xref: /openbmc/linux/fs/f2fs/segment.c (revision 7ae5c03a)
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_sb_info *sbi = F2FS_I_SB(inode);
191 	struct f2fs_inode_info *fi = F2FS_I(inode);
192 
193 	if (!f2fs_is_atomic_file(inode))
194 		return;
195 
196 	if (clean)
197 		truncate_inode_pages_final(inode->i_mapping);
198 	clear_inode_flag(fi->cow_inode, FI_COW_FILE);
199 	iput(fi->cow_inode);
200 	fi->cow_inode = NULL;
201 	release_atomic_write_cnt(inode);
202 	clear_inode_flag(inode, FI_ATOMIC_FILE);
203 
204 	spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
205 	sbi->atomic_files--;
206 	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
207 }
208 
209 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index,
210 			block_t new_addr, block_t *old_addr, bool recover)
211 {
212 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
213 	struct dnode_of_data dn;
214 	struct node_info ni;
215 	int err;
216 
217 retry:
218 	set_new_dnode(&dn, inode, NULL, NULL, 0);
219 	err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE_RA);
220 	if (err) {
221 		if (err == -ENOMEM) {
222 			f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
223 			goto retry;
224 		}
225 		return err;
226 	}
227 
228 	err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
229 	if (err) {
230 		f2fs_put_dnode(&dn);
231 		return err;
232 	}
233 
234 	if (recover) {
235 		/* dn.data_blkaddr is always valid */
236 		if (!__is_valid_data_blkaddr(new_addr)) {
237 			if (new_addr == NULL_ADDR)
238 				dec_valid_block_count(sbi, inode, 1);
239 			f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
240 			f2fs_update_data_blkaddr(&dn, new_addr);
241 		} else {
242 			f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
243 				new_addr, ni.version, true, true);
244 		}
245 	} else {
246 		blkcnt_t count = 1;
247 
248 		*old_addr = dn.data_blkaddr;
249 		f2fs_truncate_data_blocks_range(&dn, 1);
250 		dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count);
251 		inc_valid_block_count(sbi, inode, &count);
252 		f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr,
253 					ni.version, true, false);
254 	}
255 
256 	f2fs_put_dnode(&dn);
257 	return 0;
258 }
259 
260 static void __complete_revoke_list(struct inode *inode, struct list_head *head,
261 					bool revoke)
262 {
263 	struct revoke_entry *cur, *tmp;
264 
265 	list_for_each_entry_safe(cur, tmp, head, list) {
266 		if (revoke)
267 			__replace_atomic_write_block(inode, cur->index,
268 						cur->old_addr, NULL, true);
269 		list_del(&cur->list);
270 		kmem_cache_free(revoke_entry_slab, cur);
271 	}
272 }
273 
274 static int __f2fs_commit_atomic_write(struct inode *inode)
275 {
276 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
277 	struct f2fs_inode_info *fi = F2FS_I(inode);
278 	struct inode *cow_inode = fi->cow_inode;
279 	struct revoke_entry *new;
280 	struct list_head revoke_list;
281 	block_t blkaddr;
282 	struct dnode_of_data dn;
283 	pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
284 	pgoff_t off = 0, blen, index;
285 	int ret = 0, i;
286 
287 	INIT_LIST_HEAD(&revoke_list);
288 
289 	while (len) {
290 		blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len);
291 
292 		set_new_dnode(&dn, cow_inode, NULL, NULL, 0);
293 		ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA);
294 		if (ret && ret != -ENOENT) {
295 			goto out;
296 		} else if (ret == -ENOENT) {
297 			ret = 0;
298 			if (dn.max_level == 0)
299 				goto out;
300 			goto next;
301 		}
302 
303 		blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_page, cow_inode),
304 				len);
305 		index = off;
306 		for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) {
307 			blkaddr = f2fs_data_blkaddr(&dn);
308 
309 			if (!__is_valid_data_blkaddr(blkaddr)) {
310 				continue;
311 			} else if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
312 					DATA_GENERIC_ENHANCE)) {
313 				f2fs_put_dnode(&dn);
314 				ret = -EFSCORRUPTED;
315 				goto out;
316 			}
317 
318 			new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS,
319 							true, NULL);
320 
321 			ret = __replace_atomic_write_block(inode, index, blkaddr,
322 							&new->old_addr, false);
323 			if (ret) {
324 				f2fs_put_dnode(&dn);
325 				kmem_cache_free(revoke_entry_slab, new);
326 				goto out;
327 			}
328 
329 			f2fs_update_data_blkaddr(&dn, NULL_ADDR);
330 			new->index = index;
331 			list_add_tail(&new->list, &revoke_list);
332 		}
333 		f2fs_put_dnode(&dn);
334 next:
335 		off += blen;
336 		len -= blen;
337 	}
338 
339 out:
340 	if (ret)
341 		sbi->revoked_atomic_block += fi->atomic_write_cnt;
342 	else
343 		sbi->committed_atomic_block += fi->atomic_write_cnt;
344 
345 	__complete_revoke_list(inode, &revoke_list, ret ? true : false);
346 
347 	return ret;
348 }
349 
350 int f2fs_commit_atomic_write(struct inode *inode)
351 {
352 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
353 	struct f2fs_inode_info *fi = F2FS_I(inode);
354 	int err;
355 
356 	err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
357 	if (err)
358 		return err;
359 
360 	f2fs_down_write(&fi->i_gc_rwsem[WRITE]);
361 	f2fs_lock_op(sbi);
362 
363 	err = __f2fs_commit_atomic_write(inode);
364 
365 	f2fs_unlock_op(sbi);
366 	f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
367 
368 	return err;
369 }
370 
371 /*
372  * This function balances dirty node and dentry pages.
373  * In addition, it controls garbage collection.
374  */
375 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
376 {
377 	if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
378 		f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
379 		f2fs_stop_checkpoint(sbi, false);
380 	}
381 
382 	/* balance_fs_bg is able to be pending */
383 	if (need && excess_cached_nats(sbi))
384 		f2fs_balance_fs_bg(sbi, false);
385 
386 	if (!f2fs_is_checkpoint_ready(sbi))
387 		return;
388 
389 	/*
390 	 * We should do GC or end up with checkpoint, if there are so many dirty
391 	 * dir/node pages without enough free segments.
392 	 */
393 	if (has_not_enough_free_secs(sbi, 0, 0)) {
394 		if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
395 					sbi->gc_thread->f2fs_gc_task) {
396 			DEFINE_WAIT(wait);
397 
398 			prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
399 						TASK_UNINTERRUPTIBLE);
400 			wake_up(&sbi->gc_thread->gc_wait_queue_head);
401 			io_schedule();
402 			finish_wait(&sbi->gc_thread->fggc_wq, &wait);
403 		} else {
404 			struct f2fs_gc_control gc_control = {
405 				.victim_segno = NULL_SEGNO,
406 				.init_gc_type = BG_GC,
407 				.no_bg_gc = true,
408 				.should_migrate_blocks = false,
409 				.err_gc_skipped = false,
410 				.nr_free_secs = 1 };
411 			f2fs_down_write(&sbi->gc_lock);
412 			f2fs_gc(sbi, &gc_control);
413 		}
414 	}
415 }
416 
417 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi)
418 {
419 	int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2;
420 	unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS);
421 	unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA);
422 	unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES);
423 	unsigned int meta = get_pages(sbi, F2FS_DIRTY_META);
424 	unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA);
425 	unsigned int threshold = sbi->blocks_per_seg * factor *
426 					DEFAULT_DIRTY_THRESHOLD;
427 	unsigned int global_threshold = threshold * 3 / 2;
428 
429 	if (dents >= threshold || qdata >= threshold ||
430 		nodes >= threshold || meta >= threshold ||
431 		imeta >= threshold)
432 		return true;
433 	return dents + qdata + nodes + meta + imeta >  global_threshold;
434 }
435 
436 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
437 {
438 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
439 		return;
440 
441 	/* try to shrink extent cache when there is no enough memory */
442 	if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
443 		f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
444 
445 	/* check the # of cached NAT entries */
446 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
447 		f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
448 
449 	if (!f2fs_available_free_memory(sbi, FREE_NIDS))
450 		f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
451 	else
452 		f2fs_build_free_nids(sbi, false, false);
453 
454 	if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) ||
455 		excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi))
456 		goto do_sync;
457 
458 	/* there is background inflight IO or foreground operation recently */
459 	if (is_inflight_io(sbi, REQ_TIME) ||
460 		(!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem)))
461 		return;
462 
463 	/* exceed periodical checkpoint timeout threshold */
464 	if (f2fs_time_over(sbi, CP_TIME))
465 		goto do_sync;
466 
467 	/* checkpoint is the only way to shrink partial cached entries */
468 	if (f2fs_available_free_memory(sbi, NAT_ENTRIES) &&
469 		f2fs_available_free_memory(sbi, INO_ENTRIES))
470 		return;
471 
472 do_sync:
473 	if (test_opt(sbi, DATA_FLUSH) && from_bg) {
474 		struct blk_plug plug;
475 
476 		mutex_lock(&sbi->flush_lock);
477 
478 		blk_start_plug(&plug);
479 		f2fs_sync_dirty_inodes(sbi, FILE_INODE);
480 		blk_finish_plug(&plug);
481 
482 		mutex_unlock(&sbi->flush_lock);
483 	}
484 	f2fs_sync_fs(sbi->sb, true);
485 	stat_inc_bg_cp_count(sbi->stat_info);
486 }
487 
488 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
489 				struct block_device *bdev)
490 {
491 	int ret = blkdev_issue_flush(bdev);
492 
493 	trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
494 				test_opt(sbi, FLUSH_MERGE), ret);
495 	return ret;
496 }
497 
498 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
499 {
500 	int ret = 0;
501 	int i;
502 
503 	if (!f2fs_is_multi_device(sbi))
504 		return __submit_flush_wait(sbi, sbi->sb->s_bdev);
505 
506 	for (i = 0; i < sbi->s_ndevs; i++) {
507 		if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
508 			continue;
509 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
510 		if (ret)
511 			break;
512 	}
513 	return ret;
514 }
515 
516 static int issue_flush_thread(void *data)
517 {
518 	struct f2fs_sb_info *sbi = data;
519 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
520 	wait_queue_head_t *q = &fcc->flush_wait_queue;
521 repeat:
522 	if (kthread_should_stop())
523 		return 0;
524 
525 	if (!llist_empty(&fcc->issue_list)) {
526 		struct flush_cmd *cmd, *next;
527 		int ret;
528 
529 		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
530 		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
531 
532 		cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
533 
534 		ret = submit_flush_wait(sbi, cmd->ino);
535 		atomic_inc(&fcc->issued_flush);
536 
537 		llist_for_each_entry_safe(cmd, next,
538 					  fcc->dispatch_list, llnode) {
539 			cmd->ret = ret;
540 			complete(&cmd->wait);
541 		}
542 		fcc->dispatch_list = NULL;
543 	}
544 
545 	wait_event_interruptible(*q,
546 		kthread_should_stop() || !llist_empty(&fcc->issue_list));
547 	goto repeat;
548 }
549 
550 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
551 {
552 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
553 	struct flush_cmd cmd;
554 	int ret;
555 
556 	if (test_opt(sbi, NOBARRIER))
557 		return 0;
558 
559 	if (!test_opt(sbi, FLUSH_MERGE)) {
560 		atomic_inc(&fcc->queued_flush);
561 		ret = submit_flush_wait(sbi, ino);
562 		atomic_dec(&fcc->queued_flush);
563 		atomic_inc(&fcc->issued_flush);
564 		return ret;
565 	}
566 
567 	if (atomic_inc_return(&fcc->queued_flush) == 1 ||
568 	    f2fs_is_multi_device(sbi)) {
569 		ret = submit_flush_wait(sbi, ino);
570 		atomic_dec(&fcc->queued_flush);
571 
572 		atomic_inc(&fcc->issued_flush);
573 		return ret;
574 	}
575 
576 	cmd.ino = ino;
577 	init_completion(&cmd.wait);
578 
579 	llist_add(&cmd.llnode, &fcc->issue_list);
580 
581 	/*
582 	 * update issue_list before we wake up issue_flush thread, this
583 	 * smp_mb() pairs with another barrier in ___wait_event(), see
584 	 * more details in comments of waitqueue_active().
585 	 */
586 	smp_mb();
587 
588 	if (waitqueue_active(&fcc->flush_wait_queue))
589 		wake_up(&fcc->flush_wait_queue);
590 
591 	if (fcc->f2fs_issue_flush) {
592 		wait_for_completion(&cmd.wait);
593 		atomic_dec(&fcc->queued_flush);
594 	} else {
595 		struct llist_node *list;
596 
597 		list = llist_del_all(&fcc->issue_list);
598 		if (!list) {
599 			wait_for_completion(&cmd.wait);
600 			atomic_dec(&fcc->queued_flush);
601 		} else {
602 			struct flush_cmd *tmp, *next;
603 
604 			ret = submit_flush_wait(sbi, ino);
605 
606 			llist_for_each_entry_safe(tmp, next, list, llnode) {
607 				if (tmp == &cmd) {
608 					cmd.ret = ret;
609 					atomic_dec(&fcc->queued_flush);
610 					continue;
611 				}
612 				tmp->ret = ret;
613 				complete(&tmp->wait);
614 			}
615 		}
616 	}
617 
618 	return cmd.ret;
619 }
620 
621 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
622 {
623 	dev_t dev = sbi->sb->s_bdev->bd_dev;
624 	struct flush_cmd_control *fcc;
625 	int err = 0;
626 
627 	if (SM_I(sbi)->fcc_info) {
628 		fcc = SM_I(sbi)->fcc_info;
629 		if (fcc->f2fs_issue_flush)
630 			return err;
631 		goto init_thread;
632 	}
633 
634 	fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
635 	if (!fcc)
636 		return -ENOMEM;
637 	atomic_set(&fcc->issued_flush, 0);
638 	atomic_set(&fcc->queued_flush, 0);
639 	init_waitqueue_head(&fcc->flush_wait_queue);
640 	init_llist_head(&fcc->issue_list);
641 	SM_I(sbi)->fcc_info = fcc;
642 	if (!test_opt(sbi, FLUSH_MERGE))
643 		return err;
644 
645 init_thread:
646 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
647 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
648 	if (IS_ERR(fcc->f2fs_issue_flush)) {
649 		err = PTR_ERR(fcc->f2fs_issue_flush);
650 		kfree(fcc);
651 		SM_I(sbi)->fcc_info = NULL;
652 		return err;
653 	}
654 
655 	return err;
656 }
657 
658 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
659 {
660 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
661 
662 	if (fcc && fcc->f2fs_issue_flush) {
663 		struct task_struct *flush_thread = fcc->f2fs_issue_flush;
664 
665 		fcc->f2fs_issue_flush = NULL;
666 		kthread_stop(flush_thread);
667 	}
668 	if (free) {
669 		kfree(fcc);
670 		SM_I(sbi)->fcc_info = NULL;
671 	}
672 }
673 
674 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
675 {
676 	int ret = 0, i;
677 
678 	if (!f2fs_is_multi_device(sbi))
679 		return 0;
680 
681 	if (test_opt(sbi, NOBARRIER))
682 		return 0;
683 
684 	for (i = 1; i < sbi->s_ndevs; i++) {
685 		int count = DEFAULT_RETRY_IO_COUNT;
686 
687 		if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
688 			continue;
689 
690 		do {
691 			ret = __submit_flush_wait(sbi, FDEV(i).bdev);
692 			if (ret)
693 				f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
694 		} while (ret && --count);
695 
696 		if (ret) {
697 			f2fs_stop_checkpoint(sbi, false);
698 			break;
699 		}
700 
701 		spin_lock(&sbi->dev_lock);
702 		f2fs_clear_bit(i, (char *)&sbi->dirty_device);
703 		spin_unlock(&sbi->dev_lock);
704 	}
705 
706 	return ret;
707 }
708 
709 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
710 		enum dirty_type dirty_type)
711 {
712 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
713 
714 	/* need not be added */
715 	if (IS_CURSEG(sbi, segno))
716 		return;
717 
718 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
719 		dirty_i->nr_dirty[dirty_type]++;
720 
721 	if (dirty_type == DIRTY) {
722 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
723 		enum dirty_type t = sentry->type;
724 
725 		if (unlikely(t >= DIRTY)) {
726 			f2fs_bug_on(sbi, 1);
727 			return;
728 		}
729 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
730 			dirty_i->nr_dirty[t]++;
731 
732 		if (__is_large_section(sbi)) {
733 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
734 			block_t valid_blocks =
735 				get_valid_blocks(sbi, segno, true);
736 
737 			f2fs_bug_on(sbi, unlikely(!valid_blocks ||
738 					valid_blocks == CAP_BLKS_PER_SEC(sbi)));
739 
740 			if (!IS_CURSEC(sbi, secno))
741 				set_bit(secno, dirty_i->dirty_secmap);
742 		}
743 	}
744 }
745 
746 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
747 		enum dirty_type dirty_type)
748 {
749 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
750 	block_t valid_blocks;
751 
752 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
753 		dirty_i->nr_dirty[dirty_type]--;
754 
755 	if (dirty_type == DIRTY) {
756 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
757 		enum dirty_type t = sentry->type;
758 
759 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
760 			dirty_i->nr_dirty[t]--;
761 
762 		valid_blocks = get_valid_blocks(sbi, segno, true);
763 		if (valid_blocks == 0) {
764 			clear_bit(GET_SEC_FROM_SEG(sbi, segno),
765 						dirty_i->victim_secmap);
766 #ifdef CONFIG_F2FS_CHECK_FS
767 			clear_bit(segno, SIT_I(sbi)->invalid_segmap);
768 #endif
769 		}
770 		if (__is_large_section(sbi)) {
771 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
772 
773 			if (!valid_blocks ||
774 					valid_blocks == CAP_BLKS_PER_SEC(sbi)) {
775 				clear_bit(secno, dirty_i->dirty_secmap);
776 				return;
777 			}
778 
779 			if (!IS_CURSEC(sbi, secno))
780 				set_bit(secno, dirty_i->dirty_secmap);
781 		}
782 	}
783 }
784 
785 /*
786  * Should not occur error such as -ENOMEM.
787  * Adding dirty entry into seglist is not critical operation.
788  * If a given segment is one of current working segments, it won't be added.
