]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/f2fs/segment.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
UBUNTU: Ubuntu-4.15.0-96.97
[mirror_ubuntu-bionic-kernel.git] / fs / f2fs / segment.c
1 /*
2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/bio.h>
14 #include <linux/blkdev.h>
15 #include <linux/prefetch.h>
16 #include <linux/kthread.h>
17 #include <linux/swap.h>
18 #include <linux/timer.h>
19 #include <linux/freezer.h>
20 #include <linux/sched/signal.h>
21
22 #include "f2fs.h"
23 #include "segment.h"
24 #include "node.h"
25 #include "gc.h"
26 #include "trace.h"
27 #include <trace/events/f2fs.h>
28
29 #define __reverse_ffz(x) __reverse_ffs(~(x))
30
31 static struct kmem_cache *discard_entry_slab;
32 static struct kmem_cache *discard_cmd_slab;
33 static struct kmem_cache *sit_entry_set_slab;
34 static struct kmem_cache *inmem_entry_slab;
35
36 static unsigned long __reverse_ulong(unsigned char *str)
37 {
38 unsigned long tmp = 0;
39 int shift = 24, idx = 0;
40
41 #if BITS_PER_LONG == 64
42 shift = 56;
43 #endif
44 while (shift >= 0) {
45 tmp |= (unsigned long)str[idx++] << shift;
46 shift -= BITS_PER_BYTE;
47 }
48 return tmp;
49 }
50
51 /*
52 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
53 * MSB and LSB are reversed in a byte by f2fs_set_bit.
54 */
55 static inline unsigned long __reverse_ffs(unsigned long word)
56 {
57 int num = 0;
58
59 #if BITS_PER_LONG == 64
60 if ((word & 0xffffffff00000000UL) == 0)
61 num += 32;
62 else
63 word >>= 32;
64 #endif
65 if ((word & 0xffff0000) == 0)
66 num += 16;
67 else
68 word >>= 16;
69
70 if ((word & 0xff00) == 0)
71 num += 8;
72 else
73 word >>= 8;
74
75 if ((word & 0xf0) == 0)
76 num += 4;
77 else
78 word >>= 4;
79
80 if ((word & 0xc) == 0)
81 num += 2;
82 else
83 word >>= 2;
84
85 if ((word & 0x2) == 0)
86 num += 1;
87 return num;
88 }
89
90 /*
91 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
92 * f2fs_set_bit makes MSB and LSB reversed in a byte.
93 * @size must be integral times of unsigned long.
94 * Example:
95 * MSB <--> LSB
96 * f2fs_set_bit(0, bitmap) => 1000 0000
97 * f2fs_set_bit(7, bitmap) => 0000 0001
98 */
99 static unsigned long __find_rev_next_bit(const unsigned long *addr,
100 unsigned long size, unsigned long offset)
101 {
102 const unsigned long *p = addr + BIT_WORD(offset);
103 unsigned long result = size;
104 unsigned long tmp;
105
106 if (offset >= size)
107 return size;
108
109 size -= (offset & ~(BITS_PER_LONG - 1));
110 offset %= BITS_PER_LONG;
111
112 while (1) {
113 if (*p == 0)
114 goto pass;
115
116 tmp = __reverse_ulong((unsigned char *)p);
117
118 tmp &= ~0UL >> offset;
119 if (size < BITS_PER_LONG)
120 tmp &= (~0UL << (BITS_PER_LONG - size));
121 if (tmp)
122 goto found;
123 pass:
124 if (size <= BITS_PER_LONG)
125 break;
126 size -= BITS_PER_LONG;
127 offset = 0;
128 p++;
129 }
130 return result;
131 found:
132 return result - size + __reverse_ffs(tmp);
133 }
134
135 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
136 unsigned long size, unsigned long offset)
137 {
138 const unsigned long *p = addr + BIT_WORD(offset);
139 unsigned long result = size;
140 unsigned long tmp;
141
142 if (offset >= size)
143 return size;
144
145 size -= (offset & ~(BITS_PER_LONG - 1));
146 offset %= BITS_PER_LONG;
147
148 while (1) {
149 if (*p == ~0UL)
150 goto pass;
151
152 tmp = __reverse_ulong((unsigned char *)p);
153
154 if (offset)
155 tmp |= ~0UL << (BITS_PER_LONG - offset);
156 if (size < BITS_PER_LONG)
157 tmp |= ~0UL >> size;
158 if (tmp != ~0UL)
159 goto found;
160 pass:
161 if (size <= BITS_PER_LONG)
162 break;
163 size -= BITS_PER_LONG;
164 offset = 0;
165 p++;
166 }
167 return result;
168 found:
169 return result - size + __reverse_ffz(tmp);
170 }
171
172 bool need_SSR(struct f2fs_sb_info *sbi)
173 {
174 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
175 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
176 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
177
178 if (test_opt(sbi, LFS))
179 return false;
180 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
181 return true;
182
183 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
184 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
185 }
186
187 void register_inmem_page(struct inode *inode, struct page *page)
188 {
189 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
190 struct f2fs_inode_info *fi = F2FS_I(inode);
191 struct inmem_pages *new;
192
193 f2fs_trace_pid(page);
194
195 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
196 SetPagePrivate(page);
197
198 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
199
200 /* add atomic page indices to the list */
201 new->page = page;
202 INIT_LIST_HEAD(&new->list);
203
204 /* increase reference count with clean state */
205 mutex_lock(&fi->inmem_lock);
206 get_page(page);
207 list_add_tail(&new->list, &fi->inmem_pages);
208 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
209 if (list_empty(&fi->inmem_ilist))
210 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
211 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
212 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
213 mutex_unlock(&fi->inmem_lock);
214
215 trace_f2fs_register_inmem_page(page, INMEM);
216 }
217
218 static int __revoke_inmem_pages(struct inode *inode,
219 struct list_head *head, bool drop, bool recover,
220 bool trylock)
221 {
222 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
223 struct inmem_pages *cur, *tmp;
224 int err = 0;
225
226 list_for_each_entry_safe(cur, tmp, head, list) {
227 struct page *page = cur->page;
228
229 if (drop)
230 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
231
232 if (trylock) {
233 /*
234 * to avoid deadlock in between page lock and
235 * inmem_lock.
236 */
237 if (!trylock_page(page))
238 continue;
239 } else {
240 lock_page(page);
241 }
242
243 f2fs_wait_on_page_writeback(page, DATA, true);
244
245 if (recover) {
246 struct dnode_of_data dn;
247 struct node_info ni;
248
249 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
250 retry:
251 set_new_dnode(&dn, inode, NULL, NULL, 0);
252 err = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
253 if (err) {
254 if (err == -ENOMEM) {
255 congestion_wait(BLK_RW_ASYNC, HZ/50);
256 cond_resched();
257 goto retry;
258 }
259 err = -EAGAIN;
260 goto next;
261 }
262 get_node_info(sbi, dn.nid, &ni);
263 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
264 cur->old_addr, ni.version, true, true);
265 f2fs_put_dnode(&dn);
266 }
267 next:
268 /* we don't need to invalidate this in the sccessful status */
269 if (drop || recover) {
270 ClearPageUptodate(page);
271 clear_cold_data(page);
272 }
273 set_page_private(page, 0);
274 ClearPagePrivate(page);
275 f2fs_put_page(page, 1);
276
277 list_del(&cur->list);
278 kmem_cache_free(inmem_entry_slab, cur);
279 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
280 }
281 return err;
282 }
283
284 void drop_inmem_pages_all(struct f2fs_sb_info *sbi)
285 {
286 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
287 struct inode *inode;
288 struct f2fs_inode_info *fi;
289 next:
290 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
291 if (list_empty(head)) {
292 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
293 return;
294 }
295 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
296 inode = igrab(&fi->vfs_inode);
297 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
298
299 if (inode) {
300 drop_inmem_pages(inode);
301 iput(inode);
302 }
303 congestion_wait(BLK_RW_ASYNC, HZ/50);
304 cond_resched();
305 goto next;
306 }
307
308 void drop_inmem_pages(struct inode *inode)
309 {
310 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
311 struct f2fs_inode_info *fi = F2FS_I(inode);
312
313 while (!list_empty(&fi->inmem_pages)) {
314 mutex_lock(&fi->inmem_lock);
315 __revoke_inmem_pages(inode, &fi->inmem_pages,
316 true, false, true);
317
318 if (list_empty(&fi->inmem_pages)) {
319 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
320 if (!list_empty(&fi->inmem_ilist))
321 list_del_init(&fi->inmem_ilist);
322 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
323 }
324 mutex_unlock(&fi->inmem_lock);
325 }
326
327 clear_inode_flag(inode, FI_ATOMIC_FILE);
328 clear_inode_flag(inode, FI_HOT_DATA);
329 stat_dec_atomic_write(inode);
330 }
331
332 void drop_inmem_page(struct inode *inode, struct page *page)
333 {
334 struct f2fs_inode_info *fi = F2FS_I(inode);
335 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
336 struct list_head *head = &fi->inmem_pages;
337 struct inmem_pages *cur = NULL;
338
339 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
340
341 mutex_lock(&fi->inmem_lock);
342 list_for_each_entry(cur, head, list) {
343 if (cur->page == page)
