?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
drm/panfrost: Remove set but not used variable 'bo'
[linux/fpc-iii.git] / fs / f2fs / segment.c
blobcf0eb002cfd40ced0bd135ab9593f9d30df68af0
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/prefetch.h>
13 #include <linux/kthread.h>
14 #include <linux/swap.h>
15 #include <linux/timer.h>
16 #include <linux/freezer.h>
17 #include <linux/sched/signal.h>
18
19 #include "f2fs.h"
20 #include "segment.h"
21 #include "node.h"
22 #include "gc.h"
23 #include "trace.h"
24 #include <trace/events/f2fs.h>
25
26 #define __reverse_ffz(x) __reverse_ffs(~(x))
27
28 static struct kmem_cache *discard_entry_slab;
29 static struct kmem_cache *discard_cmd_slab;
30 static struct kmem_cache *sit_entry_set_slab;
31 static struct kmem_cache *inmem_entry_slab;
32
33 static unsigned long __reverse_ulong(unsigned char *str)
34 {
35 unsigned long tmp = 0;
36 int shift = 24, idx = 0;
37
38 #if BITS_PER_LONG == 64
39 shift = 56;
40 #endif
41 while (shift >= 0) {
42 tmp |= (unsigned long)str[idx++] << shift;
43 shift -= BITS_PER_BYTE;
44 }
45 return tmp;
46 }
47
48 /*
49 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
50 * MSB and LSB are reversed in a byte by f2fs_set_bit.
51 */
52 static inline unsigned long __reverse_ffs(unsigned long word)
53 {
54 int num = 0;
55
56 #if BITS_PER_LONG == 64
57 if ((word & 0xffffffff00000000UL) == 0)
58 num += 32;
59 else
60 word >>= 32;
61 #endif
62 if ((word & 0xffff0000) == 0)
63 num += 16;
64 else
65 word >>= 16;
66
67 if ((word & 0xff00) == 0)
68 num += 8;
69 else
70 word >>= 8;
71
72 if ((word & 0xf0) == 0)
73 num += 4;
74 else
75 word >>= 4;
76
77 if ((word & 0xc) == 0)
78 num += 2;
79 else
80 word >>= 2;
81
82 if ((word & 0x2) == 0)
83 num += 1;
84 return num;
85 }
86
87 /*
88 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
89 * f2fs_set_bit makes MSB and LSB reversed in a byte.
90 * @size must be integral times of unsigned long.
91 * Example:
92 * MSB <--> LSB
93 * f2fs_set_bit(0, bitmap) => 1000 0000
94 * f2fs_set_bit(7, bitmap) => 0000 0001
95 */
96 static unsigned long __find_rev_next_bit(const unsigned long *addr,
97 unsigned long size, unsigned long offset)
98 {
99 const unsigned long *p = addr + BIT_WORD(offset);
100 unsigned long result = size;
101 unsigned long tmp;
102
103 if (offset >= size)
104 return size;
105
106 size -= (offset & ~(BITS_PER_LONG - 1));
107 offset %= BITS_PER_LONG;
108
109 while (1) {
110 if (*p == 0)
111 goto pass;
112
113 tmp = __reverse_ulong((unsigned char *)p);
114
115 tmp &= ~0UL >> offset;
116 if (size < BITS_PER_LONG)
117 tmp &= (~0UL << (BITS_PER_LONG - size));
118 if (tmp)
119 goto found;
120 pass:
121 if (size <= BITS_PER_LONG)
122 break;
123 size -= BITS_PER_LONG;
124 offset = 0;
125 p++;
126 }
127 return result;
128 found:
129 return result - size + __reverse_ffs(tmp);
130 }
131
132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
133 unsigned long size, unsigned long offset)
134 {
135 const unsigned long *p = addr + BIT_WORD(offset);
136 unsigned long result = size;
137 unsigned long tmp;
138
139 if (offset >= size)
140 return size;
141
142 size -= (offset & ~(BITS_PER_LONG - 1));
143 offset %= BITS_PER_LONG;
144
145 while (1) {
146 if (*p == ~0UL)
147 goto pass;
148
149 tmp = __reverse_ulong((unsigned char *)p);
150
151 if (offset)
152 tmp |= ~0UL << (BITS_PER_LONG - offset);
153 if (size < BITS_PER_LONG)
154 tmp |= ~0UL >> size;
155 if (tmp != ~0UL)
156 goto found;
157 pass:
158 if (size <= BITS_PER_LONG)
159 break;
160 size -= BITS_PER_LONG;
161 offset = 0;
162 p++;
163 }
164 return result;
165 found:
166 return result - size + __reverse_ffz(tmp);
167 }
168
169 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
170 {
171 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
172 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
173 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
174
175 if (test_opt(sbi, LFS))
176 return false;
177 if (sbi->gc_mode == GC_URGENT)
178 return true;
179 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
180 return true;
181
182 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
183 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
184 }
185
186 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
187 {
188 struct inmem_pages *new;
189
190 f2fs_trace_pid(page);
191
192 f2fs_set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
193
194 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
195
196 /* add atomic page indices to the list */
197 new->page = page;
198 INIT_LIST_HEAD(&new->list);
199
200 /* increase reference count with clean state */
201 get_page(page);
202 mutex_lock(&F2FS_I(inode)->inmem_lock);
203 list_add_tail(&new->list, &F2FS_I(inode)->inmem_pages);
204 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
205 mutex_unlock(&F2FS_I(inode)->inmem_lock);
206
207 trace_f2fs_register_inmem_page(page, INMEM);
208 }
209
210 static int __revoke_inmem_pages(struct inode *inode,
211 struct list_head *head, bool drop, bool recover,
212 bool trylock)
213 {
214 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
215 struct inmem_pages *cur, *tmp;
216 int err = 0;
217
218 list_for_each_entry_safe(cur, tmp, head, list) {
219 struct page *page = cur->page;
220
221 if (drop)
222 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
223
224 if (trylock) {
225 /*
226 * to avoid deadlock in between page lock and
227 * inmem_lock.
228 */
229 if (!trylock_page(page))
230 continue;
231 } else {
232 lock_page(page);
233 }
234
235 f2fs_wait_on_page_writeback(page, DATA, true, true);
236
237 if (recover) {
238 struct dnode_of_data dn;
239 struct node_info ni;
240
241 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
242 retry:
243 set_new_dnode(&dn, inode, NULL, NULL, 0);
244 err = f2fs_get_dnode_of_data(&dn, page->index,
245 LOOKUP_NODE);
246 if (err) {
247 if (err == -ENOMEM) {
248 congestion_wait(BLK_RW_ASYNC, HZ/50);
249 cond_resched();
250 goto retry;
251 }
252 err = -EAGAIN;
253 goto next;
254 }
255
256 err = f2fs_get_node_info(sbi, dn.nid, &ni);
257 if (err) {
258 f2fs_put_dnode(&dn);
259 return err;
260 }
261
262 if (cur->old_addr == NEW_ADDR) {
263 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
264 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
265 } else
266 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
267 cur->old_addr, ni.version, true, true);
268 f2fs_put_dnode(&dn);
269 }
270 next:
271 /* we don't need to invalidate this in the sccessful status */
272 if (drop || recover) {
273 ClearPageUptodate(page);
274 clear_cold_data(page);
275 }
276 f2fs_clear_page_private(page);
277 f2fs_put_page(page, 1);
278
279 list_del(&cur->list);
280 kmem_cache_free(inmem_entry_slab, cur);
281 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
282 }
283 return err;
284 }
285
286 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
287 {
288 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
289 struct inode *inode;
290 struct f2fs_inode_info *fi;
291 unsigned int count = sbi->atomic_files;
292 unsigned int looped = 0;
293 next:
294 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
295 if (list_empty(head)) {
296 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
297 return;
298 }
299 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
300 inode = igrab(&fi->vfs_inode);
301 if (inode)
302 list_move_tail(&fi->inmem_ilist, head);
303 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
304
305 if (inode) {
306 if (gc_failure) {
307 if (!fi->i_gc_failures[GC_FAILURE_ATOMIC])
308 goto skip;
309 }
310 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
311 f2fs_drop_inmem_pages(inode);
312 skip:
313 iput(inode);
314 }
315 congestion_wait(BLK_RW_ASYNC, HZ/50);
316 cond_resched();
317 if (gc_failure) {
318 if (++looped >= count)
319 return;
320 }
321 goto next;
322 }
323
324 void f2fs_drop_inmem_pages(struct inode *inode)
325 {
326 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
327 struct f2fs_inode_info *fi = F2FS_I(inode);
328
329 while (!list_empty(&fi->inmem_pages)) {
330 mutex_lock(&fi->inmem_lock);
331 __revoke_inmem_pages(inode, &fi->inmem_pages,
332 true, false, true);
333 mutex_unlock(&fi->inmem_lock);
334 }
335
336 fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
337
338 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
339 if (!list_empty(&fi->inmem_ilist))
340 list_del_init(&fi->inmem_ilist);
341 if (f2fs_is_atomic_file(inode)) {
342 clear_inode_flag(inode, FI_ATOMIC_FILE);
343 sbi->atomic_files--;
344 }
345 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
346 }
347
348 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
349 {
350 struct f2fs_inode_info *fi = F2FS_I(inode);
351 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
352 struct list_head *head = &fi->inmem_pages;
353 struct inmem_pages *cur = NULL;
354
355 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
356
357 mutex_lock(&fi->inmem_lock);
358 list_for_each_entry(cur, head, list) {
359 if (cur->page == page)
360 break;
361 }
362
363 f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
364 list_del(&cur->list);
365 mutex_unlock(&fi->inmem_lock);
366
367 dec_page_count(sbi, F2FS_INMEM_PAGES);
368 kmem_cache_free(inmem_entry_slab, cur);
369
370 ClearPageUptodate(page);
371 f2fs_clear_page_private(page);
372 f2fs_put_page(page, 0);
373
374 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
375 }
376
377 static int __f2fs_commit_inmem_pages(struct inode *inode)
378 {
379 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
380 struct f2fs_inode_info *fi = F2FS_I(inode);
381 struct inmem_pages *cur, *tmp;
382 struct f2fs_io_info fio = {
383 .sbi = sbi,
384 .ino = inode->i_ino,
385 .type = DATA,
386 .op = REQ_OP_WRITE,
387 .op_flags = REQ_SYNC | REQ_PRIO,
388 .io_type = FS_DATA_IO,
389 };
390 struct list_head revoke_list;
391 bool submit_bio = false;
392 int err = 0;
393
394 INIT_LIST_HEAD(&revoke_list);
395
396 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
397 struct page *page = cur->page;
398
399 lock_page(page);
400 if (page->mapping == inode->i_mapping) {
401 trace_f2fs_commit_inmem_page(page, INMEM);
402
403 f2fs_wait_on_page_writeback(page, DATA, true, true);
404
405 set_page_dirty(page);
406 if (clear_page_dirty_for_io(page)) {
407 inode_dec_dirty_pages(inode);
408 f2fs_remove_dirty_inode(inode);
409 }
410 retry:
411 fio.page = page;
412 fio.old_blkaddr = NULL_ADDR;
413 fio.encrypted_page = NULL;
414 fio.need_lock = LOCK_DONE;
415 err = f2fs_do_write_data_page(&fio);
416 if (err) {
417 if (err == -ENOMEM) {
418 congestion_wait(BLK_RW_ASYNC, HZ/50);
419 cond_resched();
420 goto retry;
421 }
422 unlock_page(page);
423 break;
424 }
425 /* record old blkaddr for revoking */
426 cur->old_addr = fio.old_blkaddr;
427 submit_bio = true;
428 }
429 unlock_page(page);
430 list_move_tail(&cur->list, &revoke_list);
431 }
432
433 if (submit_bio)
434 f2fs_submit_merged_write_cond(sbi, inode, NULL, 0, DATA);
435
436 if (err) {
437 /*
438 * try to revoke all committed pages, but still we could fail
439 * due to no memory or other reason, if that happened, EAGAIN
440 * will be returned, which means in such case, transaction is
441 * already not integrity, caller should use journal to do the
442 * recovery or rewrite & commit last transaction. For other
443 * error number, revoking was done by filesystem itself.
444 */
445 err = __revoke_inmem_pages(inode, &revoke_list,
446 false, true, false);
447
448 /* drop all uncommitted pages */
449 __revoke_inmem_pages(inode, &fi->inmem_pages,
450 true, false, false);
451 } else {
452 __revoke_inmem_pages(inode, &revoke_list,
453 false, false, false);
454 }
455
456 return err;
457 }
458
459 int f2fs_commit_inmem_pages(struct inode *inode)
460 {
461 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
462 struct f2fs_inode_info *fi = F2FS_I(inode);
463 int err;
464
465 f2fs_balance_fs(sbi, true);
466
467 down_write(&fi->i_gc_rwsem[WRITE]);
468
469 f2fs_lock_op(sbi);
470 set_inode_flag(inode, FI_ATOMIC_COMMIT);
471
472 mutex_lock(&fi->inmem_lock);
473 err = __f2fs_commit_inmem_pages(inode);
474 mutex_unlock(&fi->inmem_lock);
475
476 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
477
478 f2fs_unlock_op(sbi);
479 up_write(&fi->i_gc_rwsem[WRITE]);
480
481 return err;
482 }
483
484 /*
485 * This function balances dirty node and dentry pages.
486 * In addition, it controls garbage collection.
487 */
488 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
489 {
490 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
491 f2fs_show_injection_info(sbi, FAULT_CHECKPOINT);
492 f2fs_stop_checkpoint(sbi, false);
493 }
494
495 /* balance_fs_bg is able to be pending */
496 if (need && excess_cached_nats(sbi))
497 f2fs_balance_fs_bg(sbi);
498
499 if (!f2fs_is_checkpoint_ready(sbi))
500 return;
501
502 /*
503 * We should do GC or end up with checkpoint, if there are so many dirty
504 * dir/node pages without enough free segments.
505 */
506 if (has_not_enough_free_secs(sbi, 0, 0)) {
507 down_write(&sbi->gc_lock);
508 f2fs_gc(sbi, false, false, NULL_SEGNO);
509 }
510 }
511
512 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
513 {
514 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
515 return;
516
517 /* try to shrink extent cache when there is no enough memory */
518 if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
519 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
520
521 /* check the # of cached NAT entries */
522 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
523 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
524
525 if (!f2fs_available_free_memory(sbi, FREE_NIDS))
526 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
527 else
528 f2fs_build_free_nids(sbi, false, false);
529
530 if (!is_idle(sbi, REQ_TIME) &&
531 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
532 return;
533
534 /* checkpoint is the only way to shrink partial cached entries */
535 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
536 !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
537 excess_prefree_segs(sbi) ||
538 excess_dirty_nats(sbi) ||
539 excess_dirty_nodes(sbi) ||
540 f2fs_time_over(sbi, CP_TIME)) {
541 if (test_opt(sbi, DATA_FLUSH)) {
542 struct blk_plug plug;
543
544 mutex_lock(&sbi->flush_lock);
545
546 blk_start_plug(&plug);
547 f2fs_sync_dirty_inodes(sbi, FILE_INODE);
548 blk_finish_plug(&plug);
549
550 mutex_unlock(&sbi->flush_lock);
551 }
552 f2fs_sync_fs(sbi->sb, true);
553 stat_inc_bg_cp_count(sbi->stat_info);
554 }
555 }
556
557 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
558 struct block_device *bdev)
559 {
560 struct bio *bio;
561 int ret;
562
563 bio = f2fs_bio_alloc(sbi, 0, false);
564 if (!bio)
565 return -ENOMEM;
566
567 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
568 bio_set_dev(bio, bdev);
569 ret = submit_bio_wait(bio);
570 bio_put(bio);
571
572 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
573 test_opt(sbi, FLUSH_MERGE), ret);
574 return ret;
575 }
576
577 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
578 {
579 int ret = 0;
580 int i;
581
582 if (!f2fs_is_multi_device(sbi))
583 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
584
585 for (i = 0; i < sbi->s_ndevs; i++) {
586 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
587 continue;
588 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
589 if (ret)
590 break;
591 }
592 return ret;
593 }
594
595 static int issue_flush_thread(void *data)
596 {
597 struct f2fs_sb_info *sbi = data;
598 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
599 wait_queue_head_t *q = &fcc->flush_wait_queue;
600 repeat:
601 if (kthread_should_stop())
602 return 0;
603
604 sb_start_intwrite(sbi->sb);
605
606 if (!llist_empty(&fcc->issue_list)) {
607 struct flush_cmd *cmd, *next;
608 int ret;
609
610 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
611 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
612
613 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
614
615 ret = submit_flush_wait(sbi, cmd->ino);
616 atomic_inc(&fcc->issued_flush);
617
618 llist_for_each_entry_safe(cmd, next,
619 fcc->dispatch_list, llnode) {
620 cmd->ret = ret;
621 complete(&cmd->wait);
622 }
623 fcc->dispatch_list = NULL;
624 }
625
626 sb_end_intwrite(sbi->sb);
627
628 wait_event_interruptible(*q,
629 kthread_should_stop() || !llist_empty(&fcc->issue_list));
630 goto repeat;
631 }
632
633 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
634 {
635 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
636 struct flush_cmd cmd;
637 int ret;
638
639 if (test_opt(sbi, NOBARRIER))
640 return 0;
641
642 if (!test_opt(sbi, FLUSH_MERGE)) {
643 atomic_inc(&fcc->queued_flush);
644 ret = submit_flush_wait(sbi, ino);
645 atomic_dec(&fcc->queued_flush);
646 atomic_inc(&fcc->issued_flush);
647 return ret;
648 }
649
650 if (atomic_inc_return(&fcc->queued_flush) == 1 ||
651 f2fs_is_multi_device(sbi)) {
652 ret = submit_flush_wait(sbi, ino);
653 atomic_dec(&fcc->queued_flush);
654
655 atomic_inc(&fcc->issued_flush);
656 return ret;
657 }
658
659 cmd.ino = ino;
660 init_completion(&cmd.wait);
661
662 llist_add(&cmd.llnode, &fcc->issue_list);
663
664 /* update issue_list before we wake up issue_flush thread */
665 smp_mb();
666
667 if (waitqueue_active(&fcc->flush_wait_queue))
668 wake_up(&fcc->flush_wait_queue);
669
670 if (fcc->f2fs_issue_flush) {
671 wait_for_completion(&cmd.wait);
672 atomic_dec(&fcc->queued_flush);
673 } else {
674 struct llist_node *list;
675
676 list = llist_del_all(&fcc->issue_list);
677 if (!list) {
678 wait_for_completion(&cmd.wait);
679 atomic_dec(&fcc->queued_flush);
680 } else {
681 struct flush_cmd *tmp, *next;
682
683 ret = submit_flush_wait(sbi, ino);
684
685 llist_for_each_entry_safe(tmp, next, list, llnode) {
686 if (tmp == &cmd) {
687 cmd.ret = ret;
688 atomic_dec(&fcc->queued_flush);
689 continue;
690 }
691 tmp->ret = ret;
692 complete(&tmp->wait);
693 }
694 }
695 }
696
697 return cmd.ret;
698 }
699
700 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
701 {
702 dev_t dev = sbi->sb->s_bdev->bd_dev;
703 struct flush_cmd_control *fcc;
704 int err = 0;
705
706 if (SM_I(sbi)->fcc_info) {
707 fcc = SM_I(sbi)->fcc_info;
708 if (fcc->f2fs_issue_flush)
709 return err;
710 goto init_thread;
711 }
712
713 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
714 if (!fcc)
715 return -ENOMEM;
716 atomic_set(&fcc->issued_flush, 0);
717 atomic_set(&fcc->queued_flush, 0);
718 init_waitqueue_head(&fcc->flush_wait_queue);
719 init_llist_head(&fcc->issue_list);
720 SM_I(sbi)->fcc_info = fcc;
721 if (!test_opt(sbi, FLUSH_MERGE))
722 return err;
723
724 init_thread:
725 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
726 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
727 if (IS_ERR(fcc->f2fs_issue_flush)) {
728 err = PTR_ERR(fcc->f2fs_issue_flush);
729 kvfree(fcc);
730 SM_I(sbi)->fcc_info = NULL;
731 return err;
732 }
733
734 return err;
735 }
736
737 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
738 {
739 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
740
741 if (fcc && fcc->f2fs_issue_flush) {
742 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
743
744 fcc->f2fs_issue_flush = NULL;
745 kthread_stop(flush_thread);
746 }
747 if (free) {
748 kvfree(fcc);
749 SM_I(sbi)->fcc_info = NULL;
750 }
751 }
752
753 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
754 {
755 int ret = 0, i;
756
757 if (!f2fs_is_multi_device(sbi))
758 return 0;
759
760 for (i = 1; i < sbi->s_ndevs; i++) {
761 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
762 continue;
763 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
764 if (ret)
765 break;
766
767 spin_lock(&sbi->dev_lock);
768 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
769 spin_unlock(&sbi->dev_lock);
770 }
771
772 return ret;
773 }
774
775 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
776 enum dirty_type dirty_type)
777 {
778 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
779
780 /* need not be added */
781 if (IS_CURSEG(sbi, segno))
782 return;
783
784 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
785 dirty_i->nr_dirty[dirty_type]++;
786
787 if (dirty_type == DIRTY) {
788 struct seg_entry *sentry = get_seg_entry(sbi, segno);
789 enum dirty_type t = sentry->type;
790
791 if (unlikely(t >= DIRTY)) {
792 f2fs_bug_on(sbi, 1);
793 return;
794 }
795 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
796 dirty_i->nr_dirty[t]++;
797 }
798 }
799
800 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
801 enum dirty_type dirty_type)
802 {
803 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
804
805 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
806 dirty_i->nr_dirty[dirty_type]--;
807
808 if (dirty_type == DIRTY) {
809 struct seg_entry *sentry = get_seg_entry(sbi, segno);
810 enum dirty_type t = sentry->type;
811
812 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
813 dirty_i->nr_dirty[t]--;
814
815 if (get_valid_blocks(sbi, segno, true) == 0) {
816 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
817 dirty_i->victim_secmap);
818 #ifdef CONFIG_F2FS_CHECK_FS
819 clear_bit(segno, SIT_I(sbi)->invalid_segmap);
820 #endif
821 }
822 }
823 }
824
825 /*
826 * Should not occur error such as -ENOMEM.