789  */
790 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
791 {
792 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
793 	unsigned short valid_blocks, ckpt_valid_blocks;
794 	unsigned int usable_blocks;
795 
796 	if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
797 		return;
798 
799 	usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
800 	mutex_lock(&dirty_i->seglist_lock);
801 
802 	valid_blocks = get_valid_blocks(sbi, segno, false);
803 	ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
804 
805 	if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
806 		ckpt_valid_blocks == usable_blocks)) {
807 		__locate_dirty_segment(sbi, segno, PRE);
808 		__remove_dirty_segment(sbi, segno, DIRTY);
809 	} else if (valid_blocks < usable_blocks) {
810 		__locate_dirty_segment(sbi, segno, DIRTY);
811 	} else {
812 		/* Recovery routine with SSR needs this */
813 		__remove_dirty_segment(sbi, segno, DIRTY);
814 	}
815 
816 	mutex_unlock(&dirty_i->seglist_lock);
817 }
818 
819 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
820 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
821 {
822 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
823 	unsigned int segno;
824 
825 	mutex_lock(&dirty_i->seglist_lock);
826 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
827 		if (get_valid_blocks(sbi, segno, false))
828 			continue;
829 		if (IS_CURSEG(sbi, segno))
830 			continue;
831 		__locate_dirty_segment(sbi, segno, PRE);
832 		__remove_dirty_segment(sbi, segno, DIRTY);
833 	}
834 	mutex_unlock(&dirty_i->seglist_lock);
835 }
836 
837 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
838 {
839 	int ovp_hole_segs =
840 		(overprovision_segments(sbi) - reserved_segments(sbi));
841 	block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
842 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
843 	block_t holes[2] = {0, 0};	/* DATA and NODE */
844 	block_t unusable;
845 	struct seg_entry *se;
846 	unsigned int segno;
847 
848 	mutex_lock(&dirty_i->seglist_lock);
849 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
850 		se = get_seg_entry(sbi, segno);
851 		if (IS_NODESEG(se->type))
852 			holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
853 							se->valid_blocks;
854 		else
855 			holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
856 							se->valid_blocks;
857 	}
858 	mutex_unlock(&dirty_i->seglist_lock);
859 
860 	unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
861 	if (unusable > ovp_holes)
862 		return unusable - ovp_holes;
863 	return 0;
864 }
865 
866 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
867 {
868 	int ovp_hole_segs =
869 		(overprovision_segments(sbi) - reserved_segments(sbi));
870 	if (unusable > F2FS_OPTION(sbi).unusable_cap)
871 		return -EAGAIN;
872 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
873 		dirty_segments(sbi) > ovp_hole_segs)
874 		return -EAGAIN;
875 	return 0;
876 }
877 
878 /* This is only used by SBI_CP_DISABLED */
879 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
880 {
881 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
882 	unsigned int segno = 0;
883 
884 	mutex_lock(&dirty_i->seglist_lock);
885 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
886 		if (get_valid_blocks(sbi, segno, false))
887 			continue;
888 		if (get_ckpt_valid_blocks(sbi, segno, false))
889 			continue;
890 		mutex_unlock(&dirty_i->seglist_lock);
891 		return segno;
892 	}
893 	mutex_unlock(&dirty_i->seglist_lock);
894 	return NULL_SEGNO;
895 }
896 
897 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
898 		struct block_device *bdev, block_t lstart,
899 		block_t start, block_t len)
900 {
901 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
902 	struct list_head *pend_list;
903 	struct discard_cmd *dc;
904 
905 	f2fs_bug_on(sbi, !len);
906 
907 	pend_list = &dcc->pend_list[plist_idx(len)];
908 
909 	dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL);
910 	INIT_LIST_HEAD(&dc->list);
911 	dc->bdev = bdev;
912 	dc->lstart = lstart;
913 	dc->start = start;
914 	dc->len = len;
915 	dc->ref = 0;
916 	dc->state = D_PREP;
917 	dc->queued = 0;
918 	dc->error = 0;
919 	init_completion(&dc->wait);
920 	list_add_tail(&dc->list, pend_list);
921 	spin_lock_init(&dc->lock);
922 	dc->bio_ref = 0;
923 	atomic_inc(&dcc->discard_cmd_cnt);
924 	dcc->undiscard_blks += len;
925 
926 	return dc;
927 }
928 
929 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
930 				struct block_device *bdev, block_t lstart,
931 				block_t start, block_t len,
932 				struct rb_node *parent, struct rb_node **p,
933 				bool leftmost)
934 {
935 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
936 	struct discard_cmd *dc;
937 
938 	dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
939 
940 	rb_link_node(&dc->rb_node, parent, p);
941 	rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
942 
943 	return dc;
944 }
945 
946 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
947 							struct discard_cmd *dc)
948 {
949 	if (dc->state == D_DONE)
950 		atomic_sub(dc->queued, &dcc->queued_discard);
951 
952 	list_del(&dc->list);
953 	rb_erase_cached(&dc->rb_node, &dcc->root);
954 	dcc->undiscard_blks -= dc->len;
955 
956 	kmem_cache_free(discard_cmd_slab, dc);
957 
958 	atomic_dec(&dcc->discard_cmd_cnt);
959 }
960 
961 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
962 							struct discard_cmd *dc)
963 {
964 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
965 	unsigned long flags;
966 
967 	trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
968 
969 	spin_lock_irqsave(&dc->lock, flags);
970 	if (dc->bio_ref) {
971 		spin_unlock_irqrestore(&dc->lock, flags);
972 		return;
973 	}
974 	spin_unlock_irqrestore(&dc->lock, flags);
975 
976 	f2fs_bug_on(sbi, dc->ref);
977 
978 	if (dc->error == -EOPNOTSUPP)
979 		dc->error = 0;
980 
981 	if (dc->error)
982 		printk_ratelimited(
983 			"%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
984 			KERN_INFO, sbi->sb->s_id,
985 			dc->lstart, dc->start, dc->len, dc->error);
986 	__detach_discard_cmd(dcc, dc);
987 }
988 
989 static void f2fs_submit_discard_endio(struct bio *bio)
990 {
991 	struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
992 	unsigned long flags;
993 
994 	spin_lock_irqsave(&dc->lock, flags);
995 	if (!dc->error)
996 		dc->error = blk_status_to_errno(bio->bi_status);
997 	dc->bio_ref--;
998 	if (!dc->bio_ref && dc->state == D_SUBMIT) {
999 		dc->state = D_DONE;
1000 		complete_all(&dc->wait);
1001 	}
1002 	spin_unlock_irqrestore(&dc->lock, flags);
1003 	bio_put(bio);
1004 }
1005 
1006 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1007 				block_t start, block_t end)
1008 {
1009 #ifdef CONFIG_F2FS_CHECK_FS
1010 	struct seg_entry *sentry;
1011 	unsigned int segno;
1012 	block_t blk = start;
1013 	unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1014 	unsigned long *map;
1015 
1016 	while (blk < end) {
1017 		segno = GET_SEGNO(sbi, blk);
1018 		sentry = get_seg_entry(sbi, segno);
1019 		offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1020 
1021 		if (end < START_BLOCK(sbi, segno + 1))
1022 			size = GET_BLKOFF_FROM_SEG0(sbi, end);
1023 		else
1024 			size = max_blocks;
1025 		map = (unsigned long *)(sentry->cur_valid_map);
1026 		offset = __find_rev_next_bit(map, size, offset);
1027 		f2fs_bug_on(sbi, offset != size);
1028 		blk = START_BLOCK(sbi, segno + 1);
1029 	}
1030 #endif
1031 }
1032 
1033 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1034 				struct discard_policy *dpolicy,
1035 				int discard_type, unsigned int granularity)
1036 {
1037 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1038 
1039 	/* common policy */
1040 	dpolicy->type = discard_type;
1041 	dpolicy->sync = true;
1042 	dpolicy->ordered = false;
1043 	dpolicy->granularity = granularity;
1044 
1045 	dpolicy->max_requests = dcc->max_discard_request;
1046 	dpolicy->io_aware_gran = MAX_PLIST_NUM;
1047 	dpolicy->timeout = false;
1048 
1049 	if (discard_type == DPOLICY_BG) {
1050 		dpolicy->min_interval = dcc->min_discard_issue_time;
1051 		dpolicy->mid_interval = dcc->mid_discard_issue_time;
1052 		dpolicy->max_interval = dcc->max_discard_issue_time;
1053 		dpolicy->io_aware = true;
1054 		dpolicy->sync = false;
1055 		dpolicy->ordered = true;
1056 		if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1057 			dpolicy->granularity = 1;
1058 			if (atomic_read(&dcc->discard_cmd_cnt))
1059 				dpolicy->max_interval =
1060 					dcc->min_discard_issue_time;
1061 		}
1062 	} else if (discard_type == DPOLICY_FORCE) {
1063 		dpolicy->min_interval = dcc->min_discard_issue_time;
1064 		dpolicy->mid_interval = dcc->mid_discard_issue_time;
1065 		dpolicy->max_interval = dcc->max_discard_issue_time;
1066 		dpolicy->io_aware = false;
1067 	} else if (discard_type == DPOLICY_FSTRIM) {
1068 		dpolicy->io_aware = false;
1069 	} else if (discard_type == DPOLICY_UMOUNT) {
1070 		dpolicy->io_aware = false;
1071 		/* we need to issue all to keep CP_TRIMMED_FLAG */
1072 		dpolicy->granularity = 1;
1073 		dpolicy->timeout = true;
1074 	}
1075 }
1076 
1077 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1078 				struct block_device *bdev, block_t lstart,
1079 				block_t start, block_t len);
1080 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1081 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1082 						struct discard_policy *dpolicy,
1083 						struct discard_cmd *dc,
1084 						unsigned int *issued)
1085 {
1086 	struct block_device *bdev = dc->bdev;
1087 	unsigned int max_discard_blocks =
1088 			SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1089 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1090 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1091 					&(dcc->fstrim_list) : &(dcc->wait_list);
1092 	blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0;
1093 	block_t lstart, start, len, total_len;
1094 	int err = 0;
1095 
1096 	if (dc->state != D_PREP)
1097 		return 0;
1098 
1099 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1100 		return 0;
1101 
1102 	trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1103 
1104 	lstart = dc->lstart;
1105 	start = dc->start;
1106 	len = dc->len;
1107 	total_len = len;
1108 
1109 	dc->len = 0;
1110 
1111 	while (total_len && *issued < dpolicy->max_requests && !err) {
1112 		struct bio *bio = NULL;
1113 		unsigned long flags;
1114 		bool last = true;
1115 
1116 		if (len > max_discard_blocks) {
1117 			len = max_discard_blocks;
1118 			last = false;
1119 		}
1120 
1121 		(*issued)++;
1122 		if (*issued == dpolicy->max_requests)
1123 			last = true;
1124 
1125 		dc->len += len;
1126 
1127 		if (time_to_inject(sbi, FAULT_DISCARD)) {
1128 			f2fs_show_injection_info(sbi, FAULT_DISCARD);
1129 			err = -EIO;
1130 			goto submit;
1131 		}
1132 		err = __blkdev_issue_discard(bdev,
1133 					SECTOR_FROM_BLOCK(start),
1134 					SECTOR_FROM_BLOCK(len),
1135 					GFP_NOFS, &bio);
1136 submit:
1137 		if (err) {
1138 			spin_lock_irqsave(&dc->lock, flags);
1139 			if (dc->state == D_PARTIAL)
1140 				dc->state = D_SUBMIT;
1141 			spin_unlock_irqrestore(&dc->lock, flags);
1142 
1143 			break;
1144 		}
1145 
1146 		f2fs_bug_on(sbi, !bio);
1147 
1148 		/*
1149 		 * should keep before submission to avoid D_DONE
1150 		 * right away
1151 		 */
1152 		spin_lock_irqsave(&dc->lock, flags);
1153 		if (last)
1154 			dc->state = D_SUBMIT;
1155 		else
1156 			dc->state = D_PARTIAL;
1157 		dc->bio_ref++;
1158 		spin_unlock_irqrestore(&dc->lock, flags);
1159 
1160 		atomic_inc(&dcc->queued_discard);
1161 		dc->queued++;
1162 		list_move_tail(&dc->list, wait_list);
1163 
1164 		/* sanity check on discard range */
1165 		__check_sit_bitmap(sbi, lstart, lstart + len);
1166 
1167 		bio->bi_private = dc;
1168 		bio->bi_end_io = f2fs_submit_discard_endio;
1169 		bio->bi_opf |= flag;
1170 		submit_bio(bio);
1171 
1172 		atomic_inc(&dcc->issued_discard);
1173 
1174 		f2fs_update_iostat(sbi, FS_DISCARD, 1);
1175 
1176 		lstart += len;
1177 		start += len;
1178 		total_len -= len;
1179 		len = total_len;
1180 	}
1181 
1182 	if (!err && len) {
1183 		dcc->undiscard_blks -= len;
1184 		__update_discard_tree_range(sbi, bdev, lstart, start, len);
1185 	}
1186 	return err;
1187 }
1188 
1189 static void __insert_discard_tree(struct f2fs_sb_info *sbi,
1190 				struct block_device *bdev, block_t lstart,
1191 				block_t start, block_t len,
1192 				struct rb_node **insert_p,
1193 				struct rb_node *insert_parent)
1194 {
1195 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1196 	struct rb_node **p;
1197 	struct rb_node *parent = NULL;
1198 	bool leftmost = true;
1199 
1200 	if (insert_p && insert_parent) {
1201 		parent = insert_parent;
1202 		p = insert_p;
1203 		goto do_insert;
1204 	}
1205 
1206 	p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1207 							lstart, &leftmost);
1208 do_insert:
1209 	__attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1210 								p, leftmost);
1211 }
1212 
1213 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1214 						struct discard_cmd *dc)
1215 {
1216 	list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1217 }
1218 
1219 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1220 				struct discard_cmd *dc, block_t blkaddr)
1221 {
1222 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1223 	struct discard_info di = dc->di;
1224 	bool modified = false;
1225 
1226 	if (dc->state == D_DONE || dc->len == 1) {
1227 		__remove_discard_cmd(sbi, dc);
1228 		return;
1229 	}
1230 
1231 	dcc->undiscard_blks -= di.len;
1232 
1233 	if (blkaddr > di.lstart) {
1234 		dc->len = blkaddr - dc->lstart;
1235 		dcc->undiscard_blks += dc->len;
1236 		__relocate_discard_cmd(dcc, dc);
1237 		modified = true;
1238 	}
1239 
1240 	if (blkaddr < di.lstart + di.len - 1) {
1241 		if (modified) {
1242 			__insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1243 					di.start + blkaddr + 1 - di.lstart,
1244 					di.lstart + di.len - 1 - blkaddr,
1245 					NULL, NULL);
1246 		} else {
1247 			dc->lstart++;
1248 			dc->len--;
1249 			dc->start++;
1250 			dcc->undiscard_blks += dc->len;
1251 			__relocate_discard_cmd(dcc, dc);
1252 		}
1253 	}
1254 }
1255 
1256 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1257 				struct block_device *bdev, block_t lstart,
1258 				block_t start, block_t len)
1259 {
1260 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1261 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1262 	struct discard_cmd *dc;
1263 	struct discard_info di = {0};
1264 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1265 	unsigned int max_discard_blocks =
1266 			SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1267 	block_t end = lstart + len;
1268 
1269 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1270 					NULL, lstart,
1271 					(struct rb_entry **)&prev_dc,
1272 					(struct rb_entry **)&next_dc,
1273 					&insert_p, &insert_parent, true, NULL);
1274 	if (dc)
1275 		prev_dc = dc;
1276 
1277 	if (!prev_dc) {
1278 		di.lstart = lstart;
1279 		di.len = next_dc ? next_dc->lstart - lstart : len;
1280 		di.len = min(di.len, len);
1281 		di.start = start;
1282 	}
1283 
1284 	while (1) {
1285 		struct rb_node *node;
1286 		bool merged = false;
1287 		struct discard_cmd *tdc = NULL;
1288 
1289 		if (prev_dc) {
1290 			di.lstart = prev_dc->lstart + prev_dc->len;
1291 			if (di.lstart < lstart)
1292 				di.lstart = lstart;
1293 			if (di.lstart >= end)
1294 				break;
1295 
1296 			if (!next_dc || next_dc->lstart > end)
1297 				di.len = end - di.lstart;
1298 			else
1299 				di.len = next_dc->lstart - di.lstart;
1300 			di.start = start + di.lstart - lstart;
1301 		}
1302 
1303 		if (!di.len)
1304 			goto next;
1305 
1306 		if (prev_dc && prev_dc->state == D_PREP &&
1307 			prev_dc->bdev == bdev &&
1308 			__is_discard_back_mergeable(&di, &prev_dc->di,
1309 							max_discard_blocks)) {
1310 			prev_dc->di.len += di.len;
1311 			dcc->undiscard_blks += di.len;
1312 			__relocate_discard_cmd(dcc, prev_dc);
1313 			di = prev_dc->di;
1314 			tdc = prev_dc;
1315 			merged = true;
1316 		}
1317 
1318 		if (next_dc && next_dc->state == D_PREP &&
1319 			next_dc->bdev == bdev &&
1320 			__is_discard_front_mergeable(&di, &next_dc->di,
1321 							max_discard_blocks)) {
1322 			next_dc->di.lstart = di.lstart;
1323 			next_dc->di.len += di.len;
1324 			next_dc->di.start = di.start;
1325 			dcc->undiscard_blks += di.len;
1326 			__relocate_discard_cmd(dcc, next_dc);
1327 			if (tdc)
1328 				__remove_discard_cmd(sbi, tdc);
1329 			merged = true;
1330 		}
1331 
1332 		if (!merged) {
1333 			__insert_discard_tree(sbi, bdev, di.lstart, di.start,
1334 							di.len, NULL, NULL);
1335 		}
1336  next:
1337 		prev_dc = next_dc;
1338 		if (!prev_dc)
1339 			break;
1340 
1341 		node = rb_next(&prev_dc->rb_node);
1342 		next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1343 	}
1344 }
1345 
1346 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1347 		struct block_device *bdev, block_t blkstart, block_t blklen)
1348 {
1349 	block_t lblkstart = blkstart;
1350 
1351 	if (!f2fs_bdev_support_discard(bdev))
1352 		return 0;
1353 
1354 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
1355 
1356 	if (f2fs_is_multi_device(sbi)) {
1357 		int devi = f2fs_target_device_index(sbi, blkstart);
1358 
1359 		blkstart -= FDEV(devi).start_blk;
1360 	}
1361 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1362 	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1363 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1364 	return 0;
1365 }
1366 
1367 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1368 					struct discard_policy *dpolicy)
1369 {
1370 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1371 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1372 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1373 	struct discard_cmd *dc;
1374 	struct blk_plug plug;
1375 	unsigned int pos = dcc->next_pos;
1376 	unsigned int issued = 0;
1377 	bool io_interrupted = false;
1378 
1379 	mutex_lock(&dcc->cmd_lock);
1380 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1381 					NULL, pos,
1382 					(struct rb_entry **)&prev_dc,
1383 					(struct rb_entry **)&next_dc,
1384 					&insert_p, &insert_parent, true, NULL);
1385 	if (!dc)
1386 		dc = next_dc;
1387 
1388 	blk_start_plug(&plug);
1389 
1390 	while (dc) {
1391 		struct rb_node *node;
1392 		int err = 0;
1393 
1394 		if (dc->state != D_PREP)
1395 			goto next;
1396 
1397 		if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1398 			io_interrupted = true;
1399 			break;
1400 		}
1401 
1402 		dcc->next_pos = dc->lstart + dc->len;
1403 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1404 
1405 		if (issued >= dpolicy->max_requests)
1406 			break;
1407 next:
1408 		node = rb_next(&dc->rb_node);
1409 		if (err)
1410 			__remove_discard_cmd(sbi, dc);
1411 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1412 	}
1413 
1414 	blk_finish_plug(&plug);
1415 
1416 	if (!dc)
1417 		dcc->next_pos = 0;
1418 
1419 	mutex_unlock(&dcc->cmd_lock);
1420 
1421 	if (!issued && io_interrupted)
1422 		issued = -1;
1423 
1424 	return issued;
1425 }
1426 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1427 					struct discard_policy *dpolicy);
1428 
1429 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1430 					struct discard_policy *dpolicy)
1431 {
1432 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1433 	struct list_head *pend_list;
1434 	struct discard_cmd *dc, *tmp;
1435 	struct blk_plug plug;
1436 	int i, issued;
1437 	bool io_interrupted = false;
1438 
1439 	if (dpolicy->timeout)
1440 		f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1441 
1442 retry:
1443 	issued = 0;
1444 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1445 		if (dpolicy->timeout &&
1446 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1447 			break;
1448 
1449 		if (i + 1 < dpolicy->granularity)
1450 			break;
1451 