344 break;
345 }
346
347 f2fs_bug_on(sbi, !cur || cur->page != page);
348 list_del(&cur->list);
349 mutex_unlock(&fi->inmem_lock);
350
351 dec_page_count(sbi, F2FS_INMEM_PAGES);
352 kmem_cache_free(inmem_entry_slab, cur);
353
354 ClearPageUptodate(page);
355 set_page_private(page, 0);
356 ClearPagePrivate(page);
357 f2fs_put_page(page, 0);
358
359 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
360 }
361
362 static int __commit_inmem_pages(struct inode *inode,
363 struct list_head *revoke_list)
364 {
365 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
366 struct f2fs_inode_info *fi = F2FS_I(inode);
367 struct inmem_pages *cur, *tmp;
368 struct f2fs_io_info fio = {
369 .sbi = sbi,
370 .ino = inode->i_ino,
371 .type = DATA,
372 .op = REQ_OP_WRITE,
373 .op_flags = REQ_SYNC | REQ_PRIO,
374 .io_type = FS_DATA_IO,
375 };
376 pgoff_t last_idx = ULONG_MAX;
377 int err = 0;
378
379 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
380 struct page *page = cur->page;
381
382 lock_page(page);
383 if (page->mapping == inode->i_mapping) {
384 trace_f2fs_commit_inmem_page(page, INMEM);
385
386 set_page_dirty(page);
387 f2fs_wait_on_page_writeback(page, DATA, true);
388 if (clear_page_dirty_for_io(page)) {
389 inode_dec_dirty_pages(inode);
390 remove_dirty_inode(inode);
391 }
392 retry:
393 fio.page = page;
394 fio.old_blkaddr = NULL_ADDR;
395 fio.encrypted_page = NULL;
396 fio.need_lock = LOCK_DONE;
397 err = do_write_data_page(&fio);
398 if (err) {
399 if (err == -ENOMEM) {
400 congestion_wait(BLK_RW_ASYNC, HZ/50);
401 cond_resched();
402 goto retry;
403 }
404 unlock_page(page);
405 break;
406 }
407 /* record old blkaddr for revoking */
408 cur->old_addr = fio.old_blkaddr;
409 last_idx = page->index;
410 }
411 unlock_page(page);
412 list_move_tail(&cur->list, revoke_list);
413 }
414
415 if (last_idx != ULONG_MAX)
416 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
417
418 if (!err)
419 __revoke_inmem_pages(inode, revoke_list, false, false, false);
420
421 return err;
422 }
423
424 int commit_inmem_pages(struct inode *inode)
425 {
426 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
427 struct f2fs_inode_info *fi = F2FS_I(inode);
428 struct list_head revoke_list;
429 int err;
430
431 INIT_LIST_HEAD(&revoke_list);
432 f2fs_balance_fs(sbi, true);
433 f2fs_lock_op(sbi);
434
435 set_inode_flag(inode, FI_ATOMIC_COMMIT);
436
437 mutex_lock(&fi->inmem_lock);
438 err = __commit_inmem_pages(inode, &revoke_list);
439 if (err) {
440 int ret;
441 /*
442 * try to revoke all committed pages, but still we could fail
443 * due to no memory or other reason, if that happened, EAGAIN
444 * will be returned, which means in such case, transaction is
445 * already not integrity, caller should use journal to do the
446 * recovery or rewrite & commit last transaction. For other
447 * error number, revoking was done by filesystem itself.
448 */
449 ret = __revoke_inmem_pages(inode, &revoke_list,
450 false, true, false);
451 if (ret)
452 err = ret;
453
454 /* drop all uncommitted pages */
455 __revoke_inmem_pages(inode, &fi->inmem_pages,
456 true, false, false);
457 }
458 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
459 if (!list_empty(&fi->inmem_ilist))
460 list_del_init(&fi->inmem_ilist);
461 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
462 mutex_unlock(&fi->inmem_lock);
463
464 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
465
466 f2fs_unlock_op(sbi);
467 return err;
468 }
469
470 /*
471 * This function balances dirty node and dentry pages.
472 * In addition, it controls garbage collection.
473 */
474 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
475 {
476 #ifdef CONFIG_F2FS_FAULT_INJECTION
477 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
478 f2fs_show_injection_info(FAULT_CHECKPOINT);
479 f2fs_stop_checkpoint(sbi, false);
480 }
481 #endif
482
483 /* balance_fs_bg is able to be pending */
484 if (need && excess_cached_nats(sbi))
485 f2fs_balance_fs_bg(sbi);
486
487 /*
488 * We should do GC or end up with checkpoint, if there are so many dirty
489 * dir/node pages without enough free segments.
490 */
491 if (has_not_enough_free_secs(sbi, 0, 0)) {
492 mutex_lock(&sbi->gc_mutex);
493 f2fs_gc(sbi, false, false, NULL_SEGNO);
494 }
495 }
496
497 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
498 {
499 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
500 return;
501
502 /* try to shrink extent cache when there is no enough memory */
503 if (!available_free_memory(sbi, EXTENT_CACHE))
504 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
505
506 /* check the # of cached NAT entries */
507 if (!available_free_memory(sbi, NAT_ENTRIES))
508 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
509
510 if (!available_free_memory(sbi, FREE_NIDS))
511 try_to_free_nids(sbi, MAX_FREE_NIDS);
512 else
513 build_free_nids(sbi, false, false);
514
515 if (!is_idle(sbi) && !excess_dirty_nats(sbi))
516 return;
517
518 /* checkpoint is the only way to shrink partial cached entries */
519 if (!available_free_memory(sbi, NAT_ENTRIES) ||
520 !available_free_memory(sbi, INO_ENTRIES) ||
521 excess_prefree_segs(sbi) ||
522 excess_dirty_nats(sbi) ||
523 f2fs_time_over(sbi, CP_TIME)) {
524 if (test_opt(sbi, DATA_FLUSH)) {
525 struct blk_plug plug;
526
527 blk_start_plug(&plug);
528 sync_dirty_inodes(sbi, FILE_INODE);
529 blk_finish_plug(&plug);
530 }
531 f2fs_sync_fs(sbi->sb, true);
532 stat_inc_bg_cp_count(sbi->stat_info);
533 }
534 }
535
536 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
537 struct block_device *bdev)
538 {
539 struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
540 int ret;
541
542 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
543 bio_set_dev(bio, bdev);
544 ret = submit_bio_wait(bio);
545 bio_put(bio);
546
547 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
548 test_opt(sbi, FLUSH_MERGE), ret);
549 return ret;
550 }
551
552 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
553 {
554 int ret = 0;
555 int i;
556
557 if (!f2fs_is_multi_device(sbi))
558 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
559
560 for (i = 0; i < sbi->s_ndevs; i++) {
561 if (!is_dirty_device(sbi, ino, i, FLUSH_INO))
562 continue;
563 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
564 if (ret)
565 break;
566 }
567 return ret;
568 }
569
570 static int issue_flush_thread(void *data)
571 {
572 struct f2fs_sb_info *sbi = data;
573 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
574 wait_queue_head_t *q = &fcc->flush_wait_queue;
575 repeat:
576 if (kthread_should_stop())
577 return 0;
578
579 sb_start_intwrite(sbi->sb);
580
581 if (!llist_empty(&fcc->issue_list)) {
582 struct flush_cmd *cmd, *next;
583 int ret;
584
585 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
586 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
587
588 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
589
590 ret = submit_flush_wait(sbi, cmd->ino);
591 atomic_inc(&fcc->issued_flush);
592
593 llist_for_each_entry_safe(cmd, next,
594 fcc->dispatch_list, llnode) {
595 cmd->ret = ret;
596 complete(&cmd->wait);
597 }
598 fcc->dispatch_list = NULL;
599 }
600
601 sb_end_intwrite(sbi->sb);
602
603 wait_event_interruptible(*q,
604 kthread_should_stop() || !llist_empty(&fcc->issue_list));
605 goto repeat;
606 }
607
608 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
609 {
610 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
611 struct flush_cmd cmd;
612 int ret;
613
614 if (test_opt(sbi, NOBARRIER))
615 return 0;
616
617 if (!test_opt(sbi, FLUSH_MERGE)) {
618 atomic_inc(&fcc->issing_flush);
619 ret = submit_flush_wait(sbi, ino);
620 atomic_dec(&fcc->issing_flush);
621 atomic_inc(&fcc->issued_flush);
622 return ret;
623 }
624
625 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
626 ret = submit_flush_wait(sbi, ino);
627 atomic_dec(&fcc->issing_flush);
628
629 atomic_inc(&fcc->issued_flush);
630 return ret;
631 }
632
633 cmd.ino = ino;
634 init_completion(&cmd.wait);
635
636 llist_add(&cmd.llnode, &fcc->issue_list);
637
638 /* update issue_list before we wake up issue_flush thread */
639 smp_mb();
640
641 if (waitqueue_active(&fcc->flush_wait_queue))
642 wake_up(&fcc->flush_wait_queue);
643
644 if (fcc->f2fs_issue_flush) {
645 wait_for_completion(&cmd.wait);
646 atomic_dec(&fcc->issing_flush);
647 } else {
648 struct llist_node *list;
649
650 list = llist_del_all(&fcc->issue_list);
651 if (!list) {
652 wait_for_completion(&cmd.wait);
653 atomic_dec(&fcc->issing_flush);
654 } else {
655 struct flush_cmd *tmp, *next;
656
657 ret = submit_flush_wait(sbi, ino);
658
659 llist_for_each_entry_safe(tmp, next, list, llnode) {
660 if (tmp == &cmd) {
661 cmd.ret = ret;
662 atomic_dec(&fcc->issing_flush);
663 continue;
664 }
