xref: /openbmc/linux/fs/f2fs/segment.c (revision 3d6af30f)
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 	release_atomic_write_cnt(inode);
196 	clear_inode_flag(inode, FI_ATOMIC_COMMITTED);
197 	clear_inode_flag(inode, FI_ATOMIC_REPLACE);
198 	clear_inode_flag(inode, FI_ATOMIC_FILE);
199 	stat_dec_atomic_inode(inode);
200 
201 	F2FS_I(inode)->atomic_write_task = NULL;
202 
203 	if (clean) {
204 		truncate_inode_pages_final(inode->i_mapping);
205 		f2fs_i_size_write(inode, fi->original_i_size);
206 		fi->original_i_size = 0;
207 	}
208 	/* avoid stale dirty inode during eviction */
209 	sync_inode_metadata(inode, 0);
210 }
211 
212 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index,
213 			block_t new_addr, block_t *old_addr, bool recover)
214 {
215 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
216 	struct dnode_of_data dn;
217 	struct node_info ni;
218 	int err;
219 
220 retry:
221 	set_new_dnode(&dn, inode, NULL, NULL, 0);
222 	err = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE);
223 	if (err) {
224 		if (err == -ENOMEM) {
225 			f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
226 			goto retry;
227 		}
228 		return err;
229 	}
230 
231 	err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
232 	if (err) {
233 		f2fs_put_dnode(&dn);
234 		return err;
235 	}
236 
237 	if (recover) {
238 		/* dn.data_blkaddr is always valid */
239 		if (!__is_valid_data_blkaddr(new_addr)) {
240 			if (new_addr == NULL_ADDR)
241 				dec_valid_block_count(sbi, inode, 1);
242 			f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
243 			f2fs_update_data_blkaddr(&dn, new_addr);
244 		} else {
245 			f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
246 				new_addr, ni.version, true, true);
247 		}
248 	} else {
249 		blkcnt_t count = 1;
250 
251 		err = inc_valid_block_count(sbi, inode, &count, true);
252 		if (err) {
253 			f2fs_put_dnode(&dn);
254 			return err;
255 		}
256 
257 		*old_addr = dn.data_blkaddr;
258 		f2fs_truncate_data_blocks_range(&dn, 1);
259 		dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count);
260 
261 		f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr,
262 					ni.version, true, false);
263 	}
264 
265 	f2fs_put_dnode(&dn);
266 
267 	trace_f2fs_replace_atomic_write_block(inode, F2FS_I(inode)->cow_inode,
268 			index, old_addr ? *old_addr : 0, new_addr, recover);
269 	return 0;
270 }
271 
272 static void __complete_revoke_list(struct inode *inode, struct list_head *head,
273 					bool revoke)
274 {
275 	struct revoke_entry *cur, *tmp;
276 	pgoff_t start_index = 0;
277 	bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE);
278 
279 	list_for_each_entry_safe(cur, tmp, head, list) {
280 		if (revoke) {
281 			__replace_atomic_write_block(inode, cur->index,
282 						cur->old_addr, NULL, true);
283 		} else if (truncate) {
284 			f2fs_truncate_hole(inode, start_index, cur->index);
285 			start_index = cur->index + 1;
286 		}
287 
288 		list_del(&cur->list);
289 		kmem_cache_free(revoke_entry_slab, cur);
290 	}
291 
292 	if (!revoke && truncate)
293 		f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false);
294 }
295 
296 static int __f2fs_commit_atomic_write(struct inode *inode)
297 {
298 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
299 	struct f2fs_inode_info *fi = F2FS_I(inode);
300 	struct inode *cow_inode = fi->cow_inode;
301 	struct revoke_entry *new;
302 	struct list_head revoke_list;
303 	block_t blkaddr;
304 	struct dnode_of_data dn;
305 	pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
306 	pgoff_t off = 0, blen, index;
307 	int ret = 0, i;
308 
309 	INIT_LIST_HEAD(&revoke_list);
310 
311 	while (len) {
312 		blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len);
313 
314 		set_new_dnode(&dn, cow_inode, NULL, NULL, 0);
315 		ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA);
316 		if (ret && ret != -ENOENT) {
317 			goto out;
318 		} else if (ret == -ENOENT) {
319 			ret = 0;
320 			if (dn.max_level == 0)
321 				goto out;
322 			goto next;
323 		}
324 
325 		blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode),
326 				len);
327 		index = off;
328 		for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) {
329 			blkaddr = f2fs_data_blkaddr(&dn);
330 
331 			if (!__is_valid_data_blkaddr(blkaddr)) {
332 				continue;
333 			} else if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
334 					DATA_GENERIC_ENHANCE)) {
335 				f2fs_put_dnode(&dn);
336 				ret = -EFSCORRUPTED;
337 				f2fs_handle_error(sbi,
338 						ERROR_INVALID_BLKADDR);
339 				goto out;
340 			}
341 
342 			new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS,
343 							true, NULL);
344 
345 			ret = __replace_atomic_write_block(inode, index, blkaddr,
346 							&new->old_addr, false);
347 			if (ret) {
348 				f2fs_put_dnode(&dn);
349 				kmem_cache_free(revoke_entry_slab, new);
350 				goto out;
351 			}
352 
353 			f2fs_update_data_blkaddr(&dn, NULL_ADDR);
354 			new->index = index;
355 			list_add_tail(&new->list, &revoke_list);
356 		}
357 		f2fs_put_dnode(&dn);
358 next:
359 		off += blen;
360 		len -= blen;
361 	}
362 
363 out:
364 	if (ret) {
365 		sbi->revoked_atomic_block += fi->atomic_write_cnt;
366 	} else {
367 		sbi->committed_atomic_block += fi->atomic_write_cnt;
368 		set_inode_flag(inode, FI_ATOMIC_COMMITTED);
369 	}
370 
371 	__complete_revoke_list(inode, &revoke_list, ret ? true : false);
372 
373 	return ret;
374 }
375 
376 int f2fs_commit_atomic_write(struct inode *inode)
377 {
378 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
379 	struct f2fs_inode_info *fi = F2FS_I(inode);
380 	int err;
381 
382 	err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
383 	if (err)
384 		return err;
385 
386 	f2fs_down_write(&fi->i_gc_rwsem[WRITE]);
387 	f2fs_lock_op(sbi);
388 
389 	err = __f2fs_commit_atomic_write(inode);
390 
391 	f2fs_unlock_op(sbi);
392 	f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
393 
394 	return err;
395 }
396 
397 /*
398  * This function balances dirty node and dentry pages.
399  * In addition, it controls garbage collection.
400  */
401 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
402 {
403 	if (time_to_inject(sbi, FAULT_CHECKPOINT))
404 		f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT);
405 
406 	/* balance_fs_bg is able to be pending */
407 	if (need && excess_cached_nats(sbi))
408 		f2fs_balance_fs_bg(sbi, false);
409 
410 	if (!f2fs_is_checkpoint_ready(sbi))
411 		return;
412 
413 	/*
414 	 * We should do GC or end up with checkpoint, if there are so many dirty
415 	 * dir/node pages without enough free segments.
416 	 */
417 	if (has_enough_free_secs(sbi, 0, 0))
418 		return;
419 
420 	if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
421 				sbi->gc_thread->f2fs_gc_task) {
422 		DEFINE_WAIT(wait);
423 
424 		prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
425 					TASK_UNINTERRUPTIBLE);
426 		wake_up(&sbi->gc_thread->gc_wait_queue_head);
427 		io_schedule();
428 		finish_wait(&sbi->gc_thread->fggc_wq, &wait);
429 	} else {
430 		struct f2fs_gc_control gc_control = {
431 			.victim_segno = NULL_SEGNO,
432 			.init_gc_type = BG_GC,
433 			.no_bg_gc = true,
434 			.should_migrate_blocks = false,
435 			.err_gc_skipped = false,
436 			.nr_free_secs = 1 };
437 		f2fs_down_write(&sbi->gc_lock);
438 		stat_inc_gc_call_count(sbi, FOREGROUND);
439 		f2fs_gc(sbi, &gc_control);
440 	}
441 }
442 
443 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi)
444 {
445 	int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2;
446 	unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS);
447 	unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA);
448 	unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES);
449 	unsigned int meta = get_pages(sbi, F2FS_DIRTY_META);
450 	unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA);
451 	unsigned int threshold = sbi->blocks_per_seg * factor *
452 					DEFAULT_DIRTY_THRESHOLD;
453 	unsigned int global_threshold = threshold * 3 / 2;
454 
455 	if (dents >= threshold || qdata >= threshold ||
456 		nodes >= threshold || meta >= threshold ||
457 		imeta >= threshold)
458 		return true;
459 	return dents + qdata + nodes + meta + imeta >  global_threshold;
460 }
461 
462 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
463 {
464 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
465 		return;
466 
467 	/* try to shrink extent cache when there is no enough memory */
468 	if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE))
469 		f2fs_shrink_read_extent_tree(sbi,
470 				READ_EXTENT_CACHE_SHRINK_NUMBER);
471 
472 	/* try to shrink age extent cache when there is no enough memory */
473 	if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE))
474 		f2fs_shrink_age_extent_tree(sbi,
475 				AGE_EXTENT_CACHE_SHRINK_NUMBER);
476 
477 	/* check the # of cached NAT entries */
478 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
479 		f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
480 
481 	if (!f2fs_available_free_memory(sbi, FREE_NIDS))
482 		f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
483 	else
484 		f2fs_build_free_nids(sbi, false, false);
485 
486 	if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) ||
487 		excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi))
488 		goto do_sync;
489 
490 	/* there is background inflight IO or foreground operation recently */
491 	if (is_inflight_io(sbi, REQ_TIME) ||
492 		(!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem)))
493 		return;
494 
495 	/* exceed periodical checkpoint timeout threshold */
496 	if (f2fs_time_over(sbi, CP_TIME))
497 		goto do_sync;
498 
499 	/* checkpoint is the only way to shrink partial cached entries */
500 	if (f2fs_available_free_memory(sbi, NAT_ENTRIES) &&
501 		f2fs_available_free_memory(sbi, INO_ENTRIES))
502 		return;
503 
504 do_sync:
505 	if (test_opt(sbi, DATA_FLUSH) && from_bg) {
506 		struct blk_plug plug;
507 
508 		mutex_lock(&sbi->flush_lock);
509 
510 		blk_start_plug(&plug);
511 		f2fs_sync_dirty_inodes(sbi, FILE_INODE, false);
512 		blk_finish_plug(&plug);
513 
514 		mutex_unlock(&sbi->flush_lock);
515 	}
516 	stat_inc_cp_call_count(sbi, BACKGROUND);
517 	f2fs_sync_fs(sbi->sb, 1);
518 }
519 
520 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
521 				struct block_device *bdev)
522 {
523 	int ret = blkdev_issue_flush(bdev);
524 
525 	trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
526 				test_opt(sbi, FLUSH_MERGE), ret);
527 	if (!ret)
528 		f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0);
529 	return ret;
530 }
531 
532 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
533 {
534 	int ret = 0;
535 	int i;
536 
537 	if (!f2fs_is_multi_device(sbi))
538 		return __submit_flush_wait(sbi, sbi->sb->s_bdev);
539 
540 	for (i = 0; i < sbi->s_ndevs; i++) {
541 		if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
542 			continue;
543 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
544 		if (ret)
545 			break;
546 	}
547 	return ret;
548 }
549 
550 static int issue_flush_thread(void *data)
551 {
552 	struct f2fs_sb_info *sbi = data;
553 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
554 	wait_queue_head_t *q = &fcc->flush_wait_queue;
555 repeat:
556 	if (kthread_should_stop())
557 		return 0;
558 
559 	if (!llist_empty(&fcc->issue_list)) {
560 		struct flush_cmd *cmd, *next;
561 		int ret;
562 
563 		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
564 		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
565 
566 		cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
567 
568 		ret = submit_flush_wait(sbi, cmd->ino);
569 		atomic_inc(&fcc->issued_flush);
570 
571 		llist_for_each_entry_safe(cmd, next,
572 					  fcc->dispatch_list, llnode) {
573 			cmd->ret = ret;
574 			complete(&cmd->wait);
575 		}
576 		fcc->dispatch_list = NULL;
577 	}
578 
579 	wait_event_interruptible(*q,
580 		kthread_should_stop() || !llist_empty(&fcc->issue_list));
581 	goto repeat;
582 }
583 
584 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
585 {
586 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
587 	struct flush_cmd cmd;
588 	int ret;
589 
590 	if (test_opt(sbi, NOBARRIER))
591 		return 0;
592 
593 	if (!test_opt(sbi, FLUSH_MERGE)) {
594 		atomic_inc(&fcc->queued_flush);
595 		ret = submit_flush_wait(sbi, ino);
596 		atomic_dec(&fcc->queued_flush);
597 		atomic_inc(&fcc->issued_flush);
598 		return ret;
599 	}
600 
601 	if (atomic_inc_return(&fcc->queued_flush) == 1 ||
602 	    f2fs_is_multi_device(sbi)) {
603 		ret = submit_flush_wait(sbi, ino);
604 		atomic_dec(&fcc->queued_flush);
605 
606 		atomic_inc(&fcc->issued_flush);
607 		return ret;
608 	}
609 
610 	cmd.ino = ino;
611 	init_completion(&cmd.wait);
612 
613 	llist_add(&cmd.llnode, &fcc->issue_list);
614 
615 	/*
616 	 * update issue_list before we wake up issue_flush thread, this
617 	 * smp_mb() pairs with another barrier in ___wait_event(), see
618 	 * more details in comments of waitqueue_active().
619 	 */
620 	smp_mb();
621 
622 	if (waitqueue_active(&fcc->flush_wait_queue))
623 		wake_up(&fcc->flush_wait_queue);
624 
625 	if (fcc->f2fs_issue_flush) {
626 		wait_for_completion(&cmd.wait);
627 		atomic_dec(&fcc->queued_flush);
628 	} else {
629 		struct llist_node *list;
630 
631 		list = llist_del_all(&fcc->issue_list);
632 		if (!list) {
633 			wait_for_completion(&cmd.wait);
634 			atomic_dec(&fcc->queued_flush);
635 		} else {
636 			struct flush_cmd *tmp, *next;
637 
638 			ret = submit_flush_wait(sbi, ino);
639 
640 			llist_for_each_entry_safe(tmp, next, list, llnode) {
641 				if (tmp == &cmd) {
642 					cmd.ret = ret;
643 					atomic_dec(&fcc->queued_flush);
644 					continue;
645 				}
646 				tmp->ret = ret;
647 				complete(&tmp->wait);
648 			}
649 		}
650 	}
651 
652 	return cmd.ret;
653 }
654 
655 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
656 {
657 	dev_t dev = sbi->sb->s_bdev->bd_dev;
658 	struct flush_cmd_control *fcc;
659 
660 	if (SM_I(sbi)->fcc_info) {
661 		fcc = SM_I(sbi)->fcc_info;
662 		if (fcc->f2fs_issue_flush)
663 			return 0;
664 		goto init_thread;
665 	}
666 
667 	fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
668 	if (!fcc)
669 		return -ENOMEM;
670 	atomic_set(&fcc->issued_flush, 0);
671 	atomic_set(&fcc->queued_flush, 0);
672 	init_waitqueue_head(&fcc->flush_wait_queue);
673 	init_llist_head(&fcc->issue_list);
674 	SM_I(sbi)->fcc_info = fcc;
675 	if (!test_opt(sbi, FLUSH_MERGE))
676 		return 0;
677 
678 init_thread:
679 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
680 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
681 	if (IS_ERR(fcc->f2fs_issue_flush)) {
682 		int err = PTR_ERR(fcc->f2fs_issue_flush);
683 
684 		fcc->f2fs_issue_flush = NULL;
685 		return err;
686 	}
687 
688 	return 0;
689 }
690 
691 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
692 {
693 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
694 
695 	if (fcc && fcc->f2fs_issue_flush) {
696 		struct task_struct *flush_thread = fcc->f2fs_issue_flush;
697 
698 		fcc->f2fs_issue_flush = NULL;
699 		kthread_stop(flush_thread);
700 	}
701 	if (free) {
702 		kfree(fcc);
703 		SM_I(sbi)->fcc_info = NULL;
704 	}
705 }
706 
707 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
708 {
709 	int ret = 0, i;
710 
711 	if (!f2fs_is_multi_device(sbi))
712 		return 0;
713 
714 	if (test_opt(sbi, NOBARRIER))
715 		return 0;
716 
717 	for (i = 1; i < sbi->s_ndevs; i++) {
718 		int count = DEFAULT_RETRY_IO_COUNT;
719 
720 		if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
721 			continue;
722 
723 		do {
724 			ret = __submit_flush_wait(sbi, FDEV(i).bdev);
725 			if (ret)
726 				f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
727 		} while (ret && --count);
728 
729 		if (ret) {
730 			f2fs_stop_checkpoint(sbi, false,
731 					STOP_CP_REASON_FLUSH_FAIL);
732 			break;
733 		}
734 
735 		spin_lock(&sbi->dev_lock);
736 		f2fs_clear_bit(i, (char *)&sbi->dirty_device);
737 		spin_unlock(&sbi->dev_lock);
738 	}
739 
740 	return ret;
741 }
742 
743 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
744 		enum dirty_type dirty_type)
745 {
746 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
747 
748 	/* need not be added */
749 	if (IS_CURSEG(sbi, segno))
750 		return;
751 
752 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
753 		dirty_i->nr_dirty[dirty_type]++;
754 
755 	if (dirty_type == DIRTY) {
756 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
757 		enum dirty_type t = sentry->type;
758 
759 		if (unlikely(t >= DIRTY)) {
760 			f2fs_bug_on(sbi, 1);
761 			return;
762 		}
763 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
764 			dirty_i->nr_dirty[t]++;
765 
766 		if (__is_large_section(sbi)) {
767 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
768 			block_t valid_blocks =
769 				get_valid_blocks(sbi, segno, true);
770 
771 			f2fs_bug_on(sbi, unlikely(!valid_blocks ||
772 					valid_blocks == CAP_BLKS_PER_SEC(sbi)));
773 
774 			if (!IS_CURSEC(sbi, secno))
775 				set_bit(secno, dirty_i->dirty_secmap);
776 		}
777 	}
778 }
779 
780 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
781 		enum dirty_type dirty_type)
782 {
783 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
784 	block_t valid_blocks;
785 
786 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
787 		dirty_i->nr_dirty[dirty_type]--;
788 
789 	if (dirty_type == DIRTY) {
790 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
791 		enum dirty_type t = sentry->type;
792 
793 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
794 			dirty_i->nr_dirty[t]--;
795 
796 		valid_blocks = get_valid_blocks(sbi, segno, true);
797 		if (valid_blocks == 0) {
798 			clear_bit(GET_SEC_FROM_SEG(sbi, segno),
799 						dirty_i->victim_secmap);
800 #ifdef CONFIG_F2FS_CHECK_FS
801 			clear_bit(segno, SIT_I(sbi)->invalid_segmap);
802 #endif
803 		}
804 		if (__is_large_section(sbi)) {
805 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
806 
807 			if (!valid_blocks ||
808 					valid_blocks == CAP_BLKS_PER_SEC(sbi)) {
809 				clear_bit(secno, dirty_i->dirty_secmap);
810 				return;
811 			}
812 
813 			if (!IS_CURSEC(sbi, secno))
814 				set_bit(secno, dirty_i->dirty_secmap);
815 		}
816 	}
817 }
818 
819 /*
820  * Should not occur error such as -ENOMEM.
821  * Adding dirty entry into seglist is not critical operation.
822  * If a given segment is one of current working segments, it won't be added.