827 * Adding dirty entry into seglist is not critical operation.
828 * If a given segment is one of current working segments, it won't be added.
829 */
830 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
831 {
832 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
833 unsigned short valid_blocks, ckpt_valid_blocks;
834
835 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
836 return;
837
838 mutex_lock(&dirty_i->seglist_lock);
839
840 valid_blocks = get_valid_blocks(sbi, segno, false);
841 ckpt_valid_blocks = get_ckpt_valid_blocks(sbi, segno);
842
843 if (valid_blocks == 0 && (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) ||
844 ckpt_valid_blocks == sbi->blocks_per_seg)) {
845 __locate_dirty_segment(sbi, segno, PRE);
846 __remove_dirty_segment(sbi, segno, DIRTY);
847 } else if (valid_blocks < sbi->blocks_per_seg) {
848 __locate_dirty_segment(sbi, segno, DIRTY);
849 } else {
850 /* Recovery routine with SSR needs this */
851 __remove_dirty_segment(sbi, segno, DIRTY);
852 }
853
854 mutex_unlock(&dirty_i->seglist_lock);
855 }
856
857 /* This moves currently empty dirty blocks to prefree. Must hold seglist_lock */
858 void f2fs_dirty_to_prefree(struct f2fs_sb_info *sbi)
859 {
860 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
861 unsigned int segno;
862
863 mutex_lock(&dirty_i->seglist_lock);
864 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
865 if (get_valid_blocks(sbi, segno, false))
866 continue;
867 if (IS_CURSEG(sbi, segno))
868 continue;
869 __locate_dirty_segment(sbi, segno, PRE);
870 __remove_dirty_segment(sbi, segno, DIRTY);
871 }
872 mutex_unlock(&dirty_i->seglist_lock);
873 }
874
875 block_t f2fs_get_unusable_blocks(struct f2fs_sb_info *sbi)
876 {
877 int ovp_hole_segs =
878 (overprovision_segments(sbi) - reserved_segments(sbi));
879 block_t ovp_holes = ovp_hole_segs << sbi->log_blocks_per_seg;
880 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
881 block_t holes[2] = {0, 0}; /* DATA and NODE */
882 block_t unusable;
883 struct seg_entry *se;
884 unsigned int segno;
885
886 mutex_lock(&dirty_i->seglist_lock);
887 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
888 se = get_seg_entry(sbi, segno);
889 if (IS_NODESEG(se->type))
890 holes[NODE] += sbi->blocks_per_seg - se->valid_blocks;
891 else
892 holes[DATA] += sbi->blocks_per_seg - se->valid_blocks;
893 }
894 mutex_unlock(&dirty_i->seglist_lock);
895
896 unusable = holes[DATA] > holes[NODE] ? holes[DATA] : holes[NODE];
897 if (unusable > ovp_holes)
898 return unusable - ovp_holes;
899 return 0;
900 }
901
902 int f2fs_disable_cp_again(struct f2fs_sb_info *sbi, block_t unusable)
903 {
904 int ovp_hole_segs =
905 (overprovision_segments(sbi) - reserved_segments(sbi));
906 if (unusable > F2FS_OPTION(sbi).unusable_cap)
907 return -EAGAIN;
908 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK) &&
909 dirty_segments(sbi) > ovp_hole_segs)
910 return -EAGAIN;
911 return 0;
912 }
913
914 /* This is only used by SBI_CP_DISABLED */
915 static unsigned int get_free_segment(struct f2fs_sb_info *sbi)
916 {
917 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
918 unsigned int segno = 0;
919
920 mutex_lock(&dirty_i->seglist_lock);
921 for_each_set_bit(segno, dirty_i->dirty_segmap[DIRTY], MAIN_SEGS(sbi)) {
922 if (get_valid_blocks(sbi, segno, false))
923 continue;
924 if (get_ckpt_valid_blocks(sbi, segno))
925 continue;
926 mutex_unlock(&dirty_i->seglist_lock);
927 return segno;
928 }
929 mutex_unlock(&dirty_i->seglist_lock);
930 return NULL_SEGNO;
931 }
932
933 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
934 struct block_device *bdev, block_t lstart,
935 block_t start, block_t len)
936 {
937 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
938 struct list_head *pend_list;
939 struct discard_cmd *dc;
940
941 f2fs_bug_on(sbi, !len);
942
943 pend_list = &dcc->pend_list[plist_idx(len)];
944
945 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
946 INIT_LIST_HEAD(&dc->list);
947 dc->bdev = bdev;
948 dc->lstart = lstart;
949 dc->start = start;
950 dc->len = len;
951 dc->ref = 0;
952 dc->state = D_PREP;
953 dc->queued = 0;
954 dc->error = 0;
955 init_completion(&dc->wait);
956 list_add_tail(&dc->list, pend_list);
957 spin_lock_init(&dc->lock);
958 dc->bio_ref = 0;
959 atomic_inc(&dcc->discard_cmd_cnt);
960 dcc->undiscard_blks += len;
961
962 return dc;
963 }
964
965 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
966 struct block_device *bdev, block_t lstart,
967 block_t start, block_t len,
968 struct rb_node *parent, struct rb_node **p,
969 bool leftmost)
970 {
971 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
972 struct discard_cmd *dc;
973
974 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
975
976 rb_link_node(&dc->rb_node, parent, p);
977 rb_insert_color_cached(&dc->rb_node, &dcc->root, leftmost);
978
979 return dc;
980 }
981
982 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
983 struct discard_cmd *dc)
984 {
985 if (dc->state == D_DONE)
986 atomic_sub(dc->queued, &dcc->queued_discard);
987
988 list_del(&dc->list);
989 rb_erase_cached(&dc->rb_node, &dcc->root);
990 dcc->undiscard_blks -= dc->len;
991
992 kmem_cache_free(discard_cmd_slab, dc);
993
994 atomic_dec(&dcc->discard_cmd_cnt);
995 }
996
997 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
998 struct discard_cmd *dc)
999 {
1000 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1001 unsigned long flags;
1002
1003 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
1004
1005 spin_lock_irqsave(&dc->lock, flags);
1006 if (dc->bio_ref) {
1007 spin_unlock_irqrestore(&dc->lock, flags);
1008 return;
1009 }
1010 spin_unlock_irqrestore(&dc->lock, flags);
1011
1012 f2fs_bug_on(sbi, dc->ref);
1013
1014 if (dc->error == -EOPNOTSUPP)
1015 dc->error = 0;
1016
1017 if (dc->error)
1018 printk_ratelimited(
1019 "%sF2FS-fs (%s): Issue discard(%u, %u, %u) failed, ret: %d",
1020 KERN_INFO, sbi->sb->s_id,
1021 dc->lstart, dc->start, dc->len, dc->error);
1022 __detach_discard_cmd(dcc, dc);
1023 }
1024
1025 static void f2fs_submit_discard_endio(struct bio *bio)
1026 {
1027 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
1028 unsigned long flags;
1029
1030 dc->error = blk_status_to_errno(bio->bi_status);
1031
1032 spin_lock_irqsave(&dc->lock, flags);
1033 dc->bio_ref--;
1034 if (!dc->bio_ref && dc->state == D_SUBMIT) {
1035 dc->state = D_DONE;
1036 complete_all(&dc->wait);
1037 }
1038 spin_unlock_irqrestore(&dc->lock, flags);
1039 bio_put(bio);
1040 }
1041
1042 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
1043 block_t start, block_t end)
1044 {
1045 #ifdef CONFIG_F2FS_CHECK_FS
1046 struct seg_entry *sentry;
1047 unsigned int segno;
1048 block_t blk = start;
1049 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
1050 unsigned long *map;
1051
1052 while (blk < end) {
1053 segno = GET_SEGNO(sbi, blk);
1054 sentry = get_seg_entry(sbi, segno);
1055 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
1056
1057 if (end < START_BLOCK(sbi, segno + 1))
1058 size = GET_BLKOFF_FROM_SEG0(sbi, end);
1059 else
1060 size = max_blocks;
1061 map = (unsigned long *)(sentry->cur_valid_map);
1062 offset = __find_rev_next_bit(map, size, offset);
1063 f2fs_bug_on(sbi, offset != size);
1064 blk = START_BLOCK(sbi, segno + 1);
1065 }
1066 #endif
1067 }
1068
1069 static void __init_discard_policy(struct f2fs_sb_info *sbi,
1070 struct discard_policy *dpolicy,
1071 int discard_type, unsigned int granularity)
1072 {
1073 /* common policy */
1074 dpolicy->type = discard_type;
1075 dpolicy->sync = true;
1076 dpolicy->ordered = false;
1077 dpolicy->granularity = granularity;
1078
1079 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1080 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1081 dpolicy->timeout = 0;
1082
1083 if (discard_type == DPOLICY_BG) {
1084 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1085 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1086 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1087 dpolicy->io_aware = true;
1088 dpolicy->sync = false;
1089 dpolicy->ordered = true;
1090 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
1091 dpolicy->granularity = 1;
1092 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1093 }
1094 } else if (discard_type == DPOLICY_FORCE) {
1095 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1096 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
1097 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1098 dpolicy->io_aware = false;
1099 } else if (discard_type == DPOLICY_FSTRIM) {
1100 dpolicy->io_aware = false;
1101 } else if (discard_type == DPOLICY_UMOUNT) {
1102 dpolicy->max_requests = UINT_MAX;
1103 dpolicy->io_aware = false;
1104 /* we need to issue all to keep CP_TRIMMED_FLAG */
1105 dpolicy->granularity = 1;
1106 }
1107 }
1108
1109 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1110 struct block_device *bdev, block_t lstart,
1111 block_t start, block_t len);
1112 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
1113 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
1114 struct discard_policy *dpolicy,
1115 struct discard_cmd *dc,
1116 unsigned int *issued)
1117 {
1118 struct block_device *bdev = dc->bdev;
1119 struct request_queue *q = bdev_get_queue(bdev);
1120 unsigned int max_discard_blocks =
1121 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1122 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1123 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1124 &(dcc->fstrim_list) : &(dcc->wait_list);
1125 int flag = dpolicy->sync ? REQ_SYNC : 0;
1126 block_t lstart, start, len, total_len;
1127 int err = 0;
1128
1129 if (dc->state != D_PREP)
1130 return 0;
1131
1132 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1133 return 0;
1134
1135 trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1136
1137 lstart = dc->lstart;
1138 start = dc->start;
1139 len = dc->len;
1140 total_len = len;
1141
1142 dc->len = 0;
1143
1144 while (total_len && *issued < dpolicy->max_requests && !err) {
1145 struct bio *bio = NULL;
1146 unsigned long flags;
1147 bool last = true;
1148
1149 if (len > max_discard_blocks) {
1150 len = max_discard_blocks;
1151 last = false;
1152 }
1153
1154 (*issued)++;
1155 if (*issued == dpolicy->max_requests)
1156 last = true;
1157
1158 dc->len += len;
1159
1160 if (time_to_inject(sbi, FAULT_DISCARD)) {
1161 f2fs_show_injection_info(sbi, FAULT_DISCARD);
1162 err = -EIO;
1163 goto submit;
1164 }
1165 err = __blkdev_issue_discard(bdev,
1166 SECTOR_FROM_BLOCK(start),
1167 SECTOR_FROM_BLOCK(len),
1168 GFP_NOFS, 0, &bio);
1169 submit:
1170 if (err) {
1171 spin_lock_irqsave(&dc->lock, flags);
1172 if (dc->state == D_PARTIAL)
1173 dc->state = D_SUBMIT;
1174 spin_unlock_irqrestore(&dc->lock, flags);
1175
1176 break;
1177 }
1178
1179 f2fs_bug_on(sbi, !bio);
1180
1181 /*
1182 * should keep before submission to avoid D_DONE
1183 * right away
1184 */
1185 spin_lock_irqsave(&dc->lock, flags);
1186 if (last)
1187 dc->state = D_SUBMIT;
1188 else
1189 dc->state = D_PARTIAL;
1190 dc->bio_ref++;
1191 spin_unlock_irqrestore(&dc->lock, flags);
1192
1193 atomic_inc(&dcc->queued_discard);
1194 dc->queued++;
1195 list_move_tail(&dc->list, wait_list);
1196
1197 /* sanity check on discard range */
1198 __check_sit_bitmap(sbi, lstart, lstart + len);
1199
1200 bio->bi_private = dc;
1201 bio->bi_end_io = f2fs_submit_discard_endio;
1202 bio->bi_opf |= flag;
1203 submit_bio(bio);
1204
1205 atomic_inc(&dcc->issued_discard);
1206
1207 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1208
1209 lstart += len;
1210 start += len;
1211 total_len -= len;
1212 len = total_len;
1213 }
1214
1215 if (!err && len)
1216 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1217 return err;
1218 }
1219
1220 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1221 struct block_device *bdev, block_t lstart,
1222 block_t start, block_t len,
1223 struct rb_node **insert_p,
1224 struct rb_node *insert_parent)
1225 {
1226 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1227 struct rb_node **p;
1228 struct rb_node *parent = NULL;
1229 struct discard_cmd *dc = NULL;
1230 bool leftmost = true;
1231
1232 if (insert_p && insert_parent) {
1233 parent = insert_parent;
1234 p = insert_p;
1235 goto do_insert;
1236 }
1237
1238 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent,
1239 lstart, &leftmost);
1240 do_insert:
1241 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent,
1242 p, leftmost);
1243 if (!dc)
1244 return NULL;
1245
1246 return dc;
1247 }
1248
1249 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1250 struct discard_cmd *dc)
1251 {
1252 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1253 }
1254
1255 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1256 struct discard_cmd *dc, block_t blkaddr)
1257 {
1258 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1259 struct discard_info di = dc->di;
1260 bool modified = false;
1261
1262 if (dc->state == D_DONE || dc->len == 1) {
1263 __remove_discard_cmd(sbi, dc);
1264 return;
1265 }
1266
1267 dcc->undiscard_blks -= di.len;
1268
1269 if (blkaddr > di.lstart) {
1270 dc->len = blkaddr - dc->lstart;
1271 dcc->undiscard_blks += dc->len;
1272 __relocate_discard_cmd(dcc, dc);
1273 modified = true;
1274 }
1275
1276 if (blkaddr < di.lstart + di.len - 1) {
1277 if (modified) {
1278 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1279 di.start + blkaddr + 1 - di.lstart,
1280 di.lstart + di.len - 1 - blkaddr,
1281 NULL, NULL);
1282 } else {
1283 dc->lstart++;
1284 dc->len--;
1285 dc->start++;
1286 dcc->undiscard_blks += dc->len;
1287 __relocate_discard_cmd(dcc, dc);
1288 }
1289 }
1290 }
1291
1292 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1293 struct block_device *bdev, block_t lstart,
1294 block_t start, block_t len)
1295 {
1296 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1297 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1298 struct discard_cmd *dc;
1299 struct discard_info di = {0};
1300 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1301 struct request_queue *q = bdev_get_queue(bdev);
1302 unsigned int max_discard_blocks =
1303 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1304 block_t end = lstart + len;
1305
1306 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1307 NULL, lstart,
1308 (struct rb_entry **)&prev_dc,
1309 (struct rb_entry **)&next_dc,
1310 &insert_p, &insert_parent, true, NULL);
1311 if (dc)
1312 prev_dc = dc;
1313
1314 if (!prev_dc) {
1315 di.lstart = lstart;
1316 di.len = next_dc ? next_dc->lstart - lstart : len;
1317 di.len = min(di.len, len);
1318 di.start = start;
1319 }
1320
1321 while (1) {
1322 struct rb_node *node;
1323 bool merged = false;
1324 struct discard_cmd *tdc = NULL;
1325
1326 if (prev_dc) {
1327 di.lstart = prev_dc->lstart + prev_dc->len;
1328 if (di.lstart < lstart)
1329 di.lstart = lstart;
1330 if (di.lstart >= end)
1331 break;
1332
1333 if (!next_dc || next_dc->lstart > end)
1334 di.len = end - di.lstart;
1335 else
1336 di.len = next_dc->lstart - di.lstart;
1337 di.start = start + di.lstart - lstart;
1338 }
1339
1340 if (!di.len)
1341 goto next;
1342
1343 if (prev_dc && prev_dc->state == D_PREP &&
1344 prev_dc->bdev == bdev &&
1345 __is_discard_back_mergeable(&di, &prev_dc->di,
1346 max_discard_blocks)) {
1347 prev_dc->di.len += di.len;
1348 dcc->undiscard_blks += di.len;
1349 __relocate_discard_cmd(dcc, prev_dc);
1350 di = prev_dc->di;
1351 tdc = prev_dc;
1352 merged = true;
1353 }
1354
1355 if (next_dc && next_dc->state == D_PREP &&
1356 next_dc->bdev == bdev &&
1357 __is_discard_front_mergeable(&di, &next_dc->di,
1358 max_discard_blocks)) {
1359 next_dc->di.lstart = di.lstart;
1360 next_dc->di.len += di.len;