665 tmp->ret = ret;
666 complete(&tmp->wait);
667 }
668 }
669 }
670
671 return cmd.ret;
672 }
673
674 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
675 {
676 dev_t dev = sbi->sb->s_bdev->bd_dev;
677 struct flush_cmd_control *fcc;
678 int err = 0;
679
680 if (SM_I(sbi)->fcc_info) {
681 fcc = SM_I(sbi)->fcc_info;
682 if (fcc->f2fs_issue_flush)
683 return err;
684 goto init_thread;
685 }
686
687 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
688 if (!fcc)
689 return -ENOMEM;
690 atomic_set(&fcc->issued_flush, 0);
691 atomic_set(&fcc->issing_flush, 0);
692 init_waitqueue_head(&fcc->flush_wait_queue);
693 init_llist_head(&fcc->issue_list);
694 SM_I(sbi)->fcc_info = fcc;
695 if (!test_opt(sbi, FLUSH_MERGE))
696 return err;
697
698 init_thread:
699 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
700 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
701 if (IS_ERR(fcc->f2fs_issue_flush)) {
702 err = PTR_ERR(fcc->f2fs_issue_flush);
703 kfree(fcc);
704 SM_I(sbi)->fcc_info = NULL;
705 return err;
706 }
707
708 return err;
709 }
710
711 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
712 {
713 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
714
715 if (fcc && fcc->f2fs_issue_flush) {
716 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
717
718 fcc->f2fs_issue_flush = NULL;
719 kthread_stop(flush_thread);
720 }
721 if (free) {
722 kfree(fcc);
723 SM_I(sbi)->fcc_info = NULL;
724 }
725 }
726
727 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
728 {
729 int ret = 0, i;
730
731 if (!sbi->s_ndevs)
732 return 0;
733
734 for (i = 1; i < sbi->s_ndevs; i++) {
735 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
736 continue;
737 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
738 if (ret)
739 break;
740
741 spin_lock(&sbi->dev_lock);
742 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
743 spin_unlock(&sbi->dev_lock);
744 }
745
746 return ret;
747 }
748
749 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
750 enum dirty_type dirty_type)
751 {
752 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
753
754 /* need not be added */
755 if (IS_CURSEG(sbi, segno))
756 return;
757
758 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
759 dirty_i->nr_dirty[dirty_type]++;
760
761 if (dirty_type == DIRTY) {
762 struct seg_entry *sentry = get_seg_entry(sbi, segno);
763 enum dirty_type t = sentry->type;
764
765 if (unlikely(t >= DIRTY)) {
766 f2fs_bug_on(sbi, 1);
767 return;
768 }
769 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
770 dirty_i->nr_dirty[t]++;
771 }
772 }
773
774 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
775 enum dirty_type dirty_type)
776 {
777 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
778
779 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
780 dirty_i->nr_dirty[dirty_type]--;
781
782 if (dirty_type == DIRTY) {
783 struct seg_entry *sentry = get_seg_entry(sbi, segno);
784 enum dirty_type t = sentry->type;
785
786 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
787 dirty_i->nr_dirty[t]--;
788
789 if (get_valid_blocks(sbi, segno, true) == 0)
790 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
791 dirty_i->victim_secmap);
792 }
793 }
794
795 /*
796 * Should not occur error such as -ENOMEM.
797 * Adding dirty entry into seglist is not critical operation.
798 * If a given segment is one of current working segments, it won't be added.
799 */
800 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
801 {
802 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
803 unsigned short valid_blocks;
804
805 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
806 return;
807
808 mutex_lock(&dirty_i->seglist_lock);
809
810 valid_blocks = get_valid_blocks(sbi, segno, false);
811
812 if (valid_blocks == 0) {
813 __locate_dirty_segment(sbi, segno, PRE);
814 __remove_dirty_segment(sbi, segno, DIRTY);
815 } else if (valid_blocks < sbi->blocks_per_seg) {
816 __locate_dirty_segment(sbi, segno, DIRTY);
817 } else {
818 /* Recovery routine with SSR needs this */
819 __remove_dirty_segment(sbi, segno, DIRTY);
820 }
821
822 mutex_unlock(&dirty_i->seglist_lock);
823 }
824
825 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
826 struct block_device *bdev, block_t lstart,
827 block_t start, block_t len)
828 {
829 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
830 struct list_head *pend_list;
831 struct discard_cmd *dc;
832
833 f2fs_bug_on(sbi, !len);
834
835 pend_list = &dcc->pend_list[plist_idx(len)];
836
837 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
838 INIT_LIST_HEAD(&dc->list);
839 dc->bdev = bdev;
840 dc->lstart = lstart;
841 dc->start = start;
842 dc->len = len;
843 dc->ref = 0;
844 dc->state = D_PREP;
845 dc->error = 0;
846 init_completion(&dc->wait);
847 list_add_tail(&dc->list, pend_list);
848 atomic_inc(&dcc->discard_cmd_cnt);
849 dcc->undiscard_blks += len;
850
851 return dc;
852 }
853
854 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
855 struct block_device *bdev, block_t lstart,
856 block_t start, block_t len,
857 struct rb_node *parent, struct rb_node **p)
858 {
859 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
860 struct discard_cmd *dc;
861
862 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
863
864 rb_link_node(&dc->rb_node, parent, p);
865 rb_insert_color(&dc->rb_node, &dcc->root);
866
867 return dc;
868 }
869
870 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
871 struct discard_cmd *dc)
872 {
873 if (dc->state == D_DONE)
874 atomic_dec(&dcc->issing_discard);
875
876 list_del(&dc->list);
877 rb_erase(&dc->rb_node, &dcc->root);
878 dcc->undiscard_blks -= dc->len;
879
880 kmem_cache_free(discard_cmd_slab, dc);
881
882 atomic_dec(&dcc->discard_cmd_cnt);
883 }
884
885 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
886 struct discard_cmd *dc)
887 {
888 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
889
890 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
891
892 f2fs_bug_on(sbi, dc->ref);
893
894 if (dc->error == -EOPNOTSUPP)
895 dc->error = 0;
896
897 if (dc->error)
898 f2fs_msg(sbi->sb, KERN_INFO,
899 "Issue discard(%u, %u, %u) failed, ret: %d",
900 dc->lstart, dc->start, dc->len, dc->error);
901 __detach_discard_cmd(dcc, dc);
902 }
903
904 static void f2fs_submit_discard_endio(struct bio *bio)
905 {
906 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
907
908 dc->error = blk_status_to_errno(bio->bi_status);
909 dc->state = D_DONE;
910 complete_all(&dc->wait);
911 bio_put(bio);
912 }
913
914 void __check_sit_bitmap(struct f2fs_sb_info *sbi,
915 block_t start, block_t end)
916 {
917 #ifdef CONFIG_F2FS_CHECK_FS
918 struct seg_entry *sentry;
919 unsigned int segno;
920 block_t blk = start;
921 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
922 unsigned long *map;
923
924 while (blk < end) {
925 segno = GET_SEGNO(sbi, blk);
926 sentry = get_seg_entry(sbi, segno);
927 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
928
929 if (end < START_BLOCK(sbi, segno + 1))
930 size = GET_BLKOFF_FROM_SEG0(sbi, end);
931 else
932 size = max_blocks;
933 map = (unsigned long *)(sentry->cur_valid_map);
934 offset = __find_rev_next_bit(map, size, offset);
935 f2fs_bug_on(sbi, offset != size);
936 blk = START_BLOCK(sbi, segno + 1);
937 }
938 #endif
939 }
940
941 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
942 static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
943 struct discard_policy *dpolicy,
944 struct discard_cmd *dc)
945 {
946 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
947 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
948 &(dcc->fstrim_list) : &(dcc->wait_list);
949 struct bio *bio = NULL;
950 int flag = dpolicy->sync ? REQ_SYNC : 0;
951
952 if (dc->state != D_PREP)
953 return;
954
955 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
956
957 dc->error = __blkdev_issue_discard(dc->bdev,
958 SECTOR_FROM_BLOCK(dc->start),
959 SECTOR_FROM_BLOCK(dc->len),
960 GFP_NOFS, 0, &bio);
961 if (!dc->error) {
962 /* should keep before submission to avoid D_DONE right away */
963 dc->state = D_SUBMIT;
964 atomic_inc(&dcc->issued_discard);
965 atomic_inc(&dcc->issing_discard);
966 if (bio) {
967 bio->bi_private = dc;
968 bio->bi_end_io = f2fs_submit_discard_endio;
969 bio->bi_opf |= flag;
970 submit_bio(bio);
971 list_move_tail(&dc->list, wait_list);
972 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
973
974 f2fs_update_iostat(sbi, FS_DISCARD, 1);
975 }
976 } else {
977 __remove_discard_cmd(sbi, dc);
978 }
979 }
980
981 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
982 struct block_device *bdev, block_t lstart,
983 block_t start, block_t len,
984 struct rb_node **insert_p,
985 struct rb_node *insert_parent)
986 {
987 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
