(&dc->lock, flags);
1312 			if (dc->state == D_PARTIAL)
1313 				dc->state = D_SUBMIT;
1314 			spin_unlock_irqrestore(&dc->lock, flags);
1315 
1316 			break;
1317 		}
1318 
1319 		f2fs_bug_on(sbi, !bio);
1320 
1321 		/*
1322 		 * should keep before submission to avoid D_DONE
1323 		 * right away
1324 		 */
1325 		spin_lock_irqsave(&dc->lock, flags);
1326 		if (last)
1327 			dc->state = D_SUBMIT;
1328 		else
1329 			dc->state = D_PARTIAL;
1330 		dc->bio_ref++;
1331 		spin_unlock_irqrestore(&dc->lock, flags);
1332 
1333 		atomic_inc(&dcc->queued_discard);
1334 		dc->queued++;
1335 		list_move_tail(&dc->list, wait_list);
1336 
1337 		/* sanity check on discard range */
1338 		__check_sit_bitmap(sbi, lstart, lstart + len);
1339 
1340 		bio->bi_private = dc;
1341 		bio->bi_end_io = f2fs_submit_discard_endio;
1342 		bio->bi_opf |= flag;
1343 		submit_bio(bio);
1344 
1345 		atomic_inc(&dcc->issued_discard);
1346 
1347 		f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE);
1348 
1349 		lstart += len;
1350 		start += len;
1351 		total_len -= len;
1352 		len = total_len;
1353 	}
1354 
1355 	if (!err && len) {
1356 		dcc->undiscard_blks -= len;
1357 		__update_discard_tree_range(sbi, bdev, lstart, start, len);
1358 	}
1359 	return err;
1360 }
1361 
1362 static void __insert_discard_cmd(struct f2fs_sb_info *sbi,
1363 				struct block_device *bdev, block_t lstart,
1364 				block_t start, block_t len)
1365 {
1366 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1367 	struct rb_node **p = &dcc->root.rb_root.rb_node;
1368 	struct rb_node *parent = NULL;
1369 	struct discard_cmd *dc;
1370 	bool leftmost = true;
1371 
1372 	/* look up rb tree to find parent node */
1373 	while (*p) {
1374 		parent = *p;
1375 		dc = rb_entry(parent, struct discard_cmd, rb_node);
1376 
1377 		if (lstart < dc->di.lstart) {
1378 			p = &(*p)->rb_left;
1379 		} else if (lstart >= dc->di.lstart + dc->di.len) {
1380 			p = &(*p)->rb_right;
1381 			leftmost = false;
1382 		} else {
1383 			f2fs_bug_on(sbi, 1);
1384 		}
1385 	}
1386 
1387 	dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1388 
1389 	rb_link_node(&dc->rb_node, parent, p);
1390 	rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1391 }
1392 
1393 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1394 						struct discard_cmd *dc)
1395 {
1396 	list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]);
1397 }
1398 
1399 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1400 				struct discard_cmd *dc, block_t blkaddr)
1401 {
1402 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1403 	struct discard_info di = dc->di;
1404 	bool modified = false;
1405 
1406 	if (dc->state == D_DONE || dc->di.len == 1) {
1407 		__remove_discard_cmd(sbi, dc);
1408 		return;
1409 	}
1410 
1411 	dcc->undiscard_blks -= di.len;
1412 
1413 	if (blkaddr > di.lstart) {
1414 		dc->di.len = blkaddr - dc->di.lstart;
1415 		dcc->undiscard_blks += dc->di.len;
1416 		__relocate_discard_cmd(dcc, dc);
1417 		modified = true;
1418 	}
1419 
1420 	if (blkaddr < di.lstart + di.len - 1) {
1421 		if (modified) {
1422 			__insert_discard_cmd(sbi, dc->bdev, blkaddr + 1,
1423 					di.start + blkaddr + 1 - di.lstart,
1424 					di.lstart + di.len - 1 - blkaddr);
1425 		} else {
1426 			dc->di.lstart++;
1427 			dc->di.len--;
1428 			dc->di.start++;
1429 			dcc->undiscard_blks += dc->di.len;
1430 			__relocate_discard_cmd(dcc, dc);
1431 		}
1432 	}
1433 }
1434 
1435 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1436 				struct block_device *bdev, block_t lstart,
1437 				block_t start, block_t len)
1438 {
1439 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1440 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1441 	struct discard_cmd *dc;
1442 	struct discard_info di = {0};
1443 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1444 	unsigned int max_discard_blocks =
1445 			SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1446 	block_t end = lstart + len;
1447 
1448 	dc = __lookup_discard_cmd_ret(&dcc->root, lstart,
1449 				&prev_dc, &next_dc, &insert_p, &insert_parent);
1450 	if (dc)
1451 		prev_dc = dc;
1452 
1453 	if (!prev_dc) {
1454 		di.lstart = lstart;
1455 		di.len = next_dc ? next_dc->di.lstart - lstart : len;
1456 		di.len = min(di.len, len);
1457 		di.start = start;
1458 	}
1459 
1460 	while (1) {
1461 		struct rb_node *node;
1462 		bool merged = false;
1463 		struct discard_cmd *tdc = NULL;
1464 
1465 		if (prev_dc) {
1466 			di.lstart = prev_dc->di.lstart + prev_dc->di.len;
1467 			if (di.lstart < lstart)
1468 				di.lstart = lstart;
1469 			if (di.lstart >= end)
1470 				break;
1471 
1472 			if (!next_dc || next_dc->di.lstart > end)
1473 				di.len = end - di.lstart;
1474 			else
1475 				di.len = next_dc->di.lstart - di.lstart;
1476 			di.start = start + di.lstart - lstart;
1477 		}
1478 
1479 		if (!di.len)
1480 			goto next;
1481 
1482 		if (prev_dc && prev_dc->state == D_PREP &&
1483 			prev_dc->bdev == bdev &&
1484 			__is_discard_back_mergeable(&di, &prev_dc->di,
1485 							max_discard_blocks)) {
1486 			prev_dc->di.len += di.len;
1487 			dcc->undiscard_blks += di.len;
1488 			__relocate_discard_cmd(dcc, prev_dc);
1489 			di = prev_dc->di;
1490 			tdc = prev_dc;
1491 			merged = true;
1492 		}
1493 
1494 		if (next_dc && next_dc->state == D_PREP &&
1495 			next_dc->bdev == bdev &&
1496 			__is_discard_front_mergeable(&di, &next_dc->di,
1497 							max_discard_blocks)) {
1498 			next_dc->di.lstart = di.lstart;
1499 			next_dc->di.len += di.len;
1500 			next_dc->di.start = di.start;
1501 			dcc->undiscard_blks += di.len;
1502 			__relocate_discard_cmd(dcc, next_dc);
1503 			if (tdc)
1504 				__remove_discard_cmd(sbi, tdc);
1505 			merged = true;
1506 		}
1507 
1508 		if (!merged)
1509 			__insert_discard_cmd(sbi, bdev,
1510 						di.lstart, di.start, di.len);
1511  next:
1512 		prev_dc = next_dc;
1513 		if (!prev_dc)
1514 			break;
1515 
1516 		node = rb_next(&prev_dc->rb_node);
1517 		next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1518 	}
1519 }
1520 
1521 #ifdef CONFIG_BLK_DEV_ZONED
1522 static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi,
1523 		struct block_device *bdev, block_t blkstart, block_t lblkstart,
1524 		block_t blklen)
1525 {
1526 	trace_f2fs_queue_reset_zone(bdev, blkstart);
1527 
1528 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1529 	__insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen);
1530 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1531 }
1532 #endif
1533 
1534 static void __queue_discard_cmd(struct f2fs_sb_info *sbi,
1535 		struct block_device *bdev, block_t blkstart, block_t blklen)
1536 {
1537 	block_t lblkstart = blkstart;
1538 
1539 	if (!f2fs_bdev_support_discard(bdev))
1540 		return;
1541 
1542 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
1543 
1544 	if (f2fs_is_multi_device(sbi)) {
1545 		int devi = f2fs_target_device_index(sbi, blkstart);
1546 
1547 		blkstart -= FDEV(devi).start_blk;
1548 	}
1549 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1550 	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1551 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1552 }
1553 
1554 static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1555 		struct discard_policy *dpolicy, int *issued)
1556 {
1557 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1558 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1559 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1560 	struct discard_cmd *dc;
1561 	struct blk_plug plug;
1562 	bool io_interrupted = false;
1563 
1564 	mutex_lock(&dcc->cmd_lock);
1565 	dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos,
1566 				&prev_dc, &next_dc, &insert_p, &insert_parent);
1567 	if (!dc)
1568 		dc = next_dc;
1569 
1570 	blk_start_plug(&plug);
1571 
1572 	while (dc) {
1573 		struct rb_node *node;
1574 		int err = 0;
1575 
1576 		if (dc->state != D_PREP)
1577 			goto next;
1578 
1579 		if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1580 			io_interrupted = true;
1581 			break;
1582 		}
1583 
1584 		dcc->next_pos = dc->di.lstart + dc->di.len;
1585 		err = __submit_discard_cmd(sbi, dpolicy, dc, issued);
1586 
1587 		if (*issued >= dpolicy->max_requests)
1588 			break;
1589 next:
1590 		node = rb_next(&dc->rb_node);
1591 		if (err)
1592 			__remove_discard_cmd(sbi, dc);
1593 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1594 	}
1595 
1596 	blk_finish_plug(&plug);
1597 
1598 	if (!dc)
1599 		dcc->next_pos = 0;
1600 
1601 	mutex_unlock(&dcc->cmd_lock);
1602 
1603 	if (!(*issued) && io_interrupted)
1604 		*issued = -1;
1605 }
1606 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1607 					struct discard_policy *dpolicy);
1608 
1609 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1610 					struct discard_policy *dpolicy)
1611 {
1612 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1613 	struct list_head *pend_list;
1614 	struct discard_cmd *dc, *tmp;
1615 	struct blk_plug plug;
1616 	int i, issued;
1617 	bool io_interrupted = false;
1618 
1619 	if (dpolicy->timeout)
1620 		f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1621 
1622 retry:
1623 	issued = 0;
1624 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1625 		if (dpolicy->timeout &&
1626 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1627 			break;
1628 
1629 		if (i + 1 < dpolicy->granularity)
1630 			break;
1631 
1632 		if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) {
1633 			__issue_discard_cmd_orderly(sbi, dpolicy, &issued);
1634 			return issued;
1635 		}
1636 
1637 		pend_list = &dcc->pend_list[i];
1638 
1639 		mutex_lock(&dcc->cmd_lock);
1640 		if (list_empty(pend_list))
1641 			goto next;
1642 		if (unlikely(dcc->rbtree_check))
1643 			f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
1644 		blk_start_plug(&plug);
1645 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1646 			f2fs_bug_on(sbi, dc->state != D_PREP);
1647 
1648 			if (dpolicy->timeout &&
1649 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1650 				break;
1651 
1652 			if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1653 						!is_idle(sbi, DISCARD_TIME)) {
1654 				io_interrupted = true;
1655 				break;
1656 			}
1657 
1658 			__submit_discard_cmd(sbi, dpolicy, dc, &issued);
1659 
1660 			if (issued >= dpolicy->max_requests)
1661 				break;
1662 		}
1663 		blk_finish_plug(&plug);
1664 next:
1665 		mutex_unlock(&dcc->cmd_lock);
1666 
1667 		if (issued >= dpolicy->max_requests || io_interrupted)
1668 			break;
1669 	}
1670 
1671 	if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1672 		__wait_all_discard_cmd(sbi, dpolicy);
1673 		goto retry;
1674 	}
1675 
1676 	if (!issued && io_interrupted)
1677 		issued = -1;
1678 
1679 	return issued;
1680 }
1681 
1682 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1683 {
1684 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1685 	struct list_head *pend_list;
1686 	struct discard_cmd *dc, *tmp;
1687 	int i;
1688 	bool dropped = false;
1689 
1690 	mutex_lock(&dcc->cmd_lock);
1691 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1692 		pend_list = &dcc->pend_list[i];
1693 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1694 			f2fs_bug_on(sbi, dc->state != D_PREP);
1695 			__remove_discard_cmd(sbi, dc);
1696 			dropped = true;
1697 		}
1698 	}
1699 	mutex_unlock(&dcc->cmd_lock);
1700 
1701 	return dropped;
1702 }
1703 
1704 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1705 {
1706 	__drop_discard_cmd(sbi);
1707 }
1708 
1709 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1710 							struct discard_cmd *dc)
1711 {
1712 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1713 	unsigned int len = 0;
1714 
1715 	wait_for_completion_io(&dc->wait);
1716 	mutex_lock(&dcc->cmd_lock);
1717 	f2fs_bug_on(sbi, dc->state != D_DONE);
1718 	dc->ref--;
1719 	if (!dc->ref) {
1720 		if (!dc->error)
1721 			len = dc->di.len;
1722 		__remove_discard_cmd(sbi, dc);
1723 	}
1724 	mutex_unlock(&dcc->cmd_lock);
1725 
1726 	return len;
1727 }
1728 
1729 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1730 						struct discard_policy *dpolicy,
1731 						block_t start, block_t end)
1732 {
1733 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1734 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1735 					&(dcc->fstrim_list) : &(dcc->wait_list);
1736 	struct discard_cmd *dc = NULL, *iter, *tmp;
1737 	unsigned int trimmed = 0;
1738 
1739 next:
1740 	dc = NULL;
1741 
1742 	mutex_lock(&dcc->cmd_lock);
1743 	list_for_each_entry_safe(iter, tmp, wait_list, list) {
1744 		if (iter->di.lstart + iter->di.len <= start ||
1745 					end <= iter->di.lstart)
1746 			continue;
1747 		if (iter->di.len < dpolicy->granularity)
1748 			continue;
1749 		if (iter->state == D_DONE && !iter->ref) {
1750 			wait_for_completion_io(&iter->wait);
1751 			if (!iter->error)
1752 				trimmed += iter->di.len;
1753 			__remove_discard_cmd(sbi, iter);
1754 		} else {
1755 			iter->ref++;
1756 			dc = iter;
1757 			break;
1758 		}
1759 	}
1760 	mutex_unlock(&dcc->cmd_lock);
1761 
1762 	if (dc) {
1763 		trimmed += __wait_one_discard_bio(sbi, dc);
1764 		goto next;
1765 	}
1766 
1767 	return trimmed;
1768 }
1769 
1770 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1771 						struct discard_policy *dpolicy)
1772 {
1773 	struct discard_policy dp;
1774 	unsigned int discard_blks;
1775 
1776 	if (dpolicy)
1777 		return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1778 
1779 	/* wait all */
1780 	__init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY);
1781 	discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1782 	__init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY);
1783 	discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1784 
1785 	return discard_blks;
1786 }
1787 
1788 /* This should be covered by global mutex, &sit_i->sentry_lock */
1789 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1790 {
1791 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1792 	struct discard_cmd *dc;
1793 	bool need_wait = false;
1794 
1795 	mutex_lock(&dcc->cmd_lock);
1796 	dc = __lookup_discard_cmd(sbi, blkaddr);
1797 #ifdef CONFIG_BLK_DEV_ZONED
1798 	if (dc && f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) {
1799 		int devi = f2fs_bdev_index(sbi, dc->bdev);
1800 
1801 		if (devi < 0) {
1802 			mutex_unlock(&dcc->cmd_lock);
1803 			return;
1804 		}
1805 
1806 		if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1807 			/* force submit zone reset */
1808 			if (dc->state == D_PREP)
1809 				__submit_zone_reset_cmd(sbi, dc, REQ_SYNC,
1810 							&dcc->wait_list, NULL);
1811 			dc->ref++;
1812 			mutex_unlock(&dcc->cmd_lock);
1813 			/* wait zone reset */
1814 			__wait_one_discard_bio(sbi, dc);
1815 			return;
1816 		}
1817 	}
1818 #endif
1819 	if (dc) {
1820 		if (dc->state == D_PREP) {
1821 			__punch_discard_cmd(sbi, dc, blkaddr);
1822 		} else {
1823 			dc->ref++;
1824 			need_wait = true;
1825 		}
1826 	}
1827 	mutex_unlock(&dcc->cmd_lock);
1828 
1829 	if (need_wait)
1830 		__wait_one_discard_bio(sbi, dc);
1831 }
1832 
1833 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1834 {
1835 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1836 
1837 	if (dcc && dcc->f2fs_issue_discard) {
1838 		struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1839 
1840 		dcc->f2fs_issue_discard = NULL;
1841 		kthread_stop(discard_thread);
1842 	}
1843 }
1844 
1845 /**
1846  * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT
1847  * @sbi: the f2fs_sb_info data for discard cmd to issue
1848  *
1849  * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped
1850  *
1851  * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false.