1452 		if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1453 			return __issue_discard_cmd_orderly(sbi, dpolicy);
1454 
1455 		pend_list = &dcc->pend_list[i];
1456 
1457 		mutex_lock(&dcc->cmd_lock);
1458 		if (list_empty(pend_list))
1459 			goto next;
1460 		if (unlikely(dcc->rbtree_check))
1461 			f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1462 							&dcc->root, false));
1463 		blk_start_plug(&plug);
1464 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1465 			f2fs_bug_on(sbi, dc->state != D_PREP);
1466 
1467 			if (dpolicy->timeout &&
1468 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1469 				break;
1470 
1471 			if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1472 						!is_idle(sbi, DISCARD_TIME)) {
1473 				io_interrupted = true;
1474 				break;
1475 			}
1476 
1477 			__submit_discard_cmd(sbi, dpolicy, dc, &issued);
1478 
1479 			if (issued >= dpolicy->max_requests)
1480 				break;
1481 		}
1482 		blk_finish_plug(&plug);
1483 next:
1484 		mutex_unlock(&dcc->cmd_lock);
1485 
1486 		if (issued >= dpolicy->max_requests || io_interrupted)
1487 			break;
1488 	}
1489 
1490 	if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1491 		__wait_all_discard_cmd(sbi, dpolicy);
1492 		goto retry;
1493 	}
1494 
1495 	if (!issued && io_interrupted)
1496 		issued = -1;
1497 
1498 	return issued;
1499 }
1500 
1501 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1502 {
1503 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1504 	struct list_head *pend_list;
1505 	struct discard_cmd *dc, *tmp;
1506 	int i;
1507 	bool dropped = false;
1508 
1509 	mutex_lock(&dcc->cmd_lock);
1510 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1511 		pend_list = &dcc->pend_list[i];
1512 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1513 			f2fs_bug_on(sbi, dc->state != D_PREP);
1514 			__remove_discard_cmd(sbi, dc);
1515 			dropped = true;
1516 		}
1517 	}
1518 	mutex_unlock(&dcc->cmd_lock);
1519 
1520 	return dropped;
1521 }
1522 
1523 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1524 {
1525 	__drop_discard_cmd(sbi);
1526 }
1527 
1528 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1529 							struct discard_cmd *dc)
1530 {
1531 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1532 	unsigned int len = 0;
1533 
1534 	wait_for_completion_io(&dc->wait);
1535 	mutex_lock(&dcc->cmd_lock);
1536 	f2fs_bug_on(sbi, dc->state != D_DONE);
1537 	dc->ref--;
1538 	if (!dc->ref) {
1539 		if (!dc->error)
1540 			len = dc->len;
1541 		__remove_discard_cmd(sbi, dc);
1542 	}
1543 	mutex_unlock(&dcc->cmd_lock);
1544 
1545 	return len;
1546 }
1547 
1548 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1549 						struct discard_policy *dpolicy,
1550 						block_t start, block_t end)
1551 {
1552 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1553 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1554 					&(dcc->fstrim_list) : &(dcc->wait_list);
1555 	struct discard_cmd *dc = NULL, *iter, *tmp;
1556 	unsigned int trimmed = 0;
1557 
1558 next:
1559 	dc = NULL;
1560 
1561 	mutex_lock(&dcc->cmd_lock);
1562 	list_for_each_entry_safe(iter, tmp, wait_list, list) {
1563 		if (iter->lstart + iter->len <= start || end <= iter->lstart)
1564 			continue;
1565 		if (iter->len < dpolicy->granularity)
1566 			continue;
1567 		if (iter->state == D_DONE && !iter->ref) {
1568 			wait_for_completion_io(&iter->wait);
1569 			if (!iter->error)
1570 				trimmed += iter->len;
1571 			__remove_discard_cmd(sbi, iter);
1572 		} else {
1573 			iter->ref++;
1574 			dc = iter;
1575 			break;
1576 		}
1577 	}
1578 	mutex_unlock(&dcc->cmd_lock);
1579 
1580 	if (dc) {
1581 		trimmed += __wait_one_discard_bio(sbi, dc);
1582 		goto next;
1583 	}
1584 
1585 	return trimmed;
1586 }
1587 
1588 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1589 						struct discard_policy *dpolicy)
1590 {
1591 	struct discard_policy dp;
1592 	unsigned int discard_blks;
1593 
1594 	if (dpolicy)
1595 		return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1596 
1597 	/* wait all */
1598 	__init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1599 	discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1600 	__init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1601 	discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1602 
1603 	return discard_blks;
1604 }
1605 
1606 /* This should be covered by global mutex, &sit_i->sentry_lock */
1607 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1608 {
1609 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1610 	struct discard_cmd *dc;
1611 	bool need_wait = false;
1612 
1613 	mutex_lock(&dcc->cmd_lock);
1614 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1615 							NULL, blkaddr);
1616 	if (dc) {
1617 		if (dc->state == D_PREP) {
1618 			__punch_discard_cmd(sbi, dc, blkaddr);
1619 		} else {
1620 			dc->ref++;
1621 			need_wait = true;
1622 		}
1623 	}
1624 	mutex_unlock(&dcc->cmd_lock);
1625 
1626 	if (need_wait)
1627 		__wait_one_discard_bio(sbi, dc);
1628 }
1629 
1630 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1631 {
1632 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1633 
1634 	if (dcc && dcc->f2fs_issue_discard) {
1635 		struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1636 
1637 		dcc->f2fs_issue_discard = NULL;
1638 		kthread_stop(discard_thread);
1639 	}
1640 }
1641 
1642 /* This comes from f2fs_put_super */
1643 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1644 {
1645 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1646 	struct discard_policy dpolicy;
1647 	bool dropped;
1648 
1649 	__init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1650 					dcc->discard_granularity);
1651 	__issue_discard_cmd(sbi, &dpolicy);
1652 	dropped = __drop_discard_cmd(sbi);
1653 
1654 	/* just to make sure there is no pending discard commands */
1655 	__wait_all_discard_cmd(sbi, NULL);
1656 
1657 	f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1658 	return dropped;
1659 }
1660 
1661 static int issue_discard_thread(void *data)
1662 {
1663 	struct f2fs_sb_info *sbi = data;
1664 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1665 	wait_queue_head_t *q = &dcc->discard_wait_queue;
1666 	struct discard_policy dpolicy;
1667 	unsigned int wait_ms = dcc->min_discard_issue_time;
1668 	int issued;
1669 
1670 	set_freezable();
1671 
1672 	do {
1673 		if (sbi->gc_mode == GC_URGENT_HIGH ||
1674 			!f2fs_available_free_memory(sbi, DISCARD_CACHE))
1675 			__init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1676 		else
1677 			__init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1678 						dcc->discard_granularity);
1679 
1680 		if (!atomic_read(&dcc->discard_cmd_cnt))
1681 		       wait_ms = dpolicy.max_interval;
1682 
1683 		wait_event_interruptible_timeout(*q,
1684 				kthread_should_stop() || freezing(current) ||
1685 				dcc->discard_wake,
1686 				msecs_to_jiffies(wait_ms));
1687 
1688 		if (dcc->discard_wake)
1689 			dcc->discard_wake = 0;
1690 
1691 		/* clean up pending candidates before going to sleep */
1692 		if (atomic_read(&dcc->queued_discard))
1693 			__wait_all_discard_cmd(sbi, NULL);
1694 
1695 		if (try_to_freeze())
1696 			continue;
1697 		if (f2fs_readonly(sbi->sb))
1698 			continue;
1699 		if (kthread_should_stop())
1700 			return 0;
1701 		if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1702 			wait_ms = dpolicy.max_interval;
1703 			continue;
1704 		}
1705 		if (!atomic_read(&dcc->discard_cmd_cnt))
1706 			continue;
1707 
1708 		sb_start_intwrite(sbi->sb);
1709 
1710 		issued = __issue_discard_cmd(sbi, &dpolicy);
1711 		if (issued > 0) {
1712 			__wait_all_discard_cmd(sbi, &dpolicy);
1713 			wait_ms = dpolicy.min_interval;
1714 		} else if (issued == -1) {
1715 			wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1716 			if (!wait_ms)
1717 				wait_ms = dpolicy.mid_interval;
1718 		} else {
1719 			wait_ms = dpolicy.max_interval;
1720 		}
1721 
1722 		sb_end_intwrite(sbi->sb);
1723 
1724 	} while (!kthread_should_stop());
1725 	return 0;
1726 }
1727 
1728 #ifdef CONFIG_BLK_DEV_ZONED
1729 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1730 		struct block_device *bdev, block_t blkstart, block_t blklen)
1731 {
1732 	sector_t sector, nr_sects;
1733 	block_t lblkstart = blkstart;
1734 	int devi = 0;
1735 
1736 	if (f2fs_is_multi_device(sbi)) {
1737 		devi = f2fs_target_device_index(sbi, blkstart);
1738 		if (blkstart < FDEV(devi).start_blk ||
1739 		    blkstart > FDEV(devi).end_blk) {
1740 			f2fs_err(sbi, "Invalid block %x", blkstart);
1741 			return -EIO;
1742 		}
1743 		blkstart -= FDEV(devi).start_blk;
1744 	}
1745 
1746 	/* For sequential zones, reset the zone write pointer */
1747 	if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1748 		sector = SECTOR_FROM_BLOCK(blkstart);
1749 		nr_sects = SECTOR_FROM_BLOCK(blklen);
1750 
1751 		if (sector & (bdev_zone_sectors(bdev) - 1) ||
1752 				nr_sects != bdev_zone_sectors(bdev)) {
1753 			f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1754 				 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1755 				 blkstart, blklen);
1756 			return -EIO;
1757 		}
1758 		trace_f2fs_issue_reset_zone(bdev, blkstart);
1759 		return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1760 					sector, nr_sects, GFP_NOFS);
1761 	}
1762 
1763 	/* For conventional zones, use regular discard if supported */
1764 	return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1765 }
1766 #endif
1767 
1768 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1769 		struct block_device *bdev, block_t blkstart, block_t blklen)
1770 {
1771 #ifdef CONFIG_BLK_DEV_ZONED
1772 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1773 		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1774 #endif
1775 	return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1776 }
1777 
1778 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1779 				block_t blkstart, block_t blklen)
1780 {
1781 	sector_t start = blkstart, len = 0;
1782 	struct block_device *bdev;
1783 	struct seg_entry *se;
1784 	unsigned int offset;
1785 	block_t i;
1786 	int err = 0;
1787 
1788 	bdev = f2fs_target_device(sbi, blkstart, NULL);
1789 
1790 	for (i = blkstart; i < blkstart + blklen; i++, len++) {
1791 		if (i != start) {
1792 			struct block_device *bdev2 =
1793 				f2fs_target_device(sbi, i, NULL);
1794 
1795 			if (bdev2 != bdev) {
1796 				err = __issue_discard_async(sbi, bdev,
1797 						start, len);
1798 				if (err)
1799 					return err;
1800 				bdev = bdev2;
1801 				start = i;
1802 				len = 0;
1803 			}
1804 		}
1805 
1806 		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1807 		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1808 
1809 		if (f2fs_block_unit_discard(sbi) &&
1810 				!f2fs_test_and_set_bit(offset, se->discard_map))
1811 			sbi->discard_blks--;
1812 	}
1813 
1814 	if (len)
1815 		err = __issue_discard_async(sbi, bdev, start, len);
1816 	return err;
1817 }
1818 
1819 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1820 							bool check_only)
1821 {
1822 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1823 	int max_blocks = sbi->blocks_per_seg;
1824 	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1825 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1826 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1827 	unsigned long *discard_map = (unsigned long *)se->discard_map;
1828 	unsigned long *dmap = SIT_I(sbi)->tmp_map;
1829 	unsigned int start = 0, end = -1;
1830 	bool force = (cpc->reason & CP_DISCARD);
1831 	struct discard_entry *de = NULL;
1832 	struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1833 	int i;
1834 
1835 	if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi) ||
1836 			!f2fs_block_unit_discard(sbi))
1837 		return false;
1838 
1839 	if (!force) {
1840 		if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1841 			SM_I(sbi)->dcc_info->nr_discards >=
1842 				SM_I(sbi)->dcc_info->max_discards)
1843 			return false;
1844 	}
1845 
1846 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1847 	for (i = 0; i < entries; i++)
1848 		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1849 				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1850 
1851 	while (force || SM_I(sbi)->dcc_info->nr_discards <=
1852 				SM_I(sbi)->dcc_info->max_discards) {
1853 		start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1854 		if (start >= max_blocks)
1855 			break;
1856 
1857 		end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1858 		if (force && start && end != max_blocks
1859 					&& (end - start) < cpc->trim_minlen)
1860 			continue;
1861 
1862 		if (check_only)
1863 			return true;
1864 
1865 		if (!de) {
1866 			de = f2fs_kmem_cache_alloc(discard_entry_slab,
1867 						GFP_F2FS_ZERO, true, NULL);
1868 			de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1869 			list_add_tail(&de->list, head);
1870 		}
1871 
1872 		for (i = start; i < end; i++)
1873 			__set_bit_le(i, (void *)de->discard_map);
1874 
1875 		SM_I(sbi)->dcc_info->nr_discards += end - start;
1876 	}
1877 	return false;
1878 }
1879 
1880 static void release_discard_addr(struct discard_entry *entry)
1881 {
1882 	list_del(&entry->list);
1883 	kmem_cache_free(discard_entry_slab, entry);
1884 }
1885 
1886 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1887 {
1888 	struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1889 	struct discard_entry *entry, *this;
1890 
1891 	/* drop caches */
1892 	list_for_each_entry_safe(entry, this, head, list)
1893 		release_discard_addr(entry);
1894 }
1895 
1896 /*
1897  * Should call f2fs_clear_prefree_segments after checkpoint is done.
1898  */
1899 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1900 {
1901 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1902 	unsigned int segno;
1903 
1904 	mutex_lock(&dirty_i->seglist_lock);
1905 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1906 		__set_test_and_free(sbi, segno, false);
1907 	mutex_unlock(&dirty_i->seglist_lock);
1908 }
1909 
1910 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1911 						struct cp_control *cpc)
1912 {
1913 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1914 	struct list_head *head = &dcc->entry_list;
1915 	struct discard_entry *entry, *this;
1916 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1917 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1918 	unsigned int start = 0, end = -1;
1919 	unsigned int secno, start_segno;
1920 	bool force = (cpc->reason & CP_DISCARD);
1921 	bool section_alignment = F2FS_OPTION(sbi).discard_unit ==
1922 						DISCARD_UNIT_SECTION;
1923 
1924 	if (f2fs_lfs_mode(sbi) && __is_large_section(sbi))
1925 		section_alignment = true;
1926 
1927 	mutex_lock(&dirty_i->seglist_lock);
1928 
1929 	while (1) {
1930 		int i;
1931 
1932 		if (section_alignment && end != -1)
1933 			end--;
1934 		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1935 		if (start >= MAIN_SEGS(sbi))
1936 			break;
1937 		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1938 								start + 1);
1939 
1940 		if (section_alignment) {
1941 			start = rounddown(start, sbi->segs_per_sec);
1942 			end = roundup(end, sbi->segs_per_sec);
1943 		}
1944 
1945 		for (i = start; i < end; i++) {
1946 			if (test_and_clear_bit(i, prefree_map))
1947 				dirty_i->nr_dirty[PRE]--;
1948 		}
1949 
1950 		if (!f2fs_realtime_discard_enable(sbi))
1951 			continue;
1952 
1953 		if (force && start >= cpc->trim_start &&
1954 					(end - 1) <= cpc->trim_end)
1955 				continue;
1956 
1957 		if (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi)) {
1958 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1959 				(end - start) << sbi->log_blocks_per_seg);
1960 			continue;
1961 		}
1962 next:
1963 		secno = GET_SEC_FROM_SEG(sbi, start);
1964 		start_segno = GET_SEG_FROM_SEC(sbi, secno);
1965 		if (!IS_CURSEC(sbi, secno) &&
1966 			!get_valid_blocks(sbi, start, true))
1967 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1968 				sbi->segs_per_sec << sbi->log_blocks_per_seg);
1969 
1970 		start = start_segno + sbi->segs_per_sec;
1971 		if (start < end)
1972 			goto next;
1973 		else
1974 			end = start - 1;
1975 	}
1976 	mutex_unlock(&dirty_i->seglist_lock);
1977 
1978 	if (!f2fs_block_unit_discard(sbi))
1979 		goto wakeup;
1980 
1981 	/* send small discards */
1982 	list_for_each_entry_safe(entry, this, head, list) {
1983 		unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1984 		bool is_valid = test_bit_le(0, entry->discard_map);
1985 
1986 find_next:
1987 		if (is_valid) {
1988 			next_pos = find_next_zero_bit_le(entry->discard_map,
1989 					sbi->blocks_per_seg, cur_pos);
1990 			len = next_pos - cur_pos;
1991 
1992 			if (f2fs_sb_has_blkzoned(sbi) ||
1993 			    (force && len < cpc->trim_minlen))
1994 				goto skip;
1995 
1996 			f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1997 									len);
1998 			total_len += len;
1999 		} else {
2000 			next_pos = find_next_bit_le(entry->discard_map,
2001 					sbi->blocks_per_seg, cur_pos);
2002 		}
2003 skip:
2004 		cur_pos = next_pos;
2005 		is_valid = !is_valid;
2006 
2007 		if (cur_pos < sbi->blocks_per_seg)
2008 			goto find_next;
2009 
2010 		release_discard_addr(entry);
2011 		dcc->nr_discards -= total_len;
2012 	}
2013 
2014 wakeup:
2015 	wake_up_discard_thread(sbi, false);
2016 }
2017 
2018 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi)
2019 {
2020 	dev_t dev = sbi->sb->s_bdev->bd_dev;
2021 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2022 	int err = 0;
2023 
2024 	if (!f2fs_realtime_discard_enable(sbi))
2025 		return 0;
2026 
2027 	dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2028 				"f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2029 	if (IS_ERR(dcc->f2fs_issue_discard))
2030 		err = PTR_ERR(dcc->f2fs_issue_discard);
2031 
2032 	return err;
2033 }
2034 
2035 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2036 {
2037 	struct discard_cmd_control *dcc;
2038 	int err = 0, i;
2039 
2040 	if (SM_I(sbi)->dcc_info) {
2041 		dcc = SM_I(sbi)->dcc_info;
2042 		goto init_thread;
2043 	}
2044 
2045 	dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2046 	if (!dcc)
2047 		return -ENOMEM;
2048 
2049 	dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2050 	if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT)
2051 		dcc->discard_granularity = sbi->blocks_per_seg;
2052 	else if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION)
2053 		dcc->discard_granularity = BLKS_PER_SEC(sbi);
2054 
2055 	INIT_LIST_HEAD(&dcc->entry_list);
2056 	for (i = 0; i < MAX_PLIST_NUM; i++)
2057 		INIT_LIST_HEAD(&dcc->pend_list[i]);
2058 	INIT_LIST_HEAD(&dcc->wait_list);
2059 	INIT_LIST_HEAD(&dcc->fstrim_list);
2060 	mutex_init(&dcc->cmd_lock);
2061 	atomic_set(&dcc->issued_discard, 0);
2062 	atomic_set(&dcc->queued_discard, 0);
2063 	atomic_set(&dcc->discard_cmd_cnt, 0);
2064 	dcc->nr_discards = 0;
2065 	dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2066 	dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST;
2067 	dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME;
2068 	dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME;
2069 	dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME;
2070 	dcc->undiscard_blks = 0;
2071 	dcc->next_pos = 0;
2072 	dcc->root = RB_ROOT_CACHED;
2073 	dcc->rbtree_check = false;
2074 
2075 	init_waitqueue_head(&dcc->discard_wait_queue);
2076 	SM_I(sbi)->dcc_info = dcc;
2077 init_thread:
2078 	err = f2fs_start_discard_thread(sbi);
2079 	if (err) {
2080 		kfree(dcc);
2081 		SM_I(sbi)->dcc_info = NULL;
2082 	}
2083 
2084 	return err;
2085 }
2086 
2087 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2088 {
2089 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2090 
2091 	if (!dcc)
2092 		return;
2093 
2094 	f2fs_stop_discard_thread(sbi);
2095 
2096 	/*
2097 	 * Recovery can cache discard commands, so in error path of
2098 	 * fill_super(), it needs to give a chance to handle them.
