xref: /linux/fs/f2fs/segment.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
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 	if (is_inode_flag_set(inode, FI_ATOMIC_DIRTIED)) {
203 		clear_inode_flag(inode, FI_ATOMIC_DIRTIED);
204 		f2fs_mark_inode_dirty_sync(inode, true);
205 	}
206 	stat_dec_atomic_inode(inode);
207 
208 	F2FS_I(inode)->atomic_write_task = NULL;
209 
210 	if (clean) {
211 		f2fs_i_size_write(inode, fi->original_i_size);
212 		fi->original_i_size = 0;
213 	}
214 	/* avoid stale dirty inode during eviction */
215 	sync_inode_metadata(inode, 0);
216 }
217 
218 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index,
219 			block_t new_addr, block_t *old_addr, bool recover)
220 {
221 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
222 	struct dnode_of_data dn;
223 	struct node_info ni;
224 	int err;
225 
226 retry:
227 	set_new_dnode(&dn, inode, NULL, NULL, 0);
228 	err = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE);
229 	if (err) {
230 		if (err == -ENOMEM) {
231 			f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
232 			goto retry;
233 		}
234 		return err;
235 	}
236 
237 	err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
238 	if (err) {
239 		f2fs_put_dnode(&dn);
240 		return err;
241 	}
242 
243 	if (recover) {
244 		/* dn.data_blkaddr is always valid */
245 		if (!__is_valid_data_blkaddr(new_addr)) {
246 			if (new_addr == NULL_ADDR)
247 				dec_valid_block_count(sbi, inode, 1);
248 			f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
249 			f2fs_update_data_blkaddr(&dn, new_addr);
250 		} else {
251 			f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
252 				new_addr, ni.version, true, true);
253 		}
254 	} else {
255 		blkcnt_t count = 1;
256 
257 		err = inc_valid_block_count(sbi, inode, &count, true);
258 		if (err) {
259 			f2fs_put_dnode(&dn);
260 			return err;
261 		}
262 
263 		*old_addr = dn.data_blkaddr;
264 		f2fs_truncate_data_blocks_range(&dn, 1);
265 		dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count);
266 
267 		f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr,
268 					ni.version, true, false);
269 	}
270 
271 	f2fs_put_dnode(&dn);
272 
273 	trace_f2fs_replace_atomic_write_block(inode, F2FS_I(inode)->cow_inode,
274 			index, old_addr ? *old_addr : 0, new_addr, recover);
275 	return 0;
276 }
277 
278 static void __complete_revoke_list(struct inode *inode, struct list_head *head,
279 					bool revoke)
280 {
281 	struct revoke_entry *cur, *tmp;
282 	pgoff_t start_index = 0;
283 	bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE);
284 
285 	list_for_each_entry_safe(cur, tmp, head, list) {
286 		if (revoke) {
287 			__replace_atomic_write_block(inode, cur->index,
288 						cur->old_addr, NULL, true);
289 		} else if (truncate) {
290 			f2fs_truncate_hole(inode, start_index, cur->index);
291 			start_index = cur->index + 1;
292 		}
293 
294 		list_del(&cur->list);
295 		kmem_cache_free(revoke_entry_slab, cur);
296 	}
297 
298 	if (!revoke && truncate)
299 		f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false);
300 }
301 
302 static int __f2fs_commit_atomic_write(struct inode *inode)
303 {
304 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
305 	struct f2fs_inode_info *fi = F2FS_I(inode);
306 	struct inode *cow_inode = fi->cow_inode;
307 	struct revoke_entry *new;
308 	struct list_head revoke_list;
309 	block_t blkaddr;
310 	struct dnode_of_data dn;
311 	pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
312 	pgoff_t off = 0, blen, index;
313 	int ret = 0, i;
314 
315 	INIT_LIST_HEAD(&revoke_list);
316 
317 	while (len) {
318 		blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len);
319 
320 		set_new_dnode(&dn, cow_inode, NULL, NULL, 0);
321 		ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA);
322 		if (ret && ret != -ENOENT) {
323 			goto out;
324 		} else if (ret == -ENOENT) {
325 			ret = 0;
326 			if (dn.max_level == 0)
327 				goto out;
328 			goto next;
329 		}
330 
331 		blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode),
332 				len);
333 		index = off;
334 		for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) {
335 			blkaddr = f2fs_data_blkaddr(&dn);
336 
337 			if (!__is_valid_data_blkaddr(blkaddr)) {
338 				continue;
339 			} else if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
340 					DATA_GENERIC_ENHANCE)) {
341 				f2fs_put_dnode(&dn);
342 				ret = -EFSCORRUPTED;
343 				goto out;
344 			}
345 
346 			new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS,
347 							true, NULL);
348 
349 			ret = __replace_atomic_write_block(inode, index, blkaddr,
350 							&new->old_addr, false);
351 			if (ret) {
352 				f2fs_put_dnode(&dn);
353 				kmem_cache_free(revoke_entry_slab, new);
354 				goto out;
355 			}
356 
357 			f2fs_update_data_blkaddr(&dn, NULL_ADDR);
358 			new->index = index;
359 			list_add_tail(&new->list, &revoke_list);
360 		}
361 		f2fs_put_dnode(&dn);
362 next:
363 		off += blen;
364 		len -= blen;
365 	}
366 
367 out:
368 	if (ret) {
369 		sbi->revoked_atomic_block += fi->atomic_write_cnt;
370 	} else {
371 		sbi->committed_atomic_block += fi->atomic_write_cnt;
372 		set_inode_flag(inode, FI_ATOMIC_COMMITTED);
373 		if (is_inode_flag_set(inode, FI_ATOMIC_DIRTIED)) {
374 			clear_inode_flag(inode, FI_ATOMIC_DIRTIED);
375 			f2fs_mark_inode_dirty_sync(inode, true);
376 		}
377 	}
378 
379 	__complete_revoke_list(inode, &revoke_list, ret ? true : false);
380 
381 	return ret;
382 }
383 
384 int f2fs_commit_atomic_write(struct inode *inode)
385 {
386 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
387 	struct f2fs_inode_info *fi = F2FS_I(inode);
388 	int err;
389 
390 	err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
391 	if (err)
392 		return err;
393 
394 	f2fs_down_write(&fi->i_gc_rwsem[WRITE]);
395 	f2fs_lock_op(sbi);
396 
397 	err = __f2fs_commit_atomic_write(inode);
398 
399 	f2fs_unlock_op(sbi);
400 	f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
401 
402 	return err;
403 }
404 
405 /*
406  * This function balances dirty node and dentry pages.
407  * In addition, it controls garbage collection.
408  */
409 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
410 {
411 	if (f2fs_cp_error(sbi))
412 		return;
413 
414 	if (time_to_inject(sbi, FAULT_CHECKPOINT))
415 		f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT);
416 
417 	/* balance_fs_bg is able to be pending */
418 	if (need && excess_cached_nats(sbi))
419 		f2fs_balance_fs_bg(sbi, false);
420 
421 	if (!f2fs_is_checkpoint_ready(sbi))
422 		return;
423 
424 	/*
425 	 * We should do GC or end up with checkpoint, if there are so many dirty
426 	 * dir/node pages without enough free segments.
427 	 */
428 	if (has_enough_free_secs(sbi, 0, 0))
429 		return;
430 
431 	if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
432 				sbi->gc_thread->f2fs_gc_task) {
433 		DEFINE_WAIT(wait);
434 
435 		prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
436 					TASK_UNINTERRUPTIBLE);
437 		wake_up(&sbi->gc_thread->gc_wait_queue_head);
438 		io_schedule();
439 		finish_wait(&sbi->gc_thread->fggc_wq, &wait);
440 	} else {
441 		struct f2fs_gc_control gc_control = {
442 			.victim_segno = NULL_SEGNO,
443 			.init_gc_type = BG_GC,
444 			.no_bg_gc = true,
445 			.should_migrate_blocks = false,
446 			.err_gc_skipped = false,
447 			.nr_free_secs = 1 };
448 		f2fs_down_write(&sbi->gc_lock);
449 		stat_inc_gc_call_count(sbi, FOREGROUND);
450 		f2fs_gc(sbi, &gc_control);
451 	}
452 }
453 
454 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi)
455 {
456 	int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2;
457 	unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS);
458 	unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA);
459 	unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES);
460 	unsigned int meta = get_pages(sbi, F2FS_DIRTY_META);
461 	unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA);
462 	unsigned int threshold =
463 		SEGS_TO_BLKS(sbi, (factor * DEFAULT_DIRTY_THRESHOLD));
464 	unsigned int global_threshold = threshold * 3 / 2;
465 
466 	if (dents >= threshold || qdata >= threshold ||
467 		nodes >= threshold || meta >= threshold ||
468 		imeta >= threshold)
469 		return true;
470 	return dents + qdata + nodes + meta + imeta >  global_threshold;
471 }
472 
473 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
474 {
475 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
476 		return;
477 
478 	/* try to shrink extent cache when there is no enough memory */
479 	if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE))
480 		f2fs_shrink_read_extent_tree(sbi,
481 				READ_EXTENT_CACHE_SHRINK_NUMBER);
482 
483 	/* try to shrink age extent cache when there is no enough memory */
484 	if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE))
485 		f2fs_shrink_age_extent_tree(sbi,
486 				AGE_EXTENT_CACHE_SHRINK_NUMBER);
487 
488 	/* check the # of cached NAT entries */
489 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
490 		f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
491 
492 	if (!f2fs_available_free_memory(sbi, FREE_NIDS))
493 		f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
494 	else
495 		f2fs_build_free_nids(sbi, false, false);
496 
497 	if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) ||
498 		excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi))
499 		goto do_sync;
500 
501 	/* there is background inflight IO or foreground operation recently */
502 	if (is_inflight_io(sbi, REQ_TIME) ||
503 		(!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem)))
504 		return;
505 
506 	/* exceed periodical checkpoint timeout threshold */
507 	if (f2fs_time_over(sbi, CP_TIME))
508 		goto do_sync;
509 
510 	/* checkpoint is the only way to shrink partial cached entries */
511 	if (f2fs_available_free_memory(sbi, NAT_ENTRIES) &&
512 		f2fs_available_free_memory(sbi, INO_ENTRIES))
513 		return;
514 
515 do_sync:
516 	if (test_opt(sbi, DATA_FLUSH) && from_bg) {
517 		struct blk_plug plug;
518 
519 		mutex_lock(&sbi->flush_lock);
520 
521 		blk_start_plug(&plug);
522 		f2fs_sync_dirty_inodes(sbi, FILE_INODE, false);
523 		blk_finish_plug(&plug);
524 
525 		mutex_unlock(&sbi->flush_lock);
526 	}
527 	stat_inc_cp_call_count(sbi, BACKGROUND);
528 	f2fs_sync_fs(sbi->sb, 1);
529 }
530 
531 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
532 				struct block_device *bdev)
533 {
534 	int ret = blkdev_issue_flush(bdev);
535 
536 	trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
537 				test_opt(sbi, FLUSH_MERGE), ret);
538 	if (!ret)
539 		f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0);
540 	return ret;
541 }
542 
543 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
544 {
545 	int ret = 0;
546 	int i;
547 
548 	if (!f2fs_is_multi_device(sbi))
549 		return __submit_flush_wait(sbi, sbi->sb->s_bdev);
550 
551 	for (i = 0; i < sbi->s_ndevs; i++) {
552 		if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
553 			continue;
554 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
555 		if (ret)
556 			break;
557 	}
558 	return ret;
559 }
560 
561 static int issue_flush_thread(void *data)
562 {
563 	struct f2fs_sb_info *sbi = data;
564 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
565 	wait_queue_head_t *q = &fcc->flush_wait_queue;
566 repeat:
567 	if (kthread_should_stop())
568 		return 0;
569 
570 	if (!llist_empty(&fcc->issue_list)) {
571 		struct flush_cmd *cmd, *next;
572 		int ret;
573 
574 		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
575 		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
576 
577 		cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
578 
579 		ret = submit_flush_wait(sbi, cmd->ino);
580 		atomic_inc(&fcc->issued_flush);
581 
582 		llist_for_each_entry_safe(cmd, next,
583 					  fcc->dispatch_list, llnode) {
584 			cmd->ret = ret;
585 			complete(&cmd->wait);
586 		}
587 		fcc->dispatch_list = NULL;
588 	}
589 
590 	wait_event_interruptible(*q,
591 		kthread_should_stop() || !llist_empty(&fcc->issue_list));
592 	goto repeat;
593 }
594 
595 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
596 {
597 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
598 	struct flush_cmd cmd;
599 	int ret;
600 
601 	if (test_opt(sbi, NOBARRIER))
602 		return 0;
603 
604 	if (!test_opt(sbi, FLUSH_MERGE)) {
605 		atomic_inc(&fcc->queued_flush);
606 		ret = submit_flush_wait(sbi, ino);
607 		atomic_dec(&fcc->queued_flush);
608 		atomic_inc(&fcc->issued_flush);
609 		return ret;
610 	}
611 
612 	if (atomic_inc_return(&fcc->queued_flush) == 1 ||
613 	    f2fs_is_multi_device(sbi)) {
614 		ret = submit_flush_wait(sbi, ino);
615 		atomic_dec(&fcc->queued_flush);
616 
617 		atomic_inc(&fcc->issued_flush);
618 		return ret;
619 	}
620 
621 	cmd.ino = ino;
622 	init_completion(&cmd.wait);
623 
624 	llist_add(&cmd.llnode, &fcc->issue_list);
625 
626 	/*
627 	 * update issue_list before we wake up issue_flush thread, this
628 	 * smp_mb() pairs with another barrier in ___wait_event(), see
629 	 * more details in comments of waitqueue_active().
630 	 */
631 	smp_mb();
632 
633 	if (waitqueue_active(&fcc->flush_wait_queue))
634 		wake_up(&fcc->flush_wait_queue);
635 
636 	if (fcc->f2fs_issue_flush) {
637 		wait_for_completion(&cmd.wait);
638 		atomic_dec(&fcc->queued_flush);
639 	} else {
640 		struct llist_node *list;
641 
642 		list = llist_del_all(&fcc->issue_list);
643 		if (!list) {
644 			wait_for_completion(&cmd.wait);
645 			atomic_dec(&fcc->queued_flush);
646 		} else {
647 			struct flush_cmd *tmp, *next;
648 
649 			ret = submit_flush_wait(sbi, ino);
650 
651 			llist_for_each_entry_safe(tmp, next, list, llnode) {
652 				if (tmp == &cmd) {
653 					cmd.ret = ret;
654 					atomic_dec(&fcc->queued_flush);
655 					continue;
656 				}
657 				tmp->ret = ret;
658 				complete(&tmp->wait);
659 			}
660 		}
661 	}
662 
663 	return cmd.ret;
664 }
665 
666 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
667 {
668 	dev_t dev = sbi->sb->s_bdev->bd_dev;
669 	struct flush_cmd_control *fcc;
670 
671 	if (SM_I(sbi)->fcc_info) {
672 		fcc = SM_I(sbi)->fcc_info;
673 		if (fcc->f2fs_issue_flush)
674 			return 0;
675 		goto init_thread;
676 	}
677 
678 	fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
679 	if (!fcc)
680 		return -ENOMEM;
681 	atomic_set(&fcc->issued_flush, 0);
682 	atomic_set(&fcc->queued_flush, 0);
683 	init_waitqueue_head(&fcc->flush_wait_queue);
684 	init_llist_head(&fcc->issue_list);
685 	SM_I(sbi)->fcc_info = fcc;
686 	if (!test_opt(sbi, FLUSH_MERGE))
687 		return 0;
688 
689 init_thread:
690 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
691 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
692 	if (IS_ERR(fcc->f2fs_issue_flush)) {
693 		int err = PTR_ERR(fcc->f2fs_issue_flush);
694 
695 		fcc->f2fs_issue_flush = NULL;
696 		return err;
697 	}
698 
699 	return 0;
700 }
701 
702 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
703 {
704 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
705 
706 	if (fcc && fcc->f2fs_issue_flush) {
707 		struct task_struct *flush_thread = fcc->f2fs_issue_flush;
708 
709 		fcc->f2fs_issue_flush = NULL;
710 		kthread_stop(flush_thread);
711 	}
712 	if (free) {
713 		kfree(fcc);
714 		SM_I(sbi)->fcc_info = NULL;
715 	}
716 }
717 
718 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
719 {
720 	int ret = 0, i;
721 
722 	if (!f2fs_is_multi_device(sbi))
723 		return 0;
724 
725 	if (test_opt(sbi, NOBARRIER))
726 		return 0;
727 
728 	for (i = 1; i < sbi->s_ndevs; i++) {
729 		int count = DEFAULT_RETRY_IO_COUNT;
730 
731 		if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
732 			continue;
733 
734 		do {
735 			ret = __submit_flush_wait(sbi, FDEV(i).bdev);
736 			if (ret)
737 				f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
738 		} while (ret && --count);
739 
740 		if (ret) {
741 			f2fs_stop_checkpoint(sbi, false,
742 					STOP_CP_REASON_FLUSH_FAIL);
743 			break;
744 		}
745 
746 		spin_lock(&sbi->dev_lock);
747 		f2fs_clear_bit(i, (char *)&sbi->dirty_device);
748 		spin_unlock(&sbi->dev_lock);
749 	}
750 
751 	return ret;
752 }
753 
754 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
755 		enum dirty_type dirty_type)
756 {
757 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
758 
759 	/* need not be added */
760 	if (IS_CURSEG(sbi, segno))
761 		return;
762 
763 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
764 		dirty_i->nr_dirty[dirty_type]++;
765 
766 	if (dirty_type == DIRTY) {
767 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
768 		enum dirty_type t = sentry->type;
769 
770 		if (unlikely(t >= DIRTY)) {
771 			f2fs_bug_on(sbi, 1);
772 			return;
773 		}
774 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
775 			dirty_i->nr_dirty[t]++;
776 
777 		if (__is_large_section(sbi)) {
778 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
779 			block_t valid_blocks =
780 				get_valid_blocks(sbi, segno, true);
781 
782 			f2fs_bug_on(sbi,
783 				(!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
784 				!valid_blocks) ||
785 				valid_blocks == CAP_BLKS_PER_SEC(sbi));
786 
787 			if (!IS_CURSEC(sbi, secno))
788 				set_bit(secno, dirty_i->dirty_secmap);
789 		}
790 	}
791 }
792 
793 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
794 		enum dirty_type dirty_type)
795 {
796 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
797 	block_t valid_blocks;
798 
799 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
800 		dirty_i->nr_dirty[dirty_type]--;
801 
802 	if (dirty_type == DIRTY) {
803 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
804 		enum dirty_type t = sentry->type;
805 
806 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
807 			dirty_i->nr_dirty[t]--;
808 
809 		valid_blocks = get_valid_blocks(sbi, segno, true);
810 		if (valid_blocks == 0) {
811 			clear_bit(GET_SEC_FROM_SEG(sbi, segno),
812 						dirty_i->victim_secmap);
813 #ifdef CONFIG_F2FS_CHECK_FS
814 			clear_bit(segno, SIT_I(sbi)->invalid_segmap);
815 #endif
816 		}
817 		if (__is_large_section(sbi)) {
818 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
819 
820 			if (!valid_blocks ||
821 					valid_blocks == CAP_BLKS_PER_SEC(sbi)) {
822 				clear_bit(secno, dirty_i->dirty_secmap);
823 				return;
824 			}
825 
826 			if (!IS_CURSEC(sbi, secno))
827 				set_bit(secno, dirty_i->dirty_secmap);
828 		}
829 	}
830 }
831 
832 /*
833  * Should not occur error such as -ENOMEM.
834  * Adding dirty entry into seglist is not critical operation.
835  * If a given segment is one of current working segments, it won't be added.
