xref: /openbmc/linux/fs/f2fs/segment.c (revision 7a829632)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * fs/f2fs/segment.c
4  *
5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6  *             http://www.samsung.com/
7  */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/bio.h>
11 #include <linux/blkdev.h>
12 #include <linux/sched/mm.h>
13 #include <linux/prefetch.h>
14 #include <linux/kthread.h>
15 #include <linux/swap.h>
16 #include <linux/timer.h>
17 #include <linux/freezer.h>
18 #include <linux/sched/signal.h>
19 #include <linux/random.h>
20 
21 #include "f2fs.h"
22 #include "segment.h"
23 #include "node.h"
24 #include "gc.h"
25 #include "iostat.h"
26 #include <trace/events/f2fs.h>
27 
28 #define __reverse_ffz(x) __reverse_ffs(~(x))
29 
30 static struct kmem_cache *discard_entry_slab;
31 static struct kmem_cache *discard_cmd_slab;
32 static struct kmem_cache *sit_entry_set_slab;
33 static struct kmem_cache *revoke_entry_slab;
34 
35 static unsigned long __reverse_ulong(unsigned char *str)
36 {
37 	unsigned long tmp = 0;
38 	int shift = 24, idx = 0;
39 
40 #if BITS_PER_LONG == 64
41 	shift = 56;
42 #endif
43 	while (shift >= 0) {
44 		tmp |= (unsigned long)str[idx++] << shift;
45 		shift -= BITS_PER_BYTE;
46 	}
47 	return tmp;
48 }
49 
50 /*
51  * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
52  * MSB and LSB are reversed in a byte by f2fs_set_bit.
53  */
54 static inline unsigned long __reverse_ffs(unsigned long word)
55 {
56 	int num = 0;
57 
58 #if BITS_PER_LONG == 64
59 	if ((word & 0xffffffff00000000UL) == 0)
60 		num += 32;
61 	else
62 		word >>= 32;
63 #endif
64 	if ((word & 0xffff0000) == 0)
65 		num += 16;
66 	else
67 		word >>= 16;
68 
69 	if ((word & 0xff00) == 0)
70 		num += 8;
71 	else
72 		word >>= 8;
73 
74 	if ((word & 0xf0) == 0)
75 		num += 4;
76 	else
77 		word >>= 4;
78 
79 	if ((word & 0xc) == 0)
80 		num += 2;
81 	else
82 		word >>= 2;
83 
84 	if ((word & 0x2) == 0)
85 		num += 1;
86 	return num;
87 }
88 
89 /*
90  * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
91  * f2fs_set_bit makes MSB and LSB reversed in a byte.
92  * @size must be integral times of unsigned long.
93  * Example:
94  *                             MSB <--> LSB
95  *   f2fs_set_bit(0, bitmap) => 1000 0000
96  *   f2fs_set_bit(7, bitmap) => 0000 0001
97  */
98 static unsigned long __find_rev_next_bit(const unsigned long *addr,
99 			unsigned long size, unsigned long offset)
100 {
101 	const unsigned long *p = addr + BIT_WORD(offset);
102 	unsigned long result = size;
103 	unsigned long tmp;
104 
105 	if (offset >= size)
106 		return size;
107 
108 	size -= (offset & ~(BITS_PER_LONG - 1));
109 	offset %= BITS_PER_LONG;
110 
111 	while (1) {
112 		if (*p == 0)
113 			goto pass;
114 
115 		tmp = __reverse_ulong((unsigned char *)p);
116 
117 		tmp &= ~0UL >> offset;
118 		if (size < BITS_PER_LONG)
119 			tmp &= (~0UL << (BITS_PER_LONG - size));
120 		if (tmp)
121 			goto found;
122 pass:
123 		if (size <= BITS_PER_LONG)
124 			break;
125 		size -= BITS_PER_LONG;
126 		offset = 0;
127 		p++;
128 	}
129 	return result;
130 found:
131 	return result - size + __reverse_ffs(tmp);
132 }
133 
134 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
135 			unsigned long size, unsigned long offset)
136 {
137 	const unsigned long *p = addr + BIT_WORD(offset);
138 	unsigned long result = size;
139 	unsigned long tmp;
140 
141 	if (offset >= size)
142 		return size;
143 
144 	size -= (offset & ~(BITS_PER_LONG - 1));
145 	offset %= BITS_PER_LONG;
146 
147 	while (1) {
148 		if (*p == ~0UL)
149 			goto pass;
150 
151 		tmp = __reverse_ulong((unsigned char *)p);
152 
153 		if (offset)
154 			tmp |= ~0UL << (BITS_PER_LONG - offset);
155 		if (size < BITS_PER_LONG)
156 			tmp |= ~0UL >> size;
157 		if (tmp != ~0UL)
158 			goto found;
159 pass:
160 		if (size <= BITS_PER_LONG)
161 			break;
162 		size -= BITS_PER_LONG;
163 		offset = 0;
164 		p++;
165 	}
166 	return result;
167 found:
168 	return result - size + __reverse_ffz(tmp);
169 }
170 
171 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
172 {
173 	int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
174 	int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
175 	int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
176 
177 	if (f2fs_lfs_mode(sbi))
178 		return false;
179 	if (sbi->gc_mode == GC_URGENT_HIGH)
180 		return true;
181 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
182 		return true;
183 
184 	return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
185 			SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
186 }
187 
188 void f2fs_abort_atomic_write(struct inode *inode, bool clean)
189 {
190 	struct f2fs_inode_info *fi = F2FS_I(inode);
191 
192 	if (!f2fs_is_atomic_file(inode))
193 		return;
194 
195 	if (clean)
196 		truncate_inode_pages_final(inode->i_mapping);
197 
198 	release_atomic_write_cnt(inode);
199 	clear_inode_flag(inode, FI_ATOMIC_COMMITTED);
200 	clear_inode_flag(inode, FI_ATOMIC_REPLACE);
201 	clear_inode_flag(inode, FI_ATOMIC_FILE);
202 	stat_dec_atomic_inode(inode);
203 
204 	F2FS_I(inode)->atomic_write_task = NULL;
205 
206 	if (clean) {
207 		f2fs_i_size_write(inode, fi->original_i_size);
208 		fi->original_i_size = 0;
209 	}
210 	/* avoid stale dirty inode during eviction */
211 	sync_inode_metadata(inode, 0);
212 }
213 
214 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index,
215 			block_t new_addr, block_t *old_addr, bool recover)
216 {
217 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
218 	struct dnode_of_data dn;
219 	struct node_info ni;
220 	int err;
221 
222 retry:
223 	set_new_dnode(&dn, inode, NULL, NULL, 0);
224 	err = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE);
225 	if (err) {
226 		if (err == -ENOMEM) {
227 			f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
228 			goto retry;
229 		}
230 		return err;
231 	}
232 
233 	err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
234 	if (err) {
235 		f2fs_put_dnode(&dn);
236 		return err;
237 	}
238 
239 	if (recover) {
240 		/* dn.data_blkaddr is always valid */
241 		if (!__is_valid_data_blkaddr(new_addr)) {
242 			if (new_addr == NULL_ADDR)
243 				dec_valid_block_count(sbi, inode, 1);
244 			f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
245 			f2fs_update_data_blkaddr(&dn, new_addr);
246 		} else {
247 			f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
248 				new_addr, ni.version, true, true);
249 		}
250 	} else {
251 		blkcnt_t count = 1;
252 
253 		err = inc_valid_block_count(sbi, inode, &count, true);
254 		if (err) {
255 			f2fs_put_dnode(&dn);
256 			return err;
257 		}
258 
259 		*old_addr = dn.data_blkaddr;
260 		f2fs_truncate_data_blocks_range(&dn, 1);
261 		dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count);
262 
263 		f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr,
264 					ni.version, true, false);
265 	}
266 
267 	f2fs_put_dnode(&dn);
268 
269 	trace_f2fs_replace_atomic_write_block(inode, F2FS_I(inode)->cow_inode,
270 			index, old_addr ? *old_addr : 0, new_addr, recover);
271 	return 0;
272 }
273 
274 static void __complete_revoke_list(struct inode *inode, struct list_head *head,
275 					bool revoke)
276 {
277 	struct revoke_entry *cur, *tmp;
278 	pgoff_t start_index = 0;
279 	bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE);
280 
281 	list_for_each_entry_safe(cur, tmp, head, list) {
282 		if (revoke) {
283 			__replace_atomic_write_block(inode, cur->index,
284 						cur->old_addr, NULL, true);
285 		} else if (truncate) {
286 			f2fs_truncate_hole(inode, start_index, cur->index);
287 			start_index = cur->index + 1;
288 		}
289 
290 		list_del(&cur->list);
291 		kmem_cache_free(revoke_entry_slab, cur);
292 	}
293 
294 	if (!revoke && truncate)
295 		f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false);
296 }
297 
298 static int __f2fs_commit_atomic_write(struct inode *inode)
299 {
300 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
301 	struct f2fs_inode_info *fi = F2FS_I(inode);
302 	struct inode *cow_inode = fi->cow_inode;
303 	struct revoke_entry *new;
304 	struct list_head revoke_list;
305 	block_t blkaddr;
306 	struct dnode_of_data dn;
307 	pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
308 	pgoff_t off = 0, blen, index;
309 	int ret = 0, i;
310 
311 	INIT_LIST_HEAD(&revoke_list);
312 
313 	while (len) {
314 		blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len);
315 
316 		set_new_dnode(&dn, cow_inode, NULL, NULL, 0);
317 		ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA);
318 		if (ret && ret != -ENOENT) {
319 			goto out;
320 		} else if (ret == -ENOENT) {
321 			ret = 0;
322 			if (dn.max_level == 0)
323 				goto out;
324 			goto next;
325 		}
326 
327 		blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode),
328 				len);
329 		index = off;
330 		for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) {
331 			blkaddr = f2fs_data_blkaddr(&dn);
332 
333 			if (!__is_valid_data_blkaddr(blkaddr)) {
334 				continue;
335 			} else if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
336 					DATA_GENERIC_ENHANCE)) {
337 				f2fs_put_dnode(&dn);
338 				ret = -EFSCORRUPTED;
339 				f2fs_handle_error(sbi,
340 						ERROR_INVALID_BLKADDR);
341 				goto out;
342 			}
343 
344 			new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS,
345 							true, NULL);
346 
347 			ret = __replace_atomic_write_block(inode, index, blkaddr,
348 							&new->old_addr, false);
349 			if (ret) {
350 				f2fs_put_dnode(&dn);
351 				kmem_cache_free(revoke_entry_slab, new);
352 				goto out;
353 			}
354 
355 			f2fs_update_data_blkaddr(&dn, NULL_ADDR);
356 			new->index = index;
357 			list_add_tail(&new->list, &revoke_list);
358 		}
359 		f2fs_put_dnode(&dn);
360 next:
361 		off += blen;
362 		len -= blen;
363 	}
364 
365 out:
366 	if (ret) {
367 		sbi->revoked_atomic_block += fi->atomic_write_cnt;
368 	} else {
369 		sbi->committed_atomic_block += fi->atomic_write_cnt;
370 		set_inode_flag(inode, FI_ATOMIC_COMMITTED);
371 	}
372 
373 	__complete_revoke_list(inode, &revoke_list, ret ? true : false);
374 
375 	return ret;
376 }
377 
378 int f2fs_commit_atomic_write(struct inode *inode)
379 {
380 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
381 	struct f2fs_inode_info *fi = F2FS_I(inode);
382 	int err;
383 
384 	err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
385 	if (err)
386 		return err;
387 
388 	f2fs_down_write(&fi->i_gc_rwsem[WRITE]);
389 	f2fs_lock_op(sbi);
390 
391 	err = __f2fs_commit_atomic_write(inode);
392 
393 	f2fs_unlock_op(sbi);
394 	f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
395 
396 	return err;
397 }
398 
399 /*
400  * This function balances dirty node and dentry pages.
401  * In addition, it controls garbage collection.
402  */
403 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
404 {
405 	if (time_to_inject(sbi, FAULT_CHECKPOINT))
406 		f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT);
407 
408 	/* balance_fs_bg is able to be pending */
409 	if (need && excess_cached_nats(sbi))
410 		f2fs_balance_fs_bg(sbi, false);
411 
412 	if (!f2fs_is_checkpoint_ready(sbi))
413 		return;
414 
415 	/*
416 	 * We should do GC or end up with checkpoint, if there are so many dirty
417 	 * dir/node pages without enough free segments.
418 	 */
419 	if (has_enough_free_secs(sbi, 0, 0))
420 		return;
421 
422 	if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
423 				sbi->gc_thread->f2fs_gc_task) {
424 		DEFINE_WAIT(wait);
425 
426 		prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
427 					TASK_UNINTERRUPTIBLE);
428 		wake_up(&sbi->gc_thread->gc_wait_queue_head);
429 		io_schedule();
430 		finish_wait(&sbi->gc_thread->fggc_wq, &wait);
431 	} else {
432 		struct f2fs_gc_control gc_control = {
433 			.victim_segno = NULL_SEGNO,
434 			.init_gc_type = BG_GC,
435 			.no_bg_gc = true,
436 			.should_migrate_blocks = false,
437 			.err_gc_skipped = false,
438 			.nr_free_secs = 1 };
439 		f2fs_down_write(&sbi->gc_lock);
440 		stat_inc_gc_call_count(sbi, FOREGROUND);
441 		f2fs_gc(sbi, &gc_control);
442 	}
443 }
444 
445 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi)
446 {
447 	int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2;
448 	unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS);
449 	unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA);
450 	unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES);
451 	unsigned int meta = get_pages(sbi, F2FS_DIRTY_META);
452 	unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA);
453 	unsigned int threshold = sbi->blocks_per_seg * factor *
454 					DEFAULT_DIRTY_THRESHOLD;
455 	unsigned int global_threshold = threshold * 3 / 2;
456 
457 	if (dents >= threshold || qdata >= threshold ||
458 		nodes >= threshold || meta >= threshold ||
459 		imeta >= threshold)
460 		return true;
461 	return dents + qdata + nodes + meta + imeta >  global_threshold;
462 }
463 
464 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
465 {
466 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
467 		return;
468 
469 	/* try to shrink extent cache when there is no enough memory */
470 	if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE))
471 		f2fs_shrink_read_extent_tree(sbi,
472 				READ_EXTENT_CACHE_SHRINK_NUMBER);
473 
474 	/* try to shrink age extent cache when there is no enough memory */
475 	if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE))
476 		f2fs_shrink_age_extent_tree(sbi,
477 				AGE_EXTENT_CACHE_SHRINK_NUMBER);
478 
479 	/* check the # of cached NAT entries */
480 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
481 		f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
482 
483 	if (!f2fs_available_free_memory(sbi, FREE_NIDS))
484 		f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
485 	else
486 		f2fs_build_free_nids(sbi, false, false);
487 
488 	if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) ||
489 		excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi))
490 		goto do_sync;
491 
492 	/* there is background inflight IO or foreground operation recently */
493 	if (is_inflight_io(sbi, REQ_TIME) ||
494 		(!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem)))
495 		return;
496 
497 	/* exceed periodical checkpoint timeout threshold */
498 	if (f2fs_time_over(sbi, CP_TIME))
499 		goto do_sync;
500 
501 	/* checkpoint is the only way to shrink partial cached entries */
502 	if (f2fs_available_free_memory(sbi, NAT_ENTRIES) &&
503 		f2fs_available_free_memory(sbi, INO_ENTRIES))
504 		return;
505 
506 do_sync:
507 	if (test_opt(sbi, DATA_FLUSH) && from_bg) {
508 		struct blk_plug plug;
509 
510 		mutex_lock(&sbi->flush_lock);
511 
512 		blk_start_plug(&plug);
513 		f2fs_sync_dirty_inodes(sbi, FILE_INODE, false);
514 		blk_finish_plug(&plug);
515 
516 		mutex_unlock(&sbi->flush_lock);
517 	}
518 	stat_inc_cp_call_count(sbi, BACKGROUND);
519 	f2fs_sync_fs(sbi->sb, 1);
520 }
521 
522 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
523 				struct block_device *bdev)
524 {
525 	int ret = blkdev_issue_flush(bdev);
526 
527 	trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
528 				test_opt(sbi, FLUSH_MERGE), ret);
529 	if (!ret)
530 		f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0);
531 	return ret;
532 }
533 
534 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
535 {
536 	int ret = 0;
537 	int i;
538 
539 	if (!f2fs_is_multi_device(sbi))
540 		return __submit_flush_wait(sbi, sbi->sb->s_bdev);
541 
542 	for (i = 0; i < sbi->s_ndevs; i++) {
543 		if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
544 			continue;
545 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
546 		if (ret)
547 			break;
548 	}
549 	return ret;
550 }
551 
552 static int issue_flush_thread(void *data)
553 {
554 	struct f2fs_sb_info *sbi = data;
555 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
556 	wait_queue_head_t *q = &fcc->flush_wait_queue;
557 repeat:
558 	if (kthread_should_stop())
559 		return 0;
560 
561 	if (!llist_empty(&fcc->issue_list)) {
562 		struct flush_cmd *cmd, *next;
563 		int ret;
564 
565 		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
566 		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
567 
568 		cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
569 
570 		ret = submit_flush_wait(sbi, cmd->ino);
571 		atomic_inc(&fcc->issued_flush);
572 
573 		llist_for_each_entry_safe(cmd, next,
574 					  fcc->dispatch_list, llnode) {
575 			cmd->ret = ret;
576 			complete(&cmd->wait);
577 		}
578 		fcc->dispatch_list = NULL;
579 	}
580 
581 	wait_event_interruptible(*q,
582 		kthread_should_stop() || !llist_empty(&fcc->issue_list));
583 	goto repeat;
584 }
585 
586 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
587 {
588 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
589 	struct flush_cmd cmd;
590 	int ret;
591 
592 	if (test_opt(sbi, NOBARRIER))
593 		return 0;
594 
595 	if (!test_opt(sbi, FLUSH_MERGE)) {
596 		atomic_inc(&fcc->queued_flush);
597 		ret = submit_flush_wait(sbi, ino);
598 		atomic_dec(&fcc->queued_flush);
599 		atomic_inc(&fcc->issued_flush);
600 		return ret;
601 	}
602 
603 	if (atomic_inc_return(&fcc->queued_flush) == 1 ||
604 	    f2fs_is_multi_device(sbi)) {
605 		ret = submit_flush_wait(sbi, ino);
606 		atomic_dec(&fcc->queued_flush);
607 
608 		atomic_inc(&fcc->issued_flush);
609 		return ret;
610 	}
611 
612 	cmd.ino = ino;
613 	init_completion(&cmd.wait);
614 
615 	llist_add(&cmd.llnode, &fcc->issue_list);
616 
617 	/*
618 	 * update issue_list before we wake up issue_flush thread, this
619 	 * smp_mb() pairs with another barrier in ___wait_event(), see
620 	 * more details in comments of waitqueue_active().