1361 next_dc->di.start = di.start;
1362 dcc->undiscard_blks += di.len;
1363 __relocate_discard_cmd(dcc, next_dc);
1364 if (tdc)
1365 __remove_discard_cmd(sbi, tdc);
1366 merged = true;
1367 }
1368
1369 if (!merged) {
1370 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1371 di.len, NULL, NULL);
1372 }
1373 next:
1374 prev_dc = next_dc;
1375 if (!prev_dc)
1376 break;
1377
1378 node = rb_next(&prev_dc->rb_node);
1379 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1380 }
1381 }
1382
1383 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1384 struct block_device *bdev, block_t blkstart, block_t blklen)
1385 {
1386 block_t lblkstart = blkstart;
1387
1388 if (!f2fs_bdev_support_discard(bdev))
1389 return 0;
1390
1391 trace_f2fs_queue_discard(bdev, blkstart, blklen);
1392
1393 if (f2fs_is_multi_device(sbi)) {
1394 int devi = f2fs_target_device_index(sbi, blkstart);
1395
1396 blkstart -= FDEV(devi).start_blk;
1397 }
1398 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1399 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1400 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1401 return 0;
1402 }
1403
1404 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1405 struct discard_policy *dpolicy)
1406 {
1407 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1408 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1409 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1410 struct discard_cmd *dc;
1411 struct blk_plug plug;
1412 unsigned int pos = dcc->next_pos;
1413 unsigned int issued = 0;
1414 bool io_interrupted = false;
1415
1416 mutex_lock(&dcc->cmd_lock);
1417 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1418 NULL, pos,
1419 (struct rb_entry **)&prev_dc,
1420 (struct rb_entry **)&next_dc,
1421 &insert_p, &insert_parent, true, NULL);
1422 if (!dc)
1423 dc = next_dc;
1424
1425 blk_start_plug(&plug);
1426
1427 while (dc) {
1428 struct rb_node *node;
1429 int err = 0;
1430
1431 if (dc->state != D_PREP)
1432 goto next;
1433
1434 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1435 io_interrupted = true;
1436 break;
1437 }
1438
1439 dcc->next_pos = dc->lstart + dc->len;
1440 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1441
1442 if (issued >= dpolicy->max_requests)
1443 break;
1444 next:
1445 node = rb_next(&dc->rb_node);
1446 if (err)
1447 __remove_discard_cmd(sbi, dc);
1448 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1449 }
1450
1451 blk_finish_plug(&plug);
1452
1453 if (!dc)
1454 dcc->next_pos = 0;
1455
1456 mutex_unlock(&dcc->cmd_lock);
1457
1458 if (!issued && io_interrupted)
1459 issued = -1;
1460
1461 return issued;
1462 }
1463
1464 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1465 struct discard_policy *dpolicy)
1466 {
1467 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1468 struct list_head *pend_list;
1469 struct discard_cmd *dc, *tmp;
1470 struct blk_plug plug;
1471 int i, issued = 0;
1472 bool io_interrupted = false;
1473
1474 if (dpolicy->timeout != 0)
1475 f2fs_update_time(sbi, dpolicy->timeout);
1476
1477 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1478 if (dpolicy->timeout != 0 &&
1479 f2fs_time_over(sbi, dpolicy->timeout))
1480 break;
1481
1482 if (i + 1 < dpolicy->granularity)
1483 break;
1484
1485 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1486 return __issue_discard_cmd_orderly(sbi, dpolicy);
1487
1488 pend_list = &dcc->pend_list[i];
1489
1490 mutex_lock(&dcc->cmd_lock);
1491 if (list_empty(pend_list))
1492 goto next;
1493 if (unlikely(dcc->rbtree_check))
1494 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1495 &dcc->root));
1496 blk_start_plug(&plug);
1497 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1498 f2fs_bug_on(sbi, dc->state != D_PREP);
1499
1500 if (dpolicy->timeout != 0 &&
1501 f2fs_time_over(sbi, dpolicy->timeout))
1502 break;
1503
1504 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1505 !is_idle(sbi, DISCARD_TIME)) {
1506 io_interrupted = true;
1507 break;
1508 }
1509
1510 __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1511
1512 if (issued >= dpolicy->max_requests)
1513 break;
1514 }
1515 blk_finish_plug(&plug);
1516 next:
1517 mutex_unlock(&dcc->cmd_lock);
1518
1519 if (issued >= dpolicy->max_requests || io_interrupted)
1520 break;
1521 }
1522
1523 if (!issued && io_interrupted)
1524 issued = -1;
1525
1526 return issued;
1527 }
1528
1529 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1530 {
1531 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1532 struct list_head *pend_list;
1533 struct discard_cmd *dc, *tmp;
1534 int i;
1535 bool dropped = false;
1536
1537 mutex_lock(&dcc->cmd_lock);
1538 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1539 pend_list = &dcc->pend_list[i];
1540 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1541 f2fs_bug_on(sbi, dc->state != D_PREP);
1542 __remove_discard_cmd(sbi, dc);
1543 dropped = true;
1544 }
1545 }
1546 mutex_unlock(&dcc->cmd_lock);
1547
1548 return dropped;
1549 }
1550
1551 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1552 {
1553 __drop_discard_cmd(sbi);
1554 }
1555
1556 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1557 struct discard_cmd *dc)
1558 {
1559 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1560 unsigned int len = 0;
1561
1562 wait_for_completion_io(&dc->wait);
1563 mutex_lock(&dcc->cmd_lock);
1564 f2fs_bug_on(sbi, dc->state != D_DONE);
1565 dc->ref--;
1566 if (!dc->ref) {
1567 if (!dc->error)
1568 len = dc->len;
1569 __remove_discard_cmd(sbi, dc);
1570 }
1571 mutex_unlock(&dcc->cmd_lock);
1572
1573 return len;
1574 }
1575
1576 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1577 struct discard_policy *dpolicy,
1578 block_t start, block_t end)
1579 {
1580 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1581 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1582 &(dcc->fstrim_list) : &(dcc->wait_list);
1583 struct discard_cmd *dc, *tmp;
1584 bool need_wait;
1585 unsigned int trimmed = 0;
1586
1587 next:
1588 need_wait = false;
1589
1590 mutex_lock(&dcc->cmd_lock);
1591 list_for_each_entry_safe(dc, tmp, wait_list, list) {
1592 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1593 continue;
1594 if (dc->len < dpolicy->granularity)
1595 continue;
1596 if (dc->state == D_DONE && !dc->ref) {
1597 wait_for_completion_io(&dc->wait);
1598 if (!dc->error)
1599 trimmed += dc->len;
1600 __remove_discard_cmd(sbi, dc);
1601 } else {
1602 dc->ref++;
1603 need_wait = true;
1604 break;
1605 }
1606 }
1607 mutex_unlock(&dcc->cmd_lock);
1608
1609 if (need_wait) {
1610 trimmed += __wait_one_discard_bio(sbi, dc);
1611 goto next;
1612 }
1613
1614 return trimmed;
1615 }
1616
1617 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1618 struct discard_policy *dpolicy)
1619 {
1620 struct discard_policy dp;
1621 unsigned int discard_blks;
1622
1623 if (dpolicy)
1624 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1625
1626 /* wait all */
1627 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1628 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1629 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1630 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1631
1632 return discard_blks;
1633 }
1634
1635 /* This should be covered by global mutex, &sit_i->sentry_lock */
1636 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1637 {
1638 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1639 struct discard_cmd *dc;
1640 bool need_wait = false;
1641
1642 mutex_lock(&dcc->cmd_lock);
1643 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1644 NULL, blkaddr);
1645 if (dc) {
1646 if (dc->state == D_PREP) {
1647 __punch_discard_cmd(sbi, dc, blkaddr);
1648 } else {
1649 dc->ref++;
1650 need_wait = true;
1651 }
1652 }
1653 mutex_unlock(&dcc->cmd_lock);
1654
1655 if (need_wait)
1656 __wait_one_discard_bio(sbi, dc);
1657 }
1658
1659 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1660 {
1661 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1662
1663 if (dcc && dcc->f2fs_issue_discard) {
1664 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1665
1666 dcc->f2fs_issue_discard = NULL;
1667 kthread_stop(discard_thread);
1668 }
1669 }
1670
1671 /* This comes from f2fs_put_super */
1672 bool f2fs_issue_discard_timeout(struct f2fs_sb_info *sbi)
1673 {
1674 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1675 struct discard_policy dpolicy;
1676 bool dropped;
1677
1678 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1679 dcc->discard_granularity);
1680 dpolicy.timeout = UMOUNT_DISCARD_TIMEOUT;
1681 __issue_discard_cmd(sbi, &dpolicy);
1682 dropped = __drop_discard_cmd(sbi);
1683
1684 /* just to make sure there is no pending discard commands */
1685 __wait_all_discard_cmd(sbi, NULL);
1686
1687 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1688 return dropped;
1689 }
1690
1691 static int issue_discard_thread(void *data)
1692 {
1693 struct f2fs_sb_info *sbi = data;
1694 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1695 wait_queue_head_t *q = &dcc->discard_wait_queue;
1696 struct discard_policy dpolicy;
1697 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1698 int issued;
1699
1700 set_freezable();
1701
1702 do {
1703 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1704 dcc->discard_granularity);
1705
1706 wait_event_interruptible_timeout(*q,
1707 kthread_should_stop() || freezing(current) ||
1708 dcc->discard_wake,
1709 msecs_to_jiffies(wait_ms));
1710
1711 if (dcc->discard_wake)
1712 dcc->discard_wake = 0;
1713
1714 /* clean up pending candidates before going to sleep */
1715 if (atomic_read(&dcc->queued_discard))
1716 __wait_all_discard_cmd(sbi, NULL);
1717
1718 if (try_to_freeze())
1719 continue;
1720 if (f2fs_readonly(sbi->sb))
1721 continue;
1722 if (kthread_should_stop())
1723 return 0;
1724 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1725 wait_ms = dpolicy.max_interval;
1726 continue;
1727 }
1728
1729 if (sbi->gc_mode == GC_URGENT)
1730 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1731
1732 sb_start_intwrite(sbi->sb);
1733
1734 issued = __issue_discard_cmd(sbi, &dpolicy);
1735 if (issued > 0) {
1736 __wait_all_discard_cmd(sbi, &dpolicy);
1737 wait_ms = dpolicy.min_interval;
1738 } else if (issued == -1){
1739 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1740 if (!wait_ms)
1741 wait_ms = dpolicy.mid_interval;
1742 } else {
1743 wait_ms = dpolicy.max_interval;
1744 }
1745
1746 sb_end_intwrite(sbi->sb);
1747
1748 } while (!kthread_should_stop());
1749 return 0;
1750 }
1751
1752 #ifdef CONFIG_BLK_DEV_ZONED
1753 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1754 struct block_device *bdev, block_t blkstart, block_t blklen)
1755 {
1756 sector_t sector, nr_sects;
1757 block_t lblkstart = blkstart;
1758 int devi = 0;
1759
1760 if (f2fs_is_multi_device(sbi)) {
1761 devi = f2fs_target_device_index(sbi, blkstart);
1762 if (blkstart < FDEV(devi).start_blk ||
1763 blkstart > FDEV(devi).end_blk) {
1764 f2fs_err(sbi, "Invalid block %x", blkstart);
1765 return -EIO;
1766 }
1767 blkstart -= FDEV(devi).start_blk;
1768 }
1769
1770 /* For sequential zones, reset the zone write pointer */
1771 if (f2fs_blkz_is_seq(sbi, devi, blkstart)) {
1772 sector = SECTOR_FROM_BLOCK(blkstart);
1773 nr_sects = SECTOR_FROM_BLOCK(blklen);
1774
1775 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1776 nr_sects != bdev_zone_sectors(bdev)) {
1777 f2fs_err(sbi, "(%d) %s: Unaligned zone reset attempted (block %x + %x)",
1778 devi, sbi->s_ndevs ? FDEV(devi).path : "",
1779 blkstart, blklen);
1780 return -EIO;
1781 }
1782 trace_f2fs_issue_reset_zone(bdev, blkstart);
1783 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1784 sector, nr_sects, GFP_NOFS);
1785 }
1786
1787 /* For conventional zones, use regular discard if supported */
1788 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1789 }
1790 #endif
1791
1792 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1793 struct block_device *bdev, block_t blkstart, block_t blklen)
1794 {
1795 #ifdef CONFIG_BLK_DEV_ZONED
1796 if (f2fs_sb_has_blkzoned(sbi) && bdev_is_zoned(bdev))
1797 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1798 #endif
1799 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1800 }
1801
1802 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1803 block_t blkstart, block_t blklen)
1804 {
1805 sector_t start = blkstart, len = 0;
1806 struct block_device *bdev;
1807 struct seg_entry *se;
1808 unsigned int offset;
1809 block_t i;
1810 int err = 0;
1811
1812 bdev = f2fs_target_device(sbi, blkstart, NULL);
1813
1814 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1815 if (i != start) {
1816 struct block_device *bdev2 =
1817 f2fs_target_device(sbi, i, NULL);
1818
1819 if (bdev2 != bdev) {
1820 err = __issue_discard_async(sbi, bdev,
1821 start, len);
1822 if (err)
1823 return err;
1824 bdev = bdev2;
1825 start = i;
1826 len = 0;
1827 }
1828 }
1829
1830 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1831 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1832
1833 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1834 sbi->discard_blks--;
1835 }
1836
1837 if (len)
1838 err = __issue_discard_async(sbi, bdev, start, len);
1839 return err;
1840 }
1841
1842 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1843 bool check_only)
1844 {
1845 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1846 int max_blocks = sbi->blocks_per_seg;
1847 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1848 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1849 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1850 unsigned long *discard_map = (unsigned long *)se->discard_map;
1851 unsigned long *dmap = SIT_I(sbi)->tmp_map;
1852 unsigned int start = 0, end = -1;
1853 bool force = (cpc->reason & CP_DISCARD);
1854 struct discard_entry *de = NULL;
1855 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1856 int i;
1857
1858 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1859 return false;
1860
1861 if (!force) {
1862 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1863 SM_I(sbi)->dcc_info->nr_discards >=
1864 SM_I(sbi)->dcc_info->max_discards)
1865 return false;
1866 }
1867
1868 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1869 for (i = 0; i < entries; i++)
1870 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1871 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1872
1873 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1874 SM_I(sbi)->dcc_info->max_discards) {
1875 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1876 if (start >= max_blocks)
1877 break;
1878
1879 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1880 if (force && start && end != max_blocks
1881 && (end - start) < cpc->trim_minlen)
1882 continue;
1883
1884 if (check_only)
1885 return true;
1886
1887 if (!de) {
1888 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1889 GFP_F2FS_ZERO);
1890 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1891 list_add_tail(&de->list, head);
1892 }
1893
1894 for (i = start; i < end; i++)
1895 __set_bit_le(i, (void *)de->discard_map);
1896
1897 SM_I(sbi)->dcc_info->nr_discards += end - start;
1898 }
1899 return false;
1900 }
1901
1902 static void release_discard_addr(struct discard_entry *entry)
1903 {
1904 list_del(&entry->list);
1905 kmem_cache_free(discard_entry_slab, entry);
1906 }
1907
1908 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1909 {
1910 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1911 struct discard_entry *entry, *this;
1912
1913 /* drop caches */
1914 list_for_each_entry_safe(entry, this, head, list)
1915 release_discard_addr(entry);
1916 }
1917
1918 /*
1919 * Should call f2fs_clear_prefree_segments after checkpoint is done.