988 struct rb_node **p;
989 struct rb_node *parent = NULL;
990 struct discard_cmd *dc = NULL;
991
992 if (insert_p && insert_parent) {
993 parent = insert_parent;
994 p = insert_p;
995 goto do_insert;
996 }
997
998 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
999 do_insert:
1000 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1001 if (!dc)
1002 return NULL;
1003
1004 return dc;
1005 }
1006
1007 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1008 struct discard_cmd *dc)
1009 {
1010 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1011 }
1012
1013 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1014 struct discard_cmd *dc, block_t blkaddr)
1015 {
1016 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1017 struct discard_info di = dc->di;
1018 bool modified = false;
1019
1020 if (dc->state == D_DONE || dc->len == 1) {
1021 __remove_discard_cmd(sbi, dc);
1022 return;
1023 }
1024
1025 dcc->undiscard_blks -= di.len;
1026
1027 if (blkaddr > di.lstart) {
1028 dc->len = blkaddr - dc->lstart;
1029 dcc->undiscard_blks += dc->len;
1030 __relocate_discard_cmd(dcc, dc);
1031 modified = true;
1032 }
1033
1034 if (blkaddr < di.lstart + di.len - 1) {
1035 if (modified) {
1036 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1037 di.start + blkaddr + 1 - di.lstart,
1038 di.lstart + di.len - 1 - blkaddr,
1039 NULL, NULL);
1040 } else {
1041 dc->lstart++;
1042 dc->len--;
1043 dc->start++;
1044 dcc->undiscard_blks += dc->len;
1045 __relocate_discard_cmd(dcc, dc);
1046 }
1047 }
1048 }
1049
1050 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1051 struct block_device *bdev, block_t lstart,
1052 block_t start, block_t len)
1053 {
1054 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1055 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1056 struct discard_cmd *dc;
1057 struct discard_info di = {0};
1058 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1059 block_t end = lstart + len;
1060
1061 mutex_lock(&dcc->cmd_lock);
1062
1063 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1064 NULL, lstart,
1065 (struct rb_entry **)&prev_dc,
1066 (struct rb_entry **)&next_dc,
1067 &insert_p, &insert_parent, true);
1068 if (dc)
1069 prev_dc = dc;
1070
1071 if (!prev_dc) {
1072 di.lstart = lstart;
1073 di.len = next_dc ? next_dc->lstart - lstart : len;
1074 di.len = min(di.len, len);
1075 di.start = start;
1076 }
1077
1078 while (1) {
1079 struct rb_node *node;
1080 bool merged = false;
1081 struct discard_cmd *tdc = NULL;
1082
1083 if (prev_dc) {
1084 di.lstart = prev_dc->lstart + prev_dc->len;
1085 if (di.lstart < lstart)
1086 di.lstart = lstart;
1087 if (di.lstart >= end)
1088 break;
1089
1090 if (!next_dc || next_dc->lstart > end)
1091 di.len = end - di.lstart;
1092 else
1093 di.len = next_dc->lstart - di.lstart;
1094 di.start = start + di.lstart - lstart;
1095 }
1096
1097 if (!di.len)
1098 goto next;
1099
1100 if (prev_dc && prev_dc->state == D_PREP &&
1101 prev_dc->bdev == bdev &&
1102 __is_discard_back_mergeable(&di, &prev_dc->di)) {
1103 prev_dc->di.len += di.len;
1104 dcc->undiscard_blks += di.len;
1105 __relocate_discard_cmd(dcc, prev_dc);
1106 di = prev_dc->di;
1107 tdc = prev_dc;
1108 merged = true;
1109 }
1110
1111 if (next_dc && next_dc->state == D_PREP &&
1112 next_dc->bdev == bdev &&
1113 __is_discard_front_mergeable(&di, &next_dc->di)) {
1114 next_dc->di.lstart = di.lstart;
1115 next_dc->di.len += di.len;
1116 next_dc->di.start = di.start;
1117 dcc->undiscard_blks += di.len;
1118 __relocate_discard_cmd(dcc, next_dc);
1119 if (tdc)
1120 __remove_discard_cmd(sbi, tdc);
1121 merged = true;
1122 }
1123
1124 if (!merged) {
1125 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1126 di.len, NULL, NULL);
1127 }
1128 next:
1129 prev_dc = next_dc;
1130 if (!prev_dc)
1131 break;
1132
1133 node = rb_next(&prev_dc->rb_node);
1134 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1135 }
1136
1137 mutex_unlock(&dcc->cmd_lock);
1138 }
1139
1140 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1141 struct block_device *bdev, block_t blkstart, block_t blklen)
1142 {
1143 block_t lblkstart = blkstart;
1144
1145 trace_f2fs_queue_discard(bdev, blkstart, blklen);
1146
1147 if (f2fs_is_multi_device(sbi)) {
1148 int devi = f2fs_target_device_index(sbi, blkstart);
1149
1150 blkstart -= FDEV(devi).start_blk;
1151 }
1152 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1153 return 0;
1154 }
1155
1156 static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
1157 struct discard_policy *dpolicy,
1158 unsigned int start, unsigned int end)
1159 {
1160 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1161 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1162 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1163 struct discard_cmd *dc;
1164 struct blk_plug plug;
1165 int issued;
1166
1167 next:
1168 issued = 0;
1169
1170 mutex_lock(&dcc->cmd_lock);
1171 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1172
1173 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1174 NULL, start,
1175 (struct rb_entry **)&prev_dc,
1176 (struct rb_entry **)&next_dc,
1177 &insert_p, &insert_parent, true);
1178 if (!dc)
1179 dc = next_dc;
1180
1181 blk_start_plug(&plug);
1182
1183 while (dc && dc->lstart <= end) {
1184 struct rb_node *node;
1185
1186 if (dc->len < dpolicy->granularity)
1187 goto skip;
1188
1189 if (dc->state != D_PREP) {
1190 list_move_tail(&dc->list, &dcc->fstrim_list);
1191 goto skip;
1192 }
1193
1194 __submit_discard_cmd(sbi, dpolicy, dc);
1195
1196 if (++issued >= dpolicy->max_requests) {
1197 start = dc->lstart + dc->len;
1198
1199 blk_finish_plug(&plug);
1200 mutex_unlock(&dcc->cmd_lock);
1201
1202 schedule();
1203
1204 goto next;
1205 }
1206 skip:
1207 node = rb_next(&dc->rb_node);
1208 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1209
1210 if (fatal_signal_pending(current))
1211 break;
1212 }
1213
1214 blk_finish_plug(&plug);
1215 mutex_unlock(&dcc->cmd_lock);
1216 }
1217
1218 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1219 struct discard_policy *dpolicy)
1220 {
1221 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1222 struct list_head *pend_list;
1223 struct discard_cmd *dc, *tmp;
1224 struct blk_plug plug;
1225 int i, iter = 0, issued = 0;
1226 bool io_interrupted = false;
1227
1228 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1229 if (i + 1 < dpolicy->granularity)
1230 break;
1231 pend_list = &dcc->pend_list[i];
1232
1233 mutex_lock(&dcc->cmd_lock);
1234 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1235 blk_start_plug(&plug);
1236 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1237 f2fs_bug_on(sbi, dc->state != D_PREP);
1238
1239 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1240 !is_idle(sbi)) {
1241 io_interrupted = true;
1242 goto skip;
1243 }
1244
1245 __submit_discard_cmd(sbi, dpolicy, dc);
1246 issued++;
1247 skip:
1248 if (++iter >= dpolicy->max_requests)
1249 break;
1250 }
1251 blk_finish_plug(&plug);
1252 mutex_unlock(&dcc->cmd_lock);
1253
1254 if (iter >= dpolicy->max_requests)
1255 break;
1256 }
1257
1258 if (!issued && io_interrupted)
1259 issued = -1;
1260
1261 return issued;
1262 }
1263
1264 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1265 {
1266 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1267 struct list_head *pend_list;
1268 struct discard_cmd *dc, *tmp;
1269 int i;
1270 bool dropped = false;
1271
1272 mutex_lock(&dcc->cmd_lock);
1273 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1274 pend_list = &dcc->pend_list[i];
1275 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1276 f2fs_bug_on(sbi, dc->state != D_PREP);
1277 __remove_discard_cmd(sbi, dc);
1278 dropped = true;
1279 }
1280 }
1281 mutex_unlock(&dcc->cmd_lock);
1282
1283 return dropped;
1284 }
1285
1286 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1287 struct discard_cmd *dc)
1288 {
1289 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1290 unsigned int len = 0;
1291
1292 wait_for_completion_io(&dc->wait);
1293 mutex_lock(&dcc->cmd_lock);
1294 f2fs_bug_on(sbi, dc->state != D_DONE);
1295 dc->ref--;
1296 if (!dc->ref) {
1297 if (!dc->error)
1298 len = dc->len;
1299 __remove_discard_cmd(sbi, dc);
1300 }
1301 mutex_unlock(&dcc->cmd_lock);
1302
1303 return len;
1304 }
1305
1306 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1307 struct discard_policy *dpolicy,
1308 block_t start, block_t end)
1309 {
1310 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1311 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1312 &(dcc->fstrim_list) : &(dcc->wait_list);
1313 struct discard_cmd *dc, *tmp;
1314 bool need_wait;
1315 unsigned int trimmed = 0;
1316
1317 next:
1318 need_wait = false;
1319
1320 mutex_lock(&dcc->cmd_lock);