621 	 */
622 	smp_mb();
623 
624 	if (waitqueue_active(&fcc->flush_wait_queue))
625 		wake_up(&fcc->flush_wait_queue);
626 
627 	if (fcc->f2fs_issue_flush) {
628 		wait_for_completion(&cmd.wait);
629 		atomic_dec(&fcc->queued_flush);
630 	} else {
631 		struct llist_node *list;
632 
633 		list = llist_del_all(&fcc->issue_list);
634 		if (!list) {
635 			wait_for_completion(&cmd.wait);
636 			atomic_dec(&fcc->queued_flush);
637 		} else {
638 			struct flush_cmd *tmp, *next;
639 
640 			ret = submit_flush_wait(sbi, ino);
641 
642 			llist_for_each_entry_safe(tmp, next, list, llnode) {
643 				if (tmp == &cmd) {
644 					cmd.ret = ret;
645 					atomic_dec(&fcc->queued_flush);
646 					continue;
647 				}
648 				tmp->ret = ret;
649 				complete(&tmp->wait);
650 			}
651 		}
652 	}
653 
654 	return cmd.ret;
655 }
656 
657 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
658 {
659 	dev_t dev = sbi->sb->s_bdev->bd_dev;
660 	struct flush_cmd_control *fcc;
661 
662 	if (SM_I(sbi)->fcc_info) {
663 		fcc = SM_I(sbi)->fcc_info;
664 		if (fcc->f2fs_issue_flush)
665 			return 0;
666 		goto init_thread;
667 	}
668 
669 	fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
670 	if (!fcc)
671 		return -ENOMEM;
672 	atomic_set(&fcc->issued_flush, 0);
673 	atomic_set(&fcc->queued_flush, 0);
674 	init_waitqueue_head(&fcc->flush_wait_queue);
675 	init_llist_head(&fcc->issue_list);
676 	SM_I(sbi)->fcc_info = fcc;
677 	if (!test_opt(sbi, FLUSH_MERGE))
678 		return 0;
679 
680 init_thread:
681 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
682 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
683 	if (IS_ERR(fcc->f2fs_issue_flush)) {
684 		int err = PTR_ERR(fcc->f2fs_issue_flush);
685 
686 		fcc->f2fs_issue_flush = NULL;
687 		return err;
688 	}
689 
690 	return 0;
691 }
692 
693 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
694 {
695 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
696 
697 	if (fcc && fcc->f2fs_issue_flush) {
698 		struct task_struct *flush_thread = fcc->f2fs_issue_flush;
699 
700 		fcc->f2fs_issue_flush = NULL;
701 		kthread_stop(flush_thread);
702 	}
703 	if (free) {
704 		kfree(fcc);
705 		SM_I(sbi)->fcc_info = NULL;
706 	}
707 }
708 
709 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
710 {
711 	int ret = 0, i;
712 
713 	if (!f2fs_is_multi_device(sbi))
714 		return 0;
715 
716 	if (test_opt(sbi, NOBARRIER))
717 		return 0;
718 
719 	for (i = 1; i < sbi->s_ndevs; i++) {
720 		int count = DEFAULT_RETRY_IO_COUNT;
721 
722 		if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
723 			continue;
724 
725 		do {
726 			ret = __submit_flush_wait(sbi, FDEV(i).bdev);
727 			if (ret)
728 				f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
729 		} while (ret && --count);
730 
731 		if (ret) {
732 			f2fs_stop_checkpoint(sbi, false,
733 					STOP_CP_REASON_FLUSH_FAIL);
734 			break;
735 		}
736 
737 		spin_lock(&sbi->dev_lock);
738 		f2fs_clear_bit(i, (char *)&sbi->dirty_device);
739 		spin_unlock(&sbi->dev_lock);
740 	}
741 
742 	return ret;
743 }
744 
745 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
746 		enum dirty_type dirty_type)
747 {
748 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
749 
750 	/* need not be added */
751 	if (IS_CURSEG(sbi, segno))
752 		return;
753 
754 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
755 		dirty_i->nr_dirty[dirty_type]++;
756 
757 	if (dirty_type == DIRTY) {
758 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
759 		enum dirty_type t = sentry->type;
760 
761 		if (unlikely(t >= DIRTY)) {
762 			f2fs_bug_on(sbi, 1);
763 			return;
764 		}
765 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
766 			dirty_i->nr_dirty[t]++;
767 
768 		if (__is_large_section(sbi)) {
769 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
770 			block_t valid_blocks =
771 				get_valid_blocks(sbi, segno, true);
772 
773 			f2fs_bug_on(sbi, unlikely(!valid_blocks ||
774 					valid_blocks == CAP_BLKS_PER_SEC(sbi)));
775 
776 			if (!IS_CURSEC(sbi, secno))
777 				set_bit(secno, dirty_i->dirty_secmap);
778 		}
779 	}
780 }
781 
782 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
783 		enum dirty_type dirty_type)
784 {
785 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
786 	block_t valid_blocks;
787 
788 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
789 		dirty_i->nr_dirty[dirty_type]--;
790 
791 	if (dirty_type == DIRTY) {
792 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
793 		enum dirty_type t = sentry->type;
794 
795 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
796 			dirty_i->nr_dirty[t]--;
797 
798 		valid_blocks = get_valid_blocks(sbi, segno, true);
799 		if (valid_blocks == 0) {
800 			clear_bit(GET_SEC_FROM_SEG(sbi, segno),
801 						dirty_i->victim_secmap);
802 #ifdef CONFIG_F2FS_CHECK_FS
803 			clear_bit(segno, SIT_I(sbi)->invalid_segmap);
804 #endif
805 		}
806 		if (__is_large_section(sbi)) {
807 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
808 
809 			if (!valid_blocks ||
810 					valid_blocks == CAP_BLKS_PER_SEC(sbi)) {
811 				clear_bit(secno, dirty_i->dirty_secmap);
812 				return;
813 			}
814 
815 			if (!IS_CURSEC(sbi, secno))
816 				set_bit(secno, dirty_i->dirty_secmap);
817 		}
818 	}
819 }
820 
821 /*
822  * Should not occur error such as -ENOMEM.
823  * Adding dirty entry into seglist is not critical operation.
824  * If a given segment is one of current working segments, it won't be added.
825  */
826 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
827 {
828 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
829 	unsigned short valid_blocks, ckpt_valid_blocks;
830 	unsigned int usable_blocks;
831 
832 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
833 		return;
834 
835 	usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
836 	mutex_lock(&dirty_i->seglist_lock);
837 
838 	valid_blocks = get_valid_blocks(sbi, segno, false);
839 	ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
840 
841 	if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
842 		ckpt_valid_blocks == usable_blocks)) {
843 		__locate_dirty_segment(sbi, segno, PRE);
844 		__remove_dirty_segment(sbi, segno, DIRTY);
845 	} else if (valid_blocks < usable_blocks) {
846 		__locate_dirty_segment(sbi, segno, DIRTY);
847 	} else {
848 		/* Recovery routine with SSR needs this */
849 		__remove_dirty_segment(sbi, segno, DIRTY);
850 	}
851 
852 	mutex_unlock(&dirty_i->seglist_lock);
853 }
854 
855 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
856 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
857 {
858 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
859 	unsigned int segno;
860 
861 	mutex_lock(&dirty_i->seglist_lock);
862 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
863 		if (get_valid_blocks(sbi, segno, false))
864 			continue;
865 		if (IS_CURSEG(sbi, segno))
866 			continue;
867 		__locate_dirty_segment(sbi, segno, PRE);
868 		__remove_dirty_segment(sbi, segno, DIRTY);
869 	}
870 	mutex_unlock(&dirty_i->seglist_lock);
871 }
872 
873 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
874 {
875 	int ovp_hole_segs =
876 		(overprovision_segments(sbi) - reserved_segments(sbi));
877 	block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
878 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
879 	block_t holes[2] = {0, 0};	/* DATA and NODE */
880 	block_t unusable;
881 	struct seg_entry *se;
882 	unsigned int segno;
883 
884 	mutex_lock(&dirty_i->seglist_lock);
885 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
886 		se = get_seg_entry(sbi, segno);
887 		if (IS_NODESEG(se->type))
888 			holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
889 							se->valid_blocks;
890 		else
891 			holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
892 							se->valid_blocks;
893 	}
894 	mutex_unlock(&dirty_i->seglist_lock);
895 
896 	unusable = max(holes[DATA], holes[NODE]);
897 	if (unusable > ovp_holes)
898 		return unusable - ovp_holes;
899 	return 0;
900 }
901 
902 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
903 {
904 	int ovp_hole_segs =
905 		(overprovision_segments(sbi) - reserved_segments(sbi));
906 	if (unusable > F2FS_OPTION(sbi).unusable_cap)
907 		return -EAGAIN;
908 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
909 		dirty_segments(sbi) > ovp_hole_segs)
910 		return -EAGAIN;
911 	return 0;
912 }
913 
914 /* This is only used by SBI_CP_DISABLED */
915 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
916 {
917 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
918 	unsigned int segno = 0;
919 
920 	mutex_lock(&dirty_i->seglist_lock);
921 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
922 		if (get_valid_blocks(sbi, segno, false))
923 			continue;
924 		if (get_ckpt_valid_blocks(sbi, segno, false))
925 			continue;
926 		mutex_unlock(&dirty_i->seglist_lock);
927 		return segno;
928 	}
929 	mutex_unlock(&dirty_i->seglist_lock);
930 	return NULL_SEGNO;
931 }
932 
933 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
934 		struct block_device *bdev, block_t lstart,
935 		block_t start, block_t len)
936 {
937 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
938 	struct list_head *pend_list;
939 	struct discard_cmd *dc;
940 
941 	f2fs_bug_on(sbi, !len);
942 
943 	pend_list = &dcc->pend_list[plist_idx(len)];
944 
945 	dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL);
946 	INIT_LIST_HEAD(&dc->list);
947 	dc->bdev = bdev;
948 	dc->di.lstart = lstart;
949 	dc->di.start = start;
950 	dc->di.len = len;
951 	dc->ref = 0;
952 	dc->state = D_PREP;
953 	dc->queued = 0;
954 	dc->error = 0;
955 	init_completion(&dc->wait);
956 	list_add_tail(&dc->list, pend_list);
957 	spin_lock_init(&dc->lock);
958 	dc->bio_ref = 0;
959 	atomic_inc(&dcc->discard_cmd_cnt);
960 	dcc->undiscard_blks += len;
961 
962 	return dc;
963 }
964 
965 static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi)
966 {
967 #ifdef CONFIG_F2FS_CHECK_FS
968 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
969 	struct rb_node *cur = rb_first_cached(&dcc->root), *next;
970 	struct discard_cmd *cur_dc, *next_dc;
971 
972 	while (cur) {
973 		next = rb_next(cur);
974 		if (!next)
975 			return true;
976 
977 		cur_dc = rb_entry(cur, struct discard_cmd, rb_node);
978 		next_dc = rb_entry(next, struct discard_cmd, rb_node);
979 
980 		if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) {
981 			f2fs_info(sbi, "broken discard_rbtree, "
982 				"cur(%u, %u) next(%u, %u)",
983 				cur_dc->di.lstart, cur_dc->di.len,
984 				next_dc->di.lstart, next_dc->di.len);
985 			return false;
986 		}
987 		cur = next;
988 	}
989 #endif
990 	return true;
991 }
992 
993 static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi,
994 						block_t blkaddr)
995 {
996 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
997 	struct rb_node *node = dcc->root.rb_root.rb_node;
998 	struct discard_cmd *dc;
999 
1000 	while (node) {
1001 		dc = rb_entry(node, struct discard_cmd, rb_node);
1002 
1003 		if (blkaddr < dc->di.lstart)
1004 			node = node->rb_left;
1005 		else if (blkaddr >= dc->di.lstart + dc->di.len)
1006 			node = node->rb_right;
1007 		else
1008 			return dc;
1009 	}
1010 	return NULL;
1011 }
1012 
1013 static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root,
1014 				block_t blkaddr,
1015 				struct discard_cmd **prev_entry,
1016 				struct discard_cmd **next_entry,
1017 				struct rb_node ***insert_p,
1018 				struct rb_node **insert_parent)
1019 {
1020 	struct rb_node **pnode = &root->rb_root.rb_node;
1021 	struct rb_node *parent = NULL, *tmp_node;
1022 	struct discard_cmd *dc;
1023 
1024 	*insert_p = NULL;
1025 	*insert_parent = NULL;
1026 	*prev_entry = NULL;
1027 	*next_entry = NULL;
1028 
1029 	if (RB_EMPTY_ROOT(&root->rb_root))
1030 		return NULL;
1031 
1032 	while (*pnode) {
1033 		parent = *pnode;
1034 		dc = rb_entry(*pnode, struct discard_cmd, rb_node);
1035 
1036 		if (blkaddr < dc->di.lstart)
1037 			pnode = &(*pnode)->rb_left;
1038 		else if (blkaddr >= dc->di.lstart + dc->di.len)
1039 			pnode = &(*pnode)->rb_right;
1040 		else
1041 			goto lookup_neighbors;
1042 	}
1043 
1044 	*insert_p = pnode;
1045 	*insert_parent = parent;
1046 
1047 	dc = rb_entry(parent, struct discard_cmd, rb_node);
1048 	tmp_node = parent;
1049 	if (parent && blkaddr > dc->di.lstart)
1050 		tmp_node = rb_next(parent);
1051 	*next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1052 
1053 	tmp_node = parent;
1054 	if (parent && blkaddr < dc->di.lstart)
1055 		tmp_node = rb_prev(parent);
1056 	*prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1057 	return NULL;
1058 
1059 lookup_neighbors:
1060 	/* lookup prev node for merging backward later */
1061 	tmp_node = rb_prev(&dc->rb_node);
1062 	*prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1063 
1064 	/* lookup next node for merging frontward later */
1065 	tmp_node = rb_next(&dc->rb_node);
1066 	*next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1067 	return dc;
1068 }
1069 
1070 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1071 							struct discard_cmd *dc)
1072 {
1073 	if (dc->state == D_DONE)
1074 		atomic_sub(dc->queued, &dcc->queued_discard);
1075 
1076 	list_del(&dc->list);
1077 	rb_erase_cached(&dc->rb_node, &dcc->root);
1078 	dcc->undiscard_blks -= dc->di.len;
1079 
1080 	kmem_cache_free(discard_cmd_slab, dc);
1081 
1082 	atomic_dec(&dcc->discard_cmd_cnt);
1083 }
1084 
1085 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1086 							struct discard_cmd *dc)
1087 {
1088 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1089 	unsigned long flags;
1090 
1091 	trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len);
1092 
1093 	spin_lock_irqsave(&dc->lock, flags);
1094 	if (dc->bio_ref) {
1095 		spin_unlock_irqrestore(&dc->lock, flags);
1096 		return;
1097 	}
1098 	spin_unlock_irqrestore(&dc->lock, flags);
1099 
1100 	f2fs_bug_on(sbi, dc->ref);
1101 
1102 	if (dc->error == -EOPNOTSUPP)
1103 		dc->error = 0;
1104 
1105 	if (dc->error)
1106 		printk_ratelimited(
1107 			"%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1108 			KERN_INFO, sbi->sb->s_id,
1109 			dc->di.lstart, dc->di.start, dc->di.len, dc->error);
1110 	__detach_discard_cmd(dcc, dc);
1111 }
1112 
1113 static void f2fs_submit_discard_endio(struct bio *bio)
1114 {
1115 	struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1116 	unsigned long flags;
1117 
1118 	spin_lock_irqsave(&dc->lock, flags);
1119 	if (!dc->error)
1120 		dc->error = blk_status_to_errno(bio->bi_status);
1121 	dc->bio_ref--;
1122 	if (!dc->bio_ref && dc->state == D_SUBMIT) {
1123 		dc->state = D_DONE;
1124 		complete_all(&dc->wait);
1125 	}
1126 	spin_unlock_irqrestore(&dc->lock, flags);
1127 	bio_put(bio);
1128 }
1129 
1130 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1131 				block_t start, block_t end)
1132 {
1133 #ifdef CONFIG_F2FS_CHECK_FS
1134 	struct seg_entry *sentry;
1135 	unsigned int segno;
1136 	block_t blk = start;
1137 	unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1138 	unsigned long *map;
1139 
1140 	while (blk < end) {
1141 		segno = GET_SEGNO(sbi, blk);
1142 		sentry = get_seg_entry(sbi, segno);
1143 		offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1144 
1145 		if (end < START_BLOCK(sbi, segno + 1))
1146 			size = GET_BLKOFF_FROM_SEG0(sbi, end);
1147 		else
1148 			size = max_blocks;
1149 		map = (unsigned long *)(sentry->cur_valid_map);
1150 		offset = __find_rev_next_bit(map, size, offset);
1151 		f2fs_bug_on(sbi, offset != size);
1152 		blk = START_BLOCK(sbi, segno + 1);
1153 	}
1154 #endif
1155 }
1156 
1157 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1158 				struct discard_policy *dpolicy,
1159 				int discard_type, unsigned int granularity)
1160 {
1161 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1162 
1163 	/* common policy */
1164 	dpolicy->type = discard_type;
1165 	dpolicy->sync = true;
1166 	dpolicy->ordered = false;
1167 	dpolicy->granularity = granularity;
1168 
1169 	dpolicy->max_requests = dcc->max_discard_request;
1170 	dpolicy->io_aware_gran = dcc->discard_io_aware_gran;
1171 	dpolicy->timeout = false;
1172 
1173 	if (discard_type == DPOLICY_BG) {
1174 		dpolicy->min_interval = dcc->min_discard_issue_time;
1175 		dpolicy->mid_interval = dcc->mid_discard_issue_time;
1176 		dpolicy->max_interval = dcc->max_discard_issue_time;
1177 		dpolicy->io_aware = true;
1178 		dpolicy->sync = false;
1179 		dpolicy->ordered = true;
1180 		if (utilization(sbi) > dcc->discard_urgent_util) {
1181 			dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1182 			if (atomic_read(&dcc->discard_cmd_cnt))
1183 				dpolicy->max_interval =
1184 					dcc->min_discard_issue_time;
1185 		}
1186 	} else if (discard_type == DPOLICY_FORCE) {
1187 		dpolicy->min_interval = dcc->min_discard_issue_time;
1188 		dpolicy->mid_interval = dcc->mid_discard_issue_time;
1189 		dpolicy->max_interval = dcc->max_discard_issue_time;
1190 		dpolicy->io_aware = false;
1191 	} else if (discard_type == DPOLICY_FSTRIM) {
1192 		dpolicy->io_aware = false;
1193 	} else if (discard_type == DPOLICY_UMOUNT) {
1194 		dpolicy->io_aware = false;
1195 		/* we need to issue all to keep CP_TRIMMED_FLAG */
1196 		dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1197 		dpolicy->timeout = true;
1198 	}
1199 }
1200 
1201 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1202 				struct block_device *bdev, block_t lstart,
1203 				block_t start, block_t len);
1204 
1205 #ifdef CONFIG_BLK_DEV_ZONED
1206 static void __submit_zone_reset_cmd(struct f2fs_sb_info *sbi,
1207 				   struct discard_cmd *dc, blk_opf_t flag,
1208 				   struct list_head *wait_list,
1209 				   unsigned int *issued)
1210 {
1211 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1212 	struct block_device *bdev = dc->bdev;
1213 	struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS);
1214 	unsigned long flags;
1215 
1216 	trace_f2fs_issue_reset_zone(bdev, dc->di.start);
1217 
1218 	spin_lock_irqsave(&dc->lock, flags);
1219 	dc->state = D_SUBMIT;
1220 	dc->bio_ref++;
1221 	spin_unlock_irqrestore(&dc->lock, flags);
1222 
1223 	if (issued)
1224 		(*issued)++;
1225 
1226 	atomic_inc(&dcc->queued_discard);
1227 	dc->queued++;
1228 	list_move_tail(&dc->list, wait_list);
1229 
1230 	/* sanity check on discard range */
1231 	__check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len);
1232 
1233 	bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start);
1234 	bio->bi_private = dc;
1235 	bio->bi_end_io = f2fs_submit_discard_endio;
1236 	submit_bio(bio);
1237 
1238 	atomic_inc(&dcc->issued_discard);
1239 	f2fs_update_iostat(sbi, NULL, FS_ZONE_RESET_IO, dc->di.len * F2FS_BLKSIZE);
1240 }
1241 #endif
1242 
1243 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1244 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1245 				struct discard_policy *dpolicy,
1246 				struct discard_cmd *dc, int *issued)
1247 {
1248 	struct block_device *bdev = dc->bdev;
1249 	unsigned int max_discard_blocks =
1250 			SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1251 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1252 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1253 					&(dcc->fstrim_list) : &(dcc->wait_list);
1254 	blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0;
1255 	block_t lstart, start, len, total_len;
1256 	int err = 0;
1257 
1258 	if (dc->state != D_PREP)
1259 		return 0;
1260 
1261 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1262 		return 0;
1263 
1264 #ifdef CONFIG_BLK_DEV_ZONED
1265 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) {
1266 		int devi = f2fs_bdev_index(sbi, bdev);
1267 
1268 		if (devi < 0)
1269 			return -EINVAL;
1270 
1271 		if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1272 			__submit_zone_reset_cmd(sbi, dc, flag,
1273 						wait_list, issued);
1274 			return 0;
1275 		}
1276 	}
1277 #endif
1278 
1279 	trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len);
1280 
1281 	lstart = dc->di.lstart;
1282 	start = dc->di.start;
1283 	len = dc->di.len;
1284 	total_len = len;
1285 
1286 	dc->di.len = 0;
1287 
1288 	while (total_len && *issued < dpolicy->max_requests && !err) {
1289 		struct bio *bio = NULL;
1290 		unsigned long flags;
1291 		bool last = true;
1292 
1293 		if (len > max_discard_blocks) {
1294 			len = max_discard_blocks;
1295 			last = false;
1296 		}
1297 
1298 		(*issued)++;
1299 		if (*issued == dpolicy->max_requests)
1300 			last = true;
1301 
1302 		dc->di.len += len;
1303 
1304 		if (time_to_inject(sbi, FAULT_DISCARD)) {
1305 			err = -EIO;
1306 		} else {
1307 			err = __blkdev_issue_discard(bdev,
1308 					SECTOR_FROM_BLOCK(start),
1309 					SECTOR_FROM_BLOCK(len),
1310 					GFP_NOFS, &bio);
1311 		}
1312 		if (err) {
1313 			spin_lock_irqsave(&dc->lock, flags);
1314 			if (dc->state == D_PARTIAL)
1315 				dc->state = D_SUBMIT;
1316 			spin_unlock_irqrestore(&dc->lock, flags);
1317 
1318 			break;
1319 		}
1320 
1321 		f2fs_bug_on(sbi, !bio);
1322 
1323 		/*
1324 		 * should keep before submission to avoid D_DONE
1325 		 * right away
1326 		 */
1327 		spin_lock_irqsave(&dc->lock, flags);
1328 		if (last)
1329 			dc->state = D_SUBMIT;
1330 		else
1331 			dc->state = D_PARTIAL;
1332 		dc->bio_ref++;
1333 		spin_unlock_irqrestore(&dc->lock, flags);
1334 
1335 		atomic_inc(&dcc->queued_discard);
1336 		dc->queued++;
1337 		list_move_tail(&dc->list, wait_list);
1338 
1339 		/* sanity check on discard range */
1340 		__check_sit_bitmap(sbi, lstart, lstart + len);
1341 
1342 		bio->bi_private = dc;
1343 		bio->bi_end_io = f2fs_submit_discard_endio;
1344 		bio->bi_opf |= flag;
1345 		submit_bio(bio);
1346 
1347 		atomic_inc(&dcc->issued_discard);
1348 
1349 		f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE);
1350 
1351 		lstart += len;
1352 		start += len;
1353 		total_len -= len;
1354 		len = total_len;
1355 	}
1356 
1357 	if (!err && len) {
1358 		dcc->undiscard_blks -= len;
1359 		__update_discard_tree_range(sbi, bdev, lstart, start, len);
1360 	}
1361 	return err;
1362 }
1363 
1364 static void __insert_discard_cmd(struct f2fs_sb_info *sbi,
1365 				struct block_device *bdev, block_t lstart,
1366 				block_t start, block_t len)
1367 {
1368 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1369 	struct rb_node **p = &dcc->root.rb_root.rb_node;
1370 	struct rb_node *parent = NULL;
1371 	struct discard_cmd *dc;
1372 	bool leftmost = true;
1373 
1374 	/* look up rb tree to find parent node */
1375 	while (*p) {
1376 		parent = *p;