836  */
837 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
838 {
839 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
840 	unsigned short valid_blocks, ckpt_valid_blocks;
841 	unsigned int usable_blocks;
842 
843 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
844 		return;
845 
846 	usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
847 	mutex_lock(&dirty_i->seglist_lock);
848 
849 	valid_blocks = get_valid_blocks(sbi, segno, false);
850 	ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
851 
852 	if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
853 		ckpt_valid_blocks == usable_blocks)) {
854 		__locate_dirty_segment(sbi, segno, PRE);
855 		__remove_dirty_segment(sbi, segno, DIRTY);
856 	} else if (valid_blocks < usable_blocks) {
857 		__locate_dirty_segment(sbi, segno, DIRTY);
858 	} else {
859 		/* Recovery routine with SSR needs this */
860 		__remove_dirty_segment(sbi, segno, DIRTY);
861 	}
862 
863 	mutex_unlock(&dirty_i->seglist_lock);
864 }
865 
866 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
867 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
868 {
869 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
870 	unsigned int segno;
871 
872 	mutex_lock(&dirty_i->seglist_lock);
873 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
874 		if (get_valid_blocks(sbi, segno, false))
875 			continue;
876 		if (IS_CURSEG(sbi, segno))
877 			continue;
878 		__locate_dirty_segment(sbi, segno, PRE);
879 		__remove_dirty_segment(sbi, segno, DIRTY);
880 	}
881 	mutex_unlock(&dirty_i->seglist_lock);
882 }
883 
884 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
885 {
886 	int ovp_hole_segs =
887 		(overprovision_segments(sbi) - reserved_segments(sbi));
888 	block_t ovp_holes = SEGS_TO_BLKS(sbi, ovp_hole_segs);
889 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
890 	block_t holes[2] = {0, 0};	/* DATA and NODE */
891 	block_t unusable;
892 	struct seg_entry *se;
893 	unsigned int segno;
894 
895 	mutex_lock(&dirty_i->seglist_lock);
896 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
897 		se = get_seg_entry(sbi, segno);
898 		if (IS_NODESEG(se->type))
899 			holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
900 							se->valid_blocks;
901 		else
902 			holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
903 							se->valid_blocks;
904 	}
905 	mutex_unlock(&dirty_i->seglist_lock);
906 
907 	unusable = max(holes[DATA], holes[NODE]);
908 	if (unusable > ovp_holes)
909 		return unusable - ovp_holes;
910 	return 0;
911 }
912 
913 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
914 {
915 	int ovp_hole_segs =
916 		(overprovision_segments(sbi) - reserved_segments(sbi));
917 
918 	if (F2FS_OPTION(sbi).unusable_cap_perc == 100)
919 		return 0;
920 	if (unusable > F2FS_OPTION(sbi).unusable_cap)
921 		return -EAGAIN;
922 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
923 		dirty_segments(sbi) > ovp_hole_segs)
924 		return -EAGAIN;
925 	if (has_not_enough_free_secs(sbi, 0, 0))
926 		return -EAGAIN;
927 	return 0;
928 }
929 
930 /* This is only used by SBI_CP_DISABLED */
931 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
932 {
933 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
934 	unsigned int segno = 0;
935 
936 	mutex_lock(&dirty_i->seglist_lock);
937 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
938 		if (get_valid_blocks(sbi, segno, false))
939 			continue;
940 		if (get_ckpt_valid_blocks(sbi, segno, false))
941 			continue;
942 		mutex_unlock(&dirty_i->seglist_lock);
943 		return segno;
944 	}
945 	mutex_unlock(&dirty_i->seglist_lock);
946 	return NULL_SEGNO;
947 }
948 
949 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
950 		struct block_device *bdev, block_t lstart,
951 		block_t start, block_t len)
952 {
953 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
954 	struct list_head *pend_list;
955 	struct discard_cmd *dc;
956 
957 	f2fs_bug_on(sbi, !len);
958 
959 	pend_list = &dcc->pend_list[plist_idx(len)];
960 
961 	dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL);
962 	INIT_LIST_HEAD(&dc->list);
963 	dc->bdev = bdev;
964 	dc->di.lstart = lstart;
965 	dc->di.start = start;
966 	dc->di.len = len;
967 	dc->ref = 0;
968 	dc->state = D_PREP;
969 	dc->queued = 0;
970 	dc->error = 0;
971 	init_completion(&dc->wait);
972 	list_add_tail(&dc->list, pend_list);
973 	spin_lock_init(&dc->lock);
974 	dc->bio_ref = 0;
975 	atomic_inc(&dcc->discard_cmd_cnt);
976 	dcc->undiscard_blks += len;
977 
978 	return dc;
979 }
980 
981 static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi)
982 {
983 #ifdef CONFIG_F2FS_CHECK_FS
984 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
985 	struct rb_node *cur = rb_first_cached(&dcc->root), *next;
986 	struct discard_cmd *cur_dc, *next_dc;
987 
988 	while (cur) {
989 		next = rb_next(cur);
990 		if (!next)
991 			return true;
992 
993 		cur_dc = rb_entry(cur, struct discard_cmd, rb_node);
994 		next_dc = rb_entry(next, struct discard_cmd, rb_node);
995 
996 		if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) {
997 			f2fs_info(sbi, "broken discard_rbtree, "
998 				"cur(%u, %u) next(%u, %u)",
999 				cur_dc->di.lstart, cur_dc->di.len,
1000 				next_dc->di.lstart, next_dc->di.len);
1001 			return false;
1002 		}
1003 		cur = next;
1004 	}
1005 #endif
1006 	return true;
1007 }
1008 
1009 static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi,
1010 						block_t blkaddr)
1011 {
1012 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1013 	struct rb_node *node = dcc->root.rb_root.rb_node;
1014 	struct discard_cmd *dc;
1015 
1016 	while (node) {
1017 		dc = rb_entry(node, struct discard_cmd, rb_node);
1018 
1019 		if (blkaddr < dc->di.lstart)
1020 			node = node->rb_left;
1021 		else if (blkaddr >= dc->di.lstart + dc->di.len)
1022 			node = node->rb_right;
1023 		else
1024 			return dc;
1025 	}
1026 	return NULL;
1027 }
1028 
1029 static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root,
1030 				block_t blkaddr,
1031 				struct discard_cmd **prev_entry,
1032 				struct discard_cmd **next_entry,
1033 				struct rb_node ***insert_p,
1034 				struct rb_node **insert_parent)
1035 {
1036 	struct rb_node **pnode = &root->rb_root.rb_node;
1037 	struct rb_node *parent = NULL, *tmp_node;
1038 	struct discard_cmd *dc;
1039 
1040 	*insert_p = NULL;
1041 	*insert_parent = NULL;
1042 	*prev_entry = NULL;
1043 	*next_entry = NULL;
1044 
1045 	if (RB_EMPTY_ROOT(&root->rb_root))
1046 		return NULL;
1047 
1048 	while (*pnode) {
1049 		parent = *pnode;
1050 		dc = rb_entry(*pnode, struct discard_cmd, rb_node);
1051 
1052 		if (blkaddr < dc->di.lstart)
1053 			pnode = &(*pnode)->rb_left;
1054 		else if (blkaddr >= dc->di.lstart + dc->di.len)
1055 			pnode = &(*pnode)->rb_right;
1056 		else
1057 			goto lookup_neighbors;
1058 	}
1059 
1060 	*insert_p = pnode;
1061 	*insert_parent = parent;
1062 
1063 	dc = rb_entry(parent, struct discard_cmd, rb_node);
1064 	tmp_node = parent;
1065 	if (parent && blkaddr > dc->di.lstart)
1066 		tmp_node = rb_next(parent);
1067 	*next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1068 
1069 	tmp_node = parent;
1070 	if (parent && blkaddr < dc->di.lstart)
1071 		tmp_node = rb_prev(parent);
1072 	*prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1073 	return NULL;
1074 
1075 lookup_neighbors:
1076 	/* lookup prev node for merging backward later */
1077 	tmp_node = rb_prev(&dc->rb_node);
1078 	*prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1079 
1080 	/* lookup next node for merging frontward later */
1081 	tmp_node = rb_next(&dc->rb_node);
1082 	*next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1083 	return dc;
1084 }
1085 
1086 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1087 							struct discard_cmd *dc)
1088 {
1089 	if (dc->state == D_DONE)
1090 		atomic_sub(dc->queued, &dcc->queued_discard);
1091 
1092 	list_del(&dc->list);
1093 	rb_erase_cached(&dc->rb_node, &dcc->root);
1094 	dcc->undiscard_blks -= dc->di.len;
1095 
1096 	kmem_cache_free(discard_cmd_slab, dc);
1097 
1098 	atomic_dec(&dcc->discard_cmd_cnt);
1099 }
1100 
1101 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1102 							struct discard_cmd *dc)
1103 {
1104 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1105 	unsigned long flags;
1106 
1107 	trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len);
1108 
1109 	spin_lock_irqsave(&dc->lock, flags);
1110 	if (dc->bio_ref) {
1111 		spin_unlock_irqrestore(&dc->lock, flags);
1112 		return;
1113 	}
1114 	spin_unlock_irqrestore(&dc->lock, flags);
1115 
1116 	f2fs_bug_on(sbi, dc->ref);
1117 
1118 	if (dc->error == -EOPNOTSUPP)
1119 		dc->error = 0;
1120 
1121 	if (dc->error)
1122 		f2fs_info_ratelimited(sbi,
1123 			"Issue discard(%u, %u, %u) failed, ret: %d",
1124 			dc->di.lstart, dc->di.start, dc->di.len, dc->error);
1125 	__detach_discard_cmd(dcc, dc);
1126 }
1127 
1128 static void f2fs_submit_discard_endio(struct bio *bio)
1129 {
1130 	struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1131 	unsigned long flags;
1132 
1133 	spin_lock_irqsave(&dc->lock, flags);
1134 	if (!dc->error)
1135 		dc->error = blk_status_to_errno(bio->bi_status);
1136 	dc->bio_ref--;
1137 	if (!dc->bio_ref && dc->state == D_SUBMIT) {
1138 		dc->state = D_DONE;
1139 		complete_all(&dc->wait);
1140 	}
1141 	spin_unlock_irqrestore(&dc->lock, flags);
1142 	bio_put(bio);
1143 }
1144 
1145 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1146 				block_t start, block_t end)
1147 {
1148 #ifdef CONFIG_F2FS_CHECK_FS
1149 	struct seg_entry *sentry;
1150 	unsigned int segno;
1151 	block_t blk = start;
1152 	unsigned long offset, size, *map;
1153 
1154 	while (blk < end) {
1155 		segno = GET_SEGNO(sbi, blk);
1156 		sentry = get_seg_entry(sbi, segno);
1157 		offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1158 
1159 		if (end < START_BLOCK(sbi, segno + 1))
1160 			size = GET_BLKOFF_FROM_SEG0(sbi, end);
1161 		else
1162 			size = BLKS_PER_SEG(sbi);
1163 		map = (unsigned long *)(sentry->cur_valid_map);
1164 		offset = __find_rev_next_bit(map, size, offset);
1165 		f2fs_bug_on(sbi, offset != size);
1166 		blk = START_BLOCK(sbi, segno + 1);
1167 	}
1168 #endif
1169 }
1170 
1171 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1172 				struct discard_policy *dpolicy,
1173 				int discard_type, unsigned int granularity)
1174 {
1175 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1176 
1177 	/* common policy */
1178 	dpolicy->type = discard_type;
1179 	dpolicy->sync = true;
1180 	dpolicy->ordered = false;
1181 	dpolicy->granularity = granularity;
1182 
1183 	dpolicy->max_requests = dcc->max_discard_request;
1184 	dpolicy->io_aware_gran = dcc->discard_io_aware_gran;
1185 	dpolicy->timeout = false;
1186 
1187 	if (discard_type == DPOLICY_BG) {
1188 		dpolicy->min_interval = dcc->min_discard_issue_time;
1189 		dpolicy->mid_interval = dcc->mid_discard_issue_time;
1190 		dpolicy->max_interval = dcc->max_discard_issue_time;
1191 		if (dcc->discard_io_aware == DPOLICY_IO_AWARE_ENABLE)
1192 			dpolicy->io_aware = true;
1193 		else if (dcc->discard_io_aware == DPOLICY_IO_AWARE_DISABLE)
1194 			dpolicy->io_aware = false;
1195 		dpolicy->sync = false;
1196 		dpolicy->ordered = true;
1197 		if (utilization(sbi) > dcc->discard_urgent_util) {
1198 			dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1199 			if (atomic_read(&dcc->discard_cmd_cnt))
1200 				dpolicy->max_interval =
1201 					dcc->min_discard_issue_time;
1202 		}
1203 	} else if (discard_type == DPOLICY_FORCE) {
1204 		dpolicy->min_interval = dcc->min_discard_issue_time;
1205 		dpolicy->mid_interval = dcc->mid_discard_issue_time;
1206 		dpolicy->max_interval = dcc->max_discard_issue_time;
1207 		dpolicy->io_aware = false;
1208 	} else if (discard_type == DPOLICY_FSTRIM) {
1209 		dpolicy->io_aware = false;
1210 	} else if (discard_type == DPOLICY_UMOUNT) {
1211 		dpolicy->io_aware = false;
1212 		/* we need to issue all to keep CP_TRIMMED_FLAG */
1213 		dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1214 		dpolicy->timeout = true;
1215 	}
1216 }
1217 
1218 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1219 				struct block_device *bdev, block_t lstart,
1220 				block_t start, block_t len);
1221 
1222 #ifdef CONFIG_BLK_DEV_ZONED
1223 static void __submit_zone_reset_cmd(struct f2fs_sb_info *sbi,
1224 				   struct discard_cmd *dc, blk_opf_t flag,
1225 				   struct list_head *wait_list,
1226 				   unsigned int *issued)
1227 {
1228 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1229 	struct block_device *bdev = dc->bdev;
1230 	struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS);
1231 	unsigned long flags;
1232 
1233 	trace_f2fs_issue_reset_zone(bdev, dc->di.start);
1234 
1235 	spin_lock_irqsave(&dc->lock, flags);
1236 	dc->state = D_SUBMIT;
1237 	dc->bio_ref++;
1238 	spin_unlock_irqrestore(&dc->lock, flags);
1239 
1240 	if (issued)
1241 		(*issued)++;
1242 
1243 	atomic_inc(&dcc->queued_discard);
1244 	dc->queued++;
1245 	list_move_tail(&dc->list, wait_list);
1246 
1247 	/* sanity check on discard range */
1248 	__check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len);
1249 
1250 	bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start);
1251 	bio->bi_private = dc;
1252 	bio->bi_end_io = f2fs_submit_discard_endio;
1253 	submit_bio(bio);
1254 
1255 	atomic_inc(&dcc->issued_discard);
1256 	f2fs_update_iostat(sbi, NULL, FS_ZONE_RESET_IO, dc->di.len * F2FS_BLKSIZE);
1257 }
1258 #endif
1259 
1260 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1261 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1262 				struct discard_policy *dpolicy,
1263 				struct discard_cmd *dc, int *issued)
1264 {
1265 	struct block_device *bdev = dc->bdev;
1266 	unsigned int max_discard_blocks =
1267 			SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1268 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1269 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1270 					&(dcc->fstrim_list) : &(dcc->wait_list);
1271 	blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0;
1272 	block_t lstart, start, len, total_len;
1273 	int err = 0;
1274 
1275 	if (dc->state != D_PREP)
1276 		return 0;
1277 
1278 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1279 		return 0;
1280 
1281 #ifdef CONFIG_BLK_DEV_ZONED
1282 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) {
1283 		int devi = f2fs_bdev_index(sbi, bdev);
1284 
1285 		if (devi < 0)
1286 			return -EINVAL;
1287 
1288 		if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1289 			__submit_zone_reset_cmd(sbi, dc, flag,
1290 						wait_list, issued);
1291 			return 0;
1292 		}
1293 	}
1294 #endif
1295 
1296 	/*
1297 	 * stop issuing discard for any of below cases:
1298 	 * 1. device is conventional zone, but it doesn't support discard.
1299 	 * 2. device is regulare device, after snapshot it doesn't support
1300 	 * discard.
1301 	 */
1302 	if (!bdev_max_discard_sectors(bdev))
1303 		return -EOPNOTSUPP;
1304 
1305 	trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len);
1306 
1307 	lstart = dc->di.lstart;
1308 	start = dc->di.start;
1309 	len = dc->di.len;
1310 	total_len = len;
1311 
1312 	dc->di.len = 0;
1313 
1314 	while (total_len && *issued < dpolicy->max_requests && !err) {
1315 		struct bio *bio = NULL;
1316 		unsigned long flags;
1317 		bool last = true;
1318 
1319 		if (len > max_discard_blocks) {
1320 			len = max_discard_blocks;
1321 			last = false;
1322 		}
1323 
1324 		(*issued)++;
1325 		if (*issued == dpolicy->max_requests)
1326 			last = true;
1327 
1328 		dc->di.len += len;
1329 
1330 		if (time_to_inject(sbi, FAULT_DISCARD)) {
1331 			err = -EIO;
1332 		} else {
1333 			err = __blkdev_issue_discard(bdev,
1334 					SECTOR_FROM_BLOCK(start),
1335 					SECTOR_FROM_BLOCK(len),
1336 					GFP_NOFS, &bio);
1337 		}
1338 		if (err) {
1339 			spin_lock_irqsave(&dc->lock, flags);
1340 			if (dc->state == D_PARTIAL)
1341 				dc->state = D_SUBMIT;
1342 			spin_unlock_irqrestore(&dc->lock, flags);
1343 
1344 			break;
1345 		}
1346 
1347 		f2fs_bug_on(sbi, !bio);
1348 
1349 		/*
1350 		 * should keep before submission to avoid D_DONE
1351 		 * right away
1352 		 */
1353 		spin_lock_irqsave(&dc->lock, flags);
1354 		if (last)
1355 			dc->state = D_SUBMIT;
1356 		else
1357 			dc->state = D_PARTIAL;
1358 		dc->bio_ref++;
1359 		spin_unlock_irqrestore(&dc->lock, flags);
1360 
1361 		atomic_inc(&dcc->queued_discard);
1362 		dc->queued++;
1363 		list_move_tail(&dc->list, wait_list);
1364 
1365 		/* sanity check on discard range */
1366 		__check_sit_bitmap(sbi, lstart, lstart + len);
1367 
1368 		bio->bi_private = dc;
1369 		bio->bi_end_io = f2fs_submit_discard_endio;
1370 		bio->bi_opf |= flag;
1371 		submit_bio(bio);
1372 
1373 		atomic_inc(&dcc->issued_discard);
1374 
1375 		f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE);
1376 
1377 		lstart += len;
1378 		start += len;
1379 		total_len -= len;
1380 		len = total_len;
1381 	}
1382 
1383 	if (!err && len) {
1384 		dcc->undiscard_blks -= len;
1385 		__update_discard_tree_range(sbi, bdev, lstart, start, len);
1386 	}
1387 	return err;
1388 }
1389 
1390 static void __insert_discard_cmd(struct f2fs_sb_info *sbi,
1391 				struct block_device *bdev, block_t lstart,
1392 				block_t start, block_t len)
1393 {
1394 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1395 	struct rb_node **p = &dcc->root.rb_root.rb_node;
1396 	struct rb_node *parent = NULL;
1397 	struct discard_cmd *dc;
1398 	bool leftmost = true;
1399 
1400 	/* look up rb tree to find parent node */
1401 	while (*p) {
1402 		parent = *p;
1403 		dc = rb_entry(parent, struct discard_cmd, rb_node);
1404 
1405 		if (lstart < dc->di.lstart) {
1406 			p = &(*p)->rb_left;
1407 		} else if (lstart >= dc->di.lstart + dc->di.len) {
1408 			p = &(*p)->rb_right;
1409 			leftmost = false;
1410 		} else {
1411 			/* Let's skip to add, if exists */
1412 			return;
1413 		}
1414 	}
1415 
1416 	dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1417 
1418 	rb_link_node(&dc->rb_node, parent, p);
1419 	rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1420 }
1421 
1422 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1423 						struct discard_cmd *dc)
1424 {
1425 	list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]);
1426 }
1427 
1428 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1429 				struct discard_cmd *dc, block_t blkaddr)
1430 {
1431 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1432 	struct discard_info di = dc->di;
1433 	bool modified = false;
1434 
1435 	if (dc->state == D_DONE || dc->di.len == 1) {
1436 		__remove_discard_cmd(sbi, dc);
1437 		return;
1438 	}
1439 
1440 	dcc->undiscard_blks -= di.len;
1441 
1442 	if (blkaddr > di.lstart) {
1443 		dc->di.len = blkaddr - dc->di.lstart;
1444 		dcc->undiscard_blks += dc->di.len;
1445 		__relocate_discard_cmd(dcc, dc);
1446 		modified = true;
1447 	}
1448 
1449 	if (blkaddr < di.lstart + di.len - 1) {
1450 		if (modified) {
1451 			__insert_discard_cmd(sbi, dc->bdev, blkaddr + 1,
1452 					di.start + blkaddr + 1 - di.lstart,
1453 					di.lstart + di.len - 1 - blkaddr);
1454 		} else {
1455 			dc->di.lstart++;
1456 			dc->di.len--;
1457 			dc->di.start++;
1458 			dcc->undiscard_blks += dc->di.len;
1459 			__relocate_discard_cmd(dcc, dc);
1460 		}
1461 	}
1462 }
1463 
1464 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1465 				struct block_device *bdev, block_t lstart,
1466 				block_t start, block_t len)
1467 {
1468 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1469 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1470 	struct discard_cmd *dc;
1471 	struct discard_info di = {0};
1472 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1473 	unsigned int max_discard_blocks =
1474 			SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1475 	block_t end = lstart + len;
1476 
1477 	dc = __lookup_discard_cmd_ret(&dcc->root, lstart,
1478 				&prev_dc, &next_dc, &insert_p, &insert_parent);
1479 	if (dc)
1480 		prev_dc = dc;
1481 
1482 	if (!prev_dc) {
1483 		di.lstart = lstart;
1484 		di.len = next_dc ? next_dc->di.lstart - lstart : len;
1485 		di.len = min(di.len, len);
1486 		di.start = start;
1487 	}
1488 
1489 	while (1) {
1490 		struct rb_node *node;
1491 		bool merged = false;
1492 		struct discard_cmd *tdc = NULL;
1493 
1494 		if (prev_dc) {
1495 			di.lstart = prev_dc->di.lstart + prev_dc->di.len;
1496 			if (di.lstart < lstart)
1497 				di.lstart = lstart;
1498 			if (di.lstart >= end)
1499 				break;
1500 
1501 			if (!next_dc || next_dc->di.lstart > end)
1502 				di.len = end - di.lstart;
1503 			else
1504 				di.len = next_dc->di.lstart - di.lstart;
1505 			di.start = start + di.lstart - lstart;
1506 		}
1507 
1508 		if (!di.len)
1509 			goto next;
1510 
1511 		if (prev_dc && prev_dc->state == D_PREP &&
1512 			prev_dc->bdev == bdev &&
1513 			__is_discard_back_mergeable(&di, &prev_dc->di,
1514 							max_discard_blocks)) {
1515 			prev_dc->di.len += di.len;
1516 			dcc->undiscard_blks += di.len;
1517 			__relocate_discard_cmd(dcc, prev_dc);
1518 			di = prev_dc->di;
1519 			tdc = prev_dc;
1520 			merged = true;
1521 		}
1522 
1523 		if (next_dc && next_dc->state == D_PREP &&
1524 			next_dc->bdev == bdev &&
1525 			__is_discard_front_mergeable(&di, &next_dc->di,
1526 							max_discard_blocks)) {
1527 			next_dc->di.lstart = di.lstart;
1528 			next_dc->di.len += di.len;
1529 			next_dc->di.start = di.start;
1530 			dcc->undiscard_blks += di.len;
1531 			__relocate_discard_cmd(dcc, next_dc);
1532 			if (tdc)
1533 				__remove_discard_cmd(sbi, tdc);
1534 			merged = true;
1535 		}
1536 
1537 		if (!merged)
1538 			__insert_discard_cmd(sbi, bdev,
1539 						di.lstart, di.start, di.len);
1540  next:
1541 		prev_dc = next_dc;
1542 		if (!prev_dc)
1543 			break;
1544 
1545 		node = rb_next(&prev_dc->rb_node);
1546 		next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1547 	}
1548 }
1549 
1550 #ifdef CONFIG_BLK_DEV_ZONED
1551 static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi,
1552 		struct block_device *bdev, block_t blkstart, block_t lblkstart,
1553 		block_t blklen)
1554 {
1555 	trace_f2fs_queue_reset_zone(bdev, blkstart);
1556 
1557 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1558 	__insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen);
1559 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1560 }
1561 #endif
1562 
1563 static void __queue_discard_cmd(struct f2fs_sb_info *sbi,
1564 		struct block_device *bdev, block_t blkstart, block_t blklen)
1565 {
1566 	block_t lblkstart = blkstart;
1567 
1568 	if (!f2fs_bdev_support_discard(bdev))
1569 		return;
1570 
1571 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
1572 
1573 	if (f2fs_is_multi_device(sbi)) {
1574 		int devi = f2fs_target_device_index(sbi, blkstart);
1575 
1576 		blkstart -= FDEV(devi).start_blk;
1577 	}
1578 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1579 	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1580 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1581 }
1582 
1583 static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1584 		struct discard_policy *dpolicy, int *issued)
1585 {
1586 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1587 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1588 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1589 	struct discard_cmd *dc;
1590 	struct blk_plug plug;
1591 	bool io_interrupted = false;
1592 
1593 	mutex_lock(&dcc->cmd_lock);
1594 	dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos,