823  */
824 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
825 {
826 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
827 	unsigned short valid_blocks, ckpt_valid_blocks;
828 	unsigned int usable_blocks;
829 
830 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
831 		return;
832 
833 	usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
834 	mutex_lock(&dirty_i->seglist_lock);
835 
836 	valid_blocks = get_valid_blocks(sbi, segno, false);
837 	ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
838 
839 	if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
840 		ckpt_valid_blocks == usable_blocks)) {
841 		__locate_dirty_segment(sbi, segno, PRE);
842 		__remove_dirty_segment(sbi, segno, DIRTY);
843 	} else if (valid_blocks < usable_blocks) {
844 		__locate_dirty_segment(sbi, segno, DIRTY);
845 	} else {
846 		/* Recovery routine with SSR needs this */
847 		__remove_dirty_segment(sbi, segno, DIRTY);
848 	}
849 
850 	mutex_unlock(&dirty_i->seglist_lock);
851 }
852 
853 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
854 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
855 {
856 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
857 	unsigned int segno;
858 
859 	mutex_lock(&dirty_i->seglist_lock);
860 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
861 		if (get_valid_blocks(sbi, segno, false))
862 			continue;
863 		if (IS_CURSEG(sbi, segno))
864 			continue;
865 		__locate_dirty_segment(sbi, segno, PRE);
866 		__remove_dirty_segment(sbi, segno, DIRTY);
867 	}
868 	mutex_unlock(&dirty_i->seglist_lock);
869 }
870 
871 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
872 {
873 	int ovp_hole_segs =
874 		(overprovision_segments(sbi) - reserved_segments(sbi));
875 	block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
876 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
877 	block_t holes[2] = {0, 0};	/* DATA and NODE */
878 	block_t unusable;
879 	struct seg_entry *se;
880 	unsigned int segno;
881 
882 	mutex_lock(&dirty_i->seglist_lock);
883 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
884 		se = get_seg_entry(sbi, segno);
885 		if (IS_NODESEG(se->type))
886 			holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
887 							se->valid_blocks;
888 		else
889 			holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
890 							se->valid_blocks;
891 	}
892 	mutex_unlock(&dirty_i->seglist_lock);
893 
894 	unusable = max(holes[DATA], holes[NODE]);
895 	if (unusable > ovp_holes)
896 		return unusable - ovp_holes;
897 	return 0;
898 }
899 
900 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
901 {
902 	int ovp_hole_segs =
903 		(overprovision_segments(sbi) - reserved_segments(sbi));
904 	if (unusable > F2FS_OPTION(sbi).unusable_cap)
905 		return -EAGAIN;
906 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
907 		dirty_segments(sbi) > ovp_hole_segs)
908 		return -EAGAIN;
909 	return 0;
910 }
911 
912 /* This is only used by SBI_CP_DISABLED */
913 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
914 {
915 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
916 	unsigned int segno = 0;
917 
918 	mutex_lock(&dirty_i->seglist_lock);
919 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
920 		if (get_valid_blocks(sbi, segno, false))
921 			continue;
922 		if (get_ckpt_valid_blocks(sbi, segno, false))
923 			continue;
924 		mutex_unlock(&dirty_i->seglist_lock);
925 		return segno;
926 	}
927 	mutex_unlock(&dirty_i->seglist_lock);
928 	return NULL_SEGNO;
929 }
930 
931 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
932 		struct block_device *bdev, block_t lstart,
933 		block_t start, block_t len)
934 {
935 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
936 	struct list_head *pend_list;
937 	struct discard_cmd *dc;
938 
939 	f2fs_bug_on(sbi, !len);
940 
941 	pend_list = &dcc->pend_list[plist_idx(len)];
942 
943 	dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL);
944 	INIT_LIST_HEAD(&dc->list);
945 	dc->bdev = bdev;
946 	dc->di.lstart = lstart;
947 	dc->di.start = start;
948 	dc->di.len = len;
949 	dc->ref = 0;
950 	dc->state = D_PREP;
951 	dc->queued = 0;
952 	dc->error = 0;
953 	init_completion(&dc->wait);
954 	list_add_tail(&dc->list, pend_list);
955 	spin_lock_init(&dc->lock);
956 	dc->bio_ref = 0;
957 	atomic_inc(&dcc->discard_cmd_cnt);
958 	dcc->undiscard_blks += len;
959 
960 	return dc;
961 }
962 
963 static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi)
964 {
965 #ifdef CONFIG_F2FS_CHECK_FS
966 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
967 	struct rb_node *cur = rb_first_cached(&dcc->root), *next;
968 	struct discard_cmd *cur_dc, *next_dc;
969 
970 	while (cur) {
971 		next = rb_next(cur);
972 		if (!next)
973 			return true;
974 
975 		cur_dc = rb_entry(cur, struct discard_cmd, rb_node);
976 		next_dc = rb_entry(next, struct discard_cmd, rb_node);
977 
978 		if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) {
979 			f2fs_info(sbi, "broken discard_rbtree, "
980 				"cur(%u, %u) next(%u, %u)",
981 				cur_dc->di.lstart, cur_dc->di.len,
982 				next_dc->di.lstart, next_dc->di.len);
983 			return false;
984 		}
985 		cur = next;
986 	}
987 #endif
988 	return true;
989 }
990 
991 static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi,
992 						block_t blkaddr)
993 {
994 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
995 	struct rb_node *node = dcc->root.rb_root.rb_node;
996 	struct discard_cmd *dc;
997 
998 	while (node) {
999 		dc = rb_entry(node, struct discard_cmd, rb_node);
1000 
1001 		if (blkaddr < dc->di.lstart)
1002 			node = node->rb_left;
1003 		else if (blkaddr >= dc->di.lstart + dc->di.len)
1004 			node = node->rb_right;
1005 		else
1006 			return dc;
1007 	}
1008 	return NULL;
1009 }
1010 
1011 static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root,
1012 				block_t blkaddr,
1013 				struct discard_cmd **prev_entry,
1014 				struct discard_cmd **next_entry,
1015 				struct rb_node ***insert_p,
1016 				struct rb_node **insert_parent)
1017 {
1018 	struct rb_node **pnode = &root->rb_root.rb_node;
1019 	struct rb_node *parent = NULL, *tmp_node;
1020 	struct discard_cmd *dc;
1021 
1022 	*insert_p = NULL;
1023 	*insert_parent = NULL;
1024 	*prev_entry = NULL;
1025 	*next_entry = NULL;
1026 
1027 	if (RB_EMPTY_ROOT(&root->rb_root))
1028 		return NULL;
1029 
1030 	while (*pnode) {
1031 		parent = *pnode;
1032 		dc = rb_entry(*pnode, struct discard_cmd, rb_node);
1033 
1034 		if (blkaddr < dc->di.lstart)
1035 			pnode = &(*pnode)->rb_left;
1036 		else if (blkaddr >= dc->di.lstart + dc->di.len)
1037 			pnode = &(*pnode)->rb_right;
1038 		else
1039 			goto lookup_neighbors;
1040 	}
1041 
1042 	*insert_p = pnode;
1043 	*insert_parent = parent;
1044 
1045 	dc = rb_entry(parent, struct discard_cmd, rb_node);
1046 	tmp_node = parent;
1047 	if (parent && blkaddr > dc->di.lstart)
1048 		tmp_node = rb_next(parent);
1049 	*next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1050 
1051 	tmp_node = parent;
1052 	if (parent && blkaddr < dc->di.lstart)
1053 		tmp_node = rb_prev(parent);
1054 	*prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1055 	return NULL;
1056 
1057 lookup_neighbors:
1058 	/* lookup prev node for merging backward later */
1059 	tmp_node = rb_prev(&dc->rb_node);
1060 	*prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1061 
1062 	/* lookup next node for merging frontward later */
1063 	tmp_node = rb_next(&dc->rb_node);
1064 	*next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1065 	return dc;
1066 }
1067 
1068 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1069 							struct discard_cmd *dc)
1070 {
1071 	if (dc->state == D_DONE)
1072 		atomic_sub(dc->queued, &dcc->queued_discard);
1073 
1074 	list_del(&dc->list);
1075 	rb_erase_cached(&dc->rb_node, &dcc->root);
1076 	dcc->undiscard_blks -= dc->di.len;
1077 
1078 	kmem_cache_free(discard_cmd_slab, dc);
1079 
1080 	atomic_dec(&dcc->discard_cmd_cnt);
1081 }
1082 
1083 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1084 							struct discard_cmd *dc)
1085 {
1086 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1087 	unsigned long flags;
1088 
1089 	trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len);
1090 
1091 	spin_lock_irqsave(&dc->lock, flags);
1092 	if (dc->bio_ref) {
1093 		spin_unlock_irqrestore(&dc->lock, flags);
1094 		return;
1095 	}
1096 	spin_unlock_irqrestore(&dc->lock, flags);
1097 
1098 	f2fs_bug_on(sbi, dc->ref);
1099 
1100 	if (dc->error == -EOPNOTSUPP)
1101 		dc->error = 0;
1102 
1103 	if (dc->error)
1104 		printk_ratelimited(
1105 			"%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1106 			KERN_INFO, sbi->sb->s_id,
1107 			dc->di.lstart, dc->di.start, dc->di.len, dc->error);
1108 	__detach_discard_cmd(dcc, dc);
1109 }
1110 
1111 static void f2fs_submit_discard_endio(struct bio *bio)
1112 {
1113 	struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1114 	unsigned long flags;
1115 
1116 	spin_lock_irqsave(&dc->lock, flags);
1117 	if (!dc->error)
1118 		dc->error = blk_status_to_errno(bio->bi_status);
1119 	dc->bio_ref--;
1120 	if (!dc->bio_ref && dc->state == D_SUBMIT) {
1121 		dc->state = D_DONE;
1122 		complete_all(&dc->wait);
1123 	}
1124 	spin_unlock_irqrestore(&dc->lock, flags);
1125 	bio_put(bio);
1126 }
1127 
1128 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1129 				block_t start, block_t end)
1130 {
1131 #ifdef CONFIG_F2FS_CHECK_FS
1132 	struct seg_entry *sentry;
1133 	unsigned int segno;
1134 	block_t blk = start;
1135 	unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1136 	unsigned long *map;
1137 
1138 	while (blk < end) {
1139 		segno = GET_SEGNO(sbi, blk);
1140 		sentry = get_seg_entry(sbi, segno);
1141 		offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1142 
1143 		if (end < START_BLOCK(sbi, segno + 1))
1144 			size = GET_BLKOFF_FROM_SEG0(sbi, end);
1145 		else
1146 			size = max_blocks;
1147 		map = (unsigned long *)(sentry->cur_valid_map);
1148 		offset = __find_rev_next_bit(map, size, offset);
1149 		f2fs_bug_on(sbi, offset != size);
1150 		blk = START_BLOCK(sbi, segno + 1);
1151 	}
1152 #endif
1153 }
1154 
1155 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1156 				struct discard_policy *dpolicy,
1157 				int discard_type, unsigned int granularity)
1158 {
1159 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1160 
1161 	/* common policy */
1162 	dpolicy->type = discard_type;
1163 	dpolicy->sync = true;
1164 	dpolicy->ordered = false;
1165 	dpolicy->granularity = granularity;
1166 
1167 	dpolicy->max_requests = dcc->max_discard_request;
1168 	dpolicy->io_aware_gran = dcc->discard_io_aware_gran;
1169 	dpolicy->timeout = false;
1170 
1171 	if (discard_type == DPOLICY_BG) {
1172 		dpolicy->min_interval = dcc->min_discard_issue_time;
1173 		dpolicy->mid_interval = dcc->mid_discard_issue_time;
1174 		dpolicy->max_interval = dcc->max_discard_issue_time;
1175 		dpolicy->io_aware = true;
1176 		dpolicy->sync = false;
1177 		dpolicy->ordered = true;
1178 		if (utilization(sbi) > dcc->discard_urgent_util) {
1179 			dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1180 			if (atomic_read(&dcc->discard_cmd_cnt))
1181 				dpolicy->max_interval =
1182 					dcc->min_discard_issue_time;
1183 		}
1184 	} else if (discard_type == DPOLICY_FORCE) {
1185 		dpolicy->min_interval = dcc->min_discard_issue_time;
1186 		dpolicy->mid_interval = dcc->mid_discard_issue_time;
1187 		dpolicy->max_interval = dcc->max_discard_issue_time;
1188 		dpolicy->io_aware = false;
1189 	} else if (discard_type == DPOLICY_FSTRIM) {
1190 		dpolicy->io_aware = false;
1191 	} else if (discard_type == DPOLICY_UMOUNT) {
1192 		dpolicy->io_aware = false;
1193 		/* we need to issue all to keep CP_TRIMMED_FLAG */
1194 		dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1195 		dpolicy->timeout = true;
1196 	}
1197 }
1198 
1199 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1200 				struct block_device *bdev, block_t lstart,
1201 				block_t start, block_t len);
1202 
1203 #ifdef CONFIG_BLK_DEV_ZONED
1204 static void __submit_zone_reset_cmd(struct f2fs_sb_info *sbi,
1205 				   struct discard_cmd *dc, blk_opf_t flag,
1206 				   struct list_head *wait_list,
1207 				   unsigned int *issued)
1208 {
1209 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1210 	struct block_device *bdev = dc->bdev;
1211 	struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS);
1212 	unsigned long flags;
1213 
1214 	trace_f2fs_issue_reset_zone(bdev, dc->di.start);
1215 
1216 	spin_lock_irqsave(&dc->lock, flags);
1217 	dc->state = D_SUBMIT;
1218 	dc->bio_ref++;
1219 	spin_unlock_irqrestore(&dc->lock, flags);
1220 
1221 	if (issued)
1222 		(*issued)++;
1223 
1224 	atomic_inc(&dcc->queued_discard);
1225 	dc->queued++;
1226 	list_move_tail(&dc->list, wait_list);
1227 
1228 	/* sanity check on discard range */
1229 	__check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len);
1230 
1231 	bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start);
1232 	bio->bi_private = dc;
1233 	bio->bi_end_io = f2fs_submit_discard_endio;
1234 	submit_bio(bio);
1235 
1236 	atomic_inc(&dcc->issued_discard);
1237 	f2fs_update_iostat(sbi, NULL, FS_ZONE_RESET_IO, dc->di.len * F2FS_BLKSIZE);
1238 }
1239 #endif
1240 
1241 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1242 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1243 				struct discard_policy *dpolicy,
1244 				struct discard_cmd *dc, int *issued)
1245 {
1246 	struct block_device *bdev = dc->bdev;
1247 	unsigned int max_discard_blocks =
1248 			SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1249 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1250 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1251 					&(dcc->fstrim_list) : &(dcc->wait_list);
1252 	blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0;
1253 	block_t lstart, start, len, total_len;
1254 	int err = 0;
1255 
1256 	if (dc->state != D_PREP)
1257 		return 0;
1258 
1259 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1260 		return 0;
1261 
1262 #ifdef CONFIG_BLK_DEV_ZONED
1263 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) {
1264 		int devi = f2fs_bdev_index(sbi, bdev);
1265 
1266 		if (devi < 0)
1267 			return -EINVAL;
1268 
1269 		if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1270 			__submit_zone_reset_cmd(sbi, dc, flag,
1271 						wait_list, issued);
1272 			return 0;
1273 		}
1274 	}
1275 #endif
1276 
1277 	trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len);
1278 
1279 	lstart = dc->di.lstart;
1280 	start = dc->di.start;
1281 	len = dc->di.len;
1282 	total_len = len;
1283 
1284 	dc->di.len = 0;
1285 
1286 	while (total_len && *issued < dpolicy->max_requests && !err) {
1287 		struct bio *bio = NULL;
1288 		unsigned long flags;
1289 		bool last = true;
1290 
1291 		if (len > max_discard_blocks) {
1292 			len = max_discard_blocks;
1293 			last = false;
1294 		}
1295 
1296 		(*issued)++;
1297 		if (*issued == dpolicy->max_requests)
1298 			last = true;
1299 
1300 		dc->di.len += len;
1301 
1302 		if (time_to_inject(sbi, FAULT_DISCARD)) {
1303 			err = -EIO;
1304 		} else {
1305 			err = __blkdev_issue_discard(bdev,
1306 					SECTOR_FROM_BLOCK(start),
1307 					SECTOR_FROM_BLOCK(len),
1308 					GFP_NOFS, &bio);
1309 		}
1310 		if (err) {
1311 			spin_lock_irqsave(&dc->lock, flags);
1312 			if (dc->state == D_PARTIAL)
1313 				dc->state = D_SUBMIT;
1314 			spin_unlock_irqrestore(&dc->lock, flags);
1315 
1316 			break;
1317 		}
1318 
1319 		f2fs_bug_on(sbi, !bio);
1320 
1321 		/*
1322 		 * should keep before submission to avoid D_DONE
1323 		 * right away
1324 		 */
1325 		spin_lock_irqsave(&dc->lock, flags);
1326 		if (last)
1327 			dc->state = D_SUBMIT;
1328 		else
1329 			dc->state = D_PARTIAL;
1330 		dc->bio_ref++;
1331 		spin_unlock_irqrestore(&dc->lock, flags);
1332 
1333 		atomic_inc(&dcc->queued_discard);
1334 		dc->queued++;
1335 		list_move_tail(&dc->list, wait_list);
1336 
1337 		/* sanity check on discard range */
1338 		__check_sit_bitmap(sbi, lstart, lstart + len);
1339 
1340 		bio->bi_private = dc;
1341 		bio->bi_end_io = f2fs_submit_discard_endio;
1342 		bio->bi_opf |= flag;
1343 		submit_bio(bio);
1344 
1345 		atomic_inc(&dcc->issued_discard);
1346 
1347 		f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE);
1348 
1349 		lstart += len;
1350 		start += len;
1351 		total_len -= len;
1352 		len = total_len;
1353 	}
1354 
1355 	if (!err && len) {
1356 		dcc->undiscard_blks -= len;
1357 		__update_discard_tree_range(sbi, bdev, lstart, start, len);
1358 	}
1359 	return err;
1360 }
1361 
1362 static void __insert_discard_cmd(struct f2fs_sb_info *sbi,
1363 				struct block_device *bdev, block_t lstart,
1364 				block_t start, block_t len)
1365 {
1366 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1367 	struct rb_node **p = &dcc->root.rb_root.rb_node;
1368 	struct rb_node *parent = NULL;
1369 	struct discard_cmd *dc;
1370 	bool leftmost = true;
1371 
1372 	/* look up rb tree to find parent node */
1373 	while (*p) {
1374 		parent = *p;
1375 		dc = rb_entry(parent, struct discard_cmd, rb_node);
1376 
1377 		if (lstart < dc->di.lstart) {
1378 			p = &(*p)->rb_left;
1379 		} else if (lstart >= dc->di.lstart + dc->di.len) {
1380 			p = &(*p)->rb_right;
1381 			leftmost = false;
1382 		} else {
1383 			f2fs_bug_on(sbi, 1);
1384 		}
1385 	}
1386 
1387 	dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1388 
1389 	rb_link_node(&dc->rb_node, parent, p);
1390 	rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1391 }
1392 
1393 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1394 						struct discard_cmd *dc)
1395 {
1396 	list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]);
1397 }
1398 
1399 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1400 				struct discard_cmd *dc, block_t blkaddr)
1401 {
1402 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1403 	struct discard_info di = dc->di;
1404 	bool modified = false;
1405 
1406 	if (dc->state == D_DONE || dc->di.len == 1) {
1407 		__remove_discard_cmd(sbi, dc);
1408 		return;
1409 	}
1410 
1411 	dcc->undiscard_blks -= di.len;
1412 
1413 	if (blkaddr > di.lstart) {
1414 		dc->di.len = blkaddr - dc->di.lstart;
1415 		dcc->undiscard_blks += dc->di.len;
1416 		__relocate_discard_cmd(dcc, dc);
1417 		modified = true;
1418 	}
1419 
1420 	if (blkaddr < di.lstart + di.len - 1) {
1421 		if (modified) {
1422 			__insert_discard_cmd(sbi, dc->bdev, blkaddr + 1,
1423 					di.start + blkaddr + 1 - di.lstart,
1424 					di.lstart + di.len - 1 - blkaddr);
1425 		} else {
1426 			dc->di.lstart++;
1427 			dc->di.len--;
1428 			dc->di.start++;
1429 			dcc->undiscard_blks += dc->di.len;
1430 			__relocate_discard_cmd(dcc, dc);
1431 		}
1432 	}
1433 }
1434 
1435 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1436 				struct block_device *bdev, block_t lstart,
1437 				block_t start, block_t len)
1438 {
1439 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1440 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1441 	struct discard_cmd *dc;
1442 	struct discard_info di = {0};
1443 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1444 	unsigned int max_discard_blocks =
1445 			SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1446 	block_t end = lstart + len;
1447 
1448 	dc = __lookup_discard_cmd_ret(&dcc->root, lstart,
1449 				&prev_dc, &next_dc, &insert_p, &insert_parent);
1450 	if (dc)
1451 		prev_dc = dc;
1452 
1453 	if (!prev_dc) {
1454 		di.lstart = lstart;
1455 		di.len = next_dc ? next_dc->di.lstart - lstart : len;
1456 		di.len = min(di.len, len);
1457 		di.start = start;
1458 	}
1459 
1460 	while (1) {
1461 		struct rb_node *node;
1462 		bool merged = false;
1463 		struct discard_cmd *tdc = NULL;
1464 
1465 		if (prev_dc) {
1466 			di.lstart = prev_dc->di.lstart + prev_dc->di.len;