1377 		dc = rb_entry(parent, struct discard_cmd, rb_node);
1378 
1379 		if (lstart < dc->di.lstart) {
1380 			p = &(*p)->rb_left;
1381 		} else if (lstart >= dc->di.lstart + dc->di.len) {
1382 			p = &(*p)->rb_right;
1383 			leftmost = false;
1384 		} else {
1385 			f2fs_bug_on(sbi, 1);
1386 		}
1387 	}
1388 
1389 	dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1390 
1391 	rb_link_node(&dc->rb_node, parent, p);
1392 	rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1393 }
1394 
1395 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1396 						struct discard_cmd *dc)
1397 {
1398 	list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]);
1399 }
1400 
1401 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1402 				struct discard_cmd *dc, block_t blkaddr)
1403 {
1404 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1405 	struct discard_info di = dc->di;
1406 	bool modified = false;
1407 
1408 	if (dc->state == D_DONE || dc->di.len == 1) {
1409 		__remove_discard_cmd(sbi, dc);
1410 		return;
1411 	}
1412 
1413 	dcc->undiscard_blks -= di.len;
1414 
1415 	if (blkaddr > di.lstart) {
1416 		dc->di.len = blkaddr - dc->di.lstart;
1417 		dcc->undiscard_blks += dc->di.len;
1418 		__relocate_discard_cmd(dcc, dc);
1419 		modified = true;
1420 	}
1421 
1422 	if (blkaddr < di.lstart + di.len - 1) {
1423 		if (modified) {
1424 			__insert_discard_cmd(sbi, dc->bdev, blkaddr + 1,
1425 					di.start + blkaddr + 1 - di.lstart,
1426 					di.lstart + di.len - 1 - blkaddr);
1427 		} else {
1428 			dc->di.lstart++;
1429 			dc->di.len--;
1430 			dc->di.start++;
1431 			dcc->undiscard_blks += dc->di.len;
1432 			__relocate_discard_cmd(dcc, dc);
1433 		}
1434 	}
1435 }
1436 
1437 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1438 				struct block_device *bdev, block_t lstart,
1439 				block_t start, block_t len)
1440 {
1441 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1442 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1443 	struct discard_cmd *dc;
1444 	struct discard_info di = {0};
1445 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1446 	unsigned int max_discard_blocks =
1447 			SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1448 	block_t end = lstart + len;
1449 
1450 	dc = __lookup_discard_cmd_ret(&dcc->root, lstart,
1451 				&prev_dc, &next_dc, &insert_p, &insert_parent);
1452 	if (dc)
1453 		prev_dc = dc;
1454 
1455 	if (!prev_dc) {
1456 		di.lstart = lstart;
1457 		di.len = next_dc ? next_dc->di.lstart - lstart : len;
1458 		di.len = min(di.len, len);
1459 		di.start = start;
1460 	}
1461 
1462 	while (1) {
1463 		struct rb_node *node;
1464 		bool merged = false;
1465 		struct discard_cmd *tdc = NULL;
1466 
1467 		if (prev_dc) {
1468 			di.lstart = prev_dc->di.lstart + prev_dc->di.len;
1469 			if (di.lstart < lstart)
1470 				di.lstart = lstart;
1471 			if (di.lstart >= end)
1472 				break;
1473 
1474 			if (!next_dc || next_dc->di.lstart > end)
1475 				di.len = end - di.lstart;
1476 			else
1477 				di.len = next_dc->di.lstart - di.lstart;
1478 			di.start = start + di.lstart - lstart;
1479 		}
1480 
1481 		if (!di.len)
1482 			goto next;
1483 
1484 		if (prev_dc && prev_dc->state == D_PREP &&
1485 			prev_dc->bdev == bdev &&
1486 			__is_discard_back_mergeable(&di, &prev_dc->di,
1487 							max_discard_blocks)) {
1488 			prev_dc->di.len += di.len;
1489 			dcc->undiscard_blks += di.len;
1490 			__relocate_discard_cmd(dcc, prev_dc);
1491 			di = prev_dc->di;
1492 			tdc = prev_dc;
1493 			merged = true;
1494 		}
1495 
1496 		if (next_dc && next_dc->state == D_PREP &&
1497 			next_dc->bdev == bdev &&
1498 			__is_discard_front_mergeable(&di, &next_dc->di,
1499 							max_discard_blocks)) {
1500 			next_dc->di.lstart = di.lstart;
1501 			next_dc->di.len += di.len;
1502 			next_dc->di.start = di.start;
1503 			dcc->undiscard_blks += di.len;
1504 			__relocate_discard_cmd(dcc, next_dc);
1505 			if (tdc)
1506 				__remove_discard_cmd(sbi, tdc);
1507 			merged = true;
1508 		}
1509 
1510 		if (!merged)
1511 			__insert_discard_cmd(sbi, bdev,
1512 						di.lstart, di.start, di.len);
1513  next:
1514 		prev_dc = next_dc;
1515 		if (!prev_dc)
1516 			break;
1517 
1518 		node = rb_next(&prev_dc->rb_node);
1519 		next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1520 	}
1521 }
1522 
1523 #ifdef CONFIG_BLK_DEV_ZONED
1524 static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi,
1525 		struct block_device *bdev, block_t blkstart, block_t lblkstart,
1526 		block_t blklen)
1527 {
1528 	trace_f2fs_queue_reset_zone(bdev, blkstart);
1529 
1530 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1531 	__insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen);
1532 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1533 }
1534 #endif
1535 
1536 static void __queue_discard_cmd(struct f2fs_sb_info *sbi,
1537 		struct block_device *bdev, block_t blkstart, block_t blklen)
1538 {
1539 	block_t lblkstart = blkstart;
1540 
1541 	if (!f2fs_bdev_support_discard(bdev))
1542 		return;
1543 
1544 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
1545 
1546 	if (f2fs_is_multi_device(sbi)) {
1547 		int devi = f2fs_target_device_index(sbi, blkstart);
1548 
1549 		blkstart -= FDEV(devi).start_blk;
1550 	}
1551 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1552 	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1553 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1554 }
1555 
1556 static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1557 		struct discard_policy *dpolicy, int *issued)
1558 {
1559 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1560 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1561 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1562 	struct discard_cmd *dc;
1563 	struct blk_plug plug;
1564 	bool io_interrupted = false;
1565 
1566 	mutex_lock(&dcc->cmd_lock);
1567 	dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos,
1568 				&prev_dc, &next_dc, &insert_p, &insert_parent);
1569 	if (!dc)
1570 		dc = next_dc;
1571 
1572 	blk_start_plug(&plug);
1573 
1574 	while (dc) {
1575 		struct rb_node *node;
1576 		int err = 0;
1577 
1578 		if (dc->state != D_PREP)
1579 			goto next;
1580 
1581 		if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1582 			io_interrupted = true;
1583 			break;
1584 		}
1585 
1586 		dcc->next_pos = dc->di.lstart + dc->di.len;
1587 		err = __submit_discard_cmd(sbi, dpolicy, dc, issued);
1588 
1589 		if (*issued >= dpolicy->max_requests)
1590 			break;
1591 next:
1592 		node = rb_next(&dc->rb_node);
1593 		if (err)
1594 			__remove_discard_cmd(sbi, dc);
1595 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1596 	}
1597 
1598 	blk_finish_plug(&plug);
1599 
1600 	if (!dc)
1601 		dcc->next_pos = 0;
1602 
1603 	mutex_unlock(&dcc->cmd_lock);
1604 
1605 	if (!(*issued) && io_interrupted)
1606 		*issued = -1;
1607 }
1608 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1609 					struct discard_policy *dpolicy);
1610 
1611 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1612 					struct discard_policy *dpolicy)
1613 {
1614 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1615 	struct list_head *pend_list;
1616 	struct discard_cmd *dc, *tmp;
1617 	struct blk_plug plug;
1618 	int i, issued;
1619 	bool io_interrupted = false;
1620 
1621 	if (dpolicy->timeout)
1622 		f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1623 
1624 retry:
1625 	issued = 0;
1626 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1627 		if (dpolicy->timeout &&
1628 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1629 			break;
1630 
1631 		if (i + 1 < dpolicy->granularity)
1632 			break;
1633 
1634 		if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) {
1635 			__issue_discard_cmd_orderly(sbi, dpolicy, &issued);
1636 			return issued;
1637 		}
1638 
1639 		pend_list = &dcc->pend_list[i];
1640 
1641 		mutex_lock(&dcc->cmd_lock);
1642 		if (list_empty(pend_list))
1643 			goto next;
1644 		if (unlikely(dcc->rbtree_check))
1645 			f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
1646 		blk_start_plug(&plug);
1647 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1648 			f2fs_bug_on(sbi, dc->state != D_PREP);
1649 
1650 			if (dpolicy->timeout &&
1651 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1652 				break;
1653 
1654 			if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1655 						!is_idle(sbi, DISCARD_TIME)) {
1656 				io_interrupted = true;
1657 				break;
1658 			}
1659 
1660 			__submit_discard_cmd(sbi, dpolicy, dc, &issued);
1661 
1662 			if (issued >= dpolicy->max_requests)
1663 				break;
1664 		}
1665 		blk_finish_plug(&plug);
1666 next:
1667 		mutex_unlock(&dcc->cmd_lock);
1668 
1669 		if (issued >= dpolicy->max_requests || io_interrupted)
1670 			break;
1671 	}
1672 
1673 	if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1674 		__wait_all_discard_cmd(sbi, dpolicy);
1675 		goto retry;
1676 	}
1677 
1678 	if (!issued && io_interrupted)
1679 		issued = -1;
1680 
1681 	return issued;
1682 }
1683 
1684 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1685 {
1686 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1687 	struct list_head *pend_list;
1688 	struct discard_cmd *dc, *tmp;
1689 	int i;
1690 	bool dropped = false;
1691 
1692 	mutex_lock(&dcc->cmd_lock);
1693 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1694 		pend_list = &dcc->pend_list[i];
1695 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1696 			f2fs_bug_on(sbi, dc->state != D_PREP);
1697 			__remove_discard_cmd(sbi, dc);
1698 			dropped = true;
1699 		}
1700 	}
1701 	mutex_unlock(&dcc->cmd_lock);
1702 
1703 	return dropped;
1704 }
1705 
1706 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1707 {
1708 	__drop_discard_cmd(sbi);
1709 }
1710 
1711 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1712 							struct discard_cmd *dc)
1713 {
1714 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1715 	unsigned int len = 0;
1716 
1717 	wait_for_completion_io(&dc->wait);
1718 	mutex_lock(&dcc->cmd_lock);
1719 	f2fs_bug_on(sbi, dc->state != D_DONE);
1720 	dc->ref--;
1721 	if (!dc->ref) {
1722 		if (!dc->error)
1723 			len = dc->di.len;
1724 		__remove_discard_cmd(sbi, dc);
1725 	}
1726 	mutex_unlock(&dcc->cmd_lock);
1727 
1728 	return len;
1729 }
1730 
1731 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1732 						struct discard_policy *dpolicy,
1733 						block_t start, block_t end)
1734 {
1735 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1736 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1737 					&(dcc->fstrim_list) : &(dcc->wait_list);
1738 	struct discard_cmd *dc = NULL, *iter, *tmp;
1739 	unsigned int trimmed = 0;
1740 
1741 next:
1742 	dc = NULL;
1743 
1744 	mutex_lock(&dcc->cmd_lock);
1745 	list_for_each_entry_safe(iter, tmp, wait_list, list) {
1746 		if (iter->di.lstart + iter->di.len <= start ||
1747 					end <= iter->di.lstart)
1748 			continue;
1749 		if (iter->di.len < dpolicy->granularity)
1750 			continue;
1751 		if (iter->state == D_DONE && !iter->ref) {
1752 			wait_for_completion_io(&iter->wait);
1753 			if (!iter->error)
1754 				trimmed += iter->di.len;
1755 			__remove_discard_cmd(sbi, iter);
1756 		} else {
1757 			iter->ref++;
1758 			dc = iter;
1759 			break;
1760 		}
1761 	}
1762 	mutex_unlock(&dcc->cmd_lock);
1763 
1764 	if (dc) {
1765 		trimmed += __wait_one_discard_bio(sbi, dc);
1766 		goto next;
1767 	}
1768 
1769 	return trimmed;
1770 }
1771 
1772 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1773 						struct discard_policy *dpolicy)
1774 {
1775 	struct discard_policy dp;
1776 	unsigned int discard_blks;
1777 
1778 	if (dpolicy)
1779 		return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1780 
1781 	/* wait all */
1782 	__init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY);
1783 	discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1784 	__init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY);
1785 	discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1786 
1787 	return discard_blks;
1788 }
1789 
1790 /* This should be covered by global mutex, &sit_i->sentry_lock */
1791 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1792 {
1793 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1794 	struct discard_cmd *dc;
1795 	bool need_wait = false;
1796 
1797 	mutex_lock(&dcc->cmd_lock);
1798 	dc = __lookup_discard_cmd(sbi, blkaddr);
1799 #ifdef CONFIG_BLK_DEV_ZONED
1800 	if (dc && f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) {
1801 		int devi = f2fs_bdev_index(sbi, dc->bdev);
1802 
1803 		if (devi < 0) {
1804 			mutex_unlock(&dcc->cmd_lock);
1805 			return;
1806 		}
1807 
1808 		if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1809 			/* force submit zone reset */
1810 			if (dc->state == D_PREP)
1811 				__submit_zone_reset_cmd(sbi, dc, REQ_SYNC,
1812 							&dcc->wait_list, NULL);
1813 			dc->ref++;
1814 			mutex_unlock(&dcc->cmd_lock);
1815 			/* wait zone reset */
1816 			__wait_one_discard_bio(sbi, dc);
1817 			return;
1818 		}
1819 	}
1820 #endif
1821 	if (dc) {
1822 		if (dc->state == D_PREP) {
1823 			__punch_discard_cmd(sbi, dc, blkaddr);
1824 		} else {
1825 			dc->ref++;
1826 			need_wait = true;
1827 		}
1828 	}
1829 	mutex_unlock(&dcc->cmd_lock);
1830 
1831 	if (need_wait)
1832 		__wait_one_discard_bio(sbi, dc);
1833 }
1834 
1835 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1836 {
1837 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1838 
1839 	if (dcc && dcc->f2fs_issue_discard) {
1840 		struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1841 
1842 		dcc->f2fs_issue_discard = NULL;
1843 		kthread_stop(discard_thread);
1844 	}
1845 }
1846 
1847 /**
1848  * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT
1849  * @sbi: the f2fs_sb_info data for discard cmd to issue
1850  *
1851  * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped
1852  *
1853  * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false.
1854  */
1855 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1856 {
1857 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1858 	struct discard_policy dpolicy;
1859 	bool dropped;
1860 
1861 	if (!atomic_read(&dcc->discard_cmd_cnt))
1862 		return true;
1863 
1864 	__init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1865 					dcc->discard_granularity);
1866 	__issue_discard_cmd(sbi, &dpolicy);
1867 	dropped = __drop_discard_cmd(sbi);
1868 
1869 	/* just to make sure there is no pending discard commands */
1870 	__wait_all_discard_cmd(sbi, NULL);
1871 
1872 	f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1873 	return !dropped;
1874 }
1875 
1876 static int issue_discard_thread(void *data)
1877 {
1878 	struct f2fs_sb_info *sbi = data;
1879 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1880 	wait_queue_head_t *q = &dcc->discard_wait_queue;
1881 	struct discard_policy dpolicy;
1882 	unsigned int wait_ms = dcc->min_discard_issue_time;
1883 	int issued;
1884 
1885 	set_freezable();
1886 
1887 	do {
1888 		wait_event_interruptible_timeout(*q,
1889 				kthread_should_stop() || freezing(current) ||
1890 				dcc->discard_wake,
1891 				msecs_to_jiffies(wait_ms));
1892 
1893 		if (sbi->gc_mode == GC_URGENT_HIGH ||
1894 			!f2fs_available_free_memory(sbi, DISCARD_CACHE))
1895 			__init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE,
1896 						MIN_DISCARD_GRANULARITY);
1897 		else
1898 			__init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1899 						dcc->discard_granularity);
1900 
1901 		if (dcc->discard_wake)
1902 			dcc->discard_wake = false;
1903 
1904 		/* clean up pending candidates before going to sleep */
1905 		if (atomic_read(&dcc->queued_discard))
1906 			__wait_all_discard_cmd(sbi, NULL);
1907 
1908 		if (try_to_freeze())
1909 			continue;
1910 		if (f2fs_readonly(sbi->sb))
1911 			continue;
1912 		if (kthread_should_stop())
1913 			return 0;
1914 		if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) ||
1915 			!atomic_read(&dcc->discard_cmd_cnt)) {
1916 			wait_ms = dpolicy.max_interval;
1917 			continue;
1918 		}
1919 
1920 		sb_start_intwrite(sbi->sb);
1921 
1922 		issued = __issue_discard_cmd(sbi, &dpolicy);
1923 		if (issued > 0) {
1924 			__wait_all_discard_cmd(sbi, &dpolicy);
1925 			wait_ms = dpolicy.min_interval;
1926 		} else if (issued == -1) {
1927 			wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1928 			if (!wait_ms)
1929 				wait_ms = dpolicy.mid_interval;
1930 		} else {
1931 			wait_ms = dpolicy.max_interval;
1932 		}
1933 		if (!atomic_read(&dcc->discard_cmd_cnt))
1934 			wait_ms = dpolicy.max_interval;
1935 
1936 		sb_end_intwrite(sbi->sb);
1937 
1938 	} while (!kthread_should_stop());
1939 	return 0;
1940 }
1941 
1942 #ifdef CONFIG_BLK_DEV_ZONED
1943 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1944 		struct block_device *bdev, block_t blkstart, block_t blklen)
1945 {
1946 	sector_t sector, nr_sects;
1947 	block_t lblkstart = blkstart;
1948 	int devi = 0;
1949 	u64 remainder = 0;
1950 
1951 	if (f2fs_is_multi_device(sbi)) {
1952 		devi = f2fs_target_device_index(sbi, blkstart);
1953 		if (blkstart < FDEV(devi).start_blk ||
1954 		    blkstart > FDEV(devi).end_blk) {
1955 			f2fs_err(sbi, "Invalid block %x", blkstart);
1956 			return -EIO;
1957 		}
1958 		blkstart -= FDEV(devi).start_blk;
1959 	}
1960 
1961 	/* For sequential zones, reset the zone write pointer */
1962 	if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1963 		sector = SECTOR_FROM_BLOCK(blkstart);
1964 		nr_sects = SECTOR_FROM_BLOCK(blklen);
1965 		div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder);
1966 
1967 		if (remainder || nr_sects != bdev_zone_sectors(bdev)) {
1968 			f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1969 				 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1970 				 blkstart, blklen);
1971 			return -EIO;
1972 		}
1973 
1974 		if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) {
1975 			trace_f2fs_issue_reset_zone(bdev, blkstart);
1976 			return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1977 						sector, nr_sects, GFP_NOFS);
1978 		}
1979 
1980 		__queue_zone_reset_cmd(sbi, bdev, blkstart, lblkstart, blklen);
1981 		return 0;
1982 	}
1983 
1984 	/* For conventional zones, use regular discard if supported */
1985 	__queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1986 	return 0;
1987 }
1988 #endif
1989 
1990 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1991 		struct block_device *bdev, block_t blkstart, block_t blklen)
1992 {
1993 #ifdef CONFIG_BLK_DEV_ZONED
1994 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1995 		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1996 #endif
1997 	__queue_discard_cmd(sbi, bdev, blkstart, blklen);
1998 	return 0;
1999 }
2000 
2001 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
2002 				block_t blkstart, block_t blklen)
2003 {
2004 	sector_t start = blkstart, len = 0;
2005 	struct block_device *bdev;
2006 	struct seg_entry *se;
2007 	unsigned int offset;
2008 	block_t i;
2009 	int err = 0;
2010 
2011 	bdev = f2fs_target_device(sbi, blkstart, NULL);
2012 
2013 	for (i = blkstart; i < blkstart + blklen; i++, len++) {
2014 		if (i != start) {
2015 			struct block_device *bdev2 =
2016 				f2fs_target_device(sbi, i, NULL);
2017 
2018 			if (bdev2 != bdev) {
2019 				err = __issue_discard_async(sbi, bdev,
2020 						start, len);
2021 				if (err)
2022 					return err;
2023 				bdev = bdev2;
2024 				start = i;
2025 				len = 0;
2026 			}
2027 		}
2028 
2029 		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
2030 		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
2031 
2032 		if (f2fs_block_unit_discard(sbi) &&
2033 				!f2fs_test_and_set_bit(offset, se->discard_map))
2034 			sbi->discard_blks--;
2035 	}
2036 
2037 	if (len)
2038 		err = __issue_discard_async(sbi, bdev, start, len);
2039 	return err;
2040 }
2041 
2042 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
2043 							bool check_only)
2044 {
2045 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2046 	int max_blocks = sbi->blocks_per_seg;
2047 	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
2048 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2049 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2050 	unsigned long *discard_map = (unsigned long *)se->discard_map;
2051 	unsigned long *dmap = SIT_I(sbi)->tmp_map;
2052 	unsigned int start = 0, end = -1;
2053 	bool force = (cpc->reason & CP_DISCARD);
2054 	struct discard_entry *de = NULL;
2055 	struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
2056 	int i;
2057 
2058 	if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi) ||
2059 			!f2fs_block_unit_discard(sbi))
2060 		return false;
2061 
2062 	if (!force) {
2063 		if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
2064 			SM_I(sbi)->dcc_info->nr_discards >=
2065 				SM_I(sbi)->dcc_info->max_discards)
2066 			return false;
2067 	}
2068 
2069 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
2070 	for (i = 0; i < entries; i++)
2071 		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
2072 				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
2073 
2074 	while (force || SM_I(sbi)->dcc_info->nr_discards <=
2075 				SM_I(sbi)->dcc_info->max_discards) {
2076 		start = __find_rev_next_bit(dmap, max_blocks, end + 1);
2077 		if (start >= max_blocks)
2078 			break;
2079 
2080 		end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
2081 		if (force && start && end != max_blocks
2082 					&& (end - start) < cpc->trim_minlen)
2083 			continue;
2084 
2085 		if (check_only)
2086 			return true;
2087 
2088 		if (!de) {
2089 			de = f2fs_kmem_cache_alloc(discard_entry_slab,
2090 						GFP_F2FS_ZERO, true, NULL);
2091 			de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
2092 			list_add_tail(&de->list, head);
2093 		}
2094 
2095 		for (i = start; i < end; i++)
2096 			__set_bit_le(i, (void *)de->discard_map);
2097 
2098 		SM_I(sbi)->dcc_info->nr_discards += end - start;
2099 	}
2100 	return false;
2101 }
2102 
2103 static void release_discard_addr(struct discard_entry *entry)
2104 {
2105 	list_del(&entry->list);
2106 	kmem_cache_free(discard_entry_slab, entry);
2107 }
2108 
2109 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
2110 {
2111 	struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
2112 	struct discard_entry *entry, *this;
2113 
2114 	/* drop caches */
2115 	list_for_each_entry_safe(entry, this, head, list)
2116 		release_discard_addr(entry);
2117 }
2118 
2119 /*
2120  * Should call f2fs_clear_prefree_segments after checkpoint is done.