2099 	 */
2100 	if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2101 		f2fs_issue_discard_timeout(sbi);
2102 
2103 	kfree(dcc);
2104 	SM_I(sbi)->dcc_info = NULL;
2105 }
2106 
2107 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2108 {
2109 	struct sit_info *sit_i = SIT_I(sbi);
2110 
2111 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2112 		sit_i->dirty_sentries++;
2113 		return false;
2114 	}
2115 
2116 	return true;
2117 }
2118 
2119 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2120 					unsigned int segno, int modified)
2121 {
2122 	struct seg_entry *se = get_seg_entry(sbi, segno);
2123 
2124 	se->type = type;
2125 	if (modified)
2126 		__mark_sit_entry_dirty(sbi, segno);
2127 }
2128 
2129 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2130 								block_t blkaddr)
2131 {
2132 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
2133 
2134 	if (segno == NULL_SEGNO)
2135 		return 0;
2136 	return get_seg_entry(sbi, segno)->mtime;
2137 }
2138 
2139 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2140 						unsigned long long old_mtime)
2141 {
2142 	struct seg_entry *se;
2143 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
2144 	unsigned long long ctime = get_mtime(sbi, false);
2145 	unsigned long long mtime = old_mtime ? old_mtime : ctime;
2146 
2147 	if (segno == NULL_SEGNO)
2148 		return;
2149 
2150 	se = get_seg_entry(sbi, segno);
2151 
2152 	if (!se->mtime)
2153 		se->mtime = mtime;
2154 	else
2155 		se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2156 						se->valid_blocks + 1);
2157 
2158 	if (ctime > SIT_I(sbi)->max_mtime)
2159 		SIT_I(sbi)->max_mtime = ctime;
2160 }
2161 
2162 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2163 {
2164 	struct seg_entry *se;
2165 	unsigned int segno, offset;
2166 	long int new_vblocks;
2167 	bool exist;
2168 #ifdef CONFIG_F2FS_CHECK_FS
2169 	bool mir_exist;
2170 #endif
2171 
2172 	segno = GET_SEGNO(sbi, blkaddr);
2173 
2174 	se = get_seg_entry(sbi, segno);
2175 	new_vblocks = se->valid_blocks + del;
2176 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2177 
2178 	f2fs_bug_on(sbi, (new_vblocks < 0 ||
2179 			(new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2180 
2181 	se->valid_blocks = new_vblocks;
2182 
2183 	/* Update valid block bitmap */
2184 	if (del > 0) {
2185 		exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2186 #ifdef CONFIG_F2FS_CHECK_FS
2187 		mir_exist = f2fs_test_and_set_bit(offset,
2188 						se->cur_valid_map_mir);
2189 		if (unlikely(exist != mir_exist)) {
2190 			f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2191 				 blkaddr, exist);
2192 			f2fs_bug_on(sbi, 1);
2193 		}
2194 #endif
2195 		if (unlikely(exist)) {
2196 			f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2197 				 blkaddr);
2198 			f2fs_bug_on(sbi, 1);
2199 			se->valid_blocks--;
2200 			del = 0;
2201 		}
2202 
2203 		if (f2fs_block_unit_discard(sbi) &&
2204 				!f2fs_test_and_set_bit(offset, se->discard_map))
2205 			sbi->discard_blks--;
2206 
2207 		/*
2208 		 * SSR should never reuse block which is checkpointed
2209 		 * or newly invalidated.
2210 		 */
2211 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2212 			if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2213 				se->ckpt_valid_blocks++;
2214 		}
2215 	} else {
2216 		exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2217 #ifdef CONFIG_F2FS_CHECK_FS
2218 		mir_exist = f2fs_test_and_clear_bit(offset,
2219 						se->cur_valid_map_mir);
2220 		if (unlikely(exist != mir_exist)) {
2221 			f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2222 				 blkaddr, exist);
2223 			f2fs_bug_on(sbi, 1);
2224 		}
2225 #endif
2226 		if (unlikely(!exist)) {
2227 			f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2228 				 blkaddr);
2229 			f2fs_bug_on(sbi, 1);
2230 			se->valid_blocks++;
2231 			del = 0;
2232 		} else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2233 			/*
2234 			 * If checkpoints are off, we must not reuse data that
2235 			 * was used in the previous checkpoint. If it was used
2236 			 * before, we must track that to know how much space we
2237 			 * really have.
2238 			 */
2239 			if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2240 				spin_lock(&sbi->stat_lock);
2241 				sbi->unusable_block_count++;
2242 				spin_unlock(&sbi->stat_lock);
2243 			}
2244 		}
2245 
2246 		if (f2fs_block_unit_discard(sbi) &&
2247 			f2fs_test_and_clear_bit(offset, se->discard_map))
2248 			sbi->discard_blks++;
2249 	}
2250 	if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2251 		se->ckpt_valid_blocks += del;
2252 
2253 	__mark_sit_entry_dirty(sbi, segno);
2254 
2255 	/* update total number of valid blocks to be written in ckpt area */
2256 	SIT_I(sbi)->written_valid_blocks += del;
2257 
2258 	if (__is_large_section(sbi))
2259 		get_sec_entry(sbi, segno)->valid_blocks += del;
2260 }
2261 
2262 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2263 {
2264 	unsigned int segno = GET_SEGNO(sbi, addr);
2265 	struct sit_info *sit_i = SIT_I(sbi);
2266 
2267 	f2fs_bug_on(sbi, addr == NULL_ADDR);
2268 	if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2269 		return;
2270 
2271 	invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2272 	f2fs_invalidate_compress_page(sbi, addr);
2273 
2274 	/* add it into sit main buffer */
2275 	down_write(&sit_i->sentry_lock);
2276 
2277 	update_segment_mtime(sbi, addr, 0);
2278 	update_sit_entry(sbi, addr, -1);
2279 
2280 	/* add it into dirty seglist */
2281 	locate_dirty_segment(sbi, segno);
2282 
2283 	up_write(&sit_i->sentry_lock);
2284 }
2285 
2286 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2287 {
2288 	struct sit_info *sit_i = SIT_I(sbi);
2289 	unsigned int segno, offset;
2290 	struct seg_entry *se;
2291 	bool is_cp = false;
2292 
2293 	if (!__is_valid_data_blkaddr(blkaddr))
2294 		return true;
2295 
2296 	down_read(&sit_i->sentry_lock);
2297 
2298 	segno = GET_SEGNO(sbi, blkaddr);
2299 	se = get_seg_entry(sbi, segno);
2300 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2301 
2302 	if (f2fs_test_bit(offset, se->ckpt_valid_map))
2303 		is_cp = true;
2304 
2305 	up_read(&sit_i->sentry_lock);
2306 
2307 	return is_cp;
2308 }
2309 
2310 /*
2311  * This function should be resided under the curseg_mutex lock
2312  */
2313 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2314 					struct f2fs_summary *sum)
2315 {
2316 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2317 	void *addr = curseg->sum_blk;
2318 
2319 	addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2320 	memcpy(addr, sum, sizeof(struct f2fs_summary));
2321 }
2322 
2323 /*
2324  * Calculate the number of current summary pages for writing
2325  */
2326 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2327 {
2328 	int valid_sum_count = 0;
2329 	int i, sum_in_page;
2330 
2331 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2332 		if (sbi->ckpt->alloc_type[i] == SSR)
2333 			valid_sum_count += sbi->blocks_per_seg;
2334 		else {
2335 			if (for_ra)
2336 				valid_sum_count += le16_to_cpu(
2337 					F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2338 			else
2339 				valid_sum_count += curseg_blkoff(sbi, i);
2340 		}
2341 	}
2342 
2343 	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2344 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2345 	if (valid_sum_count <= sum_in_page)
2346 		return 1;
2347 	else if ((valid_sum_count - sum_in_page) <=
2348 		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2349 		return 2;
2350 	return 3;
2351 }
2352 
2353 /*
2354  * Caller should put this summary page
2355  */
2356 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2357 {
2358 	if (unlikely(f2fs_cp_error(sbi)))
2359 		return ERR_PTR(-EIO);
2360 	return f2fs_get_meta_page_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2361 }
2362 
2363 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2364 					void *src, block_t blk_addr)
2365 {
2366 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2367 
2368 	memcpy(page_address(page), src, PAGE_SIZE);
2369 	set_page_dirty(page);
2370 	f2fs_put_page(page, 1);
2371 }
2372 
2373 static void write_sum_page(struct f2fs_sb_info *sbi,
2374 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
2375 {
2376 	f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2377 }
2378 
2379 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2380 						int type, block_t blk_addr)
2381 {
2382 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2383 	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2384 	struct f2fs_summary_block *src = curseg->sum_blk;
2385 	struct f2fs_summary_block *dst;
2386 
2387 	dst = (struct f2fs_summary_block *)page_address(page);
2388 	memset(dst, 0, PAGE_SIZE);
2389 
2390 	mutex_lock(&curseg->curseg_mutex);
2391 
2392 	down_read(&curseg->journal_rwsem);
2393 	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2394 	up_read(&curseg->journal_rwsem);
2395 
2396 	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2397 	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2398 
2399 	mutex_unlock(&curseg->curseg_mutex);
2400 
2401 	set_page_dirty(page);
2402 	f2fs_put_page(page, 1);
2403 }
2404 
2405 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2406 				struct curseg_info *curseg, int type)
2407 {
2408 	unsigned int segno = curseg->segno + 1;
2409 	struct free_segmap_info *free_i = FREE_I(sbi);
2410 
2411 	if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2412 		return !test_bit(segno, free_i->free_segmap);
2413 	return 0;
2414 }
2415 
2416 /*
2417  * Find a new segment from the free segments bitmap to right order
2418  * This function should be returned with success, otherwise BUG
2419  */
2420 static void get_new_segment(struct f2fs_sb_info *sbi,
2421 			unsigned int *newseg, bool new_sec, int dir)
2422 {
2423 	struct free_segmap_info *free_i = FREE_I(sbi);
2424 	unsigned int segno, secno, zoneno;
2425 	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2426 	unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2427 	unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2428 	unsigned int left_start = hint;
2429 	bool init = true;
2430 	int go_left = 0;
2431 	int i;
2432 
2433 	spin_lock(&free_i->segmap_lock);
2434 
2435 	if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2436 		segno = find_next_zero_bit(free_i->free_segmap,
2437 			GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2438 		if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2439 			goto got_it;
2440 	}
2441 find_other_zone:
2442 	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2443 	if (secno >= MAIN_SECS(sbi)) {
2444 		if (dir == ALLOC_RIGHT) {
2445 			secno = find_first_zero_bit(free_i->free_secmap,
2446 							MAIN_SECS(sbi));
2447 			f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2448 		} else {
2449 			go_left = 1;
2450 			left_start = hint - 1;
2451 		}
2452 	}
2453 	if (go_left == 0)
2454 		goto skip_left;
2455 
2456 	while (test_bit(left_start, free_i->free_secmap)) {
2457 		if (left_start > 0) {
2458 			left_start--;
2459 			continue;
2460 		}
2461 		left_start = find_first_zero_bit(free_i->free_secmap,
2462 							MAIN_SECS(sbi));
2463 		f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2464 		break;
2465 	}
2466 	secno = left_start;
2467 skip_left:
2468 	segno = GET_SEG_FROM_SEC(sbi, secno);
2469 	zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2470 
2471 	/* give up on finding another zone */
2472 	if (!init)
2473 		goto got_it;
2474 	if (sbi->secs_per_zone == 1)
2475 		goto got_it;
2476 	if (zoneno == old_zoneno)
2477 		goto got_it;
2478 	if (dir == ALLOC_LEFT) {
2479 		if (!go_left && zoneno + 1 >= total_zones)
2480 			goto got_it;
2481 		if (go_left && zoneno == 0)
2482 			goto got_it;
2483 	}
2484 	for (i = 0; i < NR_CURSEG_TYPE; i++)
2485 		if (CURSEG_I(sbi, i)->zone == zoneno)
2486 			break;
2487 
2488 	if (i < NR_CURSEG_TYPE) {
2489 		/* zone is in user, try another */
2490 		if (go_left)
2491 			hint = zoneno * sbi->secs_per_zone - 1;
2492 		else if (zoneno + 1 >= total_zones)
2493 			hint = 0;
2494 		else
2495 			hint = (zoneno + 1) * sbi->secs_per_zone;
2496 		init = false;
2497 		goto find_other_zone;
2498 	}
2499 got_it:
2500 	/* set it as dirty segment in free segmap */
2501 	f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2502 	__set_inuse(sbi, segno);
2503 	*newseg = segno;
2504 	spin_unlock(&free_i->segmap_lock);
2505 }
2506 
2507 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2508 {
2509 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2510 	struct summary_footer *sum_footer;
2511 	unsigned short seg_type = curseg->seg_type;
2512 
2513 	curseg->inited = true;
2514 	curseg->segno = curseg->next_segno;
2515 	curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2516 	curseg->next_blkoff = 0;
2517 	curseg->next_segno = NULL_SEGNO;
2518 
2519 	sum_footer = &(curseg->sum_blk->footer);
2520 	memset(sum_footer, 0, sizeof(struct summary_footer));
2521 
2522 	sanity_check_seg_type(sbi, seg_type);
2523 
2524 	if (IS_DATASEG(seg_type))
2525 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2526 	if (IS_NODESEG(seg_type))
2527 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2528 	__set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2529 }
2530 
2531 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2532 {
2533 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2534 	unsigned short seg_type = curseg->seg_type;
2535 
2536 	sanity_check_seg_type(sbi, seg_type);
2537 	if (f2fs_need_rand_seg(sbi))
2538 		return prandom_u32() % (MAIN_SECS(sbi) * sbi->segs_per_sec);
2539 
2540 	/* if segs_per_sec is large than 1, we need to keep original policy. */
2541 	if (__is_large_section(sbi))
2542 		return curseg->segno;
2543 
2544 	/* inmem log may not locate on any segment after mount */
2545 	if (!curseg->inited)
2546 		return 0;
2547 
2548 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2549 		return 0;
2550 
2551 	if (test_opt(sbi, NOHEAP) &&
2552 		(seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type)))
2553 		return 0;
2554 
2555 	if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2556 		return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2557 
2558 	/* find segments from 0 to reuse freed segments */
2559 	if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2560 		return 0;
2561 
2562 	return curseg->segno;
2563 }
2564 
2565 /*
2566  * Allocate a current working segment.
2567  * This function always allocates a free segment in LFS manner.
2568  */
2569 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2570 {
2571 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2572 	unsigned short seg_type = curseg->seg_type;
2573 	unsigned int segno = curseg->segno;
2574 	int dir = ALLOC_LEFT;
2575 
2576 	if (curseg->inited)
2577 		write_sum_page(sbi, curseg->sum_blk,
2578 				GET_SUM_BLOCK(sbi, segno));
2579 	if (seg_type == CURSEG_WARM_DATA || seg_type == CURSEG_COLD_DATA)
2580 		dir = ALLOC_RIGHT;
2581 
2582 	if (test_opt(sbi, NOHEAP))
2583 		dir = ALLOC_RIGHT;
2584 
2585 	segno = __get_next_segno(sbi, type);
2586 	get_new_segment(sbi, &segno, new_sec, dir);
2587 	curseg->next_segno = segno;
2588 	reset_curseg(sbi, type, 1);
2589 	curseg->alloc_type = LFS;
2590 	if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
2591 		curseg->fragment_remained_chunk =
2592 				prandom_u32() % sbi->max_fragment_chunk + 1;
2593 }
2594 
2595 static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2596 					int segno, block_t start)
2597 {
2598 	struct seg_entry *se = get_seg_entry(sbi, segno);
2599 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2600 	unsigned long *target_map = SIT_I(sbi)->tmp_map;
2601 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2602 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2603 	int i;
2604 
2605 	for (i = 0; i < entries; i++)
2606 		target_map[i] = ckpt_map[i] | cur_map[i];
2607 
2608 	return __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2609 }
2610 
2611 /*
2612  * If a segment is written by LFS manner, next block offset is just obtained
2613  * by increasing the current block offset. However, if a segment is written by
2614  * SSR manner, next block offset obtained by calling __next_free_blkoff
2615  */
2616 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2617 				struct curseg_info *seg)
2618 {
2619 	if (seg->alloc_type == SSR) {
2620 		seg->next_blkoff =
2621 			__next_free_blkoff(sbi, seg->segno,
2622 						seg->next_blkoff + 1);
2623 	} else {
2624 		seg->next_blkoff++;
2625 		if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK) {
2626 			/* To allocate block chunks in different sizes, use random number */
2627 			if (--seg->fragment_remained_chunk <= 0) {
2628 				seg->fragment_remained_chunk =
2629 				   prandom_u32() % sbi->max_fragment_chunk + 1;
2630 				seg->next_blkoff +=
2631 				   prandom_u32() % sbi->max_fragment_hole + 1;
2632 			}
2633 		}
2634 	}
2635 }
2636 
2637 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
2638 {
2639 	return __next_free_blkoff(sbi, segno, 0) < sbi->blocks_per_seg;
2640 }
2641 
2642 /*
2643  * This function always allocates a used segment(from dirty seglist) by SSR
2644  * manner, so it should recover the existing segment information of valid blocks
2645  */
2646 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool flush)
2647 {
2648 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2649 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2650 	unsigned int new_segno = curseg->next_segno;
2651 	struct f2fs_summary_block *sum_node;
2652 	struct page *sum_page;
2653 
2654 	if (flush)
2655 		write_sum_page(sbi, curseg->sum_blk,
2656 					GET_SUM_BLOCK(sbi, curseg->segno));
2657 
2658 	__set_test_and_inuse(sbi, new_segno);
2659 
2660 	mutex_lock(&dirty_i->seglist_lock);
2661 	__remove_dirty_segment(sbi, new_segno, PRE);
2662 	__remove_dirty_segment(sbi, new_segno, DIRTY);
2663 	mutex_unlock(&dirty_i->seglist_lock);
2664 
2665 	reset_curseg(sbi, type, 1);
2666 	curseg->alloc_type = SSR;
2667 	curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
2668 
2669 	sum_page = f2fs_get_sum_page(sbi, new_segno);
2670 	if (IS_ERR(sum_page)) {
2671 		/* GC won't be able to use stale summary pages by cp_error */
2672 		memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
2673 		return;
2674 	}
2675 	sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2676 	memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2677 	f2fs_put_page(sum_page, 1);