1852  */
1853 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1854 {
1855 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1856 	struct discard_policy dpolicy;
1857 	bool dropped;
1858 
1859 	if (!atomic_read(&dcc->discard_cmd_cnt))
1860 		return true;
1861 
1862 	__init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1863 					dcc->discard_granularity);
1864 	__issue_discard_cmd(sbi, &dpolicy);
1865 	dropped = __drop_discard_cmd(sbi);
1866 
1867 	/* just to make sure there is no pending discard commands */
1868 	__wait_all_discard_cmd(sbi, NULL);
1869 
1870 	f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1871 	return !dropped;
1872 }
1873 
1874 static int issue_discard_thread(void *data)
1875 {
1876 	struct f2fs_sb_info *sbi = data;
1877 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1878 	wait_queue_head_t *q = &dcc->discard_wait_queue;
1879 	struct discard_policy dpolicy;
1880 	unsigned int wait_ms = dcc->min_discard_issue_time;
1881 	int issued;
1882 
1883 	set_freezable();
1884 
1885 	do {
1886 		wait_event_interruptible_timeout(*q,
1887 				kthread_should_stop() || freezing(current) ||
1888 				dcc->discard_wake,
1889 				msecs_to_jiffies(wait_ms));
1890 
1891 		if (sbi->gc_mode == GC_URGENT_HIGH ||
1892 			!f2fs_available_free_memory(sbi, DISCARD_CACHE))
1893 			__init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE,
1894 						MIN_DISCARD_GRANULARITY);
1895 		else
1896 			__init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1897 						dcc->discard_granularity);
1898 
1899 		if (dcc->discard_wake)
1900 			dcc->discard_wake = false;
1901 
1902 		/* clean up pending candidates before going to sleep */
1903 		if (atomic_read(&dcc->queued_discard))
1904 			__wait_all_discard_cmd(sbi, NULL);
1905 
1906 		if (try_to_freeze())
1907 			continue;
1908 		if (f2fs_readonly(sbi->sb))
1909 			continue;
1910 		if (kthread_should_stop())
1911 			return 0;
1912 		if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) ||
1913 			!atomic_read(&dcc->discard_cmd_cnt)) {
1914 			wait_ms = dpolicy.max_interval;
1915 			continue;
1916 		}
1917 
1918 		sb_start_intwrite(sbi->sb);
1919 
1920 		issued = __issue_discard_cmd(sbi, &dpolicy);
1921 		if (issued > 0) {
1922 			__wait_all_discard_cmd(sbi, &dpolicy);
1923 			wait_ms = dpolicy.min_interval;
1924 		} else if (issued == -1) {
1925 			wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1926 			if (!wait_ms)
1927 				wait_ms = dpolicy.mid_interval;
1928 		} else {
1929 			wait_ms = dpolicy.max_interval;
1930 		}
1931 		if (!atomic_read(&dcc->discard_cmd_cnt))
1932 			wait_ms = dpolicy.max_interval;
1933 
1934 		sb_end_intwrite(sbi->sb);
1935 
1936 	} while (!kthread_should_stop());
1937 	return 0;
1938 }
1939 
1940 #ifdef CONFIG_BLK_DEV_ZONED
1941 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1942 		struct block_device *bdev, block_t blkstart, block_t blklen)
1943 {
1944 	sector_t sector, nr_sects;
1945 	block_t lblkstart = blkstart;
1946 	int devi = 0;
1947 	u64 remainder = 0;
1948 
1949 	if (f2fs_is_multi_device(sbi)) {
1950 		devi = f2fs_target_device_index(sbi, blkstart);
1951 		if (blkstart < FDEV(devi).start_blk ||
1952 		    blkstart > FDEV(devi).end_blk) {
1953 			f2fs_err(sbi, "Invalid block %x", blkstart);
1954 			return -EIO;
1955 		}
1956 		blkstart -= FDEV(devi).start_blk;
1957 	}
1958 
1959 	/* For sequential zones, reset the zone write pointer */
1960 	if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1961 		sector = SECTOR_FROM_BLOCK(blkstart);
1962 		nr_sects = SECTOR_FROM_BLOCK(blklen);
1963 		div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder);
1964 
1965 		if (remainder || nr_sects != bdev_zone_sectors(bdev)) {
1966 			f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1967 				 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1968 				 blkstart, blklen);
1969 			return -EIO;
1970 		}
1971 
1972 		if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) {
1973 			trace_f2fs_issue_reset_zone(bdev, blkstart);
1974 			return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1975 						sector, nr_sects, GFP_NOFS);
1976 		}
1977 
1978 		__queue_zone_reset_cmd(sbi, bdev, blkstart, lblkstart, blklen);
1979 		return 0;
1980 	}
1981 
1982 	/* For conventional zones, use regular discard if supported */
1983 	__queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1984 	return 0;
1985 }
1986 #endif
1987 
1988 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1989 		struct block_device *bdev, block_t blkstart, block_t blklen)
1990 {
1991 #ifdef CONFIG_BLK_DEV_ZONED
1992 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1993 		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1994 #endif
1995 	__queue_discard_cmd(sbi, bdev, blkstart, blklen);
1996 	return 0;
1997 }
1998 
1999 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
2000 				block_t blkstart, block_t blklen)
2001 {
2002 	sector_t start = blkstart, len = 0;
2003 	struct block_device *bdev;
2004 	struct seg_entry *se;
2005 	unsigned int offset;
2006 	block_t i;
2007 	int err = 0;
2008 
2009 	bdev = f2fs_target_device(sbi, blkstart, NULL);
2010 
2011 	for (i = blkstart; i < blkstart + blklen; i++, len++) {
2012 		if (i != start) {
2013 			struct block_device *bdev2 =
2014 				f2fs_target_device(sbi, i, NULL);
2015 
2016 			if (bdev2 != bdev) {
2017 				err = __issue_discard_async(sbi, bdev,
2018 						start, len);
2019 				if (err)
2020 					return err;
2021 				bdev = bdev2;
2022 				start = i;
2023 				len = 0;
2024 			}
2025 		}
2026 
2027 		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
2028 		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
2029 
2030 		if (f2fs_block_unit_discard(sbi) &&
2031 				!f2fs_test_and_set_bit(offset, se->discard_map))
2032 			sbi->discard_blks--;
2033 	}
2034 
2035 	if (len)
2036 		err = __issue_discard_async(sbi, bdev, start, len);
2037 	return err;
2038 }
2039 
2040 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
2041 							bool check_only)
2042 {
2043 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2044 	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
2045 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2046 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2047 	unsigned long *discard_map = (unsigned long *)se->discard_map;
2048 	unsigned long *dmap = SIT_I(sbi)->tmp_map;
2049 	unsigned int start = 0, end = -1;
2050 	bool force = (cpc->reason & CP_DISCARD);
2051 	struct discard_entry *de = NULL;
2052 	struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
2053 	int i;
2054 
2055 	if (se->valid_blocks == BLKS_PER_SEG(sbi) ||
2056 	    !f2fs_hw_support_discard(sbi) ||
2057 	    !f2fs_block_unit_discard(sbi))
2058 		return false;
2059 
2060 	if (!force) {
2061 		if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
2062 			SM_I(sbi)->dcc_info->nr_discards >=
2063 				SM_I(sbi)->dcc_info->max_discards)
2064 			return false;
2065 	}
2066 
2067 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
2068 	for (i = 0; i < entries; i++)
2069 		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
2070 				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
2071 
2072 	while (force || SM_I(sbi)->dcc_info->nr_discards <=
2073 				SM_I(sbi)->dcc_info->max_discards) {
2074 		start = __find_rev_next_bit(dmap, BLKS_PER_SEG(sbi), end + 1);
2075 		if (start >= BLKS_PER_SEG(sbi))
2076 			break;
2077 
2078 		end = __find_rev_next_zero_bit(dmap,
2079 						BLKS_PER_SEG(sbi), start + 1);
2080 		if (force && start && end != BLKS_PER_SEG(sbi) &&
2081 		    (end - start) < cpc->trim_minlen)
2082 			continue;
2083 
2084 		if (check_only)
2085 			return true;
2086 
2087 		if (!de) {
2088 			de = f2fs_kmem_cache_alloc(discard_entry_slab,
2089 						GFP_F2FS_ZERO, true, NULL);
2090 			de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
2091 			list_add_tail(&de->list, head);
2092 		}
2093 
2094 		for (i = start; i < end; i++)
2095 			__set_bit_le(i, (void *)de->discard_map);
2096 
2097 		SM_I(sbi)->dcc_info->nr_discards += end - start;
2098 	}
2099 	return false;
2100 }
2101 
2102 static void release_discard_addr(struct discard_entry *entry)
2103 {
2104 	list_del(&entry->list);
2105 	kmem_cache_free(discard_entry_slab, entry);
2106 }
2107 
2108 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
2109 {
2110 	struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
2111 	struct discard_entry *entry, *this;
2112 
2113 	/* drop caches */
2114 	list_for_each_entry_safe(entry, this, head, list)
2115 		release_discard_addr(entry);
2116 }
2117 
2118 /*
2119  * Should call f2fs_clear_prefree_segments after checkpoint is done.
2120  */
2121 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
2122 {
2123 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2124 	unsigned int segno;
2125 
2126 	mutex_lock(&dirty_i->seglist_lock);
2127 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
2128 		__set_test_and_free(sbi, segno, false);
2129 	mutex_unlock(&dirty_i->seglist_lock);
2130 }
2131 
2132 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
2133 						struct cp_control *cpc)
2134 {
2135 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2136 	struct list_head *head = &dcc->entry_list;
2137 	struct discard_entry *entry, *this;
2138 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2139 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
2140 	unsigned int start = 0, end = -1;
2141 	unsigned int secno, start_segno;
2142 	bool force = (cpc->reason & CP_DISCARD);
2143 	bool section_alignment = F2FS_OPTION(sbi).discard_unit ==
2144 						DISCARD_UNIT_SECTION;
2145 
2146 	if (f2fs_lfs_mode(sbi) && __is_large_section(sbi))
2147 		section_alignment = true;
2148 
2149 	mutex_lock(&dirty_i->seglist_lock);
2150 
2151 	while (1) {
2152 		int i;
2153 
2154 		if (section_alignment && end != -1)
2155 			end--;
2156 		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
2157 		if (start >= MAIN_SEGS(sbi))
2158 			break;
2159 		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
2160 								start + 1);
2161 
2162 		if (section_alignment) {
2163 			start = rounddown(start, SEGS_PER_SEC(sbi));
2164 			end = roundup(end, SEGS_PER_SEC(sbi));
2165 		}
2166 
2167 		for (i = start; i < end; i++) {
2168 			if (test_and_clear_bit(i, prefree_map))
2169 				dirty_i->nr_dirty[PRE]--;
2170 		}
2171 
2172 		if (!f2fs_realtime_discard_enable(sbi))
2173 			continue;
2174 
2175 		if (force && start >= cpc->trim_start &&
2176 					(end - 1) <= cpc->trim_end)
2177 			continue;
2178 
2179 		/* Should cover 2MB zoned device for zone-based reset */
2180 		if (!f2fs_sb_has_blkzoned(sbi) &&
2181 		    (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) {
2182 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2183 				(end - start) << sbi->log_blocks_per_seg);
2184 			continue;
2185 		}
2186 next:
2187 		secno = GET_SEC_FROM_SEG(sbi, start);
2188 		start_segno = GET_SEG_FROM_SEC(sbi, secno);
2189 		if (!IS_CURSEC(sbi, secno) &&
2190 			!get_valid_blocks(sbi, start, true))
2191 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2192 						BLKS_PER_SEC(sbi));
2193 
2194 		start = start_segno + SEGS_PER_SEC(sbi);
2195 		if (start < end)
2196 			goto next;
2197 		else
2198 			end = start - 1;
2199 	}
2200 	mutex_unlock(&dirty_i->seglist_lock);
2201 
2202 	if (!f2fs_block_unit_discard(sbi))
2203 		goto wakeup;
2204 
2205 	/* send small discards */
2206 	list_for_each_entry_safe(entry, this, head, list) {
2207 		unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2208 		bool is_valid = test_bit_le(0, entry->discard_map);
2209 
2210 find_next:
2211 		if (is_valid) {
2212 			next_pos = find_next_zero_bit_le(entry->discard_map,
2213 						BLKS_PER_SEG(sbi), cur_pos);
2214 			len = next_pos - cur_pos;
2215 
2216 			if (f2fs_sb_has_blkzoned(sbi) ||
2217 			    (force && len < cpc->trim_minlen))
2218 				goto skip;
2219 
2220 			f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2221 									len);
2222 			total_len += len;
2223 		} else {
2224 			next_pos = find_next_bit_le(entry->discard_map,
2225 						BLKS_PER_SEG(sbi), cur_pos);
2226 		}
2227 skip:
2228 		cur_pos = next_pos;
2229 		is_valid = !is_valid;
2230 
2231 		if (cur_pos < BLKS_PER_SEG(sbi))
2232 			goto find_next;
2233 
2234 		release_discard_addr(entry);
2235 		dcc->nr_discards -= total_len;
2236 	}
2237 
2238 wakeup:
2239 	wake_up_discard_thread(sbi, false);
2240 }
2241 
2242 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi)
2243 {
2244 	dev_t dev = sbi->sb->s_bdev->bd_dev;
2245 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2246 	int err = 0;
2247 
2248 	if (!f2fs_realtime_discard_enable(sbi))
2249 		return 0;
2250 
2251 	dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2252 				"f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2253 	if (IS_ERR(dcc->f2fs_issue_discard)) {
2254 		err = PTR_ERR(dcc->f2fs_issue_discard);
2255 		dcc->f2fs_issue_discard = NULL;
2256 	}
2257 
2258 	return err;
2259 }
2260 
2261 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2262 {
2263 	struct discard_cmd_control *dcc;
2264 	int err = 0, i;
2265 
2266 	if (SM_I(sbi)->dcc_info) {
2267 		dcc = SM_I(sbi)->dcc_info;
2268 		goto init_thread;
2269 	}
2270 
2271 	dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2272 	if (!dcc)
2273 		return -ENOMEM;
2274 
2275 	dcc->discard_io_aware_gran = MAX_PLIST_NUM;
2276 	dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2277 	dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY;
2278 	if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT)
2279 		dcc->discard_granularity = BLKS_PER_SEG(sbi);
2280 	else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION)
2281 		dcc->discard_granularity = BLKS_PER_SEC(sbi);
2282 
2283 	INIT_LIST_HEAD(&dcc->entry_list);
2284 	for (i = 0; i < MAX_PLIST_NUM; i++)
2285 		INIT_LIST_HEAD(&dcc->pend_list[i]);
2286 	INIT_LIST_HEAD(&dcc->wait_list);
2287 	INIT_LIST_HEAD(&dcc->fstrim_list);
2288 	mutex_init(&dcc->cmd_lock);
2289 	atomic_set(&dcc->issued_discard, 0);
2290 	atomic_set(&dcc->queued_discard, 0);
2291 	atomic_set(&dcc->discard_cmd_cnt, 0);
2292 	dcc->nr_discards = 0;
2293 	dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2294 	dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST;
2295 	dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME;
2296 	dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME;
2297 	dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME;
2298 	dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL;
2299 	dcc->undiscard_blks = 0;
2300 	dcc->next_pos = 0;
2301 	dcc->root = RB_ROOT_CACHED;
2302 	dcc->rbtree_check = false;
2303 
2304 	init_waitqueue_head(&dcc->discard_wait_queue);
2305 	SM_I(sbi)->dcc_info = dcc;
2306 init_thread:
2307 	err = f2fs_start_discard_thread(sbi);
2308 	if (err) {
2309 		kfree(dcc);
2310 		SM_I(sbi)->dcc_info = NULL;
2311 	}
2312 
2313 	return err;
2314 }
2315 
2316 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2317 {
2318 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2319 
2320 	if (!dcc)
2321 		return;
2322 
2323 	f2fs_stop_discard_thread(sbi);
2324 
2325 	/*
2326 	 * Recovery can cache discard commands, so in error path of
2327 	 * fill_super(), it needs to give a chance to handle them.