2121  */
2122 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
2123 {
2124 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2125 	unsigned int segno;
2126 
2127 	mutex_lock(&dirty_i->seglist_lock);
2128 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
2129 		__set_test_and_free(sbi, segno, false);
2130 	mutex_unlock(&dirty_i->seglist_lock);
2131 }
2132 
2133 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
2134 						struct cp_control *cpc)
2135 {
2136 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2137 	struct list_head *head = &dcc->entry_list;
2138 	struct discard_entry *entry, *this;
2139 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2140 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
2141 	unsigned int start = 0, end = -1;
2142 	unsigned int secno, start_segno;
2143 	bool force = (cpc->reason & CP_DISCARD);
2144 	bool section_alignment = F2FS_OPTION(sbi).discard_unit ==
2145 						DISCARD_UNIT_SECTION;
2146 
2147 	if (f2fs_lfs_mode(sbi) && __is_large_section(sbi))
2148 		section_alignment = true;
2149 
2150 	mutex_lock(&dirty_i->seglist_lock);
2151 
2152 	while (1) {
2153 		int i;
2154 
2155 		if (section_alignment && end != -1)
2156 			end--;
2157 		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
2158 		if (start >= MAIN_SEGS(sbi))
2159 			break;
2160 		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
2161 								start + 1);
2162 
2163 		if (section_alignment) {
2164 			start = rounddown(start, sbi->segs_per_sec);
2165 			end = roundup(end, sbi->segs_per_sec);
2166 		}
2167 
2168 		for (i = start; i < end; i++) {
2169 			if (test_and_clear_bit(i, prefree_map))
2170 				dirty_i->nr_dirty[PRE]--;
2171 		}
2172 
2173 		if (!f2fs_realtime_discard_enable(sbi))
2174 			continue;
2175 
2176 		if (force && start >= cpc->trim_start &&
2177 					(end - 1) <= cpc->trim_end)
2178 			continue;
2179 
2180 		/* Should cover 2MB zoned device for zone-based reset */
2181 		if (!f2fs_sb_has_blkzoned(sbi) &&
2182 		    (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) {
2183 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2184 				(end - start) << sbi->log_blocks_per_seg);
2185 			continue;
2186 		}
2187 next:
2188 		secno = GET_SEC_FROM_SEG(sbi, start);
2189 		start_segno = GET_SEG_FROM_SEC(sbi, secno);
2190 		if (!IS_CURSEC(sbi, secno) &&
2191 			!get_valid_blocks(sbi, start, true))
2192 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2193 				sbi->segs_per_sec << sbi->log_blocks_per_seg);
2194 
2195 		start = start_segno + sbi->segs_per_sec;
2196 		if (start < end)
2197 			goto next;
2198 		else
2199 			end = start - 1;
2200 	}
2201 	mutex_unlock(&dirty_i->seglist_lock);
2202 
2203 	if (!f2fs_block_unit_discard(sbi))
2204 		goto wakeup;
2205 
2206 	/* send small discards */
2207 	list_for_each_entry_safe(entry, this, head, list) {
2208 		unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2209 		bool is_valid = test_bit_le(0, entry->discard_map);
2210 
2211 find_next:
2212 		if (is_valid) {
2213 			next_pos = find_next_zero_bit_le(entry->discard_map,
2214 					sbi->blocks_per_seg, cur_pos);
2215 			len = next_pos - cur_pos;
2216 
2217 			if (f2fs_sb_has_blkzoned(sbi) ||
2218 			    (force && len < cpc->trim_minlen))
2219 				goto skip;
2220 
2221 			f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2222 									len);
2223 			total_len += len;
2224 		} else {
2225 			next_pos = find_next_bit_le(entry->discard_map,
2226 					sbi->blocks_per_seg, cur_pos);
2227 		}
2228 skip:
2229 		cur_pos = next_pos;
2230 		is_valid = !is_valid;
2231 
2232 		if (cur_pos < sbi->blocks_per_seg)
2233 			goto find_next;
2234 
2235 		release_discard_addr(entry);
2236 		dcc->nr_discards -= total_len;
2237 	}
2238 
2239 wakeup:
2240 	wake_up_discard_thread(sbi, false);
2241 }
2242 
2243 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi)
2244 {
2245 	dev_t dev = sbi->sb->s_bdev->bd_dev;
2246 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2247 	int err = 0;
2248 
2249 	if (!f2fs_realtime_discard_enable(sbi))
2250 		return 0;
2251 
2252 	dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2253 				"f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2254 	if (IS_ERR(dcc->f2fs_issue_discard)) {
2255 		err = PTR_ERR(dcc->f2fs_issue_discard);
2256 		dcc->f2fs_issue_discard = NULL;
2257 	}
2258 
2259 	return err;
2260 }
2261 
2262 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2263 {
2264 	struct discard_cmd_control *dcc;
2265 	int err = 0, i;
2266 
2267 	if (SM_I(sbi)->dcc_info) {
2268 		dcc = SM_I(sbi)->dcc_info;
2269 		goto init_thread;
2270 	}
2271 
2272 	dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2273 	if (!dcc)
2274 		return -ENOMEM;
2275 
2276 	dcc->discard_io_aware_gran = MAX_PLIST_NUM;
2277 	dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2278 	dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY;
2279 	if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT)
2280 		dcc->discard_granularity = sbi->blocks_per_seg;
2281 	else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION)
2282 		dcc->discard_granularity = BLKS_PER_SEC(sbi);
2283 
2284 	INIT_LIST_HEAD(&dcc->entry_list);
2285 	for (i = 0; i < MAX_PLIST_NUM; i++)
2286 		INIT_LIST_HEAD(&dcc->pend_list[i]);
2287 	INIT_LIST_HEAD(&dcc->wait_list);
2288 	INIT_LIST_HEAD(&dcc->fstrim_list);
2289 	mutex_init(&dcc->cmd_lock);
2290 	atomic_set(&dcc->issued_discard, 0);
2291 	atomic_set(&dcc->queued_discard, 0);
2292 	atomic_set(&dcc->discard_cmd_cnt, 0);
2293 	dcc->nr_discards = 0;
2294 	dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2295 	dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST;
2296 	dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME;
2297 	dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME;
2298 	dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME;
2299 	dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL;
2300 	dcc->undiscard_blks = 0;
2301 	dcc->next_pos = 0;
2302 	dcc->root = RB_ROOT_CACHED;
2303 	dcc->rbtree_check = false;
2304 
2305 	init_waitqueue_head(&dcc->discard_wait_queue);
2306 	SM_I(sbi)->dcc_info = dcc;
2307 init_thread:
2308 	err = f2fs_start_discard_thread(sbi);
2309 	if (err) {
2310 		kfree(dcc);
2311 		SM_I(sbi)->dcc_info = NULL;
2312 	}
2313 
2314 	return err;
2315 }
2316 
2317 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2318 {
2319 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2320 
2321 	if (!dcc)
2322 		return;
2323 
2324 	f2fs_stop_discard_thread(sbi);
2325 
2326 	/*
2327 	 * Recovery can cache discard commands, so in error path of
2328 	 * fill_super(), it needs to give a chance to handle them.
2329 	 */
2330 	f2fs_issue_discard_timeout(sbi);
2331 
2332 	kfree(dcc);
2333 	SM_I(sbi)->dcc_info = NULL;
2334 }
2335 
2336 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2337 {
2338 	struct sit_info *sit_i = SIT_I(sbi);
2339 
2340 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2341 		sit_i->dirty_sentries++;
2342 		return false;
2343 	}
2344 
2345 	return true;
2346 }
2347 
2348 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2349 					unsigned int segno, int modified)
2350 {
2351 	struct seg_entry *se = get_seg_entry(sbi, segno);
2352 
2353 	se->type = type;
2354 	if (modified)
2355 		__mark_sit_entry_dirty(sbi, segno);
2356 }
2357 
2358 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2359 								block_t blkaddr)
2360 {
2361 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
2362 
2363 	if (segno == NULL_SEGNO)
2364 		return 0;
2365 	return get_seg_entry(sbi, segno)->mtime;
2366 }
2367 
2368 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2369 						unsigned long long old_mtime)
2370 {
2371 	struct seg_entry *se;
2372 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
2373 	unsigned long long ctime = get_mtime(sbi, false);
2374 	unsigned long long mtime = old_mtime ? old_mtime : ctime;
2375 
2376 	if (segno == NULL_SEGNO)
2377 		return;
2378 
2379 	se = get_seg_entry(sbi, segno);
2380 
2381 	if (!se->mtime)
2382 		se->mtime = mtime;
2383 	else
2384 		se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2385 						se->valid_blocks + 1);
2386 
2387 	if (ctime > SIT_I(sbi)->max_mtime)
2388 		SIT_I(sbi)->max_mtime = ctime;
2389 }
2390 
2391 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2392 {
2393 	struct seg_entry *se;
2394 	unsigned int segno, offset;
2395 	long int new_vblocks;
2396 	bool exist;
2397 #ifdef CONFIG_F2FS_CHECK_FS
2398 	bool mir_exist;
2399 #endif
2400 
2401 	segno = GET_SEGNO(sbi, blkaddr);
2402 
2403 	se = get_seg_entry(sbi, segno);
2404 	new_vblocks = se->valid_blocks + del;
2405 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2406 
2407 	f2fs_bug_on(sbi, (new_vblocks < 0 ||
2408 			(new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2409 
2410 	se->valid_blocks = new_vblocks;
2411 
2412 	/* Update valid block bitmap */
2413 	if (del > 0) {
2414 		exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2415 #ifdef CONFIG_F2FS_CHECK_FS
2416 		mir_exist = f2fs_test_and_set_bit(offset,
2417 						se->cur_valid_map_mir);
2418 		if (unlikely(exist != mir_exist)) {
2419 			f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2420 				 blkaddr, exist);
2421 			f2fs_bug_on(sbi, 1);
2422 		}
2423 #endif
2424 		if (unlikely(exist)) {
2425 			f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2426 				 blkaddr);
2427 			f2fs_bug_on(sbi, 1);
2428 			se->valid_blocks--;
2429 			del = 0;
2430 		}
2431 
2432 		if (f2fs_block_unit_discard(sbi) &&
2433 				!f2fs_test_and_set_bit(offset, se->discard_map))
2434 			sbi->discard_blks--;
2435 
2436 		/*
2437 		 * SSR should never reuse block which is checkpointed
2438 		 * or newly invalidated.
2439 		 */
2440 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2441 			if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2442 				se->ckpt_valid_blocks++;
2443 		}
2444 	} else {
2445 		exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2446 #ifdef CONFIG_F2FS_CHECK_FS
2447 		mir_exist = f2fs_test_and_clear_bit(offset,
2448 						se->cur_valid_map_mir);
2449 		if (unlikely(exist != mir_exist)) {
2450 			f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2451 				 blkaddr, exist);
2452 			f2fs_bug_on(sbi, 1);
2453 		}
2454 #endif
2455 		if (unlikely(!exist)) {
2456 			f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2457 				 blkaddr);
2458 			f2fs_bug_on(sbi, 1);
2459 			se->valid_blocks++;
2460 			del = 0;
2461 		} else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2462 			/*
2463 			 * If checkpoints are off, we must not reuse data that
2464 			 * was used in the previous checkpoint. If it was used
2465 			 * before, we must track that to know how much space we
2466 			 * really have.
2467 			 */
2468 			if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2469 				spin_lock(&sbi->stat_lock);
2470 				sbi->unusable_block_count++;
2471 				spin_unlock(&sbi->stat_lock);
2472 			}
2473 		}
2474 
2475 		if (f2fs_block_unit_discard(sbi) &&
2476 			f2fs_test_and_clear_bit(offset, se->discard_map))
2477 			sbi->discard_blks++;
2478 	}
2479 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2480 		se->ckpt_valid_blocks += del;
2481 
2482 	__mark_sit_entry_dirty(sbi, segno);
2483 
2484 	/* update total number of valid blocks to be written in ckpt area */
2485 	SIT_I(sbi)->written_valid_blocks += del;
2486 
2487 	if (__is_large_section(sbi))
2488 		get_sec_entry(sbi, segno)->valid_blocks += del;
2489 }
2490 
2491 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2492 {
2493 	unsigned int segno = GET_SEGNO(sbi, addr);
2494 	struct sit_info *sit_i = SIT_I(sbi);
2495 
2496 	f2fs_bug_on(sbi, addr == NULL_ADDR);
2497 	if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2498 		return;
2499 
2500 	f2fs_invalidate_internal_cache(sbi, addr);
2501 
2502 	/* add it into sit main buffer */
2503 	down_write(&sit_i->sentry_lock);
2504 
2505 	update_segment_mtime(sbi, addr, 0);
2506 	update_sit_entry(sbi, addr, -1);
2507 
2508 	/* add it into dirty seglist */
2509 	locate_dirty_segment(sbi, segno);
2510 
2511 	up_write(&sit_i->sentry_lock);
2512 }
2513 
2514 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2515 {
2516 	struct sit_info *sit_i = SIT_I(sbi);
2517 	unsigned int segno, offset;
2518 	struct seg_entry *se;
2519 	bool is_cp = false;
2520 
2521 	if (!__is_valid_data_blkaddr(blkaddr))
2522 		return true;
2523 
2524 	down_read(&sit_i->sentry_lock);
2525 
2526 	segno = GET_SEGNO(sbi, blkaddr);
2527 	se = get_seg_entry(sbi, segno);
2528 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2529 
2530 	if (f2fs_test_bit(offset, se->ckpt_valid_map))
2531 		is_cp = true;
2532 
2533 	up_read(&sit_i->sentry_lock);
2534 
2535 	return is_cp;
2536 }
2537 
2538 static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type)
2539 {
2540 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2541 
2542 	if (sbi->ckpt->alloc_type[type] == SSR)
2543 		return sbi->blocks_per_seg;
2544 	return curseg->next_blkoff;
2545 }
2546 
2547 /*
2548  * Calculate the number of current summary pages for writing
2549  */
2550 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2551 {
2552 	int valid_sum_count = 0;
2553 	int i, sum_in_page;
2554 
2555 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2556 		if (sbi->ckpt->alloc_type[i] != SSR && for_ra)
2557 			valid_sum_count +=
2558 				le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2559 		else
2560 			valid_sum_count += f2fs_curseg_valid_blocks(sbi, i);
2561 	}
2562 
2563 	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2564 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2565 	if (valid_sum_count <= sum_in_page)
2566 		return 1;
2567 	else if ((valid_sum_count - sum_in_page) <=
2568 		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2569 		return 2;
2570 	return 3;
2571 }
2572 
2573 /*
2574  * Caller should put this summary page
2575  */
2576 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2577 {
2578 	if (unlikely(f2fs_cp_error(sbi)))
2579 		return ERR_PTR(-EIO);
2580 	return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2581 }
2582 
2583 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2584 					void *src, block_t blk_addr)
2585 {
2586 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2587 
2588 	memcpy(page_address(page), src, PAGE_SIZE);
2589 	set_page_dirty(page);
2590 	f2fs_put_page(page, 1);
2591 }
2592 
2593 static void write_sum_page(struct f2fs_sb_info *sbi,
2594 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
2595 {
2596 	f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2597 }
2598 
2599 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2600 						int type, block_t blk_addr)
2601 {
2602 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2603 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2604 	struct f2fs_summary_block *src = curseg->sum_blk;
2605 	struct f2fs_summary_block *dst;
2606 
2607 	dst = (struct f2fs_summary_block *)page_address(page);
2608 	memset(dst, 0, PAGE_SIZE);
2609 
2610 	mutex_lock(&curseg->curseg_mutex);
2611 
2612 	down_read(&curseg->journal_rwsem);
2613 	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2614 	up_read(&curseg->journal_rwsem);
2615 
2616 	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2617 	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2618 
2619 	mutex_unlock(&curseg->curseg_mutex);
2620 
2621 	set_page_dirty(page);
2622 	f2fs_put_page(page, 1);
2623 }
2624 
2625 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2626 				struct curseg_info *curseg, int type)
2627 {
2628 	unsigned int segno = curseg->segno + 1;
2629 	struct free_segmap_info *free_i = FREE_I(sbi);
2630 
2631 	if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2632 		return !test_bit(segno, free_i->free_segmap);
2633 	return 0;
2634 }
2635 
2636 /*
2637  * Find a new segment from the free segments bitmap to right order
2638  * This function should be returned with success, otherwise BUG
2639  */
2640 static void get_new_segment(struct f2fs_sb_info *sbi,
2641 			unsigned int *newseg, bool new_sec, int dir)
2642 {
2643 	struct free_segmap_info *free_i = FREE_I(sbi);
2644 	unsigned int segno, secno, zoneno;
2645 	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2646 	unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2647 	unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2648 	unsigned int left_start = hint;
2649 	bool init = true;
2650 	int go_left = 0;
2651 	int i;
2652 
2653 	spin_lock(&free_i->segmap_lock);
2654 
2655 	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2656 		segno = find_next_zero_bit(free_i->free_segmap,
2657 			GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2658 		if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2659 			goto got_it;
2660 	}
2661 find_other_zone:
2662 	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2663 	if (secno >= MAIN_SECS(sbi)) {
2664 		if (dir == ALLOC_RIGHT) {
2665 			secno = find_first_zero_bit(free_i->free_secmap,
2666 							MAIN_SECS(sbi));
2667 			f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2668 		} else {
2669 			go_left = 1;
2670 			left_start = hint - 1;
2671 		}
2672 	}
2673 	if (go_left == 0)
2674 		goto skip_left;
2675 
2676 	while (test_bit(left_start, free_i->free_secmap)) {
2677 		if (left_start > 0) {
2678 			left_start--;
2679 			continue;
2680 		}
2681 		left_start = find_first_zero_bit(free_i->free_secmap,
2682 							MAIN_SECS(sbi));
2683 		f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2684 		break;
2685 	}
2686 	secno = left_start;
2687 skip_left:
2688 	segno = GET_SEG_FROM_SEC(sbi, secno);
2689 	zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2690 
2691 	/* give up on finding another zone */
2692 	if (!init)
2693 		goto got_it;
2694 	if (sbi->secs_per_zone == 1)
2695 		goto got_it;
2696 	if (zoneno == old_zoneno)
2697 		goto got_it;
2698 	if (dir == ALLOC_LEFT) {
2699 		if (!go_left && zoneno + 1 >= total_zones)
2700 			goto got_it;
2701 		if (go_left && zoneno == 0)
2702 			goto got_it;
2703 	}
2704 	for (i = 0; i < NR_CURSEG_TYPE; i++)
2705 		if (CURSEG_I(sbi, i)->zone == zoneno)
2706 			break;
2707 
2708 	if (i < NR_CURSEG_TYPE) {
2709 		/* zone is in user, try another */
2710 		if (go_left)
2711 			hint = zoneno * sbi->secs_per_zone - 1;
2712 		else if (zoneno + 1 >= total_zones)
2713 			hint = 0;
2714 		else
2715 			hint = (zoneno + 1) * sbi->secs_per_zone;
2716 		init = false;
2717 		goto find_other_zone;
2718 	}
2719 got_it:
2720 	/* set it as dirty segment in free segmap */
2721 	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2722 	__set_inuse(sbi, segno);
2723 	*newseg = segno;
2724 	spin_unlock(&free_i->segmap_lock);
2725 }
2726 
2727 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2728 {
2729 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2730 	struct summary_footer *sum_footer;
2731 	unsigned short seg_type = curseg->seg_type;
2732 
2733 	curseg->inited = true;
2734 	curseg->segno = curseg->next_segno;
2735 	curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2736 	curseg->next_blkoff = 0;
2737 	curseg->next_segno = NULL_SEGNO;
2738 
2739 	sum_footer = &(curseg->sum_blk->footer);
2740 	memset(sum_footer, 0, sizeof(struct summary_footer));
2741 
2742 	sanity_check_seg_type(sbi, seg_type);
2743 
2744 	if (IS_DATASEG(seg_type))
2745 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2746 	if (IS_NODESEG(seg_type))
2747 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2748 	__set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2749 }
2750 
2751 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2752 {
2753 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2754 	unsigned short seg_type = curseg->seg_type;
2755 
2756 	sanity_check_seg_type(sbi, seg_type);
2757 	if (f2fs_need_rand_seg(sbi))
2758 		return get_random_u32_below(MAIN_SECS(sbi) * sbi->segs_per_sec);
2759 
2760 	/* if segs_per_sec is large than 1, we need to keep original policy. */
2761 	if (__is_large_section(sbi))
2762 		return curseg->segno;
2763 
2764 	/* inmem log may not locate on any segment after mount */
2765 	if (!curseg->inited)
2766 		return 0;
2767 
2768 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2769 		return 0;
2770 
2771 	if (test_opt(sbi, NOHEAP) &&
2772 		(seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)))
2773 		return 0;
2774 
2775 	if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2776 		return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2777 
2778 	/* find segments from 0 to reuse freed segments */
2779 	if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2780 		return 0;
2781 
2782 	return curseg->segno;
2783 }
2784 
2785 /*
2786  * Allocate a current working segment.