2678 }
2679 
2680 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2681 				int alloc_mode, unsigned long long age);
2682 
2683 static void get_atssr_segment(struct f2fs_sb_info *sbi, int type,
2684 					int target_type, int alloc_mode,
2685 					unsigned long long age)
2686 {
2687 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2688 
2689 	curseg->seg_type = target_type;
2690 
2691 	if (get_ssr_segment(sbi, type, alloc_mode, age)) {
2692 		struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
2693 
2694 		curseg->seg_type = se->type;
2695 		change_curseg(sbi, type, true);
2696 	} else {
2697 		/* allocate cold segment by default */
2698 		curseg->seg_type = CURSEG_COLD_DATA;
2699 		new_curseg(sbi, type, true);
2700 	}
2701 	stat_inc_seg_type(sbi, curseg);
2702 }
2703 
2704 static void __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi)
2705 {
2706 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
2707 
2708 	if (!sbi->am.atgc_enabled)
2709 		return;
2710 
2711 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
2712 
2713 	mutex_lock(&curseg->curseg_mutex);
2714 	down_write(&SIT_I(sbi)->sentry_lock);
2715 
2716 	get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC, CURSEG_COLD_DATA, SSR, 0);
2717 
2718 	up_write(&SIT_I(sbi)->sentry_lock);
2719 	mutex_unlock(&curseg->curseg_mutex);
2720 
2721 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
2722 
2723 }
2724 void f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
2725 {
2726 	__f2fs_init_atgc_curseg(sbi);
2727 }
2728 
2729 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2730 {
2731 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2732 
2733 	mutex_lock(&curseg->curseg_mutex);
2734 	if (!curseg->inited)
2735 		goto out;
2736 
2737 	if (get_valid_blocks(sbi, curseg->segno, false)) {
2738 		write_sum_page(sbi, curseg->sum_blk,
2739 				GET_SUM_BLOCK(sbi, curseg->segno));
2740 	} else {
2741 		mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2742 		__set_test_and_free(sbi, curseg->segno, true);
2743 		mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2744 	}
2745 out:
2746 	mutex_unlock(&curseg->curseg_mutex);
2747 }
2748 
2749 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
2750 {
2751 	__f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2752 
2753 	if (sbi->am.atgc_enabled)
2754 		__f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2755 }
2756 
2757 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
2758 {
2759 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2760 
2761 	mutex_lock(&curseg->curseg_mutex);
2762 	if (!curseg->inited)
2763 		goto out;
2764 	if (get_valid_blocks(sbi, curseg->segno, false))
2765 		goto out;
2766 
2767 	mutex_lock(&DIRTY_I(sbi)->seglist_lock);
2768 	__set_test_and_inuse(sbi, curseg->segno);
2769 	mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
2770 out:
2771 	mutex_unlock(&curseg->curseg_mutex);
2772 }
2773 
2774 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
2775 {
2776 	__f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
2777 
2778 	if (sbi->am.atgc_enabled)
2779 		__f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
2780 }
2781 
2782 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
2783 				int alloc_mode, unsigned long long age)
2784 {
2785 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2786 	const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2787 	unsigned segno = NULL_SEGNO;
2788 	unsigned short seg_type = curseg->seg_type;
2789 	int i, cnt;
2790 	bool reversed = false;
2791 
2792 	sanity_check_seg_type(sbi, seg_type);
2793 
2794 	/* f2fs_need_SSR() already forces to do this */
2795 	if (!v_ops->get_victim(sbi, &segno, BG_GC, seg_type, alloc_mode, age)) {
2796 		curseg->next_segno = segno;
2797 		return 1;
2798 	}
2799 
2800 	/* For node segments, let's do SSR more intensively */
2801 	if (IS_NODESEG(seg_type)) {
2802 		if (seg_type >= CURSEG_WARM_NODE) {
2803 			reversed = true;
2804 			i = CURSEG_COLD_NODE;
2805 		} else {
2806 			i = CURSEG_HOT_NODE;
2807 		}
2808 		cnt = NR_CURSEG_NODE_TYPE;
2809 	} else {
2810 		if (seg_type >= CURSEG_WARM_DATA) {
2811 			reversed = true;
2812 			i = CURSEG_COLD_DATA;
2813 		} else {
2814 			i = CURSEG_HOT_DATA;
2815 		}
2816 		cnt = NR_CURSEG_DATA_TYPE;
2817 	}
2818 
2819 	for (; cnt-- > 0; reversed ? i-- : i++) {
2820 		if (i == seg_type)
2821 			continue;
2822 		if (!v_ops->get_victim(sbi, &segno, BG_GC, i, alloc_mode, age)) {
2823 			curseg->next_segno = segno;
2824 			return 1;
2825 		}
2826 	}
2827 
2828 	/* find valid_blocks=0 in dirty list */
2829 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2830 		segno = get_free_segment(sbi);
2831 		if (segno != NULL_SEGNO) {
2832 			curseg->next_segno = segno;
2833 			return 1;
2834 		}
2835 	}
2836 	return 0;
2837 }
2838 
2839 /*
2840  * flush out current segment and replace it with new segment
2841  * This function should be returned with success, otherwise BUG
2842  */
2843 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2844 						int type, bool force)
2845 {
2846 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2847 
2848 	if (force)
2849 		new_curseg(sbi, type, true);
2850 	else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2851 					curseg->seg_type == CURSEG_WARM_NODE)
2852 		new_curseg(sbi, type, false);
2853 	else if (curseg->alloc_type == LFS &&
2854 			is_next_segment_free(sbi, curseg, type) &&
2855 			likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2856 		new_curseg(sbi, type, false);
2857 	else if (f2fs_need_SSR(sbi) &&
2858 			get_ssr_segment(sbi, type, SSR, 0))
2859 		change_curseg(sbi, type, true);
2860 	else
2861 		new_curseg(sbi, type, false);
2862 
2863 	stat_inc_seg_type(sbi, curseg);
2864 }
2865 
2866 void f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2867 					unsigned int start, unsigned int end)
2868 {
2869 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2870 	unsigned int segno;
2871 
2872 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
2873 	mutex_lock(&curseg->curseg_mutex);
2874 	down_write(&SIT_I(sbi)->sentry_lock);
2875 
2876 	segno = CURSEG_I(sbi, type)->segno;
2877 	if (segno < start || segno > end)
2878 		goto unlock;
2879 
2880 	if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
2881 		change_curseg(sbi, type, true);
2882 	else
2883 		new_curseg(sbi, type, true);
2884 
2885 	stat_inc_seg_type(sbi, curseg);
2886 
2887 	locate_dirty_segment(sbi, segno);
2888 unlock:
2889 	up_write(&SIT_I(sbi)->sentry_lock);
2890 
2891 	if (segno != curseg->segno)
2892 		f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2893 			    type, segno, curseg->segno);
2894 
2895 	mutex_unlock(&curseg->curseg_mutex);
2896 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
2897 }
2898 
2899 static void __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
2900 						bool new_sec, bool force)
2901 {
2902 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2903 	unsigned int old_segno;
2904 
2905 	if (!curseg->inited)
2906 		goto alloc;
2907 
2908 	if (force || curseg->next_blkoff ||
2909 		get_valid_blocks(sbi, curseg->segno, new_sec))
2910 		goto alloc;
2911 
2912 	if (!get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
2913 		return;
2914 alloc:
2915 	old_segno = curseg->segno;
2916 	SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
2917 	locate_dirty_segment(sbi, old_segno);
2918 }
2919 
2920 static void __allocate_new_section(struct f2fs_sb_info *sbi,
2921 						int type, bool force)
2922 {
2923 	__allocate_new_segment(sbi, type, true, force);
2924 }
2925 
2926 void f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
2927 {
2928 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
2929 	down_write(&SIT_I(sbi)->sentry_lock);
2930 	__allocate_new_section(sbi, type, force);
2931 	up_write(&SIT_I(sbi)->sentry_lock);
2932 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
2933 }
2934 
2935 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2936 {
2937 	int i;
2938 
2939 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
2940 	down_write(&SIT_I(sbi)->sentry_lock);
2941 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
2942 		__allocate_new_segment(sbi, i, false, false);
2943 	up_write(&SIT_I(sbi)->sentry_lock);
2944 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
2945 }
2946 
2947 static const struct segment_allocation default_salloc_ops = {
2948 	.allocate_segment = allocate_segment_by_default,
2949 };
2950 
2951 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2952 						struct cp_control *cpc)
2953 {
2954 	__u64 trim_start = cpc->trim_start;
2955 	bool has_candidate = false;
2956 
2957 	down_write(&SIT_I(sbi)->sentry_lock);
2958 	for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2959 		if (add_discard_addrs(sbi, cpc, true)) {
2960 			has_candidate = true;
2961 			break;
2962 		}
2963 	}
2964 	up_write(&SIT_I(sbi)->sentry_lock);
2965 
2966 	cpc->trim_start = trim_start;
2967 	return has_candidate;
2968 }
2969 
2970 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2971 					struct discard_policy *dpolicy,
2972 					unsigned int start, unsigned int end)
2973 {
2974 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2975 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2976 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
2977 	struct discard_cmd *dc;
2978 	struct blk_plug plug;
2979 	int issued;
2980 	unsigned int trimmed = 0;
2981 
2982 next:
2983 	issued = 0;
2984 
2985 	mutex_lock(&dcc->cmd_lock);
2986 	if (unlikely(dcc->rbtree_check))
2987 		f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2988 							&dcc->root, false));
2989 
2990 	dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2991 					NULL, start,
2992 					(struct rb_entry **)&prev_dc,
2993 					(struct rb_entry **)&next_dc,
2994 					&insert_p, &insert_parent, true, NULL);
2995 	if (!dc)
2996 		dc = next_dc;
2997 
2998 	blk_start_plug(&plug);
2999 
3000 	while (dc && dc->lstart <= end) {
3001 		struct rb_node *node;
3002 		int err = 0;
3003 
3004 		if (dc->len < dpolicy->granularity)
3005 			goto skip;
3006 
3007 		if (dc->state != D_PREP) {
3008 			list_move_tail(&dc->list, &dcc->fstrim_list);
3009 			goto skip;
3010 		}
3011 
3012 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3013 
3014 		if (issued >= dpolicy->max_requests) {
3015 			start = dc->lstart + dc->len;
3016 
3017 			if (err)
3018 				__remove_discard_cmd(sbi, dc);
3019 
3020 			blk_finish_plug(&plug);
3021 			mutex_unlock(&dcc->cmd_lock);
3022 			trimmed += __wait_all_discard_cmd(sbi, NULL);
3023 			f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
3024 			goto next;
3025 		}
3026 skip:
3027 		node = rb_next(&dc->rb_node);
3028 		if (err)
3029 			__remove_discard_cmd(sbi, dc);
3030 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3031 
3032 		if (fatal_signal_pending(current))
3033 			break;
3034 	}
3035 
3036 	blk_finish_plug(&plug);
3037 	mutex_unlock(&dcc->cmd_lock);
3038 
3039 	return trimmed;
3040 }
3041 
3042 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3043 {
3044 	__u64 start = F2FS_BYTES_TO_BLK(range->start);
3045 	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3046 	unsigned int start_segno, end_segno;
3047 	block_t start_block, end_block;
3048 	struct cp_control cpc;
3049 	struct discard_policy dpolicy;
3050 	unsigned long long trimmed = 0;
3051 	int err = 0;
3052 	bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3053 
3054 	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3055 		return -EINVAL;
3056 
3057 	if (end < MAIN_BLKADDR(sbi))
3058 		goto out;
3059 
3060 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3061 		f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3062 		return -EFSCORRUPTED;
3063 	}
3064 
3065 	/* start/end segment number in main_area */
3066 	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3067 	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3068 						GET_SEGNO(sbi, end);
3069 	if (need_align) {
3070 		start_segno = rounddown(start_segno, sbi->segs_per_sec);
3071 		end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
3072 	}
3073 
3074 	cpc.reason = CP_DISCARD;
3075 	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3076 	cpc.trim_start = start_segno;
3077 	cpc.trim_end = end_segno;
3078 
3079 	if (sbi->discard_blks == 0)
3080 		goto out;
3081 
3082 	f2fs_down_write(&sbi->gc_lock);
3083 	err = f2fs_write_checkpoint(sbi, &cpc);
3084 	f2fs_up_write(&sbi->gc_lock);
3085 	if (err)
3086 		goto out;
3087 
3088 	/*
3089 	 * We filed discard candidates, but actually we don't need to wait for
3090 	 * all of them, since they'll be issued in idle time along with runtime
3091 	 * discard option. User configuration looks like using runtime discard
3092 	 * or periodic fstrim instead of it.
3093 	 */
3094 	if (f2fs_realtime_discard_enable(sbi))
3095 		goto out;
3096 
3097 	start_block = START_BLOCK(sbi, start_segno);
3098 	end_block = START_BLOCK(sbi, end_segno + 1);
3099 
3100 	__init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3101 	trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3102 					start_block, end_block);
3103 
3104 	trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3105 					start_block, end_block);
3106 out:
3107 	if (!err)
3108 		range->len = F2FS_BLK_TO_BYTES(trimmed);
3109 	return err;
3110 }
3111 
3112 static bool __has_curseg_space(struct f2fs_sb_info *sbi,
3113 					struct curseg_info *curseg)
3114 {
3115 	return curseg->next_blkoff < f2fs_usable_blks_in_seg(sbi,
3116 							curseg->segno);
3117 }
3118 
3119 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
3120 {
3121 	switch (hint) {
3122 	case WRITE_LIFE_SHORT:
3123 		return CURSEG_HOT_DATA;
3124 	case WRITE_LIFE_EXTREME:
3125 		return CURSEG_COLD_DATA;
3126 	default:
3127 		return CURSEG_WARM_DATA;
3128 	}
3129 }
3130 
3131 static int __get_segment_type_2(struct f2fs_io_info *fio)
3132 {
3133 	if (fio->type == DATA)
3134 		return CURSEG_HOT_DATA;
3135 	else
3136 		return CURSEG_HOT_NODE;
3137 }
3138 
3139 static int __get_segment_type_4(struct f2fs_io_info *fio)
3140 {
3141 	if (fio->type == DATA) {
3142 		struct inode *inode = fio->page->mapping->host;
3143 
3144 		if (S_ISDIR(inode->i_mode))
3145 			return CURSEG_HOT_DATA;
3146 		else
3147 			return CURSEG_COLD_DATA;
3148 	} else {
3149 		if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3150 			return CURSEG_WARM_NODE;
3151 		else
3152 			return CURSEG_COLD_NODE;
3153 	}
3154 }
3155 
3156 static int __get_segment_type_6(struct f2fs_io_info *fio)
3157 {
3158 	if (fio->type == DATA) {
3159 		struct inode *inode = fio->page->mapping->host;
3160 
3161 		if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
3162 			return CURSEG_COLD_DATA_PINNED;
3163 
3164 		if (page_private_gcing(fio->page)) {
3165 			if (fio->sbi->am.atgc_enabled &&
3166 				(fio->io_type == FS_DATA_IO) &&
3167 				(fio->sbi->gc_mode != GC_URGENT_HIGH))
3168 				return CURSEG_ALL_DATA_ATGC;
3169 			else
3170 				return CURSEG_COLD_DATA;
3171 		}
3172 		if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3173 			return CURSEG_COLD_DATA;
3174 		if (file_is_hot(inode) ||
3175 				is_inode_flag_set(inode, FI_HOT_DATA) ||
3176 				f2fs_is_cow_file(inode))
3177 			return CURSEG_HOT_DATA;
3178 		return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3179 	} else {
3180 		if (IS_DNODE(fio->page))
3181 			return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3182 						CURSEG_HOT_NODE;
3183 		return CURSEG_COLD_NODE;
3184 	}
3185 }
3186 
3187 static int __get_segment_type(struct f2fs_io_info *fio)
3188 {
3189 	int type = 0;
3190 
3191 	switch (F2FS_OPTION(fio->sbi).active_logs) {
3192 	case 2:
3193 		type = __get_segment_type_2(fio);
3194 		break;
3195 	case 4:
3196 		type = __get_segment_type_4(fio);
3197 		break;
3198 	case 6:
3199 		type = __get_segment_type_6(fio);
3200 		break;
3201 	default:
3202 		f2fs_bug_on(fio->sbi, true);
3203 	}
3204 
3205 	if (IS_HOT(type))
3206 		fio->temp = HOT;
3207 	else if (IS_WARM(type))
3208 		fio->temp = WARM;
3209 	else
3210 		fio->temp = COLD;
3211 	return type;
3212 }
3213 
3214 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3215 		block_t old_blkaddr, block_t *new_blkaddr,
3216 		struct f2fs_summary *sum, int type,
3217 		struct f2fs_io_info *fio)
3218 {
3219 	struct sit_info *sit_i = SIT_I(sbi);
3220 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3221 	unsigned long long old_mtime;
3222 	bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3223 	struct seg_entry *se = NULL;
3224 
3225 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3226 
3227 	mutex_lock(&curseg->curseg_mutex);
3228 	down_write(&sit_i->sentry_lock);
3229 
3230 	if (from_gc) {
3231 		f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3232 		se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3233 		sanity_check_seg_type(sbi, se->type);
3234 		f2fs_bug_on(sbi, IS_NODESEG(se->type));
3235 	}
3236 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3237 
3238 	f2fs_bug_on(sbi, curseg->next_blkoff >= sbi->blocks_per_seg);
3239 
3240 	f2fs_wait_discard_bio(sbi, *new_blkaddr);
3241 
3242 	/*
3243 	 * __add_sum_entry should be resided under the curseg_mutex
3244 	 * because, this function updates a summary entry in the
3245 	 * current summary block.
3246 	 */
3247 	__add_sum_entry(sbi, type, sum);
3248 
3249 	__refresh_next_blkoff(sbi, curseg);
3250 
3251 	stat_inc_block_count(sbi, curseg);
3252 
3253 	if (from_gc) {
3254 		old_mtime = get_segment_mtime(sbi, old_blkaddr);
3255 	} else {
3256 		update_segment_mtime(sbi, old_blkaddr, 0);
3257 		old_mtime = 0;
3258 	}
3259 	update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3260 
3261 	/*
3262 	 * SIT information should be updated before segment allocation,
3263 	 * since SSR needs latest valid block information.
3264 	 */
3265 	update_sit_entry(sbi, *new_blkaddr, 1);
3266 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3267 		update_sit_entry(sbi, old_blkaddr, -1);
3268 
3269 	if (!__has_curseg_space(sbi, curseg)) {
3270 		if (from_gc)
3271 			get_atssr_segment(sbi, type, se->type,
3272 						AT_SSR, se->mtime);
3273 		else
3274 			sit_i->s_ops->allocate_segment(sbi, type, false);
3275 	}
3276 	/*
3277 	 * segment dirty status should be updated after segment allocation,
3278 	 * so we just need to update status only one time after previous
3279 	 * segment being closed.