2328 	 */
2329 	f2fs_issue_discard_timeout(sbi);
2330 
2331 	kfree(dcc);
2332 	SM_I(sbi)->dcc_info = NULL;
2333 }
2334 
2335 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2336 {
2337 	struct sit_info *sit_i = SIT_I(sbi);
2338 
2339 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2340 		sit_i->dirty_sentries++;
2341 		return false;
2342 	}
2343 
2344 	return true;
2345 }
2346 
2347 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2348 					unsigned int segno, int modified)
2349 {
2350 	struct seg_entry *se = get_seg_entry(sbi, segno);
2351 
2352 	se->type = type;
2353 	if (modified)
2354 		__mark_sit_entry_dirty(sbi, segno);
2355 }
2356 
2357 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2358 								block_t blkaddr)
2359 {
2360 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
2361 
2362 	if (segno == NULL_SEGNO)
2363 		return 0;
2364 	return get_seg_entry(sbi, segno)->mtime;
2365 }
2366 
2367 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2368 						unsigned long long old_mtime)
2369 {
2370 	struct seg_entry *se;
2371 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
2372 	unsigned long long ctime = get_mtime(sbi, false);
2373 	unsigned long long mtime = old_mtime ? old_mtime : ctime;
2374 
2375 	if (segno == NULL_SEGNO)
2376 		return;
2377 
2378 	se = get_seg_entry(sbi, segno);
2379 
2380 	if (!se->mtime)
2381 		se->mtime = mtime;
2382 	else
2383 		se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2384 						se->valid_blocks + 1);
2385 
2386 	if (ctime > SIT_I(sbi)->max_mtime)
2387 		SIT_I(sbi)->max_mtime = ctime;
2388 }
2389 
2390 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2391 {
2392 	struct seg_entry *se;
2393 	unsigned int segno, offset;
2394 	long int new_vblocks;
2395 	bool exist;
2396 #ifdef CONFIG_F2FS_CHECK_FS
2397 	bool mir_exist;
2398 #endif
2399 
2400 	segno = GET_SEGNO(sbi, blkaddr);
2401 
2402 	se = get_seg_entry(sbi, segno);
2403 	new_vblocks = se->valid_blocks + del;
2404 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2405 
2406 	f2fs_bug_on(sbi, (new_vblocks < 0 ||
2407 			(new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2408 
2409 	se->valid_blocks = new_vblocks;
2410 
2411 	/* Update valid block bitmap */
2412 	if (del > 0) {
2413 		exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2414 #ifdef CONFIG_F2FS_CHECK_FS
2415 		mir_exist = f2fs_test_and_set_bit(offset,
2416 						se->cur_valid_map_mir);
2417 		if (unlikely(exist != mir_exist)) {
2418 			f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2419 				 blkaddr, exist);
2420 			f2fs_bug_on(sbi, 1);
2421 		}
2422 #endif
2423 		if (unlikely(exist)) {
2424 			f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2425 				 blkaddr);
2426 			f2fs_bug_on(sbi, 1);
2427 			se->valid_blocks--;
2428 			del = 0;
2429 		}
2430 
2431 		if (f2fs_block_unit_discard(sbi) &&
2432 				!f2fs_test_and_set_bit(offset, se->discard_map))
2433 			sbi->discard_blks--;
2434 
2435 		/*
2436 		 * SSR should never reuse block which is checkpointed
2437 		 * or newly invalidated.
2438 		 */
2439 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2440 			if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2441 				se->ckpt_valid_blocks++;
2442 		}
2443 	} else {
2444 		exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2445 #ifdef CONFIG_F2FS_CHECK_FS
2446 		mir_exist = f2fs_test_and_clear_bit(offset,
2447 						se->cur_valid_map_mir);
2448 		if (unlikely(exist != mir_exist)) {
2449 			f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2450 				 blkaddr, exist);
2451 			f2fs_bug_on(sbi, 1);
2452 		}
2453 #endif
2454 		if (unlikely(!exist)) {
2455 			f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2456 				 blkaddr);
2457 			f2fs_bug_on(sbi, 1);
2458 			se->valid_blocks++;
2459 			del = 0;
2460 		} else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2461 			/*
2462 			 * If checkpoints are off, we must not reuse data that
2463 			 * was used in the previous checkpoint. If it was used
2464 			 * before, we must track that to know how much space we
2465 			 * really have.
2466 			 */
2467 			if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2468 				spin_lock(&sbi->stat_lock);
2469 				sbi->unusable_block_count++;
2470 				spin_unlock(&sbi->stat_lock);
2471 			}
2472 		}
2473 
2474 		if (f2fs_block_unit_discard(sbi) &&
2475 			f2fs_test_and_clear_bit(offset, se->discard_map))
2476 			sbi->discard_blks++;
2477 	}
2478 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2479 		se->ckpt_valid_blocks += del;
2480 
2481 	__mark_sit_entry_dirty(sbi, segno);
2482 
2483 	/* update total number of valid blocks to be written in ckpt area */
2484 	SIT_I(sbi)->written_valid_blocks += del;
2485 
2486 	if (__is_large_section(sbi))
2487 		get_sec_entry(sbi, segno)->valid_blocks += del;
2488 }
2489 
2490 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2491 {
2492 	unsigned int segno = GET_SEGNO(sbi, addr);
2493 	struct sit_info *sit_i = SIT_I(sbi);
2494 
2495 	f2fs_bug_on(sbi, addr == NULL_ADDR);
2496 	if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2497 		return;
2498 
2499 	f2fs_invalidate_internal_cache(sbi, addr);
2500 
2501 	/* add it into sit main buffer */
2502 	down_write(&sit_i->sentry_lock);
2503 
2504 	update_segment_mtime(sbi, addr, 0);
2505 	update_sit_entry(sbi, addr, -1);
2506 
2507 	/* add it into dirty seglist */
2508 	locate_dirty_segment(sbi, segno);
2509 
2510 	up_write(&sit_i->sentry_lock);
2511 }
2512 
2513 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2514 {
2515 	struct sit_info *sit_i = SIT_I(sbi);
2516 	unsigned int segno, offset;
2517 	struct seg_entry *se;
2518 	bool is_cp = false;
2519 
2520 	if (!__is_valid_data_blkaddr(blkaddr))
2521 		return true;
2522 
2523 	down_read(&sit_i->sentry_lock);
2524 
2525 	segno = GET_SEGNO(sbi, blkaddr);
2526 	se = get_seg_entry(sbi, segno);
2527 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2528 
2529 	if (f2fs_test_bit(offset, se->ckpt_valid_map))
2530 		is_cp = true;
2531 
2532 	up_read(&sit_i->sentry_lock);
2533 
2534 	return is_cp;
2535 }
2536 
2537 static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type)
2538 {
2539 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2540 
2541 	if (sbi->ckpt->alloc_type[type] == SSR)
2542 		return BLKS_PER_SEG(sbi);
2543 	return curseg->next_blkoff;
2544 }
2545 
2546 /*
2547  * Calculate the number of current summary pages for writing
2548  */
2549 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2550 {
2551 	int valid_sum_count = 0;
2552 	int i, sum_in_page;
2553 
2554 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2555 		if (sbi->ckpt->alloc_type[i] != SSR && for_ra)
2556 			valid_sum_count +=
2557 				le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2558 		else
2559 			valid_sum_count += f2fs_curseg_valid_blocks(sbi, i);
2560 	}
2561 
2562 	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2563 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2564 	if (valid_sum_count <= sum_in_page)
2565 		return 1;
2566 	else if ((valid_sum_count - sum_in_page) <=
2567 		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2568 		return 2;
2569 	return 3;
2570 }
2571 
2572 /*
2573  * Caller should put this summary page
2574  */
2575 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2576 {
2577 	if (unlikely(f2fs_cp_error(sbi)))
2578 		return ERR_PTR(-EIO);
2579 	return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2580 }
2581 
2582 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2583 					void *src, block_t blk_addr)
2584 {
2585 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2586 
2587 	memcpy(page_address(page), src, PAGE_SIZE);
2588 	set_page_dirty(page);
2589 	f2fs_put_page(page, 1);
2590 }
2591 
2592 static void write_sum_page(struct f2fs_sb_info *sbi,
2593 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
2594 {
2595 	f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2596 }
2597 
2598 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2599 						int type, block_t blk_addr)
2600 {
2601 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2602 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2603 	struct f2fs_summary_block *src = curseg->sum_blk;
2604 	struct f2fs_summary_block *dst;
2605 
2606 	dst = (struct f2fs_summary_block *)page_address(page);
2607 	memset(dst, 0, PAGE_SIZE);
2608 
2609 	mutex_lock(&curseg->curseg_mutex);
2610 
2611 	down_read(&curseg->journal_rwsem);
2612 	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2613 	up_read(&curseg->journal_rwsem);
2614 
2615 	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2616 	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2617 
2618 	mutex_unlock(&curseg->curseg_mutex);
2619 
2620 	set_page_dirty(page);
2621 	f2fs_put_page(page, 1);
2622 }
2623 
2624 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2625 				struct curseg_info *curseg, int type)
2626 {
2627 	unsigned int segno = curseg->segno + 1;
2628 	struct free_segmap_info *free_i = FREE_I(sbi);
2629 
2630 	if (segno < MAIN_SEGS(sbi) && segno % SEGS_PER_SEC(sbi))
2631 		return !test_bit(segno, free_i->free_segmap);
2632 	return 0;
2633 }
2634 
2635 /*
2636  * Find a new segment from the free segments bitmap to right order
2637  * This function should be returned with success, otherwise BUG
2638  */
2639 static void get_new_segment(struct f2fs_sb_info *sbi,
2640 			unsigned int *newseg, bool new_sec, bool pinning)
2641 {
2642 	struct free_segmap_info *free_i = FREE_I(sbi);
2643 	unsigned int segno, secno, zoneno;
2644 	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2645 	unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2646 	unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2647 	bool init = true;
2648 	int i;
2649 
2650 	spin_lock(&free_i->segmap_lock);
2651 
2652 	if (!new_sec && ((*newseg + 1) % SEGS_PER_SEC(sbi))) {
2653 		segno = find_next_zero_bit(free_i->free_segmap,
2654 			GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2655 		if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2656 			goto got_it;
2657 	}
2658 
2659 	/*
2660 	 * If we format f2fs on zoned storage, let's try to get pinned sections
2661 	 * from beginning of the storage, which should be a conventional one.
2662 	 */
2663 	if (f2fs_sb_has_blkzoned(sbi)) {
2664 		segno = pinning ? 0 : max(first_zoned_segno(sbi), *newseg);
2665 		hint = GET_SEC_FROM_SEG(sbi, segno);
2666 	}
2667 
2668 find_other_zone:
2669 	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2670 	if (secno >= MAIN_SECS(sbi)) {
2671 		secno = find_first_zero_bit(free_i->free_secmap,
2672 							MAIN_SECS(sbi));
2673 		f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2674 	}
2675 	segno = GET_SEG_FROM_SEC(sbi, secno);
2676 	zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2677 
2678 	/* give up on finding another zone */
2679 	if (!init)
2680 		goto got_it;
2681 	if (sbi->secs_per_zone == 1)
2682 		goto got_it;
2683 	if (zoneno == old_zoneno)
2684 		goto got_it;
2685 	for (i = 0; i < NR_CURSEG_TYPE; i++)
2686 		if (CURSEG_I(sbi, i)->zone == zoneno)
2687 			break;
2688 
2689 	if (i < NR_CURSEG_TYPE) {
2690 		/* zone is in user, try another */
2691 		if (zoneno + 1 >= total_zones)
2692 			hint = 0;
2693 		else
2694 			hint = (zoneno + 1) * sbi->secs_per_zone;
2695 		init = false;
2696 		goto find_other_zone;
2697 	}
2698 got_it:
2699 	/* set it as dirty segment in free segmap */
2700 	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2701 	__set_inuse(sbi, segno);
2702 	*newseg = segno;
2703 	spin_unlock(&free_i->segmap_lock);
2704 }
2705 
2706 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2707 {
2708 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2709 	struct summary_footer *sum_footer;
2710 	unsigned short seg_type = curseg->seg_type;
2711 
2712 	curseg->inited = true;
2713 	curseg->segno = curseg->next_segno;
2714 	curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2715 	curseg->next_blkoff = 0;
2716 	curseg->next_segno = NULL_SEGNO;
2717 
2718 	sum_footer = &(curseg->sum_blk->footer);
2719 	memset(sum_footer, 0, sizeof(struct summary_footer));
2720 
2721 	sanity_check_seg_type(sbi, seg_type);
2722 
2723 	if (IS_DATASEG(seg_type))
2724 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2725 	if (IS_NODESEG(seg_type))
2726 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2727 	__set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2728 }
2729 
2730 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2731 {
2732 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2733 	unsigned short seg_type = curseg->seg_type;
2734 
2735 	sanity_check_seg_type(sbi, seg_type);
2736 	if (f2fs_need_rand_seg(sbi))
2737 		return get_random_u32_below(MAIN_SECS(sbi) * SEGS_PER_SEC(sbi));
2738 
2739 	if (__is_large_section(sbi))
2740 		return curseg->segno;
2741 
2742 	/* inmem log may not locate on any segment after mount */
2743 	if (!curseg->inited)
2744 		return 0;
2745 
2746 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2747 		return 0;
2748 
2749 	if (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type))
2750 		return 0;
2751 
2752 	if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2753 		return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2754 
2755 	/* find segments from 0 to reuse freed segments */
2756 	if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2757 		return 0;
2758 
2759 	return curseg->segno;
2760 }
2761 
2762 /*
2763  * Allocate a current working segment.
2764  * This function always allocates a free segment in LFS manner.