1321 list_for_each_entry_safe(dc, tmp, wait_list, list) {
1322 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1323 continue;
1324 if (dc->len < dpolicy->granularity)
1325 continue;
1326 if (dc->state == D_DONE && !dc->ref) {
1327 wait_for_completion_io(&dc->wait);
1328 if (!dc->error)
1329 trimmed += dc->len;
1330 __remove_discard_cmd(sbi, dc);
1331 } else {
1332 dc->ref++;
1333 need_wait = true;
1334 break;
1335 }
1336 }
1337 mutex_unlock(&dcc->cmd_lock);
1338
1339 if (need_wait) {
1340 trimmed += __wait_one_discard_bio(sbi, dc);
1341 goto next;
1342 }
1343
1344 return trimmed;
1345 }
1346
1347 static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1348 struct discard_policy *dpolicy)
1349 {
1350 __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1351 }
1352
1353 /* This should be covered by global mutex, &sit_i->sentry_lock */
1354 void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1355 {
1356 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1357 struct discard_cmd *dc;
1358 bool need_wait = false;
1359
1360 mutex_lock(&dcc->cmd_lock);
1361 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1362 if (dc) {
1363 if (dc->state == D_PREP) {
1364 __punch_discard_cmd(sbi, dc, blkaddr);
1365 } else {
1366 dc->ref++;
1367 need_wait = true;
1368 }
1369 }
1370 mutex_unlock(&dcc->cmd_lock);
1371
1372 if (need_wait)
1373 __wait_one_discard_bio(sbi, dc);
1374 }
1375
1376 void stop_discard_thread(struct f2fs_sb_info *sbi)
1377 {
1378 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1379
1380 if (dcc && dcc->f2fs_issue_discard) {
1381 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1382
1383 dcc->f2fs_issue_discard = NULL;
1384 kthread_stop(discard_thread);
1385 }
1386 }
1387
1388 /* This comes from f2fs_put_super */
1389 bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1390 {
1391 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1392 struct discard_policy dpolicy;
1393 bool dropped;
1394
1395 init_discard_policy(&dpolicy, DPOLICY_UMOUNT, dcc->discard_granularity);
1396 __issue_discard_cmd(sbi, &dpolicy);
1397 dropped = __drop_discard_cmd(sbi);
1398 __wait_all_discard_cmd(sbi, &dpolicy);
1399
1400 return dropped;
1401 }
1402
1403 static int issue_discard_thread(void *data)
1404 {
1405 struct f2fs_sb_info *sbi = data;
1406 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1407 wait_queue_head_t *q = &dcc->discard_wait_queue;
1408 struct discard_policy dpolicy;
1409 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1410 int issued;
1411
1412 set_freezable();
1413
1414 do {
1415 init_discard_policy(&dpolicy, DPOLICY_BG,
1416 dcc->discard_granularity);
1417
1418 wait_event_interruptible_timeout(*q,
1419 kthread_should_stop() || freezing(current) ||
1420 dcc->discard_wake,
1421 msecs_to_jiffies(wait_ms));
1422 if (try_to_freeze())
1423 continue;
1424 if (kthread_should_stop())
1425 return 0;
1426
1427 if (dcc->discard_wake) {
1428 dcc->discard_wake = 0;
1429 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
1430 init_discard_policy(&dpolicy,
1431 DPOLICY_FORCE, 1);
1432 }
1433
1434 sb_start_intwrite(sbi->sb);
1435
1436 issued = __issue_discard_cmd(sbi, &dpolicy);
1437 if (issued) {
1438 __wait_all_discard_cmd(sbi, &dpolicy);
1439 wait_ms = dpolicy.min_interval;
1440 } else {
1441 wait_ms = dpolicy.max_interval;
1442 }
1443
1444 sb_end_intwrite(sbi->sb);
1445
1446 } while (!kthread_should_stop());
1447 return 0;
1448 }
1449
1450 #ifdef CONFIG_BLK_DEV_ZONED
1451 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1452 struct block_device *bdev, block_t blkstart, block_t blklen)
1453 {
1454 sector_t sector, nr_sects;
1455 block_t lblkstart = blkstart;
1456 int devi = 0;
1457
1458 if (f2fs_is_multi_device(sbi)) {
1459 devi = f2fs_target_device_index(sbi, blkstart);
1460 blkstart -= FDEV(devi).start_blk;
1461 }
1462
1463 /*
1464 * We need to know the type of the zone: for conventional zones,
1465 * use regular discard if the drive supports it. For sequential
1466 * zones, reset the zone write pointer.
1467 */
1468 switch (get_blkz_type(sbi, bdev, blkstart)) {
1469
1470 case BLK_ZONE_TYPE_CONVENTIONAL:
1471 if (!blk_queue_discard(bdev_get_queue(bdev)))
1472 return 0;
1473 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1474 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1475 case BLK_ZONE_TYPE_SEQWRITE_PREF:
1476 sector = SECTOR_FROM_BLOCK(blkstart);
1477 nr_sects = SECTOR_FROM_BLOCK(blklen);
1478
1479 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1480 nr_sects != bdev_zone_sectors(bdev)) {
1481 f2fs_msg(sbi->sb, KERN_INFO,
1482 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1483 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1484 blkstart, blklen);
1485 return -EIO;
1486 }
1487 trace_f2fs_issue_reset_zone(bdev, blkstart);
1488 return blkdev_reset_zones(bdev, sector,
1489 nr_sects, GFP_NOFS);
1490 default:
1491 /* Unknown zone type: broken device ? */
1492 return -EIO;
1493 }
1494 }
1495 #endif
1496
1497 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1498 struct block_device *bdev, block_t blkstart, block_t blklen)
1499 {
1500 #ifdef CONFIG_BLK_DEV_ZONED
1501 if (f2fs_sb_mounted_blkzoned(sbi->sb) &&
1502 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1503 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1504 #endif
1505 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1506 }
1507
1508 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1509 block_t blkstart, block_t blklen)
1510 {
1511 sector_t start = blkstart, len = 0;
1512 struct block_device *bdev;
1513 struct seg_entry *se;
1514 unsigned int offset;
1515 block_t i;
1516 int err = 0;
1517
1518 bdev = f2fs_target_device(sbi, blkstart, NULL);
1519
1520 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1521 if (i != start) {
1522 struct block_device *bdev2 =
1523 f2fs_target_device(sbi, i, NULL);
1524
1525 if (bdev2 != bdev) {
1526 err = __issue_discard_async(sbi, bdev,
1527 start, len);
1528 if (err)
1529 return err;
1530 bdev = bdev2;
1531 start = i;
1532 len = 0;
1533 }
1534 }
1535
1536 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1537 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1538
1539 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1540 sbi->discard_blks--;
1541 }
1542
1543 if (len)
1544 err = __issue_discard_async(sbi, bdev, start, len);
1545 return err;
1546 }
1547
1548 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1549 bool check_only)
1550 {
1551 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1552 int max_blocks = sbi->blocks_per_seg;
1553 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1554 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1555 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1556 unsigned long *discard_map = (unsigned long *)se->discard_map;
1557 unsigned long *dmap = SIT_I(sbi)->tmp_map;
1558 unsigned int start = 0, end = -1;
1559 bool force = (cpc->reason & CP_DISCARD);
1560 struct discard_entry *de = NULL;
1561 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1562 int i;
1563
1564 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1565 return false;
1566
1567 if (!force) {
1568 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1569 SM_I(sbi)->dcc_info->nr_discards >=
1570 SM_I(sbi)->dcc_info->max_discards)
1571 return false;
1572 }
1573
1574 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1575 for (i = 0; i < entries; i++)
1576 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1577 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1578
1579 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1580 SM_I(sbi)->dcc_info->max_discards) {
1581 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1582 if (start >= max_blocks)
1583 break;
1584
1585 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1586 if (force && start && end != max_blocks
1587 && (end - start) < cpc->trim_minlen)
1588 continue;
1589
1590 if (check_only)
1591 return true;
1592
1593 if (!de) {
1594 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1595 GFP_F2FS_ZERO);
1596 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1597 list_add_tail(&de->list, head);
1598 }
1599
1600 for (i = start; i < end; i++)
1601 __set_bit_le(i, (void *)de->discard_map);
1602
1603 SM_I(sbi)->dcc_info->nr_discards += end - start;
1604 }
1605 return false;
1606 }
1607
1608 void release_discard_addrs(struct f2fs_sb_info *sbi)
1609 {
1610 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1611 struct discard_entry *entry, *this;
1612
1613 /* drop caches */
1614 list_for_each_entry_safe(entry, this, head, list) {
1615 list_del(&entry->list);
1616 kmem_cache_free(discard_entry_slab, entry);
1617 }
1618 }
1619
1620 /*
1621 * Should call clear_prefree_segments after checkpoint is done.