2787  * This function always allocates a free segment in LFS manner.
2788  */
2789 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2790 {
2791 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2792 	unsigned short seg_type = curseg->seg_type;
2793 	unsigned int segno = curseg->segno;
2794 	int dir = ALLOC_LEFT;
2795 
2796 	if (curseg->inited)
2797 		write_sum_page(sbi, curseg->sum_blk,
2798 				GET_SUM_BLOCK(sbi, segno));
2799 	if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA)
2800 		dir = ALLOC_RIGHT;
2801 
2802 	if (test_opt(sbi, NOHEAP))
2803 		dir = ALLOC_RIGHT;
2804 
2805 	segno = __get_next_segno(sbi, type);
2806 	get_new_segment(sbi, &segno, new_sec, dir);
2807 	curseg->next_segno = segno;
2808 	reset_curseg(sbi, type, 1);
2809 	curseg->alloc_type = LFS;
2810 	if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
2811 		curseg->fragment_remained_chunk =
2812 				get_random_u32_inclusive(1, sbi->max_fragment_chunk);
2813 }
2814 
2815 static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2816 					int segno, block_t start)
2817 {
2818 	struct seg_entry *se = get_seg_entry(sbi, segno);
2819 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2820 	unsigned long *target_map = SIT_I(sbi)->tmp_map;
2821 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2822 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2823 	int i;
2824 
2825 	for (i = 0; i < entries; i++)
2826 		target_map[i] = ckpt_map[i] | cur_map[i];
2827 
2828 	return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2829 }
2830 
2831 static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi,
2832 		struct curseg_info *seg)
2833 {
2834 	return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1);
2835 }
2836 
2837 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2838 {
2839 	return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg;
2840 }
2841 
2842 /*
2843  * This function always allocates a used segment(from dirty seglist) by SSR
2844  * manner, so it should recover the existing segment information of valid blocks
2845  */
2846 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2847 {
2848 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2849 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2850 	unsigned int new_segno = curseg->next_segno;
2851 	struct f2fs_summary_block *sum_node;
2852 	struct page *sum_page;
2853 
2854 	write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, curseg->segno));
2855 
2856 	__set_test_and_inuse(sbi, new_segno);
2857 
2858 	mutex_lock(&dirty_i->seglist_lock);
2859 	__remove_dirty_segment(sbi, new_segno, PRE);
2860 	__remove_dirty_segment(sbi, new_segno, DIRTY);
2861 	mutex_unlock(&dirty_i->seglist_lock);
2862 
2863 	reset_curseg(sbi, type, 1);
2864 	curseg->alloc_type = SSR;
2865 	curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
2866 
2867 	sum_page = f2fs_get_sum_page(sbi, new_segno);
2868 	if (IS_ERR(sum_page)) {
2869 		/* GC won't be able to use stale summary pages by cp_error */
2870 		memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2871 		return;
2872 	}
2873 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2874 	memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2875 	f2fs_put_page(sum_page, 1);
2876 }
2877 
2878 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2879 				int alloc_mode, unsigned long long age);
2880 
2881 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2882 					int target_type, int alloc_mode,
2883 					unsigned long long age)
2884 {
2885 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2886 
2887 	curseg->seg_type = target_type;
2888 
2889 	if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2890 		struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2891 
2892 		curseg->seg_type = se->type;
2893 		change_curseg(sbi, type);
2894 	} else {
2895 		/* allocate cold segment by default */
2896 		curseg->seg_type = CURSEG_COLD_DATA;
2897 		new_curseg(sbi, type, true);
2898 	}
2899 	stat_inc_seg_type(sbi, curseg);
2900 }
2901 
2902 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
2903 {
2904 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2905 
2906 	if (!sbi->am.atgc_enabled)
2907 		return;
2908 
2909 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
2910 
2911 	mutex_lock(&curseg->curseg_mutex);
2912 	down_write(&SIT_I(sbi)->sentry_lock);
2913 
2914 	get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0);
2915 
2916 	up_write(&SIT_I(sbi)->sentry_lock);
2917 	mutex_unlock(&curseg->curseg_mutex);
2918 
2919 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
2920 
2921 }
2922 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2923 {
2924 	__f2fs_init_atgc_curseg(sbi);
2925 }
2926 
2927 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2928 {
2929 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2930 
2931 	mutex_lock(&curseg->curseg_mutex);
2932 	if (!curseg->inited)
2933 		goto out;
2934 
2935 	if (get_valid_blocks(sbi, curseg->segno, false)) {
2936 		write_sum_page(sbi, curseg->sum_blk,
2937 				GET_SUM_BLOCK(sbi, curseg->segno));
2938 	} else {
2939 		mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2940 		__set_test_and_free(sbi, curseg->segno, true);
2941 		mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2942 	}
2943 out:
2944 	mutex_unlock(&curseg->curseg_mutex);
2945 }
2946 
2947 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
2948 {
2949 	__f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2950 
2951 	if (sbi->am.atgc_enabled)
2952 		__f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2953 }
2954 
2955 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2956 {
2957 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2958 
2959 	mutex_lock(&curseg->curseg_mutex);
2960 	if (!curseg->inited)
2961 		goto out;
2962 	if (get_valid_blocks(sbi, curseg->segno, false))
2963 		goto out;
2964 
2965 	mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2966 	__set_test_and_inuse(sbi, curseg->segno);
2967 	mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2968 out:
2969 	mutex_unlock(&curseg->curseg_mutex);
2970 }
2971 
2972 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
2973 {
2974 	__f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2975 
2976 	if (sbi->am.atgc_enabled)
2977 		__f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2978 }
2979 
2980 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2981 				int alloc_mode, unsigned long long age)
2982 {
2983 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2984 	unsigned segno = NULL_SEGNO;
2985 	unsigned short seg_type = curseg->seg_type;
2986 	int i, cnt;
2987 	bool reversed = false;
2988 
2989 	sanity_check_seg_type(sbi, seg_type);
2990 
2991 	/* f2fs_need_SSR() already forces to do this */
2992 	if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
2993 		curseg->next_segno = segno;
2994 		return 1;
2995 	}
2996 
2997 	/* For node segments, let's do SSR more intensively */
2998 	if (IS_NODESEG(seg_type)) {
2999 		if (seg_type >= CURSEG_WARM_NODE) {
3000 			reversed = true;
3001 			i = CURSEG_COLD_NODE;
3002 		} else {
3003 			i = CURSEG_HOT_NODE;
3004 		}
3005 		cnt = NR_CURSEG_NODE_TYPE;
3006 	} else {
3007 		if (seg_type >= CURSEG_WARM_DATA) {
3008 			reversed = true;
3009 			i = CURSEG_COLD_DATA;
3010 		} else {
3011 			i = CURSEG_HOT_DATA;
3012 		}
3013 		cnt = NR_CURSEG_DATA_TYPE;
3014 	}
3015 
3016 	for (; cnt-- > 0; reversed ? i-- : i++) {
3017 		if (i == seg_type)
3018 			continue;
3019 		if (!f2fs_get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
3020 			curseg->next_segno = segno;
3021 			return 1;
3022 		}
3023 	}
3024 
3025 	/* find valid_blocks=0 in dirty list */
3026 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
3027 		segno = get_free_segment(sbi);
3028 		if (segno != NULL_SEGNO) {
3029 			curseg->next_segno = segno;
3030 			return 1;
3031 		}
3032 	}
3033 	return 0;
3034 }
3035 
3036 static bool need_new_seg(struct f2fs_sb_info *sbi, int type)
3037 {
3038 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3039 
3040 	if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
3041 	    curseg->seg_type == CURSEG_WARM_NODE)
3042 		return true;
3043 	if (curseg->alloc_type == LFS &&
3044 	    is_next_segment_free(sbi, curseg, type) &&
3045 	    likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
3046 		return true;
3047 	if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0))
3048 		return true;
3049 	return false;
3050 }
3051 
3052 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
3053 					unsigned int start, unsigned int end)
3054 {
3055 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3056 	unsigned int segno;
3057 
3058 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3059 	mutex_lock(&curseg->curseg_mutex);
3060 	down_write(&SIT_I(sbi)->sentry_lock);
3061 
3062 	segno = CURSEG_I(sbi, type)->segno;
3063 	if (segno < start || segno > end)
3064 		goto unlock;
3065 
3066 	if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
3067 		change_curseg(sbi, type);
3068 	else
3069 		new_curseg(sbi, type, true);
3070 
3071 	stat_inc_seg_type(sbi, curseg);
3072 
3073 	locate_dirty_segment(sbi, segno);
3074 unlock:
3075 	up_write(&SIT_I(sbi)->sentry_lock);
3076 
3077 	if (segno != curseg->segno)
3078 		f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
3079 			    type, segno, curseg->segno);
3080 
3081 	mutex_unlock(&curseg->curseg_mutex);
3082 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3083 }
3084 
3085 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
3086 						bool new_sec, bool force)
3087 {
3088 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3089 	unsigned int old_segno;
3090 
3091 	if (!force && curseg->inited &&
3092 	    !curseg->next_blkoff &&
3093 	    !get_valid_blocks(sbi, curseg->segno, new_sec) &&
3094 	    !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
3095 		return;
3096 
3097 	old_segno = curseg->segno;
3098 	new_curseg(sbi, type, true);
3099 	stat_inc_seg_type(sbi, curseg);
3100 	locate_dirty_segment(sbi, old_segno);
3101 }
3102 
3103 void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
3104 {
3105 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3106 	down_write(&SIT_I(sbi)->sentry_lock);
3107 	__allocate_new_segment(sbi, type, true, force);
3108 	up_write(&SIT_I(sbi)->sentry_lock);
3109 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3110 }
3111 
3112 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
3113 {
3114 	int i;
3115 
3116 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3117 	down_write(&SIT_I(sbi)->sentry_lock);
3118 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
3119 		__allocate_new_segment(sbi, i, false, false);
3120 	up_write(&SIT_I(sbi)->sentry_lock);
3121 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3122 }
3123 
3124 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
3125 						struct cp_control *cpc)
3126 {
3127 	__u64 trim_start = cpc->trim_start;
3128 	bool has_candidate = false;
3129 
3130 	down_write(&SIT_I(sbi)->sentry_lock);
3131 	for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
3132 		if (add_discard_addrs(sbi, cpc, true)) {
3133 			has_candidate = true;
3134 			break;
3135 		}
3136 	}
3137 	up_write(&SIT_I(sbi)->sentry_lock);
3138 
3139 	cpc->trim_start = trim_start;
3140 	return has_candidate;
3141 }
3142 
3143 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
3144 					struct discard_policy *dpolicy,
3145 					unsigned int start, unsigned int end)
3146 {
3147 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
3148 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
3149 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
3150 	struct discard_cmd *dc;
3151 	struct blk_plug plug;
3152 	int issued;
3153 	unsigned int trimmed = 0;
3154 
3155 next:
3156 	issued = 0;
3157 
3158 	mutex_lock(&dcc->cmd_lock);
3159 	if (unlikely(dcc->rbtree_check))
3160 		f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
3161 
3162 	dc = __lookup_discard_cmd_ret(&dcc->root, start,
3163 				&prev_dc, &next_dc, &insert_p, &insert_parent);
3164 	if (!dc)
3165 		dc = next_dc;
3166 
3167 	blk_start_plug(&plug);
3168 
3169 	while (dc && dc->di.lstart <= end) {
3170 		struct rb_node *node;
3171 		int err = 0;
3172 
3173 		if (dc->di.len < dpolicy->granularity)
3174 			goto skip;
3175 
3176 		if (dc->state != D_PREP) {
3177 			list_move_tail(&dc->list, &dcc->fstrim_list);
3178 			goto skip;
3179 		}
3180 
3181 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3182 
3183 		if (issued >= dpolicy->max_requests) {
3184 			start = dc->di.lstart + dc->di.len;
3185 
3186 			if (err)
3187 				__remove_discard_cmd(sbi, dc);
3188 
3189 			blk_finish_plug(&plug);
3190 			mutex_unlock(&dcc->cmd_lock);
3191 			trimmed += __wait_all_discard_cmd(sbi, NULL);
3192 			f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
3193 			goto next;
3194 		}
3195 skip:
3196 		node = rb_next(&dc->rb_node);
3197 		if (err)
3198 			__remove_discard_cmd(sbi, dc);
3199 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3200 
3201 		if (fatal_signal_pending(current))
3202 			break;
3203 	}
3204 
3205 	blk_finish_plug(&plug);
3206 	mutex_unlock(&dcc->cmd_lock);
3207 
3208 	return trimmed;
3209 }
3210 
3211 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3212 {
3213 	__u64 start = F2FS_BYTES_TO_BLK(range->start);
3214 	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3215 	unsigned int start_segno, end_segno;
3216 	block_t start_block, end_block;
3217 	struct cp_control cpc;
3218 	struct discard_policy dpolicy;
3219 	unsigned long long trimmed = 0;
3220 	int err = 0;
3221 	bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3222 
3223 	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3224 		return -EINVAL;
3225 
3226 	if (end < MAIN_BLKADDR(sbi))
3227 		goto out;
3228 
3229 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3230 		f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3231 		return -EFSCORRUPTED;
3232 	}
3233 
3234 	/* start/end segment number in main_area */
3235 	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3236 	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3237 						GET_SEGNO(sbi, end);
3238 	if (need_align) {
3239 		start_segno = rounddown(start_segno, sbi->segs_per_sec);
3240 		end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
3241 	}
3242 
3243 	cpc.reason = CP_DISCARD;
3244 	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3245 	cpc.trim_start = start_segno;
3246 	cpc.trim_end = end_segno;
3247 
3248 	if (sbi->discard_blks == 0)
3249 		goto out;
3250 
3251 	f2fs_down_write(&sbi->gc_lock);
3252 	stat_inc_cp_call_count(sbi, TOTAL_CALL);
3253 	err = f2fs_write_checkpoint(sbi, &cpc);
3254 	f2fs_up_write(&sbi->gc_lock);
3255 	if (err)
3256 		goto out;
3257 
3258 	/*
3259 	 * We filed discard candidates, but actually we don't need to wait for
3260 	 * all of them, since they'll be issued in idle time along with runtime
3261 	 * discard option. User configuration looks like using runtime discard
3262 	 * or periodic fstrim instead of it.
3263 	 */
3264 	if (f2fs_realtime_discard_enable(sbi))
3265 		goto out;
3266 
3267 	start_block = START_BLOCK(sbi, start_segno);
3268 	end_block = START_BLOCK(sbi, end_segno + 1);
3269 
3270 	__init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3271 	trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3272 					start_block, end_block);
3273 
3274 	trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3275 					start_block, end_block);
3276 out:
3277 	if (!err)
3278 		range->len = F2FS_BLK_TO_BYTES(trimmed);
3279 	return err;
3280 }
3281 
3282 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3283 {
3284 	switch (hint) {
3285 	case WRITE_LIFE_SHORT:
3286 		return CURSEG_HOT_DATA;
3287 	case WRITE_LIFE_EXTREME:
3288 		return CURSEG_COLD_DATA;
3289 	default:
3290 		return CURSEG_WARM_DATA;
3291 	}
3292 }
3293 
3294 static int __get_segment_type_2(struct f2fs_io_info *fio)
3295 {
3296 	if (fio->type == DATA)
3297 		return CURSEG_HOT_DATA;
3298 	else
3299 		return CURSEG_HOT_NODE;
3300 }
3301 
3302 static int __get_segment_type_4(struct f2fs_io_info *fio)
3303 {
3304 	if (fio->type == DATA) {
3305 		struct inode *inode = fio->page->mapping->host;
3306 
3307 		if (S_ISDIR(inode->i_mode))
3308 			return CURSEG_HOT_DATA;
3309 		else
3310 			return CURSEG_COLD_DATA;
3311 	} else {
3312 		if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3313 			return CURSEG_WARM_NODE;
3314 		else
3315 			return CURSEG_COLD_NODE;
3316 	}
3317 }
3318 
3319 static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs)
3320 {
3321 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3322 	struct extent_info ei = {};
3323 
3324 	if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) {
3325 		if (!ei.age)
3326 			return NO_CHECK_TYPE;
3327 		if (ei.age <= sbi->hot_data_age_threshold)
3328 			return CURSEG_HOT_DATA;
3329 		if (ei.age <= sbi->warm_data_age_threshold)
3330 			return CURSEG_WARM_DATA;
3331 		return CURSEG_COLD_DATA;
3332 	}
3333 	return NO_CHECK_TYPE;
3334 }
3335 
3336 static int __get_segment_type_6(struct f2fs_io_info *fio)
3337 {
3338 	if (fio->type == DATA) {
3339 		struct inode *inode = fio->page->mapping->host;
3340 		int type;
3341 
3342 		if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
3343 			return CURSEG_COLD_DATA_PINNED;
3344 
3345 		if (page_private_gcing(fio->page)) {
3346 			if (fio->sbi->am.atgc_enabled &&
3347 				(fio->io_type == FS_DATA_IO) &&
3348 				(fio->sbi->gc_mode != GC_URGENT_HIGH))
3349 				return CURSEG_ALL_DATA_ATGC;
3350 			else
3351 				return CURSEG_COLD_DATA;
3352 		}
3353 		if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3354 			return CURSEG_COLD_DATA;
3355 
3356 		type = __get_age_segment_type(inode, fio->page->index);
3357 		if (type != NO_CHECK_TYPE)
3358 			return type;
3359 
3360 		if (file_is_hot(inode) ||
3361 				is_inode_flag_set(inode, FI_HOT_DATA) ||
3362 				f2fs_is_cow_file(inode))
3363 			return CURSEG_HOT_DATA;
3364 		return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3365 	} else {
3366 		if (IS_DNODE(fio->page))
3367 			return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3368 						CURSEG_HOT_NODE;
3369 		return CURSEG_COLD_NODE;
3370 	}
3371 }
3372 
3373 static int __get_segment_type(struct f2fs_io_info *fio)
3374 {
3375 	int type = 0;
3376 
3377 	switch (F2FS_OPTION(fio->sbi).active_logs) {
3378 	case 2:
3379 		type = __get_segment_type_2(fio);
3380 		break;
3381 	case 4:
3382 		type = __get_segment_type_4(fio);
3383 		break;
3384 	case 6:
3385 		type = __get_segment_type_6(fio);
3386 		break;
3387 	default:
3388 		f2fs_bug_on(fio->sbi, true);
3389 	}
3390 
3391 	if (IS_HOT(type))
3392 		fio->temp = HOT;
3393 	else if (IS_WARM(type))
3394 		fio->temp = WARM;
3395 	else
3396 		fio->temp = COLD;
3397 	return type;
3398 }
3399 
3400 static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi,
3401 		struct curseg_info *seg)
3402 {
3403 	/* To allocate block chunks in different sizes, use random number */
3404 	if (--seg->fragment_remained_chunk > 0)
3405 		return;
3406 
3407 	seg->fragment_remained_chunk =
3408 		get_random_u32_inclusive(1, sbi->max_fragment_chunk);
3409 	seg->next_blkoff +=
3410 		get_random_u32_inclusive(1, sbi->max_fragment_hole);
3411 }
3412 
3413 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3414 		block_t old_blkaddr, block_t *new_blkaddr,
3415 		struct f2fs_summary *sum, int type,
3416 		struct f2fs_io_info *fio)
3417 {
3418 	struct sit_info *sit_i = SIT_I(sbi);
3419 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3420 	unsigned long long old_mtime;
3421 	bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3422 	struct seg_entry *se = NULL;
3423 	bool segment_full = false;
3424 
3425 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3426 
3427 	mutex_lock(&curseg->curseg_mutex);
3428 	down_write(&sit_i->sentry_lock);
3429 
3430 	if (from_gc) {
3431 		f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3432 		se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3433 		sanity_check_seg_type(sbi, se->type);
3434 		f2fs_bug_on(sbi, IS_NODESEG(se->type));
3435 	}
3436 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3437 
3438 	f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg);
3439 
3440 	f2fs_wait_discard_bio(sbi, *new_blkaddr);
3441 
3442 	curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3443 	if (curseg->alloc_type == SSR) {
3444 		curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg);
3445 	} else {
3446 		curseg->next_blkoff++;
3447 		if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
3448 			f2fs_randomize_chunk(sbi, curseg);
3449 	}
3450 	if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno))
3451 		segment_full = true;
3452 	stat_inc_block_count(sbi, curseg);
3453 
3454 	if (from_gc) {
3455 		old_mtime = get_segment_mtime(sbi, old_blkaddr);
3456 	} else {
3457 		update_segment_mtime(sbi, old_blkaddr, 0);
3458 		old_mtime = 0;
3459 	}
3460 	update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3461 
3462 	/*
3463 	 * SIT information should be updated before segment allocation,
3464 	 * since SSR needs latest valid block information.