2765  */
2766 static int new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2767 {
2768 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2769 	unsigned int segno = curseg->segno;
2770 	bool pinning = type == CURSEG_COLD_DATA_PINNED;
2771 
2772 	if (curseg->inited)
2773 		write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, segno));
2774 
2775 	segno = __get_next_segno(sbi, type);
2776 	get_new_segment(sbi, &segno, new_sec, pinning);
2777 	if (new_sec && pinning &&
2778 	    !f2fs_valid_pinned_area(sbi, START_BLOCK(sbi, segno))) {
2779 		__set_free(sbi, segno);
2780 		return -EAGAIN;
2781 	}
2782 
2783 	curseg->next_segno = segno;
2784 	reset_curseg(sbi, type, 1);
2785 	curseg->alloc_type = LFS;
2786 	if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
2787 		curseg->fragment_remained_chunk =
2788 				get_random_u32_inclusive(1, sbi->max_fragment_chunk);
2789 	return 0;
2790 }
2791 
2792 static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2793 					int segno, block_t start)
2794 {
2795 	struct seg_entry *se = get_seg_entry(sbi, segno);
2796 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2797 	unsigned long *target_map = SIT_I(sbi)->tmp_map;
2798 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2799 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2800 	int i;
2801 
2802 	for (i = 0; i < entries; i++)
2803 		target_map[i] = ckpt_map[i] | cur_map[i];
2804 
2805 	return __find_rev_next_zero_bit(target_map, BLKS_PER_SEG(sbi), start);
2806 }
2807 
2808 static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi,
2809 		struct curseg_info *seg)
2810 {
2811 	return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1);
2812 }
2813 
2814 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2815 {
2816 	return __next_free_blkoff(sbi, segno, 0) < BLKS_PER_SEG(sbi);
2817 }
2818 
2819 /*
2820  * This function always allocates a used segment(from dirty seglist) by SSR
2821  * manner, so it should recover the existing segment information of valid blocks
2822  */
2823 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2824 {
2825 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2826 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2827 	unsigned int new_segno = curseg->next_segno;
2828 	struct f2fs_summary_block *sum_node;
2829 	struct page *sum_page;
2830 
2831 	write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, curseg->segno));
2832 
2833 	__set_test_and_inuse(sbi, new_segno);
2834 
2835 	mutex_lock(&dirty_i->seglist_lock);
2836 	__remove_dirty_segment(sbi, new_segno, PRE);
2837 	__remove_dirty_segment(sbi, new_segno, DIRTY);
2838 	mutex_unlock(&dirty_i->seglist_lock);
2839 
2840 	reset_curseg(sbi, type, 1);
2841 	curseg->alloc_type = SSR;
2842 	curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
2843 
2844 	sum_page = f2fs_get_sum_page(sbi, new_segno);
2845 	if (IS_ERR(sum_page)) {
2846 		/* GC won't be able to use stale summary pages by cp_error */
2847 		memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2848 		return;
2849 	}
2850 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2851 	memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2852 	f2fs_put_page(sum_page, 1);
2853 }
2854 
2855 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2856 				int alloc_mode, unsigned long long age);
2857 
2858 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2859 					int target_type, int alloc_mode,
2860 					unsigned long long age)
2861 {
2862 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2863 
2864 	curseg->seg_type = target_type;
2865 
2866 	if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2867 		struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2868 
2869 		curseg->seg_type = se->type;
2870 		change_curseg(sbi, type);
2871 	} else {
2872 		/* allocate cold segment by default */
2873 		curseg->seg_type = CURSEG_COLD_DATA;
2874 		new_curseg(sbi, type, true);
2875 	}
2876 	stat_inc_seg_type(sbi, curseg);
2877 }
2878 
2879 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
2880 {
2881 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2882 
2883 	if (!sbi->am.atgc_enabled)
2884 		return;
2885 
2886 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
2887 
2888 	mutex_lock(&curseg->curseg_mutex);
2889 	down_write(&SIT_I(sbi)->sentry_lock);
2890 
2891 	get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0);
2892 
2893 	up_write(&SIT_I(sbi)->sentry_lock);
2894 	mutex_unlock(&curseg->curseg_mutex);
2895 
2896 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
2897 
2898 }
2899 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2900 {
2901 	__f2fs_init_atgc_curseg(sbi);
2902 }
2903 
2904 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2905 {
2906 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2907 
2908 	mutex_lock(&curseg->curseg_mutex);
2909 	if (!curseg->inited)
2910 		goto out;
2911 
2912 	if (get_valid_blocks(sbi, curseg->segno, false)) {
2913 		write_sum_page(sbi, curseg->sum_blk,
2914 				GET_SUM_BLOCK(sbi, curseg->segno));
2915 	} else {
2916 		mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2917 		__set_test_and_free(sbi, curseg->segno, true);
2918 		mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2919 	}
2920 out:
2921 	mutex_unlock(&curseg->curseg_mutex);
2922 }
2923 
2924 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
2925 {
2926 	__f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2927 
2928 	if (sbi->am.atgc_enabled)
2929 		__f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2930 }
2931 
2932 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2933 {
2934 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2935 
2936 	mutex_lock(&curseg->curseg_mutex);
2937 	if (!curseg->inited)
2938 		goto out;
2939 	if (get_valid_blocks(sbi, curseg->segno, false))
2940 		goto out;
2941 
2942 	mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2943 	__set_test_and_inuse(sbi, curseg->segno);
2944 	mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2945 out:
2946 	mutex_unlock(&curseg->curseg_mutex);
2947 }
2948 
2949 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
2950 {
2951 	__f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2952 
2953 	if (sbi->am.atgc_enabled)
2954 		__f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2955 }
2956 
2957 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2958 				int alloc_mode, unsigned long long age)
2959 {
2960 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2961 	unsigned segno = NULL_SEGNO;
2962 	unsigned short seg_type = curseg->seg_type;
2963 	int i, cnt;
2964 	bool reversed = false;
2965 
2966 	sanity_check_seg_type(sbi, seg_type);
2967 
2968 	/* f2fs_need_SSR() already forces to do this */
2969 	if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
2970 		curseg->next_segno = segno;
2971 		return 1;
2972 	}
2973 
2974 	/* For node segments, let's do SSR more intensively */
2975 	if (IS_NODESEG(seg_type)) {
2976 		if (seg_type >= CURSEG_WARM_NODE) {
2977 			reversed = true;
2978 			i = CURSEG_COLD_NODE;
2979 		} else {
2980 			i = CURSEG_HOT_NODE;
2981 		}
2982 		cnt = NR_CURSEG_NODE_TYPE;
2983 	} else {
2984 		if (seg_type >= CURSEG_WARM_DATA) {
2985 			reversed = true;
2986 			i = CURSEG_COLD_DATA;
2987 		} else {
2988 			i = CURSEG_HOT_DATA;
2989 		}
2990 		cnt = NR_CURSEG_DATA_TYPE;
2991 	}
2992 
2993 	for (; cnt-- > 0; reversed ? i-- : i++) {
2994 		if (i == seg_type)
2995 			continue;
2996 		if (!f2fs_get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
2997 			curseg->next_segno = segno;
2998 			return 1;
2999 		}
3000 	}
3001 
3002 	/* find valid_blocks=0 in dirty list */
3003 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
3004 		segno = get_free_segment(sbi);
3005 		if (segno != NULL_SEGNO) {
3006 			curseg->next_segno = segno;
3007 			return 1;
3008 		}
3009 	}
3010 	return 0;
3011 }
3012 
3013 static bool need_new_seg(struct f2fs_sb_info *sbi, int type)
3014 {
3015 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3016 
3017 	if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
3018 	    curseg->seg_type == CURSEG_WARM_NODE)
3019 		return true;
3020 	if (curseg->alloc_type == LFS &&
3021 	    is_next_segment_free(sbi, curseg, type) &&
3022 	    likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
3023 		return true;
3024 	if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0))
3025 		return true;
3026 	return false;
3027 }
3028 
3029 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
3030 					unsigned int start, unsigned int end)
3031 {
3032 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3033 	unsigned int segno;
3034 
3035 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3036 	mutex_lock(&curseg->curseg_mutex);
3037 	down_write(&SIT_I(sbi)->sentry_lock);
3038 
3039 	segno = CURSEG_I(sbi, type)->segno;
3040 	if (segno < start || segno > end)
3041 		goto unlock;
3042 
3043 	if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
3044 		change_curseg(sbi, type);
3045 	else
3046 		new_curseg(sbi, type, true);
3047 
3048 	stat_inc_seg_type(sbi, curseg);
3049 
3050 	locate_dirty_segment(sbi, segno);
3051 unlock:
3052 	up_write(&SIT_I(sbi)->sentry_lock);
3053 
3054 	if (segno != curseg->segno)
3055 		f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
3056 			    type, segno, curseg->segno);
3057 
3058 	mutex_unlock(&curseg->curseg_mutex);
3059 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3060 }
3061 
3062 static int __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
3063 						bool new_sec, bool force)
3064 {
3065 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3066 	unsigned int old_segno;
3067 
3068 	if (!force && curseg->inited &&
3069 	    !curseg->next_blkoff &&
3070 	    !get_valid_blocks(sbi, curseg->segno, new_sec) &&
3071 	    !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
3072 		return 0;
3073 
3074 	old_segno = curseg->segno;
3075 	if (new_curseg(sbi, type, true))
3076 		return -EAGAIN;
3077 	stat_inc_seg_type(sbi, curseg);
3078 	locate_dirty_segment(sbi, old_segno);
3079 	return 0;
3080 }
3081 
3082 int f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
3083 {
3084 	int ret;
3085 
3086 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3087 	down_write(&SIT_I(sbi)->sentry_lock);
3088 	ret = __allocate_new_segment(sbi, type, true, force);
3089 	up_write(&SIT_I(sbi)->sentry_lock);
3090 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3091 
3092 	return ret;
3093 }
3094 
3095 int f2fs_allocate_pinning_section(struct f2fs_sb_info *sbi)
3096 {
3097 	int err;
3098 	bool gc_required = true;
3099 
3100 retry:
3101 	f2fs_lock_op(sbi);
3102 	err = f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3103 	f2fs_unlock_op(sbi);
3104 
3105 	if (f2fs_sb_has_blkzoned(sbi) && err && gc_required) {
3106 		f2fs_down_write(&sbi->gc_lock);
3107 		f2fs_gc_range(sbi, 0, GET_SEGNO(sbi, FDEV(0).end_blk), true, 1);
3108 		f2fs_up_write(&sbi->gc_lock);
3109 
3110 		gc_required = false;
3111 		goto retry;
3112 	}
3113 
3114 	return err;
3115 }
3116 
3117 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
3118 {
3119 	int i;
3120 
3121 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3122 	down_write(&SIT_I(sbi)->sentry_lock);
3123 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
3124 		__allocate_new_segment(sbi, i, false, false);
3125 	up_write(&SIT_I(sbi)->sentry_lock);
3126 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3127 }
3128 
3129 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
3130 						struct cp_control *cpc)
3131 {
3132 	__u64 trim_start = cpc->trim_start;
3133 	bool has_candidate = false;
3134 
3135 	down_write(&SIT_I(sbi)->sentry_lock);
3136 	for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
3137 		if (add_discard_addrs(sbi, cpc, true)) {
3138 			has_candidate = true;
3139 			break;
3140 		}
3141 	}
3142 	up_write(&SIT_I(sbi)->sentry_lock);
3143 
3144 	cpc->trim_start = trim_start;
3145 	return has_candidate;
3146 }
3147 
3148 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
3149 					struct discard_policy *dpolicy,
3150 					unsigned int start, unsigned int end)
3151 {
3152 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
3153 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
3154 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
3155 	struct discard_cmd *dc;
3156 	struct blk_plug plug;
3157 	int issued;
3158 	unsigned int trimmed = 0;
3159 
3160 next:
3161 	issued = 0;
3162 
3163 	mutex_lock(&dcc->cmd_lock);
3164 	if (unlikely(dcc->rbtree_check))
3165 		f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
3166 
3167 	dc = __lookup_discard_cmd_ret(&dcc->root, start,
3168 				&prev_dc, &next_dc, &insert_p, &insert_parent);
3169 	if (!dc)
3170 		dc = next_dc;
3171 
3172 	blk_start_plug(&plug);
3173 
3174 	while (dc && dc->di.lstart <= end) {
3175 		struct rb_node *node;
3176 		int err = 0;
3177 
3178 		if (dc->di.len < dpolicy->granularity)
3179 			goto skip;
3180 
3181 		if (dc->state != D_PREP) {
3182 			list_move_tail(&dc->list, &dcc->fstrim_list);
3183 			goto skip;
3184 		}
3185 
3186 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3187 
3188 		if (issued >= dpolicy->max_requests) {
3189 			start = dc->di.lstart + dc->di.len;
3190 
3191 			if (err)
3192 				__remove_discard_cmd(sbi, dc);
3193 
3194 			blk_finish_plug(&plug);
3195 			mutex_unlock(&dcc->cmd_lock);
3196 			trimmed += __wait_all_discard_cmd(sbi, NULL);
3197 			f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
3198 			goto next;
3199 		}
3200 skip:
3201 		node = rb_next(&dc->rb_node);
3202 		if (err)
3203 			__remove_discard_cmd(sbi, dc);
3204 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3205 
3206 		if (fatal_signal_pending(current))
3207 			break;
3208 	}
3209 
3210 	blk_finish_plug(&plug);
3211 	mutex_unlock(&dcc->cmd_lock);
3212 
3213 	return trimmed;
3214 }
3215 
3216 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3217 {
3218 	__u64 start = F2FS_BYTES_TO_BLK(range->start);
3219 	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3220 	unsigned int start_segno, end_segno;
3221 	block_t start_block, end_block;
3222 	struct cp_control cpc;
3223 	struct discard_policy dpolicy;
3224 	unsigned long long trimmed = 0;
3225 	int err = 0;
3226 	bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3227 
3228 	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3229 		return -EINVAL;
3230 
3231 	if (end < MAIN_BLKADDR(sbi))
3232 		goto out;
3233 
3234 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3235 		f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3236 		return -EFSCORRUPTED;
3237 	}
3238 
3239 	/* start/end segment number in main_area */
3240 	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3241 	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3242 						GET_SEGNO(sbi, end);
3243 	if (need_align) {
3244 		start_segno = rounddown(start_segno, SEGS_PER_SEC(sbi));
3245 		end_segno = roundup(end_segno + 1, SEGS_PER_SEC(sbi)) - 1;
3246 	}
3247 
3248 	cpc.reason = CP_DISCARD;
3249 	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3250 	cpc.trim_start = start_segno;
3251 	cpc.trim_end = end_segno;
3252 
3253 	if (sbi->discard_blks == 0)
3254 		goto out;
3255 
3256 	f2fs_down_write(&sbi->gc_lock);
3257 	stat_inc_cp_call_count(sbi, TOTAL_CALL);
3258 	err = f2fs_write_checkpoint(sbi, &cpc);
3259 	f2fs_up_write(&sbi->gc_lock);
3260 	if (err)
3261 		goto out;
3262 
3263 	/*
3264 	 * We filed discard candidates, but actually we don't need to wait for
3265 	 * all of them, since they'll be issued in idle time along with runtime
3266 	 * discard option. User configuration looks like using runtime discard
3267 	 * or periodic fstrim instead of it.
3268 	 */
3269 	if (f2fs_realtime_discard_enable(sbi))
3270 		goto out;
3271 
3272 	start_block = START_BLOCK(sbi, start_segno);
3273 	end_block = START_BLOCK(sbi, end_segno + 1);
3274 
3275 	__init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3276 	trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3277 					start_block, end_block);
3278 
3279 	trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3280 					start_block, end_block);
3281 out:
3282 	if (!err)
3283 		range->len = F2FS_BLK_TO_BYTES(trimmed);
3284 	return err;
3285 }
3286 
3287 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3288 {
3289 	switch (hint) {
3290 	case WRITE_LIFE_SHORT:
3291 		return CURSEG_HOT_DATA;
3292 	case WRITE_LIFE_EXTREME:
3293 		return CURSEG_COLD_DATA;
3294 	default:
3295 		return CURSEG_WARM_DATA;
3296 	}
3297 }
3298 
3299 static int __get_segment_type_2(struct f2fs_io_info *fio)
3300 {
3301 	if (fio->type == DATA)
3302 		return CURSEG_HOT_DATA;
3303 	else
3304 		return CURSEG_HOT_NODE;
3305 }
3306 
3307 static int __get_segment_type_4(struct f2fs_io_info *fio)
3308 {
3309 	if (fio->type == DATA) {
3310 		struct inode *inode = fio->page->mapping->host;
3311 
3312 		if (S_ISDIR(inode->i_mode))
3313 			return CURSEG_HOT_DATA;
3314 		else
3315 			return CURSEG_COLD_DATA;
3316 	} else {
3317 		if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3318 			return CURSEG_WARM_NODE;
3319 		else
3320 			return CURSEG_COLD_NODE;
3321 	}
3322 }
3323 
3324 static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs)
3325 {
3326 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3327 	struct extent_info ei = {};
3328 
3329 	if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) {
3330 		if (!ei.age)
3331 			return NO_CHECK_TYPE;
3332 		if (ei.age <= sbi->hot_data_age_threshold)
3333 			return CURSEG_HOT_DATA;
3334 		if (ei.age <= sbi->warm_data_age_threshold)
3335 			return CURSEG_WARM_DATA;
3336 		return CURSEG_COLD_DATA;
3337 	}
3338 	return NO_CHECK_TYPE;
3339 }
3340 
3341 static int __get_segment_type_6(struct f2fs_io_info *fio)
3342 {
3343 	if (fio->type == DATA) {
3344 		struct inode *inode = fio->page->mapping->host;
3345 		int type;
3346 
3347 		if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
3348 			return CURSEG_COLD_DATA_PINNED;
3349 
3350 		if (page_private_gcing(fio->page)) {
3351 			if (fio->sbi->am.atgc_enabled &&
3352 				(fio->io_type == FS_DATA_IO) &&
3353 				(fio->sbi->gc_mode != GC_URGENT_HIGH))
3354 				return CURSEG_ALL_DATA_ATGC;
3355 			else
3356 				return CURSEG_COLD_DATA;
3357 		}
3358 		if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3359 			return CURSEG_COLD_DATA;
3360 
3361 		type = __get_age_segment_type(inode, fio->page->index);
3362 		if (type != NO_CHECK_TYPE)
3363 			return type;
3364 
3365 		if (file_is_hot(inode) ||
3366 				is_inode_flag_set(inode, FI_HOT_DATA) ||
3367 				f2fs_is_cow_file(inode))
3368 			return CURSEG_HOT_DATA;
3369 		return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3370 	} else {
3371 		if (IS_DNODE(fio->page))
3372 			return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3373 						CURSEG_HOT_NODE;
3374 		return CURSEG_COLD_NODE;
3375 	}
3376 }
3377 
3378 static int __get_segment_type(struct f2fs_io_info *fio)
3379 {
3380 	int type = 0;
3381 
3382 	switch (F2FS_OPTION(fio->sbi).active_logs) {
3383 	case 2:
3384 		type = __get_segment_type_2(fio);
3385 		break;
3386 	case 4:
3387 		type = __get_segment_type_4(fio);
3388 		break;
3389 	case 6:
3390 		type = __get_segment_type_6(fio);
3391 		break;
3392 	default:
3393 		f2fs_bug_on(fio->sbi, true);
3394 	}
3395 
3396 	if (IS_HOT(type))
3397 		fio->temp = HOT;
3398 	else if (IS_WARM(type))
3399 		fio->temp = WARM;
3400 	else
3401 		fio->temp = COLD;
3402 	return type;
3403 }
3404 
3405 static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi,
3406 		struct curseg_info *seg)
3407 {
3408 	/* To allocate block chunks in different sizes, use random number */
3409 	if (--seg->fragment_remained_chunk > 0)
3410 		return;
3411 
3412 	seg->fragment_remained_chunk =
3413 		get_random_u32_inclusive(1, sbi->max_fragment_chunk);
3414 	seg->next_blkoff +=
3415 		get_random_u32_inclusive(1, sbi->max_fragment_hole);
3416 }
3417 
3418 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3419 		block_t old_blkaddr, block_t *new_blkaddr,
3420 		struct f2fs_summary *sum, int type,
3421 		struct f2fs_io_info *fio)
3422 {
3423 	struct sit_info *sit_i = SIT_I(sbi);
3424 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3425 	unsigned long long old_mtime;
3426 	bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3427 	struct seg_entry *se = NULL;
3428 	bool segment_full = false;
3429 
3430 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3431 
3432 	mutex_lock(&curseg->curseg_mutex);
3433 	down_write(&sit_i->sentry_lock);
3434 
3435 	if (from_gc) {
3436 		f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3437 		se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3438 		sanity_check_seg_type(sbi, se->type);
3439 		f2fs_bug_on(sbi, IS_NODESEG(se->type));
3440 	}
3441 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3442 
3443 	f2fs_bug_on(sbi, curseg->next_blkoff >= BLKS_PER_SEG(sbi));
3444 
3445 	f2fs_wait_discard_bio(sbi, *new_blkaddr);
3446 
3447 	curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3448 	if (curseg->alloc_type == SSR) {
3449 		curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg);
3450 	} else {
3451 		curseg->next_blkoff++;
3452 		if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
3453 			f2fs_randomize_chunk(sbi, curseg);
3454 	}
3455 	if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno))
3456 		segment_full = true;
3457 	stat_inc_block_count(sbi, curseg);
3458 
3459 	if (from_gc) {
3460 		old_mtime = get_segment_mtime(sbi, old_blkaddr);
3461 	} else {
3462 		update_segment_mtime(sbi, old_blkaddr, 0);
3463 		old_mtime = 0;
3464 	}
3465 	update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3466 
3467 	/*
3468 	 * SIT information should be updated before segment allocation,
3469 	 * since SSR needs latest valid block information.