1622 */
1623 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1624 {
1625 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1626 unsigned int segno;
1627
1628 mutex_lock(&dirty_i->seglist_lock);
1629 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1630 __set_test_and_free(sbi, segno);
1631 mutex_unlock(&dirty_i->seglist_lock);
1632 }
1633
1634 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1635 {
1636 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1637 struct list_head *head = &dcc->entry_list;
1638 struct discard_entry *entry, *this;
1639 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1640 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1641 unsigned int start = 0, end = -1;
1642 unsigned int secno, start_segno;
1643 bool force = (cpc->reason & CP_DISCARD);
1644 bool need_align = test_opt(sbi, LFS) && sbi->segs_per_sec > 1;
1645
1646 mutex_lock(&dirty_i->seglist_lock);
1647
1648 while (1) {
1649 int i;
1650
1651 if (need_align && end != -1)
1652 end--;
1653 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1654 if (start >= MAIN_SEGS(sbi))
1655 break;
1656 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1657 start + 1);
1658
1659 if (need_align) {
1660 start = rounddown(start, sbi->segs_per_sec);
1661 end = roundup(end, sbi->segs_per_sec);
1662 }
1663
1664 for (i = start; i < end; i++) {
1665 if (test_and_clear_bit(i, prefree_map))
1666 dirty_i->nr_dirty[PRE]--;
1667 }
1668
1669 if (!test_opt(sbi, DISCARD))
1670 continue;
1671
1672 if (force && start >= cpc->trim_start &&
1673 (end - 1) <= cpc->trim_end)
1674 continue;
1675
1676 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1677 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1678 (end - start) << sbi->log_blocks_per_seg);
1679 continue;
1680 }
1681 next:
1682 secno = GET_SEC_FROM_SEG(sbi, start);
1683 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1684 if (!IS_CURSEC(sbi, secno) &&
1685 !get_valid_blocks(sbi, start, true))
1686 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1687 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1688
1689 start = start_segno + sbi->segs_per_sec;
1690 if (start < end)
1691 goto next;
1692 else
1693 end = start - 1;
1694 }
1695 mutex_unlock(&dirty_i->seglist_lock);
1696
1697 /* send small discards */
1698 list_for_each_entry_safe(entry, this, head, list) {
1699 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1700 bool is_valid = test_bit_le(0, entry->discard_map);
1701
1702 find_next:
1703 if (is_valid) {
1704 next_pos = find_next_zero_bit_le(entry->discard_map,
1705 sbi->blocks_per_seg, cur_pos);
1706 len = next_pos - cur_pos;
1707
1708 if (f2fs_sb_mounted_blkzoned(sbi->sb) ||
1709 (force && len < cpc->trim_minlen))
1710 goto skip;
1711
1712 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1713 len);
1714 total_len += len;
1715 } else {
1716 next_pos = find_next_bit_le(entry->discard_map,
1717 sbi->blocks_per_seg, cur_pos);
1718 }
1719 skip:
1720 cur_pos = next_pos;
1721 is_valid = !is_valid;
1722
1723 if (cur_pos < sbi->blocks_per_seg)
1724 goto find_next;
1725
1726 list_del(&entry->list);
1727 dcc->nr_discards -= total_len;
1728 kmem_cache_free(discard_entry_slab, entry);
1729 }
1730
1731 wake_up_discard_thread(sbi, false);
1732 }
1733
1734 void init_discard_policy(struct discard_policy *dpolicy,
1735 int discard_type, unsigned int granularity)
1736 {
1737 /* common policy */
1738 dpolicy->type = discard_type;
1739 dpolicy->sync = true;
1740 dpolicy->granularity = granularity;
1741
1742 if (discard_type == DPOLICY_BG) {
1743 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1744 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1745 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1746 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1747 dpolicy->io_aware = true;
1748 } else if (discard_type == DPOLICY_FORCE) {
1749 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
1750 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
1751 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1752 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1753 dpolicy->io_aware = true;
1754 } else if (discard_type == DPOLICY_FSTRIM) {
1755 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1756 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1757 dpolicy->io_aware = false;
1758 } else if (discard_type == DPOLICY_UMOUNT) {
1759 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
1760 dpolicy->io_aware_gran = MAX_PLIST_NUM;
1761 dpolicy->io_aware = false;
1762 }
1763 }
1764
1765 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1766 {
1767 dev_t dev = sbi->sb->s_bdev->bd_dev;
1768 struct discard_cmd_control *dcc;
1769 int err = 0, i;
1770
1771 if (SM_I(sbi)->dcc_info) {
1772 dcc = SM_I(sbi)->dcc_info;
1773 goto init_thread;
1774 }
1775
1776 dcc = kzalloc(sizeof(struct discard_cmd_control), GFP_KERNEL);
1777 if (!dcc)
1778 return -ENOMEM;
1779
1780 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
1781 INIT_LIST_HEAD(&dcc->entry_list);
1782 for (i = 0; i < MAX_PLIST_NUM; i++)
1783 INIT_LIST_HEAD(&dcc->pend_list[i]);
1784 INIT_LIST_HEAD(&dcc->wait_list);
1785 INIT_LIST_HEAD(&dcc->fstrim_list);
1786 mutex_init(&dcc->cmd_lock);
1787 atomic_set(&dcc->issued_discard, 0);
1788 atomic_set(&dcc->issing_discard, 0);
1789 atomic_set(&dcc->discard_cmd_cnt, 0);
1790 dcc->nr_discards = 0;
1791 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1792 dcc->undiscard_blks = 0;
1793 dcc->root = RB_ROOT;
1794
1795 init_waitqueue_head(&dcc->discard_wait_queue);
1796 SM_I(sbi)->dcc_info = dcc;
1797 init_thread:
1798 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1799 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1800 if (IS_ERR(dcc->f2fs_issue_discard)) {
1801 err = PTR_ERR(dcc->f2fs_issue_discard);
1802 kfree(dcc);
1803 SM_I(sbi)->dcc_info = NULL;
1804 return err;
1805 }
1806
1807 return err;
1808 }
1809
1810 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1811 {
1812 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1813
1814 if (!dcc)
1815 return;
1816
1817 stop_discard_thread(sbi);
1818
1819 kfree(dcc);
1820 SM_I(sbi)->dcc_info = NULL;
1821 }
1822
1823 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1824 {
1825 struct sit_info *sit_i = SIT_I(sbi);
1826
1827 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1828 sit_i->dirty_sentries++;
1829 return false;
1830 }
1831
1832 return true;
1833 }
1834
1835 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1836 unsigned int segno, int modified)
1837 {
1838 struct seg_entry *se = get_seg_entry(sbi, segno);
1839 se->type = type;
1840 if (modified)
1841 __mark_sit_entry_dirty(sbi, segno);
1842 }
1843
1844 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1845 {
1846 struct seg_entry *se;
1847 unsigned int segno, offset;
1848 long int new_vblocks;
1849 bool exist;
1850 #ifdef CONFIG_F2FS_CHECK_FS
1851 bool mir_exist;
1852 #endif
1853
1854 segno = GET_SEGNO(sbi, blkaddr);
1855
1856 se = get_seg_entry(sbi, segno);
1857 new_vblocks = se->valid_blocks + del;
1858 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1859
1860 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1861 (new_vblocks > sbi->blocks_per_seg)));
1862
1863 se->valid_blocks = new_vblocks;
1864 se->mtime = get_mtime(sbi);
1865 SIT_I(sbi)->max_mtime = se->mtime;
1866
1867 /* Update valid block bitmap */
1868 if (del > 0) {
1869 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
1870 #ifdef CONFIG_F2FS_CHECK_FS
1871 mir_exist = f2fs_test_and_set_bit(offset,
1872 se->cur_valid_map_mir);
1873 if (unlikely(exist != mir_exist)) {
1874 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1875 "when setting bitmap, blk:%u, old bit:%d",
1876 blkaddr, exist);
1877 f2fs_bug_on(sbi, 1);
1878 }
1879 #endif
1880 if (unlikely(exist)) {
1881 f2fs_msg(sbi->sb, KERN_ERR,
1882 "Bitmap was wrongly set, blk:%u", blkaddr);
1883 f2fs_bug_on(sbi, 1);
1884 se->valid_blocks--;
1885 del = 0;
1886 }
1887
1888 if (f2fs_discard_en(sbi) &&
1889 !f2fs_test_and_set_bit(offset, se->discard_map))
1890 sbi->discard_blks--;
1891
1892 /* don't overwrite by SSR to keep node chain */
1893 if (se->type == CURSEG_WARM_NODE) {
1894 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1895 se->ckpt_valid_blocks++;
1896 }
1897 } else {
1898 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
1899 #ifdef CONFIG_F2FS_CHECK_FS
1900 mir_exist = f2fs_test_and_clear_bit(offset,
1901 se->cur_valid_map_mir);
1902 if (unlikely(exist != mir_exist)) {
1903 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1904 "when clearing bitmap, blk:%u, old bit:%d",
1905 blkaddr, exist);
1906 f2fs_bug_on(sbi, 1);
1907 }
1908 #endif
1909 if (unlikely(!exist)) {
1910 f2fs_msg(sbi->sb, KERN_ERR,
