xref: /linux/fs/f2fs/segment.c (revision 0974f486f3dde9df1ad979d4ff341dc9c2d545f5)
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 
__reverse_ulong(unsigned char * str)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  */
__reverse_ffs(unsigned long word)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  */
__find_rev_next_bit(const unsigned long * addr,unsigned long size,unsigned long offset)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 
__find_rev_next_zero_bit(const unsigned long * addr,unsigned long size,unsigned long offset)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 
f2fs_need_SSR(struct f2fs_sb_info * sbi)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 
f2fs_abort_atomic_write(struct inode * inode,bool clean)188 void f2fs_abort_atomic_write(struct inode *inode, bool clean)
189 {
190 	struct f2fs_inode_info *fi = F2FS_I(inode);
191 
192 	if (!f2fs_is_atomic_file(inode))
193 		return;
194 
195 	if (clean)
196 		truncate_inode_pages_final(inode->i_mapping);
197 
198 	release_atomic_write_cnt(inode);
199 	clear_inode_flag(inode, FI_ATOMIC_COMMITTED);
200 	clear_inode_flag(inode, FI_ATOMIC_REPLACE);
201 	clear_inode_flag(inode, FI_ATOMIC_FILE);
202 	if (is_inode_flag_set(inode, FI_ATOMIC_DIRTIED)) {
203 		clear_inode_flag(inode, FI_ATOMIC_DIRTIED);
204 		/*
205 		 * The vfs inode keeps clean during commit, but the f2fs inode
206 		 * doesn't. So clear the dirty state after commit and let
207 		 * f2fs_mark_inode_dirty_sync ensure a consistent dirty state.
208 		 */
209 		f2fs_inode_synced(inode);
210 		f2fs_mark_inode_dirty_sync(inode, true);
211 	}
212 	stat_dec_atomic_inode(inode);
213 
214 	F2FS_I(inode)->atomic_write_task = NULL;
215 
216 	if (clean) {
217 		f2fs_i_size_write(inode, fi->original_i_size);
218 		fi->original_i_size = 0;
219 	}
220 	/* avoid stale dirty inode during eviction */
221 	sync_inode_metadata(inode, 0);
222 }
223 
__replace_atomic_write_block(struct inode * inode,pgoff_t index,block_t new_addr,block_t * old_addr,bool recover)224 static int __replace_atomic_write_block(struct inode *inode, pgoff_t index,
225 			block_t new_addr, block_t *old_addr, bool recover)
226 {
227 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
228 	struct dnode_of_data dn;
229 	struct node_info ni;
230 	int err;
231 
232 retry:
233 	set_new_dnode(&dn, inode, NULL, NULL, 0);
234 	err = f2fs_get_dnode_of_data(&dn, index, ALLOC_NODE);
235 	if (err) {
236 		if (err == -ENOMEM) {
237 			f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
238 			goto retry;
239 		}
240 		return err;
241 	}
242 
243 	err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
244 	if (err) {
245 		f2fs_put_dnode(&dn);
246 		return err;
247 	}
248 
249 	if (recover) {
250 		/* dn.data_blkaddr is always valid */
251 		if (!__is_valid_data_blkaddr(new_addr)) {
252 			if (new_addr == NULL_ADDR)
253 				dec_valid_block_count(sbi, inode, 1);
254 			f2fs_invalidate_blocks(sbi, dn.data_blkaddr, 1);
255 			f2fs_update_data_blkaddr(&dn, new_addr);
256 		} else {
257 			f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
258 				new_addr, ni.version, true, true);
259 		}
260 	} else {
261 		blkcnt_t count = 1;
262 
263 		err = inc_valid_block_count(sbi, inode, &count, true);
264 		if (err) {
265 			f2fs_put_dnode(&dn);
266 			return err;
267 		}
268 
269 		*old_addr = dn.data_blkaddr;
270 		f2fs_truncate_data_blocks_range(&dn, 1);
271 		dec_valid_block_count(sbi, F2FS_I(inode)->cow_inode, count);
272 
273 		f2fs_replace_block(sbi, &dn, dn.data_blkaddr, new_addr,
274 					ni.version, true, false);
275 	}
276 
277 	f2fs_put_dnode(&dn);
278 
279 	trace_f2fs_replace_atomic_write_block(inode, F2FS_I(inode)->cow_inode,
280 			index, old_addr ? *old_addr : 0, new_addr, recover);
281 	return 0;
282 }
283 
__complete_revoke_list(struct inode * inode,struct list_head * head,bool revoke)284 static void __complete_revoke_list(struct inode *inode, struct list_head *head,
285 					bool revoke)
286 {
287 	struct revoke_entry *cur, *tmp;
288 	pgoff_t start_index = 0;
289 	bool truncate = is_inode_flag_set(inode, FI_ATOMIC_REPLACE);
290 
291 	list_for_each_entry_safe(cur, tmp, head, list) {
292 		if (revoke) {
293 			__replace_atomic_write_block(inode, cur->index,
294 						cur->old_addr, NULL, true);
295 		} else if (truncate) {
296 			f2fs_truncate_hole(inode, start_index, cur->index);
297 			start_index = cur->index + 1;
298 		}
299 
300 		list_del(&cur->list);
301 		kmem_cache_free(revoke_entry_slab, cur);
302 	}
303 
304 	if (!revoke && truncate)
305 		f2fs_do_truncate_blocks(inode, start_index * PAGE_SIZE, false);
306 }
307 
__f2fs_commit_atomic_write(struct inode * inode)308 static int __f2fs_commit_atomic_write(struct inode *inode)
309 {
310 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
311 	struct f2fs_inode_info *fi = F2FS_I(inode);
312 	struct inode *cow_inode = fi->cow_inode;
313 	struct revoke_entry *new;
314 	struct list_head revoke_list;
315 	block_t blkaddr;
316 	struct dnode_of_data dn;
317 	pgoff_t len = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
318 	pgoff_t off = 0, blen, index;
319 	int ret = 0, i;
320 
321 	INIT_LIST_HEAD(&revoke_list);
322 
323 	while (len) {
324 		blen = min_t(pgoff_t, ADDRS_PER_BLOCK(cow_inode), len);
325 
326 		set_new_dnode(&dn, cow_inode, NULL, NULL, 0);
327 		ret = f2fs_get_dnode_of_data(&dn, off, LOOKUP_NODE_RA);
328 		if (ret && ret != -ENOENT) {
329 			goto out;
330 		} else if (ret == -ENOENT) {
331 			ret = 0;
332 			if (dn.max_level == 0)
333 				goto out;
334 			goto next;
335 		}
336 
337 		blen = min((pgoff_t)ADDRS_PER_PAGE(dn.node_folio, cow_inode),
338 				len);
339 		index = off;
340 		for (i = 0; i < blen; i++, dn.ofs_in_node++, index++) {
341 			blkaddr = f2fs_data_blkaddr(&dn);
342 
343 			if (!__is_valid_data_blkaddr(blkaddr)) {
344 				continue;
345 			} else if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
346 					DATA_GENERIC_ENHANCE)) {
347 				f2fs_put_dnode(&dn);
348 				ret = -EFSCORRUPTED;
349 				goto out;
350 			}
351 
352 			new = f2fs_kmem_cache_alloc(revoke_entry_slab, GFP_NOFS,
353 							true, NULL);
354 
355 			ret = __replace_atomic_write_block(inode, index, blkaddr,
356 							&new->old_addr, false);
357 			if (ret) {
358 				f2fs_put_dnode(&dn);
359 				kmem_cache_free(revoke_entry_slab, new);
360 				goto out;
361 			}
362 
363 			f2fs_update_data_blkaddr(&dn, NULL_ADDR);
364 			new->index = index;
365 			list_add_tail(&new->list, &revoke_list);
366 		}
367 		f2fs_put_dnode(&dn);
368 next:
369 		off += blen;
370 		len -= blen;
371 	}
372 
373 out:
374 	if (time_to_inject(sbi, FAULT_TIMEOUT))
375 		f2fs_io_schedule_timeout_killable(DEFAULT_FAULT_TIMEOUT);
376 
377 	if (ret) {
378 		sbi->revoked_atomic_block += fi->atomic_write_cnt;
379 	} else {
380 		sbi->committed_atomic_block += fi->atomic_write_cnt;
381 		set_inode_flag(inode, FI_ATOMIC_COMMITTED);
382 
383 		/*
384 		 * inode may has no FI_ATOMIC_DIRTIED flag due to no write
385 		 * before commit.
386 		 */
387 		if (is_inode_flag_set(inode, FI_ATOMIC_DIRTIED)) {
388 			/* clear atomic dirty status and set vfs dirty status */
389 			clear_inode_flag(inode, FI_ATOMIC_DIRTIED);
390 			f2fs_mark_inode_dirty_sync(inode, true);
391 		}
392 	}
393 
394 	__complete_revoke_list(inode, &revoke_list, ret ? true : false);
395 
396 	return ret;
397 }
398 
f2fs_commit_atomic_write(struct inode * inode)399 int f2fs_commit_atomic_write(struct inode *inode)
400 {
401 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
402 	struct f2fs_inode_info *fi = F2FS_I(inode);
403 	int err;
404 
405 	err = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
406 	if (err)
407 		return err;
408 
409 	f2fs_down_write(&fi->i_gc_rwsem[WRITE]);
410 	f2fs_lock_op(sbi);
411 
412 	err = __f2fs_commit_atomic_write(inode);
413 
414 	f2fs_unlock_op(sbi);
415 	f2fs_up_write(&fi->i_gc_rwsem[WRITE]);
416 
417 	return err;
418 }
419 
420 /*
421  * This function balances dirty node and dentry pages.
422  * In addition, it controls garbage collection.
423  */
f2fs_balance_fs(struct f2fs_sb_info * sbi,bool need)424 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
425 {
426 	if (f2fs_cp_error(sbi))
427 		return;
428 
429 	if (time_to_inject(sbi, FAULT_CHECKPOINT))
430 		f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_FAULT_INJECT);
431 
432 	/* balance_fs_bg is able to be pending */
433 	if (need && excess_cached_nats(sbi))
434 		f2fs_balance_fs_bg(sbi, false);
435 
436 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
437 		return;
438 
439 	/*
440 	 * We should do GC or end up with checkpoint, if there are so many dirty
441 	 * dir/node pages without enough free segments.
442 	 */
443 	if (has_enough_free_secs(sbi, 0, 0))
444 		return;
445 
446 	if (test_opt(sbi, GC_MERGE) && sbi->gc_thread &&
447 				sbi->gc_thread->f2fs_gc_task) {
448 		DEFINE_WAIT(wait);
449 
450 		prepare_to_wait(&sbi->gc_thread->fggc_wq, &wait,
451 					TASK_UNINTERRUPTIBLE);
452 		wake_up(&sbi->gc_thread->gc_wait_queue_head);
453 		io_schedule();
454 		finish_wait(&sbi->gc_thread->fggc_wq, &wait);
455 	} else {
456 		struct f2fs_gc_control gc_control = {
457 			.victim_segno = NULL_SEGNO,
458 			.init_gc_type = f2fs_sb_has_blkzoned(sbi) ?
459 				FG_GC : BG_GC,
460 			.no_bg_gc = true,
461 			.should_migrate_blocks = false,
462 			.err_gc_skipped = false,
463 			.nr_free_secs = 1 };
464 		f2fs_down_write(&sbi->gc_lock);
465 		stat_inc_gc_call_count(sbi, FOREGROUND);
466 		f2fs_gc(sbi, &gc_control);
467 	}
468 }
469 
excess_dirty_threshold(struct f2fs_sb_info * sbi)470 static inline bool excess_dirty_threshold(struct f2fs_sb_info *sbi)
471 {
472 	int factor = f2fs_rwsem_is_locked(&sbi->cp_rwsem) ? 3 : 2;
473 	unsigned int dents = get_pages(sbi, F2FS_DIRTY_DENTS);
474 	unsigned int qdata = get_pages(sbi, F2FS_DIRTY_QDATA);
475 	unsigned int nodes = get_pages(sbi, F2FS_DIRTY_NODES);
476 	unsigned int meta = get_pages(sbi, F2FS_DIRTY_META);
477 	unsigned int imeta = get_pages(sbi, F2FS_DIRTY_IMETA);
478 	unsigned int threshold =
479 		SEGS_TO_BLKS(sbi, (factor * DEFAULT_DIRTY_THRESHOLD));
480 	unsigned int global_threshold = threshold * 3 / 2;
481 
482 	if (dents >= threshold || qdata >= threshold ||
483 		nodes >= threshold || meta >= threshold ||
484 		imeta >= threshold)
485 		return true;
486 	return dents + qdata + nodes + meta + imeta >  global_threshold;
487 }
488 
f2fs_balance_fs_bg(struct f2fs_sb_info * sbi,bool from_bg)489 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi, bool from_bg)
490 {
491 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
492 		return;
493 
494 	/* try to shrink extent cache when there is no enough memory */
495 	if (!f2fs_available_free_memory(sbi, READ_EXTENT_CACHE))
496 		f2fs_shrink_read_extent_tree(sbi,
497 				READ_EXTENT_CACHE_SHRINK_NUMBER);
498 
499 	/* try to shrink age extent cache when there is no enough memory */
500 	if (!f2fs_available_free_memory(sbi, AGE_EXTENT_CACHE))
501 		f2fs_shrink_age_extent_tree(sbi,
502 				AGE_EXTENT_CACHE_SHRINK_NUMBER);
503 
504 	/* check the # of cached NAT entries */
505 	if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
506 		f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
507 
508 	if (!f2fs_available_free_memory(sbi, FREE_NIDS))
509 		f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
510 	else
511 		f2fs_build_free_nids(sbi, false, false);
512 
513 	if (excess_dirty_nats(sbi) || excess_dirty_threshold(sbi) ||
514 		excess_prefree_segs(sbi) || !f2fs_space_for_roll_forward(sbi))
515 		goto do_sync;
516 
517 	/* there is background inflight IO or foreground operation recently */
518 	if (is_inflight_io(sbi, REQ_TIME) ||
519 		(!f2fs_time_over(sbi, REQ_TIME) && f2fs_rwsem_is_locked(&sbi->cp_rwsem)))
520 		return;
521 
522 	/* exceed periodical checkpoint timeout threshold */
523 	if (f2fs_time_over(sbi, CP_TIME))
524 		goto do_sync;
525 
526 	/* checkpoint is the only way to shrink partial cached entries */
527 	if (f2fs_available_free_memory(sbi, NAT_ENTRIES) &&
528 		f2fs_available_free_memory(sbi, INO_ENTRIES))
529 		return;
530 
531 do_sync:
532 	if (test_opt(sbi, DATA_FLUSH) && from_bg) {
533 		struct blk_plug plug;
534 
535 		mutex_lock(&sbi->flush_lock);
536 
537 		blk_start_plug(&plug);
538 		f2fs_sync_dirty_inodes(sbi, FILE_INODE, false);
539 		blk_finish_plug(&plug);
540 
541 		mutex_unlock(&sbi->flush_lock);
542 	}
543 	stat_inc_cp_call_count(sbi, BACKGROUND);
544 	f2fs_sync_fs(sbi->sb, 1);
545 }
546 
__submit_flush_wait(struct f2fs_sb_info * sbi,struct block_device * bdev)547 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
548 				struct block_device *bdev)
549 {
550 	int ret = blkdev_issue_flush(bdev);
551 
552 	trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
553 				test_opt(sbi, FLUSH_MERGE), ret);
554 	if (!ret)
555 		f2fs_update_iostat(sbi, NULL, FS_FLUSH_IO, 0);
556 	return ret;
557 }
558 
submit_flush_wait(struct f2fs_sb_info * sbi,nid_t ino)559 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
560 {
561 	int ret = 0;
562 	int i;
563 
564 	if (!f2fs_is_multi_device(sbi))
565 		return __submit_flush_wait(sbi, sbi->sb->s_bdev);
566 
567 	for (i = 0; i < sbi->s_ndevs; i++) {
568 		if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
569 			continue;
570 		ret = __submit_flush_wait(sbi, FDEV(i).bdev);
571 		if (ret)
572 			break;
573 	}
574 	return ret;
575 }
576 
issue_flush_thread(void * data)577 static int issue_flush_thread(void *data)
578 {
579 	struct f2fs_sb_info *sbi = data;
580 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
581 	wait_queue_head_t *q = &fcc->flush_wait_queue;
582 repeat:
583 	if (kthread_should_stop())
584 		return 0;
585 
586 	if (!llist_empty(&fcc->issue_list)) {
587 		struct flush_cmd *cmd, *next;
588 		int ret;
589 
590 		fcc->dispatch_list = llist_del_all(&fcc->issue_list);
591 		fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
592 
593 		cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
594 
595 		ret = submit_flush_wait(sbi, cmd->ino);
596 		atomic_inc(&fcc->issued_flush);
597 
598 		llist_for_each_entry_safe(cmd, next,
599 					  fcc->dispatch_list, llnode) {
600 			cmd->ret = ret;
601 			complete(&cmd->wait);
602 		}
603 		fcc->dispatch_list = NULL;
604 	}
605 
606 	wait_event_interruptible(*q,
607 		kthread_should_stop() || !llist_empty(&fcc->issue_list));
608 	goto repeat;
609 }
610 
f2fs_issue_flush(struct f2fs_sb_info * sbi,nid_t ino)611 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
612 {
613 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
614 	struct flush_cmd cmd;
615 	int ret;
616 
617 	if (test_opt(sbi, NOBARRIER))
618 		return 0;
619 
620 	if (!test_opt(sbi, FLUSH_MERGE)) {
621 		atomic_inc(&fcc->queued_flush);
622 		ret = submit_flush_wait(sbi, ino);
623 		atomic_dec(&fcc->queued_flush);
624 		atomic_inc(&fcc->issued_flush);
625 		return ret;
626 	}
627 
628 	if (atomic_inc_return(&fcc->queued_flush) == 1 ||
629 	    f2fs_is_multi_device(sbi)) {
630 		ret = submit_flush_wait(sbi, ino);
631 		atomic_dec(&fcc->queued_flush);
632 
633 		atomic_inc(&fcc->issued_flush);
634 		return ret;
635 	}
636 
637 	cmd.ino = ino;
638 	init_completion(&cmd.wait);
639 
640 	llist_add(&cmd.llnode, &fcc->issue_list);
641 
642 	/*
643 	 * update issue_list before we wake up issue_flush thread, this
644 	 * smp_mb() pairs with another barrier in ___wait_event(), see
645 	 * more details in comments of waitqueue_active().
646 	 */
647 	smp_mb();
648 
649 	if (waitqueue_active(&fcc->flush_wait_queue))
650 		wake_up(&fcc->flush_wait_queue);
651 
652 	if (fcc->f2fs_issue_flush) {
653 		wait_for_completion(&cmd.wait);
654 		atomic_dec(&fcc->queued_flush);
655 	} else {
656 		struct llist_node *list;
657 
658 		list = llist_del_all(&fcc->issue_list);
659 		if (!list) {
660 			wait_for_completion(&cmd.wait);
661 			atomic_dec(&fcc->queued_flush);
662 		} else {
663 			struct flush_cmd *tmp, *next;
664 
665 			ret = submit_flush_wait(sbi, ino);
666 
667 			llist_for_each_entry_safe(tmp, next, list, llnode) {
668 				if (tmp == &cmd) {
669 					cmd.ret = ret;
670 					atomic_dec(&fcc->queued_flush);
671 					continue;
672 				}
673 				tmp->ret = ret;
674 				complete(&tmp->wait);
675 			}
676 		}
677 	}
678 
679 	return cmd.ret;
680 }
681 
f2fs_create_flush_cmd_control(struct f2fs_sb_info * sbi)682 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
683 {
684 	dev_t dev = sbi->sb->s_bdev->bd_dev;
685 	struct flush_cmd_control *fcc;
686 
687 	if (SM_I(sbi)->fcc_info) {
688 		fcc = SM_I(sbi)->fcc_info;
689 		if (fcc->f2fs_issue_flush)
690 			return 0;
691 		goto init_thread;
692 	}
693 
694 	fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
695 	if (!fcc)
696 		return -ENOMEM;
697 	atomic_set(&fcc->issued_flush, 0);
698 	atomic_set(&fcc->queued_flush, 0);
699 	init_waitqueue_head(&fcc->flush_wait_queue);
700 	init_llist_head(&fcc->issue_list);
701 	SM_I(sbi)->fcc_info = fcc;
702 	if (!test_opt(sbi, FLUSH_MERGE))
703 		return 0;
704 
705 init_thread:
706 	fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
707 				"f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
708 	if (IS_ERR(fcc->f2fs_issue_flush)) {
709 		int err = PTR_ERR(fcc->f2fs_issue_flush);
710 
711 		fcc->f2fs_issue_flush = NULL;
712 		return err;
713 	}
714 
715 	return 0;
716 }
717 
f2fs_destroy_flush_cmd_control(struct f2fs_sb_info * sbi,bool free)718 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
719 {
720 	struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
721 
722 	if (fcc && fcc->f2fs_issue_flush) {
723 		struct task_struct *flush_thread = fcc->f2fs_issue_flush;
724 
725 		fcc->f2fs_issue_flush = NULL;
726 		kthread_stop(flush_thread);
727 	}
728 	if (free) {
729 		kfree(fcc);
730 		SM_I(sbi)->fcc_info = NULL;
731 	}
732 }
733 
f2fs_flush_device_cache(struct f2fs_sb_info * sbi)734 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
735 {
736 	int ret = 0, i;
737 
738 	if (!f2fs_is_multi_device(sbi))
739 		return 0;
740 
741 	if (test_opt(sbi, NOBARRIER))
742 		return 0;
743 
744 	for (i = 1; i < sbi->s_ndevs; i++) {
745 		int count = DEFAULT_RETRY_IO_COUNT;
746 
747 		if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
748 			continue;
749 
750 		do {
751 			ret = __submit_flush_wait(sbi, FDEV(i).bdev);
752 			if (ret)
753 				f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
754 		} while (ret && --count);
755 
756 		if (ret) {
757 			f2fs_stop_checkpoint(sbi, false,
758 					STOP_CP_REASON_FLUSH_FAIL);
759 			break;
760 		}
761 
762 		spin_lock(&sbi->dev_lock);
763 		f2fs_clear_bit(i, (char *)&sbi->dirty_device);
764 		spin_unlock(&sbi->dev_lock);
765 	}
766 
767 	return ret;
768 }
769 
__locate_dirty_segment(struct f2fs_sb_info * sbi,unsigned int segno,enum dirty_type dirty_type)770 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
771 		enum dirty_type dirty_type)
772 {
773 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
774 
775 	/* need not be added */
776 	if (is_curseg(sbi, segno))
777 		return;
778 
779 	if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
780 		dirty_i->nr_dirty[dirty_type]++;
781 
782 	if (dirty_type == DIRTY) {
783 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
784 		enum dirty_type t = sentry->type;
785 
786 		if (unlikely(t >= DIRTY)) {
787 			f2fs_bug_on(sbi, 1);
788 			return;
789 		}
790 		if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
791 			dirty_i->nr_dirty[t]++;
792 
793 		if (__is_large_section(sbi)) {
794 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
795 			block_t valid_blocks =
796 				get_valid_blocks(sbi, segno, true);
797 
798 			f2fs_bug_on(sbi,
799 				(!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
800 				!valid_blocks) ||
801 				valid_blocks == CAP_BLKS_PER_SEC(sbi));
802 
803 			if (!is_cursec(sbi, secno))
804 				set_bit(secno, dirty_i->dirty_secmap);
805 		}
806 	}
807 }
808 
__remove_dirty_segment(struct f2fs_sb_info * sbi,unsigned int segno,enum dirty_type dirty_type)809 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
810 		enum dirty_type dirty_type)
811 {
812 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
813 	block_t valid_blocks;
814 
815 	if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
816 		dirty_i->nr_dirty[dirty_type]--;
817 
818 	if (dirty_type == DIRTY) {
819 		struct seg_entry *sentry = get_seg_entry(sbi, segno);
820 		enum dirty_type t = sentry->type;
821 
822 		if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
823 			dirty_i->nr_dirty[t]--;
824 
825 		valid_blocks = get_valid_blocks(sbi, segno, true);
826 		if (valid_blocks == 0) {
827 			clear_bit(GET_SEC_FROM_SEG(sbi, segno),
828 						dirty_i->victim_secmap);
829 #ifdef CONFIG_F2FS_CHECK_FS
830 			clear_bit(segno, SIT_I(sbi)->invalid_segmap);
831 #endif
832 		}
833 		if (__is_large_section(sbi)) {
834 			unsigned int secno = GET_SEC_FROM_SEG(sbi, segno);
835 
836 			if (!valid_blocks ||
837 					valid_blocks == CAP_BLKS_PER_SEC(sbi)) {
838 				clear_bit(secno, dirty_i->dirty_secmap);
839 				return;
840 			}
841 
842 			if (!is_cursec(sbi, secno))
843 				set_bit(secno, dirty_i->dirty_secmap);
844 		}
845 	}
846 }
847 
848 /*
849  * Should not occur error such as -ENOMEM.
850  * Adding dirty entry into seglist is not critical operation.
851  * If a given segment is one of current working segments, it won't be added.