3280 	 */
3281 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3282 	locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3283 
3284 	up_write(&sit_i->sentry_lock);
3285 
3286 	if (page && IS_NODESEG(type)) {
3287 		fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3288 
3289 		f2fs_inode_chksum_set(sbi, page);
3290 	}
3291 
3292 	if (fio) {
3293 		struct f2fs_bio_info *io;
3294 
3295 		if (F2FS_IO_ALIGNED(sbi))
3296 			fio->retry = false;
3297 
3298 		INIT_LIST_HEAD(&fio->list);
3299 		fio->in_list = true;
3300 		io = sbi->write_io[fio->type] + fio->temp;
3301 		spin_lock(&io->io_lock);
3302 		list_add_tail(&fio->list, &io->io_list);
3303 		spin_unlock(&io->io_lock);
3304 	}
3305 
3306 	mutex_unlock(&curseg->curseg_mutex);
3307 
3308 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3309 }
3310 
3311 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino,
3312 					block_t blkaddr, unsigned int blkcnt)
3313 {
3314 	if (!f2fs_is_multi_device(sbi))
3315 		return;
3316 
3317 	while (1) {
3318 		unsigned int devidx = f2fs_target_device_index(sbi, blkaddr);
3319 		unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1;
3320 
3321 		/* update device state for fsync */
3322 		f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO);
3323 
3324 		/* update device state for checkpoint */
3325 		if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3326 			spin_lock(&sbi->dev_lock);
3327 			f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3328 			spin_unlock(&sbi->dev_lock);
3329 		}
3330 
3331 		if (blkcnt <= blks)
3332 			break;
3333 		blkcnt -= blks;
3334 		blkaddr += blks;
3335 	}
3336 }
3337 
3338 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3339 {
3340 	int type = __get_segment_type(fio);
3341 	bool keep_order = (f2fs_lfs_mode(fio->sbi) && type == CURSEG_COLD_DATA);
3342 
3343 	if (keep_order)
3344 		f2fs_down_read(&fio->sbi->io_order_lock);
3345 reallocate:
3346 	f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3347 			&fio->new_blkaddr, sum, type, fio);
3348 	if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO) {
3349 		invalidate_mapping_pages(META_MAPPING(fio->sbi),
3350 					fio->old_blkaddr, fio->old_blkaddr);
3351 		f2fs_invalidate_compress_page(fio->sbi, fio->old_blkaddr);
3352 	}
3353 
3354 	/* writeout dirty page into bdev */
3355 	f2fs_submit_page_write(fio);
3356 	if (fio->retry) {
3357 		fio->old_blkaddr = fio->new_blkaddr;
3358 		goto reallocate;
3359 	}
3360 
3361 	f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1);
3362 
3363 	if (keep_order)
3364 		f2fs_up_read(&fio->sbi->io_order_lock);
3365 }
3366 
3367 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3368 					enum iostat_type io_type)
3369 {
3370 	struct f2fs_io_info fio = {
3371 		.sbi = sbi,
3372 		.type = META,
3373 		.temp = HOT,
3374 		.op = REQ_OP_WRITE,
3375 		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3376 		.old_blkaddr = page->index,
3377 		.new_blkaddr = page->index,
3378 		.page = page,
3379 		.encrypted_page = NULL,
3380 		.in_list = false,
3381 	};
3382 
3383 	if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3384 		fio.op_flags &= ~REQ_META;
3385 
3386 	set_page_writeback(page);
3387 	ClearPageError(page);
3388 	f2fs_submit_page_write(&fio);
3389 
3390 	stat_inc_meta_count(sbi, page->index);
3391 	f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3392 }
3393 
3394 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3395 {
3396 	struct f2fs_summary sum;
3397 
3398 	set_summary(&sum, nid, 0, 0);
3399 	do_write_page(&sum, fio);
3400 
3401 	f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3402 }
3403 
3404 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3405 					struct f2fs_io_info *fio)
3406 {
3407 	struct f2fs_sb_info *sbi = fio->sbi;
3408 	struct f2fs_summary sum;
3409 
3410 	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3411 	set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3412 	do_write_page(&sum, fio);
3413 	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3414 
3415 	f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3416 }
3417 
3418 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3419 {
3420 	int err;
3421 	struct f2fs_sb_info *sbi = fio->sbi;
3422 	unsigned int segno;
3423 
3424 	fio->new_blkaddr = fio->old_blkaddr;
3425 	/* i/o temperature is needed for passing down write hints */
3426 	__get_segment_type(fio);
3427 
3428 	segno = GET_SEGNO(sbi, fio->new_blkaddr);
3429 
3430 	if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3431 		set_sbi_flag(sbi, SBI_NEED_FSCK);
3432 		f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3433 			  __func__, segno);
3434 		err = -EFSCORRUPTED;
3435 		goto drop_bio;
3436 	}
3437 
3438 	if (f2fs_cp_error(sbi)) {
3439 		err = -EIO;
3440 		goto drop_bio;
3441 	}
3442 
3443 	if (fio->post_read)
3444 		invalidate_mapping_pages(META_MAPPING(sbi),
3445 				fio->new_blkaddr, fio->new_blkaddr);
3446 
3447 	stat_inc_inplace_blocks(fio->sbi);
3448 
3449 	if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE)))
3450 		err = f2fs_merge_page_bio(fio);
3451 	else
3452 		err = f2fs_submit_page_bio(fio);
3453 	if (!err) {
3454 		f2fs_update_device_state(fio->sbi, fio->ino,
3455 						fio->new_blkaddr, 1);
3456 		f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3457 	}
3458 
3459 	return err;
3460 drop_bio:
3461 	if (fio->bio && *(fio->bio)) {
3462 		struct bio *bio = *(fio->bio);
3463 
3464 		bio->bi_status = BLK_STS_IOERR;
3465 		bio_endio(bio);
3466 		*(fio->bio) = NULL;
3467 	}
3468 	return err;
3469 }
3470 
3471 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3472 						unsigned int segno)
3473 {
3474 	int i;
3475 
3476 	for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3477 		if (CURSEG_I(sbi, i)->segno == segno)
3478 			break;
3479 	}
3480 	return i;
3481 }
3482 
3483 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3484 				block_t old_blkaddr, block_t new_blkaddr,
3485 				bool recover_curseg, bool recover_newaddr,
3486 				bool from_gc)
3487 {
3488 	struct sit_info *sit_i = SIT_I(sbi);
3489 	struct curseg_info *curseg;
3490 	unsigned int segno, old_cursegno;
3491 	struct seg_entry *se;
3492 	int type;
3493 	unsigned short old_blkoff;
3494 	unsigned char old_alloc_type;
3495 
3496 	segno = GET_SEGNO(sbi, new_blkaddr);
3497 	se = get_seg_entry(sbi, segno);
3498 	type = se->type;
3499 
3500 	f2fs_down_write(&SM_I(sbi)->curseg_lock);
3501 
3502 	if (!recover_curseg) {
3503 		/* for recovery flow */
3504 		if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3505 			if (old_blkaddr == NULL_ADDR)
3506 				type = CURSEG_COLD_DATA;
3507 			else
3508 				type = CURSEG_WARM_DATA;
3509 		}
3510 	} else {
3511 		if (IS_CURSEG(sbi, segno)) {
3512 			/* se->type is volatile as SSR allocation */
3513 			type = __f2fs_get_curseg(sbi, segno);
3514 			f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3515 		} else {
3516 			type = CURSEG_WARM_DATA;
3517 		}
3518 	}
3519 
3520 	f2fs_bug_on(sbi, !IS_DATASEG(type));
3521 	curseg = CURSEG_I(sbi, type);
3522 
3523 	mutex_lock(&curseg->curseg_mutex);
3524 	down_write(&sit_i->sentry_lock);
3525 
3526 	old_cursegno = curseg->segno;
3527 	old_blkoff = curseg->next_blkoff;
3528 	old_alloc_type = curseg->alloc_type;
3529 
3530 	/* change the current segment */
3531 	if (segno != curseg->segno) {
3532 		curseg->next_segno = segno;
3533 		change_curseg(sbi, type, true);
3534 	}
3535 
3536 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3537 	__add_sum_entry(sbi, type, sum);
3538 
3539 	if (!recover_curseg || recover_newaddr) {
3540 		if (!from_gc)
3541 			update_segment_mtime(sbi, new_blkaddr, 0);
3542 		update_sit_entry(sbi, new_blkaddr, 1);
3543 	}
3544 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3545 		invalidate_mapping_pages(META_MAPPING(sbi),
3546 					old_blkaddr, old_blkaddr);
3547 		f2fs_invalidate_compress_page(sbi, old_blkaddr);
3548 		if (!from_gc)
3549 			update_segment_mtime(sbi, old_blkaddr, 0);
3550 		update_sit_entry(sbi, old_blkaddr, -1);
3551 	}
3552 
3553 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3554 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3555 
3556 	locate_dirty_segment(sbi, old_cursegno);
3557 
3558 	if (recover_curseg) {
3559 		if (old_cursegno != curseg->segno) {
3560 			curseg->next_segno = old_cursegno;
3561 			change_curseg(sbi, type, true);
3562 		}
3563 		curseg->next_blkoff = old_blkoff;
3564 		curseg->alloc_type = old_alloc_type;
3565 	}
3566 
3567 	up_write(&sit_i->sentry_lock);
3568 	mutex_unlock(&curseg->curseg_mutex);
3569 	f2fs_up_write(&SM_I(sbi)->curseg_lock);
3570 }
3571 
3572 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3573 				block_t old_addr, block_t new_addr,
3574 				unsigned char version, bool recover_curseg,
3575 				bool recover_newaddr)
3576 {
3577 	struct f2fs_summary sum;
3578 
3579 	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3580 
3581 	f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3582 					recover_curseg, recover_newaddr, false);
3583 
3584 	f2fs_update_data_blkaddr(dn, new_addr);
3585 }
3586 
3587 void f2fs_wait_on_page_writeback(struct page *page,
3588 				enum page_type type, bool ordered, bool locked)
3589 {
3590 	if (PageWriteback(page)) {
3591 		struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3592 
3593 		/* submit cached LFS IO */
3594 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3595 		/* sbumit cached IPU IO */
3596 		f2fs_submit_merged_ipu_write(sbi, NULL, page);
3597 		if (ordered) {
3598 			wait_on_page_writeback(page);
3599 			f2fs_bug_on(sbi, locked && PageWriteback(page));
3600 		} else {
3601 			wait_for_stable_page(page);
3602 		}
3603 	}
3604 }
3605 
3606 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3607 {
3608 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3609 	struct page *cpage;
3610 
3611 	if (!f2fs_post_read_required(inode))
3612 		return;
3613 
3614 	if (!__is_valid_data_blkaddr(blkaddr))
3615 		return;
3616 
3617 	cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3618 	if (cpage) {
3619 		f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3620 		f2fs_put_page(cpage, 1);
3621 	}
3622 }
3623 
3624 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3625 								block_t len)
3626 {
3627 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3628 	block_t i;
3629 
3630 	if (!f2fs_post_read_required(inode))
3631 		return;
3632 
3633 	for (i = 0; i < len; i++)
3634 		f2fs_wait_on_block_writeback(inode, blkaddr + i);
3635 
3636 	invalidate_mapping_pages(META_MAPPING(sbi), blkaddr, blkaddr + len - 1);
3637 }
3638 
3639 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3640 {
3641 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3642 	struct curseg_info *seg_i;
3643 	unsigned char *kaddr;
3644 	struct page *page;
3645 	block_t start;
3646 	int i, j, offset;
3647 
3648 	start = start_sum_block(sbi);
3649 
3650 	page = f2fs_get_meta_page(sbi, start++);
3651 	if (IS_ERR(page))
3652 		return PTR_ERR(page);
3653 	kaddr = (unsigned char *)page_address(page);
3654 
3655 	/* Step 1: restore nat cache */
3656 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3657 	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3658 
3659 	/* Step 2: restore sit cache */
3660 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3661 	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3662 	offset = 2 * SUM_JOURNAL_SIZE;
3663 
3664 	/* Step 3: restore summary entries */
3665 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3666 		unsigned short blk_off;
3667 		unsigned int segno;
3668 
3669 		seg_i = CURSEG_I(sbi, i);
3670 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3671 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3672 		seg_i->next_segno = segno;
3673 		reset_curseg(sbi, i, 0);
3674 		seg_i->alloc_type = ckpt->alloc_type[i];
3675 		seg_i->next_blkoff = blk_off;
3676 
3677 		if (seg_i->alloc_type == SSR)
3678 			blk_off = sbi->blocks_per_seg;
3679 
3680 		for (j = 0; j < blk_off; j++) {
3681 			struct f2fs_summary *s;
3682 
3683 			s = (struct f2fs_summary *)(kaddr + offset);
3684 			seg_i->sum_blk->entries[j] = *s;
3685 			offset += SUMMARY_SIZE;
3686 			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3687 						SUM_FOOTER_SIZE)
3688 				continue;
3689 
3690 			f2fs_put_page(page, 1);
3691 			page = NULL;
3692 
3693 			page = f2fs_get_meta_page(sbi, start++);
3694 			if (IS_ERR(page))
3695 				return PTR_ERR(page);
3696 			kaddr = (unsigned char *)page_address(page);
3697 			offset = 0;
3698 		}
3699 	}
3700 	f2fs_put_page(page, 1);
3701 	return 0;
3702 }
3703 
3704 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3705 {
3706 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3707 	struct f2fs_summary_block *sum;
3708 	struct curseg_info *curseg;
3709 	struct page *new;
3710 	unsigned short blk_off;
3711 	unsigned int segno = 0;
3712 	block_t blk_addr = 0;
3713 	int err = 0;
3714 
3715 	/* get segment number and block addr */
3716 	if (IS_DATASEG(type)) {
3717 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3718 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3719 							CURSEG_HOT_DATA]);
3720 		if (__exist_node_summaries(sbi))
3721 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
3722 		else
3723 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3724 	} else {
3725 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
3726 							CURSEG_HOT_NODE]);
3727 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3728 							CURSEG_HOT_NODE]);
3729 		if (__exist_node_summaries(sbi))
3730 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3731 							type - CURSEG_HOT_NODE);
3732 		else
3733 			blk_addr = GET_SUM_BLOCK(sbi, segno);
3734 	}
3735 
3736 	new = f2fs_get_meta_page(sbi, blk_addr);
3737 	if (IS_ERR(new))
3738 		return PTR_ERR(new);
3739 	sum = (struct f2fs_summary_block *)page_address(new);
3740 
3741 	if (IS_NODESEG(type)) {
3742 		if (__exist_node_summaries(sbi)) {
3743 			struct f2fs_summary *ns = &sum->entries[0];
3744 			int i;
3745 
3746 			for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3747 				ns->version = 0;
3748 				ns->ofs_in_node = 0;
3749 			}
3750 		} else {
3751 			err = f2fs_restore_node_summary(sbi, segno, sum);
3752 			if (err)
3753 				goto out;
3754 		}
3755 	}
3756 
3757 	/* set uncompleted segment to curseg */
3758 	curseg = CURSEG_I(sbi, type);
3759 	mutex_lock(&curseg->curseg_mutex);
3760 
3761 	/* update journal info */
3762 	down_write(&curseg->journal_rwsem);
3763 	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3764 	up_write(&curseg->journal_rwsem);
3765 
3766 	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3767 	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3768 	curseg->next_segno = segno;
3769 	reset_curseg(sbi, type, 0);
3770 	curseg->alloc_type = ckpt->alloc_type[type];
3771 	curseg->next_blkoff = blk_off;
3772 	mutex_unlock(&curseg->curseg_mutex);
3773 out:
3774 	f2fs_put_page(new, 1);
3775 	return err;
3776 }
3777 
3778 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3779 {
3780 	struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3781 	struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3782 	int type = CURSEG_HOT_DATA;
3783 	int err;
3784 
3785 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3786 		int npages = f2fs_npages_for_summary_flush(sbi, true);
3787 
3788 		if (npages >= 2)
3789 			f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3790 							META_CP, true);
3791 
3792 		/* restore for compacted data summary */
3793 		err = read_compacted_summaries(sbi);
3794 		if (err)
3795 			return err;
3796 		type = CURSEG_HOT_NODE;
3797 	}
3798 
3799 	if (__exist_node_summaries(sbi))
3800 		f2fs_ra_meta_pages(sbi,
3801 				sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
3802 				NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
3803 
3804 	for (; type <= CURSEG_COLD_NODE; type++) {
3805 		err = read_normal_summaries(sbi, type);
3806 		if (err)
3807 			return err;
3808 	}
3809 
3810 	/* sanity check for summary blocks */
3811 	if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3812 			sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3813 		f2fs_err(sbi, "invalid journal entries nats %u sits %u",
3814 			 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3815 		return -EINVAL;
3816 	}
3817 
3818 	return 0;
3819 }
3820 
3821 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3822 {
3823 	struct page *page;
3824 	unsigned char *kaddr;
3825 	struct f2fs_summary *summary;
3826 	struct curseg_info *seg_i;
3827 	int written_size = 0;
3828 	int i, j;
3829 
3830 	page = f2fs_grab_meta_page(sbi, blkaddr++);
3831 	kaddr = (unsigned char *)page_address(page);
3832 	memset(kaddr, 0, PAGE_SIZE);
3833 
3834 	/* Step 1: write nat cache */
3835 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3836 	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3837 	written_size += SUM_JOURNAL_SIZE;
3838 
3839 	/* Step 2: write sit cache */
3840 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3841 	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3842 	written_size += SUM_JOURNAL_SIZE;
3843 
3844 	/* Step 3: write summary entries */
3845 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3846 		unsigned short blkoff;
3847 
3848 		seg_i = CURSEG_I(sbi, i);
3849 		if (sbi->ckpt->alloc_type[i] == SSR)
3850 			blkoff = sbi->blocks_per_seg;
3851 		else
3852 			blkoff = curseg_blkoff(sbi, i);
3853 
3854 		for (j = 0; j < blkoff; j++) {
3855 			if (!page) {
3856 				page = f2fs_grab_meta_page(sbi, blkaddr++);
3857 				kaddr = (unsigned char *)page_address(page);
3858 				memset(kaddr, 0, PAGE_SIZE);
3859 				written_size = 0;
3860 			}
3861 			summary = (struct f2fs_summary *)(kaddr + written_size);
3862 			*summary = seg_i->sum_blk->entries[j];
3863 			written_size += SUMMARY_SIZE;
3864 
3865 			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3866 							SUM_FOOTER_SIZE)
3867 				continue;
3868 
3869 			set_page_dirty(page);
3870 			f2fs_put_page(page, 1);
3871 			page = NULL;
3872 		}
3873 	}
3874 	if (page) {
3875 		set_page_dirty(page);
3876 		f2fs_put_page(page, 1);
3877 	}
3878 }
3879 
3880 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3881 					block_t blkaddr, int type)
3882 {
3883 	int i, end;
3884 
3885 	if (IS_DATASEG(type))
3886 		end = type + NR_CURSEG_DATA_TYPE;
3887 	else
3888 		end = type + NR_CURSEG_NODE_TYPE;
3889 
3890 	for (i = type; i < end; i++)
3891 		write_current_sum_page(sbi, i, blkaddr + (i - type));
3892 }
3893 
3894 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3895 {
3896 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3897 		write_compacted_summaries(sbi, start_blk);
3898 	else
3899 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3900 }
3901 
3902 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3903 {
3904 	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3905 }
3906 
3907 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3908 					unsigned int val, int alloc)
3909 {
3910 	int i;
3911 
3912 	if (type == NAT_JOURNAL) {
3913 		for (i = 0; i < nats_in_cursum(journal); i++) {
3914 			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3915 				return i;
3916 		}
3917 		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3918 			return update_nats_in_cursum(journal, 1);
3919 	} else if (type == SIT_JOURNAL) {
3920 		for (i = 0; i < sits_in_cursum(journal); i++)
3921 			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3922 				return i;
3923 		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3924 			return update_sits_in_cursum(journal, 1);
3925 	}
3926 	return -1;
3927 }
3928 
3929 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3930 					unsigned int segno)
3931 {
3932 	return f2fs_get_meta_page(sbi, current_sit_addr(sbi, segno));
3933 }
3934 
3935 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3936 					unsigned int start)
3937 {
3938 	struct sit_info *sit_i = SIT_I(sbi);
3939 	struct page *page;
3940 	pgoff_t src_off, dst_off;
3941 
3942 	src_off = current_sit_addr(sbi, start);
3943 	dst_off = next_sit_addr(sbi, src_off);
3944 
3945 	page = f2fs_grab_meta_page(sbi, dst_off);
3946 	seg_info_to_sit_page(sbi, page, start);
3947 
3948 	set_page_dirty(page);
3949 	set_to_next_sit(sit_i, start);
3950 
3951 	return page;
3952 }
3953 
3954 static struct sit_entry_set *grab_sit_entry_set(void)
3955 {
3956 	struct sit_entry_set *ses =
3957 			f2fs_kmem_cache_alloc(sit_entry_set_slab,
3958 						GFP_NOFS, true, NULL);
3959 
3960 	ses->entry_cnt = 0;
3961 	INIT_LIST_HEAD(&ses->set_list);
3962 	return ses;
3963 }
3964 
3965 static void release_sit_entry_set(struct sit_entry_set *ses)
3966 {
3967 	list_del(&ses->set_list);
3968 	kmem_cache_free(sit_entry_set_slab, ses);
3969 }
3970 
3971 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3972 						struct list_head *head)
3973 {
3974 	struct sit_entry_set *next = ses;
3975 
3976 	if (list_is_last(&ses->set_list, head))
3977 		return;
3978 
3979 	list_for_each_entry_continue(next, head, set_list)
3980 		if (ses->entry_cnt <= next->entry_cnt) {
3981 			list_move_tail(&ses->set_list, &next->set_list);
3982 			return;
3983 		}
3984 
3985 	list_move_tail(&ses->set_list, head);
3986 }
3987 
3988 static void add_sit_entry(unsigned int segno, struct list_head *head)
3989 {
3990 	struct sit_entry_set *ses;
3991 	unsigned int start_segno = START_SEGNO(segno);
3992 
3993 	list_for_each_entry(ses, head, set_list) {
3994 		if (ses->start_segno == start_segno) {
3995 			ses->entry_cnt++;
3996 			adjust_sit_entry_set(ses, head);
3997 			return;
3998 		}
3999 	}
4000 
4001 	ses = grab_sit_entry_set();
4002 
4003 	ses->start_segno = start_segno;
4004 	ses->entry_cnt++;
4005 	list_add(&ses->set_list, head);
4006 }
4007 
4008 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4009 {
4010 	struct f2fs_sm_info *sm_info = SM_I(sbi);
4011 	struct list_head *set_list = &sm_info->sit_entry_set;
4012 	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4013 	unsigned int segno;
4014 
4015 	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4016 		add_sit_entry(segno, set_list);
4017 }
4018 
4019 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4020 {
4021 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4022 	struct f2fs_journal *journal = curseg->journal;
4023 	int i;
4024 
4025 	down_write(&curseg->journal_rwsem);
4026 	for (i = 0; i < sits_in_cursum(journal); i++) {
4027 		unsigned int segno;
4028 		bool dirtied;
4029 
4030 		segno = le32_to_cpu(segno_in_journal(journal, i));
4031 		dirtied = __mark_sit_entry_dirty(sbi, segno);
4032 
4033 		if (!dirtied)
4034 			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4035 	}
4036 	update_sits_in_cursum(journal, -i);
4037 	up_write(&curseg->journal_rwsem);
4038 }
4039 
4040 /*
4041  * CP calls this function, which flushes SIT entries including sit_journal,
4042  * and moves prefree segs to free segs.