1595 				&prev_dc, &next_dc, &insert_p, &insert_parent);
1596 	if (!dc)
1597 		dc = next_dc;
1598 
1599 	blk_start_plug(&plug);
1600 
1601 	while (dc) {
1602 		struct rb_node *node;
1603 		int err = 0;
1604 
1605 		if (dc->state != D_PREP)
1606 			goto next;
1607 
1608 		if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1609 			io_interrupted = true;
1610 			break;
1611 		}
1612 
1613 		dcc->next_pos = dc->di.lstart + dc->di.len;
1614 		err = __submit_discard_cmd(sbi, dpolicy, dc, issued);
1615 
1616 		if (*issued >= dpolicy->max_requests)
1617 			break;
1618 next:
1619 		node = rb_next(&dc->rb_node);
1620 		if (err)
1621 			__remove_discard_cmd(sbi, dc);
1622 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1623 	}
1624 
1625 	blk_finish_plug(&plug);
1626 
1627 	if (!dc)
1628 		dcc->next_pos = 0;
1629 
1630 	mutex_unlock(&dcc->cmd_lock);
1631 
1632 	if (!(*issued) && io_interrupted)
1633 		*issued = -1;
1634 }
1635 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1636 					struct discard_policy *dpolicy);
1637 
1638 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1639 					struct discard_policy *dpolicy)
1640 {
1641 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1642 	struct list_head *pend_list;
1643 	struct discard_cmd *dc, *tmp;
1644 	struct blk_plug plug;
1645 	int i, issued;
1646 	bool io_interrupted = false;
1647 
1648 	if (dpolicy->timeout)
1649 		f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1650 
1651 retry:
1652 	issued = 0;
1653 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1654 		if (dpolicy->timeout &&
1655 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1656 			break;
1657 
1658 		if (i + 1 < dpolicy->granularity)
1659 			break;
1660 
1661 		if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) {
1662 			__issue_discard_cmd_orderly(sbi, dpolicy, &issued);
1663 			return issued;
1664 		}
1665 
1666 		pend_list = &dcc->pend_list[i];
1667 
1668 		mutex_lock(&dcc->cmd_lock);
1669 		if (list_empty(pend_list))
1670 			goto next;
1671 		if (unlikely(dcc->rbtree_check))
1672 			f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
1673 		blk_start_plug(&plug);
1674 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1675 			f2fs_bug_on(sbi, dc->state != D_PREP);
1676 
1677 			if (dpolicy->timeout &&
1678 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1679 				break;
1680 
1681 			if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1682 						!is_idle(sbi, DISCARD_TIME)) {
1683 				io_interrupted = true;
1684 				break;
1685 			}
1686 
1687 			__submit_discard_cmd(sbi, dpolicy, dc, &issued);
1688 
1689 			if (issued >= dpolicy->max_requests)
1690 				break;
1691 		}
1692 		blk_finish_plug(&plug);
1693 next:
1694 		mutex_unlock(&dcc->cmd_lock);
1695 
1696 		if (issued >= dpolicy->max_requests || io_interrupted)
1697 			break;
1698 	}
1699 
1700 	if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1701 		__wait_all_discard_cmd(sbi, dpolicy);
1702 		goto retry;
1703 	}
1704 
1705 	if (!issued && io_interrupted)
1706 		issued = -1;
1707 
1708 	return issued;
1709 }
1710 
1711 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1712 {
1713 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1714 	struct list_head *pend_list;
1715 	struct discard_cmd *dc, *tmp;
1716 	int i;
1717 	bool dropped = false;
1718 
1719 	mutex_lock(&dcc->cmd_lock);
1720 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1721 		pend_list = &dcc->pend_list[i];
1722 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1723 			f2fs_bug_on(sbi, dc->state != D_PREP);
1724 			__remove_discard_cmd(sbi, dc);
1725 			dropped = true;
1726 		}
1727 	}
1728 	mutex_unlock(&dcc->cmd_lock);
1729 
1730 	return dropped;
1731 }
1732 
1733 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1734 {
1735 	__drop_discard_cmd(sbi);
1736 }
1737 
1738 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1739 							struct discard_cmd *dc)
1740 {
1741 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1742 	unsigned int len = 0;
1743 
1744 	wait_for_completion_io(&dc->wait);
1745 	mutex_lock(&dcc->cmd_lock);
1746 	f2fs_bug_on(sbi, dc->state != D_DONE);
1747 	dc->ref--;
1748 	if (!dc->ref) {
1749 		if (!dc->error)
1750 			len = dc->di.len;
1751 		__remove_discard_cmd(sbi, dc);
1752 	}
1753 	mutex_unlock(&dcc->cmd_lock);
1754 
1755 	return len;
1756 }
1757 
1758 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1759 						struct discard_policy *dpolicy,
1760 						block_t start, block_t end)
1761 {
1762 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1763 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1764 					&(dcc->fstrim_list) : &(dcc->wait_list);
1765 	struct discard_cmd *dc = NULL, *iter, *tmp;
1766 	unsigned int trimmed = 0;
1767 
1768 next:
1769 	dc = NULL;
1770 
1771 	mutex_lock(&dcc->cmd_lock);
1772 	list_for_each_entry_safe(iter, tmp, wait_list, list) {
1773 		if (iter->di.lstart + iter->di.len <= start ||
1774 					end <= iter->di.lstart)
1775 			continue;
1776 		if (iter->di.len < dpolicy->granularity)
1777 			continue;
1778 		if (iter->state == D_DONE && !iter->ref) {
1779 			wait_for_completion_io(&iter->wait);
1780 			if (!iter->error)
1781 				trimmed += iter->di.len;
1782 			__remove_discard_cmd(sbi, iter);
1783 		} else {
1784 			iter->ref++;
1785 			dc = iter;
1786 			break;
1787 		}
1788 	}
1789 	mutex_unlock(&dcc->cmd_lock);
1790 
1791 	if (dc) {
1792 		trimmed += __wait_one_discard_bio(sbi, dc);
1793 		goto next;
1794 	}
1795 
1796 	return trimmed;
1797 }
1798 
1799 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1800 						struct discard_policy *dpolicy)
1801 {
1802 	struct discard_policy dp;
1803 	unsigned int discard_blks;
1804 
1805 	if (dpolicy)
1806 		return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1807 
1808 	/* wait all */
1809 	__init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY);
1810 	discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1811 	__init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY);
1812 	discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1813 
1814 	return discard_blks;
1815 }
1816 
1817 /* This should be covered by global mutex, &sit_i->sentry_lock */
1818 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1819 {
1820 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1821 	struct discard_cmd *dc;
1822 	bool need_wait = false;
1823 
1824 	mutex_lock(&dcc->cmd_lock);
1825 	dc = __lookup_discard_cmd(sbi, blkaddr);
1826 #ifdef CONFIG_BLK_DEV_ZONED
1827 	if (dc && f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) {
1828 		int devi = f2fs_bdev_index(sbi, dc->bdev);
1829 
1830 		if (devi < 0) {
1831 			mutex_unlock(&dcc->cmd_lock);
1832 			return;
1833 		}
1834 
1835 		if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1836 			/* force submit zone reset */
1837 			if (dc->state == D_PREP)
1838 				__submit_zone_reset_cmd(sbi, dc, REQ_SYNC,
1839 							&dcc->wait_list, NULL);
1840 			dc->ref++;
1841 			mutex_unlock(&dcc->cmd_lock);
1842 			/* wait zone reset */
1843 			__wait_one_discard_bio(sbi, dc);
1844 			return;
1845 		}
1846 	}
1847 #endif
1848 	if (dc) {
1849 		if (dc->state == D_PREP) {
1850 			__punch_discard_cmd(sbi, dc, blkaddr);
1851 		} else {
1852 			dc->ref++;
1853 			need_wait = true;
1854 		}
1855 	}
1856 	mutex_unlock(&dcc->cmd_lock);
1857 
1858 	if (need_wait)
1859 		__wait_one_discard_bio(sbi, dc);
1860 }
1861 
1862 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1863 {
1864 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1865 
1866 	if (dcc && dcc->f2fs_issue_discard) {
1867 		struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1868 
1869 		dcc->f2fs_issue_discard = NULL;
1870 		kthread_stop(discard_thread);
1871 	}
1872 }
1873 
1874 /**
1875  * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT
1876  * @sbi: the f2fs_sb_info data for discard cmd to issue
1877  *
1878  * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped
1879  *
1880  * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false.
1881  */
1882 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1883 {
1884 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1885 	struct discard_policy dpolicy;
1886 	bool dropped;
1887 
1888 	if (!atomic_read(&dcc->discard_cmd_cnt))
1889 		return true;
1890 
1891 	__init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1892 					dcc->discard_granularity);
1893 	__issue_discard_cmd(sbi, &dpolicy);
1894 	dropped = __drop_discard_cmd(sbi);
1895 
1896 	/* just to make sure there is no pending discard commands */
1897 	__wait_all_discard_cmd(sbi, NULL);
1898 
1899 	f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1900 	return !dropped;
1901 }
1902 
1903 static int issue_discard_thread(void *data)
1904 {
1905 	struct f2fs_sb_info *sbi = data;
1906 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1907 	wait_queue_head_t *q = &dcc->discard_wait_queue;
1908 	struct discard_policy dpolicy;
1909 	unsigned int wait_ms = dcc->min_discard_issue_time;
1910 	int issued;
1911 
1912 	set_freezable();
1913 
1914 	do {
1915 		wait_event_freezable_timeout(*q,
1916 				kthread_should_stop() || dcc->discard_wake,
1917 				msecs_to_jiffies(wait_ms));
1918 
1919 		if (sbi->gc_mode == GC_URGENT_HIGH ||
1920 			!f2fs_available_free_memory(sbi, DISCARD_CACHE))
1921 			__init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE,
1922 						MIN_DISCARD_GRANULARITY);
1923 		else
1924 			__init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1925 						dcc->discard_granularity);
1926 
1927 		if (dcc->discard_wake)
1928 			dcc->discard_wake = false;
1929 
1930 		/* clean up pending candidates before going to sleep */
1931 		if (atomic_read(&dcc->queued_discard))
1932 			__wait_all_discard_cmd(sbi, NULL);
1933 
1934 		if (f2fs_readonly(sbi->sb))
1935 			continue;
1936 		if (kthread_should_stop())
1937 			return 0;
1938 		if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) ||
1939 			!atomic_read(&dcc->discard_cmd_cnt)) {
1940 			wait_ms = dpolicy.max_interval;
1941 			continue;
1942 		}
1943 
1944 		sb_start_intwrite(sbi->sb);
1945 
1946 		issued = __issue_discard_cmd(sbi, &dpolicy);
1947 		if (issued > 0) {
1948 			__wait_all_discard_cmd(sbi, &dpolicy);
1949 			wait_ms = dpolicy.min_interval;
1950 		} else if (issued == -1) {
1951 			wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1952 			if (!wait_ms)
1953 				wait_ms = dpolicy.mid_interval;
1954 		} else {
1955 			wait_ms = dpolicy.max_interval;
1956 		}
1957 		if (!atomic_read(&dcc->discard_cmd_cnt))
1958 			wait_ms = dpolicy.max_interval;
1959 
1960 		sb_end_intwrite(sbi->sb);
1961 
1962 	} while (!kthread_should_stop());
1963 	return 0;
1964 }
1965 
1966 #ifdef CONFIG_BLK_DEV_ZONED
1967 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1968 		struct block_device *bdev, block_t blkstart, block_t blklen)
1969 {
1970 	sector_t sector, nr_sects;
1971 	block_t lblkstart = blkstart;
1972 	int devi = 0;
1973 	u64 remainder = 0;
1974 
1975 	if (f2fs_is_multi_device(sbi)) {
1976 		devi = f2fs_target_device_index(sbi, blkstart);
1977 		if (blkstart < FDEV(devi).start_blk ||
1978 		    blkstart > FDEV(devi).end_blk) {
1979 			f2fs_err(sbi, "Invalid block %x", blkstart);
1980 			return -EIO;
1981 		}
1982 		blkstart -= FDEV(devi).start_blk;
1983 	}
1984 
1985 	/* For sequential zones, reset the zone write pointer */
1986 	if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1987 		sector = SECTOR_FROM_BLOCK(blkstart);
1988 		nr_sects = SECTOR_FROM_BLOCK(blklen);
1989 		div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder);
1990 
1991 		if (remainder || nr_sects != bdev_zone_sectors(bdev)) {
1992 			f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1993 				 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1994 				 blkstart, blklen);
1995 			return -EIO;
1996 		}
1997 
1998 		if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) {
1999 			unsigned int nofs_flags;
2000 			int ret;
2001 
2002 			trace_f2fs_issue_reset_zone(bdev, blkstart);
2003 			nofs_flags = memalloc_nofs_save();
2004 			ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
2005 						sector, nr_sects);
2006 			memalloc_nofs_restore(nofs_flags);
2007 			return ret;
2008 		}
2009 
2010 		__queue_zone_reset_cmd(sbi, bdev, blkstart, lblkstart, blklen);
2011 		return 0;
2012 	}
2013 
2014 	/* For conventional zones, use regular discard if supported */
2015 	__queue_discard_cmd(sbi, bdev, lblkstart, blklen);
2016 	return 0;
2017 }
2018 #endif
2019 
2020 static int __issue_discard_async(struct f2fs_sb_info *sbi,
2021 		struct block_device *bdev, block_t blkstart, block_t blklen)
2022 {
2023 #ifdef CONFIG_BLK_DEV_ZONED
2024 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
2025 		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
2026 #endif
2027 	__queue_discard_cmd(sbi, bdev, blkstart, blklen);
2028 	return 0;
2029 }
2030 
2031 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
2032 				block_t blkstart, block_t blklen)
2033 {
2034 	sector_t start = blkstart, len = 0;
2035 	struct block_device *bdev;
2036 	struct seg_entry *se;
2037 	unsigned int offset;
2038 	block_t i;
2039 	int err = 0;
2040 
2041 	bdev = f2fs_target_device(sbi, blkstart, NULL);
2042 
2043 	for (i = blkstart; i < blkstart + blklen; i++, len++) {
2044 		if (i != start) {
2045 			struct block_device *bdev2 =
2046 				f2fs_target_device(sbi, i, NULL);
2047 
2048 			if (bdev2 != bdev) {
2049 				err = __issue_discard_async(sbi, bdev,
2050 						start, len);
2051 				if (err)
2052 					return err;
2053 				bdev = bdev2;
2054 				start = i;
2055 				len = 0;
2056 			}
2057 		}
2058 
2059 		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
2060 		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
2061 
2062 		if (f2fs_block_unit_discard(sbi) &&
2063 				!f2fs_test_and_set_bit(offset, se->discard_map))
2064 			sbi->discard_blks--;
2065 	}
2066 
2067 	if (len)
2068 		err = __issue_discard_async(sbi, bdev, start, len);
2069 	return err;
2070 }
2071 
2072 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
2073 							bool check_only)
2074 {
2075 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2076 	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
2077 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2078 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2079 	unsigned long *discard_map = (unsigned long *)se->discard_map;
2080 	unsigned long *dmap = SIT_I(sbi)->tmp_map;
2081 	unsigned int start = 0, end = -1;
2082 	bool force = (cpc->reason & CP_DISCARD);
2083 	struct discard_entry *de = NULL;
2084 	struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
2085 	int i;
2086 
2087 	if (se->valid_blocks == BLKS_PER_SEG(sbi) ||
2088 	    !f2fs_hw_support_discard(sbi) ||
2089 	    !f2fs_block_unit_discard(sbi))
2090 		return false;
2091 
2092 	if (!force) {
2093 		if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
2094 			SM_I(sbi)->dcc_info->nr_discards >=
2095 				SM_I(sbi)->dcc_info->max_discards)
2096 			return false;
2097 	}
2098 
2099 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
2100 	for (i = 0; i < entries; i++)
2101 		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
2102 				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
2103 
2104 	while (force || SM_I(sbi)->dcc_info->nr_discards <=
2105 				SM_I(sbi)->dcc_info->max_discards) {
2106 		start = __find_rev_next_bit(dmap, BLKS_PER_SEG(sbi), end + 1);
2107 		if (start >= BLKS_PER_SEG(sbi))
2108 			break;
2109 
2110 		end = __find_rev_next_zero_bit(dmap,
2111 						BLKS_PER_SEG(sbi), start + 1);
2112 		if (force && start && end != BLKS_PER_SEG(sbi) &&
2113 		    (end - start) < cpc->trim_minlen)
2114 			continue;
2115 
2116 		if (check_only)
2117 			return true;
2118 
2119 		if (!de) {
2120 			de = f2fs_kmem_cache_alloc(discard_entry_slab,
2121 						GFP_F2FS_ZERO, true, NULL);
2122 			de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
2123 			list_add_tail(&de->list, head);
2124 		}
2125 
2126 		for (i = start; i < end; i++)
2127 			__set_bit_le(i, (void *)de->discard_map);
2128 
2129 		SM_I(sbi)->dcc_info->nr_discards += end - start;
2130 	}
2131 	return false;
2132 }
2133 
2134 static void release_discard_addr(struct discard_entry *entry)
2135 {
2136 	list_del(&entry->list);
2137 	kmem_cache_free(discard_entry_slab, entry);
2138 }
2139 
2140 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
2141 {
2142 	struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
2143 	struct discard_entry *entry, *this;
2144 
2145 	/* drop caches */
2146 	list_for_each_entry_safe(entry, this, head, list)
2147 		release_discard_addr(entry);
2148 }
2149 
2150 /*
2151  * Should call f2fs_clear_prefree_segments after checkpoint is done.
2152  */
2153 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
2154 {
2155 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2156 	unsigned int segno;
2157 
2158 	mutex_lock(&dirty_i->seglist_lock);
2159 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
2160 		__set_test_and_free(sbi, segno, false);
2161 	mutex_unlock(&dirty_i->seglist_lock);
2162 }
2163 
2164 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
2165 						struct cp_control *cpc)
2166 {
2167 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2168 	struct list_head *head = &dcc->entry_list;
2169 	struct discard_entry *entry, *this;
2170 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2171 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
2172 	unsigned int start = 0, end = -1;
2173 	unsigned int secno, start_segno;
2174 	bool force = (cpc->reason & CP_DISCARD);
2175 	bool section_alignment = F2FS_OPTION(sbi).discard_unit ==
2176 						DISCARD_UNIT_SECTION;
2177 
2178 	if (f2fs_lfs_mode(sbi) && __is_large_section(sbi))
2179 		section_alignment = true;
2180 
2181 	mutex_lock(&dirty_i->seglist_lock);
2182 
2183 	while (1) {
2184 		int i;
2185 
2186 		if (section_alignment && end != -1)
2187 			end--;
2188 		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
2189 		if (start >= MAIN_SEGS(sbi))
2190 			break;
2191 		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
2192 								start + 1);
2193 
2194 		if (section_alignment) {
2195 			start = rounddown(start, SEGS_PER_SEC(sbi));
2196 			end = roundup(end, SEGS_PER_SEC(sbi));
2197 		}
2198 
2199 		for (i = start; i < end; i++) {
2200 			if (test_and_clear_bit(i, prefree_map))
2201 				dirty_i->nr_dirty[PRE]--;
2202 		}
2203 
2204 		if (!f2fs_realtime_discard_enable(sbi))
2205 			continue;
2206 
2207 		if (force && start >= cpc->trim_start &&
2208 					(end - 1) <= cpc->trim_end)
2209 			continue;
2210 
2211 		/* Should cover 2MB zoned device for zone-based reset */
2212 		if (!f2fs_sb_has_blkzoned(sbi) &&
2213 		    (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) {
2214 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2215 				SEGS_TO_BLKS(sbi, end - start));
2216 			continue;
2217 		}
2218 next:
2219 		secno = GET_SEC_FROM_SEG(sbi, start);
2220 		start_segno = GET_SEG_FROM_SEC(sbi, secno);
2221 		if (!IS_CURSEC(sbi, secno) &&
2222 			!get_valid_blocks(sbi, start, true))
2223 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2224 						BLKS_PER_SEC(sbi));
2225 
2226 		start = start_segno + SEGS_PER_SEC(sbi);
2227 		if (start < end)
2228 			goto next;
2229 		else
2230 			end = start - 1;
2231 	}
2232 	mutex_unlock(&dirty_i->seglist_lock);
2233 
2234 	if (!f2fs_block_unit_discard(sbi))
2235 		goto wakeup;
2236 
2237 	/* send small discards */
2238 	list_for_each_entry_safe(entry, this, head, list) {
2239 		unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2240 		bool is_valid = test_bit_le(0, entry->discard_map);
2241 
2242 find_next:
2243 		if (is_valid) {
2244 			next_pos = find_next_zero_bit_le(entry->discard_map,
2245 						BLKS_PER_SEG(sbi), cur_pos);
2246 			len = next_pos - cur_pos;
2247 
2248 			if (f2fs_sb_has_blkzoned(sbi) ||
2249 			    (force && len < cpc->trim_minlen))
2250 				goto skip;
2251 
2252 			f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2253 									len);
2254 			total_len += len;
2255 		} else {
2256 			next_pos = find_next_bit_le(entry->discard_map,
2257 						BLKS_PER_SEG(sbi), cur_pos);
2258 		}
2259 skip:
2260 		cur_pos = next_pos;
2261 		is_valid = !is_valid;
2262 
2263 		if (cur_pos < BLKS_PER_SEG(sbi))
2264 			goto find_next;
2265 
2266 		release_discard_addr(entry);
2267 		dcc->nr_discards -= total_len;
2268 	}
2269 
2270 wakeup:
2271 	wake_up_discard_thread(sbi, false);
2272 }
2273 
2274 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi)
2275 {
2276 	dev_t dev = sbi->sb->s_bdev->bd_dev;
2277 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2278 	int err = 0;
2279 
2280 	if (f2fs_sb_has_readonly(sbi)) {
2281 		f2fs_info(sbi,
2282 			"Skip to start discard thread for readonly image");
2283 		return 0;
2284 	}
2285 
2286 	if (!f2fs_realtime_discard_enable(sbi))
2287 		return 0;
2288 
2289 	dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2290 				"f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2291 	if (IS_ERR(dcc->f2fs_issue_discard)) {
2292 		err = PTR_ERR(dcc->f2fs_issue_discard);
2293 		dcc->f2fs_issue_discard = NULL;
2294 	}
2295 
2296 	return err;
2297 }
2298 
2299 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2300 {
2301 	struct discard_cmd_control *dcc;
2302 	int err = 0, i;
2303 
2304 	if (SM_I(sbi)->dcc_info) {
2305 		dcc = SM_I(sbi)->dcc_info;
2306 		goto init_thread;
2307 	}
2308 
2309 	dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2310 	if (!dcc)
2311 		return -ENOMEM;
2312 
2313 	dcc->discard_io_aware_gran = MAX_PLIST_NUM;
2314 	dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2315 	dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY;
2316 	dcc->discard_io_aware = DPOLICY_IO_AWARE_ENABLE;
2317 	if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT)
2318 		dcc->discard_granularity = BLKS_PER_SEG(sbi);
2319 	else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION)
2320 		dcc->discard_granularity = BLKS_PER_SEC(sbi);
2321 
2322 	INIT_LIST_HEAD(&dcc->entry_list);
2323 	for (i = 0; i < MAX_PLIST_NUM; i++)
2324 		INIT_LIST_HEAD(&dcc->pend_list[i]);
2325 	INIT_LIST_HEAD(&dcc->wait_list);
2326 	INIT_LIST_HEAD(&dcc->fstrim_list);
2327 	mutex_init(&dcc->cmd_lock);
2328 	atomic_set(&dcc->issued_discard, 0);
2329 	atomic_set(&dcc->queued_discard, 0);
2330 	atomic_set(&dcc->discard_cmd_cnt, 0);
2331 	dcc->nr_discards = 0;
2332 	dcc->max_discards = SEGS_TO_BLKS(sbi, MAIN_SEGS(sbi));
2333 	dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST;
2334 	dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME;
2335 	dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME;
2336 	dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME;
2337 	dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL;
2338 	dcc->undiscard_blks = 0;
2339 	dcc->next_pos = 0;
2340 	dcc->root = RB_ROOT_CACHED;
2341 	dcc->rbtree_check = false;
2342 
2343 	init_waitqueue_head(&dcc->discard_wait_queue);
2344 	SM_I(sbi)->dcc_info = dcc;
2345 init_thread:
2346 	err = f2fs_start_discard_thread(sbi);
2347 	if (err) {
2348 		kfree(dcc);
2349 		SM_I(sbi)->dcc_info = NULL;
2350 	}
2351 
2352 	return err;
2353 }
2354 
2355 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2356 {
2357 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2358 
2359 	if (!dcc)
2360 		return;
2361 
2362 	f2fs_stop_discard_thread(sbi);
2363 
2364 	/*
2365 	 * Recovery can cache discard commands, so in error path of
2366 	 * fill_super(), it needs to give a chance to handle them.