852  */
locate_dirty_segment(struct f2fs_sb_info * sbi,unsigned int segno)853 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
854 {
855 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
856 	unsigned short valid_blocks, ckpt_valid_blocks;
857 	unsigned int usable_blocks;
858 
859 	if (segno == NULL_SEGNO || is_curseg(sbi, segno))
860 		return;
861 
862 	usable_blocks = f2fs_usable_blks_in_seg(sbi, segno);
863 	mutex_lock(&dirty_i->seglist_lock);
864 
865 	valid_blocks = get_valid_blocks(sbi, segno, false);
866 	ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno, false);
867 
868 	if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
869 		ckpt_valid_blocks == usable_blocks)) {
870 		__locate_dirty_segment(sbi, segno, PRE);
871 		__remove_dirty_segment(sbi, segno, DIRTY);
872 	} else if (valid_blocks < usable_blocks) {
873 		__locate_dirty_segment(sbi, segno, DIRTY);
874 	} else {
875 		/* Recovery routine with SSR needs this */
876 		__remove_dirty_segment(sbi, segno, DIRTY);
877 	}
878 
879 	mutex_unlock(&dirty_i->seglist_lock);
880 }
881 
882 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
f2fs_dirty_to_prefree(struct f2fs_sb_info * sbi)883 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
884 {
885 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
886 	unsigned int segno;
887 
888 	mutex_lock(&dirty_i->seglist_lock);
889 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
890 		if (get_valid_blocks(sbi, segno, false))
891 			continue;
892 		if (is_curseg(sbi, segno))
893 			continue;
894 		__locate_dirty_segment(sbi, segno, PRE);
895 		__remove_dirty_segment(sbi, segno, DIRTY);
896 	}
897 	mutex_unlock(&dirty_i->seglist_lock);
898 }
899 
f2fs_get_unusable_blocks(struct f2fs_sb_info * sbi)900 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
901 {
902 	int ovp_hole_segs =
903 		(overprovision_segments(sbi) - reserved_segments(sbi));
904 	block_t ovp_holes = SEGS_TO_BLKS(sbi, ovp_hole_segs);
905 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
906 	block_t holes[2] = {0, 0};	/* DATA and NODE */
907 	block_t unusable;
908 	struct seg_entry *se;
909 	unsigned int segno;
910 
911 	mutex_lock(&dirty_i->seglist_lock);
912 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
913 		se = get_seg_entry(sbi, segno);
914 		if (IS_NODESEG(se->type))
915 			holes[NODE] += f2fs_usable_blks_in_seg(sbi, segno) -
916 							se->valid_blocks;
917 		else
918 			holes[DATA] += f2fs_usable_blks_in_seg(sbi, segno) -
919 							se->valid_blocks;
920 	}
921 	mutex_unlock(&dirty_i->seglist_lock);
922 
923 	unusable = max(holes[DATA], holes[NODE]);
924 	if (unusable > ovp_holes)
925 		return unusable - ovp_holes;
926 	return 0;
927 }
928 
f2fs_disable_cp_again(struct f2fs_sb_info * sbi,block_t unusable)929 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
930 {
931 	int ovp_hole_segs =
932 		(overprovision_segments(sbi) - reserved_segments(sbi));
933 
934 	if (F2FS_OPTION(sbi).unusable_cap_perc == 100)
935 		return 0;
936 	if (unusable > F2FS_OPTION(sbi).unusable_cap)
937 		return -EAGAIN;
938 	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
939 		dirty_segments(sbi) > ovp_hole_segs)
940 		return -EAGAIN;
941 	if (has_not_enough_free_secs(sbi, 0, 0))
942 		return -EAGAIN;
943 	return 0;
944 }
945 
946 /* This is only used by SBI_CP_DISABLED */
get_free_segment(struct f2fs_sb_info * sbi)947 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
948 {
949 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
950 	unsigned int segno = 0;
951 
952 	mutex_lock(&dirty_i->seglist_lock);
953 	for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
954 		if (get_valid_blocks(sbi, segno, false))
955 			continue;
956 		if (get_ckpt_valid_blocks(sbi, segno, false))
957 			continue;
958 		mutex_unlock(&dirty_i->seglist_lock);
959 		return segno;
960 	}
961 	mutex_unlock(&dirty_i->seglist_lock);
962 	return NULL_SEGNO;
963 }
964 
__create_discard_cmd(struct f2fs_sb_info * sbi,struct block_device * bdev,block_t lstart,block_t start,block_t len)965 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
966 		struct block_device *bdev, block_t lstart,
967 		block_t start, block_t len)
968 {
969 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
970 	struct list_head *pend_list;
971 	struct discard_cmd *dc;
972 
973 	f2fs_bug_on(sbi, !len);
974 
975 	pend_list = &dcc->pend_list[plist_idx(len)];
976 
977 	dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS, true, NULL);
978 	INIT_LIST_HEAD(&dc->list);
979 	dc->bdev = bdev;
980 	dc->di.lstart = lstart;
981 	dc->di.start = start;
982 	dc->di.len = len;
983 	dc->ref = 0;
984 	dc->state = D_PREP;
985 	dc->queued = 0;
986 	dc->error = 0;
987 	init_completion(&dc->wait);
988 	list_add_tail(&dc->list, pend_list);
989 	spin_lock_init(&dc->lock);
990 	dc->bio_ref = 0;
991 	atomic_inc(&dcc->discard_cmd_cnt);
992 	dcc->undiscard_blks += len;
993 
994 	return dc;
995 }
996 
f2fs_check_discard_tree(struct f2fs_sb_info * sbi)997 static bool f2fs_check_discard_tree(struct f2fs_sb_info *sbi)
998 {
999 #ifdef CONFIG_F2FS_CHECK_FS
1000 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1001 	struct rb_node *cur = rb_first_cached(&dcc->root), *next;
1002 	struct discard_cmd *cur_dc, *next_dc;
1003 
1004 	while (cur) {
1005 		next = rb_next(cur);
1006 		if (!next)
1007 			return true;
1008 
1009 		cur_dc = rb_entry(cur, struct discard_cmd, rb_node);
1010 		next_dc = rb_entry(next, struct discard_cmd, rb_node);
1011 
1012 		if (cur_dc->di.lstart + cur_dc->di.len > next_dc->di.lstart) {
1013 			f2fs_info(sbi, "broken discard_rbtree, "
1014 				"cur(%u, %u) next(%u, %u)",
1015 				cur_dc->di.lstart, cur_dc->di.len,
1016 				next_dc->di.lstart, next_dc->di.len);
1017 			return false;
1018 		}
1019 		cur = next;
1020 	}
1021 #endif
1022 	return true;
1023 }
1024 
__lookup_discard_cmd(struct f2fs_sb_info * sbi,block_t blkaddr)1025 static struct discard_cmd *__lookup_discard_cmd(struct f2fs_sb_info *sbi,
1026 						block_t blkaddr)
1027 {
1028 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1029 	struct rb_node *node = dcc->root.rb_root.rb_node;
1030 	struct discard_cmd *dc;
1031 
1032 	while (node) {
1033 		dc = rb_entry(node, struct discard_cmd, rb_node);
1034 
1035 		if (blkaddr < dc->di.lstart)
1036 			node = node->rb_left;
1037 		else if (blkaddr >= dc->di.lstart + dc->di.len)
1038 			node = node->rb_right;
1039 		else
1040 			return dc;
1041 	}
1042 	return NULL;
1043 }
1044 
__lookup_discard_cmd_ret(struct rb_root_cached * root,block_t blkaddr,struct discard_cmd ** prev_entry,struct discard_cmd ** next_entry,struct rb_node *** insert_p,struct rb_node ** insert_parent)1045 static struct discard_cmd *__lookup_discard_cmd_ret(struct rb_root_cached *root,
1046 				block_t blkaddr,
1047 				struct discard_cmd **prev_entry,
1048 				struct discard_cmd **next_entry,
1049 				struct rb_node ***insert_p,
1050 				struct rb_node **insert_parent)
1051 {
1052 	struct rb_node **pnode = &root->rb_root.rb_node;
1053 	struct rb_node *parent = NULL, *tmp_node;
1054 	struct discard_cmd *dc;
1055 
1056 	*insert_p = NULL;
1057 	*insert_parent = NULL;
1058 	*prev_entry = NULL;
1059 	*next_entry = NULL;
1060 
1061 	if (RB_EMPTY_ROOT(&root->rb_root))
1062 		return NULL;
1063 
1064 	while (*pnode) {
1065 		parent = *pnode;
1066 		dc = rb_entry(*pnode, struct discard_cmd, rb_node);
1067 
1068 		if (blkaddr < dc->di.lstart)
1069 			pnode = &(*pnode)->rb_left;
1070 		else if (blkaddr >= dc->di.lstart + dc->di.len)
1071 			pnode = &(*pnode)->rb_right;
1072 		else
1073 			goto lookup_neighbors;
1074 	}
1075 
1076 	*insert_p = pnode;
1077 	*insert_parent = parent;
1078 
1079 	dc = rb_entry(parent, struct discard_cmd, rb_node);
1080 	tmp_node = parent;
1081 	if (parent && blkaddr > dc->di.lstart)
1082 		tmp_node = rb_next(parent);
1083 	*next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1084 
1085 	tmp_node = parent;
1086 	if (parent && blkaddr < dc->di.lstart)
1087 		tmp_node = rb_prev(parent);
1088 	*prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1089 	return NULL;
1090 
1091 lookup_neighbors:
1092 	/* lookup prev node for merging backward later */
1093 	tmp_node = rb_prev(&dc->rb_node);
1094 	*prev_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1095 
1096 	/* lookup next node for merging frontward later */
1097 	tmp_node = rb_next(&dc->rb_node);
1098 	*next_entry = rb_entry_safe(tmp_node, struct discard_cmd, rb_node);
1099 	return dc;
1100 }
1101 
__detach_discard_cmd(struct discard_cmd_control * dcc,struct discard_cmd * dc)1102 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
1103 							struct discard_cmd *dc)
1104 {
1105 	if (dc->state == D_DONE)
1106 		atomic_sub(dc->queued, &dcc->queued_discard);
1107 
1108 	list_del(&dc->list);
1109 	rb_erase_cached(&dc->rb_node, &dcc->root);
1110 	dcc->undiscard_blks -= dc->di.len;
1111 
1112 	kmem_cache_free(discard_cmd_slab, dc);
1113 
1114 	atomic_dec(&dcc->discard_cmd_cnt);
1115 }
1116 
__remove_discard_cmd(struct f2fs_sb_info * sbi,struct discard_cmd * dc)1117 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
1118 							struct discard_cmd *dc)
1119 {
1120 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1121 	unsigned long flags;
1122 
1123 	trace_f2fs_remove_discard(dc->bdev, dc->di.start, dc->di.len);
1124 
1125 	spin_lock_irqsave(&dc->lock, flags);
1126 	if (dc->bio_ref) {
1127 		spin_unlock_irqrestore(&dc->lock, flags);
1128 		return;
1129 	}
1130 	spin_unlock_irqrestore(&dc->lock, flags);
1131 
1132 	f2fs_bug_on(sbi, dc->ref);
1133 
1134 	if (dc->error == -EOPNOTSUPP)
1135 		dc->error = 0;
1136 
1137 	if (dc->error)
1138 		f2fs_info_ratelimited(sbi,
1139 			"Issue discard(%u, %u, %u) failed, ret: %d",
1140 			dc->di.lstart, dc->di.start, dc->di.len, dc->error);
1141 	__detach_discard_cmd(dcc, dc);
1142 }
1143 
f2fs_submit_discard_endio(struct bio * bio)1144 static void f2fs_submit_discard_endio(struct bio *bio)
1145 {
1146 	struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1147 	unsigned long flags;
1148 
1149 	spin_lock_irqsave(&dc->lock, flags);
1150 	if (!dc->error)
1151 		dc->error = blk_status_to_errno(bio->bi_status);
1152 	dc->bio_ref--;
1153 	if (!dc->bio_ref && dc->state == D_SUBMIT) {
1154 		dc->state = D_DONE;
1155 		complete_all(&dc->wait);
1156 	}
1157 	spin_unlock_irqrestore(&dc->lock, flags);
1158 	bio_put(bio);
1159 }
1160 
__check_sit_bitmap(struct f2fs_sb_info * sbi,block_t start,block_t end)1161 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1162 				block_t start, block_t end)
1163 {
1164 #ifdef CONFIG_F2FS_CHECK_FS
1165 	struct seg_entry *sentry;
1166 	unsigned int segno;
1167 	block_t blk = start;
1168 	unsigned long offset, size, *map;
1169 
1170 	while (blk < end) {
1171 		segno = GET_SEGNO(sbi, blk);
1172 		sentry = get_seg_entry(sbi, segno);
1173 		offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1174 
1175 		if (end < START_BLOCK(sbi, segno + 1))
1176 			size = GET_BLKOFF_FROM_SEG0(sbi, end);
1177 		else
1178 			size = BLKS_PER_SEG(sbi);
1179 		map = (unsigned long *)(sentry->cur_valid_map);
1180 		offset = __find_rev_next_bit(map, size, offset);
1181 		f2fs_bug_on(sbi, offset != size);
1182 		blk = START_BLOCK(sbi, segno + 1);
1183 	}
1184 #endif
1185 }
1186 
__init_discard_policy(struct f2fs_sb_info * sbi,struct discard_policy * dpolicy,int discard_type,unsigned int granularity)1187 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1188 				struct discard_policy *dpolicy,
1189 				int discard_type, unsigned int granularity)
1190 {
1191 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1192 
1193 	/* common policy */
1194 	dpolicy->type = discard_type;
1195 	dpolicy->sync = true;
1196 	dpolicy->ordered = false;
1197 	dpolicy->granularity = granularity;
1198 
1199 	dpolicy->max_requests = dcc->max_discard_request;
1200 	dpolicy->io_aware_gran = dcc->discard_io_aware_gran;
1201 	dpolicy->timeout = false;
1202 
1203 	if (discard_type == DPOLICY_BG) {
1204 		dpolicy->min_interval = dcc->min_discard_issue_time;
1205 		dpolicy->mid_interval = dcc->mid_discard_issue_time;
1206 		dpolicy->max_interval = dcc->max_discard_issue_time;
1207 		if (dcc->discard_io_aware == DPOLICY_IO_AWARE_ENABLE)
1208 			dpolicy->io_aware = true;
1209 		else if (dcc->discard_io_aware == DPOLICY_IO_AWARE_DISABLE)
1210 			dpolicy->io_aware = false;
1211 		dpolicy->sync = false;
1212 		dpolicy->ordered = true;
1213 		if (utilization(sbi) > dcc->discard_urgent_util) {
1214 			dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1215 			if (atomic_read(&dcc->discard_cmd_cnt))
1216 				dpolicy->max_interval =
1217 					dcc->min_discard_issue_time;
1218 		}
1219 	} else if (discard_type == DPOLICY_FORCE) {
1220 		dpolicy->min_interval = dcc->min_discard_issue_time;
1221 		dpolicy->mid_interval = dcc->mid_discard_issue_time;
1222 		dpolicy->max_interval = dcc->max_discard_issue_time;
1223 		dpolicy->io_aware = false;
1224 	} else if (discard_type == DPOLICY_FSTRIM) {
1225 		dpolicy->io_aware = false;
1226 	} else if (discard_type == DPOLICY_UMOUNT) {
1227 		dpolicy->io_aware = false;
1228 		/* we need to issue all to keep CP_TRIMMED_FLAG */
1229 		dpolicy->granularity = MIN_DISCARD_GRANULARITY;
1230 		dpolicy->timeout = true;
1231 	}
1232 }
1233 
1234 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1235 				struct block_device *bdev, block_t lstart,
1236 				block_t start, block_t len);
1237 
1238 #ifdef CONFIG_BLK_DEV_ZONED
__submit_zone_reset_cmd(struct f2fs_sb_info * sbi,struct discard_cmd * dc,blk_opf_t flag,struct list_head * wait_list,unsigned int * issued)1239 static void __submit_zone_reset_cmd(struct f2fs_sb_info *sbi,
1240 				   struct discard_cmd *dc, blk_opf_t flag,
1241 				   struct list_head *wait_list,
1242 				   unsigned int *issued)
1243 {
1244 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1245 	struct block_device *bdev = dc->bdev;
1246 	struct bio *bio = bio_alloc(bdev, 0, REQ_OP_ZONE_RESET | flag, GFP_NOFS);
1247 	unsigned long flags;
1248 
1249 	trace_f2fs_issue_reset_zone(bdev, dc->di.start);
1250 
1251 	spin_lock_irqsave(&dc->lock, flags);
1252 	dc->state = D_SUBMIT;
1253 	dc->bio_ref++;
1254 	spin_unlock_irqrestore(&dc->lock, flags);
1255 
1256 	if (issued)
1257 		(*issued)++;
1258 
1259 	atomic_inc(&dcc->queued_discard);
1260 	dc->queued++;
1261 	list_move_tail(&dc->list, wait_list);
1262 
1263 	/* sanity check on discard range */
1264 	__check_sit_bitmap(sbi, dc->di.lstart, dc->di.lstart + dc->di.len);
1265 
1266 	bio->bi_iter.bi_sector = SECTOR_FROM_BLOCK(dc->di.start);
1267 	bio->bi_private = dc;
1268 	bio->bi_end_io = f2fs_submit_discard_endio;
1269 	submit_bio(bio);
1270 
1271 	atomic_inc(&dcc->issued_discard);
1272 	f2fs_update_iostat(sbi, NULL, FS_ZONE_RESET_IO, dc->di.len * F2FS_BLKSIZE);
1273 }
1274 #endif
1275 
1276 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
__submit_discard_cmd(struct f2fs_sb_info * sbi,struct discard_policy * dpolicy,struct discard_cmd * dc,int * issued)1277 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1278 				struct discard_policy *dpolicy,
1279 				struct discard_cmd *dc, int *issued)
1280 {
1281 	struct block_device *bdev = dc->bdev;
1282 	unsigned int max_discard_blocks =
1283 			SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1284 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1285 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1286 					&(dcc->fstrim_list) : &(dcc->wait_list);
1287 	blk_opf_t flag = dpolicy->sync ? REQ_SYNC : 0;
1288 	block_t lstart, start, len, total_len;
1289 	int err = 0;
1290 
1291 	if (dc->state != D_PREP)
1292 		return 0;
1293 
1294 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1295 		return 0;
1296 
1297 #ifdef CONFIG_BLK_DEV_ZONED
1298 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev)) {
1299 		int devi = f2fs_bdev_index(sbi, bdev);
1300 
1301 		if (devi < 0)
1302 			return -EINVAL;
1303 
1304 		if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1305 			__submit_zone_reset_cmd(sbi, dc, flag,
1306 						wait_list, issued);
1307 			return 0;
1308 		}
1309 	}
1310 #endif
1311 
1312 	/*
1313 	 * stop issuing discard for any of below cases:
1314 	 * 1. device is conventional zone, but it doesn't support discard.
1315 	 * 2. device is regulare device, after snapshot it doesn't support
1316 	 * discard.
1317 	 */
1318 	if (!bdev_max_discard_sectors(bdev))
1319 		return -EOPNOTSUPP;
1320 
1321 	trace_f2fs_issue_discard(bdev, dc->di.start, dc->di.len);
1322 
1323 	lstart = dc->di.lstart;
1324 	start = dc->di.start;
1325 	len = dc->di.len;
1326 	total_len = len;
1327 
1328 	dc->di.len = 0;
1329 
1330 	while (total_len && *issued < dpolicy->max_requests && !err) {
1331 		struct bio *bio = NULL;
1332 		unsigned long flags;
1333 		bool last = true;
1334 
1335 		if (len > max_discard_blocks) {
1336 			len = max_discard_blocks;
1337 			last = false;
1338 		}
1339 
1340 		(*issued)++;
1341 		if (*issued == dpolicy->max_requests)
1342 			last = true;
1343 
1344 		dc->di.len += len;
1345 
1346 		if (time_to_inject(sbi, FAULT_DISCARD)) {
1347 			err = -EIO;
1348 		} else {
1349 			err = __blkdev_issue_discard(bdev,
1350 					SECTOR_FROM_BLOCK(start),
1351 					SECTOR_FROM_BLOCK(len),
1352 					GFP_NOFS, &bio);
1353 		}
1354 		if (err) {
1355 			spin_lock_irqsave(&dc->lock, flags);
1356 			if (dc->state == D_PARTIAL)
1357 				dc->state = D_SUBMIT;
1358 			spin_unlock_irqrestore(&dc->lock, flags);
1359 
1360 			break;
1361 		}
1362 
1363 		f2fs_bug_on(sbi, !bio);
1364 
1365 		/*
1366 		 * should keep before submission to avoid D_DONE
1367 		 * right away
1368 		 */
1369 		spin_lock_irqsave(&dc->lock, flags);
1370 		if (last)
1371 			dc->state = D_SUBMIT;
1372 		else
1373 			dc->state = D_PARTIAL;
1374 		dc->bio_ref++;
1375 		spin_unlock_irqrestore(&dc->lock, flags);
1376 
1377 		atomic_inc(&dcc->queued_discard);
1378 		dc->queued++;
1379 		list_move_tail(&dc->list, wait_list);
1380 
1381 		/* sanity check on discard range */
1382 		__check_sit_bitmap(sbi, lstart, lstart + len);
1383 
1384 		bio->bi_private = dc;
1385 		bio->bi_end_io = f2fs_submit_discard_endio;
1386 		bio->bi_opf |= flag;
1387 		submit_bio(bio);
1388 
1389 		atomic_inc(&dcc->issued_discard);
1390 
1391 		f2fs_update_iostat(sbi, NULL, FS_DISCARD_IO, len * F2FS_BLKSIZE);
1392 
1393 		lstart += len;
1394 		start += len;
1395 		total_len -= len;
1396 		len = total_len;
1397 	}
1398 
1399 	if (!err && len) {
1400 		dcc->undiscard_blks -= len;
1401 		__update_discard_tree_range(sbi, bdev, lstart, start, len);
1402 	}
1403 	return err;
1404 }
1405 
__insert_discard_cmd(struct f2fs_sb_info * sbi,struct block_device * bdev,block_t lstart,block_t start,block_t len)1406 static void __insert_discard_cmd(struct f2fs_sb_info *sbi,
1407 				struct block_device *bdev, block_t lstart,
1408 				block_t start, block_t len)
1409 {
1410 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1411 	struct rb_node **p = &dcc->root.rb_root.rb_node;
1412 	struct rb_node *parent = NULL;
1413 	struct discard_cmd *dc;
1414 	bool leftmost = true;
1415 
1416 	/* look up rb tree to find parent node */
1417 	while (*p) {
1418 		parent = *p;
1419 		dc = rb_entry(parent, struct discard_cmd, rb_node);
1420 
1421 		if (lstart < dc->di.lstart) {
1422 			p = &(*p)->rb_left;
1423 		} else if (lstart >= dc->di.lstart + dc->di.len) {
1424 			p = &(*p)->rb_right;
1425 			leftmost = false;
1426 		} else {
1427 			/* Let's skip to add, if exists */
1428 			return;
1429 		}
1430 	}
1431 
1432 	dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
1433 
1434 	rb_link_node(&dc->rb_node, parent, p);
1435 	rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
1436 }
1437 
__relocate_discard_cmd(struct discard_cmd_control * dcc,struct discard_cmd * dc)1438 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1439 						struct discard_cmd *dc)
1440 {
1441 	list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->di.len)]);
1442 }
1443 
__punch_discard_cmd(struct f2fs_sb_info * sbi,struct discard_cmd * dc,block_t blkaddr)1444 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1445 				struct discard_cmd *dc, block_t blkaddr)
1446 {
1447 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1448 	struct discard_info di = dc->di;
1449 	bool modified = false;
1450 
1451 	if (dc->state == D_DONE || dc->di.len == 1) {
1452 		__remove_discard_cmd(sbi, dc);
1453 		return;
1454 	}
1455 
1456 	dcc->undiscard_blks -= di.len;
1457 
1458 	if (blkaddr > di.lstart) {
1459 		dc->di.len = blkaddr - dc->di.lstart;
1460 		dcc->undiscard_blks += dc->di.len;
1461 		__relocate_discard_cmd(dcc, dc);
1462 		modified = true;
1463 	}
1464 
1465 	if (blkaddr < di.lstart + di.len - 1) {
1466 		if (modified) {
1467 			__insert_discard_cmd(sbi, dc->bdev, blkaddr + 1,
1468 					di.start + blkaddr + 1 - di.lstart,
1469 					di.lstart + di.len - 1 - blkaddr);
1470 		} else {
1471 			dc->di.lstart++;
1472 			dc->di.len--;
1473 			dc->di.start++;
1474 			dcc->undiscard_blks += dc->di.len;
1475 			__relocate_discard_cmd(dcc, dc);
1476 		}
1477 	}
1478 }
1479 
__update_discard_tree_range(struct f2fs_sb_info * sbi,struct block_device * bdev,block_t lstart,block_t start,block_t len)1480 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1481 				struct block_device *bdev, block_t lstart,
1482 				block_t start, block_t len)
1483 {
1484 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1485 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1486 	struct discard_cmd *dc;
1487 	struct discard_info di = {0};
1488 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1489 	unsigned int max_discard_blocks =
1490 			SECTOR_TO_BLOCK(bdev_max_discard_sectors(bdev));
1491 	block_t end = lstart + len;
1492 
1493 	dc = __lookup_discard_cmd_ret(&dcc->root, lstart,
1494 				&prev_dc, &next_dc, &insert_p, &insert_parent);
1495 	if (dc)
1496 		prev_dc = dc;
1497 
1498 	if (!prev_dc) {
1499 		di.lstart = lstart;
1500 		di.len = next_dc ? next_dc->di.lstart - lstart : len;
1501 		di.len = min(di.len, len);
1502 		di.start = start;
1503 	}
1504 
1505 	while (1) {
1506 		struct rb_node *node;
1507 		bool merged = false;
1508 		struct discard_cmd *tdc = NULL;
1509 
1510 		if (prev_dc) {
1511 			di.lstart = prev_dc->di.lstart + prev_dc->di.len;
1512 			if (di.lstart < lstart)
1513 				di.lstart = lstart;
1514 			if (di.lstart >= end)
1515 				break;
1516 
1517 			if (!next_dc || next_dc->di.lstart > end)
1518 				di.len = end - di.lstart;
1519 			else
1520 				di.len = next_dc->di.lstart - di.lstart;
1521 			di.start = start + di.lstart - lstart;
1522 		}
1523 
1524 		if (!di.len)
1525 			goto next;
1526 
1527 		if (prev_dc && prev_dc->state == D_PREP &&
1528 			prev_dc->bdev == bdev &&
1529 			__is_discard_back_mergeable(&di, &prev_dc->di,
1530 							max_discard_blocks)) {
1531 			prev_dc->di.len += di.len;
1532 			dcc->undiscard_blks += di.len;
1533 			__relocate_discard_cmd(dcc, prev_dc);
1534 			di = prev_dc->di;
1535 			tdc = prev_dc;
1536 			merged = true;
1537 		}
1538 
1539 		if (next_dc && next_dc->state == D_PREP &&
1540 			next_dc->bdev == bdev &&
1541 			__is_discard_front_mergeable(&di, &next_dc->di,
1542 							max_discard_blocks)) {
1543 			next_dc->di.lstart = di.lstart;
1544 			next_dc->di.len += di.len;
1545 			next_dc->di.start = di.start;
1546 			dcc->undiscard_blks += di.len;
1547 			__relocate_discard_cmd(dcc, next_dc);
1548 			if (tdc)
1549 				__remove_discard_cmd(sbi, tdc);
1550 			merged = true;
1551 		}
1552 
1553 		if (!merged)
1554 			__insert_discard_cmd(sbi, bdev,
1555 						di.lstart, di.start, di.len);
1556  next:
1557 		prev_dc = next_dc;
1558 		if (!prev_dc)
1559 			break;
1560 
1561 		node = rb_next(&prev_dc->rb_node);
1562 		next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1563 	}
1564 }
1565 
1566 #ifdef CONFIG_BLK_DEV_ZONED
__queue_zone_reset_cmd(struct f2fs_sb_info * sbi,struct block_device * bdev,block_t blkstart,block_t lblkstart,block_t blklen)1567 static void __queue_zone_reset_cmd(struct f2fs_sb_info *sbi,
1568 		struct block_device *bdev, block_t blkstart, block_t lblkstart,
1569 		block_t blklen)
1570 {
1571 	trace_f2fs_queue_reset_zone(bdev, blkstart);
1572 
1573 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1574 	__insert_discard_cmd(sbi, bdev, lblkstart, blkstart, blklen);
1575 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1576 }
1577 #endif
1578 
__queue_discard_cmd(struct f2fs_sb_info * sbi,struct block_device * bdev,block_t blkstart,block_t blklen)1579 static void __queue_discard_cmd(struct f2fs_sb_info *sbi,
1580 		struct block_device *bdev, block_t blkstart, block_t blklen)
1581 {
1582 	block_t lblkstart = blkstart;
1583 
1584 	if (!f2fs_bdev_support_discard(bdev))
1585 		return;
1586 
1587 	trace_f2fs_queue_discard(bdev, blkstart, blklen);
1588 
1589 	if (f2fs_is_multi_device(sbi)) {
1590 		int devi = f2fs_target_device_index(sbi, blkstart);
1591 
1592 		blkstart -= FDEV(devi).start_blk;
1593 	}
1594 	mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1595 	__update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1596 	mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1597 }
1598 
__issue_discard_cmd_orderly(struct f2fs_sb_info * sbi,struct discard_policy * dpolicy,int * issued)1599 static void __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1600 		struct discard_policy *dpolicy, int *issued)
1601 {