3470 	 */
3471 	update_sit_entry(sbi, *new_blkaddr, 1);
3472 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3473 		update_sit_entry(sbi, old_blkaddr, -1);
3474 
3475 	/*
3476 	 * If the current segment is full, flush it out and replace it with a
3477 	 * new segment.
3478 	 */
3479 	if (segment_full) {
3480 		if (type == CURSEG_COLD_DATA_PINNED &&
3481 		    !((curseg->segno + 1) % sbi->segs_per_sec)) {
3482 			write_sum_page(sbi, curseg->sum_blk,
3483 					GET_SUM_BLOCK(sbi, curseg->segno));
3484 			goto skip_new_segment;
3485 		}
3486 
3487 		if (from_gc) {
3488 			get_atssr_segment(sbi, type, se->type,
3489 						AT_SSR, se->mtime);
3490 		} else {
3491 			if (need_new_seg(sbi, type))
3492 				new_curseg(sbi, type, false);
3493 			else
3494 				change_curseg(sbi, type);
3495 			stat_inc_seg_type(sbi, curseg);
3496 		}
3497 	}
3498 
3499 skip_new_segment:
3500 	/*
3501 	 * segment dirty status should be updated after segment allocation,
3502 	 * so we just need to update status only one time after previous
3503 	 * segment being closed.
3504 	 */
3505 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3506 	locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3507 
3508 	if (IS_DATASEG(curseg->seg_type))
3509 		atomic64_inc(&sbi->allocated_data_blocks);
3510 
3511 	up_write(&sit_i->sentry_lock);
3512 
3513 	if (page && IS_NODESEG(curseg->seg_type)) {
3514 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3515 
3516 		f2fs_inode_chksum_set(sbi, page);
3517 	}
3518 
3519 	if (fio) {
3520 		struct f2fs_bio_info *io;
3521 
3522 		INIT_LIST_HEAD(&fio->list);
3523 		fio->in_list = 1;
3524 		io = sbi->write_io[fio->type] + fio->temp;
3525 		spin_lock(&io->io_lock);
3526 		list_add_tail(&fio->list, &io->io_list);
3527 		spin_unlock(&io->io_lock);
3528 	}
3529 
3530 	mutex_unlock(&curseg->curseg_mutex);
3531 
3532 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3533 }
3534 
3535 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino,
3536 					block_t blkaddr, unsigned int blkcnt)
3537 {
3538 	if (!f2fs_is_multi_device(sbi))
3539 		return;
3540 
3541 	while (1) {
3542 		unsigned int devidx = f2fs_target_device_index(sbi, blkaddr);
3543 		unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1;
3544 
3545 		/* update device state for fsync */
3546 		f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO);
3547 
3548 		/* update device state for checkpoint */
3549 		if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3550 			spin_lock(&sbi->dev_lock);
3551 			f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3552 			spin_unlock(&sbi->dev_lock);
3553 		}
3554 
3555 		if (blkcnt <= blks)
3556 			break;
3557 		blkcnt -= blks;
3558 		blkaddr += blks;
3559 	}
3560 }
3561 
3562 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3563 {
3564 	int type = __get_segment_type(fio);
3565 	bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3566 
3567 	if (keep_order)
3568 		f2fs_down_read(&fio->sbi->io_order_lock);
3569 
3570 	f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3571 			&fio->new_blkaddr, sum, type, fio);
3572 	if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3573 		f2fs_invalidate_internal_cache(fio->sbi, fio->old_blkaddr);
3574 
3575 	/* writeout dirty page into bdev */
3576 	f2fs_submit_page_write(fio);
3577 
3578 	f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1);
3579 
3580 	if (keep_order)
3581 		f2fs_up_read(&fio->sbi->io_order_lock);
3582 }
3583 
3584 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3585 					enum iostat_type io_type)
3586 {
3587 	struct f2fs_io_info fio = {
3588 		.sbi = sbi,
3589 		.type = META,
3590 		.temp = HOT,
3591 		.op = REQ_OP_WRITE,
3592 		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3593 		.old_blkaddr = page->index,
3594 		.new_blkaddr = page->index,
3595 		.page = page,
3596 		.encrypted_page = NULL,
3597 		.in_list = 0,
3598 	};
3599 
3600 	if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3601 		fio.op_flags &= ~REQ_META;
3602 
3603 	set_page_writeback(page);
3604 	f2fs_submit_page_write(&fio);
3605 
3606 	stat_inc_meta_count(sbi, page->index);
3607 	f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE);
3608 }
3609 
3610 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3611 {
3612 	struct f2fs_summary sum;
3613 
3614 	set_summary(&sum, nid, 0, 0);
3615 	do_write_page(&sum, fio);
3616 
3617 	f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE);
3618 }
3619 
3620 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3621 					struct f2fs_io_info *fio)
3622 {
3623 	struct f2fs_sb_info *sbi = fio->sbi;
3624 	struct f2fs_summary sum;
3625 
3626 	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3627 	if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO)
3628 		f2fs_update_age_extent_cache(dn);
3629 	set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3630 	do_write_page(&sum, fio);
3631 	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3632 
3633 	f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE);
3634 }
3635 
3636 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3637 {
3638 	int err;
3639 	struct f2fs_sb_info *sbi = fio->sbi;
3640 	unsigned int segno;
3641 
3642 	fio->new_blkaddr = fio->old_blkaddr;
3643 	/* i/o temperature is needed for passing down write hints */
3644 	__get_segment_type(fio);
3645 
3646 	segno = GET_SEGNO(sbi, fio->new_blkaddr);
3647 
3648 	if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3649 		set_sbi_flag(sbi, SBI_NEED_FSCK);
3650 		f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3651 			  __func__, segno);
3652 		err = -EFSCORRUPTED;
3653 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
3654 		goto drop_bio;
3655 	}
3656 
3657 	if (f2fs_cp_error(sbi)) {
3658 		err = -EIO;
3659 		goto drop_bio;
3660 	}
3661 
3662 	if (fio->post_read)
3663 		f2fs_truncate_meta_inode_pages(sbi, fio->new_blkaddr, 1);
3664 
3665 	stat_inc_inplace_blocks(fio->sbi);
3666 
3667 	if (fio->bio && !IS_F2FS_IPU_NOCACHE(sbi))
3668 		err = f2fs_merge_page_bio(fio);
3669 	else
3670 		err = f2fs_submit_page_bio(fio);
3671 	if (!err) {
3672 		f2fs_update_device_state(fio->sbi, fio->ino,
3673 						fio->new_blkaddr, 1);
3674 		f2fs_update_iostat(fio->sbi, fio->page->mapping->host,
3675 						fio->io_type, F2FS_BLKSIZE);
3676 	}
3677 
3678 	return err;
3679 drop_bio:
3680 	if (fio->bio && *(fio->bio)) {
3681 		struct bio *bio = *(fio->bio);
3682 
3683 		bio->bi_status = BLK_STS_IOERR;
3684 		bio_endio(bio);
3685 		*(fio->bio) = NULL;
3686 	}
3687 	return err;
3688 }
3689 
3690 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3691 						unsigned int segno)
3692 {
3693 	int i;
3694 
3695 	for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3696 		if (CURSEG_I(sbi, i)->segno == segno)
3697 			break;
3698 	}
3699 	return i;
3700 }
3701 
3702 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3703 				block_t old_blkaddr, block_t new_blkaddr,
3704 				bool recover_curseg, bool recover_newaddr,
3705 				bool from_gc)
3706 {
3707 	struct sit_info *sit_i = SIT_I(sbi);
3708 	struct curseg_info *curseg;
3709 	unsigned int segno, old_cursegno;
3710 	struct seg_entry *se;
3711 	int type;
3712 	unsigned short old_blkoff;
3713 	unsigned char old_alloc_type;
3714 
3715 	segno = GET_SEGNO(sbi, new_blkaddr);
3716 	se = get_seg_entry(sbi, segno);
3717 	type = se->type;
3718 
3719 	f2fs_down_write(&SM_I(sbi)->curseg_lock);
3720 
3721 	if (!recover_curseg) {
3722 		/* for recovery flow */
3723 		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3724 			if (old_blkaddr == NULL_ADDR)
3725 				type = CURSEG_COLD_DATA;
3726 			else
3727 				type = CURSEG_WARM_DATA;
3728 		}
3729 	} else {
3730 		if (IS_CURSEG(sbi, segno)) {
3731 			/* se->type is volatile as SSR allocation */
3732 			type = __f2fs_get_curseg(sbi, segno);
3733 			f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3734 		} else {
3735 			type = CURSEG_WARM_DATA;
3736 		}
3737 	}
3738 
3739 	f2fs_bug_on(sbi, !IS_DATASEG(type));
3740 	curseg = CURSEG_I(sbi, type);
3741 
3742 	mutex_lock(&curseg->curseg_mutex);
3743 	down_write(&sit_i->sentry_lock);
3744 
3745 	old_cursegno = curseg->segno;
3746 	old_blkoff = curseg->next_blkoff;
3747 	old_alloc_type = curseg->alloc_type;
3748 
3749 	/* change the current segment */
3750 	if (segno != curseg->segno) {
3751 		curseg->next_segno = segno;
3752 		change_curseg(sbi, type);
3753 	}
3754 
3755 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3756 	curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3757 
3758 	if (!recover_curseg || recover_newaddr) {
3759 		if (!from_gc)
3760 			update_segment_mtime(sbi, new_blkaddr, 0);
3761 		update_sit_entry(sbi, new_blkaddr, 1);
3762 	}
3763 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3764 		f2fs_invalidate_internal_cache(sbi, old_blkaddr);
3765 		if (!from_gc)
3766 			update_segment_mtime(sbi, old_blkaddr, 0);
3767 		update_sit_entry(sbi, old_blkaddr, -1);
3768 	}
3769 
3770 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3771 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3772 
3773 	locate_dirty_segment(sbi, old_cursegno);
3774 
3775 	if (recover_curseg) {
3776 		if (old_cursegno != curseg->segno) {
3777 			curseg->next_segno = old_cursegno;
3778 			change_curseg(sbi, type);
3779 		}
3780 		curseg->next_blkoff = old_blkoff;
3781 		curseg->alloc_type = old_alloc_type;
3782 	}
3783 
3784 	up_write(&sit_i->sentry_lock);
3785 	mutex_unlock(&curseg->curseg_mutex);
3786 	f2fs_up_write(&SM_I(sbi)->curseg_lock);
3787 }
3788 
3789 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3790 				block_t old_addr, block_t new_addr,
3791 				unsigned char version, bool recover_curseg,
3792 				bool recover_newaddr)
3793 {
3794 	struct f2fs_summary sum;
3795 
3796 	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3797 
3798 	f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3799 					recover_curseg, recover_newaddr, false);
3800 
3801 	f2fs_update_data_blkaddr(dn, new_addr);
3802 }
3803 
3804 void f2fs_wait_on_page_writeback(struct page *page,
3805 				enum page_type type, bool ordered, bool locked)
3806 {
3807 	if (PageWriteback(page)) {
3808 		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3809 
3810 		/* submit cached LFS IO */
3811 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3812 		/* submit cached IPU IO */
3813 		f2fs_submit_merged_ipu_write(sbi, NULL, page);
3814 		if (ordered) {
3815 			wait_on_page_writeback(page);
3816 			f2fs_bug_on(sbi, locked && PageWriteback(page));
3817 		} else {
3818 			wait_for_stable_page(page);
3819 		}
3820 	}
3821 }
3822 
3823 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3824 {
3825 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3826 	struct page *cpage;
3827 
3828 	if (!f2fs_post_read_required(inode))
3829 		return;
3830 
3831 	if (!__is_valid_data_blkaddr(blkaddr))
3832 		return;
3833 
3834 	cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3835 	if (cpage) {
3836 		f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3837 		f2fs_put_page(cpage, 1);
3838 	}
3839 }
3840 
3841 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3842 								block_t len)
3843 {
3844 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3845 	block_t i;
3846 
3847 	if (!f2fs_post_read_required(inode))
3848 		return;
3849 
3850 	for (i = 0; i < len; i++)
3851 		f2fs_wait_on_block_writeback(inode, blkaddr + i);
3852 
3853 	f2fs_truncate_meta_inode_pages(sbi, blkaddr, len);
3854 }
3855 
3856 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3857 {
3858 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3859 	struct curseg_info *seg_i;
3860 	unsigned char *kaddr;
3861 	struct page *page;
3862 	block_t start;
3863 	int i, j, offset;
3864 
3865 	start = start_sum_block(sbi);
3866 
3867 	page = f2fs_get_meta_page(sbi, start++);
3868 	if (IS_ERR(page))
3869 		return PTR_ERR(page);
3870 	kaddr = (unsigned char *)page_address(page);
3871 
3872 	/* Step 1: restore nat cache */
3873 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3874 	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3875 
3876 	/* Step 2: restore sit cache */
3877 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3878 	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3879 	offset = 2 * SUM_JOURNAL_SIZE;
3880 
3881 	/* Step 3: restore summary entries */
3882 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3883 		unsigned short blk_off;
3884 		unsigned int segno;
3885 