1920 */
1921 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1922 {
1923 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1924 unsigned int segno;
1925
1926 mutex_lock(&dirty_i->seglist_lock);
1927 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1928 __set_test_and_free(sbi, segno);
1929 mutex_unlock(&dirty_i->seglist_lock);
1930 }
1931
1932 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1933 struct cp_control *cpc)
1934 {
1935 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1936 struct list_head *head = &dcc->entry_list;
1937 struct discard_entry *entry, *this;
1938 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1939 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1940 unsigned int start = 0, end = -1;
1941 unsigned int secno, start_segno;
1942 bool force = (cpc->reason & CP_DISCARD);
1943 bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
1944
1945 mutex_lock(&dirty_i->seglist_lock);
1946
1947 while (1) {
1948 int i;
1949
1950 if (need_align && end != -1)
1951 end--;
1952 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1953 if (start >= MAIN_SEGS(sbi))
1954 break;
1955 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1956 start + 1);
1957
1958 if (need_align) {
1959 start = rounddown(start, sbi->segs_per_sec);
1960 end = roundup(end, sbi->segs_per_sec);
1961 }
1962
1963 for (i = start; i < end; i++) {
1964 if (test_and_clear_bit(i, prefree_map))
1965 dirty_i->nr_dirty[PRE]--;
1966 }
1967
1968 if (!f2fs_realtime_discard_enable(sbi))
1969 continue;
1970
1971 if (force && start >= cpc->trim_start &&
1972 (end - 1) <= cpc->trim_end)
1973 continue;
1974
1975 if (!test_opt(sbi, LFS) || !__is_large_section(sbi)) {
1976 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1977 (end - start) << sbi->log_blocks_per_seg);
1978 continue;
1979 }
1980 next:
1981 secno = GET_SEC_FROM_SEG(sbi, start);
1982 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1983 if (!IS_CURSEC(sbi, secno) &&
1984 !get_valid_blocks(sbi, start, true))
1985 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1986 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1987
1988 start = start_segno + sbi->segs_per_sec;
1989 if (start < end)
1990 goto next;
1991 else
1992 end = start - 1;
1993 }
1994 mutex_unlock(&dirty_i->seglist_lock);
1995
1996 /* send small discards */
1997 list_for_each_entry_safe(entry, this, head, list) {
1998 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1999 bool is_valid = test_bit_le(0, entry->discard_map);
2000
2001 find_next:
2002 if (is_valid) {
2003 next_pos = find_next_zero_bit_le(entry->discard_map,
2004 sbi->blocks_per_seg, cur_pos);
2005 len = next_pos - cur_pos;
2006
2007 if (f2fs_sb_has_blkzoned(sbi) ||
2008 (force && len < cpc->trim_minlen))
2009 goto skip;
2010
2011 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
2012 len);
2013 total_len += len;
2014 } else {
2015 next_pos = find_next_bit_le(entry->discard_map,
2016 sbi->blocks_per_seg, cur_pos);
2017 }
2018 skip:
2019 cur_pos = next_pos;
2020 is_valid = !is_valid;
2021
2022 if (cur_pos < sbi->blocks_per_seg)
2023 goto find_next;
2024
2025 release_discard_addr(entry);
2026 dcc->nr_discards -= total_len;
2027 }
2028
2029 wake_up_discard_thread(sbi, false);
2030 }
2031
2032 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
2033 {
2034 dev_t dev = sbi->sb->s_bdev->bd_dev;
2035 struct discard_cmd_control *dcc;
2036 int err = 0, i;
2037
2038 if (SM_I(sbi)->dcc_info) {
2039 dcc = SM_I(sbi)->dcc_info;
2040 goto init_thread;
2041 }
2042
2043 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
2044 if (!dcc)
2045 return -ENOMEM;
2046
2047 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
2048 INIT_LIST_HEAD(&dcc->entry_list);
2049 for (i = 0; i < MAX_PLIST_NUM; i++)
2050 INIT_LIST_HEAD(&dcc->pend_list[i]);
2051 INIT_LIST_HEAD(&dcc->wait_list);
2052 INIT_LIST_HEAD(&dcc->fstrim_list);
2053 mutex_init(&dcc->cmd_lock);
2054 atomic_set(&dcc->issued_discard, 0);
2055 atomic_set(&dcc->queued_discard, 0);
2056 atomic_set(&dcc->discard_cmd_cnt, 0);
2057 dcc->nr_discards = 0;
2058 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
2059 dcc->undiscard_blks = 0;
2060 dcc->next_pos = 0;
2061 dcc->root = RB_ROOT_CACHED;
2062 dcc->rbtree_check = false;
2063
2064 init_waitqueue_head(&dcc->discard_wait_queue);
2065 SM_I(sbi)->dcc_info = dcc;
2066 init_thread:
2067 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
2068 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
2069 if (IS_ERR(dcc->f2fs_issue_discard)) {
2070 err = PTR_ERR(dcc->f2fs_issue_discard);
2071 kvfree(dcc);
2072 SM_I(sbi)->dcc_info = NULL;
2073 return err;
2074 }
2075
2076 return err;
2077 }
2078
2079 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
2080 {
2081 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2082
2083 if (!dcc)
2084 return;
2085
2086 f2fs_stop_discard_thread(sbi);
2087
2088 /*
2089 * Recovery can cache discard commands, so in error path of
2090 * fill_super(), it needs to give a chance to handle them.
2091 */
2092 if (unlikely(atomic_read(&dcc->discard_cmd_cnt)))
2093 f2fs_issue_discard_timeout(sbi);
2094
2095 kvfree(dcc);
2096 SM_I(sbi)->dcc_info = NULL;
2097 }
2098
2099 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
2100 {
2101 struct sit_info *sit_i = SIT_I(sbi);
2102
2103 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
2104 sit_i->dirty_sentries++;
2105 return false;
2106 }
2107
2108 return true;
2109 }
2110
2111 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
2112 unsigned int segno, int modified)
2113 {
2114 struct seg_entry *se = get_seg_entry(sbi, segno);
2115 se->type = type;
2116 if (modified)
2117 __mark_sit_entry_dirty(sbi, segno);
2118 }
2119
2120 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
2121 {
2122 struct seg_entry *se;
2123 unsigned int segno, offset;
2124 long int new_vblocks;
2125 bool exist;
2126 #ifdef CONFIG_F2FS_CHECK_FS
2127 bool mir_exist;
2128 #endif
2129
2130 segno = GET_SEGNO(sbi, blkaddr);
2131
2132 se = get_seg_entry(sbi, segno);
2133 new_vblocks = se->valid_blocks + del;
2134 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2135
2136 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
2137 (new_vblocks > sbi->blocks_per_seg)));
2138
2139 se->valid_blocks = new_vblocks;
2140 se->mtime = get_mtime(sbi, false);
2141 if (se->mtime > SIT_I(sbi)->max_mtime)
2142 SIT_I(sbi)->max_mtime = se->mtime;
2143
2144 /* Update valid block bitmap */
2145 if (del > 0) {
2146 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2147 #ifdef CONFIG_F2FS_CHECK_FS
2148 mir_exist = f2fs_test_and_set_bit(offset,
2149 se->cur_valid_map_mir);
2150 if (unlikely(exist != mir_exist)) {
2151 f2fs_err(sbi, "Inconsistent error when setting bitmap, blk:%u, old bit:%d",
2152 blkaddr, exist);
2153 f2fs_bug_on(sbi, 1);
2154 }
2155 #endif
2156 if (unlikely(exist)) {
2157 f2fs_err(sbi, "Bitmap was wrongly set, blk:%u",
2158 blkaddr);
2159 f2fs_bug_on(sbi, 1);
2160 se->valid_blocks--;
2161 del = 0;
2162 }
2163
2164 if (!f2fs_test_and_set_bit(offset, se->discard_map))
2165 sbi->discard_blks--;
2166
2167 /*
2168 * SSR should never reuse block which is checkpointed
2169 * or newly invalidated.
2170 */
2171 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED)) {
2172 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2173 se->ckpt_valid_blocks++;
2174 }
2175 } else {
2176 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2177 #ifdef CONFIG_F2FS_CHECK_FS
2178 mir_exist = f2fs_test_and_clear_bit(offset,
2179 se->cur_valid_map_mir);
2180 if (unlikely(exist != mir_exist)) {
2181 f2fs_err(sbi, "Inconsistent error when clearing bitmap, blk:%u, old bit:%d",
2182 blkaddr, exist);
2183 f2fs_bug_on(sbi, 1);
2184 }
2185 #endif
2186 if (unlikely(!exist)) {
2187 f2fs_err(sbi, "Bitmap was wrongly cleared, blk:%u",
2188 blkaddr);
2189 f2fs_bug_on(sbi, 1);
2190 se->valid_blocks++;
2191 del = 0;
2192 } else if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2193 /*
2194 * If checkpoints are off, we must not reuse data that
2195 * was used in the previous checkpoint. If it was used
2196 * before, we must track that to know how much space we
2197 * really have.
2198 */
2199 if (f2fs_test_bit(offset, se->ckpt_valid_map)) {
2200 spin_lock(&sbi->stat_lock);
2201 sbi->unusable_block_count++;
2202 spin_unlock(&sbi->stat_lock);
2203 }
2204 }
2205
2206 if (f2fs_test_and_clear_bit(offset, se->discard_map))
2207 sbi->discard_blks++;
2208 }
2209 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2210 se->ckpt_valid_blocks += del;
2211
2212 __mark_sit_entry_dirty(sbi, segno);
2213
2214 /* update total number of valid blocks to be written in ckpt area */
2215 SIT_I(sbi)->written_valid_blocks += del;
2216
2217 if (__is_large_section(sbi))
2218 get_sec_entry(sbi, segno)->valid_blocks += del;
2219 }
2220
2221 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2222 {
2223 unsigned int segno = GET_SEGNO(sbi, addr);
2224 struct sit_info *sit_i = SIT_I(sbi);
2225
2226 f2fs_bug_on(sbi, addr == NULL_ADDR);
2227 if (addr == NEW_ADDR || addr == COMPRESS_ADDR)
2228 return;
2229
2230 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2231
2232 /* add it into sit main buffer */
2233 down_write(&sit_i->sentry_lock);
2234
2235 update_sit_entry(sbi, addr, -1);
2236
2237 /* add it into dirty seglist */
2238 locate_dirty_segment(sbi, segno);
2239
2240 up_write(&sit_i->sentry_lock);
2241 }
2242
2243 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2244 {
2245 struct sit_info *sit_i = SIT_I(sbi);
2246 unsigned int segno, offset;
2247 struct seg_entry *se;
2248 bool is_cp = false;
2249
2250 if (!__is_valid_data_blkaddr(blkaddr))
2251 return true;
2252
2253 down_read(&sit_i->sentry_lock);
2254
2255 segno = GET_SEGNO(sbi, blkaddr);
2256 se = get_seg_entry(sbi, segno);
2257 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2258
2259 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2260 is_cp = true;
2261
2262 up_read(&sit_i->sentry_lock);
2263
2264 return is_cp;
2265 }
2266
2267 /*
2268 * This function should be resided under the curseg_mutex lock
2269 */
2270 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2271 struct f2fs_summary *sum)
2272 {
2273 struct curseg_info *curseg = CURSEG_I(sbi, type);
2274 void *addr = curseg->sum_blk;
2275 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2276 memcpy(addr, sum, sizeof(struct f2fs_summary));
2277 }
2278
2279 /*
2280 * Calculate the number of current summary pages for writing
2281 */
2282 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2283 {
2284 int valid_sum_count = 0;
2285 int i, sum_in_page;
2286
2287 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2288 if (sbi->ckpt->alloc_type[i] == SSR)
2289 valid_sum_count += sbi->blocks_per_seg;
2290 else {
2291 if (for_ra)
2292 valid_sum_count += le16_to_cpu(
2293 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2294 else
2295 valid_sum_count += curseg_blkoff(sbi, i);
2296 }
2297 }
2298
2299 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2300 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2301 if (valid_sum_count <= sum_in_page)
2302 return 1;
2303 else if ((valid_sum_count - sum_in_page) <=
2304 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2305 return 2;
2306 return 3;
2307 }
2308
2309 /*
2310 * Caller should put this summary page
2311 */
2312 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2313 {
2314 return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2315 }
2316
2317 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2318 void *src, block_t blk_addr)
2319 {
2320 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2321
2322 memcpy(page_address(page), src, PAGE_SIZE);
2323 set_page_dirty(page);
2324 f2fs_put_page(page, 1);
2325 }
2326
2327 static void write_sum_page(struct f2fs_sb_info *sbi,
2328 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2329 {
2330 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2331 }
2332
2333 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2334 int type, block_t blk_addr)
2335 {
2336 struct curseg_info *curseg = CURSEG_I(sbi, type);
2337 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2338 struct f2fs_summary_block *src = curseg->sum_blk;
2339 struct f2fs_summary_block *dst;
2340
2341 dst = (struct f2fs_summary_block *)page_address(page);
2342 memset(dst, 0, PAGE_SIZE);
2343
2344 mutex_lock(&curseg->curseg_mutex);
2345
2346 down_read(&curseg->journal_rwsem);
2347 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2348 up_read(&curseg->journal_rwsem);
2349
2350 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2351 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2352
2353 mutex_unlock(&curseg->curseg_mutex);
2354
2355 set_page_dirty(page);
2356 f2fs_put_page(page, 1);
2357 }
2358
2359 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2360 {
2361 struct curseg_info *curseg = CURSEG_I(sbi, type);
2362 unsigned int segno = curseg->segno + 1;
2363 struct free_segmap_info *free_i = FREE_I(sbi);
2364
2365 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2366 return !test_bit(segno, free_i->free_segmap);
2367 return 0;
2368 }
2369
2370 /*
2371 * Find a new segment from the free segments bitmap to right order
2372 * This function should be returned with success, otherwise BUG
2373 */
2374 static void get_new_segment(struct f2fs_sb_info *sbi,
2375 unsigned int *newseg, bool new_sec, int dir)
2376 {
2377 struct free_segmap_info *free_i = FREE_I(sbi);
2378 unsigned int segno, secno, zoneno;
2379 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2380 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2381 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2382 unsigned int left_start = hint;
2383 bool init = true;
2384 int go_left = 0;
2385 int i;
2386
2387 spin_lock(&free_i->segmap_lock);
2388
2389 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2390 segno = find_next_zero_bit(free_i->free_segmap,
2391 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2392 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2393 goto got_it;
2394 }
2395 find_other_zone:
2396 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2397 if (secno >= MAIN_SECS(sbi)) {
2398 if (dir == ALLOC_RIGHT) {
2399 secno = find_next_zero_bit(free_i->free_secmap,
2400 MAIN_SECS(sbi), 0);
2401 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2402 } else {
2403 go_left = 1;
2404 left_start = hint - 1;
2405 }
2406 }
2407 if (go_left == 0)
2408 goto skip_left;
2409
2410 while (test_bit(left_start, free_i->free_secmap)) {
2411 if (left_start > 0) {
2412 left_start--;
2413 continue;
2414 }
2415 left_start = find_next_zero_bit(free_i->free_secmap,
2416 MAIN_SECS(sbi), 0);
2417 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2418 break;
2419 }
2420 secno = left_start;
2421 skip_left:
2422 segno = GET_SEG_FROM_SEC(sbi, secno);
2423 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2424
2425 /* give up on finding another zone */
2426 if (!init)
2427 goto got_it;
2428 if (sbi->secs_per_zone == 1)
2429 goto got_it;
2430 if (zoneno == old_zoneno)
2431 goto got_it;
2432 if (dir == ALLOC_LEFT) {
2433 if (!go_left && zoneno + 1 >= total_zones)
2434 goto got_it;
2435 if (go_left && zoneno == 0)
2436 goto got_it;
2437 }
2438 for (i = 0; i < NR_CURSEG_TYPE; i++)
2439 if (CURSEG_I(sbi, i)->zone == zoneno)
2440 break;
2441
2442 if (i < NR_CURSEG_TYPE) {
2443 /* zone is in user, try another */
2444 if (go_left)
2445 hint = zoneno * sbi->secs_per_zone - 1;
2446 else if (zoneno + 1 >= total_zones)
2447 hint = 0;
2448 else
2449 hint = (zoneno + 1) * sbi->secs_per_zone;
2450 init = false;
2451 goto find_other_zone;
2452 }
2453 got_it:
2454 /* set it as dirty segment in free segmap */
2455 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2456 __set_inuse(sbi, segno);
2457 *newseg = segno;
2458 spin_unlock(&free_i->segmap_lock);
2459 }
2460
2461 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2462 {
2463 struct curseg_info *curseg = CURSEG_I(sbi, type);
2464 struct summary_footer *sum_footer;
2465
2466 curseg->segno = curseg->next_segno;
2467 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2468 curseg->next_blkoff = 0;
2469 curseg->next_segno = NULL_SEGNO;
2470
2471 sum_footer = &(curseg->sum_blk->footer);
2472 memset(sum_footer, 0, sizeof(struct summary_footer));
2473 if (IS_DATASEG(type))
2474 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2475 if (IS_NODESEG(type))
2476 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2477 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2478 }
2479
2480 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2481 {
2482 /* if segs_per_sec is large than 1, we need to keep original policy. */
2483 if (__is_large_section(sbi))
2484 return CURSEG_I(sbi, type)->segno;
2485
2486 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2487 return 0;
2488
2489 if (test_opt(sbi, NOHEAP) &&
2490 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2491 return 0;
2492
2493 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2494 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2495
2496 /* find segments from 0 to reuse freed segments */
2497 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2498 return 0;
2499
2500 return CURSEG_I(sbi, type)->segno;
2501 }
2502
2503 /*
2504 * Allocate a current working segment.
2505 * This function always allocates a free segment in LFS manner.