1602 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1603 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1604 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
1605 	struct discard_cmd *dc;
1606 	struct blk_plug plug;
1607 	bool io_interrupted = false;
1608 
1609 	mutex_lock(&dcc->cmd_lock);
1610 	dc = __lookup_discard_cmd_ret(&dcc->root, dcc->next_pos,
1611 				&prev_dc, &next_dc, &insert_p, &insert_parent);
1612 	if (!dc)
1613 		dc = next_dc;
1614 
1615 	blk_start_plug(&plug);
1616 
1617 	while (dc) {
1618 		struct rb_node *node;
1619 		int err = 0;
1620 
1621 		if (dc->state != D_PREP)
1622 			goto next;
1623 
1624 		if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1625 			io_interrupted = true;
1626 			break;
1627 		}
1628 
1629 		dcc->next_pos = dc->di.lstart + dc->di.len;
1630 		err = __submit_discard_cmd(sbi, dpolicy, dc, issued);
1631 
1632 		if (*issued >= dpolicy->max_requests)
1633 			break;
1634 next:
1635 		node = rb_next(&dc->rb_node);
1636 		if (err)
1637 			__remove_discard_cmd(sbi, dc);
1638 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1639 	}
1640 
1641 	blk_finish_plug(&plug);
1642 
1643 	if (!dc)
1644 		dcc->next_pos = 0;
1645 
1646 	mutex_unlock(&dcc->cmd_lock);
1647 
1648 	if (!(*issued) && io_interrupted)
1649 		*issued = -1;
1650 }
1651 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1652 					struct discard_policy *dpolicy);
1653 
__issue_discard_cmd(struct f2fs_sb_info * sbi,struct discard_policy * dpolicy)1654 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1655 					struct discard_policy *dpolicy)
1656 {
1657 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1658 	struct list_head *pend_list;
1659 	struct discard_cmd *dc, *tmp;
1660 	struct blk_plug plug;
1661 	int i, issued;
1662 	bool io_interrupted = false;
1663 
1664 	if (dpolicy->timeout)
1665 		f2fs_update_time(sbi, UMOUNT_DISCARD_TIMEOUT);
1666 
1667 retry:
1668 	issued = 0;
1669 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1670 		if (dpolicy->timeout &&
1671 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1672 			break;
1673 
1674 		if (i + 1 < dpolicy->granularity)
1675 			break;
1676 
1677 		if (i + 1 < dcc->max_ordered_discard && dpolicy->ordered) {
1678 			__issue_discard_cmd_orderly(sbi, dpolicy, &issued);
1679 			return issued;
1680 		}
1681 
1682 		pend_list = &dcc->pend_list[i];
1683 
1684 		mutex_lock(&dcc->cmd_lock);
1685 		if (list_empty(pend_list))
1686 			goto next;
1687 		if (unlikely(dcc->rbtree_check))
1688 			f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
1689 		blk_start_plug(&plug);
1690 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1691 			f2fs_bug_on(sbi, dc->state != D_PREP);
1692 
1693 			if (dpolicy->timeout &&
1694 				f2fs_time_over(sbi, UMOUNT_DISCARD_TIMEOUT))
1695 				break;
1696 
1697 			if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1698 						!is_idle(sbi, DISCARD_TIME)) {
1699 				io_interrupted = true;
1700 				break;
1701 			}
1702 
1703 			__submit_discard_cmd(sbi, dpolicy, dc, &issued);
1704 
1705 			if (issued >= dpolicy->max_requests)
1706 				break;
1707 		}
1708 		blk_finish_plug(&plug);
1709 next:
1710 		mutex_unlock(&dcc->cmd_lock);
1711 
1712 		if (issued >= dpolicy->max_requests || io_interrupted)
1713 			break;
1714 	}
1715 
1716 	if (dpolicy->type == DPOLICY_UMOUNT && issued) {
1717 		__wait_all_discard_cmd(sbi, dpolicy);
1718 		goto retry;
1719 	}
1720 
1721 	if (!issued && io_interrupted)
1722 		issued = -1;
1723 
1724 	return issued;
1725 }
1726 
__drop_discard_cmd(struct f2fs_sb_info * sbi)1727 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1728 {
1729 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1730 	struct list_head *pend_list;
1731 	struct discard_cmd *dc, *tmp;
1732 	int i;
1733 	bool dropped = false;
1734 
1735 	mutex_lock(&dcc->cmd_lock);
1736 	for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1737 		pend_list = &dcc->pend_list[i];
1738 		list_for_each_entry_safe(dc, tmp, pend_list, list) {
1739 			f2fs_bug_on(sbi, dc->state != D_PREP);
1740 			__remove_discard_cmd(sbi, dc);
1741 			dropped = true;
1742 		}
1743 	}
1744 	mutex_unlock(&dcc->cmd_lock);
1745 
1746 	return dropped;
1747 }
1748 
f2fs_drop_discard_cmd(struct f2fs_sb_info * sbi)1749 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1750 {
1751 	__drop_discard_cmd(sbi);
1752 }
1753 
__wait_one_discard_bio(struct f2fs_sb_info * sbi,struct discard_cmd * dc)1754 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1755 							struct discard_cmd *dc)
1756 {
1757 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1758 	unsigned int len = 0;
1759 
1760 	wait_for_completion_io(&dc->wait);
1761 	mutex_lock(&dcc->cmd_lock);
1762 	f2fs_bug_on(sbi, dc->state != D_DONE);
1763 	dc->ref--;
1764 	if (!dc->ref) {
1765 		if (!dc->error)
1766 			len = dc->di.len;
1767 		__remove_discard_cmd(sbi, dc);
1768 	}
1769 	mutex_unlock(&dcc->cmd_lock);
1770 
1771 	return len;
1772 }
1773 
__wait_discard_cmd_range(struct f2fs_sb_info * sbi,struct discard_policy * dpolicy,block_t start,block_t end)1774 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1775 						struct discard_policy *dpolicy,
1776 						block_t start, block_t end)
1777 {
1778 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1779 	struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1780 					&(dcc->fstrim_list) : &(dcc->wait_list);
1781 	struct discard_cmd *dc = NULL, *iter, *tmp;
1782 	unsigned int trimmed = 0;
1783 
1784 next:
1785 	dc = NULL;
1786 
1787 	mutex_lock(&dcc->cmd_lock);
1788 	list_for_each_entry_safe(iter, tmp, wait_list, list) {
1789 		if (iter->di.lstart + iter->di.len <= start ||
1790 					end <= iter->di.lstart)
1791 			continue;
1792 		if (iter->di.len < dpolicy->granularity)
1793 			continue;
1794 		if (iter->state == D_DONE && !iter->ref) {
1795 			wait_for_completion_io(&iter->wait);
1796 			if (!iter->error)
1797 				trimmed += iter->di.len;
1798 			__remove_discard_cmd(sbi, iter);
1799 		} else {
1800 			iter->ref++;
1801 			dc = iter;
1802 			break;
1803 		}
1804 	}
1805 	mutex_unlock(&dcc->cmd_lock);
1806 
1807 	if (dc) {
1808 		trimmed += __wait_one_discard_bio(sbi, dc);
1809 		goto next;
1810 	}
1811 
1812 	return trimmed;
1813 }
1814 
__wait_all_discard_cmd(struct f2fs_sb_info * sbi,struct discard_policy * dpolicy)1815 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1816 						struct discard_policy *dpolicy)
1817 {
1818 	struct discard_policy dp;
1819 	unsigned int discard_blks;
1820 
1821 	if (dpolicy)
1822 		return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1823 
1824 	/* wait all */
1825 	__init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, MIN_DISCARD_GRANULARITY);
1826 	discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1827 	__init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, MIN_DISCARD_GRANULARITY);
1828 	discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1829 
1830 	return discard_blks;
1831 }
1832 
1833 /* This should be covered by global mutex, &sit_i->sentry_lock */
f2fs_wait_discard_bio(struct f2fs_sb_info * sbi,block_t blkaddr)1834 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1835 {
1836 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1837 	struct discard_cmd *dc;
1838 	bool need_wait = false;
1839 
1840 	mutex_lock(&dcc->cmd_lock);
1841 	dc = __lookup_discard_cmd(sbi, blkaddr);
1842 #ifdef CONFIG_BLK_DEV_ZONED
1843 	if (dc && f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(dc->bdev)) {
1844 		int devi = f2fs_bdev_index(sbi, dc->bdev);
1845 
1846 		if (devi < 0) {
1847 			mutex_unlock(&dcc->cmd_lock);
1848 			return;
1849 		}
1850 
1851 		if (f2fs_blkz_is_seq(sbi, devi, dc->di.start)) {
1852 			/* force submit zone reset */
1853 			if (dc->state == D_PREP)
1854 				__submit_zone_reset_cmd(sbi, dc, REQ_SYNC,
1855 							&dcc->wait_list, NULL);
1856 			dc->ref++;
1857 			mutex_unlock(&dcc->cmd_lock);
1858 			/* wait zone reset */
1859 			__wait_one_discard_bio(sbi, dc);
1860 			return;
1861 		}
1862 	}
1863 #endif
1864 	if (dc) {
1865 		if (dc->state == D_PREP) {
1866 			__punch_discard_cmd(sbi, dc, blkaddr);
1867 		} else {
1868 			dc->ref++;
1869 			need_wait = true;
1870 		}
1871 	}
1872 	mutex_unlock(&dcc->cmd_lock);
1873 
1874 	if (need_wait)
1875 		__wait_one_discard_bio(sbi, dc);
1876 }
1877 
f2fs_stop_discard_thread(struct f2fs_sb_info * sbi)1878 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1879 {
1880 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1881 
1882 	if (dcc && dcc->f2fs_issue_discard) {
1883 		struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1884 
1885 		dcc->f2fs_issue_discard = NULL;
1886 		kthread_stop(discard_thread);
1887 	}
1888 }
1889 
1890 /**
1891  * f2fs_issue_discard_timeout() - Issue all discard cmd within UMOUNT_DISCARD_TIMEOUT
1892  * @sbi: the f2fs_sb_info data for discard cmd to issue
1893  *
1894  * When UMOUNT_DISCARD_TIMEOUT is exceeded, all remaining discard commands will be dropped
1895  *
1896  * Return true if issued all discard cmd or no discard cmd need issue, otherwise return false.
1897  */
f2fs_issue_discard_timeout(struct f2fs_sb_info * sbi)1898 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1899 {
1900 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1901 	struct discard_policy dpolicy;
1902 	bool dropped;
1903 
1904 	if (!atomic_read(&dcc->discard_cmd_cnt))
1905 		return true;
1906 
1907 	__init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1908 					dcc->discard_granularity);
1909 	__issue_discard_cmd(sbi, &dpolicy);
1910 	dropped = __drop_discard_cmd(sbi);
1911 
1912 	/* just to make sure there is no pending discard commands */
1913 	__wait_all_discard_cmd(sbi, NULL);
1914 
1915 	f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1916 	return !dropped;
1917 }
1918 
issue_discard_thread(void * data)1919 static int issue_discard_thread(void *data)
1920 {
1921 	struct f2fs_sb_info *sbi = data;
1922 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1923 	wait_queue_head_t *q = &dcc->discard_wait_queue;
1924 	struct discard_policy dpolicy;
1925 	unsigned int wait_ms = dcc->min_discard_issue_time;
1926 	int issued;
1927 
1928 	set_freezable();
1929 
1930 	do {
1931 		wait_event_freezable_timeout(*q,
1932 				kthread_should_stop() || dcc->discard_wake,
1933 				msecs_to_jiffies(wait_ms));
1934 
1935 		if (sbi->gc_mode == GC_URGENT_HIGH ||
1936 			!f2fs_available_free_memory(sbi, DISCARD_CACHE))
1937 			__init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE,
1938 						MIN_DISCARD_GRANULARITY);
1939 		else
1940 			__init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1941 						dcc->discard_granularity);
1942 
1943 		if (dcc->discard_wake)
1944 			dcc->discard_wake = false;
1945 
1946 		/* clean up pending candidates before going to sleep */
1947 		if (atomic_read(&dcc->queued_discard))
1948 			__wait_all_discard_cmd(sbi, NULL);
1949 
1950 		if (f2fs_readonly(sbi->sb))
1951 			continue;
1952 		if (kthread_should_stop())
1953 			return 0;
1954 		if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) ||
1955 			!atomic_read(&dcc->discard_cmd_cnt)) {
1956 			wait_ms = dpolicy.max_interval;
1957 			continue;
1958 		}
1959 
1960 		sb_start_intwrite(sbi->sb);
1961 
1962 		issued = __issue_discard_cmd(sbi, &dpolicy);
1963 		if (issued > 0) {
1964 			__wait_all_discard_cmd(sbi, &dpolicy);
1965 			wait_ms = dpolicy.min_interval;
1966 		} else if (issued == -1) {
1967 			wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1968 			if (!wait_ms)
1969 				wait_ms = dpolicy.mid_interval;
1970 		} else {
1971 			wait_ms = dpolicy.max_interval;
1972 		}
1973 		if (!atomic_read(&dcc->discard_cmd_cnt))
1974 			wait_ms = dpolicy.max_interval;
1975 
1976 		sb_end_intwrite(sbi->sb);
1977 
1978 	} while (!kthread_should_stop());
1979 	return 0;
1980 }
1981 
1982 #ifdef CONFIG_BLK_DEV_ZONED
__f2fs_issue_discard_zone(struct f2fs_sb_info * sbi,struct block_device * bdev,block_t blkstart,block_t blklen)1983 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1984 		struct block_device *bdev, block_t blkstart, block_t blklen)
1985 {
1986 	sector_t sector, nr_sects;
1987 	block_t lblkstart = blkstart;
1988 	int devi = 0;
1989 	u64 remainder = 0;
1990 
1991 	if (f2fs_is_multi_device(sbi)) {
1992 		devi = f2fs_target_device_index(sbi, blkstart);
1993 		if (blkstart < FDEV(devi).start_blk ||
1994 		    blkstart > FDEV(devi).end_blk) {
1995 			f2fs_err(sbi, "Invalid block %x", blkstart);
1996 			return -EIO;
1997 		}
1998 		blkstart -= FDEV(devi).start_blk;
1999 	}
2000 
2001 	/* For sequential zones, reset the zone write pointer */
2002 	if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
2003 		sector = SECTOR_FROM_BLOCK(blkstart);
2004 		nr_sects = SECTOR_FROM_BLOCK(blklen);
2005 		div64_u64_rem(sector, bdev_zone_sectors(bdev), &remainder);
2006 
2007 		if (remainder || nr_sects != bdev_zone_sectors(bdev)) {
2008 			f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
2009 				 devi, sbi->s_ndevs ? FDEV(devi).path : "",
2010 				 blkstart, blklen);
2011 			return -EIO;
2012 		}
2013 
2014 		if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING))) {
2015 			unsigned int nofs_flags;
2016 			int ret;
2017 
2018 			trace_f2fs_issue_reset_zone(bdev, blkstart);
2019 			nofs_flags = memalloc_nofs_save();
2020 			ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
2021 						sector, nr_sects);
2022 			memalloc_nofs_restore(nofs_flags);
2023 			return ret;
2024 		}
2025 
2026 		__queue_zone_reset_cmd(sbi, bdev, blkstart, lblkstart, blklen);
2027 		return 0;
2028 	}
2029 
2030 	/* For conventional zones, use regular discard if supported */
2031 	__queue_discard_cmd(sbi, bdev, lblkstart, blklen);
2032 	return 0;
2033 }
2034 #endif
2035 
__issue_discard_async(struct f2fs_sb_info * sbi,struct block_device * bdev,block_t blkstart,block_t blklen)2036 static int __issue_discard_async(struct f2fs_sb_info *sbi,
2037 		struct block_device *bdev, block_t blkstart, block_t blklen)
2038 {
2039 #ifdef CONFIG_BLK_DEV_ZONED
2040 	if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
2041 		return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
2042 #endif
2043 	__queue_discard_cmd(sbi, bdev, blkstart, blklen);
2044 	return 0;
2045 }
2046 
f2fs_issue_discard(struct f2fs_sb_info * sbi,block_t blkstart,block_t blklen)2047 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
2048 				block_t blkstart, block_t blklen)
2049 {
2050 	sector_t start = blkstart, len = 0;
2051 	struct block_device *bdev;
2052 	struct seg_entry *se;
2053 	unsigned int offset;
2054 	block_t i;
2055 	int err = 0;
2056 
2057 	bdev = f2fs_target_device(sbi, blkstart, NULL);
2058 
2059 	for (i = blkstart; i < blkstart + blklen; i++, len++) {
2060 		if (i != start) {
2061 			struct block_device *bdev2 =
2062 				f2fs_target_device(sbi, i, NULL);
2063 
2064 			if (bdev2 != bdev) {
2065 				err = __issue_discard_async(sbi, bdev,
2066 						start, len);
2067 				if (err)
2068 					return err;
2069 				bdev = bdev2;
2070 				start = i;
2071 				len = 0;
2072 			}
2073 		}
2074 
2075 		se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
2076 		offset = GET_BLKOFF_FROM_SEG0(sbi, i);
2077 
2078 		if (f2fs_block_unit_discard(sbi) &&
2079 				!f2fs_test_and_set_bit(offset, se->discard_map))
2080 			sbi->discard_blks--;
2081 	}
2082 
2083 	if (len)
2084 		err = __issue_discard_async(sbi, bdev, start, len);
2085 	return err;
2086 }
2087 
add_discard_addrs(struct f2fs_sb_info * sbi,struct cp_control * cpc,bool check_only)2088 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
2089 							bool check_only)
2090 {
2091 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2092 	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
2093 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2094 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2095 	unsigned long *discard_map = (unsigned long *)se->discard_map;
2096 	unsigned long *dmap = SIT_I(sbi)->tmp_map;
2097 	unsigned int start = 0, end = -1;
2098 	bool force = (cpc->reason & CP_DISCARD);
2099 	struct discard_entry *de = NULL;
2100 	struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
2101 	int i;
2102 
2103 	if (se->valid_blocks == BLKS_PER_SEG(sbi) ||
2104 	    !f2fs_hw_support_discard(sbi) ||
2105 	    !f2fs_block_unit_discard(sbi))
2106 		return false;
2107 
2108 	if (!force) {
2109 		if (!f2fs_realtime_discard_enable(sbi) ||
2110 			(!se->valid_blocks &&
2111 				!is_curseg(sbi, cpc->trim_start)) ||
2112 			SM_I(sbi)->dcc_info->nr_discards >=
2113 				SM_I(sbi)->dcc_info->max_discards)
2114 			return false;
2115 	}
2116 
2117 	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
2118 	for (i = 0; i < entries; i++)
2119 		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
2120 				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
2121 
2122 	while (force || SM_I(sbi)->dcc_info->nr_discards <=
2123 				SM_I(sbi)->dcc_info->max_discards) {
2124 		start = __find_rev_next_bit(dmap, BLKS_PER_SEG(sbi), end + 1);
2125 		if (start >= BLKS_PER_SEG(sbi))
2126 			break;
2127 
2128 		end = __find_rev_next_zero_bit(dmap,
2129 						BLKS_PER_SEG(sbi), start + 1);
2130 		if (force && start && end != BLKS_PER_SEG(sbi) &&
2131 		    (end - start) < cpc->trim_minlen)
2132 			continue;
2133 
2134 		if (check_only)
2135 			return true;
2136 
2137 		if (!de) {
2138 			de = f2fs_kmem_cache_alloc(discard_entry_slab,
2139 						GFP_F2FS_ZERO, true, NULL);
2140 			de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
2141 			list_add_tail(&de->list, head);
2142 		}
2143 
2144 		for (i = start; i < end; i++)
2145 			__set_bit_le(i, (void *)de->discard_map);
2146 
2147 		SM_I(sbi)->dcc_info->nr_discards += end - start;
2148 	}
2149 	return false;
2150 }
2151 
release_discard_addr(struct discard_entry * entry)2152 static void release_discard_addr(struct discard_entry *entry)
2153 {
2154 	list_del(&entry->list);
2155 	kmem_cache_free(discard_entry_slab, entry);
2156 }
2157 
f2fs_release_discard_addrs(struct f2fs_sb_info * sbi)2158 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
2159 {
2160 	struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
2161 	struct discard_entry *entry, *this;
2162 
2163 	/* drop caches */
2164 	list_for_each_entry_safe(entry, this, head, list)
2165 		release_discard_addr(entry);
2166 }
2167 
2168 /*
2169  * Should call f2fs_clear_prefree_segments after checkpoint is done.
2170  */
set_prefree_as_free_segments(struct f2fs_sb_info * sbi)2171 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
2172 {
2173 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2174 	unsigned int segno;
2175 
2176 	mutex_lock(&dirty_i->seglist_lock);
2177 	for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
2178 		__set_test_and_free(sbi, segno, false);
2179 	mutex_unlock(&dirty_i->seglist_lock);
2180 }
2181 
f2fs_clear_prefree_segments(struct f2fs_sb_info * sbi,struct cp_control * cpc)2182 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
2183 						struct cp_control *cpc)
2184 {
2185 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2186 	struct list_head *head = &dcc->entry_list;
2187 	struct discard_entry *entry, *this;
2188 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2189 	unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
2190 	unsigned int start = 0, end = -1;
2191 	unsigned int secno, start_segno;
2192 	bool force = (cpc->reason & CP_DISCARD);
2193 	bool section_alignment = F2FS_OPTION(sbi).discard_unit ==
2194 						DISCARD_UNIT_SECTION;
2195 
2196 	if (f2fs_lfs_mode(sbi) && __is_large_section(sbi))
2197 		section_alignment = true;
2198 
2199 	mutex_lock(&dirty_i->seglist_lock);
2200 
2201 	while (1) {
2202 		int i;
2203 
2204 		if (section_alignment && end != -1)
2205 			end--;
2206 		start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
2207 		if (start >= MAIN_SEGS(sbi))
2208 			break;
2209 		end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
2210 								start + 1);
2211 
2212 		if (section_alignment) {
2213 			start = rounddown(start, SEGS_PER_SEC(sbi));
2214 			end = roundup(end, SEGS_PER_SEC(sbi));
2215 		}
2216 
2217 		for (i = start; i < end; i++) {
2218 			if (test_and_clear_bit(i, prefree_map))
2219 				dirty_i->nr_dirty[PRE]--;
2220 		}
2221 
2222 		if (!f2fs_realtime_discard_enable(sbi))
2223 			continue;
2224 
2225 		if (force && start >= cpc->trim_start &&
2226 					(end - 1) <= cpc->trim_end)
2227 			continue;
2228 
2229 		/* Should cover 2MB zoned device for zone-based reset */
2230 		if (!f2fs_sb_has_blkzoned(sbi) &&
2231 		    (!f2fs_lfs_mode(sbi) || !__is_large_section(sbi))) {
2232 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
2233 				SEGS_TO_BLKS(sbi, end - start));
2234 			continue;
2235 		}
2236 next:
2237 		secno = GET_SEC_FROM_SEG(sbi, start);
2238 		start_segno = GET_SEG_FROM_SEC(sbi, secno);
2239 		if (!is_cursec(sbi, secno) &&
2240 			!get_valid_blocks(sbi, start, true))
2241 			f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
2242 						BLKS_PER_SEC(sbi));
2243 
2244 		start = start_segno + SEGS_PER_SEC(sbi);
2245 		if (start < end)
2246 			goto next;
2247 		else
2248 			end = start - 1;
2249 	}
2250 	mutex_unlock(&dirty_i->seglist_lock);
2251 
2252 	if (!f2fs_block_unit_discard(sbi))
2253 		goto wakeup;
2254 
2255 	/* send small discards */
2256 	list_for_each_entry_safe(entry, this, head, list) {
2257 		unsigned int cur_pos = 0, next_pos, len, total_len = 0;
2258 		bool is_valid = test_bit_le(0, entry->discard_map);
2259 
2260 find_next:
2261 		if (is_valid) {
2262 			next_pos = find_next_zero_bit_le(entry->discard_map,
2263 						BLKS_PER_SEG(sbi), cur_pos);
2264 			len = next_pos - cur_pos;
2265 
2266 			if (f2fs_sb_has_blkzoned(sbi) ||
2267 			    (force && len < cpc->trim_minlen))
2268 				goto skip;
2269 
2270 			f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2271 									len);
2272 			total_len += len;
2273 		} else {
2274 			next_pos = find_next_bit_le(entry->discard_map,
2275 						BLKS_PER_SEG(sbi), cur_pos);
2276 		}
2277 skip:
2278 		cur_pos = next_pos;
2279 		is_valid = !is_valid;
2280 
2281 		if (cur_pos < BLKS_PER_SEG(sbi))
2282 			goto find_next;
2283 
2284 		release_discard_addr(entry);
2285 		dcc->nr_discards -= total_len;
2286 	}
2287 
2288 wakeup:
2289 	wake_up_discard_thread(sbi, false);
2290 }
2291 
f2fs_start_discard_thread(struct f2fs_sb_info * sbi)2292 int f2fs_start_discard_thread(struct f2fs_sb_info *sbi)
2293 {
2294 	dev_t dev = sbi->sb->s_bdev->bd_dev;
2295 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2296 	int err = 0;
2297 
2298 	if (f2fs_sb_has_readonly(sbi)) {
2299 		f2fs_info(sbi,
2300 			"Skip to start discard thread for readonly image");
2301 		return 0;
2302 	}
2303 
2304 	if (!f2fs_realtime_discard_enable(sbi))
2305 		return 0;
2306 
2307 	dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2308 				"f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2309 	if (IS_ERR(dcc->f2fs_issue_discard)) {
2310 		err = PTR_ERR(dcc->f2fs_issue_discard);
2311 		dcc->f2fs_issue_discard = NULL;
2312 	}
2313 
2314 	return err;
2315 }
2316 
create_discard_cmd_control(struct f2fs_sb_info * sbi)2317 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2318 {
2319 	struct discard_cmd_control *dcc;
2320 	int err = 0, i;
2321 
2322 	if (SM_I(sbi)->dcc_info) {
2323 		dcc = SM_I(sbi)->dcc_info;
2324 		goto init_thread;
2325 	}
2326 
2327 	dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2328 	if (!dcc)
2329 		return -ENOMEM;
2330 
2331 	dcc->discard_io_aware_gran = MAX_PLIST_NUM;
2332 	dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2333 	dcc->max_ordered_discard = DEFAULT_MAX_ORDERED_DISCARD_GRANULARITY;
2334 	dcc->discard_io_aware = DPOLICY_IO_AWARE_ENABLE;
2335 	if (F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SEGMENT ||
2336 		F2FS_OPTION(sbi).discard_unit == DISCARD_UNIT_SECTION)
2337 		dcc->discard_granularity = BLKS_PER_SEG(sbi);
2338 
2339 	INIT_LIST_HEAD(&dcc->entry_list);
2340 	for (i = 0; i < MAX_PLIST_NUM; i++)
2341 		INIT_LIST_HEAD(&dcc->pend_list[i]);
2342 	INIT_LIST_HEAD(&dcc->wait_list);
2343 	INIT_LIST_HEAD(&dcc->fstrim_list);
2344 	mutex_init(&dcc->cmd_lock);
2345 	atomic_set(&dcc->issued_discard, 0);
2346 	atomic_set(&dcc->queued_discard, 0);
2347 	atomic_set(&dcc->discard_cmd_cnt, 0);
2348 	dcc->nr_discards = 0;
2349 	dcc->max_discards = SEGS_TO_BLKS(sbi, MAIN_SEGS(sbi));
2350 	dcc->max_discard_request = DEF_MAX_DISCARD_REQUEST;
2351 	dcc->min_discard_issue_time = DEF_MIN_DISCARD_ISSUE_TIME;
2352 	dcc->mid_discard_issue_time = DEF_MID_DISCARD_ISSUE_TIME;
2353 	dcc->max_discard_issue_time = DEF_MAX_DISCARD_ISSUE_TIME;
2354 	dcc->discard_urgent_util = DEF_DISCARD_URGENT_UTIL;
2355 	dcc->undiscard_blks = 0;
2356 	dcc->next_pos = 0;
2357 	dcc->root = RB_ROOT_CACHED;
2358 	dcc->rbtree_check = false;
2359 
2360 	init_waitqueue_head(&dcc->discard_wait_queue);
2361 	SM_I(sbi)->dcc_info = dcc;
2362 init_thread:
2363 	err = f2fs_start_discard_thread(sbi);
2364 	if (err) {
2365 		kfree(dcc);
2366 		SM_I(sbi)->dcc_info = NULL;
2367 	}
2368 
2369 	return err;
2370 }
2371 
destroy_discard_cmd_control(struct f2fs_sb_info * sbi)2372 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2373 {
2374 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2375 
2376 	if (!dcc)
2377 		return;
2378 
2379 	f2fs_stop_discard_thread(sbi);
2380 
2381 	/*
2382 	 * Recovery can cache discard commands, so in error path of
2383 	 * fill_super(), it needs to give a chance to handle them.