2367 	 */
2368 	f2fs_issue_discard_timeout(sbi);
2369 
2370 	kfree(dcc);
2371 	SM_I(sbi)->dcc_info = NULL;
2372 }
2373 
2374 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2375 {
2376 	struct sit_info *sit_i = SIT_I(sbi);
2377 
2378 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2379 		sit_i->dirty_sentries++;
2380 		return false;
2381 	}
2382 
2383 	return true;
2384 }
2385 
2386 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2387 					unsigned int segno, int modified)
2388 {
2389 	struct seg_entry *se = get_seg_entry(sbi, segno);
2390 
2391 	se->type = type;
2392 	if (modified)
2393 		__mark_sit_entry_dirty(sbi, segno);
2394 }
2395 
2396 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2397 								block_t blkaddr)
2398 {
2399 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
2400 
2401 	if (segno == NULL_SEGNO)
2402 		return 0;
2403 	return get_seg_entry(sbi, segno)->mtime;
2404 }
2405 
2406 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2407 						unsigned long long old_mtime)
2408 {
2409 	struct seg_entry *se;
2410 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
2411 	unsigned long long ctime = get_mtime(sbi, false);
2412 	unsigned long long mtime = old_mtime ? old_mtime : ctime;
2413 
2414 	if (segno == NULL_SEGNO)
2415 		return;
2416 
2417 	se = get_seg_entry(sbi, segno);
2418 
2419 	if (!se->mtime)
2420 		se->mtime = mtime;
2421 	else
2422 		se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2423 						se->valid_blocks + 1);
2424 
2425 	if (ctime > SIT_I(sbi)->max_mtime)
2426 		SIT_I(sbi)->max_mtime = ctime;
2427 }
2428 
2429 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2430 {
2431 	struct seg_entry *se;
2432 	unsigned int segno, offset;
2433 	long int new_vblocks;
2434 	bool exist;
2435 #ifdef CONFIG_F2FS_CHECK_FS
2436 	bool mir_exist;
2437 #endif
2438 
2439 	segno = GET_SEGNO(sbi, blkaddr);
2440 	if (segno == NULL_SEGNO)
2441 		return;
2442 
2443 	se = get_seg_entry(sbi, segno);
2444 	new_vblocks = se->valid_blocks + del;
2445 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2446 
2447 	f2fs_bug_on(sbi, (new_vblocks < 0 ||
2448 			(new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2449 
2450 	se->valid_blocks = new_vblocks;
2451 
2452 	/* Update valid block bitmap */
2453 	if (del > 0) {
2454 		exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2455 #ifdef CONFIG_F2FS_CHECK_FS
2456 		mir_exist = f2fs_test_and_set_bit(offset,
2457 						se->cur_valid_map_mir);
2458 		if (unlikely(exist != mir_exist)) {
2459 			f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2460 				 blkaddr, exist);
2461 			f2fs_bug_on(sbi, 1);
2462 		}
2463 #endif
2464 		if (unlikely(exist)) {
2465 			f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2466 				 blkaddr);
2467 			f2fs_bug_on(sbi, 1);
2468 			se->valid_blocks--;
2469 			del = 0;
2470 		}
2471 
2472 		if (f2fs_block_unit_discard(sbi) &&
2473 				!f2fs_test_and_set_bit(offset, se->discard_map))
2474 			sbi->discard_blks--;
2475 
2476 		/*
2477 		 * SSR should never reuse block which is checkpointed
2478 		 * or newly invalidated.
2479 		 */
2480 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2481 			if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2482 				se->ckpt_valid_blocks++;
2483 		}
2484 	} else {
2485 		exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2486 #ifdef CONFIG_F2FS_CHECK_FS
2487 		mir_exist = f2fs_test_and_clear_bit(offset,
2488 						se->cur_valid_map_mir);
2489 		if (unlikely(exist != mir_exist)) {
2490 			f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2491 				 blkaddr, exist);
2492 			f2fs_bug_on(sbi, 1);
2493 		}
2494 #endif
2495 		if (unlikely(!exist)) {
2496 			f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2497 				 blkaddr);
2498 			f2fs_bug_on(sbi, 1);
2499 			se->valid_blocks++;
2500 			del = 0;
2501 		} else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2502 			/*
2503 			 * If checkpoints are off, we must not reuse data that
2504 			 * was used in the previous checkpoint. If it was used
2505 			 * before, we must track that to know how much space we
2506 			 * really have.
2507 			 */
2508 			if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2509 				spin_lock(&sbi->stat_lock);
2510 				sbi->unusable_block_count++;
2511 				spin_unlock(&sbi->stat_lock);
2512 			}
2513 		}
2514 
2515 		if (f2fs_block_unit_discard(sbi) &&
2516 			f2fs_test_and_clear_bit(offset, se->discard_map))
2517 			sbi->discard_blks++;
2518 	}
2519 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2520 		se->ckpt_valid_blocks += del;
2521 
2522 	__mark_sit_entry_dirty(sbi, segno);
2523 
2524 	/* update total number of valid blocks to be written in ckpt area */
2525 	SIT_I(sbi)->written_valid_blocks += del;
2526 
2527 	if (__is_large_section(sbi))
2528 		get_sec_entry(sbi, segno)->valid_blocks += del;
2529 }
2530 
2531 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2532 {
2533 	unsigned int segno = GET_SEGNO(sbi, addr);
2534 	struct sit_info *sit_i = SIT_I(sbi);
2535 
2536 	f2fs_bug_on(sbi, addr == NULL_ADDR);
2537 	if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2538 		return;
2539 
2540 	f2fs_invalidate_internal_cache(sbi, addr);
2541 
2542 	/* add it into sit main buffer */
2543 	down_write(&sit_i->sentry_lock);
2544 
2545 	update_segment_mtime(sbi, addr, 0);
2546 	update_sit_entry(sbi, addr, -1);
2547 
2548 	/* add it into dirty seglist */
2549 	locate_dirty_segment(sbi, segno);
2550 
2551 	up_write(&sit_i->sentry_lock);
2552 }
2553 
2554 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2555 {
2556 	struct sit_info *sit_i = SIT_I(sbi);
2557 	unsigned int segno, offset;
2558 	struct seg_entry *se;
2559 	bool is_cp = false;
2560 
2561 	if (!__is_valid_data_blkaddr(blkaddr))
2562 		return true;
2563 
2564 	down_read(&sit_i->sentry_lock);
2565 
2566 	segno = GET_SEGNO(sbi, blkaddr);
2567 	se = get_seg_entry(sbi, segno);
2568 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2569 
2570 	if (f2fs_test_bit(offset, se->ckpt_valid_map))
2571 		is_cp = true;
2572 
2573 	up_read(&sit_i->sentry_lock);
2574 
2575 	return is_cp;
2576 }
2577 
2578 static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type)
2579 {
2580 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2581 
2582 	if (sbi->ckpt->alloc_type[type] == SSR)
2583 		return BLKS_PER_SEG(sbi);
2584 	return curseg->next_blkoff;
2585 }
2586 
2587 /*
2588  * Calculate the number of current summary pages for writing
2589  */
2590 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2591 {
2592 	int valid_sum_count = 0;
2593 	int i, sum_in_page;
2594 
2595 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2596 		if (sbi->ckpt->alloc_type[i] != SSR && for_ra)
2597 			valid_sum_count +=
2598 				le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2599 		else
2600 			valid_sum_count += f2fs_curseg_valid_blocks(sbi, i);
2601 	}
2602 
2603 	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2604 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2605 	if (valid_sum_count <= sum_in_page)
2606 		return 1;
2607 	else if ((valid_sum_count - sum_in_page) <=
2608 		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2609 		return 2;
2610 	return 3;
2611 }
2612 
2613 /*
2614  * Caller should put this summary page
2615  */
2616 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2617 {
2618 	if (unlikely(f2fs_cp_error(sbi)))
2619 		return ERR_PTR(-EIO);
2620 	return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2621 }
2622 
2623 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2624 					void *src, block_t blk_addr)
2625 {
2626 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2627 
2628 	memcpy(page_address(page), src, PAGE_SIZE);
2629 	set_page_dirty(page);
2630 	f2fs_put_page(page, 1);
2631 }
2632 
2633 static void write_sum_page(struct f2fs_sb_info *sbi,
2634 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
2635 {
2636 	f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2637 }
2638 
2639 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2640 						int type, block_t blk_addr)
2641 {
2642 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2643 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2644 	struct f2fs_summary_block *src = curseg->sum_blk;
2645 	struct f2fs_summary_block *dst;
2646 
2647 	dst = (struct f2fs_summary_block *)page_address(page);
2648 	memset(dst, 0, PAGE_SIZE);
2649 
2650 	mutex_lock(&curseg->curseg_mutex);
2651 
2652 	down_read(&curseg->journal_rwsem);
2653 	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2654 	up_read(&curseg->journal_rwsem);
2655 
2656 	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2657 	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2658 
2659 	mutex_unlock(&curseg->curseg_mutex);
2660 
2661 	set_page_dirty(page);
2662 	f2fs_put_page(page, 1);
2663 }
2664 
2665 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2666 				struct curseg_info *curseg)
2667 {
2668 	unsigned int segno = curseg->segno + 1;
2669 	struct free_segmap_info *free_i = FREE_I(sbi);
2670 
2671 	if (segno < MAIN_SEGS(sbi) && segno % SEGS_PER_SEC(sbi))
2672 		return !test_bit(segno, free_i->free_segmap);
2673 	return 0;
2674 }
2675 
2676 /*
2677  * Find a new segment from the free segments bitmap to right order
2678  * This function should be returned with success, otherwise BUG
2679  */
2680 static int get_new_segment(struct f2fs_sb_info *sbi,
2681 			unsigned int *newseg, bool new_sec, bool pinning)
2682 {
2683 	struct free_segmap_info *free_i = FREE_I(sbi);
2684 	unsigned int segno, secno, zoneno;
2685 	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2686 	unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2687 	unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2688 	bool init = true;
2689 	int i;
2690 	int ret = 0;
2691 
2692 	spin_lock(&free_i->segmap_lock);
2693 
2694 	if (time_to_inject(sbi, FAULT_NO_SEGMENT)) {
2695 		ret = -ENOSPC;
2696 		goto out_unlock;
2697 	}
2698 
2699 	if (!new_sec && ((*newseg + 1) % SEGS_PER_SEC(sbi))) {
2700 		segno = find_next_zero_bit(free_i->free_segmap,
2701 			GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2702 		if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2703 			goto got_it;
2704 	}
2705 
2706 #ifdef CONFIG_BLK_DEV_ZONED
2707 	/*
2708 	 * If we format f2fs on zoned storage, let's try to get pinned sections
2709 	 * from beginning of the storage, which should be a conventional one.
2710 	 */
2711 	if (f2fs_sb_has_blkzoned(sbi)) {
2712 		/* Prioritize writing to conventional zones */
2713 		if (sbi->blkzone_alloc_policy == BLKZONE_ALLOC_PRIOR_CONV || pinning)
2714 			segno = 0;
2715 		else
2716 			segno = max(sbi->first_zoned_segno, *newseg);
2717 		hint = GET_SEC_FROM_SEG(sbi, segno);
2718 	}
2719 #endif
2720 
2721 find_other_zone:
2722 	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2723 
2724 #ifdef CONFIG_BLK_DEV_ZONED
2725 	if (secno >= MAIN_SECS(sbi) && f2fs_sb_has_blkzoned(sbi)) {
2726 		/* Write only to sequential zones */
2727 		if (sbi->blkzone_alloc_policy == BLKZONE_ALLOC_ONLY_SEQ) {
2728 			hint = GET_SEC_FROM_SEG(sbi, sbi->first_zoned_segno);
2729 			secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2730 		} else
2731 			secno = find_first_zero_bit(free_i->free_secmap,
2732 								MAIN_SECS(sbi));
2733 		if (secno >= MAIN_SECS(sbi)) {
2734 			ret = -ENOSPC;
2735 			f2fs_bug_on(sbi, 1);
2736 			goto out_unlock;
2737 		}
2738 	}
2739 #endif
2740 
2741 	if (secno >= MAIN_SECS(sbi)) {
2742 		secno = find_first_zero_bit(free_i->free_secmap,
2743 							MAIN_SECS(sbi));
2744 		if (secno >= MAIN_SECS(sbi)) {
2745 			ret = -ENOSPC;
2746 			f2fs_bug_on(sbi, 1);
2747 			goto out_unlock;
2748 		}
2749 	}
2750 	segno = GET_SEG_FROM_SEC(sbi, secno);
2751 	zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2752 
2753 	/* give up on finding another zone */
2754 	if (!init)
2755 		goto got_it;
2756 	if (sbi->secs_per_zone == 1)
2757 		goto got_it;
2758 	if (zoneno == old_zoneno)
2759 		goto got_it;
2760 	for (i = 0; i < NR_CURSEG_TYPE; i++)
2761 		if (CURSEG_I(sbi, i)->zone == zoneno)
2762 			break;
2763 
2764 	if (i < NR_CURSEG_TYPE) {
2765 		/* zone is in user, try another */
2766 		if (zoneno + 1 >= total_zones)
2767 			hint = 0;
2768 		else
2769 			hint = (zoneno + 1) * sbi->secs_per_zone;
2770 		init = false;
2771 		goto find_other_zone;
2772 	}
2773 got_it:
2774 	/* set it as dirty segment in free segmap */
2775 	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2776 
2777 	/* no free section in conventional zone */
2778 	if (new_sec && pinning &&
2779 		!f2fs_valid_pinned_area(sbi, START_BLOCK(sbi, segno))) {
2780 		ret = -EAGAIN;
2781 		goto out_unlock;
2782 	}
2783 	__set_inuse(sbi, segno);
2784 	*newseg = segno;
2785 out_unlock:
2786 	spin_unlock(&free_i->segmap_lock);
2787 
2788 	if (ret == -ENOSPC)
2789 		f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_NO_SEGMENT);
2790 	return ret;
2791 }
2792 
2793 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2794 {
2795 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2796 	struct summary_footer *sum_footer;
2797 	unsigned short seg_type = curseg->seg_type;
2798 
2799 	/* only happen when get_new_segment() fails */
2800 	if (curseg->next_segno == NULL_SEGNO)
2801 		return;
2802 
2803 	curseg->inited = true;
2804 	curseg->segno = curseg->next_segno;
2805 	curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2806 	curseg->next_blkoff = 0;
2807 	curseg->next_segno = NULL_SEGNO;
2808 
2809 	sum_footer = &(curseg->sum_blk->footer);
2810 	memset(sum_footer, 0, sizeof(struct summary_footer));
2811 
2812 	sanity_check_seg_type(sbi, seg_type);
2813 
2814 	if (IS_DATASEG(seg_type))
2815 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2816 	if (IS_NODESEG(seg_type))
2817 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2818 	__set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2819 }
2820 
2821 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2822 {
2823 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2824 	unsigned short seg_type = curseg->seg_type;
2825 
2826 	sanity_check_seg_type(sbi, seg_type);
2827 	if (__is_large_section(sbi)) {
2828 		if (f2fs_need_rand_seg(sbi)) {
2829 			unsigned int hint = GET_SEC_FROM_SEG(sbi, curseg->segno);
2830 
2831 			if (GET_SEC_FROM_SEG(sbi, curseg->segno + 1) != hint)
2832 				return curseg->segno;
2833 			return get_random_u32_inclusive(curseg->segno + 1,
2834 					GET_SEG_FROM_SEC(sbi, hint + 1) - 1);
2835 		}
2836 		return curseg->segno;
2837 	} else if (f2fs_need_rand_seg(sbi)) {
2838 		return get_random_u32_below(MAIN_SECS(sbi) * SEGS_PER_SEC(sbi));
2839 	}
2840 
2841 	/* inmem log may not locate on any segment after mount */
2842 	if (!curseg->inited)
2843 		return 0;
2844 
2845 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2846 		return 0;
2847 
2848 	if (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type))
2849 		return 0;
2850 
2851 	if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2852 		return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2853 
2854 	/* find segments from 0 to reuse freed segments */
2855 	if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2856 		return 0;
2857 
2858 	return curseg->segno;
2859 }
2860 
2861 /*
2862  * Allocate a current working segment.
2863  * This function always allocates a free segment in LFS manner.