1467 			if (di.lstart < lstart)
1468 				di.lstart = lstart;
1469 			if (di.lstart >= end)
1470 				break;
1471 
1472 			if (!next_dc || next_dc->di.lstart > end)
1473 				di.len = end - di.lstart;
1474 			else
1475 				di.len = next_dc->di.lstart - di.lstart;
1476 			di.start = start + di.lstart - lstart;
1477 		}
1478 
1479 		if (!di.len)
1480 			goto next;
1481 
1482 		if (prev_dc && prev_dc->state == D_PREP &&
1483 			prev_dc->bdev == bdev &&
1484 			__is_discard_back_mergeable(&di, &prev_dc->di,
1485 							max_discard_blocks)) {
1486 			prev_dc->di.len += di.len;
1487 			dcc->undiscard_blks += di.len;
1488 			__relocate_discard_cmd(dcc, prev_dc);
1489 			di = prev_dc->di;
1490 			tdc = prev_dc;
1491 			merged = true;
1492 		}
1493 
1494 		if (next_dc && next_dc->state == D_PREP &&
1495 			next_dc->bdev == bdev &&
1496 			__is_discard_front_mergeable(&di, &next_dc->di,
1497 							max_discard_blocks)) {
1498 			next_dc->di.lstart = di.lstart;
1499 			next_dc->di.len += di.len;
1500 			next_dc->di.start = di.start;
1501 			dcc->undiscard_blks += di.len;
1502 			__relocate_discard_cmd(dcc, next_dc);
1503 			if (tdc)
1504 				__remove_discard_cmd(sbi, tdc);
1505 			merged = true;
1506 		}
1507 
1508 		if (!merged)
1509 			__insert_discard_cmd(sbi, bdev,
1510 						di.lstart, di.start, di.len);
1511  next:
1512 		prev_dc = next_dc;
1513 		if (!prev_dc)
1514 			break;
1515 
1516 		node = rb_next(&prev_dc->rb_node);
1517 		next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1518 	}
1519 }
1520 
1521 #ifdef CONFIG_BLK_DEV_ZONED
1522 static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi,
1523 		struct block_device *bdev, block_t blkstart, block_t lblkstart,
1524 		block_t blklen)
1525 {
1526 	trace_f2fs_queue_reset_zone(bdev, blkstart);
1527 
1528 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1529 	__insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen);
1530 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1531 }
1532 #endif
1533 
1534 static void __queue_discard_cmd(struct f2fs_sb_info *sbi,
1535 		struct block_device *bdev, block_t blkstart, block_t blklen)
1536 {
1537 	block_t lblkstart = blkstart;
1538 
1539 	if (!f2fs_bdev_support_discard(bdev))
1540 		return;
1541 
1542 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
1543 
1544 	if (f2fs_is_multi_device(sbi)) {
1545 		int devi = f2fs_target_device_index(sbi, blkstart);
1546 
1547 		blkstart -= FDEV(devi).start_blk;
1548 	}
1549 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1550 	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1551 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1552 }
1553 
1554 static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1555 		struct discard_policy *dpolicy, int *issued)
1556 {
1557 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1558 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1559 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1560 	struct discard_cmd *dc;
1561 	struct blk_plug plug;
1562 	bool io_interrupted = false;
1563 
1564 	mutex_lock(&dcc->cmd_lock);
1565 	dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos,
1566 				&prev_dc, &next_dc, &insert_p, &insert_parent);
1567 	if (!dc)
1568 		dc = next_dc;
1569 
1570 	blk_start_plug(&plug);
1571 
1572 	while (dc) {
1573 		struct rb_node *node;
1574 		int err = 0;
1575 
1576 		if (dc->state != D_PREP)
1577 			goto next;
1578 
1579 		if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1580 			io_interrupted = true;
1581 			break;
1582 		}
1583 
1584 		dcc->next_pos = dc->di.lstart + dc->di.len;
1585 		err = __submit_discard_cmd(sbi, dpolicy, dc, issued);
1586 
1587 		if (*issued >= dpolicy->max_requests)
1588 			break;
1589 next:
1590 		node = rb_next(&dc->rb_node);
1591 		if (err)
1592 			__remove_discard_cmd(sbi, dc);
1593 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1594 	}
1595 
1596 	blk_finish_plug(&plug);
1597 
1598 	if (!dc)
1599 		dcc->next_pos = 0;
1600 
1601 	mutex_unlock(&dcc->cmd_lock);
1602 
1603 	if (!(*issued) && io_interrupted)
1604 		*issued = -1;
1605 }
1606 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1607 					struct discard_policy *dpolicy);
1608 
1609 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1610 					struct discard_policy *dpolicy)
1611 {
1612 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1613 	struct list_head *pend_list;
1614 	struct discard_cmd *dc, *tmp;
1615 	struct blk_plug plug;
1616 	int i, issued;
1617 	bool io_interrupted = false;
1618 
1619 	if (dpolicy->timeout)
1620 		f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1621 
1622 retry:
1623 	issued = 0;
1624 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1625 		if (dpolicy->timeout &&
1626 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1627 			break;
1628 
1629 		if (i + 1 < dpolicy->granularity)
1630 			break;
1631 
1632 		if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) {
1633 			__issue_discard_cmd_orderly(sbi, dpolicy, &issued);
1634 			return issued;
1635 		}
1636 
1637 		pend_list = &dcc->pend_list[i];
1638 
1639 		mutex_lock(&dcc->cmd_lock);
1640 		if (list_empty(pend_list))
1641 			goto next;
1642 		if (unlikely(dcc->rbtree_check))
1643 			f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
1644 		blk_start_plug(&plug);
1645 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1646 			f2fs_bug_on(sbi, dc->state != D_PREP);
1647 
1648 			if (dpolicy->timeout &&
1649 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1650 				break;
1651 
1652 			if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1653 						!is_idle(sbi, DISCARD_TIME)) {
1654 				io_interrupted = true;
1655 				break;
1656 			}
1657 
1658 			__submit_discard_cmd(sbi, dpolicy, dc, &issued);
1659 
1660 			if (issued >= dpolicy->max_requests)
1661 				break;
1662 		}
1663 		blk_finish_plug(&plug);
1664 next:
1665 		mutex_unlock(&dcc->cmd_lock);
1666 
1667 		if (issued >= dpolicy->max_requests || io_interrupted)
1668 			break;
1669 	}
1670 
1671 	if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1672 		__wait_all_discard_cmd(sbi, dpolicy);
1673 		goto retry;
1674 	}
1675 
1676 	if (!issued && io_interrupted)
1677 		issued = -1;
1678 
1679 	return issued;
1680 }
1681 
1682 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1683 {
1684 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1685 	struct list_head *pend_list;
1686 	struct discard_cmd *dc, *tmp;
1687 	int i;
1688 	bool dropped = false;
1689 
1690 	mutex_lock(&dcc->cmd_lock);
1691 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1692 		pend_list = &dcc->pend_list[i];
1693 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1694 			f2fs_bug_on(sbi, dc->state != D_PREP);
1695 			__remove_discard_cmd(sbi, dc);
1696 			dropped = true;
1697 		}
1698 	}
1699 	mutex_unlock(&dcc->cmd_lock);
1700 
1701 	return dropped;
1702 }
1703 
1704 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1705 {
1706 	__drop_discard_cmd(sbi);
1707 }
1708 
1709 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1710 							struct discard_cmd *dc)
1711 {
1712 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1713 	unsigned int len = 0;
1714 
1715 	wait_for_completion_io(&dc->wait);
1716 	mutex_lock(&dcc->cmd_lock);
1717 	f2fs_bug_on(sbi, dc->state != D_DONE);
1718 	dc->ref--;
1719 	if (!dc->ref) {
1720 		if (!dc->error)
1721 			len = dc->di.len;
1722 		__remove_discard_cmd(sbi, dc);
1723 	}
1724 	mutex_unlock(&dcc->cmd_lock);
1725 
1726 	return len;
1727 }
1728 
1729 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1730 						struct discard_policy *dpolicy,
1731 						block_t start, block_t end)
1732 {
1733 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1734 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1735 					&(dcc->fstrim_list) : &(dcc->wait_list);
1736 	struct discard_cmd *dc = NULL, *iter, *tmp;
1737 	unsigned int trimmed = 0;
1738 
1739 next:
1740 	dc = NULL;
1741 
1742 	mutex_lock(&dcc->cmd_lock);
1743 	list_for_each_entry_safe(iter, tmp, wait_list, list) {
1744 		if (iter->di.lstart + iter->di.len <= start ||
1745 					end <= iter->di.lstart)
1746 			continue;
1747 		if (iter->di.len < dpolicy->granularity)
1748 			continue;
1749 		if (iter->state == D_DONE && !iter->ref) {
1750 			wait_for_completion_io(&iter->wait);
1751 			if (!iter->error)
1752 				trimmed += iter->di.len;
1753 			__remove_discard_cmd(sbi, iter);
1754 		} else {
1755 			iter->ref++;
1756 			dc = iter;
1757 			break;
1758 		}
1759 	}
1760 	mutex_unlock(&dcc->cmd_lock);
1761 
1762 	if (dc) {
1763 		trimmed += __wait_one_discard_bio(sbi, dc);
1764 		goto next;
1765 	}
1766 
1767 	return trimmed;
1768 }
1769 
1770 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1771 						struct discard_policy *dpolicy)
1772 {
1773 	struct discard_policy dp;
1774 	unsigned int discard_blks;
1775 
1776 	if (dpolicy)
1777 		return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1778 
1779 	/* wait all */
1780 	__init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY);
1781 	discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1782 	__init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY);
1783 	discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1784 
1785 	return discard_blks;
1786 }
1787 
1788 /* This should be covered by global mutex, &sit_i->sentry_lock */
1789 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1790 {
1791 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1792 	struct discard_cmd *dc;
1793 	bool need_wait = false;
1794 
1795 	mutex_lock(&dcc->cmd_lock);
1796 	dc = __lookup_discard_cmd(sbi, blkaddr);
1797 #ifdef CONFIG_BLK_DEV_ZONED
1798 	if (dc && f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) {
1799 		int devi = f2fs_bdev_index(sbi, dc->bdev);
1800 
1801 		if (devi < 0) {
1802 			mutex_unlock(&dcc->cmd_lock);
1803 			return;
1804 		}
1805 
1806 		if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1807 			/* force submit zone reset */
1808 			if (dc->state == D_PREP)
1809 				__submit_zone_reset_cmd(sbi, dc, REQ_SYNC,
1810 							&dcc->wait_list, NULL);
1811 			dc->ref++;
1812 			mutex_unlock(&dcc->cmd_lock);
1813 			/* wait zone reset */
1814 			__wait_one_discard_bio(sbi, dc);
1815 			return;
1816 		}
1817 	}
1818 #endif
1819 	if (dc) {
1820 		if (dc->state == D_PREP) {
1821 			__punch_discard_cmd(sbi, dc, blkaddr);
1822 		} else {
1823 			dc->ref++;
1824 			need_wait = true;
1825 		}
1826 	}
1827 	mutex_unlock(&dcc->cmd_lock);
1828 
1829 	if (need_wait)
1830 		__wait_one_discard_bio(sbi, dc);
1831 }
1832 
1833 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1834 {
1835 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1836 
1837 	if (dcc && dcc->f2fs_issue_discard) {
1838 		struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1839 
1840 		dcc->f2fs_issue_discard = NULL;
1841 		kthread_stop(discard_thread);
1842 	}
1843 }
1844 
1845 /**
1846  * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT
1847  * @sbi: the f2fs_sb_info data for discard cmd to issue
1848  *
1849  * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped
1850  *
1851  * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false.
1852  */
1853 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1854 {
1855 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1856 	struct discard_policy dpolicy;
1857 	bool dropped;
1858 
1859 	if (!atomic_read(&dcc->discard_cmd_cnt))
1860 		return true;
1861 
1862 	__init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1863 					dcc->discard_granularity);
1864 	__issue_discard_cmd(sbi, &dpolicy);
1865 	dropped = __drop_discard_cmd(sbi);
1866 
1867 	/* just to make sure there is no pending discard commands */
1868 	__wait_all_discard_cmd(sbi, NULL);
1869 
1870 	f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1871 	return !dropped;
1872 }
1873 
1874 static int issue_discard_thread(void *data)
1875 {
1876 	struct f2fs_sb_info *sbi = data;
1877 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1878 	wait_queue_head_t *q = &dcc->discard_wait_queue;
1879 	struct discard_policy dpolicy;
1880 	unsigned int wait_ms = dcc->min_discard_issue_time;
1881 	int issued;
1882 
1883 	set_freezable();
1884 
1885 	do {
1886 		wait_event_interruptible_timeout(*q,
1887 				kthread_should_stop() || freezing(current) ||
1888 				dcc->discard_wake,
1889 				msecs_to_jiffies(wait_ms));
1890 
1891 		if (sbi->gc_mode == GC_URGENT_HIGH ||
1892 			!f2fs_available_free_memory(sbi, DISCARD_CACHE))
1893 			__init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE,
1894 						MIN_DISCARD_GRANULARITY);
1895 		else
1896 			__init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1897 						dcc->discard_granularity);
1898 
1899 		if (dcc->discard_wake)
1900 			dcc->discard_wake = false;
1901 
1902 		/* clean up pending candidates before going to sleep */
1903 		if (atomic_read(&dcc->queued_discard))
1904 			__wait_all_discard_cmd(sbi, NULL);
1905 
1906 		if (try_to_freeze())
1907 			continue;
1908 		if (f2fs_readonly(sbi->sb))
1909 			continue;
1910 		if (kthread_should_stop())
1911 			return 0;
1912 		if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) ||
1913 			!atomic_read(&dcc->discard_cmd_cnt)) {
1914 			wait_ms = dpolicy.max_interval;
1915 			continue;
1916 		}
1917 
1918 		sb_start_intwrite(sbi->sb);
1919 
1920 		issued = __issue_discard_cmd(sbi, &dpolicy);
1921 		if (issued > 0) {
1922 			__wait_all_discard_cmd(sbi, &dpolicy);
1923 			wait_ms = dpolicy.min_interval;
1924 		} else if (issued == -1) {
1925 			wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1926 			if (!wait_ms)
1927 				wait_ms = dpolicy.mid_interval;
1928 		} else {
1929 			wait_ms = dpolicy.max_interval;
1930 		}
1931 		if (!atomic_read(&dcc->discard_cmd_cnt))
1932 			wait_ms = dpolicy.max_interval;
1933 
1934 		sb_end_intwrite(sbi->sb);
1935 
1936 	} while (!kthread_should_stop());
1937 	return 0;
1938 }
1939 
1940 #ifdef CONFIG_BLK_DEV_ZONED
1941 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1942 		struct block_device *bdev, block_t blkstart, block_t blklen)
1943 {
1944 	sector_t sector, nr_sects;
1945 	block_t lblkstart = blkstart;
1946 	int devi = 0;
1947 	u64 remainder = 0;
1948 
1949 	if (f2fs_is_multi_device(sbi)) {
1950 		devi = f2fs_target_device_index(sbi, blkstart);
1951 		if (blkstart < FDEV(devi).start_blk ||
1952 		    blkstart > FDEV(devi).end_blk) {
1953 			f2fs_err(sbi, "Invalid block %x", blkstart);
1954 			return -EIO;
1955 		}
1956 		blkstart -= FDEV(devi).start_blk;
1957 	}
1958 
1959 	/* For sequential zones, reset the zone write pointer */
1960 	if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1961 		sector = SECTOR_FROM_BLOCK(blkstart);
1962 		nr_sects = SECTOR_FROM_BLOCK(blklen);
1963 		div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder);
1964 
1965 		if (remainder || nr_sects != bdev_zone_sectors(bdev)) {
1966 			f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1967 				 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1968 				 blkstart, blklen);
1969 			return -EIO;
1970 		}
1971 
1972 		if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) {
1973 			trace_f2fs_issue_reset_zone(bdev, blkstart);
1974 			return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1975 						sector, nr_sects, GFP_NOFS);
1976 		}
1977 
1978 		__queue_zone_reset_cmd(sbi, bdev, blkstart, lblkstart, blklen);
1979 		return 0;
1980 	}
1981 
1982 	/* For conventional zones, use regular discard if supported */
1983 	__queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1984 	return 0;
1985 }
1986 #endif
1987 
1988 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1989 		struct block_device *bdev, block_t blkstart, block_t blklen)
1990 {
1991 #ifdef CONFIG_BLK_DEV_ZONED
1992 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1993 		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1994 #endif
1995 	__queue_discard_cmd(sbi, bdev, blkstart, blklen);
1996 	return 0;
1997 }
1998 
1999 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
2000 				block_t blkstart, block_t blklen)
2001 {
2002 	sector_t start = blkstart, len = 0;
2003 	struct block_device *bdev;
2004 	struct seg_entry *se;
2005 	unsigned int offset;
2006 	block_t i;
2007 	int err = 0;
2008 
2009 	bdev = f2fs_target_device(sbi, blkstart, NULL);
2010 
2011 	for (i = blkstart; i < blkstart + blklen; i++, len++) {
2012 		if (i != start) {
2013 			struct block_device *bdev2 =
2014 				f2fs_target_device(sbi, i, NULL);
2015 
2016 			if (bdev2 != bdev) {
2017 				err = __issue_discard_async(sbi, bdev,
2018 						start, len);
2019 				if (err)
2020 					return err;
2021 				bdev = bdev2;
2022 				start = i;
2023 				len = 0;
2024 			}
2025 		}
2026 
2027 		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
2028 		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
2029 
2030 		if (f2fs_block_unit_discard(sbi) &&
2031 				!f2fs_test_and_set_bit(offset, se->discard_map))
2032 			sbi->discard_blks--;
2033 	}
2034 
2035 	if (len)
2036 		err = __issue_discard_async(sbi, bdev, start, len);
2037 	return err;
2038 }
2039 
2040 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
2041 							bool check_only)
2042 {
2043 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2044 	int max_blocks = sbi->blocks_per_seg;
2045 	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
2046 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2047 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2048 	unsigned long *discard_map = (unsigned long *)se->discard_map;
2049 	unsigned long *dmap = SIT_I(sbi)->tmp_map;
2050 	unsigned int start = 0, end = -1;
2051 	bool force = (cpc->reason & CP_DISCARD);
2052 	struct discard_entry *de = NULL;
2053 	struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
2054 	int i;
2055 
2056 	if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi) ||
2057 			!f2fs_block_unit_discard(sbi))
2058 		return false;
2059 
2060 	if (!force) {
2061 		if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
2062 			SM_I(sbi)->dcc_info->nr_discards >=
2063 				SM_I(sbi)->dcc_info->max_discards)
2064 			return false;
2065 	}
2066 
2067 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
2068 	for (i = 0; i < entries; i++)
2069 		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
2070 				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
2071 
2072 	while (force || SM_I(sbi)->dcc_info->nr_discards <=
2073 				SM_I(sbi)->dcc_info->max_discards) {
2074 		start = __find_rev_next_bit(dmap, max_blocks, end + 1);
2075 		if (start >= max_blocks)
2076 			break;
2077 
2078 		end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
2079 		if (force && start && end != max_blocks
2080 					&& (end - start) < cpc->trim_minlen)
2081 			continue;
2082 
2083 		if (check_only)
2084 			return true;
2085 
2086 		if (!de) {
2087 			de = f2fs_kmem_cache_alloc(discard_entry_slab,
2088 						GFP_F2FS_ZERO, true, NULL);
2089 			de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
2090 			list_add_tail(&de->list, head);
2091 		}
2092 
2093 		for (i = start; i < end; i++)
2094 			__set_bit_le(i, (void *)de->discard_map);
2095 
2096 		SM_I(sbi)->dcc_info->nr_discards += end - start;
2097 	}
2098 	return false;
2099 }
2100 
2101 static void release_discard_addr(struct discard_entry *entry)
2102 {
2103 	list_del(&entry->list);
2104 	kmem_cache_free(discard_entry_slab, entry);
2105 }
2106 
2107 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
2108 {
2109 	struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
2110 	struct discard_entry *entry, *this;
2111 
2112 	/* drop caches */
2113 	list_for_each_entry_safe(entry, this, head, list)
2114 		release_discard_addr(entry);
2115 }
2116 
2117 /*
2118  * Should call f2fs_clear_prefree_segments after checkpoint is done.