2384 	 */
2385 	f2fs_issue_discard_timeout(sbi);
2386 
2387 	kfree(dcc);
2388 	SM_I(sbi)->dcc_info = NULL;
2389 }
2390 
__mark_sit_entry_dirty(struct f2fs_sb_info * sbi,unsigned int segno)2391 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2392 {
2393 	struct sit_info *sit_i = SIT_I(sbi);
2394 
2395 	if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2396 		sit_i->dirty_sentries++;
2397 		return false;
2398 	}
2399 
2400 	return true;
2401 }
2402 
__set_sit_entry_type(struct f2fs_sb_info * sbi,int type,unsigned int segno,int modified)2403 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2404 					unsigned int segno, int modified)
2405 {
2406 	struct seg_entry *se = get_seg_entry(sbi, segno);
2407 
2408 	se->type = type;
2409 	if (modified)
2410 		__mark_sit_entry_dirty(sbi, segno);
2411 }
2412 
get_segment_mtime(struct f2fs_sb_info * sbi,block_t blkaddr)2413 static inline unsigned long long get_segment_mtime(struct f2fs_sb_info *sbi,
2414 								block_t blkaddr)
2415 {
2416 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
2417 
2418 	if (segno == NULL_SEGNO)
2419 		return 0;
2420 	return get_seg_entry(sbi, segno)->mtime;
2421 }
2422 
update_segment_mtime(struct f2fs_sb_info * sbi,block_t blkaddr,unsigned long long old_mtime)2423 static void update_segment_mtime(struct f2fs_sb_info *sbi, block_t blkaddr,
2424 						unsigned long long old_mtime)
2425 {
2426 	struct seg_entry *se;
2427 	unsigned int segno = GET_SEGNO(sbi, blkaddr);
2428 	unsigned long long ctime = get_mtime(sbi, false);
2429 	unsigned long long mtime = old_mtime ? old_mtime : ctime;
2430 
2431 	if (segno == NULL_SEGNO)
2432 		return;
2433 
2434 	se = get_seg_entry(sbi, segno);
2435 
2436 	if (!se->mtime)
2437 		se->mtime = mtime;
2438 	else
2439 		se->mtime = div_u64(se->mtime * se->valid_blocks + mtime,
2440 						se->valid_blocks + 1);
2441 
2442 	if (ctime > SIT_I(sbi)->max_mtime)
2443 		SIT_I(sbi)->max_mtime = ctime;
2444 }
2445 
2446 /*
2447  * NOTE: when updating multiple blocks at the same time, please ensure
2448  * that the consecutive input blocks belong to the same segment.
2449  */
update_sit_entry_for_release(struct f2fs_sb_info * sbi,struct seg_entry * se,unsigned int segno,block_t blkaddr,unsigned int offset,int del)2450 static int update_sit_entry_for_release(struct f2fs_sb_info *sbi, struct seg_entry *se,
2451 				unsigned int segno, block_t blkaddr, unsigned int offset, int del)
2452 {
2453 	bool exist;
2454 #ifdef CONFIG_F2FS_CHECK_FS
2455 	bool mir_exist;
2456 #endif
2457 	int i;
2458 	int del_count = -del;
2459 
2460 	f2fs_bug_on(sbi, GET_SEGNO(sbi, blkaddr) != GET_SEGNO(sbi, blkaddr + del_count - 1));
2461 
2462 	for (i = 0; i < del_count; i++) {
2463 		exist = f2fs_test_and_clear_bit(offset + i, se->cur_valid_map);
2464 #ifdef CONFIG_F2FS_CHECK_FS
2465 		mir_exist = f2fs_test_and_clear_bit(offset + i,
2466 						se->cur_valid_map_mir);
2467 		if (unlikely(exist != mir_exist)) {
2468 			f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2469 				blkaddr + i, exist);
2470 			f2fs_bug_on(sbi, 1);
2471 		}
2472 #endif
2473 		if (unlikely(!exist)) {
2474 			f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u", blkaddr + i);
2475 			f2fs_bug_on(sbi, 1);
2476 			se->valid_blocks++;
2477 			del += 1;
2478 		} else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2479 			/*
2480 			 * If checkpoints are off, we must not reuse data that
2481 			 * was used in the previous checkpoint. If it was used
2482 			 * before, we must track that to know how much space we
2483 			 * really have.
2484 			 */
2485 			if (f2fs_test_bit(offset + i, se->ckpt_valid_map)) {
2486 				spin_lock(&sbi->stat_lock);
2487 				sbi->unusable_block_count++;
2488 				spin_unlock(&sbi->stat_lock);
2489 			}
2490 		}
2491 
2492 		if (f2fs_block_unit_discard(sbi) &&
2493 				f2fs_test_and_clear_bit(offset + i, se->discard_map))
2494 			sbi->discard_blks++;
2495 
2496 		if (!f2fs_test_bit(offset + i, se->ckpt_valid_map)) {
2497 			se->ckpt_valid_blocks -= 1;
2498 			if (__is_large_section(sbi))
2499 				get_sec_entry(sbi, segno)->ckpt_valid_blocks -= 1;
2500 		}
2501 	}
2502 
2503 	if (__is_large_section(sbi))
2504 		sanity_check_valid_blocks(sbi, segno);
2505 
2506 	return del;
2507 }
2508 
update_sit_entry_for_alloc(struct f2fs_sb_info * sbi,struct seg_entry * se,unsigned int segno,block_t blkaddr,unsigned int offset,int del)2509 static int update_sit_entry_for_alloc(struct f2fs_sb_info *sbi, struct seg_entry *se,
2510 				unsigned int segno, block_t blkaddr, unsigned int offset, int del)
2511 {
2512 	bool exist;
2513 #ifdef CONFIG_F2FS_CHECK_FS
2514 	bool mir_exist;
2515 #endif
2516 
2517 	exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2518 #ifdef CONFIG_F2FS_CHECK_FS
2519 	mir_exist = f2fs_test_and_set_bit(offset,
2520 					se->cur_valid_map_mir);
2521 	if (unlikely(exist != mir_exist)) {
2522 		f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2523 			blkaddr, exist);
2524 		f2fs_bug_on(sbi, 1);
2525 	}
2526 #endif
2527 	if (unlikely(exist)) {
2528 		f2fs_err(sbi, "Bitmap was wrongly set, blk:%u", blkaddr);
2529 		f2fs_bug_on(sbi, 1);
2530 		se->valid_blocks--;
2531 		del = 0;
2532 	}
2533 
2534 	if (f2fs_block_unit_discard(sbi) &&
2535 			!f2fs_test_and_set_bit(offset, se->discard_map))
2536 		sbi->discard_blks--;
2537 
2538 	/*
2539 	 * SSR should never reuse block which is checkpointed
2540 	 * or newly invalidated.
2541 	 */
2542 	if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2543 		if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map)) {
2544 			se->ckpt_valid_blocks++;
2545 			if (__is_large_section(sbi))
2546 				get_sec_entry(sbi, segno)->ckpt_valid_blocks++;
2547 		}
2548 	}
2549 
2550 	if (!f2fs_test_bit(offset, se->ckpt_valid_map)) {
2551 		se->ckpt_valid_blocks += del;
2552 		if (__is_large_section(sbi))
2553 			get_sec_entry(sbi, segno)->ckpt_valid_blocks += del;
2554 	}
2555 
2556 	if (__is_large_section(sbi))
2557 		sanity_check_valid_blocks(sbi, segno);
2558 
2559 	return del;
2560 }
2561 
2562 /*
2563  * If releasing blocks, this function supports updating multiple consecutive blocks
2564  * at one time, but please note that these consecutive blocks need to belong to the
2565  * same segment.
2566  */
update_sit_entry(struct f2fs_sb_info * sbi,block_t blkaddr,int del)2567 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2568 {
2569 	struct seg_entry *se;
2570 	unsigned int segno, offset;
2571 	long int new_vblocks;
2572 
2573 	segno = GET_SEGNO(sbi, blkaddr);
2574 	if (segno == NULL_SEGNO)
2575 		return;
2576 
2577 	se = get_seg_entry(sbi, segno);
2578 	new_vblocks = se->valid_blocks + del;
2579 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2580 
2581 	f2fs_bug_on(sbi, (new_vblocks < 0 ||
2582 			(new_vblocks > f2fs_usable_blks_in_seg(sbi, segno))));
2583 
2584 	se->valid_blocks = new_vblocks;
2585 
2586 	/* Update valid block bitmap */
2587 	if (del > 0) {
2588 		del = update_sit_entry_for_alloc(sbi, se, segno, blkaddr, offset, del);
2589 	} else {
2590 		del = update_sit_entry_for_release(sbi, se, segno, blkaddr, offset, del);
2591 	}
2592 
2593 	__mark_sit_entry_dirty(sbi, segno);
2594 
2595 	/* update total number of valid blocks to be written in ckpt area */
2596 	SIT_I(sbi)->written_valid_blocks += del;
2597 
2598 	if (__is_large_section(sbi))
2599 		get_sec_entry(sbi, segno)->valid_blocks += del;
2600 }
2601 
f2fs_invalidate_blocks(struct f2fs_sb_info * sbi,block_t addr,unsigned int len)2602 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr,
2603 				unsigned int len)
2604 {
2605 	unsigned int segno = GET_SEGNO(sbi, addr);
2606 	struct sit_info *sit_i = SIT_I(sbi);
2607 	block_t addr_start = addr, addr_end = addr + len - 1;
2608 	unsigned int seg_num = GET_SEGNO(sbi, addr_end) - segno + 1;
2609 	unsigned int i = 1, max_blocks = sbi->blocks_per_seg, cnt;
2610 
2611 	f2fs_bug_on(sbi, addr == NULL_ADDR);
2612 	if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2613 		return;
2614 
2615 	f2fs_invalidate_internal_cache(sbi, addr, len);
2616 
2617 	/* add it into sit main buffer */
2618 	down_write(&sit_i->sentry_lock);
2619 
2620 	if (seg_num == 1)
2621 		cnt = len;
2622 	else
2623 		cnt = max_blocks - GET_BLKOFF_FROM_SEG0(sbi, addr);
2624 
2625 	do {
2626 		update_segment_mtime(sbi, addr_start, 0);
2627 		update_sit_entry(sbi, addr_start, -cnt);
2628 
2629 		/* add it into dirty seglist */
2630 		locate_dirty_segment(sbi, segno);
2631 
2632 		/* update @addr_start and @cnt and @segno */
2633 		addr_start = START_BLOCK(sbi, ++segno);
2634 		if (++i == seg_num)
2635 			cnt = GET_BLKOFF_FROM_SEG0(sbi, addr_end) + 1;
2636 		else
2637 			cnt = max_blocks;
2638 	} while (i <= seg_num);
2639 
2640 	up_write(&sit_i->sentry_lock);
2641 }
2642 
f2fs_is_checkpointed_data(struct f2fs_sb_info * sbi,block_t blkaddr)2643 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2644 {
2645 	struct sit_info *sit_i = SIT_I(sbi);
2646 	unsigned int segno, offset;
2647 	struct seg_entry *se;
2648 	bool is_cp = false;
2649 
2650 	if (!__is_valid_data_blkaddr(blkaddr))
2651 		return true;
2652 
2653 	down_read(&sit_i->sentry_lock);
2654 
2655 	segno = GET_SEGNO(sbi, blkaddr);
2656 	se = get_seg_entry(sbi, segno);
2657 	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2658 
2659 	if (f2fs_test_bit(offset, se->ckpt_valid_map))
2660 		is_cp = true;
2661 
2662 	up_read(&sit_i->sentry_lock);
2663 
2664 	return is_cp;
2665 }
2666 
f2fs_curseg_valid_blocks(struct f2fs_sb_info * sbi,int type)2667 static unsigned short f2fs_curseg_valid_blocks(struct f2fs_sb_info *sbi, int type)
2668 {
2669 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2670 
2671 	if (sbi->ckpt->alloc_type[type] == SSR)
2672 		return BLKS_PER_SEG(sbi);
2673 	return curseg->next_blkoff;
2674 }
2675 
2676 /*
2677  * Calculate the number of current summary pages for writing
2678  */
f2fs_npages_for_summary_flush(struct f2fs_sb_info * sbi,bool for_ra)2679 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2680 {
2681 	int valid_sum_count = 0;
2682 	int i, sum_in_page;
2683 
2684 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2685 		if (sbi->ckpt->alloc_type[i] != SSR && for_ra)
2686 			valid_sum_count +=
2687 				le16_to_cpu(F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2688 		else
2689 			valid_sum_count += f2fs_curseg_valid_blocks(sbi, i);
2690 	}
2691 
2692 	sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2693 			SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2694 	if (valid_sum_count <= sum_in_page)
2695 		return 1;
2696 	else if ((valid_sum_count - sum_in_page) <=
2697 		(PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2698 		return 2;
2699 	return 3;
2700 }
2701 
2702 /*
2703  * Caller should put this summary folio
2704  */
f2fs_get_sum_folio(struct f2fs_sb_info * sbi,unsigned int segno)2705 struct folio *f2fs_get_sum_folio(struct f2fs_sb_info *sbi, unsigned int segno)
2706 {
2707 	if (unlikely(f2fs_cp_error(sbi)))
2708 		return ERR_PTR(-EIO);
2709 	return f2fs_get_meta_folio_retry(sbi, GET_SUM_BLOCK(sbi, segno));
2710 }
2711 
f2fs_update_meta_page(struct f2fs_sb_info * sbi,void * src,block_t blk_addr)2712 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2713 					void *src, block_t blk_addr)
2714 {
2715 	struct folio *folio = f2fs_grab_meta_folio(sbi, blk_addr);
2716 
2717 	memcpy(folio_address(folio), src, PAGE_SIZE);
2718 	folio_mark_dirty(folio);
2719 	f2fs_folio_put(folio, true);
2720 }
2721 
write_sum_page(struct f2fs_sb_info * sbi,struct f2fs_summary_block * sum_blk,block_t blk_addr)2722 static void write_sum_page(struct f2fs_sb_info *sbi,
2723 			struct f2fs_summary_block *sum_blk, block_t blk_addr)
2724 {
2725 	f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2726 }
2727 
write_current_sum_page(struct f2fs_sb_info * sbi,int type,block_t blk_addr)2728 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2729 						int type, block_t blk_addr)
2730 {
2731 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2732 	struct folio *folio = f2fs_grab_meta_folio(sbi, blk_addr);
2733 	struct f2fs_summary_block *src = curseg->sum_blk;
2734 	struct f2fs_summary_block *dst;
2735 
2736 	dst = folio_address(folio);
2737 	memset(dst, 0, PAGE_SIZE);
2738 
2739 	mutex_lock(&curseg->curseg_mutex);
2740 
2741 	down_read(&curseg->journal_rwsem);
2742 	memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2743 	up_read(&curseg->journal_rwsem);
2744 
2745 	memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2746 	memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2747 
2748 	mutex_unlock(&curseg->curseg_mutex);
2749 
2750 	folio_mark_dirty(folio);
2751 	f2fs_folio_put(folio, true);
2752 }
2753 
is_next_segment_free(struct f2fs_sb_info * sbi,struct curseg_info * curseg)2754 static int is_next_segment_free(struct f2fs_sb_info *sbi,
2755 				struct curseg_info *curseg)
2756 {
2757 	unsigned int segno = curseg->segno + 1;
2758 	struct free_segmap_info *free_i = FREE_I(sbi);
2759 
2760 	if (segno < MAIN_SEGS(sbi) && segno % SEGS_PER_SEC(sbi))
2761 		return !test_bit(segno, free_i->free_segmap);
2762 	return 0;
2763 }
2764 
2765 /*
2766  * Find a new segment from the free segments bitmap to right order
2767  * This function should be returned with success, otherwise BUG
2768  */
get_new_segment(struct f2fs_sb_info * sbi,unsigned int * newseg,bool new_sec,bool pinning)2769 static int get_new_segment(struct f2fs_sb_info *sbi,
2770 			unsigned int *newseg, bool new_sec, bool pinning)
2771 {
2772 	struct free_segmap_info *free_i = FREE_I(sbi);
2773 	unsigned int segno, secno, zoneno;
2774 	unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2775 	unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2776 	unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2777 	bool init = true;
2778 	int i;
2779 	int ret = 0;
2780 
2781 	spin_lock(&free_i->segmap_lock);
2782 
2783 	if (time_to_inject(sbi, FAULT_NO_SEGMENT)) {
2784 		ret = -ENOSPC;
2785 		goto out_unlock;
2786 	}
2787 
2788 	if (!new_sec && ((*newseg + 1) % SEGS_PER_SEC(sbi))) {
2789 		segno = find_next_zero_bit(free_i->free_segmap,
2790 			GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2791 		if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2792 			goto got_it;
2793 	}
2794 
2795 #ifdef CONFIG_BLK_DEV_ZONED
2796 	/*
2797 	 * If we format f2fs on zoned storage, let's try to get pinned sections
2798 	 * from beginning of the storage, which should be a conventional one.
2799 	 */
2800 	if (f2fs_sb_has_blkzoned(sbi)) {
2801 		/* Prioritize writing to conventional zones */
2802 		if (sbi->blkzone_alloc_policy == BLKZONE_ALLOC_PRIOR_CONV || pinning)
2803 			segno = 0;
2804 		else
2805 			segno = max(sbi->first_seq_zone_segno, *newseg);
2806 		hint = GET_SEC_FROM_SEG(sbi, segno);
2807 	}
2808 #endif
2809 
2810 find_other_zone:
2811 	secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2812 
2813 #ifdef CONFIG_BLK_DEV_ZONED
2814 	if (secno >= MAIN_SECS(sbi) && f2fs_sb_has_blkzoned(sbi)) {
2815 		/* Write only to sequential zones */
2816 		if (sbi->blkzone_alloc_policy == BLKZONE_ALLOC_ONLY_SEQ) {
2817 			hint = GET_SEC_FROM_SEG(sbi, sbi->first_seq_zone_segno);
2818 			secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2819 		} else
2820 			secno = find_first_zero_bit(free_i->free_secmap,
2821 								MAIN_SECS(sbi));
2822 		if (secno >= MAIN_SECS(sbi)) {
2823 			ret = -ENOSPC;
2824 			f2fs_bug_on(sbi, 1);
2825 			goto out_unlock;
2826 		}
2827 	}
2828 #endif
2829 
2830 	if (secno >= MAIN_SECS(sbi)) {
2831 		secno = find_first_zero_bit(free_i->free_secmap,
2832 							MAIN_SECS(sbi));
2833 		if (secno >= MAIN_SECS(sbi)) {
2834 			ret = -ENOSPC;
2835 			f2fs_bug_on(sbi, !pinning);
2836 			goto out_unlock;
2837 		}
2838 	}
2839 	segno = GET_SEG_FROM_SEC(sbi, secno);
2840 	zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2841 
2842 	/* give up on finding another zone */
2843 	if (!init)
2844 		goto got_it;
2845 	if (sbi->secs_per_zone == 1)
2846 		goto got_it;
2847 	if (zoneno == old_zoneno)
2848 		goto got_it;
2849 	for (i = 0; i < NR_CURSEG_TYPE; i++)
2850 		if (CURSEG_I(sbi, i)->zone == zoneno)
2851 			break;
2852 
2853 	if (i < NR_CURSEG_TYPE) {
2854 		/* zone is in user, try another */
2855 		if (zoneno + 1 >= total_zones)
2856 			hint = 0;
2857 		else
2858 			hint = (zoneno + 1) * sbi->secs_per_zone;
2859 		init = false;
2860 		goto find_other_zone;
2861 	}
2862 got_it:
2863 	/* set it as dirty segment in free segmap */
2864 	if (test_bit(segno, free_i->free_segmap)) {
2865 		ret = -EFSCORRUPTED;
2866 		f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_CORRUPTED_FREE_BITMAP);
2867 		goto out_unlock;
2868 	}
2869 
2870 	/* no free section in conventional device or conventional zone */
2871 	if (new_sec && pinning &&
2872 		f2fs_is_sequential_zone_area(sbi, START_BLOCK(sbi, segno))) {
2873 		ret = -EAGAIN;
2874 		goto out_unlock;
2875 	}
2876 	__set_inuse(sbi, segno);
2877 	*newseg = segno;
2878 out_unlock:
2879 	spin_unlock(&free_i->segmap_lock);
2880 
2881 	if (ret == -ENOSPC && !pinning)
2882 		f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_NO_SEGMENT);
2883 	return ret;
2884 }
2885 
reset_curseg(struct f2fs_sb_info * sbi,int type,int modified)2886 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2887 {
2888 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2889 	struct summary_footer *sum_footer;
2890 	unsigned short seg_type = curseg->seg_type;
2891 
2892 	/* only happen when get_new_segment() fails */
2893 	if (curseg->next_segno == NULL_SEGNO)
2894 		return;
2895 
2896 	curseg->inited = true;
2897 	curseg->segno = curseg->next_segno;
2898 	curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2899 	curseg->next_blkoff = 0;
2900 	curseg->next_segno = NULL_SEGNO;
2901 
2902 	sum_footer = &(curseg->sum_blk->footer);
2903 	memset(sum_footer, 0, sizeof(struct summary_footer));
2904 
2905 	sanity_check_seg_type(sbi, seg_type);
2906 
2907 	if (IS_DATASEG(seg_type))
2908 		SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2909 	if (IS_NODESEG(seg_type))
2910 		SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2911 	__set_sit_entry_type(sbi, seg_type, curseg->segno, modified);
2912 }
2913 
__get_next_segno(struct f2fs_sb_info * sbi,int type)2914 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2915 {
2916 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2917 	unsigned short seg_type = curseg->seg_type;
2918 
2919 	sanity_check_seg_type(sbi, seg_type);
2920 	if (__is_large_section(sbi)) {
2921 		if (f2fs_need_rand_seg(sbi)) {
2922 			unsigned int hint = GET_SEC_FROM_SEG(sbi, curseg->segno);
2923 
2924 			if (GET_SEC_FROM_SEG(sbi, curseg->segno + 1) != hint)
2925 				return curseg->segno;
2926 			return get_random_u32_inclusive(curseg->segno + 1,
2927 					GET_SEG_FROM_SEC(sbi, hint + 1) - 1);
2928 		}
2929 		return curseg->segno;
2930 	} else if (f2fs_need_rand_seg(sbi)) {
2931 		return get_random_u32_below(MAIN_SECS(sbi) * SEGS_PER_SEC(sbi));
2932 	}
2933 
2934 	/* inmem log may not locate on any segment after mount */
2935 	if (!curseg->inited)
2936 		return 0;
2937 
2938 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2939 		return 0;
2940 
2941 	if (seg_type == CURSEG_HOT_DATA || IS_NODESEG(seg_type))
2942 		return 0;
2943 
2944 	if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2945 		return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2946 
2947 	/* find segments from 0 to reuse freed segments */
2948 	if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2949 		return 0;
2950 
2951 	return curseg->segno;
2952 }
2953 
reset_curseg_fields(struct curseg_info * curseg)2954 static void reset_curseg_fields(struct curseg_info *curseg)
2955 {
2956 	curseg->inited = false;
2957 	curseg->segno = NULL_SEGNO;
2958 	curseg->next_segno = 0;
2959 }
2960 
2961 /*
2962  * Allocate a current working segment.
2963  * This function always allocates a free segment in LFS manner.