2864  */
2865 static int new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2866 {
2867 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2868 	unsigned int segno = curseg->segno;
2869 	bool pinning = type == CURSEG_COLD_DATA_PINNED;
2870 	int ret;
2871 
2872 	if (curseg->inited)
2873 		write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, segno));
2874 
2875 	segno = __get_next_segno(sbi, type);
2876 	ret = get_new_segment(sbi, &segno, new_sec, pinning);
2877 	if (ret) {
2878 		if (ret == -ENOSPC)
2879 			curseg->segno = NULL_SEGNO;
2880 		return ret;
2881 	}
2882 
2883 	curseg->next_segno = segno;
2884 	reset_curseg(sbi, type, 1);
2885 	curseg->alloc_type = LFS;
2886 	if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
2887 		curseg->fragment_remained_chunk =
2888 				get_random_u32_inclusive(1, sbi->max_fragment_chunk);
2889 	return 0;
2890 }
2891 
2892 static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2893 					int segno, block_t start)
2894 {
2895 	struct seg_entry *se = get_seg_entry(sbi, segno);
2896 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2897 	unsigned long *target_map = SIT_I(sbi)->tmp_map;
2898 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2899 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2900 	int i;
2901 
2902 	for (i = 0; i < entries; i++)
2903 		target_map[i] = ckpt_map[i] | cur_map[i];
2904 
2905 	return __find_rev_next_zero_bit(target_map, BLKS_PER_SEG(sbi), start);
2906 }
2907 
2908 static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi,
2909 		struct curseg_info *seg)
2910 {
2911 	return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1);
2912 }
2913 
2914 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2915 {
2916 	return __next_free_blkoff(sbi, segno, 0) < BLKS_PER_SEG(sbi);
2917 }
2918 
2919 /*
2920  * This function always allocates a used segment(from dirty seglist) by SSR
2921  * manner, so it should recover the existing segment information of valid blocks
2922  */
2923 static int change_curseg(struct f2fs_sb_info *sbi, int type)
2924 {
2925 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2926 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2927 	unsigned int new_segno = curseg->next_segno;
2928 	struct f2fs_summary_block *sum_node;
2929 	struct page *sum_page;
2930 
2931 	if (curseg->inited)
2932 		write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, curseg->segno));
2933 
2934 	__set_test_and_inuse(sbi, new_segno);
2935 
2936 	mutex_lock(&dirty_i->seglist_lock);
2937 	__remove_dirty_segment(sbi, new_segno, PRE);
2938 	__remove_dirty_segment(sbi, new_segno, DIRTY);
2939 	mutex_unlock(&dirty_i->seglist_lock);
2940 
2941 	reset_curseg(sbi, type, 1);
2942 	curseg->alloc_type = SSR;
2943 	curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
2944 
2945 	sum_page = f2fs_get_sum_page(sbi, new_segno);
2946 	if (IS_ERR(sum_page)) {
2947 		/* GC won't be able to use stale summary pages by cp_error */
2948 		memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2949 		return PTR_ERR(sum_page);
2950 	}
2951 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2952 	memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2953 	f2fs_put_page(sum_page, 1);
2954 	return 0;
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 static int get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2961 					int target_type, int alloc_mode,
2962 					unsigned long long age)
2963 {
2964 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2965 	int ret = 0;
2966 
2967 	curseg->seg_type = target_type;
2968 
2969 	if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2970 		struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2971 
2972 		curseg->seg_type = se->type;
2973 		ret = change_curseg(sbi, type);
2974 	} else {
2975 		/* allocate cold segment by default */
2976 		curseg->seg_type = CURSEG_COLD_DATA;
2977 		ret = new_curseg(sbi, type, true);
2978 	}
2979 	stat_inc_seg_type(sbi, curseg);
2980 	return ret;
2981 }
2982 
2983 static int __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi, bool force)
2984 {
2985 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2986 	int ret = 0;
2987 
2988 	if (!sbi->am.atgc_enabled && !force)
2989 		return 0;
2990 
2991 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
2992 
2993 	mutex_lock(&curseg->curseg_mutex);
2994 	down_write(&SIT_I(sbi)->sentry_lock);
2995 
2996 	ret = get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC,
2997 					CURSEG_COLD_DATA, SSR, 0);
2998 
2999 	up_write(&SIT_I(sbi)->sentry_lock);
3000 	mutex_unlock(&curseg->curseg_mutex);
3001 
3002 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3003 	return ret;
3004 }
3005 
3006 int f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
3007 {
3008 	return __f2fs_init_atgc_curseg(sbi, false);
3009 }
3010 
3011 int f2fs_reinit_atgc_curseg(struct f2fs_sb_info *sbi)
3012 {
3013 	int ret;
3014 
3015 	if (!test_opt(sbi, ATGC))
3016 		return 0;
3017 	if (sbi->am.atgc_enabled)
3018 		return 0;
3019 	if (le64_to_cpu(F2FS_CKPT(sbi)->elapsed_time) <
3020 			sbi->am.age_threshold)
3021 		return 0;
3022 
3023 	ret = __f2fs_init_atgc_curseg(sbi, true);
3024 	if (!ret) {
3025 		sbi->am.atgc_enabled = true;
3026 		f2fs_info(sbi, "reenabled age threshold GC");
3027 	}
3028 	return ret;
3029 }
3030 
3031 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
3032 {
3033 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3034 
3035 	mutex_lock(&curseg->curseg_mutex);
3036 	if (!curseg->inited)
3037 		goto out;
3038 
3039 	if (get_valid_blocks(sbi, curseg->segno, false)) {
3040 		write_sum_page(sbi, curseg->sum_blk,
3041 				GET_SUM_BLOCK(sbi, curseg->segno));
3042 	} else {
3043 		mutex_lock(&DIRTY_I(sbi)->seglist_lock);
3044 		__set_test_and_free(sbi, curseg->segno, true);
3045 		mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
3046 	}
3047 out:
3048 	mutex_unlock(&curseg->curseg_mutex);
3049 }
3050 
3051 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
3052 {
3053 	__f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
3054 
3055 	if (sbi->am.atgc_enabled)
3056 		__f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
3057 }
3058 
3059 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
3060 {
3061 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3062 
3063 	mutex_lock(&curseg->curseg_mutex);
3064 	if (!curseg->inited)
3065 		goto out;
3066 	if (get_valid_blocks(sbi, curseg->segno, false))
3067 		goto out;
3068 
3069 	mutex_lock(&DIRTY_I(sbi)->seglist_lock);
3070 	__set_test_and_inuse(sbi, curseg->segno);
3071 	mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
3072 out:
3073 	mutex_unlock(&curseg->curseg_mutex);
3074 }
3075 
3076 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
3077 {
3078 	__f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
3079 
3080 	if (sbi->am.atgc_enabled)
3081 		__f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
3082 }
3083 
3084 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
3085 				int alloc_mode, unsigned long long age)
3086 {
3087 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3088 	unsigned segno = NULL_SEGNO;
3089 	unsigned short seg_type = curseg->seg_type;
3090 	int i, cnt;
3091 	bool reversed = false;
3092 
3093 	sanity_check_seg_type(sbi, seg_type);
3094 
3095 	/* f2fs_need_SSR() already forces to do this */
3096 	if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type,
3097 				alloc_mode, age, false)) {
3098 		curseg->next_segno = segno;
3099 		return 1;
3100 	}
3101 
3102 	/* For node segments, let's do SSR more intensively */
3103 	if (IS_NODESEG(seg_type)) {
3104 		if (seg_type >= CURSEG_WARM_NODE) {
3105 			reversed = true;
3106 			i = CURSEG_COLD_NODE;
3107 		} else {
3108 			i = CURSEG_HOT_NODE;
3109 		}
3110 		cnt = NR_CURSEG_NODE_TYPE;
3111 	} else {
3112 		if (seg_type >= CURSEG_WARM_DATA) {
3113 			reversed = true;
3114 			i = CURSEG_COLD_DATA;
3115 		} else {
3116 			i = CURSEG_HOT_DATA;
3117 		}
3118 		cnt = NR_CURSEG_DATA_TYPE;
3119 	}
3120 
3121 	for (; cnt-- > 0; reversed ? i-- : i++) {
3122 		if (i == seg_type)
3123 			continue;
3124 		if (!f2fs_get_victim(sbi, &segno, BG_GC, i,
3125 					alloc_mode, age, false)) {
3126 			curseg->next_segno = segno;
3127 			return 1;
3128 		}
3129 	}
3130 
3131 	/* find valid_blocks=0 in dirty list */
3132 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
3133 		segno = get_free_segment(sbi);
3134 		if (segno != NULL_SEGNO) {
3135 			curseg->next_segno = segno;
3136 			return 1;
3137 		}
3138 	}
3139 	return 0;
3140 }
3141 
3142 static bool need_new_seg(struct f2fs_sb_info *sbi, int type)
3143 {
3144 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3145 
3146 	if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
3147 	    curseg->seg_type == CURSEG_WARM_NODE)
3148 		return true;
3149 	if (curseg->alloc_type == LFS && is_next_segment_free(sbi, curseg) &&
3150 	    likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
3151 		return true;
3152 	if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0))
3153 		return true;
3154 	return false;
3155 }
3156 
3157 int f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
3158 					unsigned int start, unsigned int end)
3159 {
3160 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3161 	unsigned int segno;
3162 	int ret = 0;
3163 
3164 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3165 	mutex_lock(&curseg->curseg_mutex);
3166 	down_write(&SIT_I(sbi)->sentry_lock);
3167 
3168 	segno = CURSEG_I(sbi, type)->segno;
3169 	if (segno < start || segno > end)
3170 		goto unlock;
3171 
3172 	if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
3173 		ret = change_curseg(sbi, type);
3174 	else
3175 		ret = new_curseg(sbi, type, true);
3176 
3177 	stat_inc_seg_type(sbi, curseg);
3178 
3179 	locate_dirty_segment(sbi, segno);
3180 unlock:
3181 	up_write(&SIT_I(sbi)->sentry_lock);
3182 
3183 	if (segno != curseg->segno)
3184 		f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
3185 			    type, segno, curseg->segno);
3186 
3187 	mutex_unlock(&curseg->curseg_mutex);
3188 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3189 	return ret;
3190 }
3191 
3192 static int __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
3193 						bool new_sec, bool force)
3194 {
3195 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3196 	unsigned int old_segno;
3197 	int err = 0;
3198 
3199 	if (type == CURSEG_COLD_DATA_PINNED && !curseg->inited)
3200 		goto allocate;
3201 
3202 	if (!force && curseg->inited &&
3203 	    !curseg->next_blkoff &&
3204 	    !get_valid_blocks(sbi, curseg->segno, new_sec) &&
3205 	    !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
3206 		return 0;
3207 
3208 allocate:
3209 	old_segno = curseg->segno;
3210 	err = new_curseg(sbi, type, true);
3211 	if (err)
3212 		return err;
3213 	stat_inc_seg_type(sbi, curseg);
3214 	locate_dirty_segment(sbi, old_segno);
3215 	return 0;
3216 }
3217 
3218 int f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
3219 {
3220 	int ret;
3221 
3222 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3223 	down_write(&SIT_I(sbi)->sentry_lock);
3224 	ret = __allocate_new_segment(sbi, type, true, force);
3225 	up_write(&SIT_I(sbi)->sentry_lock);
3226 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3227 
3228 	return ret;
3229 }
3230 
3231 int f2fs_allocate_pinning_section(struct f2fs_sb_info *sbi)
3232 {
3233 	int err;
3234 	bool gc_required = true;
3235 
3236 retry:
3237 	f2fs_lock_op(sbi);
3238 	err = f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3239 	f2fs_unlock_op(sbi);
3240 
3241 	if (f2fs_sb_has_blkzoned(sbi) && err == -EAGAIN && gc_required) {
3242 		f2fs_down_write(&sbi->gc_lock);
3243 		err = f2fs_gc_range(sbi, 0, GET_SEGNO(sbi, FDEV(0).end_blk),
3244 				true, ZONED_PIN_SEC_REQUIRED_COUNT);
3245 		f2fs_up_write(&sbi->gc_lock);
3246 
3247 		gc_required = false;
3248 		if (!err)
3249 			goto retry;
3250 	}
3251 
3252 	return err;
3253 }
3254 
3255 int f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
3256 {
3257 	int i;
3258 	int err = 0;
3259 
3260 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3261 	down_write(&SIT_I(sbi)->sentry_lock);
3262 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
3263 		err += __allocate_new_segment(sbi, i, false, false);
3264 	up_write(&SIT_I(sbi)->sentry_lock);
3265 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3266 
3267 	return err;
3268 }
3269 
3270 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
3271 						struct cp_control *cpc)
3272 {
3273 	__u64 trim_start = cpc->trim_start;
3274 	bool has_candidate = false;
3275 
3276 	down_write(&SIT_I(sbi)->sentry_lock);
3277 	for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
3278 		if (add_discard_addrs(sbi, cpc, true)) {
3279 			has_candidate = true;
3280 			break;
3281 		}
3282 	}
3283 	up_write(&SIT_I(sbi)->sentry_lock);
3284 
3285 	cpc->trim_start = trim_start;
3286 	return has_candidate;
3287 }
3288 
3289 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
3290 					struct discard_policy *dpolicy,
3291 					unsigned int start, unsigned int end)
3292 {
3293 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
3294 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
3295 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
3296 	struct discard_cmd *dc;
3297 	struct blk_plug plug;
3298 	int issued;
3299 	unsigned int trimmed = 0;
3300 
3301 next:
3302 	issued = 0;
3303 
3304 	mutex_lock(&dcc->cmd_lock);
3305 	if (unlikely(dcc->rbtree_check))
3306 		f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
3307 
3308 	dc = __lookup_discard_cmd_ret(&dcc->root, start,
3309 				&prev_dc, &next_dc, &insert_p, &insert_parent);
3310 	if (!dc)
3311 		dc = next_dc;
3312 
3313 	blk_start_plug(&plug);
3314 
3315 	while (dc && dc->di.lstart <= end) {
3316 		struct rb_node *node;
3317 		int err = 0;
3318 
3319 		if (dc->di.len < dpolicy->granularity)
3320 			goto skip;
3321 
3322 		if (dc->state != D_PREP) {
3323 			list_move_tail(&dc->list, &dcc->fstrim_list);
3324 			goto skip;
3325 		}
3326 
3327 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3328 
3329 		if (issued >= dpolicy->max_requests) {
3330 			start = dc->di.lstart + dc->di.len;
3331 
3332 			if (err)
3333 				__remove_discard_cmd(sbi, dc);
3334 
3335 			blk_finish_plug(&plug);
3336 			mutex_unlock(&dcc->cmd_lock);
3337 			trimmed += __wait_all_discard_cmd(sbi, NULL);
3338 			f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
3339 			goto next;
3340 		}
3341 skip:
3342 		node = rb_next(&dc->rb_node);
3343 		if (err)
3344 			__remove_discard_cmd(sbi, dc);
3345 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3346 
3347 		if (fatal_signal_pending(current))
3348 			break;
3349 	}
3350 
3351 	blk_finish_plug(&plug);
3352 	mutex_unlock(&dcc->cmd_lock);
3353 
3354 	return trimmed;
3355 }
3356 
3357 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3358 {
3359 	__u64 start = F2FS_BYTES_TO_BLK(range->start);
3360 	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3361 	unsigned int start_segno, end_segno;
3362 	block_t start_block, end_block;
3363 	struct cp_control cpc;
3364 	struct discard_policy dpolicy;
3365 	unsigned long long trimmed = 0;
3366 	int err = 0;
3367 	bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3368 
3369 	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3370 		return -EINVAL;
3371 
3372 	if (end < MAIN_BLKADDR(sbi))
3373 		goto out;
3374 
3375 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3376 		f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3377 		return -EFSCORRUPTED;
3378 	}
3379 
3380 	/* start/end segment number in main_area */
3381 	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3382 	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3383 						GET_SEGNO(sbi, end);
3384 	if (need_align) {
3385 		start_segno = rounddown(start_segno, SEGS_PER_SEC(sbi));
3386 		end_segno = roundup(end_segno + 1, SEGS_PER_SEC(sbi)) - 1;
3387 	}
3388 
3389 	cpc.reason = CP_DISCARD;
3390 	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3391 	cpc.trim_start = start_segno;
3392 	cpc.trim_end = end_segno;
3393 
3394 	if (sbi->discard_blks == 0)
3395 		goto out;
3396 
3397 	f2fs_down_write(&sbi->gc_lock);
3398 	stat_inc_cp_call_count(sbi, TOTAL_CALL);
3399 	err = f2fs_write_checkpoint(sbi, &cpc);
3400 	f2fs_up_write(&sbi->gc_lock);
3401 	if (err)
3402 		goto out;
3403 
3404 	/*
3405 	 * We filed discard candidates, but actually we don't need to wait for
3406 	 * all of them, since they'll be issued in idle time along with runtime
3407 	 * discard option. User configuration looks like using runtime discard
3408 	 * or periodic fstrim instead of it.
3409 	 */
3410 	if (f2fs_realtime_discard_enable(sbi))
3411 		goto out;
3412 
3413 	start_block = START_BLOCK(sbi, start_segno);
3414 	end_block = START_BLOCK(sbi, end_segno + 1);
3415 
3416 	__init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3417 	trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3418 					start_block, end_block);
3419 
3420 	trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3421 					start_block, end_block);
3422 out:
3423 	if (!err)
3424 		range->len = F2FS_BLK_TO_BYTES(trimmed);
3425 	return err;
3426 }
3427 
3428 int f2fs_rw_hint_to_seg_type(struct f2fs_sb_info *sbi, enum rw_hint hint)
3429 {
3430 	if (F2FS_OPTION(sbi).active_logs == 2)
3431 		return CURSEG_HOT_DATA;
3432 	else if (F2FS_OPTION(sbi).active_logs == 4)
3433 		return CURSEG_COLD_DATA;
3434 
3435 	/* active_log == 6 */
3436 	switch (hint) {
3437 	case WRITE_LIFE_SHORT:
3438 		return CURSEG_HOT_DATA;
3439 	case WRITE_LIFE_EXTREME:
3440 		return CURSEG_COLD_DATA;
3441 	default:
3442 		return CURSEG_WARM_DATA;
3443 	}
3444 }
3445 
3446 /*
3447  * This returns write hints for each segment type. This hints will be
3448  * passed down to block layer as below by default.
3449  *
3450  * User                  F2FS                     Block
3451  * ----                  ----                     -----
3452  *                       META                     WRITE_LIFE_NONE|REQ_META
3453  *                       HOT_NODE                 WRITE_LIFE_NONE
3454  *                       WARM_NODE                WRITE_LIFE_MEDIUM
3455  *                       COLD_NODE                WRITE_LIFE_LONG
3456  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
3457  * extension list        "                        "
3458  *
3459  * -- buffered io
3460  *                       COLD_DATA                WRITE_LIFE_EXTREME
3461  *                       HOT_DATA                 WRITE_LIFE_SHORT
3462  *                       WARM_DATA                WRITE_LIFE_NOT_SET
3463  *
3464  * -- direct io
3465  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3466  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3467  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3468  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
3469  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
3470  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
3471  */
3472 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
3473 				enum page_type type, enum temp_type temp)
3474 {
3475 	switch (type) {
3476 	case DATA:
3477 		switch (temp) {
3478 		case WARM:
3479 			return WRITE_LIFE_NOT_SET;
3480 		case HOT:
3481 			return WRITE_LIFE_SHORT;
3482 		case COLD:
3483 			return WRITE_LIFE_EXTREME;
3484 		default:
3485 			return WRITE_LIFE_NONE;
3486 		}
3487 	case NODE:
3488 		switch (temp) {
3489 		case WARM:
3490 			return WRITE_LIFE_MEDIUM;
3491 		case HOT:
3492 			return WRITE_LIFE_NONE;
3493 		case COLD:
3494 			return WRITE_LIFE_LONG;
3495 		default:
3496 			return WRITE_LIFE_NONE;
3497 		}
3498 	case META:
3499 		return WRITE_LIFE_NONE;
3500 	default:
3501 		return WRITE_LIFE_NONE;
3502 	}
3503 }
3504 
3505 static int __get_segment_type_2(struct f2fs_io_info *fio)
3506 {
3507 	if (fio->type == DATA)
3508 		return CURSEG_HOT_DATA;
3509 	else
3510 		return CURSEG_HOT_NODE;
3511 }
3512 
3513 static int __get_segment_type_4(struct f2fs_io_info *fio)
3514 {
3515 	if (fio->type == DATA) {
3516 		struct inode *inode = fio->page->mapping->host;
3517 
3518 		if (S_ISDIR(inode->i_mode))
3519 			return CURSEG_HOT_DATA;
3520 		else
3521 			return CURSEG_COLD_DATA;
3522 	} else {
3523 		if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3524 			return CURSEG_WARM_NODE;
3525 		else
3526 			return CURSEG_COLD_NODE;
3527 	}
3528 }
3529 
3530 static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs)
3531 {
3532 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3533 	struct extent_info ei = {};
3534 
3535 	if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) {
3536 		if (!ei.age)
3537 			return NO_CHECK_TYPE;
3538 		if (ei.age <= sbi->hot_data_age_threshold)
3539 			return CURSEG_HOT_DATA;
3540 		if (ei.age <= sbi->warm_data_age_threshold)
3541 			return CURSEG_WARM_DATA;
3542 		return CURSEG_COLD_DATA;
3543 	}
3544 	return NO_CHECK_TYPE;
3545 }
3546 
3547 static int __get_segment_type_6(struct f2fs_io_info *fio)
3548 {
3549 	if (fio->type == DATA) {
3550 		struct inode *inode = fio->page->mapping->host;
3551 		int type;
3552 
3553 		if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
3554 			return CURSEG_COLD_DATA_PINNED;
3555 
3556 		if (page_private_gcing(fio->page)) {
3557 			if (fio->sbi->am.atgc_enabled &&
3558 				(fio->io_type == FS_DATA_IO) &&
3559 				(fio->sbi->gc_mode != GC_URGENT_HIGH) &&
3560 				__is_valid_data_blkaddr(fio->old_blkaddr) &&
3561 				!is_inode_flag_set(inode, FI_OPU_WRITE))
3562 				return CURSEG_ALL_DATA_ATGC;
3563 			else
3564 				return CURSEG_COLD_DATA;
3565 		}
3566 		if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3567 			return CURSEG_COLD_DATA;
3568 
3569 		type = __get_age_segment_type(inode,
3570 				page_folio(fio->page)->index);
3571 		if (type != NO_CHECK_TYPE)
3572 			return type;
3573 
3574 		if (file_is_hot(inode) ||
3575 				is_inode_flag_set(inode, FI_HOT_DATA) ||
3576 				f2fs_is_cow_file(inode))
3577 			return CURSEG_HOT_DATA;
3578 		return f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode),
3579 						inode->i_write_hint);
3580 	} else {
3581 		if (IS_DNODE(fio->page))
3582 			return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3583 						CURSEG_HOT_NODE;
3584 		return CURSEG_COLD_NODE;
3585 	}
3586 }
3587 
3588 enum temp_type f2fs_get_segment_temp(struct f2fs_sb_info *sbi,
3589 						enum log_type type)
3590 {
3591 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3592 	enum temp_type temp = COLD;
3593 
3594 	switch (curseg->seg_type) {
3595 	case CURSEG_HOT_NODE:
3596 	case CURSEG_HOT_DATA:
3597 		temp = HOT;
3598 		break;
3599 	case CURSEG_WARM_NODE:
3600 	case CURSEG_WARM_DATA:
3601 		temp = WARM;
3602 		break;
3603 	case CURSEG_COLD_NODE:
3604 	case CURSEG_COLD_DATA:
3605 		temp = COLD;
3606 		break;
3607 	default:
3608 		f2fs_bug_on(sbi, 1);
3609 	}
3610 
3611 	return temp;
3612 }
3613 
3614 static int __get_segment_type(struct f2fs_io_info *fio)
3615 {
3616 	enum log_type type = CURSEG_HOT_DATA;
3617 
3618 	switch (F2FS_OPTION(fio->sbi).active_logs) {
3619 	case 2:
3620 		type = __get_segment_type_2(fio);
3621 		break;
3622 	case 4:
3623 		type = __get_segment_type_4(fio);
3624 		break;
3625 	case 6:
3626 		type = __get_segment_type_6(fio);
3627 		break;
3628 	default:
3629 		f2fs_bug_on(fio->sbi, true);
3630 	}
3631 
3632 	fio->temp = f2fs_get_segment_temp(fio->sbi, type);
3633 
3634 	return type;
3635 }
3636 
3637 static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi,
3638 		struct curseg_info *seg)
3639 {
3640 	/* To allocate block chunks in different sizes, use random number */
3641 	if (--seg->fragment_remained_chunk > 0)
3642 		return;
3643 
3644 	seg->fragment_remained_chunk =
3645 		get_random_u32_inclusive(1, sbi->max_fragment_chunk);
3646 	seg->next_blkoff +=
3647 		get_random_u32_inclusive(1, sbi->max_fragment_hole);
3648 }
3649 
3650 static void reset_curseg_fields(struct curseg_info *curseg)
3651 {
3652 	curseg->inited = false;
3653 	curseg->segno = NULL_SEGNO;
3654 	curseg->next_segno = 0;
3655 }
3656 
3657 int f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3658 		block_t old_blkaddr, block_t *new_blkaddr,
3659 		struct f2fs_summary *sum, int type,
3660 		struct f2fs_io_info *fio)
3661 {
3662 	struct sit_info *sit_i = SIT_I(sbi);
3663 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3664 	unsigned long long old_mtime;
3665 	bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3666 	struct seg_entry *se = NULL;
3667 	bool segment_full = false;
3668 	int ret = 0;
3669 
3670 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3671 
3672 	mutex_lock(&curseg->curseg_mutex);
3673 	down_write(&sit_i->sentry_lock);
3674 
3675 	if (curseg->segno == NULL_SEGNO) {
3676 		ret = -ENOSPC;
3677 		goto out_err;
3678 	}
3679 
3680 	if (from_gc) {
3681 		f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3682 		se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3683 		sanity_check_seg_type(sbi, se->type);
3684 		f2fs_bug_on(sbi, IS_NODESEG(se->type));
3685 	}
3686 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3687 
3688 	f2fs_bug_on(sbi, curseg->next_blkoff >= BLKS_PER_SEG(sbi));
3689 
3690 	f2fs_wait_discard_bio(sbi, *new_blkaddr);
3691 
3692 	curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3693 	if (curseg->alloc_type == SSR) {
3694 		curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg);
3695 	} else {
3696 		curseg->next_blkoff++;
3697 		if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
3698 			f2fs_randomize_chunk(sbi, curseg);
3699 	}
3700 	if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno))
3701 		segment_full = true;
3702 	stat_inc_block_count(sbi, curseg);
3703 
3704 	if (from_gc) {
3705 		old_mtime = get_segment_mtime(sbi, old_blkaddr);
3706 	} else {
3707 		update_segment_mtime(sbi, old_blkaddr, 0);
3708 		old_mtime = 0;
3709 	}
3710 	update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3711 
3712 	/*
3713 	 * SIT information should be updated before segment allocation,
3714 	 * since SSR needs latest valid block information.