3465 	 */
3466 	update_sit_entry(sbi, *new_blkaddr, 1);
3467 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3468 		update_sit_entry(sbi, old_blkaddr, -1);
3469 
3470 	/*
3471 	 * If the current segment is full, flush it out and replace it with a
3472 	 * new segment.
3473 	 */
3474 	if (segment_full) {
3475 		if (from_gc) {
3476 			get_atssr_segment(sbi, type, se->type,
3477 						AT_SSR, se->mtime);
3478 		} else {
3479 			if (need_new_seg(sbi, type))
3480 				new_curseg(sbi, type, false);
3481 			else
3482 				change_curseg(sbi, type);
3483 			stat_inc_seg_type(sbi, curseg);
3484 		}
3485 	}
3486 	/*
3487 	 * segment dirty status should be updated after segment allocation,
3488 	 * so we just need to update status only one time after previous
3489 	 * segment being closed.
3490 	 */
3491 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3492 	locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3493 
3494 	if (IS_DATASEG(curseg->seg_type))
3495 		atomic64_inc(&sbi->allocated_data_blocks);
3496 
3497 	up_write(&sit_i->sentry_lock);
3498 
3499 	if (page && IS_NODESEG(curseg->seg_type)) {
3500 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3501 
3502 		f2fs_inode_chksum_set(sbi, page);
3503 	}
3504 
3505 	if (fio) {
3506 		struct f2fs_bio_info *io;
3507 
3508 		if (F2FS_IO_ALIGNED(sbi))
3509 			fio->retry = 0;
3510 
3511 		INIT_LIST_HEAD(&fio->list);
3512 		fio->in_list = 1;
3513 		io = sbi->write_io[fio->type] + fio->temp;
3514 		spin_lock(&io->io_lock);
3515 		list_add_tail(&fio->list, &io->io_list);
3516 		spin_unlock(&io->io_lock);
3517 	}
3518 
3519 	mutex_unlock(&curseg->curseg_mutex);
3520 
3521 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3522 }
3523 
3524 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino,
3525 					block_t blkaddr, unsigned int blkcnt)
3526 {
3527 	if (!f2fs_is_multi_device(sbi))
3528 		return;
3529 
3530 	while (1) {
3531 		unsigned int devidx = f2fs_target_device_index(sbi, blkaddr);
3532 		unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1;
3533 
3534 		/* update device state for fsync */
3535 		f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO);
3536 
3537 		/* update device state for checkpoint */
3538 		if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3539 			spin_lock(&sbi->dev_lock);
3540 			f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3541 			spin_unlock(&sbi->dev_lock);
3542 		}
3543 
3544 		if (blkcnt <= blks)
3545 			break;
3546 		blkcnt -= blks;
3547 		blkaddr += blks;
3548 	}
3549 }
3550 
3551 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3552 {
3553 	int type = __get_segment_type(fio);
3554 	bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3555 
3556 	if (keep_order)
3557 		f2fs_down_read(&fio->sbi->io_order_lock);
3558 reallocate:
3559 	f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3560 			&fio->new_blkaddr, sum, type, fio);
3561 	if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3562 		f2fs_invalidate_internal_cache(fio->sbi, fio->old_blkaddr);
3563 
3564 	/* writeout dirty page into bdev */
3565 	f2fs_submit_page_write(fio);
3566 	if (fio->retry) {
3567 		fio->old_blkaddr = fio->new_blkaddr;
3568 		goto reallocate;
3569 	}
3570 
3571 	f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1);
3572 
3573 	if (keep_order)
3574 		f2fs_up_read(&fio->sbi->io_order_lock);
3575 }
3576 
3577 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3578 					enum iostat_type io_type)
3579 {
3580 	struct f2fs_io_info fio = {
3581 		.sbi = sbi,
3582 		.type = META,
3583 		.temp = HOT,
3584 		.op = REQ_OP_WRITE,
3585 		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3586 		.old_blkaddr = page->index,
3587 		.new_blkaddr = page->index,
3588 		.page = page,
3589 		.encrypted_page = NULL,
3590 		.in_list = 0,
3591 	};
3592 
3593 	if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3594 		fio.op_flags &= ~REQ_META;
3595 
3596 	set_page_writeback(page);
3597 	f2fs_submit_page_write(&fio);
3598 
3599 	stat_inc_meta_count(sbi, page->index);
3600 	f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE);
3601 }
3602 
3603 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3604 {
3605 	struct f2fs_summary sum;
3606 
3607 	set_summary(&sum, nid, 0, 0);
3608 	do_write_page(&sum, fio);
3609 
3610 	f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE);
3611 }
3612 
3613 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3614 					struct f2fs_io_info *fio)
3615 {
3616 	struct f2fs_sb_info *sbi = fio->sbi;
3617 	struct f2fs_summary sum;
3618 
3619 	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3620 	if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO)
3621 		f2fs_update_age_extent_cache(dn);
3622 	set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3623 	do_write_page(&sum, fio);
3624 	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3625 
3626 	f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE);
3627 }
3628 
3629 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3630 {
3631 	int err;
3632 	struct f2fs_sb_info *sbi = fio->sbi;
3633 	unsigned int segno;
3634 
3635 	fio->new_blkaddr = fio->old_blkaddr;
3636 	/* i/o temperature is needed for passing down write hints */
3637 	__get_segment_type(fio);
3638 
3639 	segno = GET_SEGNO(sbi, fio->new_blkaddr);
3640 
3641 	if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3642 		set_sbi_flag(sbi, SBI_NEED_FSCK);
3643 		f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3644 			  __func__, segno);
3645 		err = -EFSCORRUPTED;
3646 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
3647 		goto drop_bio;
3648 	}
3649 
3650 	if (f2fs_cp_error(sbi)) {
3651 		err = -EIO;
3652 		goto drop_bio;
3653 	}
3654 
3655 	if (fio->post_read)
3656 		f2fs_truncate_meta_inode_pages(sbi, fio->new_blkaddr, 1);
3657 
3658 	stat_inc_inplace_blocks(fio->sbi);
3659 
3660 	if (fio->bio && !IS_F2FS_IPU_NOCACHE(sbi))
3661 		err = f2fs_merge_page_bio(fio);
3662 	else
3663 		err = f2fs_submit_page_bio(fio);
3664 	if (!err) {
3665 		f2fs_update_device_state(fio->sbi, fio->ino,
3666 						fio->new_blkaddr, 1);
3667 		f2fs_update_iostat(fio->sbi, fio->page->mapping->host,
3668 						fio->io_type, F2FS_BLKSIZE);
3669 	}
3670 
3671 	return err;
3672 drop_bio:
3673 	if (fio->bio && *(fio->bio)) {
3674 		struct bio *bio = *(fio->bio);
3675 
3676 		bio->bi_status = BLK_STS_IOERR;
3677 		bio_endio(bio);
3678 		*(fio->bio) = NULL;
3679 	}
3680 	return err;
3681 }
3682 
3683 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3684 						unsigned int segno)
3685 {
3686 	int i;
3687 
3688 	for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3689 		if (CURSEG_I(sbi, i)->segno == segno)
3690 			break;
3691 	}
3692 	return i;
3693 }
3694 
3695 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3696 				block_t old_blkaddr, block_t new_blkaddr,
3697 				bool recover_curseg, bool recover_newaddr,
3698 				bool from_gc)
3699 {
3700 	struct sit_info *sit_i = SIT_I(sbi);
3701 	struct curseg_info *curseg;
3702 	unsigned int segno, old_cursegno;
3703 	struct seg_entry *se;
3704 	int type;
3705 	unsigned short old_blkoff;
3706 	unsigned char old_alloc_type;
3707 
3708 	segno = GET_SEGNO(sbi, new_blkaddr);
3709 	se = get_seg_entry(sbi, segno);
3710 	type = se->type;
3711 
3712 	f2fs_down_write(&SM_I(sbi)->curseg_lock);
3713 
3714 	if (!recover_curseg) {
3715 		/* for recovery flow */
3716 		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3717 			if (old_blkaddr == NULL_ADDR)
3718 				type = CURSEG_COLD_DATA;
3719 			else
3720 				type = CURSEG_WARM_DATA;
3721 		}
3722 	} else {
3723 		if (IS_CURSEG(sbi, segno)) {
3724 			/* se->type is volatile as SSR allocation */
3725 			type = __f2fs_get_curseg(sbi, segno);
3726 			f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3727 		} else {
3728 			type = CURSEG_WARM_DATA;
3729 		}
3730 	}
3731 
3732 	f2fs_bug_on(sbi, !IS_DATASEG(type));
3733 	curseg = CURSEG_I(sbi, type);
3734 
3735 	mutex_lock(&curseg->curseg_mutex);
3736 	down_write(&sit_i->sentry_lock);
3737 
3738 	old_cursegno = curseg->segno;
3739 	old_blkoff = curseg->next_blkoff;
3740 	old_alloc_type = curseg->alloc_type;
3741 
3742 	/* change the current segment */
3743 	if (segno != curseg->segno) {
3744 		curseg->next_segno = segno;
3745 		change_curseg(sbi, type);
3746 	}
3747 
3748 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3749 	curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3750 
3751 	if (!recover_curseg || recover_newaddr) {
3752 		if (!from_gc)
3753 			update_segment_mtime(sbi, new_blkaddr, 0);
3754 		update_sit_entry(sbi, new_blkaddr, 1);
3755 	}
3756 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3757 		f2fs_invalidate_internal_cache(sbi, old_blkaddr);
3758 		if (!from_gc)
3759 			update_segment_mtime(sbi, old_blkaddr, 0);
3760 		update_sit_entry(sbi, old_blkaddr, -1);
3761 	}
3762 
3763 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3764 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3765 
3766 	locate_dirty_segment(sbi, old_cursegno);
3767 
3768 	if (recover_curseg) {
3769 		if (old_cursegno != curseg->segno) {
3770 			curseg->next_segno = old_cursegno;
3771 			change_curseg(sbi, type);
3772 		}
3773 		curseg->next_blkoff = old_blkoff;
3774 		curseg->alloc_type = old_alloc_type;
3775 	}
3776 
3777 	up_write(&sit_i->sentry_lock);
3778 	mutex_unlock(&curseg->curseg_mutex);
3779 	f2fs_up_write(&SM_I(sbi)->curseg_lock);
3780 }
3781 
3782 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3783 				block_t old_addr, block_t new_addr,
3784 				unsigned char version, bool recover_curseg,
3785 				bool recover_newaddr)
3786 {
3787 	struct f2fs_summary sum;
3788 
3789 	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3790 
3791 	f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3792 					recover_curseg, recover_newaddr, false);
3793 
3794 	f2fs_update_data_blkaddr(dn, new_addr);
3795 }
3796 
3797 void f2fs_wait_on_page_writeback(struct page *page,
3798 				enum page_type type, bool ordered, bool locked)
3799 {
3800 	if (PageWriteback(page)) {
3801 		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3802 
3803 		/* submit cached LFS IO */
3804 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3805 		/* submit cached IPU IO */
3806 		f2fs_submit_merged_ipu_write(sbi, NULL, page);
3807 		if (ordered) {
3808 			wait_on_page_writeback(page);
3809 			f2fs_bug_on(sbi, locked && PageWriteback(page));
3810 		} else {
3811 			wait_for_stable_page(page);
3812 		}
3813 	}
3814 }
3815 
3816 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3817 {
3818 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3819 	struct page *cpage;
3820 
3821 	if (!f2fs_post_read_required(inode))
3822 		return;
3823 
3824 	if (!__is_valid_data_blkaddr(blkaddr))
3825 		return;
3826 
3827 	cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3828 	if (cpage) {
3829 		f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3830 		f2fs_put_page(cpage, 1);
3831 	}
3832 }
3833 
3834 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3835 								block_t len)
3836 {
3837 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3838 	block_t i;
3839 
3840 	if (!f2fs_post_read_required(inode))
3841 		return;
3842 
3843 	for (i = 0; i < len; i++)
3844 		f2fs_wait_on_block_writeback(inode, blkaddr + i);
3845 
3846 	f2fs_truncate_meta_inode_pages(sbi, blkaddr, len);
3847 }
3848 
3849 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3850 {
3851 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3852 	struct curseg_info *seg_i;
3853 	unsigned char *kaddr;
3854 	struct page *page;
3855 	block_t start;
3856 	int i, j, offset;
3857 
3858 	start = start_sum_block(sbi);
3859 
3860 	page = f2fs_get_meta_page(sbi, start++);
3861 	if (IS_ERR(page))
3862 		return PTR_ERR(page);
3863 	kaddr = (unsigned char *)page_address(page);
3864 
3865 	/* Step 1: restore nat cache */
3866 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3867 	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3868 
3869 	/* Step 2: restore sit cache */
3870 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3871 	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3872 	offset = 2 * SUM_JOURNAL_SIZE;
3873 
3874 	/* Step 3: restore summary entries */
3875 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3876 		unsigned short blk_off;
3877 		unsigned int segno;
3878 
3879 		seg_i = CURSEG_I(sbi, i);
3880 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3881 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3882 		seg_i->next_segno = segno;
3883 		reset_curseg(sbi, i, 0);
3884 		seg_i->alloc_type = ckpt->alloc_type[i];
3885 		seg_i->next_blkoff = blk_off;
3886 
3887 		if (seg_i->alloc_type == SSR)
3888 			blk_off = sbi->blocks_per_seg;
3889 
3890 		for (j = 0; j < blk_off; j++) {
3891 			struct f2fs_summary *s;
3892 
3893 			s = (struct f2fs_summary *)(kaddr + offset);
3894 			seg_i->sum_blk->entries[j] = *s;
3895 			offset += SUMMARY_SIZE;
3896 			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3897 						SUM_FOOTER_SIZE)
3898 				continue;
3899 
3900 			f2fs_put_page(page, 1);
3901 			page = NULL;
3902 
3903 			page = f2fs_get_meta_page(sbi, start++);
3904 			if (IS_ERR(page))
3905 				return PTR_ERR(page);
3906 			kaddr = (unsigned char *)page_address(page);
3907 			offset = 0;
3908 		}
3909 	}
3910 	f2fs_put_page(page, 1);
3911 	return 0;
3912 }
3913 
3914 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3915 {
3916 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3917 	struct f2fs_summary_block *sum;
3918 	struct curseg_info *curseg;
3919 	struct page *new;
3920 	unsigned short blk_off;
3921 	unsigned int segno = 0;
3922 	block_t blk_addr = 0;
3923 	int err = 0;
3924 
3925 	/* get segment number and block addr */
3926 	if (IS_DATASEG(type)) {
3927 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3928 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3929 							CURSEG_HOT_DATA]);
3930 		if (__exist_node_summaries(sbi))
3931 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
3932 		else
3933 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3934 	} else {
3935 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
3936 							CURSEG_HOT_NODE]);
3937 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3938 							CURSEG_HOT_NODE]);
3939 		if (__exist_node_summaries(sbi))
3940 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3941 							type - CURSEG_HOT_NODE);
3942 		else
3943 			blk_addr = GET_SUM_BLOCK(sbi, segno);
3944 	}
3945 
3946 	new = f2fs_get_meta_page(sbi, blk_addr);
3947 	if (IS_ERR(new))
3948 		return PTR_ERR(new);
3949 	sum = (struct f2fs_summary_block *)page_address(new);
3950 
3951 	if (IS_NODESEG(type)) {
3952 		if (__exist_node_summaries(sbi)) {
3953 			struct f2fs_summary *ns = &sum->entries[0];
3954 			int i;
3955 
3956 			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3957 				ns->version = 0;
3958 				ns->ofs_in_node = 0;
3959 			}
3960 		} else {
3961 			err = f2fs_restore_node_summary(sbi, segno, sum);
3962 			if (err)
3963 				goto out;
3964 		}
3965 	}
3966 
3967 	/* set uncompleted segment to curseg */
3968 	curseg = CURSEG_I(sbi, type);
3969 	mutex_lock(&curseg->curseg_mutex);
3970 
3971 	/* update journal info */
3972 	down_write(&curseg->journal_rwsem);
3973 	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3974 	up_write(&curseg->journal_rwsem);
3975 
3976 	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3977 	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3978 	curseg->next_segno = segno;
3979 	reset_curseg(sbi, type, 0);
3980 	curseg->alloc_type = ckpt->alloc_type[type];
3981 	curseg->next_blkoff = blk_off;
3982 	mutex_unlock(&curseg->curseg_mutex);
3983 out:
3984 	f2fs_put_page(new, 1);
3985 	return err;
3986 }
3987 
3988 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3989 {
3990 	struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3991 	struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3992 	int type = CURSEG_HOT_DATA;
3993 	int err;
3994 
3995 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3996 		int npages = f2fs_npages_for_summary_flush(sbi, true);
3997 
3998 		if (npages >= 2)
3999 			f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
4000 							META_CP, true);
4001 
4002 		/* restore for compacted data summary */
4003 		err = read_compacted_summaries(sbi);
4004 		if (err)
4005 			return err;
4006 		type = CURSEG_HOT_NODE;
4007 	}
4008 
4009 	if (__exist_node_summaries(sbi))
4010 		f2fs_ra_meta_pages(sbi,
4011 				sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
4012 				NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
4013 
4014 	for (; type <= CURSEG_COLD_NODE; type++) {
4015 		err = read_normal_summaries(sbi, type);
4016 		if (err)
4017 			return err;
4018 	}
4019 
4020 	/* sanity check for summary blocks */
4021 	if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
4022 			sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
4023 		f2fs_err(sbi, "invalid journal entries nats %u sits %u",
4024 			 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
4025 		return -EINVAL;
4026 	}
4027 
4028 	return 0;
4029 }
4030 
4031 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
4032 {
4033 	struct page *page;
4034 	unsigned char *kaddr;
4035 	struct f2fs_summary *summary;
4036 	struct curseg_info *seg_i;
4037 	int written_size = 0;
4038 	int i, j;
4039 
4040 	page = f2fs_grab_meta_page(sbi, blkaddr++);
4041 	kaddr = (unsigned char *)page_address(page);
4042 	memset(kaddr, 0, PAGE_SIZE);
4043 
4044 	/* Step 1: write nat cache */
4045 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
4046 	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
4047 	written_size += SUM_JOURNAL_SIZE;
4048 
4049 	/* Step 2: write sit cache */
4050 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
4051 	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
4052 	written_size += SUM_JOURNAL_SIZE;
4053 
4054 	/* Step 3: write summary entries */
4055 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
4056 		seg_i = CURSEG_I(sbi, i);
4057 		for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); j++) {
4058 			if (!page) {
4059 				page = f2fs_grab_meta_page(sbi, blkaddr++);
4060 				kaddr = (unsigned char *)page_address(page);
4061 				memset(kaddr, 0, PAGE_SIZE);
4062 				written_size = 0;
4063 			}
4064 			summary = (struct f2fs_summary *)(kaddr + written_size);
4065 			*summary = seg_i->sum_blk->entries[j];
4066 			written_size += SUMMARY_SIZE;
4067 
4068 			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
4069 							SUM_FOOTER_SIZE)
4070 				continue;
4071 
4072 			set_page_dirty(page);
4073 			f2fs_put_page(page, 1);
4074 			page = NULL;
4075 		}
4076 	}
4077 	if (page) {
4078 		set_page_dirty(page);
4079 		f2fs_put_page(page, 1);
4080 	}
4081 }
4082 
4083 static void write_normal_summaries(struct f2fs_sb_info *sbi,
4084 					block_t blkaddr, int type)
4085 {
4086 	int i, end;
4087 
4088 	if (IS_DATASEG(type))
4089 		end = type + NR_CURSEG_DATA_TYPE;
4090 	else
4091 		end = type + NR_CURSEG_NODE_TYPE;
4092 
4093 	for (i = type; i < end; i++)
4094 		write_current_sum_page(sbi, i, blkaddr + (i - type));
4095 }
4096 
4097 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4098 {
4099 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
4100 		write_compacted_summaries(sbi, start_blk);
4101 	else
4102 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
4103 }
4104 
4105 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4106 {
4107 	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
4108 }
4109 
4110 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
4111 					unsigned int val, int alloc)
4112 {
4113 	int i;
4114 
4115 	if (type == NAT_JOURNAL) {
4116 		for (i = 0; i < nats_in_cursum(journal); i++) {
4117 			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
4118 				return i;
4119 		}
4120 		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
4121 			return update_nats_in_cursum(journal, 1);
4122 	} else if (type == SIT_JOURNAL) {
4123 		for (i = 0; i < sits_in_cursum(journal); i++)
4124 			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
4125 				return i;
4126 		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
4127 			return update_sits_in_cursum(journal, 1);
4128 	}
4129 	return -1;
4130 }
4131 
4132 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
4133 					unsigned int segno)
4134 {
4135 	return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
4136 }
4137 
4138 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
4139 					unsigned int start)
4140 {
4141 	struct sit_info *sit_i = SIT_I(sbi);
4142 	struct page *page;
4143 	pgoff_t src_off, dst_off;
4144 
4145 	src_off = current_sit_addr(sbi, start);
4146 	dst_off = next_sit_addr(sbi, src_off);
4147 
4148 	page = f2fs_grab_meta_page(sbi, dst_off);
4149 	seg_info_to_sit_page(sbi, page, start);
4150 
4151 	set_page_dirty(page);
4152 	set_to_next_sit(sit_i, start);
4153 
4154 	return page;
4155 }
4156 
4157 static struct sit_entry_set *grab_sit_entry_set(void)
4158 {
4159 	struct sit_entry_set *ses =
4160 			f2fs_kmem_cache_alloc(sit_entry_set_slab,
4161 						GFP_NOFS, true, NULL);
4162 
4163 	ses->entry_cnt = 0;
4164 	INIT_LIST_HEAD(&ses->set_list);
4165 	return ses;
4166 }
4167 
4168 static void release_sit_entry_set(struct sit_entry_set *ses)
4169 {
4170 	list_del(&ses->set_list);
4171 	kmem_cache_free(sit_entry_set_slab, ses);
4172 }
4173 
4174 static void adjust_sit_entry_set(struct sit_entry_set *ses,
4175 						struct list_head *head)
4176 {
4177 	struct sit_entry_set *next = ses;
4178 
4179 	if (list_is_last(&ses->set_list, head))
4180 		return;
4181 
4182 	list_for_each_entry_continue(next, head, set_list)
4183 		if (ses->entry_cnt <= next->entry_cnt) {
4184 			list_move_tail(&ses->set_list, &next->set_list);
4185 			return;
4186 		}
4187 
4188 	list_move_tail(&ses->set_list, head);
4189 }
4190 
4191 static void add_sit_entry(unsigned int segno, struct list_head *head)
4192 {
4193 	struct sit_entry_set *ses;
4194 	unsigned int start_segno = START_SEGNO(segno);
4195 
4196 	list_for_each_entry(ses, head, set_list) {
4197 		if (ses->start_segno == start_segno) {
4198 			ses->entry_cnt++;
4199 			adjust_sit_entry_set(ses, head);
4200 			return;
4201 		}
4202 	}
4203 
4204 	ses = grab_sit_entry_set();
4205 
4206 	ses->start_segno = start_segno;
4207 	ses->entry_cnt++;
4208 	list_add(&ses->set_list, head);
4209 }
4210 
4211 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4212 {
4213 	struct f2fs_sm_info *sm_info = SM_I(sbi);
4214 	struct list_head *set_list = &sm_info->sit_entry_set;
4215 	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4216 	unsigned int segno;
4217 
4218 	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4219 		add_sit_entry(segno, set_list);
4220 }
4221 
4222 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4223 {
4224 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4225 	struct f2fs_journal *journal = curseg->journal;
4226 	int i;
4227 
4228 	down_write(&curseg->journal_rwsem);
4229 	for (i = 0; i < sits_in_cursum(journal); i++) {
4230 		unsigned int segno;
4231 		bool dirtied;
4232 
4233 		segno = le32_to_cpu(segno_in_journal(journal, i));
4234 		dirtied = __mark_sit_entry_dirty(sbi, segno);
4235 
4236 		if (!dirtied)
4237 			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4238 	}
4239 	update_sits_in_cursum(journal, -i);
4240 	up_write(&curseg->journal_rwsem);
4241 }
4242 
4243 /*
4244  * CP calls this function, which flushes SIT entries including sit_journal,
4245  * and moves prefree segs to free segs.