1911 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1912 f2fs_bug_on(sbi, 1);
1913 se->valid_blocks++;
1914 del = 0;
1915 }
1916
1917 if (f2fs_discard_en(sbi) &&
1918 f2fs_test_and_clear_bit(offset, se->discard_map))
1919 sbi->discard_blks++;
1920 }
1921 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1922 se->ckpt_valid_blocks += del;
1923
1924 __mark_sit_entry_dirty(sbi, segno);
1925
1926 /* update total number of valid blocks to be written in ckpt area */
1927 SIT_I(sbi)->written_valid_blocks += del;
1928
1929 if (sbi->segs_per_sec > 1)
1930 get_sec_entry(sbi, segno)->valid_blocks += del;
1931 }
1932
1933 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1934 {
1935 unsigned int segno = GET_SEGNO(sbi, addr);
1936 struct sit_info *sit_i = SIT_I(sbi);
1937
1938 f2fs_bug_on(sbi, addr == NULL_ADDR);
1939 if (addr == NEW_ADDR)
1940 return;
1941
1942 /* add it into sit main buffer */
1943 down_write(&sit_i->sentry_lock);
1944
1945 update_sit_entry(sbi, addr, -1);
1946
1947 /* add it into dirty seglist */
1948 locate_dirty_segment(sbi, segno);
1949
1950 up_write(&sit_i->sentry_lock);
1951 }
1952
1953 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1954 {
1955 struct sit_info *sit_i = SIT_I(sbi);
1956 unsigned int segno, offset;
1957 struct seg_entry *se;
1958 bool is_cp = false;
1959
1960 if (!is_valid_data_blkaddr(sbi, blkaddr))
1961 return true;
1962
1963 down_read(&sit_i->sentry_lock);
1964
1965 segno = GET_SEGNO(sbi, blkaddr);
1966 se = get_seg_entry(sbi, segno);
1967 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1968
1969 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1970 is_cp = true;
1971
1972 up_read(&sit_i->sentry_lock);
1973
1974 return is_cp;
1975 }
1976
1977 /*
1978 * This function should be resided under the curseg_mutex lock
1979 */
1980 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1981 struct f2fs_summary *sum)
1982 {
1983 struct curseg_info *curseg = CURSEG_I(sbi, type);
1984 void *addr = curseg->sum_blk;
1985 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1986 memcpy(addr, sum, sizeof(struct f2fs_summary));
1987 }
1988
1989 /*
1990 * Calculate the number of current summary pages for writing
1991 */
1992 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1993 {
1994 int valid_sum_count = 0;
1995 int i, sum_in_page;
1996
1997 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1998 if (sbi->ckpt->alloc_type[i] == SSR)
1999 valid_sum_count += sbi->blocks_per_seg;
2000 else {
2001 if (for_ra)
2002 valid_sum_count += le16_to_cpu(
2003 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2004 else
2005 valid_sum_count += curseg_blkoff(sbi, i);
2006 }
2007 }
2008
2009 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2010 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2011 if (valid_sum_count <= sum_in_page)
2012 return 1;
2013 else if ((valid_sum_count - sum_in_page) <=
2014 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2015 return 2;
2016 return 3;
2017 }
2018
2019 /*
2020 * Caller should put this summary page
2021 */
2022 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2023 {
2024 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
2025 }
2026
2027 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
2028 {
2029 struct page *page = grab_meta_page(sbi, blk_addr);
2030
2031 memcpy(page_address(page), src, PAGE_SIZE);
2032 set_page_dirty(page);
2033 f2fs_put_page(page, 1);
2034 }
2035
2036 static void write_sum_page(struct f2fs_sb_info *sbi,
2037 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2038 {
2039 update_meta_page(sbi, (void *)sum_blk, blk_addr);
2040 }
2041
2042 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2043 int type, block_t blk_addr)
2044 {
2045 struct curseg_info *curseg = CURSEG_I(sbi, type);
2046 struct page *page = grab_meta_page(sbi, blk_addr);
2047 struct f2fs_summary_block *src = curseg->sum_blk;
2048 struct f2fs_summary_block *dst;
2049
2050 dst = (struct f2fs_summary_block *)page_address(page);
2051 memset(dst, 0, PAGE_SIZE);
2052
2053 mutex_lock(&curseg->curseg_mutex);
2054
2055 down_read(&curseg->journal_rwsem);
2056 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2057 up_read(&curseg->journal_rwsem);
2058
2059 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2060 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2061
2062 mutex_unlock(&curseg->curseg_mutex);
2063
2064 set_page_dirty(page);
2065 f2fs_put_page(page, 1);
2066 }
2067
2068 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2069 {
2070 struct curseg_info *curseg = CURSEG_I(sbi, type);
2071 unsigned int segno = curseg->segno + 1;
2072 struct free_segmap_info *free_i = FREE_I(sbi);
2073
2074 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2075 return !test_bit(segno, free_i->free_segmap);
2076 return 0;
2077 }
2078
2079 /*
2080 * Find a new segment from the free segments bitmap to right order
2081 * This function should be returned with success, otherwise BUG
2082 */
2083 static void get_new_segment(struct f2fs_sb_info *sbi,
2084 unsigned int *newseg, bool new_sec, int dir)
2085 {
2086 struct free_segmap_info *free_i = FREE_I(sbi);
2087 unsigned int segno, secno, zoneno;
2088 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2089 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2090 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2091 unsigned int left_start = hint;
2092 bool init = true;
2093 int go_left = 0;
2094 int i;
2095
2096 spin_lock(&free_i->segmap_lock);
2097
2098 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2099 segno = find_next_zero_bit(free_i->free_segmap,
2100 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2101 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2102 goto got_it;
2103 }
2104 find_other_zone:
2105 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2106 if (secno >= MAIN_SECS(sbi)) {
2107 if (dir == ALLOC_RIGHT) {
2108 secno = find_next_zero_bit(free_i->free_secmap,
2109 MAIN_SECS(sbi), 0);
2110 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2111 } else {
2112 go_left = 1;
2113 left_start = hint - 1;
2114 }
2115 }
2116 if (go_left == 0)
2117 goto skip_left;
2118
2119 while (test_bit(left_start, free_i->free_secmap)) {
2120 if (left_start > 0) {
2121 left_start--;
2122 continue;
2123 }
2124 left_start = find_next_zero_bit(free_i->free_secmap,
2125 MAIN_SECS(sbi), 0);
2126 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2127 break;
2128 }
2129 secno = left_start;
2130 skip_left:
2131 segno = GET_SEG_FROM_SEC(sbi, secno);
2132 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2133
2134 /* give up on finding another zone */
2135 if (!init)
2136 goto got_it;
2137 if (sbi->secs_per_zone == 1)
2138 goto got_it;
2139 if (zoneno == old_zoneno)
2140 goto got_it;
2141 if (dir == ALLOC_LEFT) {
2142 if (!go_left && zoneno + 1 >= total_zones)
2143 goto got_it;
2144 if (go_left && zoneno == 0)
2145 goto got_it;
2146 }
2147 for (i = 0; i < NR_CURSEG_TYPE; i++)
2148 if (CURSEG_I(sbi, i)->zone == zoneno)
2149 break;
2150
2151 if (i < NR_CURSEG_TYPE) {
2152 /* zone is in user, try another */
2153 if (go_left)
2154 hint = zoneno * sbi->secs_per_zone - 1;
2155 else if (zoneno + 1 >= total_zones)
2156 hint = 0;
2157 else
2158 hint = (zoneno + 1) * sbi->secs_per_zone;
2159 init = false;
2160 goto find_other_zone;
2161 }
2162 got_it:
2163 /* set it as dirty segment in free segmap */
2164 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2165 __set_inuse(sbi, segno);
2166 *newseg = segno;
2167 spin_unlock(&free_i->segmap_lock);
2168 }
2169
2170 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2171 {
2172 struct curseg_info *curseg = CURSEG_I(sbi, type);
2173 struct summary_footer *sum_footer;
2174
2175 curseg->segno = curseg->next_segno;
2176 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2177 curseg->next_blkoff = 0;
2178 curseg->next_segno = NULL_SEGNO;
2179
2180 sum_footer = &(curseg->sum_blk->footer);
2181 memset(sum_footer, 0, sizeof(struct summary_footer));
2182 if (IS_DATASEG(type))
2183 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2184 if (IS_NODESEG(type))
2185 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2186 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2187 }
2188
2189 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2190 {
2191 /* if segs_per_sec is large than 1, we need to keep original policy. */
2192 if (sbi->segs_per_sec != 1)
2193 return CURSEG_I(sbi, type)->segno;
2194
2195 if (test_opt(sbi, NOHEAP) &&
2196 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2197 return 0;
2198
2199 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2200 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2201 return CURSEG_I(sbi, type)->segno;
2202 }
2203
2204 /*
2205 * Allocate a current working segment.