3715 	 */
3716 	update_sit_entry(sbi, *new_blkaddr, 1);
3717 	update_sit_entry(sbi, old_blkaddr, -1);
3718 
3719 	/*
3720 	 * If the current segment is full, flush it out and replace it with a
3721 	 * new segment.
3722 	 */
3723 	if (segment_full) {
3724 		if (type == CURSEG_COLD_DATA_PINNED &&
3725 		    !((curseg->segno + 1) % sbi->segs_per_sec)) {
3726 			write_sum_page(sbi, curseg->sum_blk,
3727 					GET_SUM_BLOCK(sbi, curseg->segno));
3728 			reset_curseg_fields(curseg);
3729 			goto skip_new_segment;
3730 		}
3731 
3732 		if (from_gc) {
3733 			ret = get_atssr_segment(sbi, type, se->type,
3734 						AT_SSR, se->mtime);
3735 		} else {
3736 			if (need_new_seg(sbi, type))
3737 				ret = new_curseg(sbi, type, false);
3738 			else
3739 				ret = change_curseg(sbi, type);
3740 			stat_inc_seg_type(sbi, curseg);
3741 		}
3742 
3743 		if (ret)
3744 			goto out_err;
3745 	}
3746 
3747 skip_new_segment:
3748 	/*
3749 	 * segment dirty status should be updated after segment allocation,
3750 	 * so we just need to update status only one time after previous
3751 	 * segment being closed.
3752 	 */
3753 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3754 	locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3755 
3756 	if (IS_DATASEG(curseg->seg_type))
3757 		atomic64_inc(&sbi->allocated_data_blocks);
3758 
3759 	up_write(&sit_i->sentry_lock);
3760 
3761 	if (page && IS_NODESEG(curseg->seg_type)) {
3762 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3763 
3764 		f2fs_inode_chksum_set(sbi, page);
3765 	}
3766 
3767 	if (fio) {
3768 		struct f2fs_bio_info *io;
3769 
3770 		INIT_LIST_HEAD(&fio->list);
3771 		fio->in_list = 1;
3772 		io = sbi->write_io[fio->type] + fio->temp;
3773 		spin_lock(&io->io_lock);
3774 		list_add_tail(&fio->list, &io->io_list);
3775 		spin_unlock(&io->io_lock);
3776 	}
3777 
3778 	mutex_unlock(&curseg->curseg_mutex);
3779 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3780 	return 0;
3781 
3782 out_err:
3783 	*new_blkaddr = NULL_ADDR;
3784 	up_write(&sit_i->sentry_lock);
3785 	mutex_unlock(&curseg->curseg_mutex);
3786 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3787 	return ret;
3788 }
3789 
3790 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino,
3791 					block_t blkaddr, unsigned int blkcnt)
3792 {
3793 	if (!f2fs_is_multi_device(sbi))
3794 		return;
3795 
3796 	while (1) {
3797 		unsigned int devidx = f2fs_target_device_index(sbi, blkaddr);
3798 		unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1;
3799 
3800 		/* update device state for fsync */
3801 		f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO);
3802 
3803 		/* update device state for checkpoint */
3804 		if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3805 			spin_lock(&sbi->dev_lock);
3806 			f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3807 			spin_unlock(&sbi->dev_lock);
3808 		}
3809 
3810 		if (blkcnt <= blks)
3811 			break;
3812 		blkcnt -= blks;
3813 		blkaddr += blks;
3814 	}
3815 }
3816 
3817 static int log_type_to_seg_type(enum log_type type)
3818 {
3819 	int seg_type = CURSEG_COLD_DATA;
3820 
3821 	switch (type) {
3822 	case CURSEG_HOT_DATA:
3823 	case CURSEG_WARM_DATA:
3824 	case CURSEG_COLD_DATA:
3825 	case CURSEG_HOT_NODE:
3826 	case CURSEG_WARM_NODE:
3827 	case CURSEG_COLD_NODE:
3828 		seg_type = (int)type;
3829 		break;
3830 	case CURSEG_COLD_DATA_PINNED:
3831 	case CURSEG_ALL_DATA_ATGC:
3832 		seg_type = CURSEG_COLD_DATA;
3833 		break;
3834 	default:
3835 		break;
3836 	}
3837 	return seg_type;
3838 }
3839 
3840 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3841 {
3842 	enum log_type type = __get_segment_type(fio);
3843 	int seg_type = log_type_to_seg_type(type);
3844 	bool keep_order = (f2fs_lfs_mode(fio->sbi) &&
3845 				seg_type == CURSEG_COLD_DATA);
3846 
3847 	if (keep_order)
3848 		f2fs_down_read(&fio->sbi->io_order_lock);
3849 
3850 	if (f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3851 			&fio->new_blkaddr, sum, type, fio)) {
3852 		if (fscrypt_inode_uses_fs_layer_crypto(fio->page->mapping->host))
3853 			fscrypt_finalize_bounce_page(&fio->encrypted_page);
3854 		end_page_writeback(fio->page);
3855 		if (f2fs_in_warm_node_list(fio->sbi, fio->page))
3856 			f2fs_del_fsync_node_entry(fio->sbi, fio->page);
3857 		goto out;
3858 	}
3859 	if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3860 		f2fs_invalidate_internal_cache(fio->sbi, fio->old_blkaddr);
3861 
3862 	/* writeout dirty page into bdev */
3863 	f2fs_submit_page_write(fio);
3864 
3865 	f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1);
3866 out:
3867 	if (keep_order)
3868 		f2fs_up_read(&fio->sbi->io_order_lock);
3869 }
3870 
3871 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct folio *folio,
3872 					enum iostat_type io_type)
3873 {
3874 	struct f2fs_io_info fio = {
3875 		.sbi = sbi,
3876 		.type = META,
3877 		.temp = HOT,
3878 		.op = REQ_OP_WRITE,
3879 		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3880 		.old_blkaddr = folio->index,
3881 		.new_blkaddr = folio->index,
3882 		.page = folio_page(folio, 0),
3883 		.encrypted_page = NULL,
3884 		.in_list = 0,
3885 	};
3886 
3887 	if (unlikely(folio->index >= MAIN_BLKADDR(sbi)))
3888 		fio.op_flags &= ~REQ_META;
3889 
3890 	folio_start_writeback(folio);
3891 	f2fs_submit_page_write(&fio);
3892 
3893 	stat_inc_meta_count(sbi, folio->index);
3894 	f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE);
3895 }
3896 
3897 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3898 {
3899 	struct f2fs_summary sum;
3900 
3901 	set_summary(&sum, nid, 0, 0);
3902 	do_write_page(&sum, fio);
3903 
3904 	f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE);
3905 }
3906 
3907 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3908 					struct f2fs_io_info *fio)
3909 {
3910 	struct f2fs_sb_info *sbi = fio->sbi;
3911 	struct f2fs_summary sum;
3912 
3913 	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3914 	if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO)
3915 		f2fs_update_age_extent_cache(dn);
3916 	set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3917 	do_write_page(&sum, fio);
3918 	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3919 
3920 	f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE);
3921 }
3922 
3923 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3924 {
3925 	int err;
3926 	struct f2fs_sb_info *sbi = fio->sbi;
3927 	unsigned int segno;
3928 
3929 	fio->new_blkaddr = fio->old_blkaddr;
3930 	/* i/o temperature is needed for passing down write hints */
3931 	__get_segment_type(fio);
3932 
3933 	segno = GET_SEGNO(sbi, fio->new_blkaddr);
3934 
3935 	if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3936 		set_sbi_flag(sbi, SBI_NEED_FSCK);
3937 		f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3938 			  __func__, segno);
3939 		err = -EFSCORRUPTED;
3940 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
3941 		goto drop_bio;
3942 	}
3943 
3944 	if (f2fs_cp_error(sbi)) {
3945 		err = -EIO;
3946 		goto drop_bio;
3947 	}
3948 
3949 	if (fio->meta_gc)
3950 		f2fs_truncate_meta_inode_pages(sbi, fio->new_blkaddr, 1);
3951 
3952 	stat_inc_inplace_blocks(fio->sbi);
3953 
3954 	if (fio->bio && !IS_F2FS_IPU_NOCACHE(sbi))
3955 		err = f2fs_merge_page_bio(fio);
3956 	else
3957 		err = f2fs_submit_page_bio(fio);
3958 	if (!err) {
3959 		f2fs_update_device_state(fio->sbi, fio->ino,
3960 						fio->new_blkaddr, 1);
3961 		f2fs_update_iostat(fio->sbi, fio->page->mapping->host,
3962 						fio->io_type, F2FS_BLKSIZE);
3963 	}
3964 
3965 	return err;
3966 drop_bio:
3967 	if (fio->bio && *(fio->bio)) {
3968 		struct bio *bio = *(fio->bio);
3969 
3970 		bio->bi_status = BLK_STS_IOERR;
3971 		bio_endio(bio);
3972 		*(fio->bio) = NULL;
3973 	}
3974 	return err;
3975 }
3976 
3977 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3978 						unsigned int segno)
3979 {
3980 	int i;
3981 
3982 	for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3983 		if (CURSEG_I(sbi, i)->segno == segno)
3984 			break;
3985 	}
3986 	return i;
3987 }
3988 
3989 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3990 				block_t old_blkaddr, block_t new_blkaddr,
3991 				bool recover_curseg, bool recover_newaddr,
3992 				bool from_gc)
3993 {
3994 	struct sit_info *sit_i = SIT_I(sbi);
3995 	struct curseg_info *curseg;
3996 	unsigned int segno, old_cursegno;
3997 	struct seg_entry *se;
3998 	int type;
3999 	unsigned short old_blkoff;
4000 	unsigned char old_alloc_type;
4001 
4002 	segno = GET_SEGNO(sbi, new_blkaddr);
4003 	se = get_seg_entry(sbi, segno);
4004 	type = se->type;
4005 
4006 	f2fs_down_write(&SM_I(sbi)->curseg_lock);
4007 
4008 	if (!recover_curseg) {
4009 		/* for recovery flow */
4010 		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
4011 			if (old_blkaddr == NULL_ADDR)
4012 				type = CURSEG_COLD_DATA;
4013 			else
4014 				type = CURSEG_WARM_DATA;
4015 		}
4016 	} else {
4017 		if (IS_CURSEG(sbi, segno)) {
4018 			/* se->type is volatile as SSR allocation */
4019 			type = __f2fs_get_curseg(sbi, segno);
4020 			f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
4021 		} else {
4022 			type = CURSEG_WARM_DATA;
4023 		}
4024 	}
4025 
4026 	curseg = CURSEG_I(sbi, type);
4027 	f2fs_bug_on(sbi, !IS_DATASEG(curseg->seg_type));
4028 
4029 	mutex_lock(&curseg->curseg_mutex);
4030 	down_write(&sit_i->sentry_lock);
4031 
4032 	old_cursegno = curseg->segno;
4033 	old_blkoff = curseg->next_blkoff;
4034 	old_alloc_type = curseg->alloc_type;
4035 
4036 	/* change the current segment */
4037 	if (segno != curseg->segno) {
4038 		curseg->next_segno = segno;
4039 		if (change_curseg(sbi, type))
4040 			goto out_unlock;
4041 	}
4042 
4043 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
4044 	curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
4045 
4046 	if (!recover_curseg || recover_newaddr) {
4047 		if (!from_gc)
4048 			update_segment_mtime(sbi, new_blkaddr, 0);
4049 		update_sit_entry(sbi, new_blkaddr, 1);
4050 	}
4051 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
4052 		f2fs_invalidate_internal_cache(sbi, old_blkaddr);
4053 		if (!from_gc)
4054 			update_segment_mtime(sbi, old_blkaddr, 0);
4055 		update_sit_entry(sbi, old_blkaddr, -1);
4056 	}
4057 
4058 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
4059 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
4060 
4061 	locate_dirty_segment(sbi, old_cursegno);
4062 
4063 	if (recover_curseg) {
4064 		if (old_cursegno != curseg->segno) {
4065 			curseg->next_segno = old_cursegno;
4066 			if (change_curseg(sbi, type))
4067 				goto out_unlock;
4068 		}
4069 		curseg->next_blkoff = old_blkoff;
4070 		curseg->alloc_type = old_alloc_type;
4071 	}
4072 
4073 out_unlock:
4074 	up_write(&sit_i->sentry_lock);
4075 	mutex_unlock(&curseg->curseg_mutex);
4076 	f2fs_up_write(&SM_I(sbi)->curseg_lock);
4077 }
4078 
4079 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
4080 				block_t old_addr, block_t new_addr,
4081 				unsigned char version, bool recover_curseg,
4082 				bool recover_newaddr)
4083 {
4084 	struct f2fs_summary sum;
4085 
4086 	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
4087 
4088 	f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
4089 					recover_curseg, recover_newaddr, false);
4090 
4091 	f2fs_update_data_blkaddr(dn, new_addr);
4092 }
4093 
4094 void f2fs_wait_on_page_writeback(struct page *page,
4095 				enum page_type type, bool ordered, bool locked)
4096 {
4097 	if (folio_test_writeback(page_folio(page))) {
4098 		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
4099 
4100 		/* submit cached LFS IO */
4101 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
4102 		/* submit cached IPU IO */
4103 		f2fs_submit_merged_ipu_write(sbi, NULL, page);
4104 		if (ordered) {
4105 			wait_on_page_writeback(page);
4106 			f2fs_bug_on(sbi, locked &&
4107 				folio_test_writeback(page_folio(page)));
4108 		} else {
4109 			wait_for_stable_page(page);
4110 		}
4111 	}
4112 }
4113 
4114 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
4115 {
4116 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
4117 	struct page *cpage;
4118 
4119 	if (!f2fs_meta_inode_gc_required(inode))
4120 		return;
4121 
4122 	if (!__is_valid_data_blkaddr(blkaddr))
4123 		return;
4124 
4125 	cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
4126 	if (cpage) {
4127 		f2fs_wait_on_page_writeback(cpage, DATA, true, true);
4128 		f2fs_put_page(cpage, 1);
4129 	}
4130 }
4131 
4132 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
4133 								block_t len)
4134 {
4135 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
4136 	block_t i;
4137 
4138 	if (!f2fs_meta_inode_gc_required(inode))
4139 		return;
4140 
4141 	for (i = 0; i < len; i++)
4142 		f2fs_wait_on_block_writeback(inode, blkaddr + i);
4143 
4144 	f2fs_truncate_meta_inode_pages(sbi, blkaddr, len);
4145 }
4146 
4147 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
4148 {
4149 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4150 	struct curseg_info *seg_i;
4151 	unsigned char *kaddr;
4152 	struct page *page;
4153 	block_t start;
4154 	int i, j, offset;
4155 
4156 	start = start_sum_block(sbi);
4157 
4158 	page = f2fs_get_meta_page(sbi, start++);
4159 	if (IS_ERR(page))
4160 		return PTR_ERR(page);
4161 	kaddr = (unsigned char *)page_address(page);
4162 
4163 	/* Step 1: restore nat cache */
4164 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
4165 	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
4166 
4167 	/* Step 2: restore sit cache */
4168 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
4169 	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
4170 	offset = 2 * SUM_JOURNAL_SIZE;
4171 
4172 	/* Step 3: restore summary entries */
4173 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
4174 		unsigned short blk_off;
4175 		unsigned int segno;
4176 
4177 		seg_i = CURSEG_I(sbi, i);
4178 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
4179 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
4180 		seg_i->next_segno = segno;
4181 		reset_curseg(sbi, i, 0);
4182 		seg_i->alloc_type = ckpt->alloc_type[i];
4183 		seg_i->next_blkoff = blk_off;
4184 
4185 		if (seg_i->alloc_type == SSR)
4186 			blk_off = BLKS_PER_SEG(sbi);
4187 
4188 		for (j = 0; j < blk_off; j++) {
4189 			struct f2fs_summary *s;
4190 
4191 			s = (struct f2fs_summary *)(kaddr + offset);
4192 			seg_i->sum_blk->entries[j] = *s;
4193 			offset += SUMMARY_SIZE;
4194 			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
4195 						SUM_FOOTER_SIZE)
4196 				continue;
4197 
4198 			f2fs_put_page(page, 1);
4199 			page = NULL;
4200 
4201 			page = f2fs_get_meta_page(sbi, start++);
4202 			if (IS_ERR(page))
4203 				return PTR_ERR(page);
4204 			kaddr = (unsigned char *)page_address(page);
4205 			offset = 0;
4206 		}
4207 	}
4208 	f2fs_put_page(page, 1);
4209 	return 0;
4210 }
4211 
4212 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
4213 {
4214 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4215 	struct f2fs_summary_block *sum;
4216 	struct curseg_info *curseg;
4217 	struct page *new;
4218 	unsigned short blk_off;
4219 	unsigned int segno = 0;
4220 	block_t blk_addr = 0;
4221 	int err = 0;
4222 
4223 	/* get segment number and block addr */
4224 	if (IS_DATASEG(type)) {
4225 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
4226 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
4227 							CURSEG_HOT_DATA]);
4228 		if (__exist_node_summaries(sbi))
4229 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
4230 		else
4231 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
4232 	} else {
4233 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
4234 							CURSEG_HOT_NODE]);
4235 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
4236 							CURSEG_HOT_NODE]);
4237 		if (__exist_node_summaries(sbi))
4238 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
4239 							type - CURSEG_HOT_NODE);
4240 		else
4241 			blk_addr = GET_SUM_BLOCK(sbi, segno);
4242 	}
4243 
4244 	new = f2fs_get_meta_page(sbi, blk_addr);
4245 	if (IS_ERR(new))
4246 		return PTR_ERR(new);
4247 	sum = (struct f2fs_summary_block *)page_address(new);
4248 
4249 	if (IS_NODESEG(type)) {
4250 		if (__exist_node_summaries(sbi)) {
4251 			struct f2fs_summary *ns = &sum->entries[0];
4252 			int i;
4253 
4254 			for (i = 0; i < BLKS_PER_SEG(sbi); i++, ns++) {
4255 				ns->version = 0;
4256 				ns->ofs_in_node = 0;
4257 			}
4258 		} else {
4259 			err = f2fs_restore_node_summary(sbi, segno, sum);
4260 			if (err)
4261 				goto out;
4262 		}
4263 	}
4264 
4265 	/* set uncompleted segment to curseg */
4266 	curseg = CURSEG_I(sbi, type);
4267 	mutex_lock(&curseg->curseg_mutex);
4268 
4269 	/* update journal info */
4270 	down_write(&curseg->journal_rwsem);
4271 	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
4272 	up_write(&curseg->journal_rwsem);
4273 
4274 	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
4275 	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
4276 	curseg->next_segno = segno;
4277 	reset_curseg(sbi, type, 0);
4278 	curseg->alloc_type = ckpt->alloc_type[type];
4279 	curseg->next_blkoff = blk_off;
4280 	mutex_unlock(&curseg->curseg_mutex);
4281 out:
4282 	f2fs_put_page(new, 1);
4283 	return err;
4284 }
4285 
4286 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
4287 {
4288 	struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
4289 	struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
4290 	int type = CURSEG_HOT_DATA;
4291 	int err;
4292 
4293 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
4294 		int npages = f2fs_npages_for_summary_flush(sbi, true);
4295 
4296 		if (npages >= 2)
4297 			f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
4298 							META_CP, true);
4299 
4300 		/* restore for compacted data summary */
4301 		err = read_compacted_summaries(sbi);
4302 		if (err)
4303 			return err;
4304 		type = CURSEG_HOT_NODE;
4305 	}
4306 
4307 	if (__exist_node_summaries(sbi))
4308 		f2fs_ra_meta_pages(sbi,
4309 				sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
4310 				NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
4311 
4312 	for (; type <= CURSEG_COLD_NODE; type++) {
4313 		err = read_normal_summaries(sbi, type);
4314 		if (err)
4315 			return err;
4316 	}
4317 
4318 	/* sanity check for summary blocks */
4319 	if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
4320 			sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
4321 		f2fs_err(sbi, "invalid journal entries nats %u sits %u",
4322 			 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
4323 		return -EINVAL;
4324 	}
4325 
4326 	return 0;
4327 }
4328 
4329 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
4330 {
4331 	struct page *page;
4332 	unsigned char *kaddr;
4333 	struct f2fs_summary *summary;
4334 	struct curseg_info *seg_i;
4335 	int written_size = 0;
4336 	int i, j;
4337 
4338 	page = f2fs_grab_meta_page(sbi, blkaddr++);
4339 	kaddr = (unsigned char *)page_address(page);
4340 	memset(kaddr, 0, PAGE_SIZE);
4341 
4342 	/* Step 1: write nat cache */
4343 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
4344 	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
4345 	written_size += SUM_JOURNAL_SIZE;
4346 
4347 	/* Step 2: write sit cache */
4348 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
4349 	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
4350 	written_size += SUM_JOURNAL_SIZE;
4351 
4352 	/* Step 3: write summary entries */
4353 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
4354 		seg_i = CURSEG_I(sbi, i);
4355 		for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); j++) {
4356 			if (!page) {
4357 				page = f2fs_grab_meta_page(sbi, blkaddr++);
4358 				kaddr = (unsigned char *)page_address(page);
4359 				memset(kaddr, 0, PAGE_SIZE);
4360 				written_size = 0;
4361 			}
4362 			summary = (struct f2fs_summary *)(kaddr + written_size);
4363 			*summary = seg_i->sum_blk->entries[j];
4364 			written_size += SUMMARY_SIZE;
4365 
4366 			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
4367 							SUM_FOOTER_SIZE)
4368 				continue;
4369 
4370 			set_page_dirty(page);
4371 			f2fs_put_page(page, 1);
4372 			page = NULL;
4373 		}
4374 	}
4375 	if (page) {
4376 		set_page_dirty(page);
4377 		f2fs_put_page(page, 1);
4378 	}
4379 }
4380 
4381 static void write_normal_summaries(struct f2fs_sb_info *sbi,
4382 					block_t blkaddr, int type)
4383 {
4384 	int i, end;
4385 
4386 	if (IS_DATASEG(type))
4387 		end = type + NR_CURSEG_DATA_TYPE;
4388 	else
4389 		end = type + NR_CURSEG_NODE_TYPE;
4390 
4391 	for (i = type; i < end; i++)
4392 		write_current_sum_page(sbi, i, blkaddr + (i - type));
4393 }
4394 
4395 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4396 {
4397 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
4398 		write_compacted_summaries(sbi, start_blk);
4399 	else
4400 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
4401 }
4402 
4403 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4404 {
4405 	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
4406 }
4407 
4408 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
4409 					unsigned int val, int alloc)
4410 {
4411 	int i;
4412 
4413 	if (type == NAT_JOURNAL) {
4414 		for (i = 0; i < nats_in_cursum(journal); i++) {
4415 			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
4416 				return i;
4417 		}
4418 		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
4419 			return update_nats_in_cursum(journal, 1);
4420 	} else if (type == SIT_JOURNAL) {
4421 		for (i = 0; i < sits_in_cursum(journal); i++)
4422 			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
4423 				return i;
4424 		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
4425 			return update_sits_in_cursum(journal, 1);
4426 	}
4427 	return -1;
4428 }
4429 
4430 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
4431 					unsigned int segno)
4432 {
4433 	return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
4434 }
4435 
4436 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
4437 					unsigned int start)
4438 {
4439 	struct sit_info *sit_i = SIT_I(sbi);
4440 	struct page *page;
4441 	pgoff_t src_off, dst_off;
4442 
4443 	src_off = current_sit_addr(sbi, start);
4444 	dst_off = next_sit_addr(sbi, src_off);
4445 
4446 	page = f2fs_grab_meta_page(sbi, dst_off);
4447 	seg_info_to_sit_page(sbi, page, start);
4448 
4449 	set_page_dirty(page);
4450 	set_to_next_sit(sit_i, start);
4451 
4452 	return page;
4453 }
4454 
4455 static struct sit_entry_set *grab_sit_entry_set(void)
4456 {
4457 	struct sit_entry_set *ses =
4458 			f2fs_kmem_cache_alloc(sit_entry_set_slab,
4459 						GFP_NOFS, true, NULL);
4460 
4461 	ses->entry_cnt = 0;
4462 	INIT_LIST_HEAD(&ses->set_list);
4463 	return ses;
4464 }
4465 
4466 static void release_sit_entry_set(struct sit_entry_set *ses)
4467 {
4468 	list_del(&ses->set_list);
4469 	kmem_cache_free(sit_entry_set_slab, ses);
4470 }
4471 
4472 static void adjust_sit_entry_set(struct sit_entry_set *ses,
4473 						struct list_head *head)
4474 {
4475 	struct sit_entry_set *next = ses;
4476 
4477 	if (list_is_last(&ses->set_list, head))
4478 		return;
4479 
4480 	list_for_each_entry_continue(next, head, set_list)
4481 		if (ses->entry_cnt <= next->entry_cnt) {
4482 			list_move_tail(&ses->set_list, &next->set_list);
4483 			return;
4484 		}
4485 
4486 	list_move_tail(&ses->set_list, head);
4487 }
4488 
4489 static void add_sit_entry(unsigned int segno, struct list_head *head)
4490 {
4491 	struct sit_entry_set *ses;
4492 	unsigned int start_segno = START_SEGNO(segno);
4493 
4494 	list_for_each_entry(ses, head, set_list) {
4495 		if (ses->start_segno == start_segno) {
4496 			ses->entry_cnt++;
4497 			adjust_sit_entry_set(ses, head);
4498 			return;
4499 		}
4500 	}
4501 
4502 	ses = grab_sit_entry_set();
4503 
4504 	ses->start_segno = start_segno;
4505 	ses->entry_cnt++;
4506 	list_add(&ses->set_list, head);
4507 }
4508 
4509 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4510 {
4511 	struct f2fs_sm_info *sm_info = SM_I(sbi);
4512 	struct list_head *set_list = &sm_info->sit_entry_set;
4513 	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4514 	unsigned int segno;
4515 
4516 	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4517 		add_sit_entry(segno, set_list);
4518 }
4519 
4520 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4521 {
4522 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4523 	struct f2fs_journal *journal = curseg->journal;
4524 	int i;
4525 
4526 	down_write(&curseg->journal_rwsem);
4527 	for (i = 0; i < sits_in_cursum(journal); i++) {
4528 		unsigned int segno;
4529 		bool dirtied;
4530 
4531 		segno = le32_to_cpu(segno_in_journal(journal, i));
4532 		dirtied = __mark_sit_entry_dirty(sbi, segno);
4533 
4534 		if (!dirtied)
4535 			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4536 	}
4537 	update_sits_in_cursum(journal, -i);
4538 	up_write(&curseg->journal_rwsem);
4539 }
4540 
4541 /*
4542  * CP calls this function, which flushes SIT entries including sit_journal,
4543  * and moves prefree segs to free segs.