2964  */
new_curseg(struct f2fs_sb_info * sbi,int type,bool new_sec)2965 static int new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2966 {
2967 	struct curseg_info *curseg = CURSEG_I(sbi, type);
2968 	unsigned int segno = curseg->segno;
2969 	bool pinning = type == CURSEG_COLD_DATA_PINNED;
2970 	int ret;
2971 
2972 	if (curseg->inited)
2973 		write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, segno));
2974 
2975 	segno = __get_next_segno(sbi, type);
2976 	ret = get_new_segment(sbi, &segno, new_sec, pinning);
2977 	if (ret) {
2978 		if (ret == -ENOSPC)
2979 			reset_curseg_fields(curseg);
2980 		return ret;
2981 	}
2982 
2983 	curseg->next_segno = segno;
2984 	reset_curseg(sbi, type, 1);
2985 	curseg->alloc_type = LFS;
2986 	if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
2987 		curseg->fragment_remained_chunk =
2988 				get_random_u32_inclusive(1, sbi->max_fragment_chunk);
2989 	return 0;
2990 }
2991 
__next_free_blkoff(struct f2fs_sb_info * sbi,int segno,block_t start)2992 static int __next_free_blkoff(struct f2fs_sb_info *sbi,
2993 					int segno, block_t start)
2994 {
2995 	struct seg_entry *se = get_seg_entry(sbi, segno);
2996 	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2997 	unsigned long *target_map = SIT_I(sbi)->tmp_map;
2998 	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2999 	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
3000 	int i;
3001 
3002 	for (i = 0; i < entries; i++)
3003 		target_map[i] = ckpt_map[i] | cur_map[i];
3004 
3005 	return __find_rev_next_zero_bit(target_map, BLKS_PER_SEG(sbi), start);
3006 }
3007 
f2fs_find_next_ssr_block(struct f2fs_sb_info * sbi,struct curseg_info * seg)3008 static int f2fs_find_next_ssr_block(struct f2fs_sb_info *sbi,
3009 		struct curseg_info *seg)
3010 {
3011 	return __next_free_blkoff(sbi, seg->segno, seg->next_blkoff + 1);
3012 }
3013 
f2fs_segment_has_free_slot(struct f2fs_sb_info * sbi,int segno)3014 bool f2fs_segment_has_free_slot(struct f2fs_sb_info *sbi, int segno)
3015 {
3016 	return __next_free_blkoff(sbi, segno, 0) < BLKS_PER_SEG(sbi);
3017 }
3018 
3019 /*
3020  * This function always allocates a used segment(from dirty seglist) by SSR
3021  * manner, so it should recover the existing segment information of valid blocks
3022  */
change_curseg(struct f2fs_sb_info * sbi,int type)3023 static int change_curseg(struct f2fs_sb_info *sbi, int type)
3024 {
3025 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3026 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3027 	unsigned int new_segno = curseg->next_segno;
3028 	struct f2fs_summary_block *sum_node;
3029 	struct folio *sum_folio;
3030 
3031 	if (curseg->inited)
3032 		write_sum_page(sbi, curseg->sum_blk, GET_SUM_BLOCK(sbi, curseg->segno));
3033 
3034 	__set_test_and_inuse(sbi, new_segno);
3035 
3036 	mutex_lock(&dirty_i->seglist_lock);
3037 	__remove_dirty_segment(sbi, new_segno, PRE);
3038 	__remove_dirty_segment(sbi, new_segno, DIRTY);
3039 	mutex_unlock(&dirty_i->seglist_lock);
3040 
3041 	reset_curseg(sbi, type, 1);
3042 	curseg->alloc_type = SSR;
3043 	curseg->next_blkoff = __next_free_blkoff(sbi, curseg->segno, 0);
3044 
3045 	sum_folio = f2fs_get_sum_folio(sbi, new_segno);
3046 	if (IS_ERR(sum_folio)) {
3047 		/* GC won't be able to use stale summary pages by cp_error */
3048 		memset(curseg->sum_blk, 0, SUM_ENTRY_SIZE);
3049 		return PTR_ERR(sum_folio);
3050 	}
3051 	sum_node = folio_address(sum_folio);
3052 	memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
3053 	f2fs_folio_put(sum_folio, true);
3054 	return 0;
3055 }
3056 
3057 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
3058 				int alloc_mode, unsigned long long age);
3059 
get_atssr_segment(struct f2fs_sb_info * sbi,int type,int target_type,int alloc_mode,unsigned long long age)3060 static int get_atssr_segment(struct f2fs_sb_info *sbi, int type,
3061 					int target_type, int alloc_mode,
3062 					unsigned long long age)
3063 {
3064 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3065 	int ret = 0;
3066 
3067 	curseg->seg_type = target_type;
3068 
3069 	if (get_ssr_segment(sbi, type, alloc_mode, age)) {
3070 		struct seg_entry *se = get_seg_entry(sbi, curseg->next_segno);
3071 
3072 		curseg->seg_type = se->type;
3073 		ret = change_curseg(sbi, type);
3074 	} else {
3075 		/* allocate cold segment by default */
3076 		curseg->seg_type = CURSEG_COLD_DATA;
3077 		ret = new_curseg(sbi, type, true);
3078 	}
3079 	stat_inc_seg_type(sbi, curseg);
3080 	return ret;
3081 }
3082 
__f2fs_init_atgc_curseg(struct f2fs_sb_info * sbi,bool force)3083 static int __f2fs_init_atgc_curseg(struct f2fs_sb_info *sbi, bool force)
3084 {
3085 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_ALL_DATA_ATGC);
3086 	int ret = 0;
3087 
3088 	if (!sbi->am.atgc_enabled && !force)
3089 		return 0;
3090 
3091 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3092 
3093 	mutex_lock(&curseg->curseg_mutex);
3094 	down_write(&SIT_I(sbi)->sentry_lock);
3095 
3096 	ret = get_atssr_segment(sbi, CURSEG_ALL_DATA_ATGC,
3097 					CURSEG_COLD_DATA, SSR, 0);
3098 
3099 	up_write(&SIT_I(sbi)->sentry_lock);
3100 	mutex_unlock(&curseg->curseg_mutex);
3101 
3102 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3103 	return ret;
3104 }
3105 
f2fs_init_inmem_curseg(struct f2fs_sb_info * sbi)3106 int f2fs_init_inmem_curseg(struct f2fs_sb_info *sbi)
3107 {
3108 	return __f2fs_init_atgc_curseg(sbi, false);
3109 }
3110 
f2fs_reinit_atgc_curseg(struct f2fs_sb_info * sbi)3111 int f2fs_reinit_atgc_curseg(struct f2fs_sb_info *sbi)
3112 {
3113 	int ret;
3114 
3115 	if (!test_opt(sbi, ATGC))
3116 		return 0;
3117 	if (sbi->am.atgc_enabled)
3118 		return 0;
3119 	if (le64_to_cpu(F2FS_CKPT(sbi)->elapsed_time) <
3120 			sbi->am.age_threshold)
3121 		return 0;
3122 
3123 	ret = __f2fs_init_atgc_curseg(sbi, true);
3124 	if (!ret) {
3125 		sbi->am.atgc_enabled = true;
3126 		f2fs_info(sbi, "reenabled age threshold GC");
3127 	}
3128 	return ret;
3129 }
3130 
__f2fs_save_inmem_curseg(struct f2fs_sb_info * sbi,int type)3131 static void __f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi, int type)
3132 {
3133 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3134 
3135 	mutex_lock(&curseg->curseg_mutex);
3136 	if (!curseg->inited)
3137 		goto out;
3138 
3139 	if (get_valid_blocks(sbi, curseg->segno, false)) {
3140 		write_sum_page(sbi, curseg->sum_blk,
3141 				GET_SUM_BLOCK(sbi, curseg->segno));
3142 	} else {
3143 		mutex_lock(&DIRTY_I(sbi)->seglist_lock);
3144 		__set_test_and_free(sbi, curseg->segno, true);
3145 		mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
3146 	}
3147 out:
3148 	mutex_unlock(&curseg->curseg_mutex);
3149 }
3150 
f2fs_save_inmem_curseg(struct f2fs_sb_info * sbi)3151 void f2fs_save_inmem_curseg(struct f2fs_sb_info *sbi)
3152 {
3153 	__f2fs_save_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
3154 
3155 	if (sbi->am.atgc_enabled)
3156 		__f2fs_save_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
3157 }
3158 
__f2fs_restore_inmem_curseg(struct f2fs_sb_info * sbi,int type)3159 static void __f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi, int type)
3160 {
3161 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3162 
3163 	mutex_lock(&curseg->curseg_mutex);
3164 	if (!curseg->inited)
3165 		goto out;
3166 	if (get_valid_blocks(sbi, curseg->segno, false))
3167 		goto out;
3168 
3169 	mutex_lock(&DIRTY_I(sbi)->seglist_lock);
3170 	__set_test_and_inuse(sbi, curseg->segno);
3171 	mutex_unlock(&DIRTY_I(sbi)->seglist_lock);
3172 out:
3173 	mutex_unlock(&curseg->curseg_mutex);
3174 }
3175 
f2fs_restore_inmem_curseg(struct f2fs_sb_info * sbi)3176 void f2fs_restore_inmem_curseg(struct f2fs_sb_info *sbi)
3177 {
3178 	__f2fs_restore_inmem_curseg(sbi, CURSEG_COLD_DATA_PINNED);
3179 
3180 	if (sbi->am.atgc_enabled)
3181 		__f2fs_restore_inmem_curseg(sbi, CURSEG_ALL_DATA_ATGC);
3182 }
3183 
get_ssr_segment(struct f2fs_sb_info * sbi,int type,int alloc_mode,unsigned long long age)3184 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type,
3185 				int alloc_mode, unsigned long long age)
3186 {
3187 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3188 	unsigned segno = NULL_SEGNO;
3189 	unsigned short seg_type = curseg->seg_type;
3190 	int i, cnt;
3191 	bool reversed = false;
3192 
3193 	sanity_check_seg_type(sbi, seg_type);
3194 
3195 	/* f2fs_need_SSR() already forces to do this */
3196 	if (!f2fs_get_victim(sbi, &segno, BG_GC, seg_type,
3197 				alloc_mode, age, false)) {
3198 		curseg->next_segno = segno;
3199 		return 1;
3200 	}
3201 
3202 	/* For node segments, let's do SSR more intensively */
3203 	if (IS_NODESEG(seg_type)) {
3204 		if (seg_type >= CURSEG_WARM_NODE) {
3205 			reversed = true;
3206 			i = CURSEG_COLD_NODE;
3207 		} else {
3208 			i = CURSEG_HOT_NODE;
3209 		}
3210 		cnt = NR_CURSEG_NODE_TYPE;
3211 	} else {
3212 		if (seg_type >= CURSEG_WARM_DATA) {
3213 			reversed = true;
3214 			i = CURSEG_COLD_DATA;
3215 		} else {
3216 			i = CURSEG_HOT_DATA;
3217 		}
3218 		cnt = NR_CURSEG_DATA_TYPE;
3219 	}
3220 
3221 	for (; cnt-- > 0; reversed ? i-- : i++) {
3222 		if (i == seg_type)
3223 			continue;
3224 		if (!f2fs_get_victim(sbi, &segno, BG_GC, i,
3225 					alloc_mode, age, false)) {
3226 			curseg->next_segno = segno;
3227 			return 1;
3228 		}
3229 	}
3230 
3231 	/* find valid_blocks=0 in dirty list */
3232 	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
3233 		segno = get_free_segment(sbi);
3234 		if (segno != NULL_SEGNO) {
3235 			curseg->next_segno = segno;
3236 			return 1;
3237 		}
3238 	}
3239 	return 0;
3240 }
3241 
need_new_seg(struct f2fs_sb_info * sbi,int type)3242 static bool need_new_seg(struct f2fs_sb_info *sbi, int type)
3243 {
3244 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3245 
3246 	if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
3247 	    curseg->seg_type == CURSEG_WARM_NODE)
3248 		return true;
3249 	if (curseg->alloc_type == LFS && is_next_segment_free(sbi, curseg) &&
3250 	    likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
3251 		return true;
3252 	if (!f2fs_need_SSR(sbi) || !get_ssr_segment(sbi, type, SSR, 0))
3253 		return true;
3254 	return false;
3255 }
3256 
f2fs_allocate_segment_for_resize(struct f2fs_sb_info * sbi,int type,unsigned int start,unsigned int end)3257 int f2fs_allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
3258 					unsigned int start, unsigned int end)
3259 {
3260 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3261 	unsigned int segno;
3262 	int ret = 0;
3263 
3264 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3265 	mutex_lock(&curseg->curseg_mutex);
3266 	down_write(&SIT_I(sbi)->sentry_lock);
3267 
3268 	segno = CURSEG_I(sbi, type)->segno;
3269 	if (segno < start || segno > end)
3270 		goto unlock;
3271 
3272 	if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type, SSR, 0))
3273 		ret = change_curseg(sbi, type);
3274 	else
3275 		ret = new_curseg(sbi, type, true);
3276 
3277 	stat_inc_seg_type(sbi, curseg);
3278 
3279 	locate_dirty_segment(sbi, segno);
3280 unlock:
3281 	up_write(&SIT_I(sbi)->sentry_lock);
3282 
3283 	if (segno != curseg->segno)
3284 		f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
3285 			    type, segno, curseg->segno);
3286 
3287 	mutex_unlock(&curseg->curseg_mutex);
3288 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3289 	return ret;
3290 }
3291 
__allocate_new_segment(struct f2fs_sb_info * sbi,int type,bool new_sec,bool force)3292 static int __allocate_new_segment(struct f2fs_sb_info *sbi, int type,
3293 						bool new_sec, bool force)
3294 {
3295 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3296 	unsigned int old_segno;
3297 	int err = 0;
3298 
3299 	if (type == CURSEG_COLD_DATA_PINNED && !curseg->inited)
3300 		goto allocate;
3301 
3302 	if (!force && curseg->inited &&
3303 	    !curseg->next_blkoff &&
3304 	    !get_valid_blocks(sbi, curseg->segno, new_sec) &&
3305 	    !get_ckpt_valid_blocks(sbi, curseg->segno, new_sec))
3306 		return 0;
3307 
3308 allocate:
3309 	old_segno = curseg->segno;
3310 	err = new_curseg(sbi, type, true);
3311 	if (err)
3312 		return err;
3313 	stat_inc_seg_type(sbi, curseg);
3314 	locate_dirty_segment(sbi, old_segno);
3315 	return 0;
3316 }
3317 
f2fs_allocate_new_section(struct f2fs_sb_info * sbi,int type,bool force)3318 int f2fs_allocate_new_section(struct f2fs_sb_info *sbi, int type, bool force)
3319 {
3320 	int ret;
3321 
3322 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3323 	down_write(&SIT_I(sbi)->sentry_lock);
3324 	ret = __allocate_new_segment(sbi, type, true, force);
3325 	up_write(&SIT_I(sbi)->sentry_lock);
3326 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3327 
3328 	return ret;
3329 }
3330 
f2fs_allocate_pinning_section(struct f2fs_sb_info * sbi)3331 int f2fs_allocate_pinning_section(struct f2fs_sb_info *sbi)
3332 {
3333 	int err;
3334 	bool gc_required = true;
3335 
3336 retry:
3337 	f2fs_lock_op(sbi);
3338 	err = f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3339 	f2fs_unlock_op(sbi);
3340 
3341 	if (f2fs_sb_has_blkzoned(sbi) && err == -EAGAIN && gc_required) {
3342 		f2fs_down_write(&sbi->gc_lock);
3343 		err = f2fs_gc_range(sbi, 0, sbi->first_seq_zone_segno - 1,
3344 				true, ZONED_PIN_SEC_REQUIRED_COUNT);
3345 		f2fs_up_write(&sbi->gc_lock);
3346 
3347 		gc_required = false;
3348 		if (!err)
3349 			goto retry;
3350 	}
3351 
3352 	return err;
3353 }
3354 
f2fs_allocate_new_segments(struct f2fs_sb_info * sbi)3355 int f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
3356 {
3357 	int i;
3358 	int err = 0;
3359 
3360 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3361 	down_write(&SIT_I(sbi)->sentry_lock);
3362 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
3363 		err += __allocate_new_segment(sbi, i, false, false);
3364 	up_write(&SIT_I(sbi)->sentry_lock);
3365 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3366 
3367 	return err;
3368 }
3369 
f2fs_exist_trim_candidates(struct f2fs_sb_info * sbi,struct cp_control * cpc)3370 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
3371 						struct cp_control *cpc)
3372 {
3373 	__u64 trim_start = cpc->trim_start;
3374 	bool has_candidate = false;
3375 
3376 	down_write(&SIT_I(sbi)->sentry_lock);
3377 	for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
3378 		if (add_discard_addrs(sbi, cpc, true)) {
3379 			has_candidate = true;
3380 			break;
3381 		}
3382 	}
3383 	up_write(&SIT_I(sbi)->sentry_lock);
3384 
3385 	cpc->trim_start = trim_start;
3386 	return has_candidate;
3387 }
3388 
__issue_discard_cmd_range(struct f2fs_sb_info * sbi,struct discard_policy * dpolicy,unsigned int start,unsigned int end)3389 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
3390 					struct discard_policy *dpolicy,
3391 					unsigned int start, unsigned int end)
3392 {
3393 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
3394 	struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
3395 	struct rb_node **insert_p = NULL, *insert_parent = NULL;
3396 	struct discard_cmd *dc;
3397 	struct blk_plug plug;
3398 	int issued;
3399 	unsigned int trimmed = 0;
3400 
3401 next:
3402 	issued = 0;
3403 
3404 	mutex_lock(&dcc->cmd_lock);
3405 	if (unlikely(dcc->rbtree_check))
3406 		f2fs_bug_on(sbi, !f2fs_check_discard_tree(sbi));
3407 
3408 	dc = __lookup_discard_cmd_ret(&dcc->root, start,
3409 				&prev_dc, &next_dc, &insert_p, &insert_parent);
3410 	if (!dc)
3411 		dc = next_dc;
3412 
3413 	blk_start_plug(&plug);
3414 
3415 	while (dc && dc->di.lstart <= end) {
3416 		struct rb_node *node;
3417 		int err = 0;
3418 
3419 		if (dc->di.len < dpolicy->granularity)
3420 			goto skip;
3421 
3422 		if (dc->state != D_PREP) {
3423 			list_move_tail(&dc->list, &dcc->fstrim_list);
3424 			goto skip;
3425 		}
3426 
3427 		err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
3428 
3429 		if (issued >= dpolicy->max_requests) {
3430 			start = dc->di.lstart + dc->di.len;
3431 
3432 			if (err)
3433 				__remove_discard_cmd(sbi, dc);
3434 
3435 			blk_finish_plug(&plug);
3436 			mutex_unlock(&dcc->cmd_lock);
3437 			trimmed += __wait_all_discard_cmd(sbi, NULL);
3438 			f2fs_io_schedule_timeout(DEFAULT_IO_TIMEOUT);
3439 			goto next;
3440 		}
3441 skip:
3442 		node = rb_next(&dc->rb_node);
3443 		if (err)
3444 			__remove_discard_cmd(sbi, dc);
3445 		dc = rb_entry_safe(node, struct discard_cmd, rb_node);
3446 
3447 		if (fatal_signal_pending(current))
3448 			break;
3449 	}
3450 
3451 	blk_finish_plug(&plug);
3452 	mutex_unlock(&dcc->cmd_lock);
3453 
3454 	return trimmed;
3455 }
3456 
f2fs_trim_fs(struct f2fs_sb_info * sbi,struct fstrim_range * range)3457 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
3458 {
3459 	__u64 start = F2FS_BYTES_TO_BLK(range->start);
3460 	__u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
3461 	unsigned int start_segno, end_segno;
3462 	block_t start_block, end_block;
3463 	struct cp_control cpc;
3464 	struct discard_policy dpolicy;
3465 	unsigned long long trimmed = 0;
3466 	int err = 0;
3467 	bool need_align = f2fs_lfs_mode(sbi) && __is_large_section(sbi);
3468 
3469 	if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
3470 		return -EINVAL;
3471 
3472 	if (end < MAIN_BLKADDR(sbi))
3473 		goto out;
3474 
3475 	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
3476 		f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
3477 		return -EFSCORRUPTED;
3478 	}
3479 
3480 	/* start/end segment number in main_area */
3481 	start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
3482 	end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
3483 						GET_SEGNO(sbi, end);
3484 	if (need_align) {
3485 		start_segno = rounddown(start_segno, SEGS_PER_SEC(sbi));
3486 		end_segno = roundup(end_segno + 1, SEGS_PER_SEC(sbi)) - 1;
3487 	}
3488 
3489 	cpc.reason = CP_DISCARD;
3490 	cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
3491 	cpc.trim_start = start_segno;
3492 	cpc.trim_end = end_segno;
3493 
3494 	if (sbi->discard_blks == 0)
3495 		goto out;
3496 
3497 	f2fs_down_write(&sbi->gc_lock);
3498 	stat_inc_cp_call_count(sbi, TOTAL_CALL);
3499 	err = f2fs_write_checkpoint(sbi, &cpc);
3500 	f2fs_up_write(&sbi->gc_lock);
3501 	if (err)
3502 		goto out;
3503 
3504 	/*
3505 	 * We filed discard candidates, but actually we don't need to wait for
3506 	 * all of them, since they'll be issued in idle time along with runtime
3507 	 * discard option. User configuration looks like using runtime discard
3508 	 * or periodic fstrim instead of it.
3509 	 */
3510 	if (f2fs_realtime_discard_enable(sbi))
3511 		goto out;
3512 
3513 	start_block = START_BLOCK(sbi, start_segno);
3514 	end_block = START_BLOCK(sbi, end_segno + 1);
3515 
3516 	__init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
3517 	trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
3518 					start_block, end_block);
3519 
3520 	trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
3521 					start_block, end_block);
3522 out:
3523 	if (!err)
3524 		range->len = F2FS_BLK_TO_BYTES(trimmed);
3525 	return err;
3526 }
3527 
f2fs_rw_hint_to_seg_type(struct f2fs_sb_info * sbi,enum rw_hint hint)3528 int f2fs_rw_hint_to_seg_type(struct f2fs_sb_info *sbi, enum rw_hint hint)
3529 {
3530 	if (F2FS_OPTION(sbi).active_logs == 2)
3531 		return CURSEG_HOT_DATA;
3532 	else if (F2FS_OPTION(sbi).active_logs == 4)
3533 		return CURSEG_COLD_DATA;
3534 
3535 	/* active_log == 6 */
3536 	switch (hint) {
3537 	case WRITE_LIFE_SHORT:
3538 		return CURSEG_HOT_DATA;
3539 	case WRITE_LIFE_EXTREME:
3540 		return CURSEG_COLD_DATA;
3541 	default:
3542 		return CURSEG_WARM_DATA;
3543 	}
3544 }
3545 
3546 /*
3547  * This returns write hints for each segment type. This hints will be
3548  * passed down to block layer as below by default.
3549  *
3550  * User                  F2FS                     Block
3551  * ----                  ----                     -----
3552  *                       META                     WRITE_LIFE_NONE|REQ_META
3553  *                       HOT_NODE                 WRITE_LIFE_NONE
3554  *                       WARM_NODE                WRITE_LIFE_MEDIUM
3555  *                       COLD_NODE                WRITE_LIFE_LONG
3556  * ioctl(COLD)           COLD_DATA                WRITE_LIFE_EXTREME
3557  * extension list        "                        "
3558  *
3559  * -- buffered io
3560  *                       COLD_DATA                WRITE_LIFE_EXTREME
3561  *                       HOT_DATA                 WRITE_LIFE_SHORT
3562  *                       WARM_DATA                WRITE_LIFE_NOT_SET
3563  *
3564  * -- direct io
3565  * WRITE_LIFE_EXTREME    COLD_DATA                WRITE_LIFE_EXTREME
3566  * WRITE_LIFE_SHORT      HOT_DATA                 WRITE_LIFE_SHORT
3567  * WRITE_LIFE_NOT_SET    WARM_DATA                WRITE_LIFE_NOT_SET
3568  * WRITE_LIFE_NONE       "                        WRITE_LIFE_NONE
3569  * WRITE_LIFE_MEDIUM     "                        WRITE_LIFE_MEDIUM
3570  * WRITE_LIFE_LONG       "                        WRITE_LIFE_LONG
3571  */
f2fs_io_type_to_rw_hint(struct f2fs_sb_info * sbi,enum page_type type,enum temp_type temp)3572 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
3573 				enum page_type type, enum temp_type temp)
3574 {
3575 	switch (type) {
3576 	case DATA:
3577 		switch (temp) {
3578 		case WARM:
3579 			return WRITE_LIFE_NOT_SET;
3580 		case HOT:
3581 			return WRITE_LIFE_SHORT;
3582 		case COLD:
3583 			return WRITE_LIFE_EXTREME;
3584 		default:
3585 			return WRITE_LIFE_NONE;
3586 		}
3587 	case NODE:
3588 		switch (temp) {
3589 		case WARM:
3590 			return WRITE_LIFE_MEDIUM;
3591 		case HOT:
3592 			return WRITE_LIFE_NONE;
3593 		case COLD:
3594 			return WRITE_LIFE_LONG;
3595 		default:
3596 			return WRITE_LIFE_NONE;
3597 		}
3598 	case META:
3599 		return WRITE_LIFE_NONE;
3600 	default:
3601 		return WRITE_LIFE_NONE;
3602 	}
3603 }
3604 
__get_segment_type_2(struct f2fs_io_info * fio)3605 static int __get_segment_type_2(struct f2fs_io_info *fio)
3606 {
3607 	if (fio->type == DATA)
3608 		return CURSEG_HOT_DATA;
3609 	else
3610 		return CURSEG_HOT_NODE;
3611 }
3612 
__get_segment_type_4(struct f2fs_io_info * fio)3613 static int __get_segment_type_4(struct f2fs_io_info *fio)
3614 {
3615 	if (fio->type == DATA) {
3616 		struct inode *inode = fio_inode(fio);
3617 
3618 		if (S_ISDIR(inode->i_mode))
3619 			return CURSEG_HOT_DATA;
3620 		else
3621 			return CURSEG_COLD_DATA;
3622 	} else {
3623 		if (IS_DNODE(fio->folio) && is_cold_node(fio->folio))
3624 			return CURSEG_WARM_NODE;
3625 		else
3626 			return CURSEG_COLD_NODE;
3627 	}
3628 }
3629 
__get_age_segment_type(struct inode * inode,pgoff_t pgofs)3630 static int __get_age_segment_type(struct inode *inode, pgoff_t pgofs)
3631 {
3632 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3633 	struct extent_info ei = {};
3634 
3635 	if (f2fs_lookup_age_extent_cache(inode, pgofs, &ei)) {
3636 		if (!ei.age)
3637 			return NO_CHECK_TYPE;
3638 		if (ei.age <= sbi->hot_data_age_threshold)
3639 			return CURSEG_HOT_DATA;
3640 		if (ei.age <= sbi->warm_data_age_threshold)
3641 			return CURSEG_WARM_DATA;
3642 		return CURSEG_COLD_DATA;
3643 	}
3644 	return NO_CHECK_TYPE;
3645 }
3646 
__get_segment_type_6(struct f2fs_io_info * fio)3647 static int __get_segment_type_6(struct f2fs_io_info *fio)
3648 {
3649 	if (fio->type == DATA) {
3650 		struct inode *inode = fio_inode(fio);
3651 		int type;
3652 
3653 		if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
3654 			return CURSEG_COLD_DATA_PINNED;
3655 
3656 		if (page_private_gcing(fio->page)) {
3657 			if (fio->sbi->am.atgc_enabled &&
3658 				(fio->io_type == FS_DATA_IO) &&
3659 				(fio->sbi->gc_mode != GC_URGENT_HIGH) &&
3660 				__is_valid_data_blkaddr(fio->old_blkaddr) &&
3661 				!is_inode_flag_set(inode, FI_OPU_WRITE))
3662 				return CURSEG_ALL_DATA_ATGC;
3663 			else
3664 				return CURSEG_COLD_DATA;
3665 		}
3666 		if (file_is_cold(inode) || f2fs_need_compress_data(inode))
3667 			return CURSEG_COLD_DATA;
3668 
3669 		type = __get_age_segment_type(inode, fio->folio->index);
3670 		if (type != NO_CHECK_TYPE)
3671 			return type;
3672 
3673 		if (file_is_hot(inode) ||
3674 				is_inode_flag_set(inode, FI_HOT_DATA) ||
3675 				f2fs_is_cow_file(inode))
3676 			return CURSEG_HOT_DATA;
3677 		return f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode),
3678 						inode->i_write_hint);
3679 	} else {
3680 		if (IS_DNODE(fio->folio))
3681 			return is_cold_node(fio->folio) ? CURSEG_WARM_NODE :
3682 						CURSEG_HOT_NODE;
3683 		return CURSEG_COLD_NODE;
3684 	}
3685 }
3686 
f2fs_get_segment_temp(struct f2fs_sb_info * sbi,enum log_type type)3687 enum temp_type f2fs_get_segment_temp(struct f2fs_sb_info *sbi,
3688 						enum log_type type)
3689 {
3690 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3691 	enum temp_type temp = COLD;
3692 
3693 	switch (curseg->seg_type) {
3694 	case CURSEG_HOT_NODE:
3695 	case CURSEG_HOT_DATA:
3696 		temp = HOT;
3697 		break;
3698 	case CURSEG_WARM_NODE:
3699 	case CURSEG_WARM_DATA:
3700 		temp = WARM;
3701 		break;
3702 	case CURSEG_COLD_NODE:
3703 	case CURSEG_COLD_DATA:
3704 		temp = COLD;
3705 		break;
3706 	default:
3707 		f2fs_bug_on(sbi, 1);
3708 	}
3709 
3710 	return temp;
3711 }
3712 
__get_segment_type(struct f2fs_io_info * fio)3713 static int __get_segment_type(struct f2fs_io_info *fio)
3714 {
3715 	enum log_type type = CURSEG_HOT_DATA;
3716 
3717 	switch (F2FS_OPTION(fio->sbi).active_logs) {
3718 	case 2:
3719 		type = __get_segment_type_2(fio);
3720 		break;
3721 	case 4:
3722 		type = __get_segment_type_4(fio);
3723 		break;
3724 	case 6:
3725 		type = __get_segment_type_6(fio);
3726 		break;
3727 	default:
3728 		f2fs_bug_on(fio->sbi, true);
3729 	}
3730 
3731 	fio->temp = f2fs_get_segment_temp(fio->sbi, type);
3732 
3733 	return type;
3734 }
3735 
f2fs_randomize_chunk(struct f2fs_sb_info * sbi,struct curseg_info * seg)3736 static void f2fs_randomize_chunk(struct f2fs_sb_info *sbi,
3737 		struct curseg_info *seg)
3738 {
3739 	/* To allocate block chunks in different sizes, use random number */
3740 	if (--seg->fragment_remained_chunk > 0)
3741 		return;
3742 
3743 	seg->fragment_remained_chunk =
3744 		get_random_u32_inclusive(1, sbi->max_fragment_chunk);
3745 	seg->next_blkoff +=
3746 		get_random_u32_inclusive(1, sbi->max_fragment_hole);
3747 }
3748 
f2fs_allocate_data_block(struct f2fs_sb_info * sbi,struct folio * folio,block_t old_blkaddr,block_t * new_blkaddr,struct f2fs_summary * sum,int type,struct f2fs_io_info * fio)3749 int f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct folio *folio,
3750 		block_t old_blkaddr, block_t *new_blkaddr,
3751 		struct f2fs_summary *sum, int type,
3752 		struct f2fs_io_info *fio)
3753 {
3754 	struct sit_info *sit_i = SIT_I(sbi);
3755 	struct curseg_info *curseg = CURSEG_I(sbi, type);
3756 	unsigned long long old_mtime;
3757 	bool from_gc = (type == CURSEG_ALL_DATA_ATGC);
3758 	struct seg_entry *se = NULL;
3759 	bool segment_full = false;
3760 	int ret = 0;
3761 
3762 	f2fs_down_read(&SM_I(sbi)->curseg_lock);
3763 
3764 	mutex_lock(&curseg->curseg_mutex);
3765 	down_write(&sit_i->sentry_lock);
3766 
3767 	if (curseg->segno == NULL_SEGNO) {
3768 		ret = -ENOSPC;
3769 		goto out_err;
3770 	}
3771 
3772 	if (from_gc) {
3773 		f2fs_bug_on(sbi, GET_SEGNO(sbi, old_blkaddr) == NULL_SEGNO);
3774 		se = get_seg_entry(sbi, GET_SEGNO(sbi, old_blkaddr));
3775 		sanity_check_seg_type(sbi, se->type);
3776 		f2fs_bug_on(sbi, IS_NODESEG(se->type));
3777 	}
3778 	*new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3779 
3780 	f2fs_bug_on(sbi, curseg->next_blkoff >= BLKS_PER_SEG(sbi));
3781 
3782 	f2fs_wait_discard_bio(sbi, *new_blkaddr);
3783 
3784 	curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
3785 	if (curseg->alloc_type == SSR) {
3786 		curseg->next_blkoff = f2fs_find_next_ssr_block(sbi, curseg);
3787 	} else {
3788 		curseg->next_blkoff++;
3789 		if (F2FS_OPTION(sbi).fs_mode == FS_MODE_FRAGMENT_BLK)
3790 			f2fs_randomize_chunk(sbi, curseg);
3791 	}
3792 	if (curseg->next_blkoff >= f2fs_usable_blks_in_seg(sbi, curseg->segno))
3793 		segment_full = true;
3794 	stat_inc_block_count(sbi, curseg);
3795 
3796 	if (from_gc) {
3797 		old_mtime = get_segment_mtime(sbi, old_blkaddr);
3798 	} else {
3799 		update_segment_mtime(sbi, old_blkaddr, 0);
3800 		old_mtime = 0;
3801 	}
3802 	update_segment_mtime(sbi, *new_blkaddr, old_mtime);
3803 
3804 	/*
3805 	 * SIT information should be updated before segment allocation,
3806 	 * since SSR needs latest valid block information.