3886 		seg_i = CURSEG_I(sbi, i);
3887 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3888 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3889 		seg_i->next_segno = segno;
3890 		reset_curseg(sbi, i, 0);
3891 		seg_i->alloc_type = ckpt->alloc_type[i];
3892 		seg_i->next_blkoff = blk_off;
3893 
3894 		if (seg_i->alloc_type == SSR)
3895 			blk_off = BLKS_PER_SEG(sbi);
3896 
3897 		for (j = 0; j < blk_off; j++) {
3898 			struct f2fs_summary *s;
3899 
3900 			s = (struct f2fs_summary *)(kaddr + offset);
3901 			seg_i->sum_blk->entries[j] = *s;
3902 			offset += SUMMARY_SIZE;
3903 			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3904 						SUM_FOOTER_SIZE)
3905 				continue;
3906 
3907 			f2fs_put_page(page, 1);
3908 			page = NULL;
3909 
3910 			page = f2fs_get_meta_page(sbi, start++);
3911 			if (IS_ERR(page))
3912 				return PTR_ERR(page);
3913 			kaddr = (unsigned char *)page_address(page);
3914 			offset = 0;
3915 		}
3916 	}
3917 	f2fs_put_page(page, 1);
3918 	return 0;
3919 }
3920 
3921 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3922 {
3923 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3924 	struct f2fs_summary_block *sum;
3925 	struct curseg_info *curseg;
3926 	struct page *new;
3927 	unsigned short blk_off;
3928 	unsigned int segno = 0;
3929 	block_t blk_addr = 0;
3930 	int err = 0;
3931 
3932 	/* get segment number and block addr */
3933 	if (IS_DATASEG(type)) {
3934 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3935 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3936 							CURSEG_HOT_DATA]);
3937 		if (__exist_node_summaries(sbi))
3938 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
3939 		else
3940 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3941 	} else {
3942 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
3943 							CURSEG_HOT_NODE]);
3944 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3945 							CURSEG_HOT_NODE]);
3946 		if (__exist_node_summaries(sbi))
3947 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3948 							type - CURSEG_HOT_NODE);
3949 		else
3950 			blk_addr = GET_SUM_BLOCK(sbi, segno);
3951 	}
3952 
3953 	new = f2fs_get_meta_page(sbi, blk_addr);
3954 	if (IS_ERR(new))
3955 		return PTR_ERR(new);
3956 	sum = (struct f2fs_summary_block *)page_address(new);
3957 
3958 	if (IS_NODESEG(type)) {
3959 		if (__exist_node_summaries(sbi)) {
3960 			struct f2fs_summary *ns = &sum->entries[0];
3961 			int i;
3962 
3963 			for (i = 0; i < BLKS_PER_SEG(sbi); i++, ns++) {
3964 				ns->version = 0;
3965 				ns->ofs_in_node = 0;
3966 			}
3967 		} else {
3968 			err = f2fs_restore_node_summary(sbi, segno, sum);
3969 			if (err)
3970 				goto out;
3971 		}
3972 	}
3973 
3974 	/* set uncompleted segment to curseg */
3975 	curseg = CURSEG_I(sbi, type);
3976 	mutex_lock(&curseg->curseg_mutex);
3977 
3978 	/* update journal info */
3979 	down_write(&curseg->journal_rwsem);
3980 	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3981 	up_write(&curseg->journal_rwsem);
3982 
3983 	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3984 	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3985 	curseg->next_segno = segno;
3986 	reset_curseg(sbi, type, 0);
3987 	curseg->alloc_type = ckpt->alloc_type[type];
3988 	curseg->next_blkoff = blk_off;
3989 	mutex_unlock(&curseg->curseg_mutex);
3990 out:
3991 	f2fs_put_page(new, 1);
3992 	return err;
3993 }
3994 
3995 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3996 {
3997 	struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3998 	struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3999 	int type = CURSEG_HOT_DATA;
4000 	int err;
4001 
4002 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
4003 		int npages = f2fs_npages_for_summary_flush(sbi, true);
4004 
4005 		if (npages >= 2)
4006 			f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
4007 							META_CP, true);
4008 
4009 		/* restore for compacted data summary */
4010 		err = read_compacted_summaries(sbi);
4011 		if (err)
4012 			return err;
4013 		type = CURSEG_HOT_NODE;
4014 	}
4015 
4016 	if (__exist_node_summaries(sbi))
4017 		f2fs_ra_meta_pages(sbi,
4018 				sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
4019 				NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
4020 
4021 	for (; type <= CURSEG_COLD_NODE; type++) {
4022 		err = read_normal_summaries(sbi, type);
4023 		if (err)
4024 			return err;
4025 	}
4026 
4027 	/* sanity check for summary blocks */
4028 	if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
4029 			sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
4030 		f2fs_err(sbi, "invalid journal entries nats %u sits %u",
4031 			 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
4032 		return -EINVAL;
4033 	}
4034 
4035 	return 0;
4036 }
4037 
4038 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
4039 {
4040 	struct page *page;
4041 	unsigned char *kaddr;
4042 	struct f2fs_summary *summary;
4043 	struct curseg_info *seg_i;
4044 	int written_size = 0;
4045 	int i, j;
4046 
4047 	page = f2fs_grab_meta_page(sbi, blkaddr++);
4048 	kaddr = (unsigned char *)page_address(page);
4049 	memset(kaddr, 0, PAGE_SIZE);
4050 
4051 	/* Step 1: write nat cache */
4052 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
4053 	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
4054 	written_size += SUM_JOURNAL_SIZE;
4055 
4056 	/* Step 2: write sit cache */
4057 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
4058 	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
4059 	written_size += SUM_JOURNAL_SIZE;
4060 
4061 	/* Step 3: write summary entries */
4062 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
4063 		seg_i = CURSEG_I(sbi, i);
4064 		for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); j++) {
4065 			if (!page) {
4066 				page = f2fs_grab_meta_page(sbi, blkaddr++);
4067 				kaddr = (unsigned char *)page_address(page);
4068 				memset(kaddr, 0, PAGE_SIZE);
4069 				written_size = 0;
4070 			}
4071 			summary = (struct f2fs_summary *)(kaddr + written_size);
4072 			*summary = seg_i->sum_blk->entries[j];
4073 			written_size += SUMMARY_SIZE;
4074 
4075 			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
4076 							SUM_FOOTER_SIZE)
4077 				continue;
4078 
4079 			set_page_dirty(page);
4080 			f2fs_put_page(page, 1);
4081 			page = NULL;
4082 		}
4083 	}
4084 	if (page) {
4085 		set_page_dirty(page);
4086 		f2fs_put_page(page, 1);
4087 	}
4088 }
4089 
4090 static void write_normal_summaries(struct f2fs_sb_info *sbi,
4091 					block_t blkaddr, int type)
4092 {
4093 	int i, end;
4094 
4095 	if (IS_DATASEG(type))
4096 		end = type + NR_CURSEG_DATA_TYPE;
4097 	else
4098 		end = type + NR_CURSEG_NODE_TYPE;
4099 
4100 	for (i = type; i < end; i++)
4101 		write_current_sum_page(sbi, i, blkaddr + (i - type));
4102 }
4103 
4104 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4105 {
4106 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
4107 		write_compacted_summaries(sbi, start_blk);
4108 	else
4109 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
4110 }
4111 
4112 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4113 {
4114 	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
4115 }
4116 
4117 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
4118 					unsigned int val, int alloc)
4119 {
4120 	int i;
4121 
4122 	if (type == NAT_JOURNAL) {
4123 		for (i = 0; i < nats_in_cursum(journal); i++) {
4124 			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
4125 				return i;
4126 		}
4127 		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
4128 			return update_nats_in_cursum(journal, 1);
4129 	} else if (type == SIT_JOURNAL) {
4130 		for (i = 0; i < sits_in_cursum(journal); i++)
4131 			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
4132 				return i;
4133 		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
4134 			return update_sits_in_cursum(journal, 1);
4135 	}
4136 	return -1;
4137 }
4138 
4139 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
4140 					unsigned int segno)
4141 {
4142 	return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
4143 }
4144 
4145 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
4146 					unsigned int start)
4147 {
4148 	struct sit_info *sit_i = SIT_I(sbi);
4149 	struct page *page;
4150 	pgoff_t src_off, dst_off;
4151 
4152 	src_off = current_sit_addr(sbi, start);
4153 	dst_off = next_sit_addr(sbi, src_off);
4154 
4155 	page = f2fs_grab_meta_page(sbi, dst_off);
4156 	seg_info_to_sit_page(sbi, page, start);
4157 
4158 	set_page_dirty(page);
4159 	set_to_next_sit(sit_i, start);
4160 
4161 	return page;
4162 }
4163 
4164 static struct sit_entry_set *grab_sit_entry_set(void)
4165 {
4166 	struct sit_entry_set *ses =
4167 			f2fs_kmem_cache_alloc(sit_entry_set_slab,
4168 						GFP_NOFS, true, NULL);
4169 
4170 	ses->entry_cnt = 0;
4171 	INIT_LIST_HEAD(&ses->set_list);
4172 	return ses;
4173 }
4174 
4175 static void release_sit_entry_set(struct sit_entry_set *ses)
4176 {
4177 	list_del(&ses->set_list);
4178 	kmem_cache_free(sit_entry_set_slab, ses);
4179 }
4180 
4181 static void adjust_sit_entry_set(struct sit_entry_set *ses,
4182 						struct list_head *head)
4183 {
4184 	struct sit_entry_set *next = ses;
4185 
4186 	if (list_is_last(&ses->set_list, head))
4187 		return;
4188 
4189 	list_for_each_entry_continue(next, head, set_list)
4190 		if (ses->entry_cnt <= next->entry_cnt) {
4191 			list_move_tail(&ses->set_list, &next->set_list);
4192 			return;
4193 		}
4194 
4195 	list_move_tail(&ses->set_list, head);
4196 }
4197 
4198 static void add_sit_entry(unsigned int segno, struct list_head *head)
4199 {
4200 	struct sit_entry_set *ses;
4201 	unsigned int start_segno = START_SEGNO(segno);
4202 
4203 	list_for_each_entry(ses, head, set_list) {
4204 		if (ses->start_segno == start_segno) {
4205 			ses->entry_cnt++;
4206 			adjust_sit_entry_set(ses, head);
4207 			return;
4208 		}
4209 	}
4210 
4211 	ses = grab_sit_entry_set();
4212 
4213 	ses->start_segno = start_segno;
4214 	ses->entry_cnt++;
4215 	list_add(&ses->set_list, head);
4216 }
4217 
4218 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4219 {
4220 	struct f2fs_sm_info *sm_info = SM_I(sbi);
4221 	struct list_head *set_list = &sm_info->sit_entry_set;
4222 	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4223 	unsigned int segno;
4224 
4225 	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4226 		add_sit_entry(segno, set_list);
4227 }
4228 
4229 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4230 {
4231 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4232 	struct f2fs_journal *journal = curseg->journal;
4233 	int i;
4234 
4235 	down_write(&curseg->journal_rwsem);
4236 	for (i = 0; i < sits_in_cursum(journal); i++) {
4237 		unsigned int segno;
4238 		bool dirtied;
4239 
4240 		segno = le32_to_cpu(segno_in_journal(journal, i));
4241 		dirtied = __mark_sit_entry_dirty(sbi, segno);
4242 
4243 		if (!dirtied)
4244 			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4245 	}
4246 	update_sits_in_cursum(journal, -i);
4247 	up_write(&curseg->journal_rwsem);
4248 }
4249 
4250 /*
4251  * CP calls this function, which flushes SIT entries including sit_journal,
4252  * and moves prefree segs to free segs.
4253  */
4254 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4255 {
4256 	struct sit_info *sit_i = SIT_I(sbi);
4257 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4258 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4259 	struct f2fs_journal *journal = curseg->journal;
4260 	struct sit_entry_set *ses, *tmp;
4261 	struct list_head *head = &SM_I(sbi)->sit_entry_set;
4262 	bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4263 	struct seg_entry *se;
4264 
4265 	down_write(&sit_i->sentry_lock);
4266 
4267 	if (!sit_i->dirty_sentries)
4268 		goto out;
4269 
4270 	/*
4271 	 * add and account sit entries of dirty bitmap in sit entry
4272 	 * set temporarily
4273 	 */
4274 	add_sits_in_set(sbi);
4275 
4276 	/*
4277 	 * if there are no enough space in journal to store dirty sit
4278 	 * entries, remove all entries from journal and add and account
4279 	 * them in sit entry set.
4280 	 */
4281 	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4282 								!to_journal)
4283 		remove_sits_in_journal(sbi);
4284 
4285 	/*
4286 	 * there are two steps to flush sit entries:
4287 	 * #1, flush sit entries to journal in current cold data summary block.
4288 	 * #2, flush sit entries to sit page.