4544  */
4545 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4546 {
4547 	struct sit_info *sit_i = SIT_I(sbi);
4548 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4549 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4550 	struct f2fs_journal *journal = curseg->journal;
4551 	struct sit_entry_set *ses, *tmp;
4552 	struct list_head *head = &SM_I(sbi)->sit_entry_set;
4553 	bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4554 	struct seg_entry *se;
4555 
4556 	down_write(&sit_i->sentry_lock);
4557 
4558 	if (!sit_i->dirty_sentries)
4559 		goto out;
4560 
4561 	/*
4562 	 * add and account sit entries of dirty bitmap in sit entry
4563 	 * set temporarily
4564 	 */
4565 	add_sits_in_set(sbi);
4566 
4567 	/*
4568 	 * if there are no enough space in journal to store dirty sit
4569 	 * entries, remove all entries from journal and add and account
4570 	 * them in sit entry set.
4571 	 */
4572 	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4573 								!to_journal)
4574 		remove_sits_in_journal(sbi);
4575 
4576 	/*
4577 	 * there are two steps to flush sit entries:
4578 	 * #1, flush sit entries to journal in current cold data summary block.
4579 	 * #2, flush sit entries to sit page.
4580 	 */
4581 	list_for_each_entry_safe(ses, tmp, head, set_list) {
4582 		struct page *page = NULL;
4583 		struct f2fs_sit_block *raw_sit = NULL;
4584 		unsigned int start_segno = ses->start_segno;
4585 		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4586 						(unsigned long)MAIN_SEGS(sbi));
4587 		unsigned int segno = start_segno;
4588 
4589 		if (to_journal &&
4590 			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4591 			to_journal = false;
4592 
4593 		if (to_journal) {
4594 			down_write(&curseg->journal_rwsem);
4595 		} else {
4596 			page = get_next_sit_page(sbi, start_segno);
4597 			raw_sit = page_address(page);
4598 		}
4599 
4600 		/* flush dirty sit entries in region of current sit set */
4601 		for_each_set_bit_from(segno, bitmap, end) {
4602 			int offset, sit_offset;
4603 
4604 			se = get_seg_entry(sbi, segno);
4605 #ifdef CONFIG_F2FS_CHECK_FS
4606 			if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4607 						SIT_VBLOCK_MAP_SIZE))
4608 				f2fs_bug_on(sbi, 1);
4609 #endif
4610 
4611 			/* add discard candidates */
4612 			if (!(cpc->reason & CP_DISCARD)) {
4613 				cpc->trim_start = segno;
4614 				add_discard_addrs(sbi, cpc, false);
4615 			}
4616 
4617 			if (to_journal) {
4618 				offset = f2fs_lookup_journal_in_cursum(journal,
4619 							SIT_JOURNAL, segno, 1);
4620 				f2fs_bug_on(sbi, offset < 0);
4621 				segno_in_journal(journal, offset) =
4622 							cpu_to_le32(segno);
4623 				seg_info_to_raw_sit(se,
4624 					&sit_in_journal(journal, offset));
4625 				check_block_count(sbi, segno,
4626 					&sit_in_journal(journal, offset));
4627 			} else {
4628 				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4629 				seg_info_to_raw_sit(se,
4630 						&raw_sit->entries[sit_offset]);
4631 				check_block_count(sbi, segno,
4632 						&raw_sit->entries[sit_offset]);
4633 			}
4634 
4635 			__clear_bit(segno, bitmap);
4636 			sit_i->dirty_sentries--;
4637 			ses->entry_cnt--;
4638 		}
4639 
4640 		if (to_journal)
4641 			up_write(&curseg->journal_rwsem);
4642 		else
4643 			f2fs_put_page(page, 1);
4644 
4645 		f2fs_bug_on(sbi, ses->entry_cnt);
4646 		release_sit_entry_set(ses);
4647 	}
4648 
4649 	f2fs_bug_on(sbi, !list_empty(head));
4650 	f2fs_bug_on(sbi, sit_i->dirty_sentries);
4651 out:
4652 	if (cpc->reason & CP_DISCARD) {
4653 		__u64 trim_start = cpc->trim_start;
4654 
4655 		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4656 			add_discard_addrs(sbi, cpc, false);
4657 
4658 		cpc->trim_start = trim_start;
4659 	}
4660 	up_write(&sit_i->sentry_lock);
4661 
4662 	set_prefree_as_free_segments(sbi);
4663 }
4664 
4665 static int build_sit_info(struct f2fs_sb_info *sbi)
4666 {
4667 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4668 	struct sit_info *sit_i;
4669 	unsigned int sit_segs, start;
4670 	char *src_bitmap, *bitmap;
4671 	unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4672 	unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0;
4673 
4674 	/* allocate memory for SIT information */
4675 	sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4676 	if (!sit_i)
4677 		return -ENOMEM;
4678 
4679 	SM_I(sbi)->sit_info = sit_i;
4680 
4681 	sit_i->sentries =
4682 		f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4683 					      MAIN_SEGS(sbi)),
4684 			      GFP_KERNEL);
4685 	if (!sit_i->sentries)
4686 		return -ENOMEM;
4687 
4688 	main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4689 	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4690 								GFP_KERNEL);
4691 	if (!sit_i->dirty_sentries_bitmap)
4692 		return -ENOMEM;
4693 
4694 #ifdef CONFIG_F2FS_CHECK_FS
4695 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map);
4696 #else
4697 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map);
4698 #endif
4699 	sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4700 	if (!sit_i->bitmap)
4701 		return -ENOMEM;
4702 
4703 	bitmap = sit_i->bitmap;
4704 
4705 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4706 		sit_i->sentries[start].cur_valid_map = bitmap;
4707 		bitmap += SIT_VBLOCK_MAP_SIZE;
4708 
4709 		sit_i->sentries[start].ckpt_valid_map = bitmap;
4710 		bitmap += SIT_VBLOCK_MAP_SIZE;
4711 
4712 #ifdef CONFIG_F2FS_CHECK_FS
4713 		sit_i->sentries[start].cur_valid_map_mir = bitmap;
4714 		bitmap += SIT_VBLOCK_MAP_SIZE;
4715 #endif
4716 
4717 		if (discard_map) {
4718 			sit_i->sentries[start].discard_map = bitmap;
4719 			bitmap += SIT_VBLOCK_MAP_SIZE;
4720 		}
4721 	}
4722 
4723 	sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4724 	if (!sit_i->tmp_map)
4725 		return -ENOMEM;
4726 
4727 	if (__is_large_section(sbi)) {
4728 		sit_i->sec_entries =
4729 			f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4730 						      MAIN_SECS(sbi)),
4731 				      GFP_KERNEL);
4732 		if (!sit_i->sec_entries)
4733 			return -ENOMEM;
4734 	}
4735 
4736 	/* get information related with SIT */
4737 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4738 
4739 	/* setup SIT bitmap from ckeckpoint pack */
4740 	sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4741 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4742 
4743 	sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4744 	if (!sit_i->sit_bitmap)
4745 		return -ENOMEM;
4746 
4747 #ifdef CONFIG_F2FS_CHECK_FS
4748 	sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4749 					sit_bitmap_size, GFP_KERNEL);
4750 	if (!sit_i->sit_bitmap_mir)
4751 		return -ENOMEM;
4752 
4753 	sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4754 					main_bitmap_size, GFP_KERNEL);
4755 	if (!sit_i->invalid_segmap)
4756 		return -ENOMEM;
4757 #endif
4758 
4759 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4760 	sit_i->sit_blocks = SEGS_TO_BLKS(sbi, sit_segs);
4761 	sit_i->written_valid_blocks = 0;
4762 	sit_i->bitmap_size = sit_bitmap_size;
4763 	sit_i->dirty_sentries = 0;
4764 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4765 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4766 	sit_i->mounted_time = ktime_get_boottime_seconds();
4767 	init_rwsem(&sit_i->sentry_lock);
4768 	return 0;
4769 }
4770 
4771 static int build_free_segmap(struct f2fs_sb_info *sbi)
4772 {
4773 	struct free_segmap_info *free_i;
4774 	unsigned int bitmap_size, sec_bitmap_size;
4775 
4776 	/* allocate memory for free segmap information */
4777 	free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4778 	if (!free_i)
4779 		return -ENOMEM;
4780 
4781 	SM_I(sbi)->free_info = free_i;
4782 
4783 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4784 	free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4785 	if (!free_i->free_segmap)
4786 		return -ENOMEM;
4787 
4788 	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4789 	free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4790 	if (!free_i->free_secmap)
4791 		return -ENOMEM;
4792 
4793 	/* set all segments as dirty temporarily */
4794 	memset(free_i->free_segmap, 0xff, bitmap_size);
4795 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4796 
4797 	/* init free segmap information */
4798 	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4799 	free_i->free_segments = 0;
4800 	free_i->free_sections = 0;
4801 	spin_lock_init(&free_i->segmap_lock);
4802 	return 0;
4803 }
4804 
4805 static int build_curseg(struct f2fs_sb_info *sbi)
4806 {
4807 	struct curseg_info *array;
4808 	int i;
4809 
4810 	array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4811 					sizeof(*array)), GFP_KERNEL);
4812 	if (!array)
4813 		return -ENOMEM;
4814 
4815 	SM_I(sbi)->curseg_array = array;
4816 
4817 	for (i = 0; i < NO_CHECK_TYPE; i++) {
4818 		mutex_init(&array[i].curseg_mutex);
4819 		array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4820 		if (!array[i].sum_blk)
4821 			return -ENOMEM;
4822 		init_rwsem(&array[i].journal_rwsem);
4823 		array[i].journal = f2fs_kzalloc(sbi,
4824 				sizeof(struct f2fs_journal), GFP_KERNEL);
4825 		if (!array[i].journal)
4826 			return -ENOMEM;
4827 		array[i].seg_type = log_type_to_seg_type(i);
4828 		reset_curseg_fields(&array[i]);
4829 	}
4830 	return restore_curseg_summaries(sbi);
4831 }
4832 
4833 static int build_sit_entries(struct f2fs_sb_info *sbi)
4834 {
4835 	struct sit_info *sit_i = SIT_I(sbi);
4836 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4837 	struct f2fs_journal *journal = curseg->journal;
4838 	struct seg_entry *se;
4839 	struct f2fs_sit_entry sit;
4840 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
4841 	unsigned int i, start, end;
4842 	unsigned int readed, start_blk = 0;
4843 	int err = 0;
4844 	block_t sit_valid_blocks[2] = {0, 0};
4845 
4846 	do {
4847 		readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4848 							META_SIT, true);
4849 
4850 		start = start_blk * sit_i->sents_per_block;
4851 		end = (start_blk + readed) * sit_i->sents_per_block;
4852 
4853 		for (; start < end && start < MAIN_SEGS(sbi); start++) {
4854 			struct f2fs_sit_block *sit_blk;
4855 			struct page *page;
4856 
4857 			se = &sit_i->sentries[start];
4858 			page = get_current_sit_page(sbi, start);
4859 			if (IS_ERR(page))
4860 				return PTR_ERR(page);
4861 			sit_blk = (struct f2fs_sit_block *)page_address(page);
4862 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4863 			f2fs_put_page(page, 1);
4864 
4865 			err = check_block_count(sbi, start, &sit);
4866 			if (err)
4867 				return err;
4868 			seg_info_from_raw_sit(se, &sit);
4869 
4870 			if (se->type >= NR_PERSISTENT_LOG) {
4871 				f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4872 							se->type, start);
4873 				f2fs_handle_error(sbi,
4874 						ERROR_INCONSISTENT_SUM_TYPE);
4875 				return -EFSCORRUPTED;
4876 			}
4877 
4878 			sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4879 
4880 			if (!f2fs_block_unit_discard(sbi))
4881 				goto init_discard_map_done;
4882 
4883 			/* build discard map only one time */
4884 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4885 				memset(se->discard_map, 0xff,
4886 						SIT_VBLOCK_MAP_SIZE);
4887 				goto init_discard_map_done;
4888 			}
4889 			memcpy(se->discard_map, se->cur_valid_map,
4890 						SIT_VBLOCK_MAP_SIZE);
4891 			sbi->discard_blks += BLKS_PER_SEG(sbi) -
4892 						se->valid_blocks;
4893 init_discard_map_done:
4894 			if (__is_large_section(sbi))
4895 				get_sec_entry(sbi, start)->valid_blocks +=
4896 							se->valid_blocks;
4897 		}
4898 		start_blk += readed;
4899 	} while (start_blk < sit_blk_cnt);
4900 
4901 	down_read(&curseg->journal_rwsem);
4902 	for (i = 0; i < sits_in_cursum(journal); i++) {
4903 		unsigned int old_valid_blocks;
4904 
4905 		start = le32_to_cpu(segno_in_journal(journal, i));
4906 		if (start >= MAIN_SEGS(sbi)) {
4907 			f2fs_err(sbi, "Wrong journal entry on segno %u",
4908 				 start);
4909 			err = -EFSCORRUPTED;
4910 			f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL);
4911 			break;
4912 		}
4913 
4914 		se = &sit_i->sentries[start];
4915 		sit = sit_in_journal(journal, i);
4916 
4917 		old_valid_blocks = se->valid_blocks;
4918 
4919 		sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks;
4920 
4921 		err = check_block_count(sbi, start, &sit);
4922 		if (err)
4923 			break;
4924 		seg_info_from_raw_sit(se, &sit);
4925 
4926 		if (se->type >= NR_PERSISTENT_LOG) {
4927 			f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4928 							se->type, start);
4929 			err = -EFSCORRUPTED;
4930 			f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
4931 			break;
4932 		}
4933 
4934 		sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4935 
4936 		if (f2fs_block_unit_discard(sbi)) {
4937 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4938 				memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4939 			} else {
4940 				memcpy(se->discard_map, se->cur_valid_map,
4941 							SIT_VBLOCK_MAP_SIZE);
4942 				sbi->discard_blks += old_valid_blocks;
4943 				sbi->discard_blks -= se->valid_blocks;
4944 			}
4945 		}
4946 
4947 		if (__is_large_section(sbi)) {
4948 			get_sec_entry(sbi, start)->valid_blocks +=
4949 							se->valid_blocks;
4950 			get_sec_entry(sbi, start)->valid_blocks -=
4951 							old_valid_blocks;
4952 		}
4953 	}
4954 	up_read(&curseg->journal_rwsem);
4955 
4956 	if (err)
4957 		return err;
4958 
4959 	if (sit_valid_blocks[NODE] != valid_node_count(sbi)) {
4960 		f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4961 			 sit_valid_blocks[NODE], valid_node_count(sbi));
4962 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT);
4963 		return -EFSCORRUPTED;
4964 	}
4965 
4966 	if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] >
4967 				valid_user_blocks(sbi)) {
4968 		f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u",
4969 			 sit_valid_blocks[DATA], sit_valid_blocks[NODE],
4970 			 valid_user_blocks(sbi));
4971 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT);
4972 		return -EFSCORRUPTED;
4973 	}
4974 
4975 	return 0;
4976 }
4977 
4978 static void init_free_segmap(struct f2fs_sb_info *sbi)
4979 {
4980 	unsigned int start;
4981 	int type;
4982 	struct seg_entry *sentry;
4983 
4984 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4985 		if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4986 			continue;
4987 		sentry = get_seg_entry(sbi, start);
4988 		if (!sentry->valid_blocks)
4989 			__set_free(sbi, start);
4990 		else
4991 			SIT_I(sbi)->written_valid_blocks +=
4992 						sentry->valid_blocks;
4993 	}
4994 
4995 	/* set use the current segments */
4996 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4997 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4998 
4999 		__set_test_and_inuse(sbi, curseg_t->segno);
5000 	}
5001 }
5002 
5003 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
5004 {
5005 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5006 	struct free_segmap_info *free_i = FREE_I(sbi);
5007 	unsigned int segno = 0, offset = 0, secno;
5008 	block_t valid_blocks, usable_blks_in_seg;
5009 
5010 	while (1) {
5011 		/* find dirty segment based on free segmap */
5012 		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
5013 		if (segno >= MAIN_SEGS(sbi))
5014 			break;
5015 		offset = segno + 1;
5016 		valid_blocks = get_valid_blocks(sbi, segno, false);
5017 		usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
5018 		if (valid_blocks == usable_blks_in_seg || !valid_blocks)
5019 			continue;
5020 		if (valid_blocks > usable_blks_in_seg) {
5021 			f2fs_bug_on(sbi, 1);
5022 			continue;
5023 		}
5024 		mutex_lock(&dirty_i->seglist_lock);
5025 		__locate_dirty_segment(sbi, segno, DIRTY);
5026 		mutex_unlock(&dirty_i->seglist_lock);
5027 	}
5028 
5029 	if (!__is_large_section(sbi))
5030 		return;
5031 
5032 	mutex_lock(&dirty_i->seglist_lock);
5033 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) {
5034 		valid_blocks = get_valid_blocks(sbi, segno, true);
5035 		secno = GET_SEC_FROM_SEG(sbi, segno);
5036 
5037 		if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi))
5038 			continue;
5039 		if (IS_CURSEC(sbi, secno))
5040 			continue;
5041 		set_bit(secno, dirty_i->dirty_secmap);
5042 	}
5043 	mutex_unlock(&dirty_i->seglist_lock);
5044 }
5045 
5046 static int init_victim_secmap(struct f2fs_sb_info *sbi)
5047 {
5048 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5049 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
5050 
5051 	dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
5052 	if (!dirty_i->victim_secmap)
5053 		return -ENOMEM;
5054 
5055 	dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
5056 	if (!dirty_i->pinned_secmap)
5057 		return -ENOMEM;
5058 
5059 	dirty_i->pinned_secmap_cnt = 0;
5060 	dirty_i->enable_pin_section = true;
5061 	return 0;
5062 }
5063 
5064 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
5065 {
5066 	struct dirty_seglist_info *dirty_i;
5067 	unsigned int bitmap_size, i;
5068 
5069 	/* allocate memory for dirty segments list information */
5070 	dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
5071 								GFP_KERNEL);
5072 	if (!dirty_i)
5073 		return -ENOMEM;
5074 
5075 	SM_I(sbi)->dirty_info = dirty_i;
5076 	mutex_init(&dirty_i->seglist_lock);
5077 
5078 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
5079 
5080 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
5081 		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
5082 								GFP_KERNEL);
5083 		if (!dirty_i->dirty_segmap[i])
5084 			return -ENOMEM;
5085 	}
5086 
5087 	if (__is_large_section(sbi)) {
5088 		bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
5089 		dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
5090 						bitmap_size, GFP_KERNEL);
5091 		if (!dirty_i->dirty_secmap)
5092 			return -ENOMEM;
5093 	}
5094 
5095 	init_dirty_segmap(sbi);
5096 	return init_victim_secmap(sbi);
5097 }
5098 
5099 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
5100 {
5101 	int i;
5102 
5103 	/*
5104 	 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
5105 	 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
5106 	 */
5107 	for (i = 0; i < NR_PERSISTENT_LOG; i++) {
5108 		struct curseg_info *curseg = CURSEG_I(sbi, i);
5109 		struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
5110 		unsigned int blkofs = curseg->next_blkoff;
5111 
5112 		if (f2fs_sb_has_readonly(sbi) &&
5113 			i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE)
5114 			continue;
5115 
5116 		sanity_check_seg_type(sbi, curseg->seg_type);
5117 
5118 		if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) {
5119 			f2fs_err(sbi,
5120 				 "Current segment has invalid alloc_type:%d",
5121 				 curseg->alloc_type);
5122 			f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
5123 			return -EFSCORRUPTED;
5124 		}
5125 
5126 		if (f2fs_test_bit(blkofs, se->cur_valid_map))
5127 			goto out;
5128 
5129 		if (curseg->alloc_type == SSR)
5130 			continue;
5131 
5132 		for (blkofs += 1; blkofs < BLKS_PER_SEG(sbi); blkofs++) {
5133 			if (!f2fs_test_bit(blkofs, se->cur_valid_map))
5134 				continue;
5135 out:
5136 			f2fs_err(sbi,
5137 				 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
5138 				 i, curseg->segno, curseg->alloc_type,
5139 				 curseg->next_blkoff, blkofs);
5140 			f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
5141 			return -EFSCORRUPTED;
5142 		}
5143 	}
5144 	return 0;
5145 }
5146 
5147 #ifdef CONFIG_BLK_DEV_ZONED
5148 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
5149 				    struct f2fs_dev_info *fdev,
5150 				    struct blk_zone *zone)
5151 {
5152 	unsigned int zone_segno;
5153 	block_t zone_block, valid_block_cnt;
5154 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
5155 	int ret;
5156 	unsigned int nofs_flags;
5157 
5158 	if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5159 		return 0;
5160 
5161 	zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
5162 	zone_segno = GET_SEGNO(sbi, zone_block);
5163 
5164 	/*
5165 	 * Skip check of zones cursegs point to, since
5166 	 * fix_curseg_write_pointer() checks them.