3807 	 */
3808 	update_sit_entry(sbi, *new_blkaddr, 1);
3809 	update_sit_entry(sbi, old_blkaddr, -1);
3810 
3811 	/*
3812 	 * If the current segment is full, flush it out and replace it with a
3813 	 * new segment.
3814 	 */
3815 	if (segment_full) {
3816 		if (type == CURSEG_COLD_DATA_PINNED &&
3817 		    !((curseg->segno + 1) % sbi->segs_per_sec)) {
3818 			write_sum_page(sbi, curseg->sum_blk,
3819 					GET_SUM_BLOCK(sbi, curseg->segno));
3820 			reset_curseg_fields(curseg);
3821 			goto skip_new_segment;
3822 		}
3823 
3824 		if (from_gc) {
3825 			ret = get_atssr_segment(sbi, type, se->type,
3826 						AT_SSR, se->mtime);
3827 		} else {
3828 			if (need_new_seg(sbi, type))
3829 				ret = new_curseg(sbi, type, false);
3830 			else
3831 				ret = change_curseg(sbi, type);
3832 			stat_inc_seg_type(sbi, curseg);
3833 		}
3834 
3835 		if (ret)
3836 			goto out_err;
3837 	}
3838 
3839 skip_new_segment:
3840 	/*
3841 	 * segment dirty status should be updated after segment allocation,
3842 	 * so we just need to update status only one time after previous
3843 	 * segment being closed.
3844 	 */
3845 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3846 	locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3847 
3848 	if (IS_DATASEG(curseg->seg_type))
3849 		atomic64_inc(&sbi->allocated_data_blocks);
3850 
3851 	up_write(&sit_i->sentry_lock);
3852 
3853 	if (folio && IS_NODESEG(curseg->seg_type)) {
3854 		fill_node_footer_blkaddr(folio, NEXT_FREE_BLKADDR(sbi, curseg));
3855 
3856 		f2fs_inode_chksum_set(sbi, folio);
3857 	}
3858 
3859 	if (fio) {
3860 		struct f2fs_bio_info *io;
3861 
3862 		INIT_LIST_HEAD(&fio->list);
3863 		fio->in_list = 1;
3864 		io = sbi->write_io[fio->type] + fio->temp;
3865 		spin_lock(&io->io_lock);
3866 		list_add_tail(&fio->list, &io->io_list);
3867 		spin_unlock(&io->io_lock);
3868 	}
3869 
3870 	mutex_unlock(&curseg->curseg_mutex);
3871 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3872 	return 0;
3873 
3874 out_err:
3875 	*new_blkaddr = NULL_ADDR;
3876 	up_write(&sit_i->sentry_lock);
3877 	mutex_unlock(&curseg->curseg_mutex);
3878 	f2fs_up_read(&SM_I(sbi)->curseg_lock);
3879 	return ret;
3880 }
3881 
f2fs_update_device_state(struct f2fs_sb_info * sbi,nid_t ino,block_t blkaddr,unsigned int blkcnt)3882 void f2fs_update_device_state(struct f2fs_sb_info *sbi, nid_t ino,
3883 					block_t blkaddr, unsigned int blkcnt)
3884 {
3885 	if (!f2fs_is_multi_device(sbi))
3886 		return;
3887 
3888 	while (1) {
3889 		unsigned int devidx = f2fs_target_device_index(sbi, blkaddr);
3890 		unsigned int blks = FDEV(devidx).end_blk - blkaddr + 1;
3891 
3892 		/* update device state for fsync */
3893 		f2fs_set_dirty_device(sbi, ino, devidx, FLUSH_INO);
3894 
3895 		/* update device state for checkpoint */
3896 		if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3897 			spin_lock(&sbi->dev_lock);
3898 			f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3899 			spin_unlock(&sbi->dev_lock);
3900 		}
3901 
3902 		if (blkcnt <= blks)
3903 			break;
3904 		blkcnt -= blks;
3905 		blkaddr += blks;
3906 	}
3907 }
3908 
log_type_to_seg_type(enum log_type type)3909 static int log_type_to_seg_type(enum log_type type)
3910 {
3911 	int seg_type = CURSEG_COLD_DATA;
3912 
3913 	switch (type) {
3914 	case CURSEG_HOT_DATA:
3915 	case CURSEG_WARM_DATA:
3916 	case CURSEG_COLD_DATA:
3917 	case CURSEG_HOT_NODE:
3918 	case CURSEG_WARM_NODE:
3919 	case CURSEG_COLD_NODE:
3920 		seg_type = (int)type;
3921 		break;
3922 	case CURSEG_COLD_DATA_PINNED:
3923 	case CURSEG_ALL_DATA_ATGC:
3924 		seg_type = CURSEG_COLD_DATA;
3925 		break;
3926 	default:
3927 		break;
3928 	}
3929 	return seg_type;
3930 }
3931 
do_write_page(struct f2fs_summary * sum,struct f2fs_io_info * fio)3932 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3933 {
3934 	struct folio *folio = fio->folio;
3935 	enum log_type type = __get_segment_type(fio);
3936 	int seg_type = log_type_to_seg_type(type);
3937 	bool keep_order = (f2fs_lfs_mode(fio->sbi) &&
3938 				seg_type == CURSEG_COLD_DATA);
3939 
3940 	if (keep_order)
3941 		f2fs_down_read(&fio->sbi->io_order_lock);
3942 
3943 	if (f2fs_allocate_data_block(fio->sbi, folio, fio->old_blkaddr,
3944 			&fio->new_blkaddr, sum, type, fio)) {
3945 		if (fscrypt_inode_uses_fs_layer_crypto(folio->mapping->host))
3946 			fscrypt_finalize_bounce_page(&fio->encrypted_page);
3947 		folio_end_writeback(folio);
3948 		if (f2fs_in_warm_node_list(fio->sbi, folio))
3949 			f2fs_del_fsync_node_entry(fio->sbi, folio);
3950 		f2fs_bug_on(fio->sbi, !is_set_ckpt_flags(fio->sbi,
3951 							CP_ERROR_FLAG));
3952 		goto out;
3953 	}
3954 
3955 	f2fs_bug_on(fio->sbi, !f2fs_is_valid_blkaddr_raw(fio->sbi,
3956 				fio->new_blkaddr, DATA_GENERIC_ENHANCE));
3957 
3958 	if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3959 		f2fs_invalidate_internal_cache(fio->sbi, fio->old_blkaddr, 1);
3960 
3961 	/* writeout dirty page into bdev */
3962 	f2fs_submit_page_write(fio);
3963 
3964 	f2fs_update_device_state(fio->sbi, fio->ino, fio->new_blkaddr, 1);
3965 out:
3966 	if (keep_order)
3967 		f2fs_up_read(&fio->sbi->io_order_lock);
3968 }
3969 
f2fs_do_write_meta_page(struct f2fs_sb_info * sbi,struct folio * folio,enum iostat_type io_type)3970 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct folio *folio,
3971 					enum iostat_type io_type)
3972 {
3973 	struct f2fs_io_info fio = {
3974 		.sbi = sbi,
3975 		.type = META,
3976 		.temp = HOT,
3977 		.op = REQ_OP_WRITE,
3978 		.op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3979 		.old_blkaddr = folio->index,
3980 		.new_blkaddr = folio->index,
3981 		.folio = folio,
3982 		.encrypted_page = NULL,
3983 		.in_list = 0,
3984 	};
3985 
3986 	if (unlikely(folio->index >= MAIN_BLKADDR(sbi)))
3987 		fio.op_flags &= ~REQ_META;
3988 
3989 	folio_start_writeback(folio);
3990 	f2fs_submit_page_write(&fio);
3991 
3992 	stat_inc_meta_count(sbi, folio->index);
3993 	f2fs_update_iostat(sbi, NULL, io_type, F2FS_BLKSIZE);
3994 }
3995 
f2fs_do_write_node_page(unsigned int nid,struct f2fs_io_info * fio)3996 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3997 {
3998 	struct f2fs_summary sum;
3999 
4000 	set_summary(&sum, nid, 0, 0);
4001 	do_write_page(&sum, fio);
4002 
4003 	f2fs_update_iostat(fio->sbi, NULL, fio->io_type, F2FS_BLKSIZE);
4004 }
4005 
f2fs_outplace_write_data(struct dnode_of_data * dn,struct f2fs_io_info * fio)4006 void f2fs_outplace_write_data(struct dnode_of_data *dn,
4007 					struct f2fs_io_info *fio)
4008 {
4009 	struct f2fs_sb_info *sbi = fio->sbi;
4010 	struct f2fs_summary sum;
4011 
4012 	f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
4013 	if (fio->io_type == FS_DATA_IO || fio->io_type == FS_CP_DATA_IO)
4014 		f2fs_update_age_extent_cache(dn);
4015 	set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
4016 	do_write_page(&sum, fio);
4017 	f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
4018 
4019 	f2fs_update_iostat(sbi, dn->inode, fio->io_type, F2FS_BLKSIZE);
4020 }
4021 
f2fs_inplace_write_data(struct f2fs_io_info * fio)4022 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
4023 {
4024 	int err;
4025 	struct f2fs_sb_info *sbi = fio->sbi;
4026 	unsigned int segno;
4027 
4028 	fio->new_blkaddr = fio->old_blkaddr;
4029 	/* i/o temperature is needed for passing down write hints */
4030 	__get_segment_type(fio);
4031 
4032 	segno = GET_SEGNO(sbi, fio->new_blkaddr);
4033 
4034 	if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
4035 		set_sbi_flag(sbi, SBI_NEED_FSCK);
4036 		f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
4037 			  __func__, segno);
4038 		err = -EFSCORRUPTED;
4039 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
4040 		goto drop_bio;
4041 	}
4042 
4043 	if (f2fs_cp_error(sbi)) {
4044 		err = -EIO;
4045 		goto drop_bio;
4046 	}
4047 
4048 	if (fio->meta_gc)
4049 		f2fs_truncate_meta_inode_pages(sbi, fio->new_blkaddr, 1);
4050 
4051 	stat_inc_inplace_blocks(fio->sbi);
4052 
4053 	if (fio->bio && !IS_F2FS_IPU_NOCACHE(sbi))
4054 		err = f2fs_merge_page_bio(fio);
4055 	else
4056 		err = f2fs_submit_page_bio(fio);
4057 	if (!err) {
4058 		f2fs_update_device_state(fio->sbi, fio->ino,
4059 						fio->new_blkaddr, 1);
4060 		f2fs_update_iostat(fio->sbi, fio_inode(fio),
4061 						fio->io_type, F2FS_BLKSIZE);
4062 	}
4063 
4064 	return err;
4065 drop_bio:
4066 	if (fio->bio && *(fio->bio)) {
4067 		struct bio *bio = *(fio->bio);
4068 
4069 		bio->bi_status = BLK_STS_IOERR;
4070 		bio_endio(bio);
4071 		*(fio->bio) = NULL;
4072 	}
4073 	return err;
4074 }
4075 
__f2fs_get_curseg(struct f2fs_sb_info * sbi,unsigned int segno)4076 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
4077 						unsigned int segno)
4078 {
4079 	int i;
4080 
4081 	for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
4082 		if (CURSEG_I(sbi, i)->segno == segno)
4083 			break;
4084 	}
4085 	return i;
4086 }
4087 
f2fs_do_replace_block(struct f2fs_sb_info * sbi,struct f2fs_summary * sum,block_t old_blkaddr,block_t new_blkaddr,bool recover_curseg,bool recover_newaddr,bool from_gc)4088 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
4089 				block_t old_blkaddr, block_t new_blkaddr,
4090 				bool recover_curseg, bool recover_newaddr,
4091 				bool from_gc)
4092 {
4093 	struct sit_info *sit_i = SIT_I(sbi);
4094 	struct curseg_info *curseg;
4095 	unsigned int segno, old_cursegno;
4096 	struct seg_entry *se;
4097 	int type;
4098 	unsigned short old_blkoff;
4099 	unsigned char old_alloc_type;
4100 
4101 	segno = GET_SEGNO(sbi, new_blkaddr);
4102 	se = get_seg_entry(sbi, segno);
4103 	type = se->type;
4104 
4105 	f2fs_down_write(&SM_I(sbi)->curseg_lock);
4106 
4107 	if (!recover_curseg) {
4108 		/* for recovery flow */
4109 		if (se->valid_blocks == 0 && !is_curseg(sbi, segno)) {
4110 			if (old_blkaddr == NULL_ADDR)
4111 				type = CURSEG_COLD_DATA;
4112 			else
4113 				type = CURSEG_WARM_DATA;
4114 		}
4115 	} else {
4116 		if (is_curseg(sbi, segno)) {
4117 			/* se->type is volatile as SSR allocation */
4118 			type = __f2fs_get_curseg(sbi, segno);
4119 			f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
4120 		} else {
4121 			type = CURSEG_WARM_DATA;
4122 		}
4123 	}
4124 
4125 	curseg = CURSEG_I(sbi, type);
4126 	f2fs_bug_on(sbi, !IS_DATASEG(curseg->seg_type));
4127 
4128 	mutex_lock(&curseg->curseg_mutex);
4129 	down_write(&sit_i->sentry_lock);
4130 
4131 	old_cursegno = curseg->segno;
4132 	old_blkoff = curseg->next_blkoff;
4133 	old_alloc_type = curseg->alloc_type;
4134 
4135 	/* change the current segment */
4136 	if (segno != curseg->segno) {
4137 		curseg->next_segno = segno;
4138 		if (change_curseg(sbi, type))
4139 			goto out_unlock;
4140 	}
4141 
4142 	curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
4143 	curseg->sum_blk->entries[curseg->next_blkoff] = *sum;
4144 
4145 	if (!recover_curseg || recover_newaddr) {
4146 		if (!from_gc)
4147 			update_segment_mtime(sbi, new_blkaddr, 0);
4148 		update_sit_entry(sbi, new_blkaddr, 1);
4149 	}
4150 	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
4151 		f2fs_invalidate_internal_cache(sbi, old_blkaddr, 1);
4152 		if (!from_gc)
4153 			update_segment_mtime(sbi, old_blkaddr, 0);
4154 		update_sit_entry(sbi, old_blkaddr, -1);
4155 	}
4156 
4157 	locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
4158 	locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
4159 
4160 	locate_dirty_segment(sbi, old_cursegno);
4161 
4162 	if (recover_curseg) {
4163 		if (old_cursegno != curseg->segno) {
4164 			curseg->next_segno = old_cursegno;
4165 			if (change_curseg(sbi, type))
4166 				goto out_unlock;
4167 		}
4168 		curseg->next_blkoff = old_blkoff;
4169 		curseg->alloc_type = old_alloc_type;
4170 	}
4171 
4172 out_unlock:
4173 	up_write(&sit_i->sentry_lock);
4174 	mutex_unlock(&curseg->curseg_mutex);
4175 	f2fs_up_write(&SM_I(sbi)->curseg_lock);
4176 }
4177 
f2fs_replace_block(struct f2fs_sb_info * sbi,struct dnode_of_data * dn,block_t old_addr,block_t new_addr,unsigned char version,bool recover_curseg,bool recover_newaddr)4178 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
4179 				block_t old_addr, block_t new_addr,
4180 				unsigned char version, bool recover_curseg,
4181 				bool recover_newaddr)
4182 {
4183 	struct f2fs_summary sum;
4184 
4185 	set_summary(&sum, dn->nid, dn->ofs_in_node, version);
4186 
4187 	f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
4188 					recover_curseg, recover_newaddr, false);
4189 
4190 	f2fs_update_data_blkaddr(dn, new_addr);
4191 }
4192 
f2fs_folio_wait_writeback(struct folio * folio,enum page_type type,bool ordered,bool locked)4193 void f2fs_folio_wait_writeback(struct folio *folio, enum page_type type,
4194 		bool ordered, bool locked)
4195 {
4196 	if (folio_test_writeback(folio)) {
4197 		struct f2fs_sb_info *sbi = F2FS_F_SB(folio);
4198 
4199 		/* submit cached LFS IO */
4200 		f2fs_submit_merged_write_cond(sbi, NULL, folio, 0, type);
4201 		/* submit cached IPU IO */
4202 		f2fs_submit_merged_ipu_write(sbi, NULL, folio);
4203 		if (ordered) {
4204 			folio_wait_writeback(folio);
4205 			f2fs_bug_on(sbi, locked && folio_test_writeback(folio));
4206 		} else {
4207 			folio_wait_stable(folio);
4208 		}
4209 	}
4210 }
4211 
f2fs_wait_on_block_writeback(struct inode * inode,block_t blkaddr)4212 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
4213 {
4214 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
4215 	struct folio *cfolio;
4216 
4217 	if (!f2fs_meta_inode_gc_required(inode))
4218 		return;
4219 
4220 	if (!__is_valid_data_blkaddr(blkaddr))
4221 		return;
4222 
4223 	cfolio = filemap_lock_folio(META_MAPPING(sbi), blkaddr);
4224 	if (!IS_ERR(cfolio)) {
4225 		f2fs_folio_wait_writeback(cfolio, DATA, true, true);
4226 		f2fs_folio_put(cfolio, true);
4227 	}
4228 }
4229 
f2fs_wait_on_block_writeback_range(struct inode * inode,block_t blkaddr,block_t len)4230 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
4231 								block_t len)
4232 {
4233 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
4234 	block_t i;
4235 
4236 	if (!f2fs_meta_inode_gc_required(inode))
4237 		return;
4238 
4239 	for (i = 0; i < len; i++)
4240 		f2fs_wait_on_block_writeback(inode, blkaddr + i);
4241 
4242 	f2fs_truncate_meta_inode_pages(sbi, blkaddr, len);
4243 }
4244 
read_compacted_summaries(struct f2fs_sb_info * sbi)4245 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
4246 {
4247 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4248 	struct curseg_info *seg_i;
4249 	unsigned char *kaddr;
4250 	struct folio *folio;
4251 	block_t start;
4252 	int i, j, offset;
4253 
4254 	start = start_sum_block(sbi);
4255 
4256 	folio = f2fs_get_meta_folio(sbi, start++);
4257 	if (IS_ERR(folio))
4258 		return PTR_ERR(folio);
4259 	kaddr = folio_address(folio);
4260 
4261 	/* Step 1: restore nat cache */
4262 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
4263 	memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
4264 
4265 	/* Step 2: restore sit cache */
4266 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
4267 	memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
4268 	offset = 2 * SUM_JOURNAL_SIZE;
4269 
4270 	/* Step 3: restore summary entries */
4271 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
4272 		unsigned short blk_off;
4273 		unsigned int segno;
4274 
4275 		seg_i = CURSEG_I(sbi, i);
4276 		segno = le32_to_cpu(ckpt->cur_data_segno[i]);
4277 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
4278 		seg_i->next_segno = segno;
4279 		reset_curseg(sbi, i, 0);
4280 		seg_i->alloc_type = ckpt->alloc_type[i];
4281 		seg_i->next_blkoff = blk_off;
4282 
4283 		if (seg_i->alloc_type == SSR)
4284 			blk_off = BLKS_PER_SEG(sbi);
4285 
4286 		for (j = 0; j < blk_off; j++) {
4287 			struct f2fs_summary *s;
4288 
4289 			s = (struct f2fs_summary *)(kaddr + offset);
4290 			seg_i->sum_blk->entries[j] = *s;
4291 			offset += SUMMARY_SIZE;
4292 			if (offset + SUMMARY_SIZE <= PAGE_SIZE -
4293 						SUM_FOOTER_SIZE)
4294 				continue;
4295 
4296 			f2fs_folio_put(folio, true);
4297 
4298 			folio = f2fs_get_meta_folio(sbi, start++);
4299 			if (IS_ERR(folio))
4300 				return PTR_ERR(folio);
4301 			kaddr = folio_address(folio);
4302 			offset = 0;
4303 		}
4304 	}
4305 	f2fs_folio_put(folio, true);
4306 	return 0;
4307 }
4308 
read_normal_summaries(struct f2fs_sb_info * sbi,int type)4309 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
4310 {
4311 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4312 	struct f2fs_summary_block *sum;
4313 	struct curseg_info *curseg;
4314 	struct folio *new;
4315 	unsigned short blk_off;
4316 	unsigned int segno = 0;
4317 	block_t blk_addr = 0;
4318 	int err = 0;
4319 
4320 	/* get segment number and block addr */
4321 	if (IS_DATASEG(type)) {
4322 		segno = le32_to_cpu(ckpt->cur_data_segno[type]);
4323 		blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
4324 							CURSEG_HOT_DATA]);
4325 		if (__exist_node_summaries(sbi))
4326 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type);
4327 		else
4328 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
4329 	} else {
4330 		segno = le32_to_cpu(ckpt->cur_node_segno[type -
4331 							CURSEG_HOT_NODE]);
4332 		blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
4333 							CURSEG_HOT_NODE]);
4334 		if (__exist_node_summaries(sbi))
4335 			blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
4336 							type - CURSEG_HOT_NODE);
4337 		else
4338 			blk_addr = GET_SUM_BLOCK(sbi, segno);
4339 	}
4340 
4341 	new = f2fs_get_meta_folio(sbi, blk_addr);
4342 	if (IS_ERR(new))
4343 		return PTR_ERR(new);
4344 	sum = folio_address(new);
4345 
4346 	if (IS_NODESEG(type)) {
4347 		if (__exist_node_summaries(sbi)) {
4348 			struct f2fs_summary *ns = &sum->entries[0];
4349 			int i;
4350 
4351 			for (i = 0; i < BLKS_PER_SEG(sbi); i++, ns++) {
4352 				ns->version = 0;
4353 				ns->ofs_in_node = 0;
4354 			}
4355 		} else {
4356 			err = f2fs_restore_node_summary(sbi, segno, sum);
4357 			if (err)
4358 				goto out;
4359 		}
4360 	}
4361 
4362 	/* set uncompleted segment to curseg */
4363 	curseg = CURSEG_I(sbi, type);
4364 	mutex_lock(&curseg->curseg_mutex);
4365 
4366 	/* update journal info */
4367 	down_write(&curseg->journal_rwsem);
4368 	memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
4369 	up_write(&curseg->journal_rwsem);
4370 
4371 	memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
4372 	memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
4373 	curseg->next_segno = segno;
4374 	reset_curseg(sbi, type, 0);
4375 	curseg->alloc_type = ckpt->alloc_type[type];
4376 	curseg->next_blkoff = blk_off;
4377 	mutex_unlock(&curseg->curseg_mutex);
4378 out:
4379 	f2fs_folio_put(new, true);
4380 	return err;
4381 }
4382 
restore_curseg_summaries(struct f2fs_sb_info * sbi)4383 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
4384 {
4385 	struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
4386 	struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
4387 	int type = CURSEG_HOT_DATA;
4388 	int err;
4389 
4390 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
4391 		int npages = f2fs_npages_for_summary_flush(sbi, true);
4392 
4393 		if (npages >= 2)
4394 			f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
4395 							META_CP, true);
4396 
4397 		/* restore for compacted data summary */
4398 		err = read_compacted_summaries(sbi);
4399 		if (err)
4400 			return err;
4401 		type = CURSEG_HOT_NODE;
4402 	}
4403 
4404 	if (__exist_node_summaries(sbi))
4405 		f2fs_ra_meta_pages(sbi,
4406 				sum_blk_addr(sbi, NR_CURSEG_PERSIST_TYPE, type),
4407 				NR_CURSEG_PERSIST_TYPE - type, META_CP, true);
4408 
4409 	for (; type <= CURSEG_COLD_NODE; type++) {
4410 		err = read_normal_summaries(sbi, type);
4411 		if (err)
4412 			return err;
4413 	}
4414 
4415 	/* sanity check for summary blocks */
4416 	if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
4417 			sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
4418 		f2fs_err(sbi, "invalid journal entries nats %u sits %u",
4419 			 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
4420 		return -EINVAL;
4421 	}
4422 
4423 	return 0;
4424 }
4425 
write_compacted_summaries(struct f2fs_sb_info * sbi,block_t blkaddr)4426 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
4427 {
4428 	struct folio *folio;
4429 	unsigned char *kaddr;
4430 	struct f2fs_summary *summary;
4431 	struct curseg_info *seg_i;
4432 	int written_size = 0;
4433 	int i, j;
4434 
4435 	folio = f2fs_grab_meta_folio(sbi, blkaddr++);
4436 	kaddr = folio_address(folio);
4437 	memset(kaddr, 0, PAGE_SIZE);
4438 
4439 	/* Step 1: write nat cache */
4440 	seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
4441 	memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
4442 	written_size += SUM_JOURNAL_SIZE;
4443 
4444 	/* Step 2: write sit cache */
4445 	seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
4446 	memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
4447 	written_size += SUM_JOURNAL_SIZE;
4448 
4449 	/* Step 3: write summary entries */
4450 	for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
4451 		seg_i = CURSEG_I(sbi, i);
4452 		for (j = 0; j < f2fs_curseg_valid_blocks(sbi, i); j++) {
4453 			if (!folio) {
4454 				folio = f2fs_grab_meta_folio(sbi, blkaddr++);
4455 				kaddr = folio_address(folio);
4456 				memset(kaddr, 0, PAGE_SIZE);
4457 				written_size = 0;
4458 			}
4459 			summary = (struct f2fs_summary *)(kaddr + written_size);
4460 			*summary = seg_i->sum_blk->entries[j];
4461 			written_size += SUMMARY_SIZE;
4462 
4463 			if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
4464 							SUM_FOOTER_SIZE)
4465 				continue;
4466 
4467 			folio_mark_dirty(folio);
4468 			f2fs_folio_put(folio, true);
4469 			folio = NULL;
4470 		}
4471 	}
4472 	if (folio) {
4473 		folio_mark_dirty(folio);
4474 		f2fs_folio_put(folio, true);
4475 	}
4476 }
4477 
write_normal_summaries(struct f2fs_sb_info * sbi,block_t blkaddr,int type)4478 static void write_normal_summaries(struct f2fs_sb_info *sbi,
4479 					block_t blkaddr, int type)
4480 {
4481 	int i, end;
4482 
4483 	if (IS_DATASEG(type))
4484 		end = type + NR_CURSEG_DATA_TYPE;
4485 	else
4486 		end = type + NR_CURSEG_NODE_TYPE;
4487 
4488 	for (i = type; i < end; i++)
4489 		write_current_sum_page(sbi, i, blkaddr + (i - type));
4490 }
4491 
f2fs_write_data_summaries(struct f2fs_sb_info * sbi,block_t start_blk)4492 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4493 {
4494 	if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
4495 		write_compacted_summaries(sbi, start_blk);
4496 	else
4497 		write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
4498 }
4499 
f2fs_write_node_summaries(struct f2fs_sb_info * sbi,block_t start_blk)4500 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
4501 {
4502 	write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
4503 }
4504 
f2fs_lookup_journal_in_cursum(struct f2fs_journal * journal,int type,unsigned int val,int alloc)4505 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
4506 					unsigned int val, int alloc)
4507 {
4508 	int i;
4509 
4510 	if (type == NAT_JOURNAL) {
4511 		for (i = 0; i < nats_in_cursum(journal); i++) {
4512 			if (le32_to_cpu(nid_in_journal(journal, i)) == val)
4513 				return i;
4514 		}
4515 		if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
4516 			return update_nats_in_cursum(journal, 1);
4517 	} else if (type == SIT_JOURNAL) {
4518 		for (i = 0; i < sits_in_cursum(journal); i++)
4519 			if (le32_to_cpu(segno_in_journal(journal, i)) == val)
4520 				return i;
4521 		if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
4522 			return update_sits_in_cursum(journal, 1);
4523 	}
4524 	return -1;
4525 }
4526 
get_current_sit_folio(struct f2fs_sb_info * sbi,unsigned int segno)4527 static struct folio *get_current_sit_folio(struct f2fs_sb_info *sbi,
4528 					unsigned int segno)
4529 {
4530 	return f2fs_get_meta_folio(sbi, current_sit_addr(sbi, segno));
4531 }
4532 
get_next_sit_folio(struct f2fs_sb_info * sbi,unsigned int start)4533 static struct folio *get_next_sit_folio(struct f2fs_sb_info *sbi,
4534 					unsigned int start)
4535 {
4536 	struct sit_info *sit_i = SIT_I(sbi);
4537 	struct folio *folio;
4538 	pgoff_t src_off, dst_off;
4539 
4540 	src_off = current_sit_addr(sbi, start);
4541 	dst_off = next_sit_addr(sbi, src_off);
4542 