2119  */
2120 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
2121 {
2122 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2123 	unsigned int segno;
2124 
2125 	mutex_lock(&dirty_i->seglist_lock);
2126 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
2127 		__set_test_and_free(sbi, segno, false);
2128 	mutex_unlock(&dirty_i->seglist_lock);
2129 }
2130 
2131 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
2132 						struct cp_control *cpc)
2133 {
2134 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2135 	struct list_head *head = &dcc->entry_list;
2136 	struct discard_entry *entry, *this;
2137 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2138 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
2139 	unsigned int start = 0, end = -1;
2140 	unsigned int secno, start_segno;
2141 	bool force = (cpc->reason & CP_DISCARD);
2142 	bool section_alignment = F2FS_OPTION(sbi).discard_unit ==
2143 						DISCARD_UNIT_SECTION;
2144 
2145 	if (f2fs_lfs_mode(sbi) && __is_large_section(sbi))
2146 		section_alignment = true;
2147 
2148 	mutex_lock(&dirty_i->seglist_lock);
2149 
2150 	while (1) {
2151 		int i;
2152 
2153 		if (section_alignment && end != -1)
2154 			end--;
2155 		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
2156 		if (start >= MAIN_SEGS(sbi))
2157 			break;
2158 		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
2159 								start + 1);
2160 
2161 		if (section_alignment) {
2162 			start = rounddown(start, sbi->segs_per_sec);
2163 			end = roundup(end, sbi->segs_per_sec);
2164 		}
2165 
2166 		for (i = start; i < end; i++) {
2167 			if (test_and_clear_bit(i, prefree_map))
2168 				dirty_i->nr_dirty[PRE]--;
2169 		}
2170 
2171 		if (!f2fs_realtime_discard_enable(sbi))
2172 			continue;
2173 
2174 		if (force && start >= cpc->trim_start &&
2175 					(end - 1) <= cpc->trim_end)
2176 			continue;
2177 
2178 		/* Should cover 2MB zoned device for zone-based reset */
2179 		if (!f2fs_sb_has_blkzoned(sbi) &&
2180 		    (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) {
2181 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2182 				(end - start) << sbi->log_blocks_per_seg);
2183 			continue;
2184 		}
2185 next:
2186 		secno = GET_SEC_FROM_SEG(sbi, start);
2187 		start_segno = GET_SEG_FROM_SEC(sbi, secno);
2188 		if (!IS_CURSEC(sbi, secno) &&
2189 			!get_valid_blocks(sbi, start, true))
2190 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2191 				sbi->segs_per_sec << sbi->log_blocks_per_seg);
2192 
2193 		start = start_segno + sbi->segs_per_sec;
2194 		if (start < end)
2195 			goto next;
2196 		else
2197 			end = start - 1;
2198 	}
2199 	mutex_unlock(&dirty_i->seglist_lock);
2200 
2201 	if (!f2fs_block_unit_discard(sbi))
2202 		goto wakeup;
2203 
2204 	/* send small discards */
2205 	list_for_each_entry_safe(entry, this, head, list) {
2206 		unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2207 		bool is_valid = test_bit_le(0, entry->discard_map);
2208 
2209 find_next:
2210 		if (is_valid) {
2211 			next_pos = find_next_zero_bit_le(entry->discard_map,
2212 					sbi->blocks_per_seg, cur_pos);
2213 			len = next_pos - cur_pos;
2214 
2215 			if (f2fs_sb_has_blkzoned(sbi) ||
2216 			    (force && len < cpc->trim_minlen))
2217 				goto skip;
2218 
2219 			f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2220 									len);
2221 			total_len += len;
2222 		} else {
2223 			next_pos = find_next_bit_le(entry->discard_map,
2224 					sbi->blocks_per_seg, cur_pos);
2225 		}
2226 skip:
2227 		cur_pos = next_pos;
2228 		is_valid = !is_valid;
2229 
2230 		if (cur_pos < sbi->blocks_per_seg)
2231 			goto find_next;
2232 
2233 		release_discard_addr(entry);
2234 		dcc->nr_discards -= total_len;
2235 	}
2236 
2237 wakeup:
2238 	wake_up_discard_thread(sbi, false);
2239 }
2240 
2241 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi)
2242 {
2243 	dev_t dev = sbi->sb->s_bdev->bd_dev;
2244 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2245 	int err = 0;
2246 
2247 	if (!f2fs_realtime_discard_enable(sbi))
2248 		return 0;
2249 
2250 	dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2251 				"f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2252 	if (IS_ERR(dcc->f2fs_issue_discard)) {
2253 		err = PTR_ERR(dcc->f2fs_issue_discard);
2254 		dcc->f2fs_issue_discard = NULL;
2255 	}
2256 
2257 	return err;
2258 }
2259 
2260 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2261 {
2262 	struct discard_cmd_control *dcc;
2263 	int err = 0, i;
2264 
2265 	if (SM_I(sbi)->dcc_info) {
2266 		dcc = SM_I(sbi)->dcc_info;
2267 		goto init_thread;
2268 	}
2269 
2270 	dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2271 	if (!dcc)
2272 		return -ENOMEM;
2273 
2274 	dcc->discard_io_aware_gran = MAX_PLIST_NUM;
2275 	dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2276 	dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY;
2277 	if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT)
2278 		dcc->discard_granularity = sbi->blocks_per_seg;
2279 	else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION)
2280 		dcc->discard_granularity = BLKS_PER_SEC(sbi);
2281 
2282 	INIT_LIST_HEAD(&dcc->entry_list);
2283 	for (i = 0; i < MAX_PLIST_NUM; i++)
2284 		INIT_LIST_HEAD(&dcc->pend_list[i]);
2285 	INIT_LIST_HEAD(&dcc->wait_list);
2286 	INIT_LIST_HEAD(&dcc->fstrim_list);
2287 	mutex_init(&dcc->cmd_lock);
2288 	atomic_set(&dcc->issued_discard, 0);
2289 	atomic_set(&dcc->queued_discard, 0);
2290 	atomic_set(&dcc->discard_cmd_cnt, 0);
2291 	dcc->nr_discards = 0;
2292 	dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2293 	dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST;
2294 	dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME;
2295 	dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME;
2296 	dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME;
2297 	dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL;
2298 	dcc->undiscard_blks = 0;
2299 	dcc->next_pos = 0;
2300 	dcc->root = RB_ROOT_CACHED;
2301 	dcc->rbtree_check = false;
2302 
2303 	init_waitqueue_head(&dcc->discard_wait_queue);
2304 	SM_I(sbi)->dcc_info = dcc;
2305 init_thread:
2306 	err = f2fs_start_discard_thread(sbi);
2307 	if (err) {
2308 		kfree(dcc);
2309 		SM_I(sbi)->dcc_info = NULL;
2310 	}
2311 
2312 	return err;
2313 }
2314 
2315 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2316 {
2317 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2318 
2319 	if (!dcc)
2320 		return;
2321 
2322 	f2fs_stop_discard_thread(sbi);
2323 
2324 	/*
2325 	 * Recovery can cache discard commands, so in error path of
2326 	 * fill_super(), it needs to give a chance to handle them.
2327 	 */
2328 	f2fs_issue_discard_timeout(sbi);
2329 
2330 	kfree(dcc);
2331 	SM_I(sbi)->dcc_info = NULL;
2332 }
2333 
2334 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2335 {
2336 	struct sit_info *sit_i = SIT_I(sbi);
2337 
2338 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2339 		sit_i->dirty_sentries++;
2340 		return false;
2341 	}
2342 
2343 	return true;
2344 }
2345 
2346 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2347 					unsigned int segno, int modified)
2348 {
2349 	struct seg_entry *se = get_seg_entry(sbi, segno);
2350 
2351 	se->type = type;
2352 	if (modified)
2353 		__mark_sit_entry_dirty(sbi, segno);
2354 }
2355 
2356 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2357 								block_t blkaddr)
2358 {
2359 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
2360 
2361 	if (segno == NULL_SEGNO)
2362 		return 0;
2363 	return get_seg_entry(sbi, segno)->mtime;
2364 }
2365 
2366 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2367 						unsigned long long old_mtime)
2368 {
2369 	struct seg_entry *se;
2370 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
2371 	unsigned long long ctime = get_mtime(sbi, false);
2372 	unsigned long long mtime = old_mtime ? old_mtime : ctime;
2373 
2374 	if (segno == NULL_SEGNO)
2375 		return;
2376 
2377 	se = get_seg_entry(sbi, segno);
2378 
2379 	if (!se->mtime)
2380 		se->mtime = mtime;
2381 	else
2382 		se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2383 						se->valid_blocks + 1);
2384 
2385 	if (ctime > SIT_I(sbi)->max_mtime)
2386 		SIT_I(sbi)->max_mtime = ctime;
2387 }
2388 
2389 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2390 {
2391 	struct seg_entry *se;
2392 	unsigned int segno, offset;
2393 	long int new_vblocks;
2394 	bool exist;
2395 #ifdef CONFIG_F2FS_CHECK_FS
2396 	bool mir_exist;
2397 #endif
2398 
2399 	segno = GET_SEGNO(sbi, blkaddr);
2400 
2401 	se = get_seg_entry(sbi, segno);
2402 	new_vblocks = se->valid_blocks + del;
2403 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2404 
2405 	f2fs_bug_on(sbi, (new_vblocks < 0 ||
2406 			(new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2407 
2408 	se->valid_blocks = new_vblocks;
2409 
2410 	/* Update valid block bitmap */
2411 	if (del > 0) {
2412 		exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2413 #ifdef CONFIG_F2FS_CHECK_FS
2414 		mir_exist = f2fs_test_and_set_bit(offset,
2415 						se->cur_valid_map_mir);
2416 		if (unlikely(exist != mir_exist)) {
2417 			f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2418 				 blkaddr, exist);
2419 			f2fs_bug_on(sbi, 1);
2420 		}
2421 #endif
2422 		if (unlikely(exist)) {
2423 			f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2424 				 blkaddr);
2425 			f2fs_bug_on(sbi, 1);
2426 			se->valid_blocks--;
2427 			del = 0;
2428 		}
2429 
2430 		if (f2fs_block_unit_discard(sbi) &&
2431 				!f2fs_test_and_set_bit(offset, se->discard_map))
2432 			sbi->discard_blks--;
2433 
2434 		/*
2435 		 * SSR should never reuse block which is checkpointed
2436 		 * or newly invalidated.
2437 		 */
2438 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2439 			if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2440 				se->ckpt_valid_blocks++;
2441 		}
2442 	} else {
2443 		exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2444 #ifdef CONFIG_F2FS_CHECK_FS
2445 		mir_exist = f2fs_test_and_clear_bit(offset,
2446 						se->cur_valid_map_mir);
2447 		if (unlikely(exist != mir_exist)) {
2448 			f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2449 				 blkaddr, exist);
2450 			f2fs_bug_on(sbi, 1);
2451 		}
2452 #endif
2453 		if (unlikely(!exist)) {
2454 			f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2455 				 blkaddr);
2456 			f2fs_bug_on(sbi, 1);
2457 			se->valid_blocks++;
2458 			del = 0;
2459 		} else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2460 			/*
2461 			 * If checkpoints are off, we must not reuse data that
2462 			 * was used in the previous checkpoint. If it was used
2463 			 * before, we must track that to know how much space we
2464 			 * really have.
2465 			 */
2466 			if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2467 				spin_lock(&sbi->stat_lock);
2468 				sbi->unusable_block_count++;
2469 				spin_unlock(&sbi->stat_lock);
2470 			}
2471 		}
2472 
2473 		if (f2fs_block_unit_discard(sbi) &&
2474 			f2fs_test_and_clear_bit(offset, se->discard_map))
2475 			sbi->discard_blks++;
2476 	}
2477 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2478 		se->ckpt_valid_blocks += del;
2479 
2480 	__mark_sit_entry_dirty(sbi, segno);
2481 
2482 	/* update total number of valid blocks to be written in ckpt area */
2483 	SIT_I(sbi)->written_valid_blocks += del;
2484 
2485 	if (__is_large_section(sbi))
2486 		get_sec_entry(sbi, segno)->valid_blocks += del;
2487 }
2488 
2489 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2490 {
2491 	unsigned int segno = GET_SEGNO(sbi, addr);
2492 	struct sit_info *sit_i = SIT_I(sbi);
2493 
2494 	f2fs_bug_on(sbi, addr == NULL_ADDR);
2495 	if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2496 		return;
2497 
2498 	f2fs_invalidate_internal_cache(sbi, addr);
2499 
2500 	/* add it into sit main buffer */
2501 	down_write(&sit_i->sentry_lock);
2502 
2503 	update_segment_mtime(sbi, addr, 0);
2504 	update_sit_entry(sbi, addr, -1);
2505 
2506 	/* add it into dirty seglist */
2507 	locate_dirty_segment(sbi, segno);
2508 
2509 	up_write(&sit_i->sentry_lock);
2510 }
2511 
2512 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2513 {
2514 	struct sit_info *sit_i = SIT_I(sbi);
2515 	unsigned int segno, offset;
2516 	struct seg_entry *se;
2517 	bool is_cp = false;
2518 
2519 	if (!__is_valid_data_blkaddr(blkaddr))
2520 		return true;
2521 
2522 	down_read(&sit_i->sentry_lock);
2523 
2524 	segno = GET_SEGNO(sbi, blkaddr);
2525 	se = get_seg_entry(sbi, segno);
2526 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2527 
2528 	if (f2fs_test_bit(offset, se->ckpt_valid_map))
2529 		is_cp = true;
2530 
2531 	up_read(&sit_i->sentry_lock);
2532 
2533 	return is_cp;
2534 }
2535 
2536 static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type)
2537 {
2538 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2539 
2540 	if (sbi->ckpt->alloc_type[type] == SSR)
2541 		return sbi->blocks_per_seg;
2542 	return curseg->next_blkoff;
2543 }
2544 
2545 /*
2546  * Calculate the number of current summary pages for writing
2547  */
2548 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2549 {
2550 	int valid_sum_count = 0;
2551 	int i, sum_in_page;
2552 
2553 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2554 		if (sbi->ckpt->alloc_type[i] != SSR && for_ra)
2555 			valid_sum_count +=
2556 				le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2557 		else
2558 			valid_sum_count += f2fs_curseg_valid_blocks(sbi, i);
2559 	}
2560 
2561 	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2562 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2563 	if (valid_sum_count <= sum_in_page)
2564 		return 1;
2565 	else if ((valid_sum_count - sum_in_page) <=
2566 		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2567 		return 2;
2568 	return 3;
2569 }
2570 
2571 /*
2572  * Caller should put this summary page
2573  */
2574 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2575 {
2576 	if (unlikely(f2fs_cp_error(sbi)))
2577 		return ERR_PTR(-EIO);
2578 	return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2579 }
2580 
2581 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2582 					void *src, block_t blk_addr)
2583 {
2584 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2585 
2586 	memcpy(page_address(page), src, PAGE_SIZE);
2587 	set_page_dirty(page);
2588 	f2fs_put_page(page, 1);
2589 }
2590 
2591 static void write_sum_page(struct f2fs_sb_info *sbi,
2592 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
2593 {
2594 	f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2595 }
2596 
2597 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2598 						int type, block_t blk_addr)
2599 {
2600 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2601 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2602 	struct f2fs_summary_block *src = curseg->sum_blk;
2603 	struct f2fs_summary_block *dst;
2604 
2605 	dst = (struct f2fs_summary_block *)page_address(page);
2606 	memset(dst, 0, PAGE_SIZE);
2607 
2608 	mutex_lock(&curseg->curseg_mutex);
2609 
2610 	down_read(&curseg->journal_rwsem);
2611 	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2612 	up_read(&curseg->journal_rwsem);
2613 
2614 	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2615 	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2616 
2617 	mutex_unlock(&curseg->curseg_mutex);
2618 
2619 	set_page_dirty(page);
2620 	f2fs_put_page(page, 1);
2621 }
2622 
2623 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2624 				struct curseg_info *curseg, int type)
2625 {
2626 	unsigned int segno = curseg->segno + 1;
2627 	struct free_segmap_info *free_i = FREE_I(sbi);
2628 
2629 	if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2630 		return !test_bit(segno, free_i->free_segmap);
2631 	return 0;
2632 }
2633 
2634 /*
2635  * Find a new segment from the free segments bitmap to right order
2636  * This function should be returned with success, otherwise BUG
2637  */
2638 static void get_new_segment(struct f2fs_sb_info *sbi,
2639 			unsigned int *newseg, bool new_sec, int dir)
2640 {
2641 	struct free_segmap_info *free_i = FREE_I(sbi);
2642 	unsigned int segno, secno, zoneno;
2643 	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2644 	unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2645 	unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2646 	unsigned int left_start = hint;
2647 	bool init = true;
2648 	int go_left = 0;
2649 	int i;
2650 
2651 	spin_lock(&free_i->segmap_lock);
2652 
2653 	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2654 		segno = find_next_zero_bit(free_i->free_segmap,
2655 			GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2656 		if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2657 			goto got_it;
2658 	}
2659 find_other_zone:
2660 	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2661 	if (secno >= MAIN_SECS(sbi)) {
2662 		if (dir == ALLOC_RIGHT) {
2663 			secno = find_first_zero_bit(free_i->free_secmap,
2664 							MAIN_SECS(sbi));
2665 			f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2666 		} else {
2667 			go_left = 1;
2668 			left_start = hint - 1;
2669 		}
2670 	}
2671 	if (go_left == 0)
2672 		goto skip_left;
2673 
2674 	while (test_bit(left_start, free_i->free_secmap)) {
2675 		if (left_start > 0) {
2676 			left_start--;
2677 			continue;
2678 		}
2679 		left_start = find_first_zero_bit(free_i->free_secmap,
2680 							MAIN_SECS(sbi));
2681 		f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2682 		break;
2683 	}
2684 	secno = left_start;
2685 skip_left:
2686 	segno = GET_SEG_FROM_SEC(sbi, secno);
2687 	zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2688 
2689 	/* give up on finding another zone */
2690 	if (!init)
2691 		goto got_it;
2692 	if (sbi->secs_per_zone == 1)
2693 		goto got_it;
2694 	if (zoneno == old_zoneno)
2695 		goto got_it;
2696 	if (dir == ALLOC_LEFT) {
2697 		if (!go_left && zoneno + 1 >= total_zones)
2698 			goto got_it;
2699 		if (go_left && zoneno == 0)
2700 			goto got_it;
2701 	}
2702 	for (i = 0; i < NR_CURSEG_TYPE; i++)
2703 		if (CURSEG_I(sbi, i)->zone == zoneno)
2704 			break;
2705 
2706 	if (i < NR_CURSEG_TYPE) {
2707 		/* zone is in user, try another */
2708 		if (go_left)
2709 			hint = zoneno * sbi->secs_per_zone - 1;
2710 		else if (zoneno + 1 >= total_zones)
2711 			hint = 0;
2712 		else
2713 			hint = (zoneno + 1) * sbi->secs_per_zone;
2714 		init = false;
2715 		goto find_other_zone;
2716 	}
2717 got_it:
2718 	/* set it as dirty segment in free segmap */
2719 	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2720 	__set_inuse(sbi, segno);
2721 	*newseg = segno;
2722 	spin_unlock(&free_i->segmap_lock);
2723 }
2724 
2725 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2726 {
2727 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2728 	struct summary_footer *sum_footer;
2729 	unsigned short seg_type = curseg->seg_type;
2730 
2731 	curseg->inited = true;
2732 	curseg->segno = curseg->next_segno;
2733 	curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2734 	curseg->next_blkoff = 0;
2735 	curseg->next_segno = NULL_SEGNO;
2736 
2737 	sum_footer = &(curseg->sum_blk->footer);
2738 	memset(sum_footer, 0, sizeof(struct summary_footer));
2739 
2740 	sanity_check_seg_type(sbi, seg_type);
2741 
2742 	if (IS_DATASEG(seg_type))
2743 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2744 	if (IS_NODESEG(seg_type))
2745 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2746 	__set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2747 }
2748 
2749 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2750 {
2751 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2752 	unsigned short seg_type = curseg->seg_type;
2753 
2754 	sanity_check_seg_type(sbi, seg_type);
2755 	if (f2fs_need_rand_seg(sbi))
2756 		return get_random_u32_below(MAIN_SECS(sbi) * sbi->segs_per_sec);
2757 
2758 	/* if segs_per_sec is large than 1, we need to keep original policy. */
2759 	if (__is_large_section(sbi))
2760 		return curseg->segno;
2761 
2762 	/* inmem log may not locate on any segment after mount */
2763 	if (!curseg->inited)
2764 		return 0;
2765 
2766 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2767 		return 0;
2768 
2769 	if (test_opt(sbi, NOHEAP) &&
2770 		(seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)))
2771 		return 0;
2772 
2773 	if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2774 		return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2775 
2776 	/* find segments from 0 to reuse freed segments */
2777 	if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2778 		return 0;
2779 
2780 	return curseg->segno;
2781 }
2782 
2783 /*
2784  * Allocate a current working segment.
2785  * This function always allocates a free segment in LFS manner.