4043  */
4044 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4045 {
4046 	struct sit_info *sit_i = SIT_I(sbi);
4047 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4048 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4049 	struct f2fs_journal *journal = curseg->journal;
4050 	struct sit_entry_set *ses, *tmp;
4051 	struct list_head *head = &SM_I(sbi)->sit_entry_set;
4052 	bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4053 	struct seg_entry *se;
4054 
4055 	down_write(&sit_i->sentry_lock);
4056 
4057 	if (!sit_i->dirty_sentries)
4058 		goto out;
4059 
4060 	/*
4061 	 * add and account sit entries of dirty bitmap in sit entry
4062 	 * set temporarily
4063 	 */
4064 	add_sits_in_set(sbi);
4065 
4066 	/*
4067 	 * if there are no enough space in journal to store dirty sit
4068 	 * entries, remove all entries from journal and add and account
4069 	 * them in sit entry set.
4070 	 */
4071 	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4072 								!to_journal)
4073 		remove_sits_in_journal(sbi);
4074 
4075 	/*
4076 	 * there are two steps to flush sit entries:
4077 	 * #1, flush sit entries to journal in current cold data summary block.
4078 	 * #2, flush sit entries to sit page.
4079 	 */
4080 	list_for_each_entry_safe(ses, tmp, head, set_list) {
4081 		struct page *page = NULL;
4082 		struct f2fs_sit_block *raw_sit = NULL;
4083 		unsigned int start_segno = ses->start_segno;
4084 		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4085 						(unsigned long)MAIN_SEGS(sbi));
4086 		unsigned int segno = start_segno;
4087 
4088 		if (to_journal &&
4089 			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4090 			to_journal = false;
4091 
4092 		if (to_journal) {
4093 			down_write(&curseg->journal_rwsem);
4094 		} else {
4095 			page = get_next_sit_page(sbi, start_segno);
4096 			raw_sit = page_address(page);
4097 		}
4098 
4099 		/* flush dirty sit entries in region of current sit set */
4100 		for_each_set_bit_from(segno, bitmap, end) {
4101 			int offset, sit_offset;
4102 
4103 			se = get_seg_entry(sbi, segno);
4104 #ifdef CONFIG_F2FS_CHECK_FS
4105 			if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4106 						SIT_VBLOCK_MAP_SIZE))
4107 				f2fs_bug_on(sbi, 1);
4108 #endif
4109 
4110 			/* add discard candidates */
4111 			if (!(cpc->reason & CP_DISCARD)) {
4112 				cpc->trim_start = segno;
4113 				add_discard_addrs(sbi, cpc, false);
4114 			}
4115 
4116 			if (to_journal) {
4117 				offset = f2fs_lookup_journal_in_cursum(journal,
4118 							SIT_JOURNAL, segno, 1);
4119 				f2fs_bug_on(sbi, offset < 0);
4120 				segno_in_journal(journal, offset) =
4121 							cpu_to_le32(segno);
4122 				seg_info_to_raw_sit(se,
4123 					&sit_in_journal(journal, offset));
4124 				check_block_count(sbi, segno,
4125 					&sit_in_journal(journal, offset));
4126 			} else {
4127 				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4128 				seg_info_to_raw_sit(se,
4129 						&raw_sit->entries[sit_offset]);
4130 				check_block_count(sbi, segno,
4131 						&raw_sit->entries[sit_offset]);
4132 			}
4133 
4134 			__clear_bit(segno, bitmap);
4135 			sit_i->dirty_sentries--;
4136 			ses->entry_cnt--;
4137 		}
4138 
4139 		if (to_journal)
4140 			up_write(&curseg->journal_rwsem);
4141 		else
4142 			f2fs_put_page(page, 1);
4143 
4144 		f2fs_bug_on(sbi, ses->entry_cnt);
4145 		release_sit_entry_set(ses);
4146 	}
4147 
4148 	f2fs_bug_on(sbi, !list_empty(head));
4149 	f2fs_bug_on(sbi, sit_i->dirty_sentries);
4150 out:
4151 	if (cpc->reason & CP_DISCARD) {
4152 		__u64 trim_start = cpc->trim_start;
4153 
4154 		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4155 			add_discard_addrs(sbi, cpc, false);
4156 
4157 		cpc->trim_start = trim_start;
4158 	}
4159 	up_write(&sit_i->sentry_lock);
4160 
4161 	set_prefree_as_free_segments(sbi);
4162 }
4163 
4164 static int build_sit_info(struct f2fs_sb_info *sbi)
4165 {
4166 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4167 	struct sit_info *sit_i;
4168 	unsigned int sit_segs, start;
4169 	char *src_bitmap, *bitmap;
4170 	unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4171 	unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0;
4172 
4173 	/* allocate memory for SIT information */
4174 	sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4175 	if (!sit_i)
4176 		return -ENOMEM;
4177 
4178 	SM_I(sbi)->sit_info = sit_i;
4179 
4180 	sit_i->sentries =
4181 		f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4182 					      MAIN_SEGS(sbi)),
4183 			      GFP_KERNEL);
4184 	if (!sit_i->sentries)
4185 		return -ENOMEM;
4186 
4187 	main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4188 	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4189 								GFP_KERNEL);
4190 	if (!sit_i->dirty_sentries_bitmap)
4191 		return -ENOMEM;
4192 
4193 #ifdef CONFIG_F2FS_CHECK_FS
4194 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map);
4195 #else
4196 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map);
4197 #endif
4198 	sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4199 	if (!sit_i->bitmap)
4200 		return -ENOMEM;
4201 
4202 	bitmap = sit_i->bitmap;
4203 
4204 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4205 		sit_i->sentries[start].cur_valid_map = bitmap;
4206 		bitmap += SIT_VBLOCK_MAP_SIZE;
4207 
4208 		sit_i->sentries[start].ckpt_valid_map = bitmap;
4209 		bitmap += SIT_VBLOCK_MAP_SIZE;
4210 
4211 #ifdef CONFIG_F2FS_CHECK_FS
4212 		sit_i->sentries[start].cur_valid_map_mir = bitmap;
4213 		bitmap += SIT_VBLOCK_MAP_SIZE;
4214 #endif
4215 
4216 		if (discard_map) {
4217 			sit_i->sentries[start].discard_map = bitmap;
4218 			bitmap += SIT_VBLOCK_MAP_SIZE;
4219 		}
4220 	}
4221 
4222 	sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4223 	if (!sit_i->tmp_map)
4224 		return -ENOMEM;
4225 
4226 	if (__is_large_section(sbi)) {
4227 		sit_i->sec_entries =
4228 			f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4229 						      MAIN_SECS(sbi)),
4230 				      GFP_KERNEL);
4231 		if (!sit_i->sec_entries)
4232 			return -ENOMEM;
4233 	}
4234 
4235 	/* get information related with SIT */
4236 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4237 
4238 	/* setup SIT bitmap from ckeckpoint pack */
4239 	sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4240 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4241 
4242 	sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4243 	if (!sit_i->sit_bitmap)
4244 		return -ENOMEM;
4245 
4246 #ifdef CONFIG_F2FS_CHECK_FS
4247 	sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4248 					sit_bitmap_size, GFP_KERNEL);
4249 	if (!sit_i->sit_bitmap_mir)
4250 		return -ENOMEM;
4251 
4252 	sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4253 					main_bitmap_size, GFP_KERNEL);
4254 	if (!sit_i->invalid_segmap)
4255 		return -ENOMEM;
4256 #endif
4257 
4258 	/* init SIT information */
4259 	sit_i->s_ops = &default_salloc_ops;
4260 
4261 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4262 	sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4263 	sit_i->written_valid_blocks = 0;
4264 	sit_i->bitmap_size = sit_bitmap_size;
4265 	sit_i->dirty_sentries = 0;
4266 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4267 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4268 	sit_i->mounted_time = ktime_get_boottime_seconds();
4269 	init_rwsem(&sit_i->sentry_lock);
4270 	return 0;
4271 }
4272 
4273 static int build_free_segmap(struct f2fs_sb_info *sbi)
4274 {
4275 	struct free_segmap_info *free_i;
4276 	unsigned int bitmap_size, sec_bitmap_size;
4277 
4278 	/* allocate memory for free segmap information */
4279 	free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4280 	if (!free_i)
4281 		return -ENOMEM;
4282 
4283 	SM_I(sbi)->free_info = free_i;
4284 
4285 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4286 	free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4287 	if (!free_i->free_segmap)
4288 		return -ENOMEM;
4289 
4290 	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4291 	free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4292 	if (!free_i->free_secmap)
4293 		return -ENOMEM;
4294 
4295 	/* set all segments as dirty temporarily */
4296 	memset(free_i->free_segmap, 0xff, bitmap_size);
4297 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4298 
4299 	/* init free segmap information */
4300 	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4301 	free_i->free_segments = 0;
4302 	free_i->free_sections = 0;
4303 	spin_lock_init(&free_i->segmap_lock);
4304 	return 0;
4305 }
4306 
4307 static int build_curseg(struct f2fs_sb_info *sbi)
4308 {
4309 	struct curseg_info *array;
4310 	int i;
4311 
4312 	array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4313 					sizeof(*array)), GFP_KERNEL);
4314 	if (!array)
4315 		return -ENOMEM;
4316 
4317 	SM_I(sbi)->curseg_array = array;
4318 
4319 	for (i = 0; i < NO_CHECK_TYPE; i++) {
4320 		mutex_init(&array[i].curseg_mutex);
4321 		array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4322 		if (!array[i].sum_blk)
4323 			return -ENOMEM;
4324 		init_rwsem(&array[i].journal_rwsem);
4325 		array[i].journal = f2fs_kzalloc(sbi,
4326 				sizeof(struct f2fs_journal), GFP_KERNEL);
4327 		if (!array[i].journal)
4328 			return -ENOMEM;
4329 		if (i < NR_PERSISTENT_LOG)
4330 			array[i].seg_type = CURSEG_HOT_DATA + i;
4331 		else if (i == CURSEG_COLD_DATA_PINNED)
4332 			array[i].seg_type = CURSEG_COLD_DATA;
4333 		else if (i == CURSEG_ALL_DATA_ATGC)
4334 			array[i].seg_type = CURSEG_COLD_DATA;
4335 		array[i].segno = NULL_SEGNO;
4336 		array[i].next_blkoff = 0;
4337 		array[i].inited = false;
4338 	}
4339 	return restore_curseg_summaries(sbi);
4340 }
4341 
4342 static int build_sit_entries(struct f2fs_sb_info *sbi)
4343 {
4344 	struct sit_info *sit_i = SIT_I(sbi);
4345 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4346 	struct f2fs_journal *journal = curseg->journal;
4347 	struct seg_entry *se;
4348 	struct f2fs_sit_entry sit;
4349 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
4350 	unsigned int i, start, end;
4351 	unsigned int readed, start_blk = 0;
4352 	int err = 0;
4353 	block_t sit_valid_blocks[2] = {0, 0};
4354 
4355 	do {
4356 		readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4357 							META_SIT, true);
4358 
4359 		start = start_blk * sit_i->sents_per_block;
4360 		end = (start_blk + readed) * sit_i->sents_per_block;
4361 
4362 		for (; start < end && start < MAIN_SEGS(sbi); start++) {
4363 			struct f2fs_sit_block *sit_blk;
4364 			struct page *page;
4365 
4366 			se = &sit_i->sentries[start];
4367 			page = get_current_sit_page(sbi, start);
4368 			if (IS_ERR(page))
4369 				return PTR_ERR(page);
4370 			sit_blk = (struct f2fs_sit_block *)page_address(page);
4371 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4372 			f2fs_put_page(page, 1);
4373 
4374 			err = check_block_count(sbi, start, &sit);
4375 			if (err)
4376 				return err;
4377 			seg_info_from_raw_sit(se, &sit);
4378 
4379 			if (se->type >= NR_PERSISTENT_LOG) {
4380 				f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4381 							se->type, start);
4382 				return -EFSCORRUPTED;
4383 			}
4384 
4385 			sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4386 
4387 			if (f2fs_block_unit_discard(sbi)) {
4388 				/* build discard map only one time */
4389 				if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4390 					memset(se->discard_map, 0xff,
4391 						SIT_VBLOCK_MAP_SIZE);
4392 				} else {
4393 					memcpy(se->discard_map,
4394 						se->cur_valid_map,
4395 						SIT_VBLOCK_MAP_SIZE);
4396 					sbi->discard_blks +=
4397 						sbi->blocks_per_seg -
4398 						se->valid_blocks;
4399 				}
4400 			}
4401 
4402 			if (__is_large_section(sbi))
4403 				get_sec_entry(sbi, start)->valid_blocks +=
4404 							se->valid_blocks;
4405 		}
4406 		start_blk += readed;
4407 	} while (start_blk < sit_blk_cnt);
4408 
4409 	down_read(&curseg->journal_rwsem);
4410 	for (i = 0; i < sits_in_cursum(journal); i++) {
4411 		unsigned int old_valid_blocks;
4412 
4413 		start = le32_to_cpu(segno_in_journal(journal, i));
4414 		if (start >= MAIN_SEGS(sbi)) {
4415 			f2fs_err(sbi, "Wrong journal entry on segno %u",
4416 				 start);
4417 			err = -EFSCORRUPTED;
4418 			break;
4419 		}
4420 
4421 		se = &sit_i->sentries[start];
4422 		sit = sit_in_journal(journal, i);
4423 
4424 		old_valid_blocks = se->valid_blocks;
4425 
4426 		sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks;
4427 
4428 		err = check_block_count(sbi, start, &sit);
4429 		if (err)
4430 			break;
4431 		seg_info_from_raw_sit(se, &sit);
4432 
4433 		if (se->type >= NR_PERSISTENT_LOG) {
4434 			f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4435 							se->type, start);
4436 			err = -EFSCORRUPTED;
4437 			break;
4438 		}
4439 
4440 		sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4441 
4442 		if (f2fs_block_unit_discard(sbi)) {
4443 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4444 				memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4445 			} else {
4446 				memcpy(se->discard_map, se->cur_valid_map,
4447 							SIT_VBLOCK_MAP_SIZE);
4448 				sbi->discard_blks += old_valid_blocks;
4449 				sbi->discard_blks -= se->valid_blocks;
4450 			}
4451 		}
4452 
4453 		if (__is_large_section(sbi)) {
4454 			get_sec_entry(sbi, start)->valid_blocks +=
4455 							se->valid_blocks;
4456 			get_sec_entry(sbi, start)->valid_blocks -=
4457 							old_valid_blocks;
4458 		}
4459 	}
4460 	up_read(&curseg->journal_rwsem);
4461 
4462 	if (err)
4463 		return err;
4464 
4465 	if (sit_valid_blocks[NODE] != valid_node_count(sbi)) {
4466 		f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4467 			 sit_valid_blocks[NODE], valid_node_count(sbi));
4468 		return -EFSCORRUPTED;
4469 	}
4470 
4471 	if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] >
4472 				valid_user_blocks(sbi)) {
4473 		f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u",
4474 			 sit_valid_blocks[DATA], sit_valid_blocks[NODE],
4475 			 valid_user_blocks(sbi));
4476 		return -EFSCORRUPTED;
4477 	}
4478 
4479 	return 0;
4480 }
4481 
4482 static void init_free_segmap(struct f2fs_sb_info *sbi)
4483 {
4484 	unsigned int start;
4485 	int type;
4486 	struct seg_entry *sentry;
4487 
4488 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4489 		if (f2fs_usable_blks_in_seg(sbi, start) == 0)
4490 			continue;
4491 		sentry = get_seg_entry(sbi, start);
4492 		if (!sentry->valid_blocks)
4493 			__set_free(sbi, start);
4494 		else
4495 			SIT_I(sbi)->written_valid_blocks +=
4496 						sentry->valid_blocks;
4497 	}
4498 
4499 	/* set use the current segments */
4500 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4501 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4502 
4503 		__set_test_and_inuse(sbi, curseg_t->segno);
4504 	}
4505 }
4506 
4507 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4508 {
4509 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4510 	struct free_segmap_info *free_i = FREE_I(sbi);
4511 	unsigned int segno = 0, offset = 0, secno;
4512 	block_t valid_blocks, usable_blks_in_seg;
4513 
4514 	while (1) {
4515 		/* find dirty segment based on free segmap */
4516 		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4517 		if (segno >= MAIN_SEGS(sbi))
4518 			break;
4519 		offset = segno + 1;
4520 		valid_blocks = get_valid_blocks(sbi, segno, false);
4521 		usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
4522 		if (valid_blocks == usable_blks_in_seg || !valid_blocks)
4523 			continue;
4524 		if (valid_blocks > usable_blks_in_seg) {
4525 			f2fs_bug_on(sbi, 1);
4526 			continue;
4527 		}
4528 		mutex_lock(&dirty_i->seglist_lock);
4529 		__locate_dirty_segment(sbi, segno, DIRTY);
4530 		mutex_unlock(&dirty_i->seglist_lock);
4531 	}
4532 
4533 	if (!__is_large_section(sbi))
4534 		return;
4535 
4536 	mutex_lock(&dirty_i->seglist_lock);
4537 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4538 		valid_blocks = get_valid_blocks(sbi, segno, true);
4539 		secno = GET_SEC_FROM_SEG(sbi, segno);
4540 
4541 		if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi))
4542 			continue;
4543 		if (IS_CURSEC(sbi, secno))
4544 			continue;
4545 		set_bit(secno, dirty_i->dirty_secmap);
4546 	}
4547 	mutex_unlock(&dirty_i->seglist_lock);
4548 }
4549 
4550 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4551 {
4552 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4553 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4554 
4555 	dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4556 	if (!dirty_i->victim_secmap)
4557 		return -ENOMEM;
4558 
4559 	dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4560 	if (!dirty_i->pinned_secmap)
4561 		return -ENOMEM;
4562 
4563 	dirty_i->pinned_secmap_cnt = 0;
4564 	dirty_i->enable_pin_section = true;
4565 	return 0;
4566 }
4567 
4568 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4569 {
4570 	struct dirty_seglist_info *dirty_i;
4571 	unsigned int bitmap_size, i;
4572 
4573 	/* allocate memory for dirty segments list information */
4574 	dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4575 								GFP_KERNEL);
4576 	if (!dirty_i)
4577 		return -ENOMEM;
4578 
4579 	SM_I(sbi)->dirty_info = dirty_i;
4580 	mutex_init(&dirty_i->seglist_lock);
4581 
4582 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4583 
4584 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
4585 		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4586 								GFP_KERNEL);
4587 		if (!dirty_i->dirty_segmap[i])
4588 			return -ENOMEM;
4589 	}
4590 
4591 	if (__is_large_section(sbi)) {
4592 		bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4593 		dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
4594 						bitmap_size, GFP_KERNEL);
4595 		if (!dirty_i->dirty_secmap)
4596 			return -ENOMEM;
4597 	}
4598 
4599 	init_dirty_segmap(sbi);
4600 	return init_victim_secmap(sbi);
4601 }
4602 
4603 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4604 {
4605 	int i;
4606 
4607 	/*
4608 	 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4609 	 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4610 	 */
4611 	for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4612 		struct curseg_info *curseg = CURSEG_I(sbi, i);
4613 		struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4614 		unsigned int blkofs = curseg->next_blkoff;
4615 
4616 		if (f2fs_sb_has_readonly(sbi) &&
4617 			i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE)
4618 			continue;
4619 
4620 		sanity_check_seg_type(sbi, curseg->seg_type);
4621 
4622 		if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) {
4623 			f2fs_err(sbi,
4624 				 "Current segment has invalid alloc_type:%d",
4625 				 curseg->alloc_type);
4626 			return -EFSCORRUPTED;
4627 		}
4628 
4629 		if (f2fs_test_bit(blkofs, se->cur_valid_map))
4630 			goto out;
4631 
4632 		if (curseg->alloc_type == SSR)
4633 			continue;
4634 
4635 		for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4636 			if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4637 				continue;
4638 out:
4639 			f2fs_err(sbi,
4640 				 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4641 				 i, curseg->segno, curseg->alloc_type,
4642 				 curseg->next_blkoff, blkofs);
4643 			return -EFSCORRUPTED;
4644 		}
4645 	}
4646 	return 0;
4647 }
4648 
4649 #ifdef CONFIG_BLK_DEV_ZONED
4650 
4651 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4652 				    struct f2fs_dev_info *fdev,
4653 				    struct blk_zone *zone)
4654 {
4655 	unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4656 	block_t zone_block, wp_block, last_valid_block;
4657 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4658 	int i, s, b, ret;
4659 	struct seg_entry *se;
4660 
4661 	if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4662 		return 0;
4663 
4664 	wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4665 	wp_segno = GET_SEGNO(sbi, wp_block);
4666 	wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4667 	zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4668 	zone_segno = GET_SEGNO(sbi, zone_block);
4669 	zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4670 
4671 	if (zone_segno >= MAIN_SEGS(sbi))
4672 		return 0;
4673 
4674 	/*
4675 	 * Skip check of zones cursegs point to, since
4676 	 * fix_curseg_write_pointer() checks them.