4543 	folio = f2fs_grab_meta_folio(sbi, dst_off);
4544 	seg_info_to_sit_folio(sbi, folio, start);
4545 
4546 	folio_mark_dirty(folio);
4547 	set_to_next_sit(sit_i, start);
4548 
4549 	return folio;
4550 }
4551 
grab_sit_entry_set(void)4552 static struct sit_entry_set *grab_sit_entry_set(void)
4553 {
4554 	struct sit_entry_set *ses =
4555 			f2fs_kmem_cache_alloc(sit_entry_set_slab,
4556 						GFP_NOFS, true, NULL);
4557 
4558 	ses->entry_cnt = 0;
4559 	INIT_LIST_HEAD(&ses->set_list);
4560 	return ses;
4561 }
4562 
release_sit_entry_set(struct sit_entry_set * ses)4563 static void release_sit_entry_set(struct sit_entry_set *ses)
4564 {
4565 	list_del(&ses->set_list);
4566 	kmem_cache_free(sit_entry_set_slab, ses);
4567 }
4568 
adjust_sit_entry_set(struct sit_entry_set * ses,struct list_head * head)4569 static void adjust_sit_entry_set(struct sit_entry_set *ses,
4570 						struct list_head *head)
4571 {
4572 	struct sit_entry_set *next = ses;
4573 
4574 	if (list_is_last(&ses->set_list, head))
4575 		return;
4576 
4577 	list_for_each_entry_continue(next, head, set_list)
4578 		if (ses->entry_cnt <= next->entry_cnt) {
4579 			list_move_tail(&ses->set_list, &next->set_list);
4580 			return;
4581 		}
4582 
4583 	list_move_tail(&ses->set_list, head);
4584 }
4585 
add_sit_entry(unsigned int segno,struct list_head * head)4586 static void add_sit_entry(unsigned int segno, struct list_head *head)
4587 {
4588 	struct sit_entry_set *ses;
4589 	unsigned int start_segno = START_SEGNO(segno);
4590 
4591 	list_for_each_entry(ses, head, set_list) {
4592 		if (ses->start_segno == start_segno) {
4593 			ses->entry_cnt++;
4594 			adjust_sit_entry_set(ses, head);
4595 			return;
4596 		}
4597 	}
4598 
4599 	ses = grab_sit_entry_set();
4600 
4601 	ses->start_segno = start_segno;
4602 	ses->entry_cnt++;
4603 	list_add(&ses->set_list, head);
4604 }
4605 
add_sits_in_set(struct f2fs_sb_info * sbi)4606 static void add_sits_in_set(struct f2fs_sb_info *sbi)
4607 {
4608 	struct f2fs_sm_info *sm_info = SM_I(sbi);
4609 	struct list_head *set_list = &sm_info->sit_entry_set;
4610 	unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
4611 	unsigned int segno;
4612 
4613 	for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
4614 		add_sit_entry(segno, set_list);
4615 }
4616 
remove_sits_in_journal(struct f2fs_sb_info * sbi)4617 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
4618 {
4619 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4620 	struct f2fs_journal *journal = curseg->journal;
4621 	int i;
4622 
4623 	down_write(&curseg->journal_rwsem);
4624 	for (i = 0; i < sits_in_cursum(journal); i++) {
4625 		unsigned int segno;
4626 		bool dirtied;
4627 
4628 		segno = le32_to_cpu(segno_in_journal(journal, i));
4629 		dirtied = __mark_sit_entry_dirty(sbi, segno);
4630 
4631 		if (!dirtied)
4632 			add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
4633 	}
4634 	update_sits_in_cursum(journal, -i);
4635 	up_write(&curseg->journal_rwsem);
4636 }
4637 
4638 /*
4639  * CP calls this function, which flushes SIT entries including sit_journal,
4640  * and moves prefree segs to free segs.
4641  */
f2fs_flush_sit_entries(struct f2fs_sb_info * sbi,struct cp_control * cpc)4642 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
4643 {
4644 	struct sit_info *sit_i = SIT_I(sbi);
4645 	unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
4646 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4647 	struct f2fs_journal *journal = curseg->journal;
4648 	struct sit_entry_set *ses, *tmp;
4649 	struct list_head *head = &SM_I(sbi)->sit_entry_set;
4650 	bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
4651 	struct seg_entry *se;
4652 
4653 	down_write(&sit_i->sentry_lock);
4654 
4655 	if (!sit_i->dirty_sentries)
4656 		goto out;
4657 
4658 	/*
4659 	 * add and account sit entries of dirty bitmap in sit entry
4660 	 * set temporarily
4661 	 */
4662 	add_sits_in_set(sbi);
4663 
4664 	/*
4665 	 * if there are no enough space in journal to store dirty sit
4666 	 * entries, remove all entries from journal and add and account
4667 	 * them in sit entry set.
4668 	 */
4669 	if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
4670 								!to_journal)
4671 		remove_sits_in_journal(sbi);
4672 
4673 	/*
4674 	 * there are two steps to flush sit entries:
4675 	 * #1, flush sit entries to journal in current cold data summary block.
4676 	 * #2, flush sit entries to sit page.
4677 	 */
4678 	list_for_each_entry_safe(ses, tmp, head, set_list) {
4679 		struct folio *folio = NULL;
4680 		struct f2fs_sit_block *raw_sit = NULL;
4681 		unsigned int start_segno = ses->start_segno;
4682 		unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
4683 						(unsigned long)MAIN_SEGS(sbi));
4684 		unsigned int segno = start_segno;
4685 
4686 		if (to_journal &&
4687 			!__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
4688 			to_journal = false;
4689 
4690 		if (to_journal) {
4691 			down_write(&curseg->journal_rwsem);
4692 		} else {
4693 			folio = get_next_sit_folio(sbi, start_segno);
4694 			raw_sit = folio_address(folio);
4695 		}
4696 
4697 		/* flush dirty sit entries in region of current sit set */
4698 		for_each_set_bit_from(segno, bitmap, end) {
4699 			int offset, sit_offset;
4700 
4701 			se = get_seg_entry(sbi, segno);
4702 #ifdef CONFIG_F2FS_CHECK_FS
4703 			if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
4704 						SIT_VBLOCK_MAP_SIZE))
4705 				f2fs_bug_on(sbi, 1);
4706 #endif
4707 
4708 			/* add discard candidates */
4709 			if (!(cpc->reason & CP_DISCARD)) {
4710 				cpc->trim_start = segno;
4711 				add_discard_addrs(sbi, cpc, false);
4712 			}
4713 
4714 			if (to_journal) {
4715 				offset = f2fs_lookup_journal_in_cursum(journal,
4716 							SIT_JOURNAL, segno, 1);
4717 				f2fs_bug_on(sbi, offset < 0);
4718 				segno_in_journal(journal, offset) =
4719 							cpu_to_le32(segno);
4720 				seg_info_to_raw_sit(se,
4721 					&sit_in_journal(journal, offset));
4722 				check_block_count(sbi, segno,
4723 					&sit_in_journal(journal, offset));
4724 			} else {
4725 				sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
4726 				seg_info_to_raw_sit(se,
4727 						&raw_sit->entries[sit_offset]);
4728 				check_block_count(sbi, segno,
4729 						&raw_sit->entries[sit_offset]);
4730 			}
4731 
4732 			/* update ckpt_valid_block */
4733 			if (__is_large_section(sbi)) {
4734 				set_ckpt_valid_blocks(sbi, segno);
4735 				sanity_check_valid_blocks(sbi, segno);
4736 			}
4737 
4738 			__clear_bit(segno, bitmap);
4739 			sit_i->dirty_sentries--;
4740 			ses->entry_cnt--;
4741 		}
4742 
4743 		if (to_journal)
4744 			up_write(&curseg->journal_rwsem);
4745 		else
4746 			f2fs_folio_put(folio, true);
4747 
4748 		f2fs_bug_on(sbi, ses->entry_cnt);
4749 		release_sit_entry_set(ses);
4750 	}
4751 
4752 	f2fs_bug_on(sbi, !list_empty(head));
4753 	f2fs_bug_on(sbi, sit_i->dirty_sentries);
4754 out:
4755 	if (cpc->reason & CP_DISCARD) {
4756 		__u64 trim_start = cpc->trim_start;
4757 
4758 		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
4759 			add_discard_addrs(sbi, cpc, false);
4760 
4761 		cpc->trim_start = trim_start;
4762 	}
4763 	up_write(&sit_i->sentry_lock);
4764 
4765 	set_prefree_as_free_segments(sbi);
4766 }
4767 
build_sit_info(struct f2fs_sb_info * sbi)4768 static int build_sit_info(struct f2fs_sb_info *sbi)
4769 {
4770 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4771 	struct sit_info *sit_i;
4772 	unsigned int sit_segs, start;
4773 	char *src_bitmap, *bitmap;
4774 	unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
4775 	unsigned int discard_map = f2fs_block_unit_discard(sbi) ? 1 : 0;
4776 
4777 	/* allocate memory for SIT information */
4778 	sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
4779 	if (!sit_i)
4780 		return -ENOMEM;
4781 
4782 	SM_I(sbi)->sit_info = sit_i;
4783 
4784 	sit_i->sentries =
4785 		f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
4786 					      MAIN_SEGS(sbi)),
4787 			      GFP_KERNEL);
4788 	if (!sit_i->sentries)
4789 		return -ENOMEM;
4790 
4791 	main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4792 	sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
4793 								GFP_KERNEL);
4794 	if (!sit_i->dirty_sentries_bitmap)
4795 		return -ENOMEM;
4796 
4797 #ifdef CONFIG_F2FS_CHECK_FS
4798 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (3 + discard_map);
4799 #else
4800 	bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * (2 + discard_map);
4801 #endif
4802 	sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4803 	if (!sit_i->bitmap)
4804 		return -ENOMEM;
4805 
4806 	bitmap = sit_i->bitmap;
4807 
4808 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
4809 		sit_i->sentries[start].cur_valid_map = bitmap;
4810 		bitmap += SIT_VBLOCK_MAP_SIZE;
4811 
4812 		sit_i->sentries[start].ckpt_valid_map = bitmap;
4813 		bitmap += SIT_VBLOCK_MAP_SIZE;
4814 
4815 #ifdef CONFIG_F2FS_CHECK_FS
4816 		sit_i->sentries[start].cur_valid_map_mir = bitmap;
4817 		bitmap += SIT_VBLOCK_MAP_SIZE;
4818 #endif
4819 
4820 		if (discard_map) {
4821 			sit_i->sentries[start].discard_map = bitmap;
4822 			bitmap += SIT_VBLOCK_MAP_SIZE;
4823 		}
4824 	}
4825 
4826 	sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4827 	if (!sit_i->tmp_map)
4828 		return -ENOMEM;
4829 
4830 	if (__is_large_section(sbi)) {
4831 		sit_i->sec_entries =
4832 			f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4833 						      MAIN_SECS(sbi)),
4834 				      GFP_KERNEL);
4835 		if (!sit_i->sec_entries)
4836 			return -ENOMEM;
4837 	}
4838 
4839 	/* get information related with SIT */
4840 	sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4841 
4842 	/* setup SIT bitmap from ckeckpoint pack */
4843 	sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4844 	src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4845 
4846 	sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4847 	if (!sit_i->sit_bitmap)
4848 		return -ENOMEM;
4849 
4850 #ifdef CONFIG_F2FS_CHECK_FS
4851 	sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4852 					sit_bitmap_size, GFP_KERNEL);
4853 	if (!sit_i->sit_bitmap_mir)
4854 		return -ENOMEM;
4855 
4856 	sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4857 					main_bitmap_size, GFP_KERNEL);
4858 	if (!sit_i->invalid_segmap)
4859 		return -ENOMEM;
4860 #endif
4861 
4862 	sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4863 	sit_i->sit_blocks = SEGS_TO_BLKS(sbi, sit_segs);
4864 	sit_i->written_valid_blocks = 0;
4865 	sit_i->bitmap_size = sit_bitmap_size;
4866 	sit_i->dirty_sentries = 0;
4867 	sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4868 	sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4869 	sit_i->mounted_time = ktime_get_boottime_seconds();
4870 	init_rwsem(&sit_i->sentry_lock);
4871 	return 0;
4872 }
4873 
build_free_segmap(struct f2fs_sb_info * sbi)4874 static int build_free_segmap(struct f2fs_sb_info *sbi)
4875 {
4876 	struct free_segmap_info *free_i;
4877 	unsigned int bitmap_size, sec_bitmap_size;
4878 
4879 	/* allocate memory for free segmap information */
4880 	free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4881 	if (!free_i)
4882 		return -ENOMEM;
4883 
4884 	SM_I(sbi)->free_info = free_i;
4885 
4886 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4887 	free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4888 	if (!free_i->free_segmap)
4889 		return -ENOMEM;
4890 
4891 	sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4892 	free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4893 	if (!free_i->free_secmap)
4894 		return -ENOMEM;
4895 
4896 	/* set all segments as dirty temporarily */
4897 	memset(free_i->free_segmap, 0xff, bitmap_size);
4898 	memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4899 
4900 	/* init free segmap information */
4901 	free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4902 	free_i->free_segments = 0;
4903 	free_i->free_sections = 0;
4904 	spin_lock_init(&free_i->segmap_lock);
4905 	return 0;
4906 }
4907 
build_curseg(struct f2fs_sb_info * sbi)4908 static int build_curseg(struct f2fs_sb_info *sbi)
4909 {
4910 	struct curseg_info *array;
4911 	int i;
4912 
4913 	array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE,
4914 					sizeof(*array)), GFP_KERNEL);
4915 	if (!array)
4916 		return -ENOMEM;
4917 
4918 	SM_I(sbi)->curseg_array = array;
4919 
4920 	for (i = 0; i < NO_CHECK_TYPE; i++) {
4921 		mutex_init(&array[i].curseg_mutex);
4922 		array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4923 		if (!array[i].sum_blk)
4924 			return -ENOMEM;
4925 		init_rwsem(&array[i].journal_rwsem);
4926 		array[i].journal = f2fs_kzalloc(sbi,
4927 				sizeof(struct f2fs_journal), GFP_KERNEL);
4928 		if (!array[i].journal)
4929 			return -ENOMEM;
4930 		array[i].seg_type = log_type_to_seg_type(i);
4931 		reset_curseg_fields(&array[i]);
4932 	}
4933 	return restore_curseg_summaries(sbi);
4934 }
4935 
build_sit_entries(struct f2fs_sb_info * sbi)4936 static int build_sit_entries(struct f2fs_sb_info *sbi)
4937 {
4938 	struct sit_info *sit_i = SIT_I(sbi);
4939 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4940 	struct f2fs_journal *journal = curseg->journal;
4941 	struct seg_entry *se;
4942 	struct f2fs_sit_entry sit;
4943 	int sit_blk_cnt = SIT_BLK_CNT(sbi);
4944 	unsigned int i, start, end;
4945 	unsigned int readed, start_blk = 0;
4946 	int err = 0;
4947 	block_t sit_valid_blocks[2] = {0, 0};
4948 
4949 	do {
4950 		readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_VECS,
4951 							META_SIT, true);
4952 
4953 		start = start_blk * sit_i->sents_per_block;
4954 		end = (start_blk + readed) * sit_i->sents_per_block;
4955 
4956 		for (; start < end && start < MAIN_SEGS(sbi); start++) {
4957 			struct f2fs_sit_block *sit_blk;
4958 			struct folio *folio;
4959 
4960 			se = &sit_i->sentries[start];
4961 			folio = get_current_sit_folio(sbi, start);
4962 			if (IS_ERR(folio))
4963 				return PTR_ERR(folio);
4964 			sit_blk = folio_address(folio);
4965 			sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4966 			f2fs_folio_put(folio, true);
4967 
4968 			err = check_block_count(sbi, start, &sit);
4969 			if (err)
4970 				return err;
4971 			seg_info_from_raw_sit(se, &sit);
4972 
4973 			if (se->type >= NR_PERSISTENT_LOG) {
4974 				f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
4975 							se->type, start);
4976 				f2fs_handle_error(sbi,
4977 						ERROR_INCONSISTENT_SUM_TYPE);
4978 				return -EFSCORRUPTED;
4979 			}
4980 
4981 			sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
4982 
4983 			if (!f2fs_block_unit_discard(sbi))
4984 				goto init_discard_map_done;
4985 
4986 			/* build discard map only one time */
4987 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4988 				memset(se->discard_map, 0xff,
4989 						SIT_VBLOCK_MAP_SIZE);
4990 				goto init_discard_map_done;
4991 			}
4992 			memcpy(se->discard_map, se->cur_valid_map,
4993 						SIT_VBLOCK_MAP_SIZE);
4994 			sbi->discard_blks += BLKS_PER_SEG(sbi) -
4995 						se->valid_blocks;
4996 init_discard_map_done:
4997 			if (__is_large_section(sbi))
4998 				get_sec_entry(sbi, start)->valid_blocks +=
4999 							se->valid_blocks;
5000 		}
5001 		start_blk += readed;
5002 	} while (start_blk < sit_blk_cnt);
5003 
5004 	down_read(&curseg->journal_rwsem);
5005 	for (i = 0; i < sits_in_cursum(journal); i++) {
5006 		unsigned int old_valid_blocks;
5007 
5008 		start = le32_to_cpu(segno_in_journal(journal, i));
5009 		if (start >= MAIN_SEGS(sbi)) {
5010 			f2fs_err(sbi, "Wrong journal entry on segno %u",
5011 				 start);
5012 			err = -EFSCORRUPTED;
5013 			f2fs_handle_error(sbi, ERROR_CORRUPTED_JOURNAL);
5014 			break;
5015 		}
5016 
5017 		se = &sit_i->sentries[start];
5018 		sit = sit_in_journal(journal, i);
5019 
5020 		old_valid_blocks = se->valid_blocks;
5021 
5022 		sit_valid_blocks[SE_PAGETYPE(se)] -= old_valid_blocks;
5023 
5024 		err = check_block_count(sbi, start, &sit);
5025 		if (err)
5026 			break;
5027 		seg_info_from_raw_sit(se, &sit);
5028 
5029 		if (se->type >= NR_PERSISTENT_LOG) {
5030 			f2fs_err(sbi, "Invalid segment type: %u, segno: %u",
5031 							se->type, start);
5032 			err = -EFSCORRUPTED;
5033 			f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUM_TYPE);
5034 			break;
5035 		}
5036 
5037 		sit_valid_blocks[SE_PAGETYPE(se)] += se->valid_blocks;
5038 
5039 		if (f2fs_block_unit_discard(sbi)) {
5040 			if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
5041 				memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
5042 			} else {
5043 				memcpy(se->discard_map, se->cur_valid_map,
5044 							SIT_VBLOCK_MAP_SIZE);
5045 				sbi->discard_blks += old_valid_blocks;
5046 				sbi->discard_blks -= se->valid_blocks;
5047 			}
5048 		}
5049 
5050 		if (__is_large_section(sbi)) {
5051 			get_sec_entry(sbi, start)->valid_blocks +=
5052 							se->valid_blocks;
5053 			get_sec_entry(sbi, start)->valid_blocks -=
5054 							old_valid_blocks;
5055 		}
5056 	}
5057 	up_read(&curseg->journal_rwsem);
5058 
5059 	/* update ckpt_valid_block */
5060 	if (__is_large_section(sbi)) {
5061 		unsigned int segno;
5062 
5063 		for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) {
5064 			set_ckpt_valid_blocks(sbi, segno);
5065 			sanity_check_valid_blocks(sbi, segno);
5066 		}
5067 	}
5068 
5069 	if (err)
5070 		return err;
5071 
5072 	if (sit_valid_blocks[NODE] != valid_node_count(sbi)) {
5073 		f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
5074 			 sit_valid_blocks[NODE], valid_node_count(sbi));
5075 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_NODE_COUNT);
5076 		return -EFSCORRUPTED;
5077 	}
5078 
5079 	if (sit_valid_blocks[DATA] + sit_valid_blocks[NODE] >
5080 				valid_user_blocks(sbi)) {
5081 		f2fs_err(sbi, "SIT is corrupted data# %u %u vs %u",
5082 			 sit_valid_blocks[DATA], sit_valid_blocks[NODE],
5083 			 valid_user_blocks(sbi));
5084 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_BLOCK_COUNT);
5085 		return -EFSCORRUPTED;
5086 	}
5087 
5088 	return 0;
5089 }
5090 
init_free_segmap(struct f2fs_sb_info * sbi)5091 static void init_free_segmap(struct f2fs_sb_info *sbi)
5092 {
5093 	unsigned int start;
5094 	int type;
5095 	struct seg_entry *sentry;
5096 
5097 	for (start = 0; start < MAIN_SEGS(sbi); start++) {
5098 		if (f2fs_usable_blks_in_seg(sbi, start) == 0)
5099 			continue;
5100 		sentry = get_seg_entry(sbi, start);
5101 		if (!sentry->valid_blocks)
5102 			__set_free(sbi, start);
5103 		else
5104 			SIT_I(sbi)->written_valid_blocks +=
5105 						sentry->valid_blocks;
5106 	}
5107 
5108 	/* set use the current segments */
5109 	for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
5110 		struct curseg_info *curseg_t = CURSEG_I(sbi, type);
5111 
5112 		__set_test_and_inuse(sbi, curseg_t->segno);
5113 	}
5114 }
5115 
init_dirty_segmap(struct f2fs_sb_info * sbi)5116 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
5117 {
5118 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5119 	struct free_segmap_info *free_i = FREE_I(sbi);
5120 	unsigned int segno = 0, offset = 0, secno;
5121 	block_t valid_blocks, usable_blks_in_seg;
5122 
5123 	while (1) {
5124 		/* find dirty segment based on free segmap */
5125 		segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
5126 		if (segno >= MAIN_SEGS(sbi))
5127 			break;
5128 		offset = segno + 1;
5129 		valid_blocks = get_valid_blocks(sbi, segno, false);
5130 		usable_blks_in_seg = f2fs_usable_blks_in_seg(sbi, segno);
5131 		if (valid_blocks == usable_blks_in_seg || !valid_blocks)
5132 			continue;
5133 		if (valid_blocks > usable_blks_in_seg) {
5134 			f2fs_bug_on(sbi, 1);
5135 			continue;
5136 		}
5137 		mutex_lock(&dirty_i->seglist_lock);
5138 		__locate_dirty_segment(sbi, segno, DIRTY);
5139 		mutex_unlock(&dirty_i->seglist_lock);
5140 	}
5141 
5142 	if (!__is_large_section(sbi))
5143 		return;
5144 
5145 	mutex_lock(&dirty_i->seglist_lock);
5146 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) {
5147 		valid_blocks = get_valid_blocks(sbi, segno, true);
5148 		secno = GET_SEC_FROM_SEG(sbi, segno);
5149 
5150 		if (!valid_blocks || valid_blocks == CAP_BLKS_PER_SEC(sbi))
5151 			continue;
5152 		if (is_cursec(sbi, secno))
5153 			continue;
5154 		set_bit(secno, dirty_i->dirty_secmap);
5155 	}
5156 	mutex_unlock(&dirty_i->seglist_lock);
5157 }
5158 
init_victim_secmap(struct f2fs_sb_info * sbi)5159 static int init_victim_secmap(struct f2fs_sb_info *sbi)
5160 {
5161 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5162 	unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
5163 
5164 	dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
5165 	if (!dirty_i->victim_secmap)
5166 		return -ENOMEM;
5167 
5168 	dirty_i->pinned_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
5169 	if (!dirty_i->pinned_secmap)
5170 		return -ENOMEM;
5171 
5172 	dirty_i->pinned_secmap_cnt = 0;
5173 	dirty_i->enable_pin_section = true;
5174 	return 0;
5175 }
5176 
build_dirty_segmap(struct f2fs_sb_info * sbi)5177 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
5178 {
5179 	struct dirty_seglist_info *dirty_i;
5180 	unsigned int bitmap_size, i;
5181 
5182 	/* allocate memory for dirty segments list information */
5183 	dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
5184 								GFP_KERNEL);
5185 	if (!dirty_i)
5186 		return -ENOMEM;
5187 
5188 	SM_I(sbi)->dirty_info = dirty_i;
5189 	mutex_init(&dirty_i->seglist_lock);
5190 
5191 	bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
5192 
5193 	for (i = 0; i < NR_DIRTY_TYPE; i++) {
5194 		dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
5195 								GFP_KERNEL);
5196 		if (!dirty_i->dirty_segmap[i])
5197 			return -ENOMEM;
5198 	}
5199 
5200 	if (__is_large_section(sbi)) {
5201 		bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
5202 		dirty_i->dirty_secmap = f2fs_kvzalloc(sbi,
5203 						bitmap_size, GFP_KERNEL);
5204 		if (!dirty_i->dirty_secmap)
5205 			return -ENOMEM;
5206 	}
5207 
5208 	init_dirty_segmap(sbi);
5209 	return init_victim_secmap(sbi);
5210 }
5211 
sanity_check_curseg(struct f2fs_sb_info * sbi)5212 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
5213 {
5214 	int i;
5215 
5216 	/*
5217 	 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
5218 	 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
5219 	 */
5220 	for (i = 0; i < NR_PERSISTENT_LOG; i++) {
5221 		struct curseg_info *curseg = CURSEG_I(sbi, i);
5222 		struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
5223 		unsigned int blkofs = curseg->next_blkoff;
5224 
5225 		if (f2fs_sb_has_readonly(sbi) &&
5226 			i != CURSEG_HOT_DATA && i != CURSEG_HOT_NODE)
5227 			continue;
5228 
5229 		sanity_check_seg_type(sbi, curseg->seg_type);
5230 
5231 		if (curseg->alloc_type != LFS && curseg->alloc_type != SSR) {
5232 			f2fs_err(sbi,
5233 				 "Current segment has invalid alloc_type:%d",
5234 				 curseg->alloc_type);
5235 			f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
5236 			return -EFSCORRUPTED;
5237 		}
5238 
5239 		if (f2fs_test_bit(blkofs, se->cur_valid_map))
5240 			goto out;
5241 
5242 		if (curseg->alloc_type == SSR)
5243 			continue;
5244 
5245 		for (blkofs += 1; blkofs < BLKS_PER_SEG(sbi); blkofs++) {
5246 			if (!f2fs_test_bit(blkofs, se->cur_valid_map))
5247 				continue;
5248 out:
5249 			f2fs_err(sbi,
5250 				 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
5251 				 i, curseg->segno, curseg->alloc_type,
5252 				 curseg->next_blkoff, blkofs);
5253 			f2fs_handle_error(sbi, ERROR_INVALID_CURSEG);
5254 			return -EFSCORRUPTED;
5255 		}
5256 	}
5257 	return 0;
5258 }
5259 
5260 #ifdef CONFIG_BLK_DEV_ZONED
check_zone_write_pointer(struct f2fs_sb_info * sbi,struct f2fs_dev_info * fdev,struct blk_zone * zone)5261 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
5262 				    struct f2fs_dev_info *fdev,
5263 				    struct blk_zone *zone)
5264 {
5265 	unsigned int zone_segno;
5266 	block_t zone_block, valid_block_cnt;
5267 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
5268 	int ret;
5269 	unsigned int nofs_flags;
5270 
5271 	if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5272 		return 0;
5273 
5274 	zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
5275 	zone_segno = GET_SEGNO(sbi, zone_block);
5276 
5277 	/*
5278 	 * Skip check of zones cursegs point to, since
5279 	 * fix_curseg_write_pointer() checks them.