2786  */
2787 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2788 {
2789 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2790 	unsigned short seg_type = curseg->seg_type;
2791 	unsigned int segno = curseg->segno;
2792 	int dir = ALLOC_LEFT;
2793 
2794 	if (curseg->inited)
2795 		write_sum_page(sbi, curseg->sum_blk,
2796 				GET_SUM_BLOCK(sbi, segno));
2797 	if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA)
2798 		dir = ALLOC_RIGHT;
2799 
2800 	if (test_opt(sbi, NOHEAP))
2801 		dir = ALLOC_RIGHT;
2802 
2803 	segno = __get_next_segno(sbi, type);
2804 	get_new_segment(sbi, &segno, new_sec, dir);
2805 	curseg->next_segno = segno;
2806 	reset_curseg(sbi, type, 1);
2807 	curseg->alloc_type = LFS;
2808 	if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
2809 		curseg->fragment_remained_chunk =
2810 				get_random_u32_inclusive(1, sbi->max_fragment_chunk);
2811 }
2812 
2813 static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2814 					int segno, block_t start)
2815 {
2816 	struct seg_entry *se = get_seg_entry(sbi, segno);
2817 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2818 	unsigned long *target_map = SIT_I(sbi)->tmp_map;
2819 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2820 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2821 	int i;
2822 
2823 	for (i = 0; i < entries; i++)
2824 		target_map[i] = ckpt_map[i] | cur_map[i];
2825 
2826 	return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2827 }
2828 
2829 static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi,
2830 		struct curseg_info *seg)
2831 {
2832 	return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1);
2833 }
2834 
2835 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2836 {
2837 	return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg;
2838 }
2839 
2840 /*
2841  * This function always allocates a used segment(from dirty seglist) by SSR
2842  * manner, so it should recover the existing segment information of valid blocks
2843  */
2844 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2845 {
2846 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2847 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2848 	unsigned int new_segno = curseg->next_segno;
2849 	struct f2fs_summary_block *sum_node;
2850 	struct page *sum_page;
2851 
2852 	write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, curseg->segno));
2853 
2854 	__set_test_and_inuse(sbi, new_segno);
2855 
2856 	mutex_lock(&dirty_i->seglist_lock);
2857 	__remove_dirty_segment(sbi, new_segno, PRE);
2858 	__remove_dirty_segment(sbi, new_segno, DIRTY);
2859 	mutex_unlock(&dirty_i->seglist_lock);
2860 
2861 	reset_curseg(sbi, type, 1);
2862 	curseg->alloc_type = SSR;
2863 	curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
2864 
2865 	sum_page = f2fs_get_sum_page(sbi, new_segno);
2866 	if (IS_ERR(sum_page)) {
2867 		/* GC won't be able to use stale summary pages by cp_error */
2868 		memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2869 		return;
2870 	}
2871 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2872 	memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2873 	f2fs_put_page(sum_page, 1);
2874 }
2875 
2876 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2877 				int alloc_mode, unsigned long long age);
2878 
2879 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2880 					int target_type, int alloc_mode,
2881 					unsigned long long age)
2882 {
2883 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2884 
2885 	curseg->seg_type = target_type;
2886 
2887 	if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2888 		struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2889 
2890 		curseg->seg_type = se->type;
2891 		change_curseg(sbi, type);
2892 	} else {
2893 		/* allocate cold segment by default */
2894 		curseg->seg_type = CURSEG_COLD_DATA;
2895 		new_curseg(sbi, type, true);
2896 	}
2897 	stat_inc_seg_type(sbi, curseg);
2898 }
2899 
2900 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
2901 {
2902 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2903 
2904 	if (!sbi->am.atgc_enabled)
2905 		return;
2906 
2907 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
2908 
2909 	mutex_lock(&curseg->curseg_mutex);
2910 	down_write(&SIT_I(sbi)->sentry_lock);
2911 
2912 	get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0);
2913 
2914 	up_write(&SIT_I(sbi)->sentry_lock);
2915 	mutex_unlock(&curseg->curseg_mutex);
2916 
2917 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
2918 
2919 }
2920 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2921 {
2922 	__f2fs_init_atgc_curseg(sbi);
2923 }
2924 
2925 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2926 {
2927 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2928 
2929 	mutex_lock(&curseg->curseg_mutex);
2930 	if (!curseg->inited)
2931 		goto out;
2932 
2933 	if (get_valid_blocks(sbi, curseg->segno, false)) {
2934 		write_sum_page(sbi, curseg->sum_blk,
2935 				GET_SUM_BLOCK(sbi, curseg->segno));
2936 	} else {
2937 		mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2938 		__set_test_and_free(sbi, curseg->segno, true);
2939 		mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2940 	}
2941 out:
2942 	mutex_unlock(&curseg->curseg_mutex);
2943 }
2944 
2945 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
2946 {
2947 	__f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2948 
2949 	if (sbi->am.atgc_enabled)
2950 		__f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2951 }
2952 
2953 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2954 {
2955 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2956 
2957 	mutex_lock(&curseg->curseg_mutex);
2958 	if (!curseg->inited)
2959 		goto out;
2960 	if (get_valid_blocks(sbi, curseg->segno, false))
2961 		goto out;
2962 
2963 	mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2964 	__set_test_and_inuse(sbi, curseg->segno);
2965 	mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2966 out:
2967 	mutex_unlock(&curseg->curseg_mutex);
2968 }
2969 
2970 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
2971 {
2972 	__f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2973 
2974 	if (sbi->am.atgc_enabled)
2975 		__f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2976 }
2977 
2978 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2979 				int alloc_mode, unsigned long long age)
2980 {
2981 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2982 	unsigned segno = NULL_SEGNO;
2983 	unsigned short seg_type = curseg->seg_type;
2984 	int i, cnt;
2985 	bool reversed = false;
2986 
2987 	sanity_check_seg_type(sbi, seg_type);
2988 
2989 	/* f2fs_need_SSR() already forces to do this */
2990 	if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
2991 		curseg->next_segno = segno;
2992 		return 1;
2993 	}
2994 
2995 	/* For node segments, let's do SSR more intensively */
2996 	if (IS_NODESEG(seg_type)) {
2997 		if (seg_type >= CURSEG_WARM_NODE) {
2998 			reversed = true;
2999 			i = CURSEG_COLD_NODE;
3000 		} else {
3001 			i = CURSEG_HOT_NODE;
3002 		}
3003 		cnt = NR_CURSEG_NODE_TYPE;
3004 	} else {
3005 		if (seg_type >= CURSEG_WARM_DATA) {
3006 			reversed = true;
3007 			i = CURSEG_COLD_DATA;
3008 		} else {
3009 			i = CURSEG_HOT_DATA;
3010 		}
3011 		cnt = NR_CURSEG_DATA_TYPE;
3012 	}
3013 
3014 	for (; cnt-- > 0; reversed ? i-- : i++) {
3015 		if (i == seg_type)
3016 			continue;
3017 		if (!f2fs_get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
3018 			curseg->next_segno = segno;
3019 			return 1;
3020 		}
3021 	}
3022 
3023 	/* find valid_blocks=0 in dirty list */
3024 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
3025 		segno = get_free_segment(sbi);
3026 		if (segno != NULL_SEGNO) {
3027 			curseg->next_segno = segno;
3028 			return 1;
3029 		}
3030 	}
3031 	return 0;
3032 }
3033 
3034 static bool need_new_seg(struct f2fs_sb_info *sbi, int type)
3035 {
3036 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3037 
3038 	if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
3039 	    curseg->seg_type == CURSEG_WARM_NODE)
3040 		return true;
3041 	if (curseg->alloc_type == LFS &&
3042 	    is_next_segment_free(sbi, curseg, type) &&
3043 	    likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
3044 		return true;
3045 	if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0))
3046 		return true;
3047 	return false;
3048 }
3049 
3050 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
3051 					unsigned int start, unsigned int end)
3052 {
3053 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3054 	unsigned int segno;
3055 
3056 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3057 	mutex_lock(&curseg->curseg_mutex);
3058 	down_write(&SIT_I(sbi)->sentry_lock);
3059 
3060 	segno = CURSEG_I(sbi, type)->segno;
3061 	if (segno < start || segno > end)
3062 		goto unlock;
3063 
3064 	if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
3065 		change_curseg(sbi, type);
3066 	else
3067 		new_curseg(sbi, type, true);
3068 
3069 	stat_inc_seg_type(sbi, curseg);
3070 
3071 	locate_dirty_segment(sbi, segno);
3072 unlock:
3073 	up_write(&SIT_I(sbi)->sentry_lock);
3074 
3075 	if (segno != curseg->segno)
3076 		f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
3077 			    type, segno, curseg->segno);
3078 
3079 	mutex_unlock(&curseg->curseg_mutex);
3080 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3081 }
3082 
3083 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
3084 						bool new_sec, bool force)
3085 {
3086 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3087 	unsigned int old_segno;
3088 
3089 	if (!force && curseg->inited &&
3090 	    !curseg->next_blkoff &&
3091 	    !get_valid_blocks(sbi, curseg->segno, new_sec) &&
3092 	    !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
3093 		return;
3094 
3095 	old_segno = curseg->segno;
3096 	new_curseg(sbi, type, true);
3097 	stat_inc_seg_type(sbi, curseg);
3098 	locate_dirty_segment(sbi, old_segno);
3099 }
3100 
3101 void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
3102 {
3103 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3104 	down_write(&SIT_I(sbi)->sentry_lock);
3105 	__allocate_new_segment(sbi, type, true, force);
3106 	up_write(&SIT_I(sbi)->sentry_lock);
3107 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3108 }
3109 
3110 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
3111 {
3112 	int i;
3113 
3114 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3115 	down_write(&SIT_I(sbi)->sentry_lock);
3116 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
3117 		__allocate_new_segment(sbi, i, false, false);
3118 	up_write(&SIT_I(sbi)->sentry_lock);
3119 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3120 }
3121 
3122 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
3123 						struct cp_control *cpc)
3124 {
3125 	__u64 trim_start = cpc->trim_start;
3126 	bool has_candidate = false;
3127 
3128 	down_write(&SIT_I(sbi)->sentry_lock);
3129 	for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
3130 		if (add_discard_addrs(sbi, cpc, true)) {
3131 			has_candidate = true;
3132 			break;
3133 		}
3134 	}
3135 	up_write(&SIT_I(sbi)->sentry_lock);
3136 
3137 	cpc->trim_start = trim_start;
3138 	return has_candidate;
3139 }
3140 
3141 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
3142 					struct discard_policy *dpolicy,
3143 					unsigned int start, unsigned int end)
3144 {
3145 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
3146 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
3147 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
3148 	struct discard_cmd *dc;
3149 	struct blk_plug plug;
3150 	int issued;
3151 	unsigned int trimmed = 0;
3152 
3153 next:
3154 	issued = 0;
3155 
3156 	mutex_lock(&dcc->cmd_lock);
3157 	if (unlikely(dcc->rbtree_check))
3158 		f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
3159 
3160 	dc = __lookup_discard_cmd_ret(&dcc->root, start,
3161 				&prev_dc, &next_dc, &insert_p, &insert_parent);
3162 	if (!dc)
3163 		dc = next_dc;
3164 
3165 	blk_start_plug(&plug);
3166 
3167 	while (dc && dc->di.lstart <= end) {
3168 		struct rb_node *node;
3169 		int err = 0;
3170 
3171 		if (dc->di.len < dpolicy->granularity)
3172 			goto skip;
3173 
3174 		if (dc->state != D_PREP) {
3175 			list_move_tail(&dc->list, &dcc->fstrim_list);
3176 			goto skip;
3177 		}
3178 
3179 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3180 
3181 		if (issued >= dpolicy->max_requests) {
3182 			start = dc->di.lstart + dc->di.len;
3183 
3184 			if (err)
3185 				__remove_discard_cmd(sbi, dc);
3186 
3187 			blk_finish_plug(&plug);
3188 			mutex_unlock(&dcc->cmd_lock);
3189 			trimmed += __wait_all_discard_cmd(sbi, NULL);
3190 			f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
3191 			goto next;
3192 		}
3193 skip:
3194 		node = rb_next(&dc->rb_node);
3195 		if (err)
3196 			__remove_discard_cmd(sbi, dc);
3197 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3198 
3199 		if (fatal_signal_pending(current))
3200 			break;
3201 	}
3202 
3203 	blk_finish_plug(&plug);
3204 	mutex_unlock(&dcc->cmd_lock);
3205 
3206 	return trimmed;
3207 }
3208 
3209 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3210 {
3211 	__u64 start = F2FS_BYTES_TO_BLK(range->start);
3212 	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3213 	unsigned int start_segno, end_segno;
3214 	block_t start_block, end_block;
3215 	struct cp_control cpc;
3216 	struct discard_policy dpolicy;
3217 	unsigned long long trimmed = 0;
3218 	int err = 0;
3219 	bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3220 
3221 	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3222 		return -EINVAL;
3223 
3224 	if (end < MAIN_BLKADDR(sbi))
3225 		goto out;
3226 
3227 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3228 		f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3229 		return -EFSCORRUPTED;
3230 	}
3231 
3232 	/* start/end segment number in main_area */
3233 	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3234 	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3235 						GET_SEGNO(sbi, end);
3236 	if (need_align) {
3237 		start_segno = rounddown(start_segno, sbi->segs_per_sec);
3238 		end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
3239 	}
3240 
3241 	cpc.reason = CP_DISCARD;
3242 	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3243 	cpc.trim_start = start_segno;
3244 	cpc.trim_end = end_segno;
3245 
3246 	if (sbi->discard_blks == 0)
3247 		goto out;
3248 
3249 	f2fs_down_write(&sbi->gc_lock);
3250 	stat_inc_cp_call_count(sbi, TOTAL_CALL);
3251 	err = f2fs_write_checkpoint(sbi, &cpc);
3252 	f2fs_up_write(&sbi->gc_lock);
3253 	if (err)
3254 		goto out;
3255 
3256 	/*
3257 	 * We filed discard candidates, but actually we don't need to wait for
3258 	 * all of them, since they'll be issued in idle time along with runtime
3259 	 * discard option. User configuration looks like using runtime discard
3260 	 * or periodic fstrim instead of it.
3261 	 */
3262 	if (f2fs_realtime_discard_enable(sbi))
3263 		goto out;
3264 
3265 	start_block = START_BLOCK(sbi, start_segno);
3266 	end_block = START_BLOCK(sbi, end_segno + 1);
3267 
3268 	__init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3269 	trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3270 					start_block, end_block);
3271 
3272 	trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3273 					start_block, end_block);
3274 out:
3275 	if (!err)
3276 		range->len = F2FS_BLK_TO_BYTES(trimmed);
3277 	return err;
3278 }
3279 
3280 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3281 {
3282 	switch (hint) {
3283 	case WRITE_LIFE_SHORT:
3284 		return CURSEG_HOT_DATA;
3285 	case WRITE_LIFE_EXTREME:
3286 		return CURSEG_COLD_DATA;
3287 	default:
3288 		return CURSEG_WARM_DATA;
3289 	}
3290 }
3291 
3292 static int __get_segment_type_2(struct f2fs_io_info *fio)
3293 {
3294 	if (fio->type == DATA)
3295 		return CURSEG_HOT_DATA;
3296 	else
3297 		return CURSEG_HOT_NODE;
3298 }
3299 
3300 static int __get_segment_type_4(struct f2fs_io_info *fio)
3301 {
3302 	if (fio->type == DATA) {
3303 		struct inode *inode = fio->page->mapping->host;
3304 
3305 		if (S_ISDIR(inode->i_mode))
3306 			return CURSEG_HOT_DATA;
3307 		else
3308 			return CURSEG_COLD_DATA;
3309 	} else {
3310 		if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3311 			return CURSEG_WARM_NODE;
3312 		else
3313 			return CURSEG_COLD_NODE;
3314 	}
3315 }
3316 
3317 static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs)
3318 {
3319 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3320 	struct extent_info ei = {};
3321 
3322 	if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) {
3323 		if (!ei.age)
3324 			return NO_CHECK_TYPE;
3325 		if (ei.age <= sbi->hot_data_age_threshold)
3326 			return CURSEG_HOT_DATA;
3327 		if (ei.age <= sbi->warm_data_age_threshold)
3328 			return CURSEG_WARM_DATA;
3329 		return CURSEG_COLD_DATA;
3330 	}
3331 	return NO_CHECK_TYPE;
3332 }
3333 
3334 static int __get_segment_type_6(struct f2fs_io_info *fio)
3335 {
3336 	if (fio->type == DATA) {
3337 		struct inode *inode = fio->page->mapping->host;
3338 		int type;
3339 
3340 		if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
3341 			return CURSEG_COLD_DATA_PINNED;
3342 
3343 		if (page_private_gcing(fio->page)) {
3344 			if (fio->sbi->am.atgc_enabled &&
3345 				(fio->io_type == FS_DATA_IO) &&
3346 				(fio->sbi->gc_mode != GC_URGENT_HIGH))
3347 				return CURSEG_ALL_DATA_ATGC;
3348 			else
3349 				return CURSEG_COLD_DATA;
3350 		}
3351 		if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3352 			return CURSEG_COLD_DATA;
3353 
3354 		type = __get_age_segment_type(inode, fio->page->index);
3355 		if (type != NO_CHECK_TYPE)
3356 			return type;
3357 
3358 		if (file_is_hot(inode) ||
3359 				is_inode_flag_set(inode, FI_HOT_DATA) ||
3360 				f2fs_is_cow_file(inode))
3361 			return CURSEG_HOT_DATA;
3362 		return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3363 	} else {
3364 		if (IS_DNODE(fio->page))
3365 			return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3366 						CURSEG_HOT_NODE;
3367 		return CURSEG_COLD_NODE;
3368 	}
3369 }
3370 
3371 static int __get_segment_type(struct f2fs_io_info *fio)
3372 {
3373 	int type = 0;
3374 
3375 	switch (F2FS_OPTION(fio->sbi).active_logs) {
3376 	case 2:
3377 		type = __get_segment_type_2(fio);
3378 		break;
3379 	case 4:
3380 		type = __get_segment_type_4(fio);
3381 		break;
3382 	case 6:
3383 		type = __get_segment_type_6(fio);
3384 		break;
3385 	default:
3386 		f2fs_bug_on(fio->sbi, true);
3387 	}
3388 
3389 	if (IS_HOT(type))
3390 		fio->temp = HOT;
3391 	else if (IS_WARM(type))
3392 		fio->temp = WARM;
3393 	else
3394 		fio->temp = COLD;
3395 	return type;
3396 }
3397 
3398 static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi,
3399 		struct curseg_info *seg)
3400 {
3401 	/* To allocate block chunks in different sizes, use random number */
3402 	if (--seg->fragment_remained_chunk > 0)
3403 		return;
3404 
3405 	seg->fragment_remained_chunk =
3406 		get_random_u32_inclusive(1, sbi->max_fragment_chunk);
3407 	seg->next_blkoff +=
3408 		get_random_u32_inclusive(1, sbi->max_fragment_hole);
3409 }
3410 
3411 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3412 		block_t old_blkaddr, block_t *new_blkaddr,
3413 		struct f2fs_summary *sum, int type,
3414 		struct f2fs_io_info *fio)
3415 {
3416 	struct sit_info *sit_i = SIT_I(sbi);
3417 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3418 	unsigned long long old_mtime;
3419 	bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3420 	struct seg_entry *se = NULL;
3421 	bool segment_full = false;
3422 
3423 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3424 
3425 	mutex_lock(&curseg->curseg_mutex);
3426 	down_write(&sit_i->sentry_lock);
3427 
3428 	if (from_gc) {
3429 		f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3430 		se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3431 		sanity_check_seg_type(sbi, se->type);
3432 		f2fs_bug_on(sbi, IS_NODESEG(se->type));
3433 	}
3434 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3435 
3436 	f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg);
3437 
3438 	f2fs_wait_discard_bio(sbi, *new_blkaddr);
3439 
3440 	curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3441 	if (curseg->alloc_type == SSR) {
3442 		curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg);
3443 	} else {
3444 		curseg->next_blkoff++;
3445 		if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
3446 			f2fs_randomize_chunk(sbi, curseg);
3447 	}
3448 	if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno))
3449 		segment_full = true;
3450 	stat_inc_block_count(sbi, curseg);
3451 
3452 	if (from_gc) {
3453 		old_mtime = get_segment_mtime(sbi, old_blkaddr);
3454 	} else {
3455 		update_segment_mtime(sbi, old_blkaddr, 0);
3456 		old_mtime = 0;
3457 	}
3458 	update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3459 
3460 	/*
3461 	 * SIT information should be updated before segment allocation,
3462 	 * since SSR needs latest valid block information.
3463 	 */
3464 	update_sit_entry(sbi, *new_blkaddr, 1);
3465 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3466 		update_sit_entry(sbi, old_blkaddr, -1);
3467 
3468 	/*
3469 	 * If the current segment is full, flush it out and replace it with a
3470 	 * new segment.