4246  */
4247 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4248 {
4249 	struct sit_info *sit_i = SIT_I(sbi);
4250 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4251 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4252 	struct f2fs_journal *journal = curseg->journal;
4253 	struct sit_entry_set *ses, *tmp;
4254 	struct list_head *head = &SM_I(sbi)->sit_entry_set;
4255 	bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4256 	struct seg_entry *se;
4257 
4258 	down_write(&sit_i->sentry_lock);
4259 
4260 	if (!sit_i->dirty_sentries)
4261 		goto out;
4262 
4263 	/*
4264 	 * add and account sit entries of dirty bitmap in sit entry
4265 	 * set temporarily
4266 	 */
4267 	add_sits_in_set(sbi);
4268 
4269 	/*
4270 	 * if there are no enough space in journal to store dirty sit
4271 	 * entries, remove all entries from journal and add and account
4272 	 * them in sit entry set.
4273 	 */
4274 	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4275 								!to_journal)
4276 		remove_sits_in_journal(sbi);
4277 
4278 	/*
4279 	 * there are two steps to flush sit entries:
4280 	 * #1, flush sit entries to journal in current cold data summary block.
4281 	 * #2, flush sit entries to sit page.
4282 	 */
4283 	list_for_each_entry_safe(ses, tmp, head, set_list) {
4284 		struct page *page = NULL;
4285 		struct f2fs_sit_block *raw_sit = NULL;
4286 		unsigned int start_segno = ses->start_segno;
4287 		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4288 						(unsigned long)MAIN_SEGS(sbi));
4289 		unsigned int segno = start_segno;
4290 
4291 		if (to_journal &&
4292 			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4293 			to_journal = false;
4294 
4295 		if (to_journal) {
4296 			down_write(&curseg->journal_rwsem);
4297 		} else {
4298 			page = get_next_sit_page(sbi, start_segno);
4299 			raw_sit = page_address(page);
4300 		}
4301 
4302 		/* flush dirty sit entries in region of current sit set */
4303 		for_each_set_bit_from(segno, bitmap, end) {
4304 			int offset, sit_offset;
4305 
4306 			se = get_seg_entry(sbi, segno);
4307 #ifdef CONFIG_F2FS_CHECK_FS
4308 			if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4309 						SIT_VBLOCK_MAP_SIZE))
4310 				f2fs_bug_on(sbi, 1);
4311 #endif
4312 
4313 			/* add discard candidates */
4314 			if (!(cpc->reason & CP_DISCARD)) {
4315 				cpc->trim_start = segno;
4316 				add_discard_addrs(sbi, cpc, false);
4317 			}
4318 
4319 			if (to_journal) {
4320 				offset = f2fs_lookup_journal_in_cursum(journal,
4321 							SIT_JOURNAL, segno, 1);
4322 				f2fs_bug_on(sbi, offset < 0);
4323 				segno_in_journal(journal, offset) =
4324 							cpu_to_le32(segno);
4325 				seg_info_to_raw_sit(se,
4326 					&sit_in_journal(journal, offset));
4327 				check_block_count(sbi, segno,
4328 					&sit_in_journal(journal, offset));
4329 			} else {
4330 				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4331 				seg_info_to_raw_sit(se,
4332 						&raw_sit->entries[sit_offset]);
4333 				check_block_count(sbi, segno,
4334 						&raw_sit->entries[sit_offset]);
4335 			}
4336 
4337 			__clear_bit(segno, bitmap);
4338 			sit_i->dirty_sentries--;
4339 			ses->entry_cnt--;
4340 		}
4341 
4342 		if (to_journal)
4343 			up_write(&curseg->journal_rwsem);
4344 		else
4345 			f2fs_put_page(page, 1);
4346 
4347 		f2fs_bug_on(sbi, ses->entry_cnt);
4348 		release_sit_entry_set(ses);
4349 	}
4350 
4351 	f2fs_bug_on(sbi, !list_empty(head));
4352 	f2fs_bug_on(sbi, sit_i->dirty_sentries);
4353 out:
4354 	if (cpc->reason & CP_DISCARD) {
4355 		__u64 trim_start = cpc->trim_start;
4356 
4357 		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4358 			add_discard_addrs(sbi, cpc, false);
4359 
4360 		cpc->trim_start = trim_start;
4361 	}
4362 	up_write(&sit_i->sentry_lock);
4363 
4364 	set_prefree_as_free_segments(sbi);
4365 }
4366 
4367 static int build_sit_info(struct f2fs_sb_info *sbi)
4368 {
4369 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4370 	struct sit_info *sit_i;
4371 	unsigned int sit_segs, start;
4372 	char *src_bitmap, *bitmap;
4373 	unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4374 	unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0;
4375 
4376 	/* allocate memory for SIT information */
4377 	sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4378 	if (!sit_i)
4379 		return -ENOMEM;
4380 
4381 	SM_I(sbi)->sit_info = sit_i;
4382 
4383 	sit_i->sentries =
4384 		f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4385 					      MAIN_SEGS(sbi)),
4386 			      GFP_KERNEL);
4387 	if (!sit_i->sentries)
4388 		return -ENOMEM;
4389 
4390 	main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4391 	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4392 								GFP_KERNEL);
4393 	if (!sit_i->dirty_sentries_bitmap)
4394 		return -ENOMEM;
4395 
4396 #ifdef CONFIG_F2FS_CHECK_FS
4397 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map);
4398 #else
4399 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map);
4400 #endif
4401 	sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4402 	if (!sit_i->bitmap)
4403 		return -ENOMEM;
4404 
4405 	bitmap = sit_i->bitmap;
4406 
4407 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4408 		sit_i->sentries[start].cur_valid_map = bitmap;
4409 		bitmap += SIT_VBLOCK_MAP_SIZE;
4410 
4411 		sit_i->sentries[start].ckpt_valid_map = bitmap;
4412 		bitmap += SIT_VBLOCK_MAP_SIZE;
4413 
4414 #ifdef CONFIG_F2FS_CHECK_FS
4415 		sit_i->sentries[start].cur_valid_map_mir = bitmap;
4416 		bitmap += SIT_VBLOCK_MAP_SIZE;
4417 #endif
4418 
4419 		if (discard_map) {
4420 			sit_i->sentries[start].discard_map = bitmap;
4421 			bitmap += SIT_VBLOCK_MAP_SIZE;
4422 		}
4423 	}
4424 
4425 	sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4426 	if (!sit_i->tmp_map)
4427 		return -ENOMEM;
4428 
4429 	if (__is_large_section(sbi)) {
4430 		sit_i->sec_entries =
4431 			f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4432 						      MAIN_SECS(sbi)),
4433 				      GFP_KERNEL);
4434 		if (!sit_i->sec_entries)
4435 			return -ENOMEM;
4436 	}
4437 
4438 	/* get information related with SIT */
4439 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4440 
4441 	/* setup SIT bitmap from ckeckpoint pack */
4442 	sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4443 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4444 
4445 	sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4446 	if (!sit_i->sit_bitmap)
4447 		return -ENOMEM;
4448 
4449 #ifdef CONFIG_F2FS_CHECK_FS
4450 	sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4451 					sit_bitmap_size, GFP_KERNEL);
4452 	if (!sit_i->sit_bitmap_mir)
4453 		return -ENOMEM;
4454 
4455 	sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4456 					main_bitmap_size, GFP_KERNEL);
4457 	if (!sit_i->invalid_segmap)
4458 		return -ENOMEM;
4459 #endif
4460 
4461 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4462 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4463 	sit_i->written_valid_blocks = 0;
4464 	sit_i->bitmap_size = sit_bitmap_size;
4465 	sit_i->dirty_sentries = 0;
4466 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4467 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4468 	sit_i->mounted_time = ktime_get_boottime_seconds();
4469 	init_rwsem(&sit_i->sentry_lock);
4470 	return 0;
4471 }
4472 
4473 static int build_free_segmap(struct f2fs_sb_info *sbi)
4474 {
4475 	struct free_segmap_info *free_i;
4476 	unsigned int bitmap_size, sec_bitmap_size;
4477 
4478 	/* allocate memory for free segmap information */
4479 	free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4480 	if (!free_i)
4481 		return -ENOMEM;
4482 
4483 	SM_I(sbi)->free_info = free_i;
4484 
4485 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4486 	free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4487 	if (!free_i->free_segmap)
4488 		return -ENOMEM;
4489 
4490 	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4491 	free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4492 	if (!free_i->free_secmap)
4493 		return -ENOMEM;
4494 
4495 	/* set all segments as dirty temporarily */
4496 	memset(free_i->free_segmap, 0xff, bitmap_size);
4497 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4498 
4499 	/* init free segmap information */
4500 	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4501 	free_i->free_segments = 0;
4502 	free_i->free_sections = 0;
4503 	spin_lock_init(&free_i->segmap_lock);
4504 	return 0;
4505 }
4506 
4507 static int build_curseg(struct f2fs_sb_info *sbi)
4508 {
4509 	struct curseg_info *array;
4510 	int i;
4511 
4512 	array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4513 					sizeof(*array)), GFP_KERNEL);
4514 	if (!array)
4515 		return -ENOMEM;
4516 
4517 	SM_I(sbi)->curseg_array = array;
4518 
4519 	for (i = 0; i < NO_CHECK_TYPE; i++) {
4520 		mutex_init(&array[i].curseg_mutex);
4521 		array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4522 		if (!array[i].sum_blk)
4523 			return -ENOMEM;
4524 		init_rwsem(&array[i].journal_rwsem);
4525 		array[i].journal = f2fs_kzalloc(sbi,
4526 				sizeof(struct f2fs_journal), GFP_KERNEL);
4527 		if (!array[i].journal)
4528 			return -ENOMEM;
4529 		if (i < NR_PERSISTENT_LOG)
4530 			array[i].seg_type = CURSEG_HOT_DATA + i;
4531 		else if (i == CURSEG_COLD_DATA_PINNED)
4532 			array[i].seg_type = CURSEG_COLD_DATA;
4533 		else if (i == CURSEG_ALL_DATA_ATGC)
4534 			array[i].seg_type = CURSEG_COLD_DATA;
4535 		array[i].segno = NULL_SEGNO;
4536 		array[i].next_blkoff = 0;
4537 		array[i].inited = false;
4538 	}
4539 	return restore_curseg_summaries(sbi);
4540 }
4541 
4542 static int build_sit_entries(struct f2fs_sb_info *sbi)
4543 {
4544 	struct sit_info *sit_i = SIT_I(sbi);
4545 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4546 	struct f2fs_journal *journal = curseg->journal;
4547 	struct seg_entry *se;
4548 	struct f2fs_sit_entry sit;
4549 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
4550 	unsigned int i, start, end;
4551 	unsigned int readed, start_blk = 0;
4552 	int err = 0;
4553 	block_t sit_valid_blocks[2] = {0, 0};
4554 
4555 	do {
4556 		readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4557 							META_SIT, true);
4558 
4559 		start = start_blk * sit_i->sents_per_block;
4560 		end = (start_blk + readed) * sit_i->sents_per_block;
4561 
4562 		for (; start < end && start < MAIN_SEGS(sbi); start++) {
4563 			struct f2fs_sit_block *sit_blk;
4564 			struct page *page;
4565 
4566 			se = &sit_i->sentries[start];
4567 			page = get_current_sit_page(sbi, start);
4568 			if (IS_ERR(page))
4569 				return PTR_ERR(page);
4570 			sit_blk = (struct f2fs_sit_block *)page_address(page);
4571 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4572 			f2fs_put_page(page, 1);
4573 
4574 			err = check_block_count(sbi, start, &sit);
4575 			if (err)
4576 				return err;
4577 			seg_info_from_raw_sit(se, &sit);
4578 
4579 			if (se->type >= NR_PERSISTENT_LOG) {
4580 				f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4581 							se->type, start);
4582 				f2fs_handle_error(sbi,
4583 						ERROR_INCONSISTENT_SUM_TYPE);
4584 				return -EFSCORRUPTED;
4585 			}
4586 
4587 			sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4588 
4589 			if (f2fs_block_unit_discard(sbi)) {
4590 				/* build discard map only one time */
4591 				if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4592 					memset(se->discard_map, 0xff,
4593 						SIT_VBLOCK_MAP_SIZE);
4594 				} else {
4595 					memcpy(se->discard_map,
4596 						se->cur_valid_map,
4597 						SIT_VBLOCK_MAP_SIZE);
4598 					sbi->discard_blks +=
4599 						sbi->blocks_per_seg -
4600 						se->valid_blocks;
4601 				}
4602 			}
4603 
4604 			if (__is_large_section(sbi))
4605 				get_sec_entry(sbi, start)->valid_blocks +=
4606 							se->valid_blocks;
4607 		}
4608 		start_blk += readed;
4609 	} while (start_blk < sit_blk_cnt);
4610 
4611 	down_read(&curseg->journal_rwsem);
4612 	for (i = 0; i < sits_in_cursum(journal); i++) {
4613 		unsigned int old_valid_blocks;
4614 
4615 		start = le32_to_cpu(segno_in_journal(journal, i));
4616 		if (start >= MAIN_SEGS(sbi)) {
4617 			f2fs_err(sbi, "Wrong journal entry on segno %u",
4618 				 start);
4619 			err = -EFSCORRUPTED;
4620 			f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL);
4621 			break;
4622 		}
4623 
4624 		se = &sit_i->sentries[start];
4625 		sit = sit_in_journal(journal, i);
4626 
4627 		old_valid_blocks = se->valid_blocks;
4628 
4629 		sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks;
4630 
4631 		err = check_block_count(sbi, start, &sit);
4632 		if (err)
4633 			break;
4634 		seg_info_from_raw_sit(se, &sit);
4635 
4636 		if (se->type >= NR_PERSISTENT_LOG) {
4637 			f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4638 							se->type, start);
4639 			err = -EFSCORRUPTED;
4640 			f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
4641 			break;
4642 		}
4643 
4644 		sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4645 
4646 		if (f2fs_block_unit_discard(sbi)) {
4647 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4648 				memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4649 			} else {
4650 				memcpy(se->discard_map, se->cur_valid_map,
4651 							SIT_VBLOCK_MAP_SIZE);
4652 				sbi->discard_blks += old_valid_blocks;
4653 				sbi->discard_blks -= se->valid_blocks;
4654 			}
4655 		}
4656 
4657 		if (__is_large_section(sbi)) {
4658 			get_sec_entry(sbi, start)->valid_blocks +=
4659 							se->valid_blocks;
4660 			get_sec_entry(sbi, start)->valid_blocks -=
4661 							old_valid_blocks;
4662 		}
4663 	}
4664 	up_read(&curseg->journal_rwsem);
4665 
4666 	if (err)
4667 		return err;
4668 
4669 	if (sit_valid_blocks[NODE] != valid_node_count(sbi)) {
4670 		f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4671 			 sit_valid_blocks[NODE], valid_node_count(sbi));
4672 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT);
4673 		return -EFSCORRUPTED;
4674 	}
4675 
4676 	if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] >
4677 				valid_user_blocks(sbi)) {
4678 		f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u",
4679 			 sit_valid_blocks[DATA], sit_valid_blocks[NODE],
4680 			 valid_user_blocks(sbi));
4681 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT);
4682 		return -EFSCORRUPTED;
4683 	}
4684 
4685 	return 0;
4686 }
4687 
4688 static void init_free_segmap(struct f2fs_sb_info *sbi)
4689 {
4690 	unsigned int start;
4691 	int type;
4692 	struct seg_entry *sentry;
4693 
4694 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4695 		if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4696 			continue;
4697 		sentry = get_seg_entry(sbi, start);
4698 		if (!sentry->valid_blocks)
4699 			__set_free(sbi, start);
4700 		else
4701 			SIT_I(sbi)->written_valid_blocks +=
4702 						sentry->valid_blocks;
4703 	}
4704 
4705 	/* set use the current segments */
4706 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4707 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4708 
4709 		__set_test_and_inuse(sbi, curseg_t->segno);
4710 	}
4711 }
4712 
4713 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4714 {
4715 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4716 	struct free_segmap_info *free_i = FREE_I(sbi);
4717 	unsigned int segno = 0, offset = 0, secno;
4718 	block_t valid_blocks, usable_blks_in_seg;
4719 
4720 	while (1) {
4721 		/* find dirty segment based on free segmap */
4722 		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4723 		if (segno >= MAIN_SEGS(sbi))
4724 			break;
4725 		offset = segno + 1;
4726 		valid_blocks = get_valid_blocks(sbi, segno, false);
4727 		usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4728 		if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4729 			continue;
4730 		if (valid_blocks > usable_blks_in_seg) {
4731 			f2fs_bug_on(sbi, 1);
4732 			continue;
4733 		}
4734 		mutex_lock(&dirty_i->seglist_lock);
4735 		__locate_dirty_segment(sbi, segno, DIRTY);
4736 		mutex_unlock(&dirty_i->seglist_lock);
4737 	}
4738 
4739 	if (!__is_large_section(sbi))
4740 		return;
4741 
4742 	mutex_lock(&dirty_i->seglist_lock);
4743 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4744 		valid_blocks = get_valid_blocks(sbi, segno, true);
4745 		secno = GET_SEC_FROM_SEG(sbi, segno);
4746 
4747 		if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi))
4748 			continue;
4749 		if (IS_CURSEC(sbi, secno))
4750 			continue;
4751 		set_bit(secno, dirty_i->dirty_secmap);
4752 	}
4753 	mutex_unlock(&dirty_i->seglist_lock);
4754 }
4755 
4756 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4757 {
4758 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4759 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4760 
4761 	dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4762 	if (!dirty_i->victim_secmap)
4763 		return -ENOMEM;
4764 
4765 	dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4766 	if (!dirty_i->pinned_secmap)
4767 		return -ENOMEM;
4768 
4769 	dirty_i->pinned_secmap_cnt = 0;
4770 	dirty_i->enable_pin_section = true;
4771 	return 0;
4772 }
4773 
4774 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4775 {
4776 	struct dirty_seglist_info *dirty_i;
4777 	unsigned int bitmap_size, i;
4778 
4779 	/* allocate memory for dirty segments list information */
4780 	dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4781 								GFP_KERNEL);
4782 	if (!dirty_i)
4783 		return -ENOMEM;
4784 
4785 	SM_I(sbi)->dirty_info = dirty_i;
4786 	mutex_init(&dirty_i->seglist_lock);
4787 
4788 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4789 
4790 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
4791 		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4792 								GFP_KERNEL);
4793 		if (!dirty_i->dirty_segmap[i])
4794 			return -ENOMEM;
4795 	}
4796 
4797 	if (__is_large_section(sbi)) {
4798 		bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4799 		dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4800 						bitmap_size, GFP_KERNEL);
4801 		if (!dirty_i->dirty_secmap)
4802 			return -ENOMEM;
4803 	}
4804 
4805 	init_dirty_segmap(sbi);
4806 	return init_victim_secmap(sbi);
4807 }
4808 
4809 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4810 {
4811 	int i;
4812 
4813 	/*
4814 	 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4815 	 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4816 	 */
4817 	for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4818 		struct curseg_info *curseg = CURSEG_I(sbi, i);
4819 		struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4820 		unsigned int blkofs = curseg->next_blkoff;
4821 
4822 		if (f2fs_sb_has_readonly(sbi) &&
4823 			i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE)
4824 			continue;
4825 
4826 		sanity_check_seg_type(sbi, curseg->seg_type);
4827 
4828 		if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) {
4829 			f2fs_err(sbi,
4830 				 "Current segment has invalid alloc_type:%d",
4831 				 curseg->alloc_type);
4832 			f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4833 			return -EFSCORRUPTED;
4834 		}
4835 
4836 		if (f2fs_test_bit(blkofs, se->cur_valid_map))
4837 			goto out;
4838 
4839 		if (curseg->alloc_type == SSR)
4840 			continue;
4841 
4842 		for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4843 			if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4844 				continue;
4845 out:
4846 			f2fs_err(sbi,
4847 				 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4848 				 i, curseg->segno, curseg->alloc_type,
4849 				 curseg->next_blkoff, blkofs);
4850 			f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
4851 			return -EFSCORRUPTED;
4852 		}
4853 	}
4854 	return 0;
4855 }
4856 
4857 #ifdef CONFIG_BLK_DEV_ZONED
4858 
4859 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4860 				    struct f2fs_dev_info *fdev,
4861 				    struct blk_zone *zone)
4862 {
4863 	unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4864 	block_t zone_block, wp_block, last_valid_block;
4865 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4866 	int i, s, b, ret;
4867 	struct seg_entry *se;
4868 
4869 	if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4870 		return 0;
4871 
4872 	wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4873 	wp_segno = GET_SEGNO(sbi, wp_block);
4874 	wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4875 	zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4876 	zone_segno = GET_SEGNO(sbi, zone_block);
4877 	zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4878 
4879 	if (zone_segno >= MAIN_SEGS(sbi))
4880 		return 0;
4881 
4882 	/*
4883 	 * Skip check of zones cursegs point to, since
4884 	 * fix_curseg_write_pointer() checks them.