4677 	 */
4678 	for (i = 0; i < NO_CHECK_TYPE; i++)
4679 		if (zone_secno == GET_SEC_FROM_SEG(sbi,
4680 						   CURSEG_I(sbi, i)->segno))
4681 			return 0;
4682 
4683 	/*
4684 	 * Get last valid block of the zone.
4685 	 */
4686 	last_valid_block = zone_block - 1;
4687 	for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4688 		segno = zone_segno + s;
4689 		se = get_seg_entry(sbi, segno);
4690 		for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4691 			if (f2fs_test_bit(b, se->cur_valid_map)) {
4692 				last_valid_block = START_BLOCK(sbi, segno) + b;
4693 				break;
4694 			}
4695 		if (last_valid_block >= zone_block)
4696 			break;
4697 	}
4698 
4699 	/*
4700 	 * If last valid block is beyond the write pointer, report the
4701 	 * inconsistency. This inconsistency does not cause write error
4702 	 * because the zone will not be selected for write operation until
4703 	 * it get discarded. Just report it.
4704 	 */
4705 	if (last_valid_block >= wp_block) {
4706 		f2fs_notice(sbi, "Valid block beyond write pointer: "
4707 			    "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4708 			    GET_SEGNO(sbi, last_valid_block),
4709 			    GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4710 			    wp_segno, wp_blkoff);
4711 		return 0;
4712 	}
4713 
4714 	/*
4715 	 * If there is no valid block in the zone and if write pointer is
4716 	 * not at zone start, reset the write pointer.
4717 	 */
4718 	if (last_valid_block + 1 == zone_block && zone->wp != zone->start) {
4719 		f2fs_notice(sbi,
4720 			    "Zone without valid block has non-zero write "
4721 			    "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4722 			    wp_segno, wp_blkoff);
4723 		ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4724 					zone->len >> log_sectors_per_block);
4725 		if (ret) {
4726 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4727 				 fdev->path, ret);
4728 			return ret;
4729 		}
4730 	}
4731 
4732 	return 0;
4733 }
4734 
4735 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4736 						  block_t zone_blkaddr)
4737 {
4738 	int i;
4739 
4740 	for (i = 0; i < sbi->s_ndevs; i++) {
4741 		if (!bdev_is_zoned(FDEV(i).bdev))
4742 			continue;
4743 		if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4744 				zone_blkaddr <= FDEV(i).end_blk))
4745 			return &FDEV(i);
4746 	}
4747 
4748 	return NULL;
4749 }
4750 
4751 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4752 			      void *data)
4753 {
4754 	memcpy(data, zone, sizeof(struct blk_zone));
4755 	return 0;
4756 }
4757 
4758 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4759 {
4760 	struct curseg_info *cs = CURSEG_I(sbi, type);
4761 	struct f2fs_dev_info *zbd;
4762 	struct blk_zone zone;
4763 	unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4764 	block_t cs_zone_block, wp_block;
4765 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4766 	sector_t zone_sector;
4767 	int err;
4768 
4769 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4770 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4771 
4772 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
4773 	if (!zbd)
4774 		return 0;
4775 
4776 	/* report zone for the sector the curseg points to */
4777 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4778 		<< log_sectors_per_block;
4779 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4780 				  report_one_zone_cb, &zone);
4781 	if (err != 1) {
4782 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4783 			 zbd->path, err);
4784 		return err;
4785 	}
4786 
4787 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4788 		return 0;
4789 
4790 	wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
4791 	wp_segno = GET_SEGNO(sbi, wp_block);
4792 	wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4793 	wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
4794 
4795 	if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
4796 		wp_sector_off == 0)
4797 		return 0;
4798 
4799 	f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
4800 		    "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
4801 		    type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
4802 
4803 	f2fs_notice(sbi, "Assign new section to curseg[%d]: "
4804 		    "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
4805 
4806 	f2fs_allocate_new_section(sbi, type, true);
4807 
4808 	/* check consistency of the zone curseg pointed to */
4809 	if (check_zone_write_pointer(sbi, zbd, &zone))
4810 		return -EIO;
4811 
4812 	/* check newly assigned zone */
4813 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4814 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4815 
4816 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
4817 	if (!zbd)
4818 		return 0;
4819 
4820 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4821 		<< log_sectors_per_block;
4822 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4823 				  report_one_zone_cb, &zone);
4824 	if (err != 1) {
4825 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4826 			 zbd->path, err);
4827 		return err;
4828 	}
4829 
4830 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4831 		return 0;
4832 
4833 	if (zone.wp != zone.start) {
4834 		f2fs_notice(sbi,
4835 			    "New zone for curseg[%d] is not yet discarded. "
4836 			    "Reset the zone: curseg[0x%x,0x%x]",
4837 			    type, cs->segno, cs->next_blkoff);
4838 		err = __f2fs_issue_discard_zone(sbi, zbd->bdev,
4839 				zone_sector >> log_sectors_per_block,
4840 				zone.len >> log_sectors_per_block);
4841 		if (err) {
4842 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4843 				 zbd->path, err);
4844 			return err;
4845 		}
4846 	}
4847 
4848 	return 0;
4849 }
4850 
4851 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4852 {
4853 	int i, ret;
4854 
4855 	for (i = 0; i < NR_PERSISTENT_LOG; i++) {
4856 		ret = fix_curseg_write_pointer(sbi, i);
4857 		if (ret)
4858 			return ret;
4859 	}
4860 
4861 	return 0;
4862 }
4863 
4864 struct check_zone_write_pointer_args {
4865 	struct f2fs_sb_info *sbi;
4866 	struct f2fs_dev_info *fdev;
4867 };
4868 
4869 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
4870 				      void *data)
4871 {
4872 	struct check_zone_write_pointer_args *args;
4873 
4874 	args = (struct check_zone_write_pointer_args *)data;
4875 
4876 	return check_zone_write_pointer(args->sbi, args->fdev, zone);
4877 }
4878 
4879 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4880 {
4881 	int i, ret;
4882 	struct check_zone_write_pointer_args args;
4883 
4884 	for (i = 0; i < sbi->s_ndevs; i++) {
4885 		if (!bdev_is_zoned(FDEV(i).bdev))
4886 			continue;
4887 
4888 		args.sbi = sbi;
4889 		args.fdev = &FDEV(i);
4890 		ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
4891 					  check_zone_write_pointer_cb, &args);
4892 		if (ret < 0)
4893 			return ret;
4894 	}
4895 
4896 	return 0;
4897 }
4898 
4899 static bool is_conv_zone(struct f2fs_sb_info *sbi, unsigned int zone_idx,
4900 						unsigned int dev_idx)
4901 {
4902 	if (!bdev_is_zoned(FDEV(dev_idx).bdev))
4903 		return true;
4904 	return !test_bit(zone_idx, FDEV(dev_idx).blkz_seq);
4905 }
4906 
4907 /* Return the zone index in the given device */
4908 static unsigned int get_zone_idx(struct f2fs_sb_info *sbi, unsigned int secno,
4909 					int dev_idx)
4910 {
4911 	block_t sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
4912 
4913 	return (sec_start_blkaddr - FDEV(dev_idx).start_blk) >>
4914 						sbi->log_blocks_per_blkz;
4915 }
4916 
4917 /*
4918  * Return the usable segments in a section based on the zone's
4919  * corresponding zone capacity. Zone is equal to a section.
4920  */
4921 static inline unsigned int f2fs_usable_zone_segs_in_sec(
4922 		struct f2fs_sb_info *sbi, unsigned int segno)
4923 {
4924 	unsigned int dev_idx, zone_idx;
4925 
4926 	dev_idx = f2fs_target_device_index(sbi, START_BLOCK(sbi, segno));
4927 	zone_idx = get_zone_idx(sbi, GET_SEC_FROM_SEG(sbi, segno), dev_idx);
4928 
4929 	/* Conventional zone's capacity is always equal to zone size */
4930 	if (is_conv_zone(sbi, zone_idx, dev_idx))
4931 		return sbi->segs_per_sec;
4932 
4933 	if (!sbi->unusable_blocks_per_sec)
4934 		return sbi->segs_per_sec;
4935 
4936 	/* Get the segment count beyond zone capacity block */
4937 	return sbi->segs_per_sec - (sbi->unusable_blocks_per_sec >>
4938 						sbi->log_blocks_per_seg);
4939 }
4940 
4941 /*
4942  * Return the number of usable blocks in a segment. The number of blocks
4943  * returned is always equal to the number of blocks in a segment for
4944  * segments fully contained within a sequential zone capacity or a
4945  * conventional zone. For segments partially contained in a sequential
4946  * zone capacity, the number of usable blocks up to the zone capacity
4947  * is returned. 0 is returned in all other cases.
4948  */
4949 static inline unsigned int f2fs_usable_zone_blks_in_seg(
4950 			struct f2fs_sb_info *sbi, unsigned int segno)
4951 {
4952 	block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
4953 	unsigned int zone_idx, dev_idx, secno;
4954 
4955 	secno = GET_SEC_FROM_SEG(sbi, segno);
4956 	seg_start = START_BLOCK(sbi, segno);
4957 	dev_idx = f2fs_target_device_index(sbi, seg_start);
4958 	zone_idx = get_zone_idx(sbi, secno, dev_idx);
4959 
4960 	/*
4961 	 * Conventional zone's capacity is always equal to zone size,
4962 	 * so, blocks per segment is unchanged.
4963 	 */
4964 	if (is_conv_zone(sbi, zone_idx, dev_idx))
4965 		return sbi->blocks_per_seg;
4966 
4967 	if (!sbi->unusable_blocks_per_sec)
4968 		return sbi->blocks_per_seg;
4969 
4970 	sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
4971 	sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi);
4972 
4973 	/*
4974 	 * If segment starts before zone capacity and spans beyond
4975 	 * zone capacity, then usable blocks are from seg start to
4976 	 * zone capacity. If the segment starts after the zone capacity,
4977 	 * then there are no usable blocks.
4978 	 */
4979 	if (seg_start >= sec_cap_blkaddr)
4980 		return 0;
4981 	if (seg_start + sbi->blocks_per_seg > sec_cap_blkaddr)
4982 		return sec_cap_blkaddr - seg_start;
4983 
4984 	return sbi->blocks_per_seg;
4985 }
4986 #else
4987 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4988 {
4989 	return 0;
4990 }
4991 
4992 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4993 {
4994 	return 0;
4995 }
4996 
4997 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
4998 							unsigned int segno)
4999 {
5000 	return 0;
5001 }
5002 
5003 static inline unsigned int f2fs_usable_zone_segs_in_sec(struct f2fs_sb_info *sbi,
5004 							unsigned int segno)
5005 {
5006 	return 0;
5007 }
5008 #endif
5009 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5010 					unsigned int segno)
5011 {
5012 	if (f2fs_sb_has_blkzoned(sbi))
5013 		return f2fs_usable_zone_blks_in_seg(sbi, segno);
5014 
5015 	return sbi->blocks_per_seg;
5016 }
5017 
5018 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi,
5019 					unsigned int segno)
5020 {
5021 	if (f2fs_sb_has_blkzoned(sbi))
5022 		return f2fs_usable_zone_segs_in_sec(sbi, segno);
5023 
5024 	return sbi->segs_per_sec;
5025 }
5026 
5027 /*
5028  * Update min, max modified time for cost-benefit GC algorithm
5029  */
5030 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5031 {
5032 	struct sit_info *sit_i = SIT_I(sbi);
5033 	unsigned int segno;
5034 
5035 	down_write(&sit_i->sentry_lock);
5036 
5037 	sit_i->min_mtime = ULLONG_MAX;
5038 
5039 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
5040 		unsigned int i;
5041 		unsigned long long mtime = 0;
5042 
5043 		for (i = 0; i < sbi->segs_per_sec; i++)
5044 			mtime += get_seg_entry(sbi, segno + i)->mtime;
5045 
5046 		mtime = div_u64(mtime, sbi->segs_per_sec);
5047 
5048 		if (sit_i->min_mtime > mtime)
5049 			sit_i->min_mtime = mtime;
5050 	}
5051 	sit_i->max_mtime = get_mtime(sbi, false);
5052 	sit_i->dirty_max_mtime = 0;
5053 	up_write(&sit_i->sentry_lock);
5054 }
5055 
5056 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5057 {
5058 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5059 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5060 	struct f2fs_sm_info *sm_info;
5061 	int err;
5062 
5063 	sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5064 	if (!sm_info)
5065 		return -ENOMEM;
5066 
5067 	/* init sm info */
5068 	sbi->sm_info = sm_info;
5069 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5070 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5071 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5072 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5073 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5074 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5075 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5076 	sm_info->rec_prefree_segments = sm_info->main_segments *
5077 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5078 	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5079 		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5080 
5081 	if (!f2fs_lfs_mode(sbi))
5082 		sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
5083 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5084 	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5085 	sm_info->min_seq_blocks = sbi->blocks_per_seg;
5086 	sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5087 	sm_info->min_ssr_sections = reserved_sections(sbi);
5088 
5089 	INIT_LIST_HEAD(&sm_info->sit_entry_set);
5090 
5091 	init_f2fs_rwsem(&sm_info->curseg_lock);
5092 
5093 	if (!f2fs_readonly(sbi->sb)) {
5094 		err = f2fs_create_flush_cmd_control(sbi);
5095 		if (err)
5096 			return err;
5097 	}
5098 
5099 	err = create_discard_cmd_control(sbi);
5100 	if (err)
5101 		return err;
5102 
5103 	err = build_sit_info(sbi);
5104 	if (err)
5105 		return err;
5106 	err = build_free_segmap(sbi);
5107 	if (err)
5108 		return err;
5109 	err = build_curseg(sbi);
5110 	if (err)
5111 		return err;
5112 
5113 	/* reinit free segmap based on SIT */
5114 	err = build_sit_entries(sbi);
5115 	if (err)
5116 		return err;
5117 
5118 	init_free_segmap(sbi);
5119 	err = build_dirty_segmap(sbi);
5120 	if (err)
5121 		return err;
5122 
5123 	err = sanity_check_curseg(sbi);
5124 	if (err)
5125 		return err;
5126 
5127 	init_min_max_mtime(sbi);
5128 	return 0;
5129 }
5130 
5131 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5132 		enum dirty_type dirty_type)
5133 {
5134 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5135 
5136 	mutex_lock(&dirty_i->seglist_lock);
5137 	kvfree(dirty_i->dirty_segmap[dirty_type]);
5138 	dirty_i->nr_dirty[dirty_type] = 0;
5139 	mutex_unlock(&dirty_i->seglist_lock);
5140 }
5141 
5142 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5143 {
5144 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5145 
5146 	kvfree(dirty_i->pinned_secmap);
5147 	kvfree(dirty_i->victim_secmap);
5148 }
5149 
5150 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5151 {
5152 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5153 	int i;
5154 
5155 	if (!dirty_i)
5156 		return;
5157 
5158 	/* discard pre-free/dirty segments list */
5159 	for (i = 0; i < NR_DIRTY_TYPE; i++)
5160 		discard_dirty_segmap(sbi, i);
5161 
5162 	if (__is_large_section(sbi)) {
5163 		mutex_lock(&dirty_i->seglist_lock);
5164 		kvfree(dirty_i->dirty_secmap);
5165 		mutex_unlock(&dirty_i->seglist_lock);
5166 	}
5167 
5168 	destroy_victim_secmap(sbi);
5169 	SM_I(sbi)->dirty_info = NULL;
5170 	kfree(dirty_i);
5171 }
5172 
5173 static void destroy_curseg(struct f2fs_sb_info *sbi)
5174 {
5175 	struct curseg_info *array = SM_I(sbi)->curseg_array;
5176 	int i;
5177 
5178 	if (!array)
5179 		return;
5180 	SM_I(sbi)->curseg_array = NULL;
5181 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
5182 		kfree(array[i].sum_blk);
5183 		kfree(array[i].journal);
5184 	}
5185 	kfree(array);
5186 }
5187 
5188 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5189 {
5190 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5191 
5192 	if (!free_i)
5193 		return;
5194 	SM_I(sbi)->free_info = NULL;
5195 	kvfree(free_i->free_segmap);
5196 	kvfree(free_i->free_secmap);
5197 	kfree(free_i);
5198 }
5199 
5200 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5201 {
5202 	struct sit_info *sit_i = SIT_I(sbi);
5203 
5204 	if (!sit_i)
5205 		return;
5206 
5207 	if (sit_i->sentries)
5208 		kvfree(sit_i->bitmap);
5209 	kfree(sit_i->tmp_map);
5210 
5211 	kvfree(sit_i->sentries);
5212 	kvfree(sit_i->sec_entries);
5213 	kvfree(sit_i->dirty_sentries_bitmap);
5214 
5215 	SM_I(sbi)->sit_info = NULL;
5216 	kvfree(sit_i->sit_bitmap);
5217 #ifdef CONFIG_F2FS_CHECK_FS
5218 	kvfree(sit_i->sit_bitmap_mir);
5219 	kvfree(sit_i->invalid_segmap);
5220 #endif
5221 	kfree(sit_i);
5222 }
5223 
5224 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5225 {
5226 	struct f2fs_sm_info *sm_info = SM_I(sbi);
5227 
5228 	if (!sm_info)
5229 		return;
5230 	f2fs_destroy_flush_cmd_control(sbi, true);
5231 	destroy_discard_cmd_control(sbi);
5232 	destroy_dirty_segmap(sbi);
5233 	destroy_curseg(sbi);
5234 	destroy_free_segmap(sbi);
5235 	destroy_sit_info(sbi);
5236 	sbi->sm_info = NULL;
5237 	kfree(sm_info);
5238 }
5239 
5240 int __init f2fs_create_segment_manager_caches(void)
5241 {
5242 	discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5243 			sizeof(struct discard_entry));
5244 	if (!discard_entry_slab)
5245 		goto fail;
5246 
5247 	discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5248 			sizeof(struct discard_cmd));
5249 	if (!discard_cmd_slab)
5250 		goto destroy_discard_entry;
5251 
5252 	sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5253 			sizeof(struct sit_entry_set));
5254 	if (!sit_entry_set_slab)
5255 		goto destroy_discard_cmd;
5256 
5257 	revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry",
5258 			sizeof(struct revoke_entry));
5259 	if (!revoke_entry_slab)
5260 		goto destroy_sit_entry_set;
5261 	return 0;
5262 
5263 destroy_sit_entry_set:
5264 	kmem_cache_destroy(sit_entry_set_slab);
5265 destroy_discard_cmd:
5266 	kmem_cache_destroy(discard_cmd_slab);
5267 destroy_discard_entry:
5268 	kmem_cache_destroy(discard_entry_slab);
5269 fail:
5270 	return -ENOMEM;
5271 }
5272 
5273 void f2fs_destroy_segment_manager_caches(void)
5274 {
5275 	kmem_cache_destroy(sit_entry_set_slab);
5276 	kmem_cache_destroy(discard_cmd_slab);
5277 	kmem_cache_destroy(discard_entry_slab);
5278 	kmem_cache_destroy(revoke_entry_slab);
5279 }
5280