2506 */
2507 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2508 {
2509 struct curseg_info *curseg = CURSEG_I(sbi, type);
2510 unsigned int segno = curseg->segno;
2511 int dir = ALLOC_LEFT;
2512
2513 write_sum_page(sbi, curseg->sum_blk,
2514 GET_SUM_BLOCK(sbi, segno));
2515 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2516 dir = ALLOC_RIGHT;
2517
2518 if (test_opt(sbi, NOHEAP))
2519 dir = ALLOC_RIGHT;
2520
2521 segno = __get_next_segno(sbi, type);
2522 get_new_segment(sbi, &segno, new_sec, dir);
2523 curseg->next_segno = segno;
2524 reset_curseg(sbi, type, 1);
2525 curseg->alloc_type = LFS;
2526 }
2527
2528 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2529 struct curseg_info *seg, block_t start)
2530 {
2531 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2532 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2533 unsigned long *target_map = SIT_I(sbi)->tmp_map;
2534 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2535 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2536 int i, pos;
2537
2538 for (i = 0; i < entries; i++)
2539 target_map[i] = ckpt_map[i] | cur_map[i];
2540
2541 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2542
2543 seg->next_blkoff = pos;
2544 }
2545
2546 /*
2547 * If a segment is written by LFS manner, next block offset is just obtained
2548 * by increasing the current block offset. However, if a segment is written by
2549 * SSR manner, next block offset obtained by calling __next_free_blkoff
2550 */
2551 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2552 struct curseg_info *seg)
2553 {
2554 if (seg->alloc_type == SSR)
2555 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2556 else
2557 seg->next_blkoff++;
2558 }
2559
2560 /*
2561 * This function always allocates a used segment(from dirty seglist) by SSR
2562 * manner, so it should recover the existing segment information of valid blocks
2563 */
2564 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2565 {
2566 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2567 struct curseg_info *curseg = CURSEG_I(sbi, type);
2568 unsigned int new_segno = curseg->next_segno;
2569 struct f2fs_summary_block *sum_node;
2570 struct page *sum_page;
2571
2572 write_sum_page(sbi, curseg->sum_blk,
2573 GET_SUM_BLOCK(sbi, curseg->segno));
2574 __set_test_and_inuse(sbi, new_segno);
2575
2576 mutex_lock(&dirty_i->seglist_lock);
2577 __remove_dirty_segment(sbi, new_segno, PRE);
2578 __remove_dirty_segment(sbi, new_segno, DIRTY);
2579 mutex_unlock(&dirty_i->seglist_lock);
2580
2581 reset_curseg(sbi, type, 1);
2582 curseg->alloc_type = SSR;
2583 __next_free_blkoff(sbi, curseg, 0);
2584
2585 sum_page = f2fs_get_sum_page(sbi, new_segno);
2586 f2fs_bug_on(sbi, IS_ERR(sum_page));
2587 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2588 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2589 f2fs_put_page(sum_page, 1);
2590 }
2591
2592 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2593 {
2594 struct curseg_info *curseg = CURSEG_I(sbi, type);
2595 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2596 unsigned segno = NULL_SEGNO;
2597 int i, cnt;
2598 bool reversed = false;
2599
2600 /* f2fs_need_SSR() already forces to do this */
2601 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2602 curseg->next_segno = segno;
2603 return 1;
2604 }
2605
2606 /* For node segments, let's do SSR more intensively */
2607 if (IS_NODESEG(type)) {
2608 if (type >= CURSEG_WARM_NODE) {
2609 reversed = true;
2610 i = CURSEG_COLD_NODE;
2611 } else {
2612 i = CURSEG_HOT_NODE;
2613 }
2614 cnt = NR_CURSEG_NODE_TYPE;
2615 } else {
2616 if (type >= CURSEG_WARM_DATA) {
2617 reversed = true;
2618 i = CURSEG_COLD_DATA;
2619 } else {
2620 i = CURSEG_HOT_DATA;
2621 }
2622 cnt = NR_CURSEG_DATA_TYPE;
2623 }
2624
2625 for (; cnt-- > 0; reversed ? i-- : i++) {
2626 if (i == type)
2627 continue;
2628 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2629 curseg->next_segno = segno;
2630 return 1;
2631 }
2632 }
2633
2634 /* find valid_blocks=0 in dirty list */
2635 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
2636 segno = get_free_segment(sbi);
2637 if (segno != NULL_SEGNO) {
2638 curseg->next_segno = segno;
2639 return 1;
2640 }
2641 }
2642 return 0;
2643 }
2644
2645 /*
2646 * flush out current segment and replace it with new segment
2647 * This function should be returned with success, otherwise BUG
2648 */
2649 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2650 int type, bool force)
2651 {
2652 struct curseg_info *curseg = CURSEG_I(sbi, type);
2653
2654 if (force)
2655 new_curseg(sbi, type, true);
2656 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2657 type == CURSEG_WARM_NODE)
2658 new_curseg(sbi, type, false);
2659 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type) &&
2660 likely(!is_sbi_flag_set(sbi, SBI_CP_DISABLED)))
2661 new_curseg(sbi, type, false);
2662 else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2663 change_curseg(sbi, type);
2664 else
2665 new_curseg(sbi, type, false);
2666
2667 stat_inc_seg_type(sbi, curseg);
2668 }
2669
2670 void allocate_segment_for_resize(struct f2fs_sb_info *sbi, int type,
2671 unsigned int start, unsigned int end)
2672 {
2673 struct curseg_info *curseg = CURSEG_I(sbi, type);
2674 unsigned int segno;
2675
2676 down_read(&SM_I(sbi)->curseg_lock);
2677 mutex_lock(&curseg->curseg_mutex);
2678 down_write(&SIT_I(sbi)->sentry_lock);
2679
2680 segno = CURSEG_I(sbi, type)->segno;
2681 if (segno < start || segno > end)
2682 goto unlock;
2683
2684 if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2685 change_curseg(sbi, type);
2686 else
2687 new_curseg(sbi, type, true);
2688
2689 stat_inc_seg_type(sbi, curseg);
2690
2691 locate_dirty_segment(sbi, segno);
2692 unlock:
2693 up_write(&SIT_I(sbi)->sentry_lock);
2694
2695 if (segno != curseg->segno)
2696 f2fs_notice(sbi, "For resize: curseg of type %d: %u ==> %u",
2697 type, segno, curseg->segno);
2698
2699 mutex_unlock(&curseg->curseg_mutex);
2700 up_read(&SM_I(sbi)->curseg_lock);
2701 }
2702
2703 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi, int type)
2704 {
2705 struct curseg_info *curseg;
2706 unsigned int old_segno;
2707 int i;
2708
2709 down_write(&SIT_I(sbi)->sentry_lock);
2710
2711 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2712 if (type != NO_CHECK_TYPE && i != type)
2713 continue;
2714
2715 curseg = CURSEG_I(sbi, i);
2716 if (type == NO_CHECK_TYPE || curseg->next_blkoff ||
2717 get_valid_blocks(sbi, curseg->segno, false) ||
2718 get_ckpt_valid_blocks(sbi, curseg->segno)) {
2719 old_segno = curseg->segno;
2720 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2721 locate_dirty_segment(sbi, old_segno);
2722 }
2723 }
2724
2725 up_write(&SIT_I(sbi)->sentry_lock);
2726 }
2727
2728 static const struct segment_allocation default_salloc_ops = {
2729 .allocate_segment = allocate_segment_by_default,
2730 };
2731
2732 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2733 struct cp_control *cpc)
2734 {
2735 __u64 trim_start = cpc->trim_start;
2736 bool has_candidate = false;
2737
2738 down_write(&SIT_I(sbi)->sentry_lock);
2739 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2740 if (add_discard_addrs(sbi, cpc, true)) {
2741 has_candidate = true;
2742 break;
2743 }
2744 }
2745 up_write(&SIT_I(sbi)->sentry_lock);
2746
2747 cpc->trim_start = trim_start;
2748 return has_candidate;
2749 }
2750
2751 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2752 struct discard_policy *dpolicy,
2753 unsigned int start, unsigned int end)
2754 {
2755 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2756 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2757 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2758 struct discard_cmd *dc;
2759 struct blk_plug plug;
2760 int issued;
2761 unsigned int trimmed = 0;
2762
2763 next:
2764 issued = 0;
2765
2766 mutex_lock(&dcc->cmd_lock);
2767 if (unlikely(dcc->rbtree_check))
2768 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2769 &dcc->root));
2770
2771 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2772 NULL, start,
2773 (struct rb_entry **)&prev_dc,
2774 (struct rb_entry **)&next_dc,
2775 &insert_p, &insert_parent, true, NULL);
2776 if (!dc)
2777 dc = next_dc;
2778
2779 blk_start_plug(&plug);
2780
2781 while (dc && dc->lstart <= end) {
2782 struct rb_node *node;
2783 int err = 0;
2784
2785 if (dc->len < dpolicy->granularity)
2786 goto skip;
2787
2788 if (dc->state != D_PREP) {
2789 list_move_tail(&dc->list, &dcc->fstrim_list);
2790 goto skip;
2791 }
2792
2793 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2794
2795 if (issued >= dpolicy->max_requests) {
2796 start = dc->lstart + dc->len;
2797
2798 if (err)
2799 __remove_discard_cmd(sbi, dc);
2800
2801 blk_finish_plug(&plug);
2802 mutex_unlock(&dcc->cmd_lock);
2803 trimmed += __wait_all_discard_cmd(sbi, NULL);
2804 congestion_wait(BLK_RW_ASYNC, HZ/50);
2805 goto next;
2806 }
2807 skip:
2808 node = rb_next(&dc->rb_node);
2809 if (err)
2810 __remove_discard_cmd(sbi, dc);
2811 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2812
2813 if (fatal_signal_pending(current))
2814 break;
2815 }
2816
2817 blk_finish_plug(&plug);
2818 mutex_unlock(&dcc->cmd_lock);
2819
2820 return trimmed;
2821 }
2822
2823 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2824 {
2825 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2826 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2827 unsigned int start_segno, end_segno;
2828 block_t start_block, end_block;
2829 struct cp_control cpc;
2830 struct discard_policy dpolicy;
2831 unsigned long long trimmed = 0;
2832 int err = 0;
2833 bool need_align = test_opt(sbi, LFS) && __is_large_section(sbi);
2834
2835 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2836 return -EINVAL;
2837
2838 if (end < MAIN_BLKADDR(sbi))
2839 goto out;
2840
2841 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2842 f2fs_warn(sbi, "Found FS corruption, run fsck to fix.");
2843 return -EFSCORRUPTED;
2844 }
2845
2846 /* start/end segment number in main_area */
2847 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2848 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2849 GET_SEGNO(sbi, end);
2850 if (need_align) {
2851 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2852 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2853 }
2854
2855 cpc.reason = CP_DISCARD;
2856 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2857 cpc.trim_start = start_segno;
2858 cpc.trim_end = end_segno;
2859
2860 if (sbi->discard_blks == 0)
2861 goto out;
2862
2863 down_write(&sbi->gc_lock);
2864 err = f2fs_write_checkpoint(sbi, &cpc);
2865 up_write(&sbi->gc_lock);
2866 if (err)
2867 goto out;
2868
2869 /*
2870 * We filed discard candidates, but actually we don't need to wait for
2871 * all of them, since they'll be issued in idle time along with runtime
2872 * discard option. User configuration looks like using runtime discard
2873 * or periodic fstrim instead of it.
2874 */
2875 if (f2fs_realtime_discard_enable(sbi))
2876 goto out;
2877
2878 start_block = START_BLOCK(sbi, start_segno);
2879 end_block = START_BLOCK(sbi, end_segno + 1);
2880
2881 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2882 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2883 start_block, end_block);
2884
2885 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2886 start_block, end_block);
2887 out:
2888 if (!err)
2889 range->len = F2FS_BLK_TO_BYTES(trimmed);
2890 return err;
2891 }
2892
2893 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2894 {
2895 struct curseg_info *curseg = CURSEG_I(sbi, type);
2896 if (curseg->next_blkoff < sbi->blocks_per_seg)
2897 return true;
2898 return false;
2899 }
2900
2901 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2902 {
2903 switch (hint) {
2904 case WRITE_LIFE_SHORT:
2905 return CURSEG_HOT_DATA;
2906 case WRITE_LIFE_EXTREME:
2907 return CURSEG_COLD_DATA;
2908 default:
2909 return CURSEG_WARM_DATA;
2910 }
2911 }
2912
2913 /* This returns write hints for each segment type. This hints will be
2914 * passed down to block layer. There are mapping tables which depend on
2915 * the mount option 'whint_mode'.
2916 *
2917 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2918 *
2919 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2920 *
2921 * User F2FS Block
2922 * ---- ---- -----
2923 * META WRITE_LIFE_NOT_SET
2924 * HOT_NODE "
2925 * WARM_NODE "
2926 * COLD_NODE "
2927 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2928 * extension list " "
2929 *
2930 * -- buffered io
2931 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2932 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2933 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2934 * WRITE_LIFE_NONE " "
2935 * WRITE_LIFE_MEDIUM " "
2936 * WRITE_LIFE_LONG " "
2937 *
2938 * -- direct io
2939 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2940 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2941 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2942 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2943 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2944 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2945 *
2946 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2947 *
2948 * User F2FS Block
2949 * ---- ---- -----
2950 * META WRITE_LIFE_MEDIUM;
2951 * HOT_NODE WRITE_LIFE_NOT_SET
2952 * WARM_NODE "
2953 * COLD_NODE WRITE_LIFE_NONE
2954 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2955 * extension list " "
2956 *
2957 * -- buffered io
2958 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2959 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2960 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
2961 * WRITE_LIFE_NONE " "
2962 * WRITE_LIFE_MEDIUM " "
2963 * WRITE_LIFE_LONG " "
2964 *
2965 * -- direct io
2966 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2967 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2968 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2969 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2970 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2971 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2972 */
2973
2974 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2975 enum page_type type, enum temp_type temp)
2976 {
2977 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2978 if (type == DATA) {
2979 if (temp == WARM)
2980 return WRITE_LIFE_NOT_SET;
2981 else if (temp == HOT)
2982 return WRITE_LIFE_SHORT;
2983 else if (temp == COLD)
2984 return WRITE_LIFE_EXTREME;
2985 } else {
2986 return WRITE_LIFE_NOT_SET;
2987 }
2988 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2989 if (type == DATA) {
2990 if (temp == WARM)
2991 return WRITE_LIFE_LONG;
2992 else if (temp == HOT)
2993 return WRITE_LIFE_SHORT;
2994 else if (temp == COLD)
2995 return WRITE_LIFE_EXTREME;
2996 } else if (type == NODE) {
2997 if (temp == WARM || temp == HOT)
2998 return WRITE_LIFE_NOT_SET;
2999 else if (temp == COLD)
3000 return WRITE_LIFE_NONE;
3001 } else if (type == META) {
3002 return WRITE_LIFE_MEDIUM;
3003 }
3004 }
3005 return WRITE_LIFE_NOT_SET;
3006 }
3007
3008 static int __get_segment_type_2(struct f2fs_io_info *fio)
3009 {
3010 if (fio->type == DATA)
3011 return CURSEG_HOT_DATA;
3012 else
3013 return CURSEG_HOT_NODE;
3014 }
3015
3016 static int __get_segment_type_4(struct f2fs_io_info *fio)
3017 {
3018 if (fio->type == DATA) {
3019 struct inode *inode = fio->page->mapping->host;
3020
3021 if (S_ISDIR(inode->i_mode))
3022 return CURSEG_HOT_DATA;
3023 else
3024 return CURSEG_COLD_DATA;
3025 } else {
3026 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
3027 return CURSEG_WARM_NODE;
3028 else
3029 return CURSEG_COLD_NODE;
3030 }
3031 }
3032
3033 static int __get_segment_type_6(struct f2fs_io_info *fio)
3034 {
3035 if (fio->type == DATA) {
3036 struct inode *inode = fio->page->mapping->host;
3037
3038 if (is_cold_data(fio->page) || file_is_cold(inode) ||
3039 f2fs_compressed_file(inode))
3040 return CURSEG_COLD_DATA;
3041 if (file_is_hot(inode) ||
3042 is_inode_flag_set(inode, FI_HOT_DATA) ||
3043 f2fs_is_atomic_file(inode) ||
3044 f2fs_is_volatile_file(inode))
3045 return CURSEG_HOT_DATA;
3046 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
3047 } else {
3048 if (IS_DNODE(fio->page))
3049 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
3050 CURSEG_HOT_NODE;
3051 return CURSEG_COLD_NODE;
3052 }
3053 }
3054
3055 static int __get_segment_type(struct f2fs_io_info *fio)
3056 {
3057 int type = 0;
3058
3059 switch (F2FS_OPTION(fio->sbi).active_logs) {
3060 case 2:
3061 type = __get_segment_type_2(fio);
3062 break;
3063 case 4:
3064 type = __get_segment_type_4(fio);
3065 break;
3066 case 6:
3067 type = __get_segment_type_6(fio);
3068 break;
3069 default:
3070 f2fs_bug_on(fio->sbi, true);
3071 }
3072
3073 if (IS_HOT(type))
3074 fio->temp = HOT;
3075 else if (IS_WARM(type))
3076 fio->temp = WARM;
3077 else
3078 fio->temp = COLD;
3079 return type;
3080 }
3081
3082 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
3083 block_t old_blkaddr, block_t *new_blkaddr,
3084 struct f2fs_summary *sum, int type,
3085 struct f2fs_io_info *fio, bool add_list)
3086 {
3087 struct sit_info *sit_i = SIT_I(sbi);
3088 struct curseg_info *curseg = CURSEG_I(sbi, type);
3089 bool put_pin_sem = false;
3090
3091 if (type == CURSEG_COLD_DATA) {
3092 /* GC during CURSEG_COLD_DATA_PINNED allocation */
3093 if (down_read_trylock(&sbi->pin_sem)) {
3094 put_pin_sem = true;
3095 } else {
3096 type = CURSEG_WARM_DATA;
3097 curseg = CURSEG_I(sbi, type);
3098 }
3099 } else if (type == CURSEG_COLD_DATA_PINNED) {
3100 type = CURSEG_COLD_DATA;
3101 }
3102
3103 down_read(&SM_I(sbi)->curseg_lock);
3104
3105 mutex_lock(&curseg->curseg_mutex);
3106 down_write(&sit_i->sentry_lock);
3107
3108 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
3109
3110 f2fs_wait_discard_bio(sbi, *new_blkaddr);
3111
3112 /*
3113 * __add_sum_entry should be resided under the curseg_mutex
3114 * because, this function updates a summary entry in the
3115 * current summary block.
3116 */
3117 __add_sum_entry(sbi, type, sum);
3118
3119 __refresh_next_blkoff(sbi, curseg);
3120
3121 stat_inc_block_count(sbi, curseg);
3122
3123 /*
3124 * SIT information should be updated before segment allocation,
3125 * since SSR needs latest valid block information.