4885 	 */
4886 	for (i = 0; i < NO_CHECK_TYPE; i++)
4887 		if (zone_secno == GET_SEC_FROM_SEG(sbi,
4888 						   CURSEG_I(sbi, i)->segno))
4889 			return 0;
4890 
4891 	/*
4892 	 * Get last valid block of the zone.
4893 	 */
4894 	last_valid_block = zone_block - 1;
4895 	for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4896 		segno = zone_segno + s;
4897 		se = get_seg_entry(sbi, segno);
4898 		for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4899 			if (f2fs_test_bit(b, se->cur_valid_map)) {
4900 				last_valid_block = START_BLOCK(sbi, segno) + b;
4901 				break;
4902 			}
4903 		if (last_valid_block >= zone_block)
4904 			break;
4905 	}
4906 
4907 	/*
4908 	 * The write pointer matches with the valid blocks or
4909 	 * already points to the end of the zone.
4910 	 */
4911 	if ((last_valid_block + 1 == wp_block) ||
4912 			(zone->wp == zone->start + zone->len))
4913 		return 0;
4914 
4915 	if (last_valid_block + 1 == zone_block) {
4916 		/*
4917 		 * If there is no valid block in the zone and if write pointer
4918 		 * is not at zone start, reset the write pointer.
4919 		 */
4920 		f2fs_notice(sbi,
4921 			    "Zone without valid block has non-zero write "
4922 			    "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4923 			    wp_segno, wp_blkoff);
4924 		ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4925 					zone->len >> log_sectors_per_block);
4926 		if (ret)
4927 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4928 				 fdev->path, ret);
4929 
4930 		return ret;
4931 	}
4932 
4933 	/*
4934 	 * If there are valid blocks and the write pointer doesn't
4935 	 * match with them, we need to report the inconsistency and
4936 	 * fill the zone till the end to close the zone. This inconsistency
4937 	 * does not cause write error because the zone will not be selected
4938 	 * for write operation until it get discarded.
4939 	 */
4940 	f2fs_notice(sbi, "Valid blocks are not aligned with write pointer: "
4941 		    "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4942 		    GET_SEGNO(sbi, last_valid_block),
4943 		    GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4944 		    wp_segno, wp_blkoff);
4945 
4946 	ret = blkdev_zone_mgmt(fdev->bdev, REQ_OP_ZONE_FINISH,
4947 				zone->start, zone->len, GFP_NOFS);
4948 	if (ret == -EOPNOTSUPP) {
4949 		ret = blkdev_issue_zeroout(fdev->bdev, zone->wp,
4950 					zone->len - (zone->wp - zone->start),
4951 					GFP_NOFS, 0);
4952 		if (ret)
4953 			f2fs_err(sbi, "Fill up zone failed: %s (errno=%d)",
4954 					fdev->path, ret);
4955 	} else if (ret) {
4956 		f2fs_err(sbi, "Finishing zone failed: %s (errno=%d)",
4957 				fdev->path, ret);
4958 	}
4959 
4960 	return ret;
4961 }
4962 
4963 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4964 						  block_t zone_blkaddr)
4965 {
4966 	int i;
4967 
4968 	for (i = 0; i < sbi->s_ndevs; i++) {
4969 		if (!bdev_is_zoned(FDEV(i).bdev))
4970 			continue;
4971 		if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4972 				zone_blkaddr <= FDEV(i).end_blk))
4973 			return &FDEV(i);
4974 	}
4975 
4976 	return NULL;
4977 }
4978 
4979 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4980 			      void *data)
4981 {
4982 	memcpy(data, zone, sizeof(struct blk_zone));
4983 	return 0;
4984 }
4985 
4986 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4987 {
4988 	struct curseg_info *cs = CURSEG_I(sbi, type);
4989 	struct f2fs_dev_info *zbd;
4990 	struct blk_zone zone;
4991 	unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4992 	block_t cs_zone_block, wp_block;
4993 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4994 	sector_t zone_sector;
4995 	int err;
4996 
4997 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4998 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4999 
5000 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
5001 	if (!zbd)
5002 		return 0;
5003 
5004 	/* report zone for the sector the curseg points to */
5005 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5006 		<< log_sectors_per_block;
5007 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5008 				  report_one_zone_cb, &zone);
5009 	if (err != 1) {
5010 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5011 			 zbd->path, err);
5012 		return err;
5013 	}
5014 
5015 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5016 		return 0;
5017 
5018 	wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
5019 	wp_segno = GET_SEGNO(sbi, wp_block);
5020 	wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
5021 	wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
5022 
5023 	if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
5024 		wp_sector_off == 0)
5025 		return 0;
5026 
5027 	f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
5028 		    "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
5029 		    type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
5030 
5031 	f2fs_notice(sbi, "Assign new section to curseg[%d]: "
5032 		    "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
5033 
5034 	f2fs_allocate_new_section(sbi, type, true);
5035 
5036 	/* check consistency of the zone curseg pointed to */
5037 	if (check_zone_write_pointer(sbi, zbd, &zone))
5038 		return -EIO;
5039 
5040 	/* check newly assigned zone */
5041 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
5042 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
5043 
5044 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
5045 	if (!zbd)
5046 		return 0;
5047 
5048 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5049 		<< log_sectors_per_block;
5050 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5051 				  report_one_zone_cb, &zone);
5052 	if (err != 1) {
5053 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5054 			 zbd->path, err);
5055 		return err;
5056 	}
5057 
5058 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5059 		return 0;
5060 
5061 	if (zone.wp != zone.start) {
5062 		f2fs_notice(sbi,
5063 			    "New zone for curseg[%d] is not yet discarded. "
5064 			    "Reset the zone: curseg[0x%x,0x%x]",
5065 			    type, cs->segno, cs->next_blkoff);
5066 		err = __f2fs_issue_discard_zone(sbi, zbd->bdev,	cs_zone_block,
5067 					zone.len >> log_sectors_per_block);
5068 		if (err) {
5069 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
5070 				 zbd->path, err);
5071 			return err;
5072 		}
5073 	}
5074 
5075 	return 0;
5076 }
5077 
5078 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5079 {
5080 	int i, ret;
5081 
5082 	for (i = 0; i < NR_PERSISTENT_LOG; i++) {
5083 		ret = fix_curseg_write_pointer(sbi, i);
5084 		if (ret)
5085 			return ret;
5086 	}
5087 
5088 	return 0;
5089 }
5090 
5091 struct check_zone_write_pointer_args {
5092 	struct f2fs_sb_info *sbi;
5093 	struct f2fs_dev_info *fdev;
5094 };
5095 
5096 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
5097 				      void *data)
5098 {
5099 	struct check_zone_write_pointer_args *args;
5100 
5101 	args = (struct check_zone_write_pointer_args *)data;
5102 
5103 	return check_zone_write_pointer(args->sbi, args->fdev, zone);
5104 }
5105 
5106 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5107 {
5108 	int i, ret;
5109 	struct check_zone_write_pointer_args args;
5110 
5111 	for (i = 0; i < sbi->s_ndevs; i++) {
5112 		if (!bdev_is_zoned(FDEV(i).bdev))
5113 			continue;
5114 
5115 		args.sbi = sbi;
5116 		args.fdev = &FDEV(i);
5117 		ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
5118 					  check_zone_write_pointer_cb, &args);
5119 		if (ret < 0)
5120 			return ret;
5121 	}
5122 
5123 	return 0;
5124 }
5125 
5126 /*
5127  * Return the number of usable blocks in a segment. The number of blocks
5128  * returned is always equal to the number of blocks in a segment for
5129  * segments fully contained within a sequential zone capacity or a
5130  * conventional zone. For segments partially contained in a sequential
5131  * zone capacity, the number of usable blocks up to the zone capacity
5132  * is returned. 0 is returned in all other cases.
5133  */
5134 static inline unsigned int f2fs_usable_zone_blks_in_seg(
5135 			struct f2fs_sb_info *sbi, unsigned int segno)
5136 {
5137 	block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
5138 	unsigned int secno;
5139 
5140 	if (!sbi->unusable_blocks_per_sec)
5141 		return sbi->blocks_per_seg;
5142 
5143 	secno = GET_SEC_FROM_SEG(sbi, segno);
5144 	seg_start = START_BLOCK(sbi, segno);
5145 	sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5146 	sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi);
5147 
5148 	/*
5149 	 * If segment starts before zone capacity and spans beyond
5150 	 * zone capacity, then usable blocks are from seg start to
5151 	 * zone capacity. If the segment starts after the zone capacity,
5152 	 * then there are no usable blocks.
5153 	 */
5154 	if (seg_start >= sec_cap_blkaddr)
5155 		return 0;
5156 	if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr)
5157 		return sec_cap_blkaddr - seg_start;
5158 
5159 	return sbi->blocks_per_seg;
5160 }
5161 #else
5162 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5163 {
5164 	return 0;
5165 }
5166 
5167 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
5168 {
5169 	return 0;
5170 }
5171 
5172 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5173 							unsigned int segno)
5174 {
5175 	return 0;
5176 }
5177 
5178 #endif
5179 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5180 					unsigned int segno)
5181 {
5182 	if (f2fs_sb_has_blkzoned(sbi))
5183 		return f2fs_usable_zone_blks_in_seg(sbi, segno);
5184 
5185 	return sbi->blocks_per_seg;
5186 }
5187 
5188 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5189 					unsigned int segno)
5190 {
5191 	if (f2fs_sb_has_blkzoned(sbi))
5192 		return CAP_SEGS_PER_SEC(sbi);
5193 
5194 	return sbi->segs_per_sec;
5195 }
5196 
5197 /*
5198  * Update min, max modified time for cost-benefit GC algorithm
5199  */
5200 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5201 {
5202 	struct sit_info *sit_i = SIT_I(sbi);
5203 	unsigned int segno;
5204 
5205 	down_write(&sit_i->sentry_lock);
5206 
5207 	sit_i->min_mtime = ULLONG_MAX;
5208 
5209 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
5210 		unsigned int i;
5211 		unsigned long long mtime = 0;
5212 
5213 		for (i = 0; i < sbi->segs_per_sec; i++)
5214 			mtime += get_seg_entry(sbi, segno + i)->mtime;
5215 
5216 		mtime = div_u64(mtime, sbi->segs_per_sec);
5217 
5218 		if (sit_i->min_mtime > mtime)
5219 			sit_i->min_mtime = mtime;
5220 	}
5221 	sit_i->max_mtime = get_mtime(sbi, false);
5222 	sit_i->dirty_max_mtime = 0;
5223 	up_write(&sit_i->sentry_lock);
5224 }
5225 
5226 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5227 {
5228 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5229 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5230 	struct f2fs_sm_info *sm_info;
5231 	int err;
5232 
5233 	sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5234 	if (!sm_info)
5235 		return -ENOMEM;
5236 
5237 	/* init sm info */
5238 	sbi->sm_info = sm_info;
5239 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5240 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5241 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5242 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5243 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5244 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5245 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5246 	sm_info->rec_prefree_segments = sm_info->main_segments *
5247 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5248 	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5249 		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5250 
5251 	if (!f2fs_lfs_mode(sbi))
5252 		sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC);
5253 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5254 	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5255 	sm_info->min_seq_blocks = sbi->blocks_per_seg;
5256 	sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5257 	sm_info->min_ssr_sections = reserved_sections(sbi);
5258 
5259 	INIT_LIST_HEAD(&sm_info->sit_entry_set);
5260 
5261 	init_f2fs_rwsem(&sm_info->curseg_lock);
5262 
5263 	err = f2fs_create_flush_cmd_control(sbi);
5264 	if (err)
5265 		return err;
5266 
5267 	err = create_discard_cmd_control(sbi);
5268 	if (err)
5269 		return err;
5270 
5271 	err = build_sit_info(sbi);
5272 	if (err)
5273 		return err;
5274 	err = build_free_segmap(sbi);
5275 	if (err)
5276 		return err;
5277 	err = build_curseg(sbi);
5278 	if (err)
5279 		return err;
5280 
5281 	/* reinit free segmap based on SIT */
5282 	err = build_sit_entries(sbi);
5283 	if (err)
5284 		return err;
5285 
5286 	init_free_segmap(sbi);
5287 	err = build_dirty_segmap(sbi);
5288 	if (err)
5289 		return err;
5290 
5291 	err = sanity_check_curseg(sbi);
5292 	if (err)
5293 		return err;
5294 
5295 	init_min_max_mtime(sbi);
5296 	return 0;
5297 }
5298 
5299 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5300 		enum dirty_type dirty_type)
5301 {
5302 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5303 
5304 	mutex_lock(&dirty_i->seglist_lock);
5305 	kvfree(dirty_i->dirty_segmap[dirty_type]);
5306 	dirty_i->nr_dirty[dirty_type] = 0;
5307 	mutex_unlock(&dirty_i->seglist_lock);
5308 }
5309 
5310 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5311 {
5312 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5313 
5314 	kvfree(dirty_i->pinned_secmap);
5315 	kvfree(dirty_i->victim_secmap);
5316 }
5317 
5318 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5319 {
5320 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5321 	int i;
5322 
5323 	if (!dirty_i)
5324 		return;
5325 
5326 	/* discard pre-free/dirty segments list */
5327 	for (i = 0; i < NR_DIRTY_TYPE; i++)
5328 		discard_dirty_segmap(sbi, i);
5329 
5330 	if (__is_large_section(sbi)) {
5331 		mutex_lock(&dirty_i->seglist_lock);
5332 		kvfree(dirty_i->dirty_secmap);
5333 		mutex_unlock(&dirty_i->seglist_lock);
5334 	}
5335 
5336 	destroy_victim_secmap(sbi);
5337 	SM_I(sbi)->dirty_info = NULL;
5338 	kfree(dirty_i);
5339 }
5340 
5341 static void destroy_curseg(struct f2fs_sb_info *sbi)
5342 {
5343 	struct curseg_info *array = SM_I(sbi)->curseg_array;
5344 	int i;
5345 
5346 	if (!array)
5347 		return;
5348 	SM_I(sbi)->curseg_array = NULL;
5349 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
5350 		kfree(array[i].sum_blk);
5351 		kfree(array[i].journal);
5352 	}
5353 	kfree(array);
5354 }
5355 
5356 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5357 {
5358 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5359 
5360 	if (!free_i)
5361 		return;
5362 	SM_I(sbi)->free_info = NULL;
5363 	kvfree(free_i->free_segmap);
5364 	kvfree(free_i->free_secmap);
5365 	kfree(free_i);
5366 }
5367 
5368 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5369 {
5370 	struct sit_info *sit_i = SIT_I(sbi);
5371 
5372 	if (!sit_i)
5373 		return;
5374 
5375 	if (sit_i->sentries)
5376 		kvfree(sit_i->bitmap);
5377 	kfree(sit_i->tmp_map);
5378 
5379 	kvfree(sit_i->sentries);
5380 	kvfree(sit_i->sec_entries);
5381 	kvfree(sit_i->dirty_sentries_bitmap);
5382 
5383 	SM_I(sbi)->sit_info = NULL;
5384 	kvfree(sit_i->sit_bitmap);
5385 #ifdef CONFIG_F2FS_CHECK_FS
5386 	kvfree(sit_i->sit_bitmap_mir);
5387 	kvfree(sit_i->invalid_segmap);
5388 #endif
5389 	kfree(sit_i);
5390 }
5391 
5392 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5393 {
5394 	struct f2fs_sm_info *sm_info = SM_I(sbi);
5395 
5396 	if (!sm_info)
5397 		return;
5398 	f2fs_destroy_flush_cmd_control(sbi, true);
5399 	destroy_discard_cmd_control(sbi);
5400 	destroy_dirty_segmap(sbi);
5401 	destroy_curseg(sbi);
5402 	destroy_free_segmap(sbi);
5403 	destroy_sit_info(sbi);
5404 	sbi->sm_info = NULL;
5405 	kfree(sm_info);
5406 }
5407 
5408 int __init f2fs_create_segment_manager_caches(void)
5409 {
5410 	discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5411 			sizeof(struct discard_entry));
5412 	if (!discard_entry_slab)
5413 		goto fail;
5414 
5415 	discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5416 			sizeof(struct discard_cmd));
5417 	if (!discard_cmd_slab)
5418 		goto destroy_discard_entry;
5419 
5420 	sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5421 			sizeof(struct sit_entry_set));
5422 	if (!sit_entry_set_slab)
5423 		goto destroy_discard_cmd;
5424 
5425 	revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry",
5426 			sizeof(struct revoke_entry));
5427 	if (!revoke_entry_slab)
5428 		goto destroy_sit_entry_set;
5429 	return 0;
5430 
5431 destroy_sit_entry_set:
5432 	kmem_cache_destroy(sit_entry_set_slab);
5433 destroy_discard_cmd:
5434 	kmem_cache_destroy(discard_cmd_slab);
5435 destroy_discard_entry:
5436 	kmem_cache_destroy(discard_entry_slab);
5437 fail:
5438 	return -ENOMEM;
5439 }
5440 
5441 void f2fs_destroy_segment_manager_caches(void)
5442 {
5443 	kmem_cache_destroy(sit_entry_set_slab);
5444 	kmem_cache_destroy(discard_cmd_slab);
5445 	kmem_cache_destroy(discard_entry_slab);
5446 	kmem_cache_destroy(revoke_entry_slab);
5447 }
5448