3471 	 */
3472 	if (segment_full) {
3473 		if (from_gc) {
3474 			get_atssr_segment(sbi, type, se->type,
3475 						AT_SSR, se->mtime);
3476 		} else {
3477 			if (need_new_seg(sbi, type))
3478 				new_curseg(sbi, type, false);
3479 			else
3480 				change_curseg(sbi, type);
3481 			stat_inc_seg_type(sbi, curseg);
3482 		}
3483 	}
3484 	/*
3485 	 * segment dirty status should be updated after segment allocation,
3486 	 * so we just need to update status only one time after previous
3487 	 * segment being closed.
3488 	 */
3489 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3490 	locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3491 
3492 	if (IS_DATASEG(curseg->seg_type))
3493 		atomic64_inc(&sbi->allocated_data_blocks);
3494 
3495 	up_write(&sit_i->sentry_lock);
3496 
3497 	if (page && IS_NODESEG(curseg->seg_type)) {
3498 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3499 
3500 		f2fs_inode_chksum_set(sbi, page);
3501 	}
3502 
3503 	if (fio) {
3504 		struct f2fs_bio_info *io;
3505 
3506 		if (F2FS_IO_ALIGNED(sbi))
3507 			fio->retry = 0;
3508 
3509 		INIT_LIST_HEAD(&fio->list);
3510 		fio->in_list = 1;
3511 		io = sbi->write_io[fio->type] + fio->temp;
3512 		spin_lock(&io->io_lock);
3513 		list_add_tail(&fio->list, &io->io_list);
3514 		spin_unlock(&io->io_lock);
3515 	}
3516 
3517 	mutex_unlock(&curseg->curseg_mutex);
3518 
3519 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3520 }
3521 
3522 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino,
3523 					block_t blkaddr, unsigned int blkcnt)
3524 {
3525 	if (!f2fs_is_multi_device(sbi))
3526 		return;
3527 
3528 	while (1) {
3529 		unsigned int devidx = f2fs_target_device_index(sbi, blkaddr);
3530 		unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1;
3531 
3532 		/* update device state for fsync */
3533 		f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO);
3534 
3535 		/* update device state for checkpoint */
3536 		if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3537 			spin_lock(&sbi->dev_lock);
3538 			f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3539 			spin_unlock(&sbi->dev_lock);
3540 		}
3541 
3542 		if (blkcnt <= blks)
3543 			break;
3544 		blkcnt -= blks;
3545 		blkaddr += blks;
3546 	}
3547 }
3548 
3549 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3550 {
3551 	int type = __get_segment_type(fio);
3552 	bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3553 
3554 	if (keep_order)
3555 		f2fs_down_read(&fio->sbi->io_order_lock);
3556 reallocate:
3557 	f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3558 			&fio->new_blkaddr, sum, type, fio);
3559 	if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3560 		f2fs_invalidate_internal_cache(fio->sbi, fio->old_blkaddr);
3561 
3562 	/* writeout dirty page into bdev */
3563 	f2fs_submit_page_write(fio);
3564 	if (fio->retry) {
3565 		fio->old_blkaddr = fio->new_blkaddr;
3566 		goto reallocate;
3567 	}
3568 
3569 	f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1);
3570 
3571 	if (keep_order)
3572 		f2fs_up_read(&fio->sbi->io_order_lock);
3573 }
3574 
3575 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3576 					enum iostat_type io_type)
3577 {
3578 	struct f2fs_io_info fio = {
3579 		.sbi = sbi,
3580 		.type = META,
3581 		.temp = HOT,
3582 		.op = REQ_OP_WRITE,
3583 		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3584 		.old_blkaddr = page->index,
3585 		.new_blkaddr = page->index,
3586 		.page = page,
3587 		.encrypted_page = NULL,
3588 		.in_list = 0,
3589 	};
3590 
3591 	if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3592 		fio.op_flags &= ~REQ_META;
3593 
3594 	set_page_writeback(page);
3595 	f2fs_submit_page_write(&fio);
3596 
3597 	stat_inc_meta_count(sbi, page->index);
3598 	f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE);
3599 }
3600 
3601 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3602 {
3603 	struct f2fs_summary sum;
3604 
3605 	set_summary(&sum, nid, 0, 0);
3606 	do_write_page(&sum, fio);
3607 
3608 	f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE);
3609 }
3610 
3611 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3612 					struct f2fs_io_info *fio)
3613 {
3614 	struct f2fs_sb_info *sbi = fio->sbi;
3615 	struct f2fs_summary sum;
3616 
3617 	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3618 	if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO)
3619 		f2fs_update_age_extent_cache(dn);
3620 	set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3621 	do_write_page(&sum, fio);
3622 	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3623 
3624 	f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE);
3625 }
3626 
3627 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3628 {
3629 	int err;
3630 	struct f2fs_sb_info *sbi = fio->sbi;
3631 	unsigned int segno;
3632 
3633 	fio->new_blkaddr = fio->old_blkaddr;
3634 	/* i/o temperature is needed for passing down write hints */
3635 	__get_segment_type(fio);
3636 
3637 	segno = GET_SEGNO(sbi, fio->new_blkaddr);
3638 
3639 	if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3640 		set_sbi_flag(sbi, SBI_NEED_FSCK);
3641 		f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3642 			  __func__, segno);
3643 		err = -EFSCORRUPTED;
3644 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
3645 		goto drop_bio;
3646 	}
3647 
3648 	if (f2fs_cp_error(sbi)) {
3649 		err = -EIO;
3650 		goto drop_bio;
3651 	}
3652 
3653 	if (fio->post_read)
3654 		f2fs_truncate_meta_inode_pages(sbi, fio->new_blkaddr, 1);
3655 
3656 	stat_inc_inplace_blocks(fio->sbi);
3657 
3658 	if (fio->bio && !IS_F2FS_IPU_NOCACHE(sbi))
3659 		err = f2fs_merge_page_bio(fio);
3660 	else
3661 		err = f2fs_submit_page_bio(fio);
3662 	if (!err) {
3663 		f2fs_update_device_state(fio->sbi, fio->ino,
3664 						fio->new_blkaddr, 1);
3665 		f2fs_update_iostat(fio->sbi, fio->page->mapping->host,
3666 						fio->io_type, F2FS_BLKSIZE);
3667 	}
3668 
3669 	return err;
3670 drop_bio:
3671 	if (fio->bio && *(fio->bio)) {
3672 		struct bio *bio = *(fio->bio);
3673 
3674 		bio->bi_status = BLK_STS_IOERR;
3675 		bio_endio(bio);
3676 		*(fio->bio) = NULL;
3677 	}
3678 	return err;
3679 }
3680 
3681 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3682 						unsigned int segno)
3683 {
3684 	int i;
3685 
3686 	for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3687 		if (CURSEG_I(sbi, i)->segno == segno)
3688 			break;
3689 	}
3690 	return i;
3691 }
3692 
3693 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3694 				block_t old_blkaddr, block_t new_blkaddr,
3695 				bool recover_curseg, bool recover_newaddr,
3696 				bool from_gc)
3697 {
3698 	struct sit_info *sit_i = SIT_I(sbi);
3699 	struct curseg_info *curseg;
3700 	unsigned int segno, old_cursegno;
3701 	struct seg_entry *se;
3702 	int type;
3703 	unsigned short old_blkoff;
3704 	unsigned char old_alloc_type;
3705 
3706 	segno = GET_SEGNO(sbi, new_blkaddr);
3707 	se = get_seg_entry(sbi, segno);
3708 	type = se->type;
3709 
3710 	f2fs_down_write(&SM_I(sbi)->curseg_lock);
3711 
3712 	if (!recover_curseg) {
3713 		/* for recovery flow */
3714 		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3715 			if (old_blkaddr == NULL_ADDR)
3716 				type = CURSEG_COLD_DATA;
3717 			else
3718 				type = CURSEG_WARM_DATA;
3719 		}
3720 	} else {
3721 		if (IS_CURSEG(sbi, segno)) {
3722 			/* se->type is volatile as SSR allocation */
3723 			type = __f2fs_get_curseg(sbi, segno);
3724 			f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3725 		} else {
3726 			type = CURSEG_WARM_DATA;
3727 		}
3728 	}
3729 
3730 	f2fs_bug_on(sbi, !IS_DATASEG(type));
3731 	curseg = CURSEG_I(sbi, type);
3732 
3733 	mutex_lock(&curseg->curseg_mutex);
3734 	down_write(&sit_i->sentry_lock);
3735 
3736 	old_cursegno = curseg->segno;
3737 	old_blkoff = curseg->next_blkoff;
3738 	old_alloc_type = curseg->alloc_type;
3739 
3740 	/* change the current segment */
3741 	if (segno != curseg->segno) {
3742 		curseg->next_segno = segno;
3743 		change_curseg(sbi, type);
3744 	}
3745 
3746 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3747 	curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3748 
3749 	if (!recover_curseg || recover_newaddr) {
3750 		if (!from_gc)
3751 			update_segment_mtime(sbi, new_blkaddr, 0);
3752 		update_sit_entry(sbi, new_blkaddr, 1);
3753 	}
3754 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3755 		f2fs_invalidate_internal_cache(sbi, old_blkaddr);
3756 		if (!from_gc)
3757 			update_segment_mtime(sbi, old_blkaddr, 0);
3758 		update_sit_entry(sbi, old_blkaddr, -1);
3759 	}
3760 
3761 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3762 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3763 
3764 	locate_dirty_segment(sbi, old_cursegno);
3765 
3766 	if (recover_curseg) {
3767 		if (old_cursegno != curseg->segno) {
3768 			curseg->next_segno = old_cursegno;
3769 			change_curseg(sbi, type);
3770 		}
3771 		curseg->next_blkoff = old_blkoff;
3772 		curseg->alloc_type = old_alloc_type;
3773 	}
3774 
3775 	up_write(&sit_i->sentry_lock);
3776 	mutex_unlock(&curseg->curseg_mutex);
3777 	f2fs_up_write(&SM_I(sbi)->curseg_lock);
3778 }
3779 
3780 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3781 				block_t old_addr, block_t new_addr,
3782 				unsigned char version, bool recover_curseg,
3783 				bool recover_newaddr)
3784 {
3785 	struct f2fs_summary sum;
3786 
3787 	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3788 
3789 	f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3790 					recover_curseg, recover_newaddr, false);
3791 
3792 	f2fs_update_data_blkaddr(dn, new_addr);
3793 }
3794 
3795 void f2fs_wait_on_page_writeback(struct page *page,
3796 				enum page_type type, bool ordered, bool locked)
3797 {
3798 	if (PageWriteback(page)) {
3799 		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3800 
3801 		/* submit cached LFS IO */
3802 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3803 		/* submit cached IPU IO */
3804 		f2fs_submit_merged_ipu_write(sbi, NULL, page);
3805 		if (ordered) {
3806 			wait_on_page_writeback(page);
3807 			f2fs_bug_on(sbi, locked && PageWriteback(page));
3808 		} else {
3809 			wait_for_stable_page(page);
3810 		}
3811 	}
3812 }
3813 
3814 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3815 {
3816 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3817 	struct page *cpage;
3818 
3819 	if (!f2fs_post_read_required(inode))
3820 		return;
3821 
3822 	if (!__is_valid_data_blkaddr(blkaddr))
3823 		return;
3824 
3825 	cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3826 	if (cpage) {
3827 		f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3828 		f2fs_put_page(cpage, 1);
3829 	}
3830 }
3831 
3832 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3833 								block_t len)
3834 {
3835 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3836 	block_t i;
3837 
3838 	if (!f2fs_post_read_required(inode))
3839 		return;
3840 
3841 	for (i = 0; i < len; i++)
3842 		f2fs_wait_on_block_writeback(inode, blkaddr + i);
3843 
3844 	f2fs_truncate_meta_inode_pages(sbi, blkaddr, len);
3845 }
3846 
3847 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3848 {
3849 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3850 	struct curseg_info *seg_i;
3851 	unsigned char *kaddr;
3852 	struct page *page;
3853 	block_t start;
3854 	int i, j, offset;
3855 
3856 	start = start_sum_block(sbi);
3857 
3858 	page = f2fs_get_meta_page(sbi, start++);
3859 	if (IS_ERR(page))
3860 		return PTR_ERR(page);
3861 	kaddr = (unsigned char *)page_address(page);
3862 
3863 	/* Step 1: restore nat cache */
3864 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3865 	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3866 
3867 	/* Step 2: restore sit cache */
3868 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3869 	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3870 	offset = 2 * SUM_JOURNAL_SIZE;
3871 
3872 	/* Step 3: restore summary entries */
3873 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3874 		unsigned short blk_off;
3875 		unsigned int segno;
3876 
3877 		seg_i = CURSEG_I(sbi, i);
3878 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3879 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3880 		seg_i->next_segno = segno;
3881 		reset_curseg(sbi, i, 0);
3882 		seg_i->alloc_type = ckpt->alloc_type[i];
3883 		seg_i->next_blkoff = blk_off;
3884 
3885 		if (seg_i->alloc_type == SSR)
3886 			blk_off = sbi->blocks_per_seg;
3887 
3888 		for (j = 0; j < blk_off; j++) {
3889 			struct f2fs_summary *s;
3890 
3891 			s = (struct f2fs_summary *)(kaddr + offset);
3892 			seg_i->sum_blk->entries[j] = *s;
3893 			offset += SUMMARY_SIZE;
3894 			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3895 						SUM_FOOTER_SIZE)
3896 				continue;
3897 
3898 			f2fs_put_page(page, 1);
3899 			page = NULL;
3900 
3901 			page = f2fs_get_meta_page(sbi, start++);
3902 			if (IS_ERR(page))
3903 				return PTR_ERR(page);
3904 			kaddr = (unsigned char *)page_address(page);
3905 			offset = 0;
3906 		}
3907 	}
3908 	f2fs_put_page(page, 1);
3909 	return 0;
3910 }
3911 
3912 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3913 {
3914 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3915 	struct f2fs_summary_block *sum;
3916 	struct curseg_info *curseg;
3917 	struct page *new;
3918 	unsigned short blk_off;
3919 	unsigned int segno = 0;
3920 	block_t blk_addr = 0;
3921 	int err = 0;
3922 
3923 	/* get segment number and block addr */
3924 	if (IS_DATASEG(type)) {
3925 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3926 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3927 							CURSEG_HOT_DATA]);
3928 		if (__exist_node_summaries(sbi))
3929 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
3930 		else
3931 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3932 	} else {
3933 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
3934 							CURSEG_HOT_NODE]);
3935 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3936 							CURSEG_HOT_NODE]);
3937 		if (__exist_node_summaries(sbi))
3938 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3939 							type - CURSEG_HOT_NODE);
3940 		else
3941 			blk_addr = GET_SUM_BLOCK(sbi, segno);
3942 	}
3943 
3944 	new = f2fs_get_meta_page(sbi, blk_addr);
3945 	if (IS_ERR(new))
3946 		return PTR_ERR(new);
3947 	sum = (struct f2fs_summary_block *)page_address(new);
3948 
3949 	if (IS_NODESEG(type)) {
3950 		if (__exist_node_summaries(sbi)) {
3951 			struct f2fs_summary *ns = &sum->entries[0];
3952 			int i;
3953 
3954 			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3955 				ns->version = 0;
3956 				ns->ofs_in_node = 0;
3957 			}
3958 		} else {
3959 			err = f2fs_restore_node_summary(sbi, segno, sum);
3960 			if (err)
3961 				goto out;
3962 		}
3963 	}
3964 
3965 	/* set uncompleted segment to curseg */
3966 	curseg = CURSEG_I(sbi, type);
3967 	mutex_lock(&curseg->curseg_mutex);
3968 
3969 	/* update journal info */
3970 	down_write(&curseg->journal_rwsem);
3971 	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3972 	up_write(&curseg->journal_rwsem);
3973 
3974 	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3975 	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3976 	curseg->next_segno = segno;
3977 	reset_curseg(sbi, type, 0);
3978 	curseg->alloc_type = ckpt->alloc_type[type];
3979 	curseg->next_blkoff = blk_off;
3980 	mutex_unlock(&curseg->curseg_mutex);
3981 out:
3982 	f2fs_put_page(new, 1);
3983 	return err;
3984 }
3985 
3986 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3987 {
3988 	struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3989 	struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3990 	int type = CURSEG_HOT_DATA;
3991 	int err;
3992 
3993 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3994 		int npages = f2fs_npages_for_summary_flush(sbi, true);
3995 
3996 		if (npages >= 2)
3997 			f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3998 							META_CP, true);
3999 
4000 		/* restore for compacted data summary */
4001 		err = read_compacted_summaries(sbi);
4002 		if (err)
4003 			return err;
4004 		type = CURSEG_HOT_NODE;
4005 	}
4006 
4007 	if (__exist_node_summaries(sbi))
4008 		f2fs_ra_meta_pages(sbi,
4009 				sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
4010 				NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
4011 
4012 	for (; type <= CURSEG_COLD_NODE; type++) {
4013 		err = read_normal_summaries(sbi, type);
4014 		if (err)
4015 			return err;
4016 	}
4017 
4018 	/* sanity check for summary blocks */
4019 	if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
4020 			sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
4021 		f2fs_err(sbi, "invalid journal entries nats %u sits %u",
4022 			 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
4023 		return -EINVAL;
4024 	}
4025 
4026 	return 0;
4027 }
4028 
4029 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
4030 {
4031 	struct page *page;
4032 	unsigned char *kaddr;
4033 	struct f2fs_summary *summary;
4034 	struct curseg_info *seg_i;
4035 	int written_size = 0;
4036 	int i, j;
4037 
4038 	page = f2fs_grab_meta_page(sbi, blkaddr++);
4039 	kaddr = (unsigned char *)page_address(page);
4040 	memset(kaddr, 0, PAGE_SIZE);
4041 
4042 	/* Step 1: write nat cache */
4043 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
4044 	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
4045 	written_size += SUM_JOURNAL_SIZE;
4046 
4047 	/* Step 2: write sit cache */
4048 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
4049 	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
4050 	written_size += SUM_JOURNAL_SIZE;
4051 
4052 	/* Step 3: write summary entries */
4053 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
4054 		seg_i = CURSEG_I(sbi, i);
4055 		for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); j++) {
4056 			if (!page) {
4057 				page = f2fs_grab_meta_page(sbi, blkaddr++);
4058 				kaddr = (unsigned char *)page_address(page);
4059 				memset(kaddr, 0, PAGE_SIZE);
4060 				written_size = 0;
4061 			}
4062 			summary = (struct f2fs_summary *)(kaddr + written_size);
4063 			*summary = seg_i->sum_blk->entries[j];
4064 			written_size += SUMMARY_SIZE;
4065 
4066 			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
4067 							SUM_FOOTER_SIZE)
4068 				continue;
4069 
4070 			set_page_dirty(page);
4071 			f2fs_put_page(page, 1);
4072 			page = NULL;
4073 		}
4074 	}
4075 	if (page) {
4076 		set_page_dirty(page);
4077 		f2fs_put_page(page, 1);
4078 	}
4079 }
4080 
4081 static void write_normal_summaries(struct f2fs_sb_info *sbi,
4082 					block_t blkaddr, int type)
4083 {
4084 	int i, end;
4085 
4086 	if (IS_DATASEG(type))
4087 		end = type + NR_CURSEG_DATA_TYPE;
4088 	else
4089 		end = type + NR_CURSEG_NODE_TYPE;
4090 
4091 	for (i = type; i < end; i++)
4092 		write_current_sum_page(sbi, i, blkaddr + (i - type));
4093 }
4094 
4095 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4096 {
4097 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
4098 		write_compacted_summaries(sbi, start_blk);
4099 	else
4100 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
4101 }
4102 
4103 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4104 {
4105 	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
4106 }
4107 
4108 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
4109 					unsigned int val, int alloc)
4110 {
4111 	int i;
4112 
4113 	if (type == NAT_JOURNAL) {
4114 		for (i = 0; i < nats_in_cursum(journal); i++) {
4115 			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
4116 				return i;
4117 		}
4118 		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
4119 			return update_nats_in_cursum(journal, 1);
4120 	} else if (type == SIT_JOURNAL) {
4121 		for (i = 0; i < sits_in_cursum(journal); i++)
4122 			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
4123 				return i;
4124 		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
4125 			return update_sits_in_cursum(journal, 1);
4126 	}
4127 	return -1;
4128 }
4129 
4130 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
4131 					unsigned int segno)
4132 {
4133 	return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
4134 }
4135 
4136 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
4137 					unsigned int start)
4138 {
4139 	struct sit_info *sit_i = SIT_I(sbi);
4140 	struct page *page;
4141 	pgoff_t src_off, dst_off;
4142 
4143 	src_off = current_sit_addr(sbi, start);
4144 	dst_off = next_sit_addr(sbi, src_off);
4145 
4146 	page = f2fs_grab_meta_page(sbi, dst_off);
4147 	seg_info_to_sit_page(sbi, page, start);
4148 
4149 	set_page_dirty(page);
4150 	set_to_next_sit(sit_i, start);
4151 
4152 	return page;
4153 }
4154 
4155 static struct sit_entry_set *grab_sit_entry_set(void)
4156 {
4157 	struct sit_entry_set *ses =
4158 			f2fs_kmem_cache_alloc(sit_entry_set_slab,
4159 						GFP_NOFS, true, NULL);
4160 
4161 	ses->entry_cnt = 0;
4162 	INIT_LIST_HEAD(&ses->set_list);
4163 	return ses;
4164 }
4165 
4166 static void release_sit_entry_set(struct sit_entry_set *ses)
4167 {
4168 	list_del(&ses->set_list);
4169 	kmem_cache_free(sit_entry_set_slab, ses);
4170 }
4171 
4172 static void adjust_sit_entry_set(struct sit_entry_set *ses,
4173 						struct list_head *head)
4174 {
4175 	struct sit_entry_set *next = ses;
4176 
4177 	if (list_is_last(&ses->set_list, head))
4178 		return;
4179 
4180 	list_for_each_entry_continue(next, head, set_list)
4181 		if (ses->entry_cnt <= next->entry_cnt) {
4182 			list_move_tail(&ses->set_list, &next->set_list);
4183 			return;
4184 		}
4185 
4186 	list_move_tail(&ses->set_list, head);
4187 }
4188 
4189 static void add_sit_entry(unsigned int segno, struct list_head *head)
4190 {
4191 	struct sit_entry_set *ses;
4192 	unsigned int start_segno = START_SEGNO(segno);
4193 
4194 	list_for_each_entry(ses, head, set_list) {
4195 		if (ses->start_segno == start_segno) {
4196 			ses->entry_cnt++;
4197 			adjust_sit_entry_set(ses, head);
4198 			return;
4199 		}
4200 	}
4201 
4202 	ses = grab_sit_entry_set();
4203 
4204 	ses->start_segno = start_segno;
4205 	ses->entry_cnt++;
4206 	list_add(&ses->set_list, head);
4207 }
4208 
4209 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4210 {
4211 	struct f2fs_sm_info *sm_info = SM_I(sbi);
4212 	struct list_head *set_list = &sm_info->sit_entry_set;
4213 	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4214 	unsigned int segno;
4215 
4216 	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4217 		add_sit_entry(segno, set_list);
4218 }
4219 
4220 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4221 {
4222 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4223 	struct f2fs_journal *journal = curseg->journal;
4224 	int i;
4225 
4226 	down_write(&curseg->journal_rwsem);
4227 	for (i = 0; i < sits_in_cursum(journal); i++) {
4228 		unsigned int segno;
4229 		bool dirtied;
4230 
4231 		segno = le32_to_cpu(segno_in_journal(journal, i));
4232 		dirtied = __mark_sit_entry_dirty(sbi, segno);
4233 
4234 		if (!dirtied)
4235 			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4236 	}
4237 	update_sits_in_cursum(journal, -i);
4238 	up_write(&curseg->journal_rwsem);
4239 }
4240 
4241 /*
4242  * CP calls this function, which flushes SIT entries including sit_journal,
4243  * and moves prefree segs to free segs.