3126 */
3127 update_sit_entry(sbi, *new_blkaddr, 1);
3128 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3129 update_sit_entry(sbi, old_blkaddr, -1);
3130
3131 if (!__has_curseg_space(sbi, type))
3132 sit_i->s_ops->allocate_segment(sbi, type, false);
3133
3134 /*
3135 * segment dirty status should be updated after segment allocation,
3136 * so we just need to update status only one time after previous
3137 * segment being closed.
3138 */
3139 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3140 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
3141
3142 up_write(&sit_i->sentry_lock);
3143
3144 if (page && IS_NODESEG(type)) {
3145 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
3146
3147 f2fs_inode_chksum_set(sbi, page);
3148 }
3149
3150 if (F2FS_IO_ALIGNED(sbi))
3151 fio->retry = false;
3152
3153 if (add_list) {
3154 struct f2fs_bio_info *io;
3155
3156 INIT_LIST_HEAD(&fio->list);
3157 fio->in_list = true;
3158 io = sbi->write_io[fio->type] + fio->temp;
3159 spin_lock(&io->io_lock);
3160 list_add_tail(&fio->list, &io->io_list);
3161 spin_unlock(&io->io_lock);
3162 }
3163
3164 mutex_unlock(&curseg->curseg_mutex);
3165
3166 up_read(&SM_I(sbi)->curseg_lock);
3167
3168 if (put_pin_sem)
3169 up_read(&sbi->pin_sem);
3170 }
3171
3172 static void update_device_state(struct f2fs_io_info *fio)
3173 {
3174 struct f2fs_sb_info *sbi = fio->sbi;
3175 unsigned int devidx;
3176
3177 if (!f2fs_is_multi_device(sbi))
3178 return;
3179
3180 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
3181
3182 /* update device state for fsync */
3183 f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
3184
3185 /* update device state for checkpoint */
3186 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
3187 spin_lock(&sbi->dev_lock);
3188 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
3189 spin_unlock(&sbi->dev_lock);
3190 }
3191 }
3192
3193 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
3194 {
3195 int type = __get_segment_type(fio);
3196 bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
3197
3198 if (keep_order)
3199 down_read(&fio->sbi->io_order_lock);
3200 reallocate:
3201 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
3202 &fio->new_blkaddr, sum, type, fio, true);
3203 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
3204 invalidate_mapping_pages(META_MAPPING(fio->sbi),
3205 fio->old_blkaddr, fio->old_blkaddr);
3206
3207 /* writeout dirty page into bdev */
3208 f2fs_submit_page_write(fio);
3209 if (fio->retry) {
3210 fio->old_blkaddr = fio->new_blkaddr;
3211 goto reallocate;
3212 }
3213
3214 update_device_state(fio);
3215
3216 if (keep_order)
3217 up_read(&fio->sbi->io_order_lock);
3218 }
3219
3220 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
3221 enum iostat_type io_type)
3222 {
3223 struct f2fs_io_info fio = {
3224 .sbi = sbi,
3225 .type = META,
3226 .temp = HOT,
3227 .op = REQ_OP_WRITE,
3228 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3229 .old_blkaddr = page->index,
3230 .new_blkaddr = page->index,
3231 .page = page,
3232 .encrypted_page = NULL,
3233 .in_list = false,
3234 };
3235
3236 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3237 fio.op_flags &= ~REQ_META;
3238
3239 set_page_writeback(page);
3240 ClearPageError(page);
3241 f2fs_submit_page_write(&fio);
3242
3243 stat_inc_meta_count(sbi, page->index);
3244 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3245 }
3246
3247 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3248 {
3249 struct f2fs_summary sum;
3250
3251 set_summary(&sum, nid, 0, 0);
3252 do_write_page(&sum, fio);
3253
3254 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3255 }
3256
3257 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3258 struct f2fs_io_info *fio)
3259 {
3260 struct f2fs_sb_info *sbi = fio->sbi;
3261 struct f2fs_summary sum;
3262
3263 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3264 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3265 do_write_page(&sum, fio);
3266 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3267
3268 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3269 }
3270
3271 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3272 {
3273 int err;
3274 struct f2fs_sb_info *sbi = fio->sbi;
3275 unsigned int segno;
3276
3277 fio->new_blkaddr = fio->old_blkaddr;
3278 /* i/o temperature is needed for passing down write hints */
3279 __get_segment_type(fio);
3280
3281 segno = GET_SEGNO(sbi, fio->new_blkaddr);
3282
3283 if (!IS_DATASEG(get_seg_entry(sbi, segno)->type)) {
3284 set_sbi_flag(sbi, SBI_NEED_FSCK);
3285 f2fs_warn(sbi, "%s: incorrect segment(%u) type, run fsck to fix.",
3286 __func__, segno);
3287 return -EFSCORRUPTED;
3288 }
3289
3290 stat_inc_inplace_blocks(fio->sbi);
3291
3292 if (fio->bio && !(SM_I(sbi)->ipu_policy & (1 << F2FS_IPU_NOCACHE)))
3293 err = f2fs_merge_page_bio(fio);
3294 else
3295 err = f2fs_submit_page_bio(fio);
3296 if (!err) {
3297 update_device_state(fio);
3298 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3299 }
3300
3301 return err;
3302 }
3303
3304 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3305 unsigned int segno)
3306 {
3307 int i;
3308
3309 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3310 if (CURSEG_I(sbi, i)->segno == segno)
3311 break;
3312 }
3313 return i;
3314 }
3315
3316 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3317 block_t old_blkaddr, block_t new_blkaddr,
3318 bool recover_curseg, bool recover_newaddr)
3319 {
3320 struct sit_info *sit_i = SIT_I(sbi);
3321 struct curseg_info *curseg;
3322 unsigned int segno, old_cursegno;
3323 struct seg_entry *se;
3324 int type;
3325 unsigned short old_blkoff;
3326
3327 segno = GET_SEGNO(sbi, new_blkaddr);
3328 se = get_seg_entry(sbi, segno);
3329 type = se->type;
3330
3331 down_write(&SM_I(sbi)->curseg_lock);
3332
3333 if (!recover_curseg) {
3334 /* for recovery flow */
3335 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3336 if (old_blkaddr == NULL_ADDR)
3337 type = CURSEG_COLD_DATA;
3338 else
3339 type = CURSEG_WARM_DATA;
3340 }
3341 } else {
3342 if (IS_CURSEG(sbi, segno)) {
3343 /* se->type is volatile as SSR allocation */
3344 type = __f2fs_get_curseg(sbi, segno);
3345 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3346 } else {
3347 type = CURSEG_WARM_DATA;
3348 }
3349 }
3350
3351 f2fs_bug_on(sbi, !IS_DATASEG(type));
3352 curseg = CURSEG_I(sbi, type);
3353
3354 mutex_lock(&curseg->curseg_mutex);
3355 down_write(&sit_i->sentry_lock);
3356
3357 old_cursegno = curseg->segno;
3358 old_blkoff = curseg->next_blkoff;
3359
3360 /* change the current segment */
3361 if (segno != curseg->segno) {
3362 curseg->next_segno = segno;
3363 change_curseg(sbi, type);
3364 }
3365
3366 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3367 __add_sum_entry(sbi, type, sum);
3368
3369 if (!recover_curseg || recover_newaddr)
3370 update_sit_entry(sbi, new_blkaddr, 1);
3371 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3372 invalidate_mapping_pages(META_MAPPING(sbi),
3373 old_blkaddr, old_blkaddr);
3374 update_sit_entry(sbi, old_blkaddr, -1);
3375 }
3376
3377 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3378 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3379
3380 locate_dirty_segment(sbi, old_cursegno);
3381
3382 if (recover_curseg) {
3383 if (old_cursegno != curseg->segno) {
3384 curseg->next_segno = old_cursegno;
3385 change_curseg(sbi, type);
3386 }
3387 curseg->next_blkoff = old_blkoff;
3388 }
3389
3390 up_write(&sit_i->sentry_lock);
3391 mutex_unlock(&curseg->curseg_mutex);
3392 up_write(&SM_I(sbi)->curseg_lock);
3393 }
3394
3395 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3396 block_t old_addr, block_t new_addr,
3397 unsigned char version, bool recover_curseg,
3398 bool recover_newaddr)
3399 {
3400 struct f2fs_summary sum;
3401
3402 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3403
3404 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3405 recover_curseg, recover_newaddr);
3406
3407 f2fs_update_data_blkaddr(dn, new_addr);
3408 }
3409
3410 void f2fs_wait_on_page_writeback(struct page *page,
3411 enum page_type type, bool ordered, bool locked)
3412 {
3413 if (PageWriteback(page)) {
3414 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3415
3416 /* submit cached LFS IO */
3417 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, type);
3418 /* sbumit cached IPU IO */
3419 f2fs_submit_merged_ipu_write(sbi, NULL, page);
3420 if (ordered) {
3421 wait_on_page_writeback(page);
3422 f2fs_bug_on(sbi, locked && PageWriteback(page));
3423 } else {
3424 wait_for_stable_page(page);
3425 }
3426 }
3427 }
3428
3429 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3430 {
3431 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3432 struct page *cpage;
3433
3434 if (!f2fs_post_read_required(inode))
3435 return;
3436
3437 if (!__is_valid_data_blkaddr(blkaddr))
3438 return;
3439
3440 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3441 if (cpage) {
3442 f2fs_wait_on_page_writeback(cpage, DATA, true, true);
3443 f2fs_put_page(cpage, 1);
3444 }
3445 }
3446
3447 void f2fs_wait_on_block_writeback_range(struct inode *inode, block_t blkaddr,
3448 block_t len)
3449 {
3450 block_t i;
3451
3452 for (i = 0; i < len; i++)
3453 f2fs_wait_on_block_writeback(inode, blkaddr + i);
3454 }
3455
3456 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3457 {
3458 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3459 struct curseg_info *seg_i;
3460 unsigned char *kaddr;
3461 struct page *page;
3462 block_t start;
3463 int i, j, offset;
3464
3465 start = start_sum_block(sbi);
3466
3467 page = f2fs_get_meta_page(sbi, start++);
3468 if (IS_ERR(page))
3469 return PTR_ERR(page);
3470 kaddr = (unsigned char *)page_address(page);
3471
3472 /* Step 1: restore nat cache */
3473 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3474 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3475
3476 /* Step 2: restore sit cache */
3477 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3478 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3479 offset = 2 * SUM_JOURNAL_SIZE;
3480
3481 /* Step 3: restore summary entries */
3482 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3483 unsigned short blk_off;
3484 unsigned int segno;
3485
3486 seg_i = CURSEG_I(sbi, i);
3487 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3488 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3489 seg_i->next_segno = segno;
3490 reset_curseg(sbi, i, 0);
3491 seg_i->alloc_type = ckpt->alloc_type[i];
3492 seg_i->next_blkoff = blk_off;
3493
3494 if (seg_i->alloc_type == SSR)
3495 blk_off = sbi->blocks_per_seg;
3496
3497 for (j = 0; j < blk_off; j++) {
3498 struct f2fs_summary *s;
3499 s = (struct f2fs_summary *)(kaddr + offset);
3500 seg_i->sum_blk->entries[j] = *s;
3501 offset += SUMMARY_SIZE;
3502 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3503 SUM_FOOTER_SIZE)
3504 continue;
3505
3506 f2fs_put_page(page, 1);
3507 page = NULL;
3508
3509 page = f2fs_get_meta_page(sbi, start++);
3510 if (IS_ERR(page))
3511 return PTR_ERR(page);
3512 kaddr = (unsigned char *)page_address(page);
3513 offset = 0;
3514 }
3515 }
3516 f2fs_put_page(page, 1);
3517 return 0;
3518 }
3519
3520 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3521 {
3522 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3523 struct f2fs_summary_block *sum;
3524 struct curseg_info *curseg;
3525 struct page *new;
3526 unsigned short blk_off;
3527 unsigned int segno = 0;
3528 block_t blk_addr = 0;
3529 int err = 0;
3530
3531 /* get segment number and block addr */
3532 if (IS_DATASEG(type)) {
3533 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3534 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3535 CURSEG_HOT_DATA]);
3536 if (__exist_node_summaries(sbi))
3537 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3538 else
3539 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3540 } else {
3541 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3542 CURSEG_HOT_NODE]);
3543 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3544 CURSEG_HOT_NODE]);
3545 if (__exist_node_summaries(sbi))
3546 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3547 type - CURSEG_HOT_NODE);
3548 else
3549 blk_addr = GET_SUM_BLOCK(sbi, segno);
3550 }
3551
3552 new = f2fs_get_meta_page(sbi, blk_addr);
3553 if (IS_ERR(new))
3554 return PTR_ERR(new);
3555 sum = (struct f2fs_summary_block *)page_address(new);
3556
3557 if (IS_NODESEG(type)) {
3558 if (__exist_node_summaries(sbi)) {
3559 struct f2fs_summary *ns = &sum->entries[0];
3560 int i;
3561 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3562 ns->version = 0;
3563 ns->ofs_in_node = 0;
3564 }
3565 } else {
3566 err = f2fs_restore_node_summary(sbi, segno, sum);
3567 if (err)
3568 goto out;
3569 }
3570 }
3571
3572 /* set uncompleted segment to curseg */
3573 curseg = CURSEG_I(sbi, type);
3574 mutex_lock(&curseg->curseg_mutex);
3575
3576 /* update journal info */
3577 down_write(&curseg->journal_rwsem);
3578 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3579 up_write(&curseg->journal_rwsem);
3580
3581 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3582 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3583 curseg->next_segno = segno;
3584 reset_curseg(sbi, type, 0);
3585 curseg->alloc_type = ckpt->alloc_type[type];
3586 curseg->next_blkoff = blk_off;
3587 mutex_unlock(&curseg->curseg_mutex);
3588 out:
3589 f2fs_put_page(new, 1);
3590 return err;
3591 }
3592
3593 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3594 {
3595 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3596 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3597 int type = CURSEG_HOT_DATA;
3598 int err;
3599
3600 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3601 int npages = f2fs_npages_for_summary_flush(sbi, true);
3602
3603 if (npages >= 2)
3604 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3605 META_CP, true);
3606
3607 /* restore for compacted data summary */
3608 err = read_compacted_summaries(sbi);
3609 if (err)
3610 return err;
3611 type = CURSEG_HOT_NODE;
3612 }
3613
3614 if (__exist_node_summaries(sbi))
3615 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3616 NR_CURSEG_TYPE - type, META_CP, true);
3617
3618 for (; type <= CURSEG_COLD_NODE; type++) {
3619 err = read_normal_summaries(sbi, type);
3620 if (err)
3621 return err;
3622 }
3623
3624 /* sanity check for summary blocks */
3625 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3626 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES) {
3627 f2fs_err(sbi, "invalid journal entries nats %u sits %u\n",
3628 nats_in_cursum(nat_j), sits_in_cursum(sit_j));
3629 return -EINVAL;
3630 }
3631
3632 return 0;
3633 }
3634
3635 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3636 {
3637 struct page *page;
3638 unsigned char *kaddr;
3639 struct f2fs_summary *summary;
3640 struct curseg_info *seg_i;
3641 int written_size = 0;
3642 int i, j;
3643
3644 page = f2fs_grab_meta_page(sbi, blkaddr++);
3645 kaddr = (unsigned char *)page_address(page);
3646 memset(kaddr, 0, PAGE_SIZE);
3647
3648 /* Step 1: write nat cache */
3649 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3650 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3651 written_size += SUM_JOURNAL_SIZE;
3652
3653 /* Step 2: write sit cache */
3654 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3655 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3656 written_size += SUM_JOURNAL_SIZE;
3657
3658 /* Step 3: write summary entries */
3659 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3660 unsigned short blkoff;
3661 seg_i = CURSEG_I(sbi, i);
3662 if (sbi->ckpt->alloc_type[i] == SSR)
3663 blkoff = sbi->blocks_per_seg;
3664 else
3665 blkoff = curseg_blkoff(sbi, i);
3666
3667 for (j = 0; j < blkoff; j++) {
3668 if (!page) {
3669 page = f2fs_grab_meta_page(sbi, blkaddr++);
3670 kaddr = (unsigned char *)page_address(page);
3671 memset(kaddr, 0, PAGE_SIZE);
3672 written_size = 0;
3673 }
3674 summary = (struct f2fs_summary *)(kaddr + written_size);
3675 *summary = seg_i->sum_blk->entries[j];
3676 written_size += SUMMARY_SIZE;
3677
3678 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3679 SUM_FOOTER_SIZE)
3680 continue;
3681
3682 set_page_dirty(page);
3683 f2fs_put_page(page, 1);
3684 page = NULL;
3685 }
3686 }
3687 if (page) {
3688 set_page_dirty(page);
3689 f2fs_put_page(page, 1);
3690 }
3691 }
3692
3693 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3694 block_t blkaddr, int type)
3695 {
3696 int i, end;
3697 if (IS_DATASEG(type))
3698 end = type + NR_CURSEG_DATA_TYPE;
3699 else
3700 end = type + NR_CURSEG_NODE_TYPE;
3701
3702 for (i = type; i < end; i++)
3703 write_current_sum_page(sbi, i, blkaddr + (i - type));
3704 }
3705
3706 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3707 {
3708 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3709 write_compacted_summaries(sbi, start_blk);
3710 else
3711 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3712 }
3713
3714 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3715 {
3716 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3717 }
3718
3719 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3720 unsigned int val, int alloc)
3721 {
3722 int i;
3723
3724 if (type == NAT_JOURNAL) {
3725 for (i = 0; i < nats_in_cursum(journal); i++) {
3726 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3727 return i;
3728 }
3729 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3730 return update_nats_in_cursum(journal, 1);
3731 } else if (type == SIT_JOURNAL) {
3732 for (i = 0; i < sits_in_cursum(journal); i++)
3733 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3734 return i;
3735 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3736 return update_sits_in_cursum(journal, 1);
3737 }
3738 return -1;
3739 }
3740
3741 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3742 unsigned int segno)
3743 {
3744 return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3745 }
3746
3747 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3748 unsigned int start)
3749 {
3750 struct sit_info *sit_i = SIT_I(sbi);
3751 struct page *page;
3752 pgoff_t src_off, dst_off;
3753
3754 src_off = current_sit_addr(sbi, start);
3755 dst_off = next_sit_addr(sbi, src_off);
3756
3757 page = f2fs_grab_meta_page(sbi, dst_off);
3758 seg_info_to_sit_page(sbi, page, start);
3759
3760 set_page_dirty(page);
3761 set_to_next_sit(sit_i, start);
3762
3763 return page;
3764 }
3765
3766 static struct sit_entry_set *grab_sit_entry_set(void)
3767 {
3768 struct sit_entry_set *ses =
3769 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3770
3771 ses->entry_cnt = 0;
3772 INIT_LIST_HEAD(&ses->set_list);
3773 return ses;
3774 }
3775
3776 static void release_sit_entry_set(struct sit_entry_set *ses)
3777 {
3778 list_del(&ses->set_list);
3779 kmem_cache_free(sit_entry_set_slab, ses);
3780 }
3781
3782 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3783 struct list_head *head)
3784 {
3785 struct sit_entry_set *next = ses;
3786
3787 if (list_is_last(&ses->set_list, head))
3788 return;
3789
3790 list_for_each_entry_continue(next, head, set_list)
3791 if (ses->entry_cnt <= next->entry_cnt)
3792 break;
3793
3794 list_move_tail(&ses->set_list, &next->set_list);
3795 }
3796
3797 static void add_sit_entry(unsigned int segno, struct list_head *head)
3798 {
3799 struct sit_entry_set *ses;
3800 unsigned int start_segno = START_SEGNO(segno);
3801
3802 list_for_each_entry(ses, head, set_list) {
3803 if (ses->start_segno == start_segno) {
3804 ses->entry_cnt++;
3805 adjust_sit_entry_set(ses, head);
3806 return;
3807 }
3808 }
3809
3810 ses = grab_sit_entry_set();
3811
3812 ses->start_segno = start_segno;
3813 ses->entry_cnt++;
3814 list_add(&ses->set_list, head);
3815 }
3816
3817 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3818 {
3819 struct f2fs_sm_info *sm_info = SM_I(sbi);
3820 struct list_head *set_list = &sm_info->sit_entry_set;
3821 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3822 unsigned int segno;
3823
3824 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3825 add_sit_entry(segno, set_list);
3826 }
3827
3828 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3829 {
3830 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3831 struct f2fs_journal *journal = curseg->journal;
3832 int i;
3833
3834 down_write(&curseg->journal_rwsem);
3835 for (i = 0; i < sits_in_cursum(journal); i++) {
3836 unsigned int segno;
3837 bool dirtied;
3838
3839 segno = le32_to_cpu(segno_in_journal(journal, i));
3840 dirtied = __mark_sit_entry_dirty(sbi, segno);
3841
3842 if (!dirtied)
3843 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3844 }
3845 update_sits_in_cursum(journal, -i);
3846 up_write(&curseg->journal_rwsem);
3847 }
3848
3849 /*
3850 * CP calls this function, which flushes SIT entries including sit_journal,
3851 * and moves prefree segs to free segs.