4289 	 */
4290 	list_for_each_entry_safe(ses, tmp, head, set_list) {
4291 		struct page *page = NULL;
4292 		struct f2fs_sit_block *raw_sit = NULL;
4293 		unsigned int start_segno = ses->start_segno;
4294 		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4295 						(unsigned long)MAIN_SEGS(sbi));
4296 		unsigned int segno = start_segno;
4297 
4298 		if (to_journal &&
4299 			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4300 			to_journal = false;
4301 
4302 		if (to_journal) {
4303 			down_write(&curseg->journal_rwsem);
4304 		} else {
4305 			page = get_next_sit_page(sbi, start_segno);
4306 			raw_sit = page_address(page);
4307 		}
4308 
4309 		/* flush dirty sit entries in region of current sit set */
4310 		for_each_set_bit_from(segno, bitmap, end) {
4311 			int offset, sit_offset;
4312 
4313 			se = get_seg_entry(sbi, segno);
4314 #ifdef CONFIG_F2FS_CHECK_FS
4315 			if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4316 						SIT_VBLOCK_MAP_SIZE))
4317 				f2fs_bug_on(sbi, 1);
4318 #endif
4319 
4320 			/* add discard candidates */
4321 			if (!(cpc->reason & CP_DISCARD)) {
4322 				cpc->trim_start = segno;
4323 				add_discard_addrs(sbi, cpc, false);
4324 			}
4325 
4326 			if (to_journal) {
4327 				offset = f2fs_lookup_journal_in_cursum(journal,
4328 							SIT_JOURNAL, segno, 1);
4329 				f2fs_bug_on(sbi, offset < 0);
4330 				segno_in_journal(journal, offset) =
4331 							cpu_to_le32(segno);
4332 				seg_info_to_raw_sit(se,
4333 					&sit_in_journal(journal, offset));
4334 				check_block_count(sbi, segno,
4335 					&sit_in_journal(journal, offset));
4336 			} else {
4337 				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4338 				seg_info_to_raw_sit(se,
4339 						&raw_sit->entries[sit_offset]);
4340 				check_block_count(sbi, segno,
4341 						&raw_sit->entries[sit_offset]);
4342 			}
4343 
4344 			__clear_bit(segno, bitmap);
4345 			sit_i->dirty_sentries--;
4346 			ses->entry_cnt--;
4347 		}
4348 
4349 		if (to_journal)
4350 			up_write(&curseg->journal_rwsem);
4351 		else
4352 			f2fs_put_page(page, 1);
4353 
4354 		f2fs_bug_on(sbi, ses->entry_cnt);
4355 		release_sit_entry_set(ses);
4356 	}
4357 
4358 	f2fs_bug_on(sbi, !list_empty(head));
4359 	f2fs_bug_on(sbi, sit_i->dirty_sentries);
4360 out:
4361 	if (cpc->reason & CP_DISCARD) {
4362 		__u64 trim_start = cpc->trim_start;
4363 
4364 		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4365 			add_discard_addrs(sbi, cpc, false);
4366 
4367 		cpc->trim_start = trim_start;
4368 	}
4369 	up_write(&sit_i->sentry_lock);
4370 
4371 	set_prefree_as_free_segments(sbi);
4372 }
4373 
4374 static int build_sit_info(struct f2fs_sb_info *sbi)
4375 {
4376 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4377 	struct sit_info *sit_i;
4378 	unsigned int sit_segs, start;
4379 	char *src_bitmap, *bitmap;
4380 	unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4381 	unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0;
4382 
4383 	/* allocate memory for SIT information */
4384 	sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4385 	if (!sit_i)
4386 		return -ENOMEM;
4387 
4388 	SM_I(sbi)->sit_info = sit_i;
4389 
4390 	sit_i->sentries =
4391 		f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4392 					      MAIN_SEGS(sbi)),
4393 			      GFP_KERNEL);
4394 	if (!sit_i->sentries)
4395 		return -ENOMEM;
4396 
4397 	main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4398 	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4399 								GFP_KERNEL);
4400 	if (!sit_i->dirty_sentries_bitmap)
4401 		return -ENOMEM;
4402 
4403 #ifdef CONFIG_F2FS_CHECK_FS
4404 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map);
4405 #else
4406 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map);
4407 #endif
4408 	sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4409 	if (!sit_i->bitmap)
4410 		return -ENOMEM;
4411 
4412 	bitmap = sit_i->bitmap;
4413 
4414 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4415 		sit_i->sentries[start].cur_valid_map = bitmap;
4416 		bitmap += SIT_VBLOCK_MAP_SIZE;
4417 
4418 		sit_i->sentries[start].ckpt_valid_map = bitmap;
4419 		bitmap += SIT_VBLOCK_MAP_SIZE;
4420 
4421 #ifdef CONFIG_F2FS_CHECK_FS
4422 		sit_i->sentries[start].cur_valid_map_mir = bitmap;
4423 		bitmap += SIT_VBLOCK_MAP_SIZE;
4424 #endif
4425 
4426 		if (discard_map) {
4427 			sit_i->sentries[start].discard_map = bitmap;
4428 			bitmap += SIT_VBLOCK_MAP_SIZE;
4429 		}
4430 	}
4431 
4432 	sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4433 	if (!sit_i->tmp_map)
4434 		return -ENOMEM;
4435 
4436 	if (__is_large_section(sbi)) {
4437 		sit_i->sec_entries =
4438 			f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4439 						      MAIN_SECS(sbi)),
4440 				      GFP_KERNEL);
4441 		if (!sit_i->sec_entries)
4442 			return -ENOMEM;
4443 	}
4444 
4445 	/* get information related with SIT */
4446 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4447 
4448 	/* setup SIT bitmap from ckeckpoint pack */
4449 	sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4450 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4451 
4452 	sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4453 	if (!sit_i->sit_bitmap)
4454 		return -ENOMEM;
4455 
4456 #ifdef CONFIG_F2FS_CHECK_FS
4457 	sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4458 					sit_bitmap_size, GFP_KERNEL);
4459 	if (!sit_i->sit_bitmap_mir)
4460 		return -ENOMEM;
4461 
4462 	sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4463 					main_bitmap_size, GFP_KERNEL);
4464 	if (!sit_i->invalid_segmap)
4465 		return -ENOMEM;
4466 #endif
4467 
4468 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4469 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4470 	sit_i->written_valid_blocks = 0;
4471 	sit_i->bitmap_size = sit_bitmap_size;
4472 	sit_i->dirty_sentries = 0;
4473 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4474 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4475 	sit_i->mounted_time = ktime_get_boottime_seconds();
4476 	init_rwsem(&sit_i->sentry_lock);
4477 	return 0;
4478 }
4479 
4480 static int build_free_segmap(struct f2fs_sb_info *sbi)
4481 {
4482 	struct free_segmap_info *free_i;
4483 	unsigned int bitmap_size, sec_bitmap_size;
4484 
4485 	/* allocate memory for free segmap information */
4486 	free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4487 	if (!free_i)
4488 		return -ENOMEM;
4489 
4490 	SM_I(sbi)->free_info = free_i;
4491 
4492 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4493 	free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4494 	if (!free_i->free_segmap)
4495 		return -ENOMEM;
4496 
4497 	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4498 	free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4499 	if (!free_i->free_secmap)
4500 		return -ENOMEM;
4501 
4502 	/* set all segments as dirty temporarily */
4503 	memset(free_i->free_segmap, 0xff, bitmap_size);
4504 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4505 
4506 	/* init free segmap information */
4507 	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4508 	free_i->free_segments = 0;
4509 	free_i->free_sections = 0;
4510 	spin_lock_init(&free_i->segmap_lock);
4511 	return 0;
4512 }
4513 
4514 static int build_curseg(struct f2fs_sb_info *sbi)
4515 {
4516 	struct curseg_info *array;
4517 	int i;
4518 
4519 	array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4520 					sizeof(*array)), GFP_KERNEL);
4521 	if (!array)
4522 		return -ENOMEM;
4523 
4524 	SM_I(sbi)->curseg_array = array;
4525 
4526 	for (i = 0; i < NO_CHECK_TYPE; i++) {
4527 		mutex_init(&array[i].curseg_mutex);
4528 		array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4529 		if (!array[i].sum_blk)
4530 			return -ENOMEM;
4531 		init_rwsem(&array[i].journal_rwsem);
4532 		array[i].journal = f2fs_kzalloc(sbi,
4533 				sizeof(struct f2fs_journal), GFP_KERNEL);
4534 		if (!array[i].journal)
4535 			return -ENOMEM;
4536 		if (i < NR_PERSISTENT_LOG)
4537 			array[i].seg_type = CURSEG_HOT_DATA + i;
4538 		else if (i == CURSEG_COLD_DATA_PINNED)
4539 			array[i].seg_type = CURSEG_COLD_DATA;
4540 		else if (i == CURSEG_ALL_DATA_ATGC)
4541 			array[i].seg_type = CURSEG_COLD_DATA;
4542 		array[i].segno = NULL_SEGNO;
4543 		array[i].next_blkoff = 0;
4544 		array[i].inited = false;
4545 	}
4546 	return restore_curseg_summaries(sbi);
4547 }
4548 
4549 static int build_sit_entries(struct f2fs_sb_info *sbi)
4550 {
4551 	struct sit_info *sit_i = SIT_I(sbi);
4552 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4553 	struct f2fs_journal *journal = curseg->journal;
4554 	struct seg_entry *se;
4555 	struct f2fs_sit_entry sit;
4556 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
4557 	unsigned int i, start, end;
4558 	unsigned int readed, start_blk = 0;
4559 	int err = 0;
4560 	block_t sit_valid_blocks[2] = {0, 0};
4561 
4562 	do {
4563 		readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4564 							META_SIT, true);
4565 
4566 		start = start_blk * sit_i->sents_per_block;
4567 		end = (start_blk + readed) * sit_i->sents_per_block;
4568 
4569 		for (; start < end && start < MAIN_SEGS(sbi); start++) {
4570 			struct f2fs_sit_block *sit_blk;
4571 			struct page *page;
4572 
4573 			se = &sit_i->sentries[start];
4574 			page = get_current_sit_page(sbi, start);
4575 			if (IS_ERR(page))
4576 				return PTR_ERR(page);
4577 			sit_blk = (struct f2fs_sit_block *)page_address(page);
4578 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4579 			f2fs_put_page(page, 1);
4580 
4581 			err = check_block_count(sbi, start, &sit);
4582 			if (err)
4583 				return err;
4584 			seg_info_from_raw_sit(se, &sit);
4585 
4586 			if (se->type >= NR_PERSISTENT_LOG) {
4587 				f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4588 							se->type, start);
4589 				f2fs_handle_error(sbi,
4590 						ERROR_INCONSISTENT_SUM_TYPE);
4591 				return -EFSCORRUPTED;
4592 			}
4593 
4594 			sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4595 
4596 			if (!f2fs_block_unit_discard(sbi))
4597 				goto init_discard_map_done;
4598 
4599 			/* build discard map only one time */
4600 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4601 				memset(se->discard_map, 0xff,
4602 						SIT_VBLOCK_MAP_SIZE);
4603 				goto init_discard_map_done;
4604 			}
4605 			memcpy(se->discard_map, se->cur_valid_map,
4606 						SIT_VBLOCK_MAP_SIZE);
4607 			sbi->discard_blks += BLKS_PER_SEG(sbi) -
4608 						se->valid_blocks;
4609 init_discard_map_done:
4610 			if (__is_large_section(sbi))
4611 				get_sec_entry(sbi, start)->valid_blocks +=
4612 							se->valid_blocks;
4613 		}
4614 		start_blk += readed;
4615 	} while (start_blk < sit_blk_cnt);
4616 
4617 	down_read(&curseg->journal_rwsem);
4618 	for (i = 0; i < sits_in_cursum(journal); i++) {
4619 		unsigned int old_valid_blocks;
4620 
4621 		start = le32_to_cpu(segno_in_journal(journal, i));
4622 		if (start >= MAIN_SEGS(sbi)) {
4623 			f2fs_err(sbi, "Wrong journal entry on segno %u",
4624 				 start);
4625 			err = -EFSCORRUPTED;
4626 			f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL);
4627 			break;
4628 		}
4629 
4630 		se = &sit_i->sentries[start];
4631 		sit = sit_in_journal(journal, i);
4632 
4633 		old_valid_blocks = se->valid_blocks;
4634 
4635 		sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks;
4636 
4637 		err = check_block_count(sbi, start, &sit);
4638 		if (err)
4639 			break;
4640 		seg_info_from_raw_sit(se, &sit);
4641 
4642 		if (se->type >= NR_PERSISTENT_LOG) {
4643 			f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4644 							se->type, start);
4645 			err = -EFSCORRUPTED;
4646 			f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
4647 			break;
4648 		}
4649 
4650 		sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4651 
4652 		if (f2fs_block_unit_discard(sbi)) {
4653 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4654 				memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4655 			} else {
4656 				memcpy(se->discard_map, se->cur_valid_map,
4657 							SIT_VBLOCK_MAP_SIZE);
4658 				sbi->discard_blks += old_valid_blocks;
4659 				sbi->discard_blks -= se->valid_blocks;
4660 			}
4661 		}
4662 
4663 		if (__is_large_section(sbi)) {
4664 			get_sec_entry(sbi, start)->valid_blocks +=
4665 							se->valid_blocks;
4666 			get_sec_entry(sbi, start)->valid_blocks -=
4667 							old_valid_blocks;
4668 		}
4669 	}
4670 	up_read(&curseg->journal_rwsem);
4671 
4672 	if (err)
4673 		return err;
4674 
4675 	if (sit_valid_blocks[NODE] != valid_node_count(sbi)) {
4676 		f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4677 			 sit_valid_blocks[NODE], valid_node_count(sbi));
4678 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT);
4679 		return -EFSCORRUPTED;
4680 	}
4681 
4682 	if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] >
4683 				valid_user_blocks(sbi)) {
4684 		f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u",
4685 			 sit_valid_blocks[DATA], sit_valid_blocks[NODE],
4686 			 valid_user_blocks(sbi));
4687 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT);
4688 		return -EFSCORRUPTED;
4689 	}
4690 
4691 	return 0;
4692 }
4693 
4694 static void init_free_segmap(struct f2fs_sb_info *sbi)
4695 {
4696 	unsigned int start;
4697 	int type;
4698 	struct seg_entry *sentry;
4699 
4700 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4701 		if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4702 			continue;
4703 		sentry = get_seg_entry(sbi, start);
4704 		if (!sentry->valid_blocks)
4705 			__set_free(sbi, start);
4706 		else
4707 			SIT_I(sbi)->written_valid_blocks +=
4708 						sentry->valid_blocks;
4709 	}
4710 
4711 	/* set use the current segments */
4712 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4713 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4714 
4715 		__set_test_and_inuse(sbi, curseg_t->segno);
4716 	}
4717 }
4718 
4719 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4720 {
4721 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4722 	struct free_segmap_info *free_i = FREE_I(sbi);
4723 	unsigned int segno = 0, offset = 0, secno;
4724 	block_t valid_blocks, usable_blks_in_seg;
4725 
4726 	while (1) {
4727 		/* find dirty segment based on free segmap */
4728 		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4729 		if (segno >= MAIN_SEGS(sbi))
4730 			break;
4731 		offset = segno + 1;
4732 		valid_blocks = get_valid_blocks(sbi, segno, false);
4733 		usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4734 		if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4735 			continue;
4736 		if (valid_blocks > usable_blks_in_seg) {
4737 			f2fs_bug_on(sbi, 1);
4738 			continue;
4739 		}
4740 		mutex_lock(&dirty_i->seglist_lock);
4741 		__locate_dirty_segment(sbi, segno, DIRTY);
4742 		mutex_unlock(&dirty_i->seglist_lock);
4743 	}
4744 
4745 	if (!__is_large_section(sbi))
4746 		return;
4747 
4748 	mutex_lock(&dirty_i->seglist_lock);
4749 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) {
4750 		valid_blocks = get_valid_blocks(sbi, segno, true);
4751 		secno = GET_SEC_FROM_SEG(sbi, segno);
4752 
4753 		if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi))
4754 			continue;
4755 		if (IS_CURSEC(sbi, secno))
4756 			continue;
4757 		set_bit(secno, dirty_i->dirty_secmap);
4758 	}
4759 	mutex_unlock(&dirty_i->seglist_lock);
4760 }
4761 
4762 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4763 {
4764 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4765 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4766 
4767 	dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4768 	if (!dirty_i->victim_secmap)
4769 		return -ENOMEM;
4770 
4771 	dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4772 	if (!dirty_i->pinned_secmap)
4773 		return -ENOMEM;
4774 
4775 	dirty_i->pinned_secmap_cnt = 0;
4776 	dirty_i->enable_pin_section = true;
4777 	return 0;
4778 }
4779 
4780 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4781 {
4782 	struct dirty_seglist_info *dirty_i;
4783 	unsigned int bitmap_size, i;
4784 
4785 	/* allocate memory for dirty segments list information */
4786 	dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4787 								GFP_KERNEL);
4788 	if (!dirty_i)
4789 		return -ENOMEM;
4790 
4791 	SM_I(sbi)->dirty_info = dirty_i;
4792 	mutex_init(&dirty_i->seglist_lock);
4793 
4794 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4795 
4796 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
4797 		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4798 								GFP_KERNEL);
4799 		if (!dirty_i->dirty_segmap[i])
4800 			return -ENOMEM;
4801 	}
4802 
4803 	if (__is_large_section(sbi)) {
4804 		bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4805 		dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4806 						bitmap_size, GFP_KERNEL);
4807 		if (!dirty_i->dirty_secmap)
4808 			return -ENOMEM;
4809 	}
4810 
4811 	init_dirty_segmap(sbi);
4812 	return init_victim_secmap(sbi);
4813 }
4814 
4815 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4816 {
4817 	int i;
4818 
4819 	/*
4820 	 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4821 	 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4822 	 */
4823 	for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4824 		struct curseg_info *curseg = CURSEG_I(sbi, i);
4825 		struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4826 		unsigned int blkofs = curseg->next_blkoff;
4827 
4828 		if (f2fs_sb_has_readonly(sbi) &&
4829 			i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE)
4830 			continue;
4831 
4832 		sanity_check_seg_type(sbi, curseg->seg_type);
4833 
4834 		if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) {
4835 			f2fs_err(sbi,
4836 				 "Current segment has invalid alloc_type:%d",
4837 				 curseg->alloc_type);
4838 			f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4839 			return -EFSCORRUPTED;
4840 		}
4841 
4842 		if (f2fs_test_bit(blkofs, se->cur_valid_map))
4843 			goto out;
4844 
4845 		if (curseg->alloc_type == SSR)
4846 			continue;
4847 
4848 		for (blkofs += 1; blkofs < BLKS_PER_SEG(sbi); blkofs++) {
4849 			if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4850 				continue;
4851 out:
4852 			f2fs_err(sbi,
4853 				 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4854 				 i, curseg->segno, curseg->alloc_type,
4855 				 curseg->next_blkoff, blkofs);
4856 			f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4857 			return -EFSCORRUPTED;
4858 		}
4859 	}
4860 	return 0;
4861 }
4862 
4863 #ifdef CONFIG_BLK_DEV_ZONED
4864 
4865 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4866 				    struct f2fs_dev_info *fdev,
4867 				    struct blk_zone *zone)
4868 {
4869 	unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4870 	block_t zone_block, wp_block, last_valid_block;
4871 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4872 	int i, s, b, ret;
4873 	struct seg_entry *se;
4874 
4875 	if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4876 		return 0;
4877 
4878 	wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4879 	wp_segno = GET_SEGNO(sbi, wp_block);
4880 	wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4881 	zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4882 	zone_segno = GET_SEGNO(sbi, zone_block);
4883 	zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4884 
4885 	if (zone_segno >= MAIN_SEGS(sbi))
4886 		return 0;
4887 
4888 	/*
4889 	 * Skip check of zones cursegs point to, since
4890 	 * fix_curseg_write_pointer() checks them.
4891 	 */
4892 	for (i = 0; i < NO_CHECK_TYPE; i++)
4893 		if (zone_secno == GET_SEC_FROM_SEG(sbi,
4894 						   CURSEG_I(sbi, i)->segno))
4895 			return 0;
4896 
4897 	/*
4898 	 * Get last valid block of the zone.
4899 	 */
4900 	last_valid_block = zone_block - 1;
4901 	for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4902 		segno = zone_segno + s;
4903 		se = get_seg_entry(sbi, segno);
4904 		for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4905 			if (f2fs_test_bit(b, se->cur_valid_map)) {
4906 				last_valid_block = START_BLOCK(sbi, segno) + b;
4907 				break;
4908 			}
4909 		if (last_valid_block >= zone_block)
4910 			break;
4911 	}
4912 
4913 	/*
4914 	 * The write pointer matches with the valid blocks or
4915 	 * already points to the end of the zone.