4244  */
4245 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4246 {
4247 	struct sit_info *sit_i = SIT_I(sbi);
4248 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4249 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4250 	struct f2fs_journal *journal = curseg->journal;
4251 	struct sit_entry_set *ses, *tmp;
4252 	struct list_head *head = &SM_I(sbi)->sit_entry_set;
4253 	bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4254 	struct seg_entry *se;
4255 
4256 	down_write(&sit_i->sentry_lock);
4257 
4258 	if (!sit_i->dirty_sentries)
4259 		goto out;
4260 
4261 	/*
4262 	 * add and account sit entries of dirty bitmap in sit entry
4263 	 * set temporarily
4264 	 */
4265 	add_sits_in_set(sbi);
4266 
4267 	/*
4268 	 * if there are no enough space in journal to store dirty sit
4269 	 * entries, remove all entries from journal and add and account
4270 	 * them in sit entry set.
4271 	 */
4272 	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4273 								!to_journal)
4274 		remove_sits_in_journal(sbi);
4275 
4276 	/*
4277 	 * there are two steps to flush sit entries:
4278 	 * #1, flush sit entries to journal in current cold data summary block.
4279 	 * #2, flush sit entries to sit page.
4280 	 */
4281 	list_for_each_entry_safe(ses, tmp, head, set_list) {
4282 		struct page *page = NULL;
4283 		struct f2fs_sit_block *raw_sit = NULL;
4284 		unsigned int start_segno = ses->start_segno;
4285 		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4286 						(unsigned long)MAIN_SEGS(sbi));
4287 		unsigned int segno = start_segno;
4288 
4289 		if (to_journal &&
4290 			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4291 			to_journal = false;
4292 
4293 		if (to_journal) {
4294 			down_write(&curseg->journal_rwsem);
4295 		} else {
4296 			page = get_next_sit_page(sbi, start_segno);
4297 			raw_sit = page_address(page);
4298 		}
4299 
4300 		/* flush dirty sit entries in region of current sit set */
4301 		for_each_set_bit_from(segno, bitmap, end) {
4302 			int offset, sit_offset;
4303 
4304 			se = get_seg_entry(sbi, segno);
4305 #ifdef CONFIG_F2FS_CHECK_FS
4306 			if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4307 						SIT_VBLOCK_MAP_SIZE))
4308 				f2fs_bug_on(sbi, 1);
4309 #endif
4310 
4311 			/* add discard candidates */
4312 			if (!(cpc->reason & CP_DISCARD)) {
4313 				cpc->trim_start = segno;
4314 				add_discard_addrs(sbi, cpc, false);
4315 			}
4316 
4317 			if (to_journal) {
4318 				offset = f2fs_lookup_journal_in_cursum(journal,
4319 							SIT_JOURNAL, segno, 1);
4320 				f2fs_bug_on(sbi, offset < 0);
4321 				segno_in_journal(journal, offset) =
4322 							cpu_to_le32(segno);
4323 				seg_info_to_raw_sit(se,
4324 					&sit_in_journal(journal, offset));
4325 				check_block_count(sbi, segno,
4326 					&sit_in_journal(journal, offset));
4327 			} else {
4328 				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4329 				seg_info_to_raw_sit(se,
4330 						&raw_sit->entries[sit_offset]);
4331 				check_block_count(sbi, segno,
4332 						&raw_sit->entries[sit_offset]);
4333 			}
4334 
4335 			__clear_bit(segno, bitmap);
4336 			sit_i->dirty_sentries--;
4337 			ses->entry_cnt--;
4338 		}
4339 
4340 		if (to_journal)
4341 			up_write(&curseg->journal_rwsem);
4342 		else
4343 			f2fs_put_page(page, 1);
4344 
4345 		f2fs_bug_on(sbi, ses->entry_cnt);
4346 		release_sit_entry_set(ses);
4347 	}
4348 
4349 	f2fs_bug_on(sbi, !list_empty(head));
4350 	f2fs_bug_on(sbi, sit_i->dirty_sentries);
4351 out:
4352 	if (cpc->reason & CP_DISCARD) {
4353 		__u64 trim_start = cpc->trim_start;
4354 
4355 		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4356 			add_discard_addrs(sbi, cpc, false);
4357 
4358 		cpc->trim_start = trim_start;
4359 	}
4360 	up_write(&sit_i->sentry_lock);
4361 
4362 	set_prefree_as_free_segments(sbi);
4363 }
4364 
4365 static int build_sit_info(struct f2fs_sb_info *sbi)
4366 {
4367 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4368 	struct sit_info *sit_i;
4369 	unsigned int sit_segs, start;
4370 	char *src_bitmap, *bitmap;
4371 	unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4372 	unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0;
4373 
4374 	/* allocate memory for SIT information */
4375 	sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4376 	if (!sit_i)
4377 		return -ENOMEM;
4378 
4379 	SM_I(sbi)->sit_info = sit_i;
4380 
4381 	sit_i->sentries =
4382 		f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4383 					      MAIN_SEGS(sbi)),
4384 			      GFP_KERNEL);
4385 	if (!sit_i->sentries)
4386 		return -ENOMEM;
4387 
4388 	main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4389 	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4390 								GFP_KERNEL);
4391 	if (!sit_i->dirty_sentries_bitmap)
4392 		return -ENOMEM;
4393 
4394 #ifdef CONFIG_F2FS_CHECK_FS
4395 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map);
4396 #else
4397 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map);
4398 #endif
4399 	sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4400 	if (!sit_i->bitmap)
4401 		return -ENOMEM;
4402 
4403 	bitmap = sit_i->bitmap;
4404 
4405 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4406 		sit_i->sentries[start].cur_valid_map = bitmap;
4407 		bitmap += SIT_VBLOCK_MAP_SIZE;
4408 
4409 		sit_i->sentries[start].ckpt_valid_map = bitmap;
4410 		bitmap += SIT_VBLOCK_MAP_SIZE;
4411 
4412 #ifdef CONFIG_F2FS_CHECK_FS
4413 		sit_i->sentries[start].cur_valid_map_mir = bitmap;
4414 		bitmap += SIT_VBLOCK_MAP_SIZE;
4415 #endif
4416 
4417 		if (discard_map) {
4418 			sit_i->sentries[start].discard_map = bitmap;
4419 			bitmap += SIT_VBLOCK_MAP_SIZE;
4420 		}
4421 	}
4422 
4423 	sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4424 	if (!sit_i->tmp_map)
4425 		return -ENOMEM;
4426 
4427 	if (__is_large_section(sbi)) {
4428 		sit_i->sec_entries =
4429 			f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4430 						      MAIN_SECS(sbi)),
4431 				      GFP_KERNEL);
4432 		if (!sit_i->sec_entries)
4433 			return -ENOMEM;
4434 	}
4435 
4436 	/* get information related with SIT */
4437 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4438 
4439 	/* setup SIT bitmap from ckeckpoint pack */
4440 	sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4441 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4442 
4443 	sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4444 	if (!sit_i->sit_bitmap)
4445 		return -ENOMEM;
4446 
4447 #ifdef CONFIG_F2FS_CHECK_FS
4448 	sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4449 					sit_bitmap_size, GFP_KERNEL);
4450 	if (!sit_i->sit_bitmap_mir)
4451 		return -ENOMEM;
4452 
4453 	sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4454 					main_bitmap_size, GFP_KERNEL);
4455 	if (!sit_i->invalid_segmap)
4456 		return -ENOMEM;
4457 #endif
4458 
4459 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4460 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4461 	sit_i->written_valid_blocks = 0;
4462 	sit_i->bitmap_size = sit_bitmap_size;
4463 	sit_i->dirty_sentries = 0;
4464 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4465 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4466 	sit_i->mounted_time = ktime_get_boottime_seconds();
4467 	init_rwsem(&sit_i->sentry_lock);
4468 	return 0;
4469 }
4470 
4471 static int build_free_segmap(struct f2fs_sb_info *sbi)
4472 {
4473 	struct free_segmap_info *free_i;
4474 	unsigned int bitmap_size, sec_bitmap_size;
4475 
4476 	/* allocate memory for free segmap information */
4477 	free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4478 	if (!free_i)
4479 		return -ENOMEM;
4480 
4481 	SM_I(sbi)->free_info = free_i;
4482 
4483 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4484 	free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4485 	if (!free_i->free_segmap)
4486 		return -ENOMEM;
4487 
4488 	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4489 	free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4490 	if (!free_i->free_secmap)
4491 		return -ENOMEM;
4492 
4493 	/* set all segments as dirty temporarily */
4494 	memset(free_i->free_segmap, 0xff, bitmap_size);
4495 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4496 
4497 	/* init free segmap information */
4498 	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4499 	free_i->free_segments = 0;
4500 	free_i->free_sections = 0;
4501 	spin_lock_init(&free_i->segmap_lock);
4502 	return 0;
4503 }
4504 
4505 static int build_curseg(struct f2fs_sb_info *sbi)
4506 {
4507 	struct curseg_info *array;
4508 	int i;
4509 
4510 	array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4511 					sizeof(*array)), GFP_KERNEL);
4512 	if (!array)
4513 		return -ENOMEM;
4514 
4515 	SM_I(sbi)->curseg_array = array;
4516 
4517 	for (i = 0; i < NO_CHECK_TYPE; i++) {
4518 		mutex_init(&array[i].curseg_mutex);
4519 		array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4520 		if (!array[i].sum_blk)
4521 			return -ENOMEM;
4522 		init_rwsem(&array[i].journal_rwsem);
4523 		array[i].journal = f2fs_kzalloc(sbi,
4524 				sizeof(struct f2fs_journal), GFP_KERNEL);
4525 		if (!array[i].journal)
4526 			return -ENOMEM;
4527 		if (i < NR_PERSISTENT_LOG)
4528 			array[i].seg_type = CURSEG_HOT_DATA + i;
4529 		else if (i == CURSEG_COLD_DATA_PINNED)
4530 			array[i].seg_type = CURSEG_COLD_DATA;
4531 		else if (i == CURSEG_ALL_DATA_ATGC)
4532 			array[i].seg_type = CURSEG_COLD_DATA;
4533 		array[i].segno = NULL_SEGNO;
4534 		array[i].next_blkoff = 0;
4535 		array[i].inited = false;
4536 	}
4537 	return restore_curseg_summaries(sbi);
4538 }
4539 
4540 static int build_sit_entries(struct f2fs_sb_info *sbi)
4541 {
4542 	struct sit_info *sit_i = SIT_I(sbi);
4543 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4544 	struct f2fs_journal *journal = curseg->journal;
4545 	struct seg_entry *se;
4546 	struct f2fs_sit_entry sit;
4547 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
4548 	unsigned int i, start, end;
4549 	unsigned int readed, start_blk = 0;
4550 	int err = 0;
4551 	block_t sit_valid_blocks[2] = {0, 0};
4552 
4553 	do {
4554 		readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4555 							META_SIT, true);
4556 
4557 		start = start_blk * sit_i->sents_per_block;
4558 		end = (start_blk + readed) * sit_i->sents_per_block;
4559 
4560 		for (; start < end && start < MAIN_SEGS(sbi); start++) {
4561 			struct f2fs_sit_block *sit_blk;
4562 			struct page *page;
4563 
4564 			se = &sit_i->sentries[start];
4565 			page = get_current_sit_page(sbi, start);
4566 			if (IS_ERR(page))
4567 				return PTR_ERR(page);
4568 			sit_blk = (struct f2fs_sit_block *)page_address(page);
4569 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4570 			f2fs_put_page(page, 1);
4571 
4572 			err = check_block_count(sbi, start, &sit);
4573 			if (err)
4574 				return err;
4575 			seg_info_from_raw_sit(se, &sit);
4576 
4577 			if (se->type >= NR_PERSISTENT_LOG) {
4578 				f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4579 							se->type, start);
4580 				f2fs_handle_error(sbi,
4581 						ERROR_INCONSISTENT_SUM_TYPE);
4582 				return -EFSCORRUPTED;
4583 			}
4584 
4585 			sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4586 
4587 			if (f2fs_block_unit_discard(sbi)) {
4588 				/* build discard map only one time */
4589 				if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4590 					memset(se->discard_map, 0xff,
4591 						SIT_VBLOCK_MAP_SIZE);
4592 				} else {
4593 					memcpy(se->discard_map,
4594 						se->cur_valid_map,
4595 						SIT_VBLOCK_MAP_SIZE);
4596 					sbi->discard_blks +=
4597 						sbi->blocks_per_seg -
4598 						se->valid_blocks;
4599 				}
4600 			}
4601 
4602 			if (__is_large_section(sbi))
4603 				get_sec_entry(sbi, start)->valid_blocks +=
4604 							se->valid_blocks;
4605 		}
4606 		start_blk += readed;
4607 	} while (start_blk < sit_blk_cnt);
4608 
4609 	down_read(&curseg->journal_rwsem);
4610 	for (i = 0; i < sits_in_cursum(journal); i++) {
4611 		unsigned int old_valid_blocks;
4612 
4613 		start = le32_to_cpu(segno_in_journal(journal, i));
4614 		if (start >= MAIN_SEGS(sbi)) {
4615 			f2fs_err(sbi, "Wrong journal entry on segno %u",
4616 				 start);
4617 			err = -EFSCORRUPTED;
4618 			f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL);
4619 			break;
4620 		}
4621 
4622 		se = &sit_i->sentries[start];
4623 		sit = sit_in_journal(journal, i);
4624 
4625 		old_valid_blocks = se->valid_blocks;
4626 
4627 		sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks;
4628 
4629 		err = check_block_count(sbi, start, &sit);
4630 		if (err)
4631 			break;
4632 		seg_info_from_raw_sit(se, &sit);
4633 
4634 		if (se->type >= NR_PERSISTENT_LOG) {
4635 			f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4636 							se->type, start);
4637 			err = -EFSCORRUPTED;
4638 			f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
4639 			break;
4640 		}
4641 
4642 		sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4643 
4644 		if (f2fs_block_unit_discard(sbi)) {
4645 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4646 				memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4647 			} else {
4648 				memcpy(se->discard_map, se->cur_valid_map,
4649 							SIT_VBLOCK_MAP_SIZE);
4650 				sbi->discard_blks += old_valid_blocks;
4651 				sbi->discard_blks -= se->valid_blocks;
4652 			}
4653 		}
4654 
4655 		if (__is_large_section(sbi)) {
4656 			get_sec_entry(sbi, start)->valid_blocks +=
4657 							se->valid_blocks;
4658 			get_sec_entry(sbi, start)->valid_blocks -=
4659 							old_valid_blocks;
4660 		}
4661 	}
4662 	up_read(&curseg->journal_rwsem);
4663 
4664 	if (err)
4665 		return err;
4666 
4667 	if (sit_valid_blocks[NODE] != valid_node_count(sbi)) {
4668 		f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4669 			 sit_valid_blocks[NODE], valid_node_count(sbi));
4670 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT);
4671 		return -EFSCORRUPTED;
4672 	}
4673 
4674 	if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] >
4675 				valid_user_blocks(sbi)) {
4676 		f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u",
4677 			 sit_valid_blocks[DATA], sit_valid_blocks[NODE],
4678 			 valid_user_blocks(sbi));
4679 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT);
4680 		return -EFSCORRUPTED;
4681 	}
4682 
4683 	return 0;
4684 }
4685 
4686 static void init_free_segmap(struct f2fs_sb_info *sbi)
4687 {
4688 	unsigned int start;
4689 	int type;
4690 	struct seg_entry *sentry;
4691 
4692 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4693 		if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4694 			continue;
4695 		sentry = get_seg_entry(sbi, start);
4696 		if (!sentry->valid_blocks)
4697 			__set_free(sbi, start);
4698 		else
4699 			SIT_I(sbi)->written_valid_blocks +=
4700 						sentry->valid_blocks;
4701 	}
4702 
4703 	/* set use the current segments */
4704 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4705 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4706 
4707 		__set_test_and_inuse(sbi, curseg_t->segno);
4708 	}
4709 }
4710 
4711 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4712 {
4713 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4714 	struct free_segmap_info *free_i = FREE_I(sbi);
4715 	unsigned int segno = 0, offset = 0, secno;
4716 	block_t valid_blocks, usable_blks_in_seg;
4717 
4718 	while (1) {
4719 		/* find dirty segment based on free segmap */
4720 		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4721 		if (segno >= MAIN_SEGS(sbi))
4722 			break;
4723 		offset = segno + 1;
4724 		valid_blocks = get_valid_blocks(sbi, segno, false);
4725 		usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4726 		if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4727 			continue;
4728 		if (valid_blocks > usable_blks_in_seg) {
4729 			f2fs_bug_on(sbi, 1);
4730 			continue;
4731 		}
4732 		mutex_lock(&dirty_i->seglist_lock);
4733 		__locate_dirty_segment(sbi, segno, DIRTY);
4734 		mutex_unlock(&dirty_i->seglist_lock);
4735 	}
4736 
4737 	if (!__is_large_section(sbi))
4738 		return;
4739 
4740 	mutex_lock(&dirty_i->seglist_lock);
4741 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4742 		valid_blocks = get_valid_blocks(sbi, segno, true);
4743 		secno = GET_SEC_FROM_SEG(sbi, segno);
4744 
4745 		if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi))
4746 			continue;
4747 		if (IS_CURSEC(sbi, secno))
4748 			continue;
4749 		set_bit(secno, dirty_i->dirty_secmap);
4750 	}
4751 	mutex_unlock(&dirty_i->seglist_lock);
4752 }
4753 
4754 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4755 {
4756 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4757 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4758 
4759 	dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4760 	if (!dirty_i->victim_secmap)
4761 		return -ENOMEM;
4762 
4763 	dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4764 	if (!dirty_i->pinned_secmap)
4765 		return -ENOMEM;
4766 
4767 	dirty_i->pinned_secmap_cnt = 0;
4768 	dirty_i->enable_pin_section = true;
4769 	return 0;
4770 }
4771 
4772 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4773 {
4774 	struct dirty_seglist_info *dirty_i;
4775 	unsigned int bitmap_size, i;
4776 
4777 	/* allocate memory for dirty segments list information */
4778 	dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4779 								GFP_KERNEL);
4780 	if (!dirty_i)
4781 		return -ENOMEM;
4782 
4783 	SM_I(sbi)->dirty_info = dirty_i;
4784 	mutex_init(&dirty_i->seglist_lock);
4785 
4786 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4787 
4788 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
4789 		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4790 								GFP_KERNEL);
4791 		if (!dirty_i->dirty_segmap[i])
4792 			return -ENOMEM;
4793 	}
4794 
4795 	if (__is_large_section(sbi)) {
4796 		bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4797 		dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4798 						bitmap_size, GFP_KERNEL);
4799 		if (!dirty_i->dirty_secmap)
4800 			return -ENOMEM;
4801 	}
4802 
4803 	init_dirty_segmap(sbi);
4804 	return init_victim_secmap(sbi);
4805 }
4806 
4807 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4808 {
4809 	int i;
4810 
4811 	/*
4812 	 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4813 	 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4814 	 */
4815 	for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4816 		struct curseg_info *curseg = CURSEG_I(sbi, i);
4817 		struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4818 		unsigned int blkofs = curseg->next_blkoff;
4819 
4820 		if (f2fs_sb_has_readonly(sbi) &&
4821 			i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE)
4822 			continue;
4823 
4824 		sanity_check_seg_type(sbi, curseg->seg_type);
4825 
4826 		if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) {
4827 			f2fs_err(sbi,
4828 				 "Current segment has invalid alloc_type:%d",
4829 				 curseg->alloc_type);
4830 			f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4831 			return -EFSCORRUPTED;
4832 		}
4833 
4834 		if (f2fs_test_bit(blkofs, se->cur_valid_map))
4835 			goto out;
4836 
4837 		if (curseg->alloc_type == SSR)
4838 			continue;
4839 
4840 		for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4841 			if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4842 				continue;
4843 out:
4844 			f2fs_err(sbi,
4845 				 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4846 				 i, curseg->segno, curseg->alloc_type,
4847 				 curseg->next_blkoff, blkofs);
4848 			f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4849 			return -EFSCORRUPTED;
4850 		}
4851 	}
4852 	return 0;
4853 }
4854 
4855 #ifdef CONFIG_BLK_DEV_ZONED
4856 
4857 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4858 				    struct f2fs_dev_info *fdev,
4859 				    struct blk_zone *zone)
4860 {
4861 	unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4862 	block_t zone_block, wp_block, last_valid_block;
4863 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4864 	int i, s, b, ret;
4865 	struct seg_entry *se;
4866 
4867 	if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4868 		return 0;
4869 
4870 	wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4871 	wp_segno = GET_SEGNO(sbi, wp_block);
4872 	wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4873 	zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4874 	zone_segno = GET_SEGNO(sbi, zone_block);
4875 	zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4876 
4877 	if (zone_segno >= MAIN_SEGS(sbi))
4878 		return 0;
4879 
4880 	/*
4881 	 * Skip check of zones cursegs point to, since
4882 	 * fix_curseg_write_pointer() checks them.