5280 	 */
5281 	if (zone_segno >= MAIN_SEGS(sbi))
5282 		return 0;
5283 
5284 	/*
5285 	 * Get # of valid block of the zone.
5286 	 */
5287 	valid_block_cnt = get_valid_blocks(sbi, zone_segno, true);
5288 	if (is_cursec(sbi, GET_SEC_FROM_SEG(sbi, zone_segno))) {
5289 		f2fs_notice(sbi, "Open zones: valid block[0x%x,0x%x] cond[%s]",
5290 				zone_segno, valid_block_cnt,
5291 				blk_zone_cond_str(zone->cond));
5292 		return 0;
5293 	}
5294 
5295 	if ((!valid_block_cnt && zone->cond == BLK_ZONE_COND_EMPTY) ||
5296 	    (valid_block_cnt && zone->cond == BLK_ZONE_COND_FULL))
5297 		return 0;
5298 
5299 	if (!valid_block_cnt) {
5300 		f2fs_notice(sbi, "Zone without valid block has non-zero write "
5301 			    "pointer. Reset the write pointer: cond[%s]",
5302 			    blk_zone_cond_str(zone->cond));
5303 		ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
5304 					zone->len >> log_sectors_per_block);
5305 		if (ret)
5306 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
5307 				 fdev->path, ret);
5308 		return ret;
5309 	}
5310 
5311 	/*
5312 	 * If there are valid blocks and the write pointer doesn't match
5313 	 * with them, we need to report the inconsistency and fill
5314 	 * the zone till the end to close the zone. This inconsistency
5315 	 * does not cause write error because the zone will not be
5316 	 * selected for write operation until it get discarded.
5317 	 */
5318 	f2fs_notice(sbi, "Valid blocks are not aligned with write "
5319 		    "pointer: valid block[0x%x,0x%x] cond[%s]",
5320 		    zone_segno, valid_block_cnt, blk_zone_cond_str(zone->cond));
5321 
5322 	nofs_flags = memalloc_nofs_save();
5323 	ret = blkdev_zone_mgmt(fdev->bdev, REQ_OP_ZONE_FINISH,
5324 				zone->start, zone->len);
5325 	memalloc_nofs_restore(nofs_flags);
5326 	if (ret == -EOPNOTSUPP) {
5327 		ret = blkdev_issue_zeroout(fdev->bdev, zone->wp,
5328 					zone->len - (zone->wp - zone->start),
5329 					GFP_NOFS, 0);
5330 		if (ret)
5331 			f2fs_err(sbi, "Fill up zone failed: %s (errno=%d)",
5332 					fdev->path, ret);
5333 	} else if (ret) {
5334 		f2fs_err(sbi, "Finishing zone failed: %s (errno=%d)",
5335 				fdev->path, ret);
5336 	}
5337 
5338 	return ret;
5339 }
5340 
get_target_zoned_dev(struct f2fs_sb_info * sbi,block_t zone_blkaddr)5341 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
5342 						  block_t zone_blkaddr)
5343 {
5344 	int i;
5345 
5346 	for (i = 0; i < sbi->s_ndevs; i++) {
5347 		if (!bdev_is_zoned(FDEV(i).bdev))
5348 			continue;
5349 		if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
5350 				zone_blkaddr <= FDEV(i).end_blk))
5351 			return &FDEV(i);
5352 	}
5353 
5354 	return NULL;
5355 }
5356 
report_one_zone_cb(struct blk_zone * zone,unsigned int idx,void * data)5357 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
5358 			      void *data)
5359 {
5360 	memcpy(data, zone, sizeof(struct blk_zone));
5361 	return 0;
5362 }
5363 
do_fix_curseg_write_pointer(struct f2fs_sb_info * sbi,int type)5364 static int do_fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
5365 {
5366 	struct curseg_info *cs = CURSEG_I(sbi, type);
5367 	struct f2fs_dev_info *zbd;
5368 	struct blk_zone zone;
5369 	unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
5370 	block_t cs_zone_block, wp_block;
5371 	unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
5372 	sector_t zone_sector;
5373 	int err;
5374 
5375 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
5376 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
5377 
5378 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
5379 	if (!zbd)
5380 		return 0;
5381 
5382 	/* report zone for the sector the curseg points to */
5383 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5384 		<< log_sectors_per_block;
5385 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5386 				  report_one_zone_cb, &zone);
5387 	if (err != 1) {
5388 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5389 			 zbd->path, err);
5390 		return err;
5391 	}
5392 
5393 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5394 		return 0;
5395 
5396 	/*
5397 	 * When safely unmounted in the previous mount, we could use current
5398 	 * segments. Otherwise, allocate new sections.
5399 	 */
5400 	if (is_set_ckpt_flags(sbi, CP_UMOUNT_FLAG)) {
5401 		wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
5402 		wp_segno = GET_SEGNO(sbi, wp_block);
5403 		wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
5404 		wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
5405 
5406 		if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
5407 				wp_sector_off == 0)
5408 			return 0;
5409 
5410 		f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
5411 			    "curseg[0x%x,0x%x] wp[0x%x,0x%x]", type, cs->segno,
5412 			    cs->next_blkoff, wp_segno, wp_blkoff);
5413 	}
5414 
5415 	/* Allocate a new section if it's not new. */
5416 	if (cs->next_blkoff ||
5417 	    cs->segno != GET_SEG_FROM_SEC(sbi, GET_ZONE_FROM_SEC(sbi, cs_section))) {
5418 		unsigned int old_segno = cs->segno, old_blkoff = cs->next_blkoff;
5419 
5420 		f2fs_allocate_new_section(sbi, type, true);
5421 		f2fs_notice(sbi, "Assign new section to curseg[%d]: "
5422 				"[0x%x,0x%x] -> [0x%x,0x%x]",
5423 				type, old_segno, old_blkoff,
5424 				cs->segno, cs->next_blkoff);
5425 	}
5426 
5427 	/* check consistency of the zone curseg pointed to */
5428 	if (check_zone_write_pointer(sbi, zbd, &zone))
5429 		return -EIO;
5430 
5431 	/* check newly assigned zone */
5432 	cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
5433 	cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
5434 
5435 	zbd = get_target_zoned_dev(sbi, cs_zone_block);
5436 	if (!zbd)
5437 		return 0;
5438 
5439 	zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
5440 		<< log_sectors_per_block;
5441 	err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
5442 				  report_one_zone_cb, &zone);
5443 	if (err != 1) {
5444 		f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
5445 			 zbd->path, err);
5446 		return err;
5447 	}
5448 
5449 	if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
5450 		return 0;
5451 
5452 	if (zone.wp != zone.start) {
5453 		f2fs_notice(sbi,
5454 			    "New zone for curseg[%d] is not yet discarded. "
5455 			    "Reset the zone: curseg[0x%x,0x%x]",
5456 			    type, cs->segno, cs->next_blkoff);
5457 		err = __f2fs_issue_discard_zone(sbi, zbd->bdev,	cs_zone_block,
5458 					zone.len >> log_sectors_per_block);
5459 		if (err) {
5460 			f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
5461 				 zbd->path, err);
5462 			return err;
5463 		}
5464 	}
5465 
5466 	return 0;
5467 }
5468 
fix_curseg_write_pointer(struct f2fs_sb_info * sbi)5469 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
5470 {
5471 	int i, ret;
5472 
5473 	for (i = 0; i < NR_PERSISTENT_LOG; i++) {
5474 		ret = do_fix_curseg_write_pointer(sbi, i);
5475 		if (ret)
5476 			return ret;
5477 	}
5478 
5479 	return 0;
5480 }
5481 
5482 struct check_zone_write_pointer_args {
5483 	struct f2fs_sb_info *sbi;
5484 	struct f2fs_dev_info *fdev;
5485 };
5486 
check_zone_write_pointer_cb(struct blk_zone * zone,unsigned int idx,void * data)5487 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
5488 				      void *data)
5489 {
5490 	struct check_zone_write_pointer_args *args;
5491 
5492 	args = (struct check_zone_write_pointer_args *)data;
5493 
5494 	return check_zone_write_pointer(args->sbi, args->fdev, zone);
5495 }
5496 
check_write_pointer(struct f2fs_sb_info * sbi)5497 static int check_write_pointer(struct f2fs_sb_info *sbi)
5498 {
5499 	int i, ret;
5500 	struct check_zone_write_pointer_args args;
5501 
5502 	for (i = 0; i < sbi->s_ndevs; i++) {
5503 		if (!bdev_is_zoned(FDEV(i).bdev))
5504 			continue;
5505 
5506 		args.sbi = sbi;
5507 		args.fdev = &FDEV(i);
5508 		ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
5509 					  check_zone_write_pointer_cb, &args);
5510 		if (ret < 0)
5511 			return ret;
5512 	}
5513 
5514 	return 0;
5515 }
5516 
f2fs_check_and_fix_write_pointer(struct f2fs_sb_info * sbi)5517 int f2fs_check_and_fix_write_pointer(struct f2fs_sb_info *sbi)
5518 {
5519 	int ret;
5520 
5521 	if (!f2fs_sb_has_blkzoned(sbi) || f2fs_readonly(sbi->sb) ||
5522 	    f2fs_hw_is_readonly(sbi))
5523 		return 0;
5524 
5525 	f2fs_notice(sbi, "Checking entire write pointers");
5526 	ret = fix_curseg_write_pointer(sbi);
5527 	if (!ret)
5528 		ret = check_write_pointer(sbi);
5529 	return ret;
5530 }
5531 
5532 /*
5533  * Return the number of usable blocks in a segment. The number of blocks
5534  * returned is always equal to the number of blocks in a segment for
5535  * segments fully contained within a sequential zone capacity or a
5536  * conventional zone. For segments partially contained in a sequential
5537  * zone capacity, the number of usable blocks up to the zone capacity
5538  * is returned. 0 is returned in all other cases.
5539  */
f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info * sbi,unsigned int segno)5540 static inline unsigned int f2fs_usable_zone_blks_in_seg(
5541 			struct f2fs_sb_info *sbi, unsigned int segno)
5542 {
5543 	block_t seg_start, sec_start_blkaddr, sec_cap_blkaddr;
5544 	unsigned int secno;
5545 
5546 	if (!sbi->unusable_blocks_per_sec)
5547 		return BLKS_PER_SEG(sbi);
5548 
5549 	secno = GET_SEC_FROM_SEG(sbi, segno);
5550 	seg_start = START_BLOCK(sbi, segno);
5551 	sec_start_blkaddr = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, secno));
5552 	sec_cap_blkaddr = sec_start_blkaddr + CAP_BLKS_PER_SEC(sbi);
5553 
5554 	/*
5555 	 * If segment starts before zone capacity and spans beyond
5556 	 * zone capacity, then usable blocks are from seg start to
5557 	 * zone capacity. If the segment starts after the zone capacity,
5558 	 * then there are no usable blocks.
5559 	 */
5560 	if (seg_start >= sec_cap_blkaddr)
5561 		return 0;
5562 	if (seg_start + BLKS_PER_SEG(sbi) > sec_cap_blkaddr)
5563 		return sec_cap_blkaddr - seg_start;
5564 
5565 	return BLKS_PER_SEG(sbi);
5566 }
5567 #else
f2fs_check_and_fix_write_pointer(struct f2fs_sb_info * sbi)5568 int f2fs_check_and_fix_write_pointer(struct f2fs_sb_info *sbi)
5569 {
5570 	return 0;
5571 }
5572 
f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info * sbi,unsigned int segno)5573 static inline unsigned int f2fs_usable_zone_blks_in_seg(struct f2fs_sb_info *sbi,
5574 							unsigned int segno)
5575 {
5576 	return 0;
5577 }
5578 
5579 #endif
f2fs_usable_blks_in_seg(struct f2fs_sb_info * sbi,unsigned int segno)5580 unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
5581 					unsigned int segno)
5582 {
5583 	if (f2fs_sb_has_blkzoned(sbi))
5584 		return f2fs_usable_zone_blks_in_seg(sbi, segno);
5585 
5586 	return BLKS_PER_SEG(sbi);
5587 }
5588 
f2fs_usable_segs_in_sec(struct f2fs_sb_info * sbi)5589 unsigned int f2fs_usable_segs_in_sec(struct f2fs_sb_info *sbi)
5590 {
5591 	if (f2fs_sb_has_blkzoned(sbi))
5592 		return CAP_SEGS_PER_SEC(sbi);
5593 
5594 	return SEGS_PER_SEC(sbi);
5595 }
5596 
f2fs_get_section_mtime(struct f2fs_sb_info * sbi,unsigned int segno)5597 unsigned long long f2fs_get_section_mtime(struct f2fs_sb_info *sbi,
5598 	unsigned int segno)
5599 {
5600 	unsigned int usable_segs_per_sec = f2fs_usable_segs_in_sec(sbi);
5601 	unsigned int secno = 0, start = 0;
5602 	unsigned int total_valid_blocks = 0;
5603 	unsigned long long mtime = 0;
5604 	unsigned int i = 0;
5605 
5606 	secno = GET_SEC_FROM_SEG(sbi, segno);
5607 	start = GET_SEG_FROM_SEC(sbi, secno);
5608 
5609 	if (!__is_large_section(sbi)) {
5610 		mtime = get_seg_entry(sbi, start + i)->mtime;
5611 		goto out;
5612 	}
5613 
5614 	for (i = 0; i < usable_segs_per_sec; i++) {
5615 		/* for large section, only check the mtime of valid segments */
5616 		struct seg_entry *se = get_seg_entry(sbi, start+i);
5617 
5618 		mtime += se->mtime * se->valid_blocks;
5619 		total_valid_blocks += se->valid_blocks;
5620 	}
5621 
5622 	if (total_valid_blocks == 0)
5623 		return INVALID_MTIME;
5624 
5625 	mtime = div_u64(mtime, total_valid_blocks);
5626 out:
5627 	if (unlikely(mtime == INVALID_MTIME))
5628 		mtime -= 1;
5629 	return mtime;
5630 }
5631 
5632 /*
5633  * Update min, max modified time for cost-benefit GC algorithm
5634  */
init_min_max_mtime(struct f2fs_sb_info * sbi)5635 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
5636 {
5637 	struct sit_info *sit_i = SIT_I(sbi);
5638 	unsigned int segno;
5639 
5640 	down_write(&sit_i->sentry_lock);
5641 
5642 	sit_i->min_mtime = ULLONG_MAX;
5643 
5644 	for (segno = 0; segno < MAIN_SEGS(sbi); segno += SEGS_PER_SEC(sbi)) {
5645 		unsigned long long mtime = 0;
5646 
5647 		mtime = f2fs_get_section_mtime(sbi, segno);
5648 
5649 		if (sit_i->min_mtime > mtime)
5650 			sit_i->min_mtime = mtime;
5651 	}
5652 	sit_i->max_mtime = get_mtime(sbi, false);
5653 	sit_i->dirty_max_mtime = 0;
5654 	up_write(&sit_i->sentry_lock);
5655 }
5656 
f2fs_build_segment_manager(struct f2fs_sb_info * sbi)5657 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
5658 {
5659 	struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
5660 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
5661 	struct f2fs_sm_info *sm_info;
5662 	int err;
5663 
5664 	sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
5665 	if (!sm_info)
5666 		return -ENOMEM;
5667 
5668 	/* init sm info */
5669 	sbi->sm_info = sm_info;
5670 	sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
5671 	sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
5672 	sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
5673 	sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
5674 	sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
5675 	sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
5676 	sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
5677 	sm_info->rec_prefree_segments = sm_info->main_segments *
5678 					DEF_RECLAIM_PREFREE_SEGMENTS / 100;
5679 	if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
5680 		sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
5681 
5682 	if (!f2fs_lfs_mode(sbi))
5683 		sm_info->ipu_policy = BIT(F2FS_IPU_FSYNC);
5684 	sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
5685 	sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
5686 	sm_info->min_seq_blocks = BLKS_PER_SEG(sbi);
5687 	sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
5688 	sm_info->min_ssr_sections = reserved_sections(sbi);
5689 
5690 	INIT_LIST_HEAD(&sm_info->sit_entry_set);
5691 
5692 	init_f2fs_rwsem(&sm_info->curseg_lock);
5693 
5694 	err = f2fs_create_flush_cmd_control(sbi);
5695 	if (err)
5696 		return err;
5697 
5698 	err = create_discard_cmd_control(sbi);
5699 	if (err)
5700 		return err;
5701 
5702 	err = build_sit_info(sbi);
5703 	if (err)
5704 		return err;
5705 	err = build_free_segmap(sbi);
5706 	if (err)
5707 		return err;
5708 	err = build_curseg(sbi);
5709 	if (err)
5710 		return err;
5711 
5712 	/* reinit free segmap based on SIT */
5713 	err = build_sit_entries(sbi);
5714 	if (err)
5715 		return err;
5716 
5717 	init_free_segmap(sbi);
5718 	err = build_dirty_segmap(sbi);
5719 	if (err)
5720 		return err;
5721 
5722 	err = sanity_check_curseg(sbi);
5723 	if (err)
5724 		return err;
5725 
5726 	init_min_max_mtime(sbi);
5727 	return 0;
5728 }
5729 
discard_dirty_segmap(struct f2fs_sb_info * sbi,enum dirty_type dirty_type)5730 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
5731 		enum dirty_type dirty_type)
5732 {
5733 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5734 
5735 	mutex_lock(&dirty_i->seglist_lock);
5736 	kvfree(dirty_i->dirty_segmap[dirty_type]);
5737 	dirty_i->nr_dirty[dirty_type] = 0;
5738 	mutex_unlock(&dirty_i->seglist_lock);
5739 }
5740 
destroy_victim_secmap(struct f2fs_sb_info * sbi)5741 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
5742 {
5743 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5744 
5745 	kvfree(dirty_i->pinned_secmap);
5746 	kvfree(dirty_i->victim_secmap);
5747 }
5748 
destroy_dirty_segmap(struct f2fs_sb_info * sbi)5749 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
5750 {
5751 	struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
5752 	int i;
5753 
5754 	if (!dirty_i)
5755 		return;
5756 
5757 	/* discard pre-free/dirty segments list */
5758 	for (i = 0; i < NR_DIRTY_TYPE; i++)
5759 		discard_dirty_segmap(sbi, i);
5760 
5761 	if (__is_large_section(sbi)) {
5762 		mutex_lock(&dirty_i->seglist_lock);
5763 		kvfree(dirty_i->dirty_secmap);
5764 		mutex_unlock(&dirty_i->seglist_lock);
5765 	}
5766 
5767 	destroy_victim_secmap(sbi);
5768 	SM_I(sbi)->dirty_info = NULL;
5769 	kfree(dirty_i);
5770 }
5771 
destroy_curseg(struct f2fs_sb_info * sbi)5772 static void destroy_curseg(struct f2fs_sb_info *sbi)
5773 {
5774 	struct curseg_info *array = SM_I(sbi)->curseg_array;
5775 	int i;
5776 
5777 	if (!array)
5778 		return;
5779 	SM_I(sbi)->curseg_array = NULL;
5780 	for (i = 0; i < NR_CURSEG_TYPE; i++) {
5781 		kfree(array[i].sum_blk);
5782 		kfree(array[i].journal);
5783 	}
5784 	kfree(array);
5785 }
5786 
destroy_free_segmap(struct f2fs_sb_info * sbi)5787 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
5788 {
5789 	struct free_segmap_info *free_i = SM_I(sbi)->free_info;
5790 
5791 	if (!free_i)
5792 		return;
5793 	SM_I(sbi)->free_info = NULL;
5794 	kvfree(free_i->free_segmap);
5795 	kvfree(free_i->free_secmap);
5796 	kfree(free_i);
5797 }
5798 
destroy_sit_info(struct f2fs_sb_info * sbi)5799 static void destroy_sit_info(struct f2fs_sb_info *sbi)
5800 {
5801 	struct sit_info *sit_i = SIT_I(sbi);
5802 
5803 	if (!sit_i)
5804 		return;
5805 
5806 	if (sit_i->sentries)
5807 		kvfree(sit_i->bitmap);
5808 	kfree(sit_i->tmp_map);
5809 
5810 	kvfree(sit_i->sentries);
5811 	kvfree(sit_i->sec_entries);
5812 	kvfree(sit_i->dirty_sentries_bitmap);
5813 
5814 	SM_I(sbi)->sit_info = NULL;
5815 	kfree(sit_i->sit_bitmap);
5816 #ifdef CONFIG_F2FS_CHECK_FS
5817 	kfree(sit_i->sit_bitmap_mir);
5818 	kvfree(sit_i->invalid_segmap);
5819 #endif
5820 	kfree(sit_i);
5821 }
5822 
f2fs_destroy_segment_manager(struct f2fs_sb_info * sbi)5823 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
5824 {
5825 	struct f2fs_sm_info *sm_info = SM_I(sbi);
5826 
5827 	if (!sm_info)
5828 		return;
5829 	f2fs_destroy_flush_cmd_control(sbi, true);
5830 	destroy_discard_cmd_control(sbi);
5831 	destroy_dirty_segmap(sbi);
5832 	destroy_curseg(sbi);
5833 	destroy_free_segmap(sbi);
5834 	destroy_sit_info(sbi);
5835 	sbi->sm_info = NULL;
5836 	kfree(sm_info);
5837 }
5838 
f2fs_create_segment_manager_caches(void)5839 int __init f2fs_create_segment_manager_caches(void)
5840 {
5841 	discard_entry_slab = f2fs_kmem_cache_create("f2fs_discard_entry",
5842 			sizeof(struct discard_entry));
5843 	if (!discard_entry_slab)
5844 		goto fail;
5845 
5846 	discard_cmd_slab = f2fs_kmem_cache_create("f2fs_discard_cmd",
5847 			sizeof(struct discard_cmd));
5848 	if (!discard_cmd_slab)
5849 		goto destroy_discard_entry;
5850 
5851 	sit_entry_set_slab = f2fs_kmem_cache_create("f2fs_sit_entry_set",
5852 			sizeof(struct sit_entry_set));
5853 	if (!sit_entry_set_slab)
5854 		goto destroy_discard_cmd;
5855 
5856 	revoke_entry_slab = f2fs_kmem_cache_create("f2fs_revoke_entry",
5857 			sizeof(struct revoke_entry));
5858 	if (!revoke_entry_slab)
5859 		goto destroy_sit_entry_set;
5860 	return 0;
5861 
5862 destroy_sit_entry_set:
5863 	kmem_cache_destroy(sit_entry_set_slab);
5864 destroy_discard_cmd:
5865 	kmem_cache_destroy(discard_cmd_slab);
5866 destroy_discard_entry:
5867 	kmem_cache_destroy(discard_entry_slab);
5868 fail:
5869 	return -ENOMEM;
5870 }
5871 
f2fs_destroy_segment_manager_caches(void)5872 void f2fs_destroy_segment_manager_caches(void)
5873 {
5874 	kmem_cache_destroy(sit_entry_set_slab);
5875 	kmem_cache_destroy(discard_cmd_slab);
5876 	kmem_cache_destroy(discard_entry_slab);
5877 	kmem_cache_destroy(revoke_entry_slab);
5878 }
5879