3852 */
3853 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3854 {
3855 struct sit_info *sit_i = SIT_I(sbi);
3856 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3857 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3858 struct f2fs_journal *journal = curseg->journal;
3859 struct sit_entry_set *ses, *tmp;
3860 struct list_head *head = &SM_I(sbi)->sit_entry_set;
3861 bool to_journal = !is_sbi_flag_set(sbi, SBI_IS_RESIZEFS);
3862 struct seg_entry *se;
3863
3864 down_write(&sit_i->sentry_lock);
3865
3866 if (!sit_i->dirty_sentries)
3867 goto out;
3868
3869 /*
3870 * add and account sit entries of dirty bitmap in sit entry
3871 * set temporarily
3872 */
3873 add_sits_in_set(sbi);
3874
3875 /*
3876 * if there are no enough space in journal to store dirty sit
3877 * entries, remove all entries from journal and add and account
3878 * them in sit entry set.
3879 */
3880 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL) ||
3881 !to_journal)
3882 remove_sits_in_journal(sbi);
3883
3884 /*
3885 * there are two steps to flush sit entries:
3886 * #1, flush sit entries to journal in current cold data summary block.
3887 * #2, flush sit entries to sit page.
3888 */
3889 list_for_each_entry_safe(ses, tmp, head, set_list) {
3890 struct page *page = NULL;
3891 struct f2fs_sit_block *raw_sit = NULL;
3892 unsigned int start_segno = ses->start_segno;
3893 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3894 (unsigned long)MAIN_SEGS(sbi));
3895 unsigned int segno = start_segno;
3896
3897 if (to_journal &&
3898 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3899 to_journal = false;
3900
3901 if (to_journal) {
3902 down_write(&curseg->journal_rwsem);
3903 } else {
3904 page = get_next_sit_page(sbi, start_segno);
3905 raw_sit = page_address(page);
3906 }
3907
3908 /* flush dirty sit entries in region of current sit set */
3909 for_each_set_bit_from(segno, bitmap, end) {
3910 int offset, sit_offset;
3911
3912 se = get_seg_entry(sbi, segno);
3913 #ifdef CONFIG_F2FS_CHECK_FS
3914 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3915 SIT_VBLOCK_MAP_SIZE))
3916 f2fs_bug_on(sbi, 1);
3917 #endif
3918
3919 /* add discard candidates */
3920 if (!(cpc->reason & CP_DISCARD)) {
3921 cpc->trim_start = segno;
3922 add_discard_addrs(sbi, cpc, false);
3923 }
3924
3925 if (to_journal) {
3926 offset = f2fs_lookup_journal_in_cursum(journal,
3927 SIT_JOURNAL, segno, 1);
3928 f2fs_bug_on(sbi, offset < 0);
3929 segno_in_journal(journal, offset) =
3930 cpu_to_le32(segno);
3931 seg_info_to_raw_sit(se,
3932 &sit_in_journal(journal, offset));
3933 check_block_count(sbi, segno,
3934 &sit_in_journal(journal, offset));
3935 } else {
3936 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3937 seg_info_to_raw_sit(se,
3938 &raw_sit->entries[sit_offset]);
3939 check_block_count(sbi, segno,
3940 &raw_sit->entries[sit_offset]);
3941 }
3942
3943 __clear_bit(segno, bitmap);
3944 sit_i->dirty_sentries--;
3945 ses->entry_cnt--;
3946 }
3947
3948 if (to_journal)
3949 up_write(&curseg->journal_rwsem);
3950 else
3951 f2fs_put_page(page, 1);
3952
3953 f2fs_bug_on(sbi, ses->entry_cnt);
3954 release_sit_entry_set(ses);
3955 }
3956
3957 f2fs_bug_on(sbi, !list_empty(head));
3958 f2fs_bug_on(sbi, sit_i->dirty_sentries);
3959 out:
3960 if (cpc->reason & CP_DISCARD) {
3961 __u64 trim_start = cpc->trim_start;
3962
3963 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3964 add_discard_addrs(sbi, cpc, false);
3965
3966 cpc->trim_start = trim_start;
3967 }
3968 up_write(&sit_i->sentry_lock);
3969
3970 set_prefree_as_free_segments(sbi);
3971 }
3972
3973 static int build_sit_info(struct f2fs_sb_info *sbi)
3974 {
3975 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3976 struct sit_info *sit_i;
3977 unsigned int sit_segs, start;
3978 char *src_bitmap, *bitmap;
3979 unsigned int bitmap_size, main_bitmap_size, sit_bitmap_size;
3980
3981 /* allocate memory for SIT information */
3982 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3983 if (!sit_i)
3984 return -ENOMEM;
3985
3986 SM_I(sbi)->sit_info = sit_i;
3987
3988 sit_i->sentries =
3989 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3990 MAIN_SEGS(sbi)),
3991 GFP_KERNEL);
3992 if (!sit_i->sentries)
3993 return -ENOMEM;
3994
3995 main_bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3996 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, main_bitmap_size,
3997 GFP_KERNEL);
3998 if (!sit_i->dirty_sentries_bitmap)
3999 return -ENOMEM;
4000
4001 #ifdef CONFIG_F2FS_CHECK_FS
4002 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 4;
4003 #else
4004 bitmap_size = MAIN_SEGS(sbi) * SIT_VBLOCK_MAP_SIZE * 3;
4005 #endif
4006 sit_i->bitmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4007 if (!sit_i->bitmap)
4008 return -ENOMEM;
4009
4010 bitmap = sit_i->bitmap;
4011
4012 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4013 sit_i->sentries[start].cur_valid_map = bitmap;
4014 bitmap += SIT_VBLOCK_MAP_SIZE;
4015
4016 sit_i->sentries[start].ckpt_valid_map = bitmap;
4017 bitmap += SIT_VBLOCK_MAP_SIZE;
4018
4019 #ifdef CONFIG_F2FS_CHECK_FS
4020 sit_i->sentries[start].cur_valid_map_mir = bitmap;
4021 bitmap += SIT_VBLOCK_MAP_SIZE;
4022 #endif
4023
4024 sit_i->sentries[start].discard_map = bitmap;
4025 bitmap += SIT_VBLOCK_MAP_SIZE;
4026 }
4027
4028 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
4029 if (!sit_i->tmp_map)
4030 return -ENOMEM;
4031
4032 if (__is_large_section(sbi)) {
4033 sit_i->sec_entries =
4034 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
4035 MAIN_SECS(sbi)),
4036 GFP_KERNEL);
4037 if (!sit_i->sec_entries)
4038 return -ENOMEM;
4039 }
4040
4041 /* get information related with SIT */
4042 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
4043
4044 /* setup SIT bitmap from ckeckpoint pack */
4045 sit_bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
4046 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
4047
4048 sit_i->sit_bitmap = kmemdup(src_bitmap, sit_bitmap_size, GFP_KERNEL);
4049 if (!sit_i->sit_bitmap)
4050 return -ENOMEM;
4051
4052 #ifdef CONFIG_F2FS_CHECK_FS
4053 sit_i->sit_bitmap_mir = kmemdup(src_bitmap,
4054 sit_bitmap_size, GFP_KERNEL);
4055 if (!sit_i->sit_bitmap_mir)
4056 return -ENOMEM;
4057
4058 sit_i->invalid_segmap = f2fs_kvzalloc(sbi,
4059 main_bitmap_size, GFP_KERNEL);
4060 if (!sit_i->invalid_segmap)
4061 return -ENOMEM;
4062 #endif
4063
4064 /* init SIT information */
4065 sit_i->s_ops = &default_salloc_ops;
4066
4067 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
4068 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
4069 sit_i->written_valid_blocks = 0;
4070 sit_i->bitmap_size = sit_bitmap_size;
4071 sit_i->dirty_sentries = 0;
4072 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
4073 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
4074 sit_i->mounted_time = ktime_get_real_seconds();
4075 init_rwsem(&sit_i->sentry_lock);
4076 return 0;
4077 }
4078
4079 static int build_free_segmap(struct f2fs_sb_info *sbi)
4080 {
4081 struct free_segmap_info *free_i;
4082 unsigned int bitmap_size, sec_bitmap_size;
4083
4084 /* allocate memory for free segmap information */
4085 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
4086 if (!free_i)
4087 return -ENOMEM;
4088
4089 SM_I(sbi)->free_info = free_i;
4090
4091 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4092 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
4093 if (!free_i->free_segmap)
4094 return -ENOMEM;
4095
4096 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4097 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
4098 if (!free_i->free_secmap)
4099 return -ENOMEM;
4100
4101 /* set all segments as dirty temporarily */
4102 memset(free_i->free_segmap, 0xff, bitmap_size);
4103 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
4104
4105 /* init free segmap information */
4106 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
4107 free_i->free_segments = 0;
4108 free_i->free_sections = 0;
4109 spin_lock_init(&free_i->segmap_lock);
4110 return 0;
4111 }
4112
4113 static int build_curseg(struct f2fs_sb_info *sbi)
4114 {
4115 struct curseg_info *array;
4116 int i;
4117
4118 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
4119 GFP_KERNEL);
4120 if (!array)
4121 return -ENOMEM;
4122
4123 SM_I(sbi)->curseg_array = array;
4124
4125 for (i = 0; i < NR_CURSEG_TYPE; i++) {
4126 mutex_init(&array[i].curseg_mutex);
4127 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
4128 if (!array[i].sum_blk)
4129 return -ENOMEM;
4130 init_rwsem(&array[i].journal_rwsem);
4131 array[i].journal = f2fs_kzalloc(sbi,
4132 sizeof(struct f2fs_journal), GFP_KERNEL);
4133 if (!array[i].journal)
4134 return -ENOMEM;
4135 array[i].segno = NULL_SEGNO;
4136 array[i].next_blkoff = 0;
4137 }
4138 return restore_curseg_summaries(sbi);
4139 }
4140
4141 static int build_sit_entries(struct f2fs_sb_info *sbi)
4142 {
4143 struct sit_info *sit_i = SIT_I(sbi);
4144 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
4145 struct f2fs_journal *journal = curseg->journal;
4146 struct seg_entry *se;
4147 struct f2fs_sit_entry sit;
4148 int sit_blk_cnt = SIT_BLK_CNT(sbi);
4149 unsigned int i, start, end;
4150 unsigned int readed, start_blk = 0;
4151 int err = 0;
4152 block_t total_node_blocks = 0;
4153
4154 do {
4155 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
4156 META_SIT, true);
4157
4158 start = start_blk * sit_i->sents_per_block;
4159 end = (start_blk + readed) * sit_i->sents_per_block;
4160
4161 for (; start < end && start < MAIN_SEGS(sbi); start++) {
4162 struct f2fs_sit_block *sit_blk;
4163 struct page *page;
4164
4165 se = &sit_i->sentries[start];
4166 page = get_current_sit_page(sbi, start);
4167 if (IS_ERR(page))
4168 return PTR_ERR(page);
4169 sit_blk = (struct f2fs_sit_block *)page_address(page);
4170 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
4171 f2fs_put_page(page, 1);
4172
4173 err = check_block_count(sbi, start, &sit);
4174 if (err)
4175 return err;
4176 seg_info_from_raw_sit(se, &sit);
4177 if (IS_NODESEG(se->type))
4178 total_node_blocks += se->valid_blocks;
4179
4180 /* build discard map only one time */
4181 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4182 memset(se->discard_map, 0xff,
4183 SIT_VBLOCK_MAP_SIZE);
4184 } else {
4185 memcpy(se->discard_map,
4186 se->cur_valid_map,
4187 SIT_VBLOCK_MAP_SIZE);
4188 sbi->discard_blks +=
4189 sbi->blocks_per_seg -
4190 se->valid_blocks;
4191 }
4192
4193 if (__is_large_section(sbi))
4194 get_sec_entry(sbi, start)->valid_blocks +=
4195 se->valid_blocks;
4196 }
4197 start_blk += readed;
4198 } while (start_blk < sit_blk_cnt);
4199
4200 down_read(&curseg->journal_rwsem);
4201 for (i = 0; i < sits_in_cursum(journal); i++) {
4202 unsigned int old_valid_blocks;
4203
4204 start = le32_to_cpu(segno_in_journal(journal, i));
4205 if (start >= MAIN_SEGS(sbi)) {
4206 f2fs_err(sbi, "Wrong journal entry on segno %u",
4207 start);
4208 err = -EFSCORRUPTED;
4209 break;
4210 }
4211
4212 se = &sit_i->sentries[start];
4213 sit = sit_in_journal(journal, i);
4214
4215 old_valid_blocks = se->valid_blocks;
4216 if (IS_NODESEG(se->type))
4217 total_node_blocks -= old_valid_blocks;
4218
4219 err = check_block_count(sbi, start, &sit);
4220 if (err)
4221 break;
4222 seg_info_from_raw_sit(se, &sit);
4223 if (IS_NODESEG(se->type))
4224 total_node_blocks += se->valid_blocks;
4225
4226 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
4227 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
4228 } else {
4229 memcpy(se->discard_map, se->cur_valid_map,
4230 SIT_VBLOCK_MAP_SIZE);
4231 sbi->discard_blks += old_valid_blocks;
4232 sbi->discard_blks -= se->valid_blocks;
4233 }
4234
4235 if (__is_large_section(sbi)) {
4236 get_sec_entry(sbi, start)->valid_blocks +=
4237 se->valid_blocks;
4238 get_sec_entry(sbi, start)->valid_blocks -=
4239 old_valid_blocks;
4240 }
4241 }
4242 up_read(&curseg->journal_rwsem);
4243
4244 if (!err && total_node_blocks != valid_node_count(sbi)) {
4245 f2fs_err(sbi, "SIT is corrupted node# %u vs %u",
4246 total_node_blocks, valid_node_count(sbi));
4247 err = -EFSCORRUPTED;
4248 }
4249
4250 return err;
4251 }
4252
4253 static void init_free_segmap(struct f2fs_sb_info *sbi)
4254 {
4255 unsigned int start;
4256 int type;
4257
4258 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4259 struct seg_entry *sentry = get_seg_entry(sbi, start);
4260 if (!sentry->valid_blocks)
4261 __set_free(sbi, start);
4262 else
4263 SIT_I(sbi)->written_valid_blocks +=
4264 sentry->valid_blocks;
4265 }
4266
4267 /* set use the current segments */
4268 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4269 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4270 __set_test_and_inuse(sbi, curseg_t->segno);
4271 }
4272 }
4273
4274 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4275 {
4276 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4277 struct free_segmap_info *free_i = FREE_I(sbi);
4278 unsigned int segno = 0, offset = 0;
4279 unsigned short valid_blocks;
4280
4281 while (1) {
4282 /* find dirty segment based on free segmap */
4283 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4284 if (segno >= MAIN_SEGS(sbi))
4285 break;
4286 offset = segno + 1;
4287 valid_blocks = get_valid_blocks(sbi, segno, false);
4288 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4289 continue;
4290 if (valid_blocks > sbi->blocks_per_seg) {
4291 f2fs_bug_on(sbi, 1);
4292 continue;
4293 }
4294 mutex_lock(&dirty_i->seglist_lock);
4295 __locate_dirty_segment(sbi, segno, DIRTY);
4296 mutex_unlock(&dirty_i->seglist_lock);
4297 }
4298 }
4299
4300 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4301 {
4302 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4303 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4304
4305 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4306 if (!dirty_i->victim_secmap)
4307 return -ENOMEM;
4308 return 0;
4309 }
4310
4311 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4312 {
4313 struct dirty_seglist_info *dirty_i;
4314 unsigned int bitmap_size, i;
4315
4316 /* allocate memory for dirty segments list information */
4317 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4318 GFP_KERNEL);
4319 if (!dirty_i)
4320 return -ENOMEM;
4321
4322 SM_I(sbi)->dirty_info = dirty_i;
4323 mutex_init(&dirty_i->seglist_lock);
4324
4325 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4326
4327 for (i = 0; i < NR_DIRTY_TYPE; i++) {
4328 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4329 GFP_KERNEL);
4330 if (!dirty_i->dirty_segmap[i])
4331 return -ENOMEM;
4332 }
4333
4334 init_dirty_segmap(sbi);
4335 return init_victim_secmap(sbi);
4336 }
4337
4338 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
4339 {
4340 int i;
4341
4342 /*
4343 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
4344 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
4345 */
4346 for (i = 0; i < NO_CHECK_TYPE; i++) {
4347 struct curseg_info *curseg = CURSEG_I(sbi, i);
4348 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
4349 unsigned int blkofs = curseg->next_blkoff;
4350
4351 if (f2fs_test_bit(blkofs, se->cur_valid_map))
4352 goto out;
4353
4354 if (curseg->alloc_type == SSR)
4355 continue;
4356
4357 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
4358 if (!f2fs_test_bit(blkofs, se->cur_valid_map))
4359 continue;
4360 out:
4361 f2fs_err(sbi,
4362 "Current segment's next free block offset is inconsistent with bitmap, logtype:%u, segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
4363 i, curseg->segno, curseg->alloc_type,
4364 curseg->next_blkoff, blkofs);
4365 return -EFSCORRUPTED;
4366 }
4367 }
4368 return 0;
4369 }
4370
4371 #ifdef CONFIG_BLK_DEV_ZONED
4372
4373 static int check_zone_write_pointer(struct f2fs_sb_info *sbi,
4374 struct f2fs_dev_info *fdev,
4375 struct blk_zone *zone)
4376 {
4377 unsigned int wp_segno, wp_blkoff, zone_secno, zone_segno, segno;
4378 block_t zone_block, wp_block, last_valid_block;
4379 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4380 int i, s, b, ret;
4381 struct seg_entry *se;
4382
4383 if (zone->type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4384 return 0;
4385
4386 wp_block = fdev->start_blk + (zone->wp >> log_sectors_per_block);
4387 wp_segno = GET_SEGNO(sbi, wp_block);
4388 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4389 zone_block = fdev->start_blk + (zone->start >> log_sectors_per_block);
4390 zone_segno = GET_SEGNO(sbi, zone_block);
4391 zone_secno = GET_SEC_FROM_SEG(sbi, zone_segno);
4392
4393 if (zone_segno >= MAIN_SEGS(sbi))
4394 return 0;
4395
4396 /*
4397 * Skip check of zones cursegs point to, since
4398 * fix_curseg_write_pointer() checks them.