4883 	 */
4884 	for (i = 0; i < NO_CHECK_TYPE; i++)
4885 		if (zone_secno == GET_SEC_FROM_SEG(sbi,
4886 						   CURSEG_I(sbi, i)->segno))
4887 			return 0;
4888 
4889 	/*
4890 	 * Get last valid block of the zone.
4891 	 */
4892 	last_valid_block = zone_block - 1;
4893 	for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4894 		segno = zone_segno + s;
4895 		se = get_seg_entry(sbi, segno);
4896 		for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4897 			if (f2fs_test_bit(b, se->cur_valid_map)) {
4898 				last_valid_block = START_BLOCK(sbi, segno) + b;
4899 				break;
4900 			}
4901 		if (last_valid_block >= zone_block)
4902 			break;
4903 	}
4904 
4905 	/*
4906 	 * The write pointer matches with the valid blocks or
4907 	 * already points to the end of the zone.
4908 	 */
4909 	if ((last_valid_block + 1 == wp_block) ||
4910 			(zone->wp == zone->start + zone->len))
4911 		return 0;
4912 
4913 	if (last_valid_block + 1 == zone_block) {
4914 		/*
4915 		 * If there is no valid block in the zone and if write pointer
4916 		 * is not at zone start, reset the write pointer.
4917 		 */
4918 		f2fs_notice(sbi,
4919 			    "Zone without valid block has non-zero write "
4920 			    "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4921 			    wp_segno, wp_blkoff);
4922 		ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4923 					zone->len >> log_sectors_per_block);
4924 		if (ret)
4925 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4926 				 fdev->path, ret);
4927 
4928 		return ret;
4929 	}
4930 
4931 	/*
4932 	 * If there are valid blocks and the write pointer doesn't
4933 	 * match with them, we need to report the inconsistency and
4934 	 * fill the zone till the end to close the zone. This inconsistency
4935 	 * does not cause write error because the zone will not be selected
4936 	 * for write operation until it get discarded.
4937 	 */
4938 	f2fs_notice(sbi, "Valid blocks are not aligned with write pointer: "
4939 		    "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4940 		    GET_SEGNO(sbi, last_valid_block),
4941 		    GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4942 		    wp_segno, wp_blkoff);
4943 
4944 	ret = blkdev_zone_mgmt(fdev->bdev, REQ_OP_ZONE_FINISH,
4945 				zone->start, zone->len, GFP_NOFS);
4946 	if (ret == -EOPNOTSUPP) {
4947 		ret = blkdev_issue_zeroout(fdev->bdev, zone->wp,
4948 					zone->len - (zone->wp - zone->start),
4949 					GFP_NOFS, 0);
4950 		if (ret)
4951 			f2fs_err(sbi, "Fill up zone failed: %s (errno=%d)",
4952 					fdev->path, ret);
4953 	} else if (ret) {
4954 		f2fs_err(sbi, "Finishing zone failed: %s (errno=%d)",
4955 				fdev->path, ret);
4956 	}
4957 
4958 	return ret;
4959 }
4960 
4961 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4962 						  block_t zone_blkaddr)
4963 {
4964 	int i;
4965 
4966 	for (i = 0; i < sbi->s_ndevs; i++) {
4967 		if (!bdev_is_zoned(FDEV(i).bdev))
4968 			continue;
4969 		if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4970 				zone_blkaddr <= FDEV(i).end_blk))
4971 			return &FDEV(i);
4972 	}
4973 
4974 	return NULL;
4975 }
4976 
4977 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4978 			      void *data)
4979 {
4980 	memcpy(data, zone, sizeof(struct blk_zone));
4981 	return 0;
4982 }
4983 
4984 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4985 {
4986 	struct curseg_info *cs = CURSEG_I(sbi, type);
4987 	struct f2fs_dev_info *zbd;
4988 	struct blk_zone zone;
4989 	unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4990 	block_t cs_zone_block, wp_block;
4991 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4992 	sector_t zone_sector;
4993 	int err;
4994 
4995 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4996 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4997 
4998 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
4999 	if (!zbd)
5000 		return 0;
5001 
5002 	/* report zone for the sector the curseg points to */
5003 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5004 		<< log_sectors_per_block;
5005 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5006 				  report_one_zone_cb, &zone);
5007 	if (err != 1) {
5008 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5009 			 zbd->path, err);
5010 		return err;
5011 	}
5012 
5013 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5014 		return 0;
5015 
5016 	wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
5017 	wp_segno = GET_SEGNO(sbi, wp_block);
5018 	wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
5019 	wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
5020 
5021 	if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
5022 		wp_sector_off == 0)
5023 		return 0;
5024 
5025 	f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
5026 		    "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
5027 		    type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
5028 
5029 	f2fs_notice(sbi, "Assign new section to curseg[%d]: "
5030 		    "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
5031 
5032 	f2fs_allocate_new_section(sbi, type, true);
5033 
5034 	/* check consistency of the zone curseg pointed to */
5035 	if (check_zone_write_pointer(sbi, zbd, &zone))
5036 		return -EIO;
5037 
5038 	/* check newly assigned zone */
5039 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
5040 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
5041 
5042 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
5043 	if (!zbd)
5044 		return 0;
5045 
5046 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5047 		<< log_sectors_per_block;
5048 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5049 				  report_one_zone_cb, &zone);
5050 	if (err != 1) {
5051 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5052 			 zbd->path, err);
5053 		return err;
5054 	}
5055 
5056 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5057 		return 0;
5058 
5059 	if (zone.wp != zone.start) {
5060 		f2fs_notice(sbi,
5061 			    "New zone for curseg[%d] is not yet discarded. "
5062 			    "Reset the zone: curseg[0x%x,0x%x]",
5063 			    type, cs->segno, cs->next_blkoff);
5064 		err = __f2fs_issue_discard_zone(sbi, zbd->bdev,	cs_zone_block,
5065 					zone.len >> log_sectors_per_block);
5066 		if (err) {
5067 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
5068 				 zbd->path, err);
5069 			return err;
5070 		}
5071 	}
5072 
5073 	return 0;
5074 }
5075 
5076 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5077 {
5078 	int i, ret;
5079 
5080 	for (i = 0; i < NR_PERSISTENT_LOG; i++) {
5081 		ret = fix_curseg_write_pointer(sbi, i);
5082 		if (ret)
5083 			return ret;
5084 	}
5085 
5086 	return 0;
5087 }
5088 
5089 struct check_zone_write_pointer_args {
5090 	struct f2fs_sb_info *sbi;
5091 	struct f2fs_dev_info *fdev;
5092 };
5093 
5094 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
5095 				      void *data)
5096 {
5097 	struct check_zone_write_pointer_args *args;
5098 
5099 	args = (struct check_zone_write_pointer_args *)data;
5100 
5101 	return check_zone_write_pointer(args->sbi, args->fdev, zone);
5102 }
5103 
5104 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5105 {
5106 	int i, ret;
5107 	struct check_zone_write_pointer_args args;
5108 
5109 	for (i = 0; i < sbi->s_ndevs; i++) {
5110 		if (!bdev_is_zoned(FDEV(i).bdev))
5111 			continue;
5112 
5113 		args.sbi = sbi;
5114 		args.fdev = &FDEV(i);
5115 		ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
5116 					  check_zone_write_pointer_cb, &args);
5117 		if (ret < 0)
5118 			return ret;
5119 	}
5120 
5121 	return 0;
5122 }
5123 
5124 /*
5125  * Return the number of usable blocks in a segment. The number of blocks
5126  * returned is always equal to the number of blocks in a segment for
5127  * segments fully contained within a sequential zone capacity or a
5128  * conventional zone. For segments partially contained in a sequential
5129  * zone capacity, the number of usable blocks up to the zone capacity
5130  * is returned. 0 is returned in all other cases.
5131  */
5132 static inline unsigned int f2fs_usable_zone_blks_in_seg(
5133 			struct f2fs_sb_info *sbi, unsigned int segno)
5134 {
5135 	block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
5136 	unsigned int secno;
5137 
5138 	if (!sbi->unusable_blocks_per_sec)
5139 		return sbi->blocks_per_seg;
5140 
5141 	secno = GET_SEC_FROM_SEG(sbi, segno);
5142 	seg_start = START_BLOCK(sbi, segno);
5143 	sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5144 	sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi);
5145 
5146 	/*
5147 	 * If segment starts before zone capacity and spans beyond
5148 	 * zone capacity, then usable blocks are from seg start to
5149 	 * zone capacity. If the segment starts after the zone capacity,
5150 	 * then there are no usable blocks.
5151 	 */
5152 	if (seg_start >= sec_cap_blkaddr)
5153 		return 0;
5154 	if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr)
5155 		return sec_cap_blkaddr - seg_start;
5156 
5157 	return sbi->blocks_per_seg;
5158 }
5159 #else
5160 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5161 {
5162 	return 0;
5163 }
5164 
5165 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5166 {
5167 	return 0;
5168 }
5169 
5170 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5171 							unsigned int segno)
5172 {
5173 	return 0;
5174 }
5175 
5176 #endif
5177 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5178 					unsigned int segno)
5179 {
5180 	if (f2fs_sb_has_blkzoned(sbi))
5181 		return f2fs_usable_zone_blks_in_seg(sbi, segno);
5182 
5183 	return sbi->blocks_per_seg;
5184 }
5185 
5186 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5187 					unsigned int segno)
5188 {
5189 	if (f2fs_sb_has_blkzoned(sbi))
5190 		return CAP_SEGS_PER_SEC(sbi);
5191 
5192 	return sbi->segs_per_sec;
5193 }
5194 
5195 /*
5196  * Update min, max modified time for cost-benefit GC algorithm
5197  */
5198 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5199 {
5200 	struct sit_info *sit_i = SIT_I(sbi);
5201 	unsigned int segno;
5202 
5203 	down_write(&sit_i->sentry_lock);
5204 
5205 	sit_i->min_mtime = ULLONG_MAX;
5206 
5207 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
5208 		unsigned int i;
5209 		unsigned long long mtime = 0;
5210 
5211 		for (i = 0; i < sbi->segs_per_sec; i++)
5212 			mtime += get_seg_entry(sbi, segno + i)->mtime;
5213 
5214 		mtime = div_u64(mtime, sbi->segs_per_sec);
5215 
5216 		if (sit_i->min_mtime > mtime)
5217 			sit_i->min_mtime = mtime;
5218 	}
5219 	sit_i->max_mtime = get_mtime(sbi, false);
5220 	sit_i->dirty_max_mtime = 0;
5221 	up_write(&sit_i->sentry_lock);
5222 }
5223 
5224 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5225 {
5226 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5227 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5228 	struct f2fs_sm_info *sm_info;
5229 	int err;
5230 
5231 	sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5232 	if (!sm_info)
5233 		return -ENOMEM;
5234 
5235 	/* init sm info */
5236 	sbi->sm_info = sm_info;
5237 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5238 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5239 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5240 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5241 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5242 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5243 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5244 	sm_info->rec_prefree_segments = sm_info->main_segments *
5245 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5246 	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5247 		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5248 
5249 	if (!f2fs_lfs_mode(sbi))
5250 		sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC);
5251 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5252 	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5253 	sm_info->min_seq_blocks = sbi->blocks_per_seg;
5254 	sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5255 	sm_info->min_ssr_sections = reserved_sections(sbi);
5256 
5257 	INIT_LIST_HEAD(&sm_info->sit_entry_set);
5258 
5259 	init_f2fs_rwsem(&sm_info->curseg_lock);
5260 
5261 	err = f2fs_create_flush_cmd_control(sbi);
5262 	if (err)
5263 		return err;
5264 
5265 	err = create_discard_cmd_control(sbi);
5266 	if (err)
5267 		return err;
5268 
5269 	err = build_sit_info(sbi);
5270 	if (err)
5271 		return err;
5272 	err = build_free_segmap(sbi);
5273 	if (err)
5274 		return err;
5275 	err = build_curseg(sbi);
5276 	if (err)
5277 		return err;
5278 
5279 	/* reinit free segmap based on SIT */
5280 	err = build_sit_entries(sbi);
5281 	if (err)
5282 		return err;
5283 
5284 	init_free_segmap(sbi);
5285 	err = build_dirty_segmap(sbi);
5286 	if (err)
5287 		return err;
5288 
5289 	err = sanity_check_curseg(sbi);
5290 	if (err)
5291 		return err;
5292 
5293 	init_min_max_mtime(sbi);
5294 	return 0;
5295 }
5296 
5297 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5298 		enum dirty_type dirty_type)
5299 {
5300 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5301 
5302 	mutex_lock(&dirty_i->seglist_lock);
5303 	kvfree(dirty_i->dirty_segmap[dirty_type]);
5304 	dirty_i->nr_dirty[dirty_type] = 0;
5305 	mutex_unlock(&dirty_i->seglist_lock);
5306 }
5307 
5308 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5309 {
5310 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5311 
5312 	kvfree(dirty_i->pinned_secmap);
5313 	kvfree(dirty_i->victim_secmap);
5314 }
5315 
5316 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5317 {
5318 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5319 	int i;
5320 
5321 	if (!dirty_i)
5322 		return;
5323 
5324 	/* discard pre-free/dirty segments list */
5325 	for (i = 0; i < NR_DIRTY_TYPE; i++)
5326 		discard_dirty_segmap(sbi, i);
5327 
5328 	if (__is_large_section(sbi)) {
5329 		mutex_lock(&dirty_i->seglist_lock);
5330 		kvfree(dirty_i->dirty_secmap);
5331 		mutex_unlock(&dirty_i->seglist_lock);
5332 	}
5333 
5334 	destroy_victim_secmap(sbi);
5335 	SM_I(sbi)->dirty_info = NULL;
5336 	kfree(dirty_i);
5337 }
5338 
5339 static void destroy_curseg(struct f2fs_sb_info *sbi)
5340 {
5341 	struct curseg_info *array = SM_I(sbi)->curseg_array;
5342 	int i;
5343 
5344 	if (!array)
5345 		return;
5346 	SM_I(sbi)->curseg_array = NULL;
5347 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
5348 		kfree(array[i].sum_blk);
5349 		kfree(array[i].journal);
5350 	}
5351 	kfree(array);
5352 }
5353 
5354 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5355 {
5356 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5357 
5358 	if (!free_i)
5359 		return;
5360 	SM_I(sbi)->free_info = NULL;
5361 	kvfree(free_i->free_segmap);
5362 	kvfree(free_i->free_secmap);
5363 	kfree(free_i);
5364 }
5365 
5366 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5367 {
5368 	struct sit_info *sit_i = SIT_I(sbi);
5369 
5370 	if (!sit_i)
5371 		return;
5372 
5373 	if (sit_i->sentries)
5374 		kvfree(sit_i->bitmap);
5375 	kfree(sit_i->tmp_map);
5376 
5377 	kvfree(sit_i->sentries);
5378 	kvfree(sit_i->sec_entries);
5379 	kvfree(sit_i->dirty_sentries_bitmap);
5380 
5381 	SM_I(sbi)->sit_info = NULL;
5382 	kvfree(sit_i->sit_bitmap);
5383 #ifdef CONFIG_F2FS_CHECK_FS
5384 	kvfree(sit_i->sit_bitmap_mir);
5385 	kvfree(sit_i->invalid_segmap);
5386 #endif
5387 	kfree(sit_i);
5388 }
5389 
5390 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5391 {
5392 	struct f2fs_sm_info *sm_info = SM_I(sbi);
5393 
5394 	if (!sm_info)
5395 		return;
5396 	f2fs_destroy_flush_cmd_control(sbi, true);
5397 	destroy_discard_cmd_control(sbi);
5398 	destroy_dirty_segmap(sbi);
5399 	destroy_curseg(sbi);
5400 	destroy_free_segmap(sbi);
5401 	destroy_sit_info(sbi);
5402 	sbi->sm_info = NULL;
5403 	kfree(sm_info);
5404 }
5405 
5406 int __init f2fs_create_segment_manager_caches(void)
5407 {
5408 	discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5409 			sizeof(struct discard_entry));
5410 	if (!discard_entry_slab)
5411 		goto fail;
5412 
5413 	discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5414 			sizeof(struct discard_cmd));
5415 	if (!discard_cmd_slab)
5416 		goto destroy_discard_entry;
5417 
5418 	sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5419 			sizeof(struct sit_entry_set));
5420 	if (!sit_entry_set_slab)
5421 		goto destroy_discard_cmd;
5422 
5423 	revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry",
5424 			sizeof(struct revoke_entry));
5425 	if (!revoke_entry_slab)
5426 		goto destroy_sit_entry_set;
5427 	return 0;
5428 
5429 destroy_sit_entry_set:
5430 	kmem_cache_destroy(sit_entry_set_slab);
5431 destroy_discard_cmd:
5432 	kmem_cache_destroy(discard_cmd_slab);
5433 destroy_discard_entry:
5434 	kmem_cache_destroy(discard_entry_slab);
5435 fail:
5436 	return -ENOMEM;
5437 }
5438 
5439 void f2fs_destroy_segment_manager_caches(void)
5440 {
5441 	kmem_cache_destroy(sit_entry_set_slab);
5442 	kmem_cache_destroy(discard_cmd_slab);
5443 	kmem_cache_destroy(discard_entry_slab);
5444 	kmem_cache_destroy(revoke_entry_slab);
5445 }
5446