4916 	 */
4917 	if ((last_valid_block + 1 == wp_block) ||
4918 			(zone->wp == zone->start + zone->len))
4919 		return 0;
4920 
4921 	if (last_valid_block + 1 == zone_block) {
4922 		/*
4923 		 * If there is no valid block in the zone and if write pointer
4924 		 * is not at zone start, reset the write pointer.
4925 		 */
4926 		f2fs_notice(sbi,
4927 			    "Zone without valid block has non-zero write "
4928 			    "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4929 			    wp_segno, wp_blkoff);
4930 		ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4931 					zone->len >> log_sectors_per_block);
4932 		if (ret)
4933 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4934 				 fdev->path, ret);
4935 
4936 		return ret;
4937 	}
4938 
4939 	/*
4940 	 * If there are valid blocks and the write pointer doesn't
4941 	 * match with them, we need to report the inconsistency and
4942 	 * fill the zone till the end to close the zone. This inconsistency
4943 	 * does not cause write error because the zone will not be selected
4944 	 * for write operation until it get discarded.
4945 	 */
4946 	f2fs_notice(sbi, "Valid blocks are not aligned with write pointer: "
4947 		    "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4948 		    GET_SEGNO(sbi, last_valid_block),
4949 		    GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4950 		    wp_segno, wp_blkoff);
4951 
4952 	ret = blkdev_zone_mgmt(fdev->bdev, REQ_OP_ZONE_FINISH,
4953 				zone->start, zone->len, GFP_NOFS);
4954 	if (ret == -EOPNOTSUPP) {
4955 		ret = blkdev_issue_zeroout(fdev->bdev, zone->wp,
4956 					zone->len - (zone->wp - zone->start),
4957 					GFP_NOFS, 0);
4958 		if (ret)
4959 			f2fs_err(sbi, "Fill up zone failed: %s (errno=%d)",
4960 					fdev->path, ret);
4961 	} else if (ret) {
4962 		f2fs_err(sbi, "Finishing zone failed: %s (errno=%d)",
4963 				fdev->path, ret);
4964 	}
4965 
4966 	return ret;
4967 }
4968 
4969 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4970 						  block_t zone_blkaddr)
4971 {
4972 	int i;
4973 
4974 	for (i = 0; i < sbi->s_ndevs; i++) {
4975 		if (!bdev_is_zoned(FDEV(i).bdev))
4976 			continue;
4977 		if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4978 				zone_blkaddr <= FDEV(i).end_blk))
4979 			return &FDEV(i);
4980 	}
4981 
4982 	return NULL;
4983 }
4984 
4985 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4986 			      void *data)
4987 {
4988 	memcpy(data, zone, sizeof(struct blk_zone));
4989 	return 0;
4990 }
4991 
4992 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4993 {
4994 	struct curseg_info *cs = CURSEG_I(sbi, type);
4995 	struct f2fs_dev_info *zbd;
4996 	struct blk_zone zone;
4997 	unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4998 	block_t cs_zone_block, wp_block;
4999 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
5000 	sector_t zone_sector;
5001 	int err;
5002 
5003 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
5004 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
5005 
5006 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
5007 	if (!zbd)
5008 		return 0;
5009 
5010 	/* report zone for the sector the curseg points to */
5011 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5012 		<< log_sectors_per_block;
5013 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5014 				  report_one_zone_cb, &zone);
5015 	if (err != 1) {
5016 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5017 			 zbd->path, err);
5018 		return err;
5019 	}
5020 
5021 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5022 		return 0;
5023 
5024 	wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
5025 	wp_segno = GET_SEGNO(sbi, wp_block);
5026 	wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
5027 	wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
5028 
5029 	if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
5030 		wp_sector_off == 0)
5031 		return 0;
5032 
5033 	f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
5034 		    "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
5035 		    type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
5036 
5037 	f2fs_notice(sbi, "Assign new section to curseg[%d]: "
5038 		    "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
5039 
5040 	f2fs_allocate_new_section(sbi, type, true);
5041 
5042 	/* check consistency of the zone curseg pointed to */
5043 	if (check_zone_write_pointer(sbi, zbd, &zone))
5044 		return -EIO;
5045 
5046 	/* check newly assigned zone */
5047 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
5048 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
5049 
5050 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
5051 	if (!zbd)
5052 		return 0;
5053 
5054 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5055 		<< log_sectors_per_block;
5056 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5057 				  report_one_zone_cb, &zone);
5058 	if (err != 1) {
5059 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5060 			 zbd->path, err);
5061 		return err;
5062 	}
5063 
5064 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5065 		return 0;
5066 
5067 	if (zone.wp != zone.start) {
5068 		f2fs_notice(sbi,
5069 			    "New zone for curseg[%d] is not yet discarded. "
5070 			    "Reset the zone: curseg[0x%x,0x%x]",
5071 			    type, cs->segno, cs->next_blkoff);
5072 		err = __f2fs_issue_discard_zone(sbi, zbd->bdev,	cs_zone_block,
5073 					zone.len >> log_sectors_per_block);
5074 		if (err) {
5075 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
5076 				 zbd->path, err);
5077 			return err;
5078 		}
5079 	}
5080 
5081 	return 0;
5082 }
5083 
5084 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5085 {
5086 	int i, ret;
5087 
5088 	for (i = 0; i < NR_PERSISTENT_LOG; i++) {
5089 		ret = fix_curseg_write_pointer(sbi, i);
5090 		if (ret)
5091 			return ret;
5092 	}
5093 
5094 	return 0;
5095 }
5096 
5097 struct check_zone_write_pointer_args {
5098 	struct f2fs_sb_info *sbi;
5099 	struct f2fs_dev_info *fdev;
5100 };
5101 
5102 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
5103 				      void *data)
5104 {
5105 	struct check_zone_write_pointer_args *args;
5106 
5107 	args = (struct check_zone_write_pointer_args *)data;
5108 
5109 	return check_zone_write_pointer(args->sbi, args->fdev, zone);
5110 }
5111 
5112 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5113 {
5114 	int i, ret;
5115 	struct check_zone_write_pointer_args args;
5116 
5117 	for (i = 0; i < sbi->s_ndevs; i++) {
5118 		if (!bdev_is_zoned(FDEV(i).bdev))
5119 			continue;
5120 
5121 		args.sbi = sbi;
5122 		args.fdev = &FDEV(i);
5123 		ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
5124 					  check_zone_write_pointer_cb, &args);
5125 		if (ret < 0)
5126 			return ret;
5127 	}
5128 
5129 	return 0;
5130 }
5131 
5132 /*
5133  * Return the number of usable blocks in a segment. The number of blocks
5134  * returned is always equal to the number of blocks in a segment for
5135  * segments fully contained within a sequential zone capacity or a
5136  * conventional zone. For segments partially contained in a sequential
5137  * zone capacity, the number of usable blocks up to the zone capacity
5138  * is returned. 0 is returned in all other cases.
5139  */
5140 static inline unsigned int f2fs_usable_zone_blks_in_seg(
5141 			struct f2fs_sb_info *sbi, unsigned int segno)
5142 {
5143 	block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
5144 	unsigned int secno;
5145 
5146 	if (!sbi->unusable_blocks_per_sec)
5147 		return BLKS_PER_SEG(sbi);
5148 
5149 	secno = GET_SEC_FROM_SEG(sbi, segno);
5150 	seg_start = START_BLOCK(sbi, segno);
5151 	sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5152 	sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi);
5153 
5154 	/*
5155 	 * If segment starts before zone capacity and spans beyond
5156 	 * zone capacity, then usable blocks are from seg start to
5157 	 * zone capacity. If the segment starts after the zone capacity,
5158 	 * then there are no usable blocks.
5159 	 */
5160 	if (seg_start >= sec_cap_blkaddr)
5161 		return 0;
5162 	if (seg_start + BLKS_PER_SEG(sbi) > sec_cap_blkaddr)
5163 		return sec_cap_blkaddr - seg_start;
5164 
5165 	return BLKS_PER_SEG(sbi);
5166 }
5167 #else
5168 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5169 {
5170 	return 0;
5171 }
5172 
5173 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5174 {
5175 	return 0;
5176 }
5177 
5178 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5179 							unsigned int segno)
5180 {
5181 	return 0;
5182 }
5183 
5184 #endif
5185 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5186 					unsigned int segno)
5187 {
5188 	if (f2fs_sb_has_blkzoned(sbi))
5189 		return f2fs_usable_zone_blks_in_seg(sbi, segno);
5190 
5191 	return BLKS_PER_SEG(sbi);
5192 }
5193 
5194 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5195 					unsigned int segno)
5196 {
5197 	if (f2fs_sb_has_blkzoned(sbi))
5198 		return CAP_SEGS_PER_SEC(sbi);
5199 
5200 	return SEGS_PER_SEC(sbi);
5201 }
5202 
5203 /*
5204  * Update min, max modified time for cost-benefit GC algorithm
5205  */
5206 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5207 {
5208 	struct sit_info *sit_i = SIT_I(sbi);
5209 	unsigned int segno;
5210 
5211 	down_write(&sit_i->sentry_lock);
5212 
5213 	sit_i->min_mtime = ULLONG_MAX;
5214 
5215 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) {
5216 		unsigned int i;
5217 		unsigned long long mtime = 0;
5218 
5219 		for (i = 0; i < SEGS_PER_SEC(sbi); i++)
5220 			mtime += get_seg_entry(sbi, segno + i)->mtime;
5221 
5222 		mtime = div_u64(mtime, SEGS_PER_SEC(sbi));
5223 
5224 		if (sit_i->min_mtime > mtime)
5225 			sit_i->min_mtime = mtime;
5226 	}
5227 	sit_i->max_mtime = get_mtime(sbi, false);
5228 	sit_i->dirty_max_mtime = 0;
5229 	up_write(&sit_i->sentry_lock);
5230 }
5231 
5232 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5233 {
5234 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5235 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5236 	struct f2fs_sm_info *sm_info;
5237 	int err;
5238 
5239 	sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5240 	if (!sm_info)
5241 		return -ENOMEM;
5242 
5243 	/* init sm info */
5244 	sbi->sm_info = sm_info;
5245 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5246 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5247 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5248 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5249 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5250 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5251 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5252 	sm_info->rec_prefree_segments = sm_info->main_segments *
5253 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5254 	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5255 		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5256 
5257 	if (!f2fs_lfs_mode(sbi))
5258 		sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC);
5259 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5260 	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5261 	sm_info->min_seq_blocks = BLKS_PER_SEG(sbi);
5262 	sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5263 	sm_info->min_ssr_sections = reserved_sections(sbi);
5264 
5265 	INIT_LIST_HEAD(&sm_info->sit_entry_set);
5266 
5267 	init_f2fs_rwsem(&sm_info->curseg_lock);
5268 
5269 	err = f2fs_create_flush_cmd_control(sbi);
5270 	if (err)
5271 		return err;
5272 
5273 	err = create_discard_cmd_control(sbi);
5274 	if (err)
5275 		return err;
5276 
5277 	err = build_sit_info(sbi);
5278 	if (err)
5279 		return err;
5280 	err = build_free_segmap(sbi);
5281 	if (err)
5282 		return err;
5283 	err = build_curseg(sbi);
5284 	if (err)
5285 		return err;
5286 
5287 	/* reinit free segmap based on SIT */
5288 	err = build_sit_entries(sbi);
5289 	if (err)
5290 		return err;
5291 
5292 	init_free_segmap(sbi);
5293 	err = build_dirty_segmap(sbi);
5294 	if (err)
5295 		return err;
5296 
5297 	err = sanity_check_curseg(sbi);
5298 	if (err)
5299 		return err;
5300 
5301 	init_min_max_mtime(sbi);
5302 	return 0;
5303 }
5304 
5305 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5306 		enum dirty_type dirty_type)
5307 {
5308 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5309 
5310 	mutex_lock(&dirty_i->seglist_lock);
5311 	kvfree(dirty_i->dirty_segmap[dirty_type]);
5312 	dirty_i->nr_dirty[dirty_type] = 0;
5313 	mutex_unlock(&dirty_i->seglist_lock);
5314 }
5315 
5316 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5317 {
5318 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5319 
5320 	kvfree(dirty_i->pinned_secmap);
5321 	kvfree(dirty_i->victim_secmap);
5322 }
5323 
5324 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5325 {
5326 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5327 	int i;
5328 
5329 	if (!dirty_i)
5330 		return;
5331 
5332 	/* discard pre-free/dirty segments list */
5333 	for (i = 0; i < NR_DIRTY_TYPE; i++)
5334 		discard_dirty_segmap(sbi, i);
5335 
5336 	if (__is_large_section(sbi)) {
5337 		mutex_lock(&dirty_i->seglist_lock);
5338 		kvfree(dirty_i->dirty_secmap);
5339 		mutex_unlock(&dirty_i->seglist_lock);
5340 	}
5341 
5342 	destroy_victim_secmap(sbi);
5343 	SM_I(sbi)->dirty_info = NULL;
5344 	kfree(dirty_i);
5345 }
5346 
5347 static void destroy_curseg(struct f2fs_sb_info *sbi)
5348 {
5349 	struct curseg_info *array = SM_I(sbi)->curseg_array;
5350 	int i;
5351 
5352 	if (!array)
5353 		return;
5354 	SM_I(sbi)->curseg_array = NULL;
5355 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
5356 		kfree(array[i].sum_blk);
5357 		kfree(array[i].journal);
5358 	}
5359 	kfree(array);
5360 }
5361 
5362 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5363 {
5364 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5365 
5366 	if (!free_i)
5367 		return;
5368 	SM_I(sbi)->free_info = NULL;
5369 	kvfree(free_i->free_segmap);
5370 	kvfree(free_i->free_secmap);
5371 	kfree(free_i);
5372 }
5373 
5374 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5375 {
5376 	struct sit_info *sit_i = SIT_I(sbi);
5377 
5378 	if (!sit_i)
5379 		return;
5380 
5381 	if (sit_i->sentries)
5382 		kvfree(sit_i->bitmap);
5383 	kfree(sit_i->tmp_map);
5384 
5385 	kvfree(sit_i->sentries);
5386 	kvfree(sit_i->sec_entries);
5387 	kvfree(sit_i->dirty_sentries_bitmap);
5388 
5389 	SM_I(sbi)->sit_info = NULL;
5390 	kvfree(sit_i->sit_bitmap);
5391 #ifdef CONFIG_F2FS_CHECK_FS
5392 	kvfree(sit_i->sit_bitmap_mir);
5393 	kvfree(sit_i->invalid_segmap);
5394 #endif
5395 	kfree(sit_i);
5396 }
5397 
5398 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5399 {
5400 	struct f2fs_sm_info *sm_info = SM_I(sbi);
5401 
5402 	if (!sm_info)
5403 		return;
5404 	f2fs_destroy_flush_cmd_control(sbi, true);
5405 	destroy_discard_cmd_control(sbi);
5406 	destroy_dirty_segmap(sbi);
5407 	destroy_curseg(sbi);
5408 	destroy_free_segmap(sbi);
5409 	destroy_sit_info(sbi);
5410 	sbi->sm_info = NULL;
5411 	kfree(sm_info);
5412 }
5413 
5414 int __init f2fs_create_segment_manager_caches(void)
5415 {
5416 	discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5417 			sizeof(struct discard_entry));
5418 	if (!discard_entry_slab)
5419 		goto fail;
5420 
5421 	discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5422 			sizeof(struct discard_cmd));
5423 	if (!discard_cmd_slab)
5424 		goto destroy_discard_entry;
5425 
5426 	sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5427 			sizeof(struct sit_entry_set));
5428 	if (!sit_entry_set_slab)
5429 		goto destroy_discard_cmd;
5430 
5431 	revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry",
5432 			sizeof(struct revoke_entry));
5433 	if (!revoke_entry_slab)
5434 		goto destroy_sit_entry_set;
5435 	return 0;
5436 
5437 destroy_sit_entry_set:
5438 	kmem_cache_destroy(sit_entry_set_slab);
5439 destroy_discard_cmd:
5440 	kmem_cache_destroy(discard_cmd_slab);
5441 destroy_discard_entry:
5442 	kmem_cache_destroy(discard_entry_slab);
5443 fail:
5444 	return -ENOMEM;
5445 }
5446 
5447 void f2fs_destroy_segment_manager_caches(void)
5448 {
5449 	kmem_cache_destroy(sit_entry_set_slab);
5450 	kmem_cache_destroy(discard_cmd_slab);
5451 	kmem_cache_destroy(discard_entry_slab);
5452 	kmem_cache_destroy(revoke_entry_slab);
5453 }
5454