5167 	 */
5168 	if (zone_segno >= MAIN_SEGS(sbi))
5169 		return 0;
5170 
5171 	/*
5172 	 * Get # of valid block of the zone.
5173 	 */
5174 	valid_block_cnt = get_valid_blocks(sbi, zone_segno, true);
5175 	if (IS_CURSEC(sbi, GET_SEC_FROM_SEG(sbi, zone_segno))) {
5176 		f2fs_notice(sbi, "Open zones: valid block[0x%x,0x%x] cond[%s]",
5177 				zone_segno, valid_block_cnt,
5178 				blk_zone_cond_str(zone->cond));
5179 		return 0;
5180 	}
5181 
5182 	if ((!valid_block_cnt && zone->cond == BLK_ZONE_COND_EMPTY) ||
5183 	    (valid_block_cnt && zone->cond == BLK_ZONE_COND_FULL))
5184 		return 0;
5185 
5186 	if (!valid_block_cnt) {
5187 		f2fs_notice(sbi, "Zone without valid block has non-zero write "
5188 			    "pointer. Reset the write pointer: cond[%s]",
5189 			    blk_zone_cond_str(zone->cond));
5190 		ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
5191 					zone->len >> log_sectors_per_block);
5192 		if (ret)
5193 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
5194 				 fdev->path, ret);
5195 		return ret;
5196 	}
5197 
5198 	/*
5199 	 * If there are valid blocks and the write pointer doesn't match
5200 	 * with them, we need to report the inconsistency and fill
5201 	 * the zone till the end to close the zone. This inconsistency
5202 	 * does not cause write error because the zone will not be
5203 	 * selected for write operation until it get discarded.
5204 	 */
5205 	f2fs_notice(sbi, "Valid blocks are not aligned with write "
5206 		    "pointer: valid block[0x%x,0x%x] cond[%s]",
5207 		    zone_segno, valid_block_cnt, blk_zone_cond_str(zone->cond));
5208 
5209 	nofs_flags = memalloc_nofs_save();
5210 	ret = blkdev_zone_mgmt(fdev->bdev, REQ_OP_ZONE_FINISH,
5211 				zone->start, zone->len);
5212 	memalloc_nofs_restore(nofs_flags);
5213 	if (ret == -EOPNOTSUPP) {
5214 		ret = blkdev_issue_zeroout(fdev->bdev, zone->wp,
5215 					zone->len - (zone->wp - zone->start),
5216 					GFP_NOFS, 0);
5217 		if (ret)
5218 			f2fs_err(sbi, "Fill up zone failed: %s (errno=%d)",
5219 					fdev->path, ret);
5220 	} else if (ret) {
5221 		f2fs_err(sbi, "Finishing zone failed: %s (errno=%d)",
5222 				fdev->path, ret);
5223 	}
5224 
5225 	return ret;
5226 }
5227 
5228 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
5229 						  block_t zone_blkaddr)
5230 {
5231 	int i;
5232 
5233 	for (i = 0; i < sbi->s_ndevs; i++) {
5234 		if (!bdev_is_zoned(FDEV(i).bdev))
5235 			continue;
5236 		if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
5237 				zone_blkaddr <= FDEV(i).end_blk))
5238 			return &FDEV(i);
5239 	}
5240 
5241 	return NULL;
5242 }
5243 
5244 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
5245 			      void *data)
5246 {
5247 	memcpy(data, zone, sizeof(struct blk_zone));
5248 	return 0;
5249 }
5250 
5251 static int do_fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
5252 {
5253 	struct curseg_info *cs = CURSEG_I(sbi, type);
5254 	struct f2fs_dev_info *zbd;
5255 	struct blk_zone zone;
5256 	unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
5257 	block_t cs_zone_block, wp_block;
5258 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
5259 	sector_t zone_sector;
5260 	int err;
5261 
5262 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
5263 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
5264 
5265 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
5266 	if (!zbd)
5267 		return 0;
5268 
5269 	/* report zone for the sector the curseg points to */
5270 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5271 		<< log_sectors_per_block;
5272 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5273 				  report_one_zone_cb, &zone);
5274 	if (err != 1) {
5275 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5276 			 zbd->path, err);
5277 		return err;
5278 	}
5279 
5280 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5281 		return 0;
5282 
5283 	/*
5284 	 * When safely unmounted in the previous mount, we could use current
5285 	 * segments. Otherwise, allocate new sections.
5286 	 */
5287 	if (is_set_ckpt_flags(sbi, CP_UMOUNT_FLAG)) {
5288 		wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
5289 		wp_segno = GET_SEGNO(sbi, wp_block);
5290 		wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
5291 		wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
5292 
5293 		if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
5294 				wp_sector_off == 0)
5295 			return 0;
5296 
5297 		f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
5298 			    "curseg[0x%x,0x%x] wp[0x%x,0x%x]", type, cs->segno,
5299 			    cs->next_blkoff, wp_segno, wp_blkoff);
5300 	}
5301 
5302 	/* Allocate a new section if it's not new. */
5303 	if (cs->next_blkoff ||
5304 	    cs->segno != GET_SEG_FROM_SEC(sbi, GET_ZONE_FROM_SEC(sbi, cs_section))) {
5305 		unsigned int old_segno = cs->segno, old_blkoff = cs->next_blkoff;
5306 
5307 		f2fs_allocate_new_section(sbi, type, true);
5308 		f2fs_notice(sbi, "Assign new section to curseg[%d]: "
5309 				"[0x%x,0x%x] -> [0x%x,0x%x]",
5310 				type, old_segno, old_blkoff,
5311 				cs->segno, cs->next_blkoff);
5312 	}
5313 
5314 	/* check consistency of the zone curseg pointed to */
5315 	if (check_zone_write_pointer(sbi, zbd, &zone))
5316 		return -EIO;
5317 
5318 	/* check newly assigned zone */
5319 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
5320 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
5321 
5322 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
5323 	if (!zbd)
5324 		return 0;
5325 
5326 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5327 		<< log_sectors_per_block;
5328 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5329 				  report_one_zone_cb, &zone);
5330 	if (err != 1) {
5331 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5332 			 zbd->path, err);
5333 		return err;
5334 	}
5335 
5336 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5337 		return 0;
5338 
5339 	if (zone.wp != zone.start) {
5340 		f2fs_notice(sbi,
5341 			    "New zone for curseg[%d] is not yet discarded. "
5342 			    "Reset the zone: curseg[0x%x,0x%x]",
5343 			    type, cs->segno, cs->next_blkoff);
5344 		err = __f2fs_issue_discard_zone(sbi, zbd->bdev,	cs_zone_block,
5345 					zone.len >> log_sectors_per_block);
5346 		if (err) {
5347 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
5348 				 zbd->path, err);
5349 			return err;
5350 		}
5351 	}
5352 
5353 	return 0;
5354 }
5355 
5356 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5357 {
5358 	int i, ret;
5359 
5360 	for (i = 0; i < NR_PERSISTENT_LOG; i++) {
5361 		ret = do_fix_curseg_write_pointer(sbi, i);
5362 		if (ret)
5363 			return ret;
5364 	}
5365 
5366 	return 0;
5367 }
5368 
5369 struct check_zone_write_pointer_args {
5370 	struct f2fs_sb_info *sbi;
5371 	struct f2fs_dev_info *fdev;
5372 };
5373 
5374 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
5375 				      void *data)
5376 {
5377 	struct check_zone_write_pointer_args *args;
5378 
5379 	args = (struct check_zone_write_pointer_args *)data;
5380 
5381 	return check_zone_write_pointer(args->sbi, args->fdev, zone);
5382 }
5383 
5384 static int check_write_pointer(struct f2fs_sb_info *sbi)
5385 {
5386 	int i, ret;
5387 	struct check_zone_write_pointer_args args;
5388 
5389 	for (i = 0; i < sbi->s_ndevs; i++) {
5390 		if (!bdev_is_zoned(FDEV(i).bdev))
5391 			continue;
5392 
5393 		args.sbi = sbi;
5394 		args.fdev = &FDEV(i);
5395 		ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
5396 					  check_zone_write_pointer_cb, &args);
5397 		if (ret < 0)
5398 			return ret;
5399 	}
5400 
5401 	return 0;
5402 }
5403 
5404 int f2fs_check_and_fix_write_pointer(struct f2fs_sb_info *sbi)
5405 {
5406 	int ret;
5407 
5408 	if (!f2fs_sb_has_blkzoned(sbi) || f2fs_readonly(sbi->sb))
5409 		return 0;
5410 
5411 	f2fs_notice(sbi, "Checking entire write pointers");
5412 	ret = fix_curseg_write_pointer(sbi);
5413 	if (!ret)
5414 		ret = check_write_pointer(sbi);
5415 	return ret;
5416 }
5417 
5418 /*
5419  * Return the number of usable blocks in a segment. The number of blocks
5420  * returned is always equal to the number of blocks in a segment for
5421  * segments fully contained within a sequential zone capacity or a
5422  * conventional zone. For segments partially contained in a sequential
5423  * zone capacity, the number of usable blocks up to the zone capacity
5424  * is returned. 0 is returned in all other cases.
5425  */
5426 static inline unsigned int f2fs_usable_zone_blks_in_seg(
5427 			struct f2fs_sb_info *sbi, unsigned int segno)
5428 {
5429 	block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
5430 	unsigned int secno;
5431 
5432 	if (!sbi->unusable_blocks_per_sec)
5433 		return BLKS_PER_SEG(sbi);
5434 
5435 	secno = GET_SEC_FROM_SEG(sbi, segno);
5436 	seg_start = START_BLOCK(sbi, segno);
5437 	sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5438 	sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi);
5439 
5440 	/*
5441 	 * If segment starts before zone capacity and spans beyond
5442 	 * zone capacity, then usable blocks are from seg start to
5443 	 * zone capacity. If the segment starts after the zone capacity,
5444 	 * then there are no usable blocks.
5445 	 */
5446 	if (seg_start >= sec_cap_blkaddr)
5447 		return 0;
5448 	if (seg_start + BLKS_PER_SEG(sbi) > sec_cap_blkaddr)
5449 		return sec_cap_blkaddr - seg_start;
5450 
5451 	return BLKS_PER_SEG(sbi);
5452 }
5453 #else
5454 int f2fs_check_and_fix_write_pointer(struct f2fs_sb_info *sbi)
5455 {
5456 	return 0;
5457 }
5458 
5459 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5460 							unsigned int segno)
5461 {
5462 	return 0;
5463 }
5464 
5465 #endif
5466 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5467 					unsigned int segno)
5468 {
5469 	if (f2fs_sb_has_blkzoned(sbi))
5470 		return f2fs_usable_zone_blks_in_seg(sbi, segno);
5471 
5472 	return BLKS_PER_SEG(sbi);
5473 }
5474 
5475 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi)
5476 {
5477 	if (f2fs_sb_has_blkzoned(sbi))
5478 		return CAP_SEGS_PER_SEC(sbi);
5479 
5480 	return SEGS_PER_SEC(sbi);
5481 }
5482 
5483 unsigned long long f2fs_get_section_mtime(struct f2fs_sb_info *sbi,
5484 	unsigned int segno)
5485 {
5486 	unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi);
5487 	unsigned int secno = 0, start = 0;
5488 	unsigned int total_valid_blocks = 0;
5489 	unsigned long long mtime = 0;
5490 	unsigned int i = 0;
5491 
5492 	secno = GET_SEC_FROM_SEG(sbi, segno);
5493 	start = GET_SEG_FROM_SEC(sbi, secno);
5494 
5495 	if (!__is_large_section(sbi))
5496 		return get_seg_entry(sbi, start + i)->mtime;
5497 
5498 	for (i = 0; i < usable_segs_per_sec; i++) {
5499 		/* for large section, only check the mtime of valid segments */
5500 		struct seg_entry *se = get_seg_entry(sbi, start+i);
5501 
5502 		mtime += se->mtime * se->valid_blocks;
5503 		total_valid_blocks += se->valid_blocks;
5504 	}
5505 
5506 	if (total_valid_blocks == 0)
5507 		return INVALID_MTIME;
5508 
5509 	return div_u64(mtime, total_valid_blocks);
5510 }
5511 
5512 /*
5513  * Update min, max modified time for cost-benefit GC algorithm
5514  */
5515 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5516 {
5517 	struct sit_info *sit_i = SIT_I(sbi);
5518 	unsigned int segno;
5519 
5520 	down_write(&sit_i->sentry_lock);
5521 
5522 	sit_i->min_mtime = ULLONG_MAX;
5523 
5524 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) {
5525 		unsigned long long mtime = 0;
5526 
5527 		mtime = f2fs_get_section_mtime(sbi, segno);
5528 
5529 		if (sit_i->min_mtime > mtime)
5530 			sit_i->min_mtime = mtime;
5531 	}
5532 	sit_i->max_mtime = get_mtime(sbi, false);
5533 	sit_i->dirty_max_mtime = 0;
5534 	up_write(&sit_i->sentry_lock);
5535 }
5536 
5537 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5538 {
5539 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5540 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5541 	struct f2fs_sm_info *sm_info;
5542 	int err;
5543 
5544 	sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5545 	if (!sm_info)
5546 		return -ENOMEM;
5547 
5548 	/* init sm info */
5549 	sbi->sm_info = sm_info;
5550 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5551 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5552 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5553 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5554 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5555 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5556 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5557 	sm_info->rec_prefree_segments = sm_info->main_segments *
5558 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5559 	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5560 		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5561 
5562 	if (!f2fs_lfs_mode(sbi))
5563 		sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC);
5564 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5565 	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5566 	sm_info->min_seq_blocks = BLKS_PER_SEG(sbi);
5567 	sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5568 	sm_info->min_ssr_sections = reserved_sections(sbi);
5569 
5570 	INIT_LIST_HEAD(&sm_info->sit_entry_set);
5571 
5572 	init_f2fs_rwsem(&sm_info->curseg_lock);
5573 
5574 	err = f2fs_create_flush_cmd_control(sbi);
5575 	if (err)
5576 		return err;
5577 
5578 	err = create_discard_cmd_control(sbi);
5579 	if (err)
5580 		return err;
5581 
5582 	err = build_sit_info(sbi);
5583 	if (err)
5584 		return err;
5585 	err = build_free_segmap(sbi);
5586 	if (err)
5587 		return err;
5588 	err = build_curseg(sbi);
5589 	if (err)
5590 		return err;
5591 
5592 	/* reinit free segmap based on SIT */
5593 	err = build_sit_entries(sbi);
5594 	if (err)
5595 		return err;
5596 
5597 	init_free_segmap(sbi);
5598 	err = build_dirty_segmap(sbi);
5599 	if (err)
5600 		return err;
5601 
5602 	err = sanity_check_curseg(sbi);
5603 	if (err)
5604 		return err;
5605 
5606 	init_min_max_mtime(sbi);
5607 	return 0;
5608 }
5609 
5610 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5611 		enum dirty_type dirty_type)
5612 {
5613 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5614 
5615 	mutex_lock(&dirty_i->seglist_lock);
5616 	kvfree(dirty_i->dirty_segmap[dirty_type]);
5617 	dirty_i->nr_dirty[dirty_type] = 0;
5618 	mutex_unlock(&dirty_i->seglist_lock);
5619 }
5620 
5621 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5622 {
5623 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5624 
5625 	kvfree(dirty_i->pinned_secmap);
5626 	kvfree(dirty_i->victim_secmap);
5627 }
5628 
5629 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5630 {
5631 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5632 	int i;
5633 
5634 	if (!dirty_i)
5635 		return;
5636 
5637 	/* discard pre-free/dirty segments list */
5638 	for (i = 0; i < NR_DIRTY_TYPE; i++)
5639 		discard_dirty_segmap(sbi, i);
5640 
5641 	if (__is_large_section(sbi)) {
5642 		mutex_lock(&dirty_i->seglist_lock);
5643 		kvfree(dirty_i->dirty_secmap);
5644 		mutex_unlock(&dirty_i->seglist_lock);
5645 	}
5646 
5647 	destroy_victim_secmap(sbi);
5648 	SM_I(sbi)->dirty_info = NULL;
5649 	kfree(dirty_i);
5650 }
5651 
5652 static void destroy_curseg(struct f2fs_sb_info *sbi)
5653 {
5654 	struct curseg_info *array = SM_I(sbi)->curseg_array;
5655 	int i;
5656 
5657 	if (!array)
5658 		return;
5659 	SM_I(sbi)->curseg_array = NULL;
5660 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
5661 		kfree(array[i].sum_blk);
5662 		kfree(array[i].journal);
5663 	}
5664 	kfree(array);
5665 }
5666 
5667 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5668 {
5669 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5670 
5671 	if (!free_i)
5672 		return;
5673 	SM_I(sbi)->free_info = NULL;
5674 	kvfree(free_i->free_segmap);
5675 	kvfree(free_i->free_secmap);
5676 	kfree(free_i);
5677 }
5678 
5679 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5680 {
5681 	struct sit_info *sit_i = SIT_I(sbi);
5682 
5683 	if (!sit_i)
5684 		return;
5685 
5686 	if (sit_i->sentries)
5687 		kvfree(sit_i->bitmap);
5688 	kfree(sit_i->tmp_map);
5689 
5690 	kvfree(sit_i->sentries);
5691 	kvfree(sit_i->sec_entries);
5692 	kvfree(sit_i->dirty_sentries_bitmap);
5693 
5694 	SM_I(sbi)->sit_info = NULL;
5695 	kvfree(sit_i->sit_bitmap);
5696 #ifdef CONFIG_F2FS_CHECK_FS
5697 	kvfree(sit_i->sit_bitmap_mir);
5698 	kvfree(sit_i->invalid_segmap);
5699 #endif
5700 	kfree(sit_i);
5701 }
5702 
5703 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5704 {
5705 	struct f2fs_sm_info *sm_info = SM_I(sbi);
5706 
5707 	if (!sm_info)
5708 		return;
5709 	f2fs_destroy_flush_cmd_control(sbi, true);
5710 	destroy_discard_cmd_control(sbi);
5711 	destroy_dirty_segmap(sbi);
5712 	destroy_curseg(sbi);
5713 	destroy_free_segmap(sbi);
5714 	destroy_sit_info(sbi);
5715 	sbi->sm_info = NULL;
5716 	kfree(sm_info);
5717 }
5718 
5719 int __init f2fs_create_segment_manager_caches(void)
5720 {
5721 	discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5722 			sizeof(struct discard_entry));
5723 	if (!discard_entry_slab)
5724 		goto fail;
5725 
5726 	discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5727 			sizeof(struct discard_cmd));
5728 	if (!discard_cmd_slab)
5729 		goto destroy_discard_entry;
5730 
5731 	sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5732 			sizeof(struct sit_entry_set));
5733 	if (!sit_entry_set_slab)
5734 		goto destroy_discard_cmd;
5735 
5736 	revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry",
5737 			sizeof(struct revoke_entry));
5738 	if (!revoke_entry_slab)
5739 		goto destroy_sit_entry_set;
5740 	return 0;
5741 
5742 destroy_sit_entry_set:
5743 	kmem_cache_destroy(sit_entry_set_slab);
5744 destroy_discard_cmd:
5745 	kmem_cache_destroy(discard_cmd_slab);
5746 destroy_discard_entry:
5747 	kmem_cache_destroy(discard_entry_slab);
5748 fail:
5749 	return -ENOMEM;
5750 }
5751 
5752 void f2fs_destroy_segment_manager_caches(void)
5753 {
5754 	kmem_cache_destroy(sit_entry_set_slab);
5755 	kmem_cache_destroy(discard_cmd_slab);
5756 	kmem_cache_destroy(discard_entry_slab);
5757 	kmem_cache_destroy(revoke_entry_slab);
5758 }
5759