4399 */
4400 for (i = 0; i < NO_CHECK_TYPE; i++)
4401 if (zone_secno == GET_SEC_FROM_SEG(sbi,
4402 CURSEG_I(sbi, i)->segno))
4403 return 0;
4404
4405 /*
4406 * Get last valid block of the zone.
4407 */
4408 last_valid_block = zone_block - 1;
4409 for (s = sbi->segs_per_sec - 1; s >= 0; s--) {
4410 segno = zone_segno + s;
4411 se = get_seg_entry(sbi, segno);
4412 for (b = sbi->blocks_per_seg - 1; b >= 0; b--)
4413 if (f2fs_test_bit(b, se->cur_valid_map)) {
4414 last_valid_block = START_BLOCK(sbi, segno) + b;
4415 break;
4416 }
4417 if (last_valid_block >= zone_block)
4418 break;
4419 }
4420
4421 /*
4422 * If last valid block is beyond the write pointer, report the
4423 * inconsistency. This inconsistency does not cause write error
4424 * because the zone will not be selected for write operation until
4425 * it get discarded. Just report it.
4426 */
4427 if (last_valid_block >= wp_block) {
4428 f2fs_notice(sbi, "Valid block beyond write pointer: "
4429 "valid block[0x%x,0x%x] wp[0x%x,0x%x]",
4430 GET_SEGNO(sbi, last_valid_block),
4431 GET_BLKOFF_FROM_SEG0(sbi, last_valid_block),
4432 wp_segno, wp_blkoff);
4433 return 0;
4434 }
4435
4436 /*
4437 * If there is no valid block in the zone and if write pointer is
4438 * not at zone start, reset the write pointer.
4439 */
4440 if (last_valid_block + 1 == zone_block && zone->wp != zone->start) {
4441 f2fs_notice(sbi,
4442 "Zone without valid block has non-zero write "
4443 "pointer. Reset the write pointer: wp[0x%x,0x%x]",
4444 wp_segno, wp_blkoff);
4445 ret = __f2fs_issue_discard_zone(sbi, fdev->bdev, zone_block,
4446 zone->len >> log_sectors_per_block);
4447 if (ret) {
4448 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4449 fdev->path, ret);
4450 return ret;
4451 }
4452 }
4453
4454 return 0;
4455 }
4456
4457 static struct f2fs_dev_info *get_target_zoned_dev(struct f2fs_sb_info *sbi,
4458 block_t zone_blkaddr)
4459 {
4460 int i;
4461
4462 for (i = 0; i < sbi->s_ndevs; i++) {
4463 if (!bdev_is_zoned(FDEV(i).bdev))
4464 continue;
4465 if (sbi->s_ndevs == 1 || (FDEV(i).start_blk <= zone_blkaddr &&
4466 zone_blkaddr <= FDEV(i).end_blk))
4467 return &FDEV(i);
4468 }
4469
4470 return NULL;
4471 }
4472
4473 static int report_one_zone_cb(struct blk_zone *zone, unsigned int idx,
4474 void *data) {
4475 memcpy(data, zone, sizeof(struct blk_zone));
4476 return 0;
4477 }
4478
4479 static int fix_curseg_write_pointer(struct f2fs_sb_info *sbi, int type)
4480 {
4481 struct curseg_info *cs = CURSEG_I(sbi, type);
4482 struct f2fs_dev_info *zbd;
4483 struct blk_zone zone;
4484 unsigned int cs_section, wp_segno, wp_blkoff, wp_sector_off;
4485 block_t cs_zone_block, wp_block;
4486 unsigned int log_sectors_per_block = sbi->log_blocksize - SECTOR_SHIFT;
4487 sector_t zone_sector;
4488 int err;
4489
4490 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4491 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4492
4493 zbd = get_target_zoned_dev(sbi, cs_zone_block);
4494 if (!zbd)
4495 return 0;
4496
4497 /* report zone for the sector the curseg points to */
4498 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4499 << log_sectors_per_block;
4500 err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4501 report_one_zone_cb, &zone);
4502 if (err != 1) {
4503 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4504 zbd->path, err);
4505 return err;
4506 }
4507
4508 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4509 return 0;
4510
4511 wp_block = zbd->start_blk + (zone.wp >> log_sectors_per_block);
4512 wp_segno = GET_SEGNO(sbi, wp_block);
4513 wp_blkoff = wp_block - START_BLOCK(sbi, wp_segno);
4514 wp_sector_off = zone.wp & GENMASK(log_sectors_per_block - 1, 0);
4515
4516 if (cs->segno == wp_segno && cs->next_blkoff == wp_blkoff &&
4517 wp_sector_off == 0)
4518 return 0;
4519
4520 f2fs_notice(sbi, "Unaligned curseg[%d] with write pointer: "
4521 "curseg[0x%x,0x%x] wp[0x%x,0x%x]",
4522 type, cs->segno, cs->next_blkoff, wp_segno, wp_blkoff);
4523
4524 f2fs_notice(sbi, "Assign new section to curseg[%d]: "
4525 "curseg[0x%x,0x%x]", type, cs->segno, cs->next_blkoff);
4526 allocate_segment_by_default(sbi, type, true);
4527
4528 /* check consistency of the zone curseg pointed to */
4529 if (check_zone_write_pointer(sbi, zbd, &zone))
4530 return -EIO;
4531
4532 /* check newly assigned zone */
4533 cs_section = GET_SEC_FROM_SEG(sbi, cs->segno);
4534 cs_zone_block = START_BLOCK(sbi, GET_SEG_FROM_SEC(sbi, cs_section));
4535
4536 zbd = get_target_zoned_dev(sbi, cs_zone_block);
4537 if (!zbd)
4538 return 0;
4539
4540 zone_sector = (sector_t)(cs_zone_block - zbd->start_blk)
4541 << log_sectors_per_block;
4542 err = blkdev_report_zones(zbd->bdev, zone_sector, 1,
4543 report_one_zone_cb, &zone);
4544 if (err != 1) {
4545 f2fs_err(sbi, "Report zone failed: %s errno=(%d)",
4546 zbd->path, err);
4547 return err;
4548 }
4549
4550 if (zone.type != BLK_ZONE_TYPE_SEQWRITE_REQ)
4551 return 0;
4552
4553 if (zone.wp != zone.start) {
4554 f2fs_notice(sbi,
4555 "New zone for curseg[%d] is not yet discarded. "
4556 "Reset the zone: curseg[0x%x,0x%x]",
4557 type, cs->segno, cs->next_blkoff);
4558 err = __f2fs_issue_discard_zone(sbi, zbd->bdev,
4559 zone_sector >> log_sectors_per_block,
4560 zone.len >> log_sectors_per_block);
4561 if (err) {
4562 f2fs_err(sbi, "Discard zone failed: %s (errno=%d)",
4563 zbd->path, err);
4564 return err;
4565 }
4566 }
4567
4568 return 0;
4569 }
4570
4571 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4572 {
4573 int i, ret;
4574
4575 for (i = 0; i < NO_CHECK_TYPE; i++) {
4576 ret = fix_curseg_write_pointer(sbi, i);
4577 if (ret)
4578 return ret;
4579 }
4580
4581 return 0;
4582 }
4583
4584 struct check_zone_write_pointer_args {
4585 struct f2fs_sb_info *sbi;
4586 struct f2fs_dev_info *fdev;
4587 };
4588
4589 static int check_zone_write_pointer_cb(struct blk_zone *zone, unsigned int idx,
4590 void *data) {
4591 struct check_zone_write_pointer_args *args;
4592 args = (struct check_zone_write_pointer_args *)data;
4593
4594 return check_zone_write_pointer(args->sbi, args->fdev, zone);
4595 }
4596
4597 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4598 {
4599 int i, ret;
4600 struct check_zone_write_pointer_args args;
4601
4602 for (i = 0; i < sbi->s_ndevs; i++) {
4603 if (!bdev_is_zoned(FDEV(i).bdev))
4604 continue;
4605
4606 args.sbi = sbi;
4607 args.fdev = &FDEV(i);
4608 ret = blkdev_report_zones(FDEV(i).bdev, 0, BLK_ALL_ZONES,
4609 check_zone_write_pointer_cb, &args);
4610 if (ret < 0)
4611 return ret;
4612 }
4613
4614 return 0;
4615 }
4616 #else
4617 int f2fs_fix_curseg_write_pointer(struct f2fs_sb_info *sbi)
4618 {
4619 return 0;
4620 }
4621
4622 int f2fs_check_write_pointer(struct f2fs_sb_info *sbi)
4623 {
4624 return 0;
4625 }
4626 #endif
4627
4628 /*
4629 * Update min, max modified time for cost-benefit GC algorithm
4630 */
4631 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4632 {
4633 struct sit_info *sit_i = SIT_I(sbi);
4634 unsigned int segno;
4635
4636 down_write(&sit_i->sentry_lock);
4637
4638 sit_i->min_mtime = ULLONG_MAX;
4639
4640 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4641 unsigned int i;
4642 unsigned long long mtime = 0;
4643
4644 for (i = 0; i < sbi->segs_per_sec; i++)
4645 mtime += get_seg_entry(sbi, segno + i)->mtime;
4646
4647 mtime = div_u64(mtime, sbi->segs_per_sec);
4648
4649 if (sit_i->min_mtime > mtime)
4650 sit_i->min_mtime = mtime;
4651 }
4652 sit_i->max_mtime = get_mtime(sbi, false);
4653 up_write(&sit_i->sentry_lock);
4654 }
4655
4656 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4657 {
4658 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4659 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4660 struct f2fs_sm_info *sm_info;
4661 int err;
4662
4663 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4664 if (!sm_info)
4665 return -ENOMEM;
4666
4667 /* init sm info */
4668 sbi->sm_info = sm_info;
4669 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4670 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4671 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4672 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4673 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4674 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4675 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4676 sm_info->rec_prefree_segments = sm_info->main_segments *
4677 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4678 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4679 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4680
4681 if (!test_opt(sbi, LFS))
4682 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4683 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4684 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4685 sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4686 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4687 sm_info->min_ssr_sections = reserved_sections(sbi);
4688
4689 INIT_LIST_HEAD(&sm_info->sit_entry_set);
4690
4691 init_rwsem(&sm_info->curseg_lock);
4692
4693 if (!f2fs_readonly(sbi->sb)) {
4694 err = f2fs_create_flush_cmd_control(sbi);
4695 if (err)
4696 return err;
4697 }
4698
4699 err = create_discard_cmd_control(sbi);
4700 if (err)
4701 return err;
4702
4703 err = build_sit_info(sbi);
4704 if (err)
4705 return err;
4706 err = build_free_segmap(sbi);
4707 if (err)
4708 return err;
4709 err = build_curseg(sbi);
4710 if (err)
4711 return err;
4712
4713 /* reinit free segmap based on SIT */
4714 err = build_sit_entries(sbi);
4715 if (err)
4716 return err;
4717
4718 init_free_segmap(sbi);
4719 err = build_dirty_segmap(sbi);
4720 if (err)
4721 return err;
4722
4723 err = sanity_check_curseg(sbi);
4724 if (err)
4725 return err;
4726
4727 init_min_max_mtime(sbi);
4728 return 0;
4729 }
4730
4731 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4732 enum dirty_type dirty_type)
4733 {
4734 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4735
4736 mutex_lock(&dirty_i->seglist_lock);
4737 kvfree(dirty_i->dirty_segmap[dirty_type]);
4738 dirty_i->nr_dirty[dirty_type] = 0;
4739 mutex_unlock(&dirty_i->seglist_lock);
4740 }
4741
4742 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4743 {
4744 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4745 kvfree(dirty_i->victim_secmap);
4746 }
4747
4748 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4749 {
4750 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4751 int i;
4752
4753 if (!dirty_i)
4754 return;
4755
4756 /* discard pre-free/dirty segments list */
4757 for (i = 0; i < NR_DIRTY_TYPE; i++)
4758 discard_dirty_segmap(sbi, i);
4759
4760 destroy_victim_secmap(sbi);
4761 SM_I(sbi)->dirty_info = NULL;
4762 kvfree(dirty_i);
4763 }
4764
4765 static void destroy_curseg(struct f2fs_sb_info *sbi)
4766 {
4767 struct curseg_info *array = SM_I(sbi)->curseg_array;
4768 int i;
4769
4770 if (!array)
4771 return;
4772 SM_I(sbi)->curseg_array = NULL;
4773 for (i = 0; i < NR_CURSEG_TYPE; i++) {
4774 kvfree(array[i].sum_blk);
4775 kvfree(array[i].journal);
4776 }
4777 kvfree(array);
4778 }
4779
4780 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4781 {
4782 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4783 if (!free_i)
4784 return;
4785 SM_I(sbi)->free_info = NULL;
4786 kvfree(free_i->free_segmap);
4787 kvfree(free_i->free_secmap);
4788 kvfree(free_i);
4789 }
4790
4791 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4792 {
4793 struct sit_info *sit_i = SIT_I(sbi);
4794
4795 if (!sit_i)
4796 return;
4797
4798 if (sit_i->sentries)
4799 kvfree(sit_i->bitmap);
4800 kvfree(sit_i->tmp_map);
4801
4802 kvfree(sit_i->sentries);
4803 kvfree(sit_i->sec_entries);
4804 kvfree(sit_i->dirty_sentries_bitmap);
4805
4806 SM_I(sbi)->sit_info = NULL;
4807 kvfree(sit_i->sit_bitmap);
4808 #ifdef CONFIG_F2FS_CHECK_FS
4809 kvfree(sit_i->sit_bitmap_mir);
4810 kvfree(sit_i->invalid_segmap);
4811 #endif
4812 kvfree(sit_i);
4813 }
4814
4815 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4816 {
4817 struct f2fs_sm_info *sm_info = SM_I(sbi);
4818
4819 if (!sm_info)
4820 return;
4821 f2fs_destroy_flush_cmd_control(sbi, true);
4822 destroy_discard_cmd_control(sbi);
4823 destroy_dirty_segmap(sbi);
4824 destroy_curseg(sbi);
4825 destroy_free_segmap(sbi);
4826 destroy_sit_info(sbi);
4827 sbi->sm_info = NULL;
4828 kvfree(sm_info);
4829 }
4830
4831 int __init f2fs_create_segment_manager_caches(void)
4832 {
4833 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4834 sizeof(struct discard_entry));
4835 if (!discard_entry_slab)
4836 goto fail;
4837
4838 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4839 sizeof(struct discard_cmd));
4840 if (!discard_cmd_slab)
4841 goto destroy_discard_entry;
4842
4843 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4844 sizeof(struct sit_entry_set));
4845 if (!sit_entry_set_slab)
4846 goto destroy_discard_cmd;
4847
4848 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4849 sizeof(struct inmem_pages));
4850 if (!inmem_entry_slab)
4851 goto destroy_sit_entry_set;
4852 return 0;
4853
4854 destroy_sit_entry_set:
4855 kmem_cache_destroy(sit_entry_set_slab);
4856 destroy_discard_cmd:
4857 kmem_cache_destroy(discard_cmd_slab);
4858 destroy_discard_entry:
4859 kmem_cache_destroy(discard_entry_slab);
4860 fail:
4861 return -ENOMEM;
4862 }
4863
4864 void f2fs_destroy_segment_manager_caches(void)
4865 {
4866 kmem_cache_destroy(sit_entry_set_slab);
4867 kmem_cache_destroy(discard_cmd_slab);
4868 kmem_cache_destroy(discard_entry_slab);
4869 kmem_cache_destroy(inmem_entry_slab);
4870 }
Linux with added FPC-III support