2206 * This function always allocates a free segment in LFS manner.
2207 */
2208 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2209 {
2210 struct curseg_info *curseg = CURSEG_I(sbi, type);
2211 unsigned int segno = curseg->segno;
2212 int dir = ALLOC_LEFT;
2213
2214 write_sum_page(sbi, curseg->sum_blk,
2215 GET_SUM_BLOCK(sbi, segno));
2216 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2217 dir = ALLOC_RIGHT;
2218
2219 if (test_opt(sbi, NOHEAP))
2220 dir = ALLOC_RIGHT;
2221
2222 segno = __get_next_segno(sbi, type);
2223 get_new_segment(sbi, &segno, new_sec, dir);
2224 curseg->next_segno = segno;
2225 reset_curseg(sbi, type, 1);
2226 curseg->alloc_type = LFS;
2227 }
2228
2229 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2230 struct curseg_info *seg, block_t start)
2231 {
2232 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2233 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2234 unsigned long *target_map = SIT_I(sbi)->tmp_map;
2235 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2236 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2237 int i, pos;
2238
2239 for (i = 0; i < entries; i++)
2240 target_map[i] = ckpt_map[i] | cur_map[i];
2241
2242 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2243
2244 seg->next_blkoff = pos;
2245 }
2246
2247 /*
2248 * If a segment is written by LFS manner, next block offset is just obtained
2249 * by increasing the current block offset. However, if a segment is written by
2250 * SSR manner, next block offset obtained by calling __next_free_blkoff
2251 */
2252 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2253 struct curseg_info *seg)
2254 {
2255 if (seg->alloc_type == SSR)
2256 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2257 else
2258 seg->next_blkoff++;
2259 }
2260
2261 /*
2262 * This function always allocates a used segment(from dirty seglist) by SSR
2263 * manner, so it should recover the existing segment information of valid blocks
2264 */
2265 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2266 {
2267 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2268 struct curseg_info *curseg = CURSEG_I(sbi, type);
2269 unsigned int new_segno = curseg->next_segno;
2270 struct f2fs_summary_block *sum_node;
2271 struct page *sum_page;
2272
2273 write_sum_page(sbi, curseg->sum_blk,
2274 GET_SUM_BLOCK(sbi, curseg->segno));
2275 __set_test_and_inuse(sbi, new_segno);
2276
2277 mutex_lock(&dirty_i->seglist_lock);
2278 __remove_dirty_segment(sbi, new_segno, PRE);
2279 __remove_dirty_segment(sbi, new_segno, DIRTY);
2280 mutex_unlock(&dirty_i->seglist_lock);
2281
2282 reset_curseg(sbi, type, 1);
2283 curseg->alloc_type = SSR;
2284 __next_free_blkoff(sbi, curseg, 0);
2285
2286 sum_page = get_sum_page(sbi, new_segno);
2287 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2288 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2289 f2fs_put_page(sum_page, 1);
2290 }
2291
2292 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2293 {
2294 struct curseg_info *curseg = CURSEG_I(sbi, type);
2295 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2296 unsigned segno = NULL_SEGNO;
2297 int i, cnt;
2298 bool reversed = false;
2299
2300 /* need_SSR() already forces to do this */
2301 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2302 curseg->next_segno = segno;
2303 return 1;
2304 }
2305
2306 /* For node segments, let's do SSR more intensively */
2307 if (IS_NODESEG(type)) {
2308 if (type >= CURSEG_WARM_NODE) {
2309 reversed = true;
2310 i = CURSEG_COLD_NODE;
2311 } else {
2312 i = CURSEG_HOT_NODE;
2313 }
2314 cnt = NR_CURSEG_NODE_TYPE;
2315 } else {
2316 if (type >= CURSEG_WARM_DATA) {
2317 reversed = true;
2318 i = CURSEG_COLD_DATA;
2319 } else {
2320 i = CURSEG_HOT_DATA;
2321 }
2322 cnt = NR_CURSEG_DATA_TYPE;
2323 }
2324
2325 for (; cnt-- > 0; reversed ? i-- : i++) {
2326 if (i == type)
2327 continue;
2328 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2329 curseg->next_segno = segno;
2330 return 1;
2331 }
2332 }
2333 return 0;
2334 }
2335
2336 /*
2337 * flush out current segment and replace it with new segment
2338 * This function should be returned with success, otherwise BUG
2339 */
2340 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2341 int type, bool force)
2342 {
2343 struct curseg_info *curseg = CURSEG_I(sbi, type);
2344
2345 if (force)
2346 new_curseg(sbi, type, true);
2347 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2348 type == CURSEG_WARM_NODE)
2349 new_curseg(sbi, type, false);
2350 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2351 new_curseg(sbi, type, false);
2352 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
2353 change_curseg(sbi, type);
2354 else
2355 new_curseg(sbi, type, false);
2356
2357 stat_inc_seg_type(sbi, curseg);
2358 }
2359
2360 void allocate_new_segments(struct f2fs_sb_info *sbi)
2361 {
2362 struct curseg_info *curseg;
2363 unsigned int old_segno;
2364 int i;
2365
2366 down_write(&SIT_I(sbi)->sentry_lock);
2367
2368 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2369 curseg = CURSEG_I(sbi, i);
2370 old_segno = curseg->segno;
2371 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2372 locate_dirty_segment(sbi, old_segno);
2373 }
2374
2375 up_write(&SIT_I(sbi)->sentry_lock);
2376 }
2377
2378 static const struct segment_allocation default_salloc_ops = {
2379 .allocate_segment = allocate_segment_by_default,
2380 };
2381
2382 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2383 {
2384 __u64 trim_start = cpc->trim_start;
2385 bool has_candidate = false;
2386
2387 down_write(&SIT_I(sbi)->sentry_lock);
2388 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2389 if (add_discard_addrs(sbi, cpc, true)) {
2390 has_candidate = true;
2391 break;
2392 }
2393 }
2394 up_write(&SIT_I(sbi)->sentry_lock);
2395
2396 cpc->trim_start = trim_start;
2397 return has_candidate;
2398 }
2399
2400 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2401 {
2402 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2403 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2404 unsigned int start_segno, end_segno, cur_segno;
2405 block_t start_block, end_block;
2406 struct cp_control cpc;
2407 struct discard_policy dpolicy;
2408 unsigned long long trimmed = 0;
2409 int err = 0;
2410 bool need_align = test_opt(sbi, LFS) && sbi->segs_per_sec > 1;
2411
2412 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2413 return -EINVAL;
2414
2415 if (end <= MAIN_BLKADDR(sbi))
2416 goto out;
2417
2418 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2419 f2fs_msg(sbi->sb, KERN_WARNING,
2420 "Found FS corruption, run fsck to fix.");
2421 err = -EFSCORRUPTED;
2422 goto out;
2423 }
2424
2425 /* start/end segment number in main_area */
2426 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2427 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2428 GET_SEGNO(sbi, end);
2429 if (need_align) {
2430 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2431 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2432 }
2433
2434 cpc.reason = CP_DISCARD;
2435 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2436
2437 /* do checkpoint to issue discard commands safely */
2438 for (cur_segno = start_segno; cur_segno <= end_segno;
2439 cur_segno = cpc.trim_end + 1) {
2440 cpc.trim_start = cur_segno;
2441
2442 if (sbi->discard_blks == 0)
2443 break;
2444 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
2445 cpc.trim_end = end_segno;
2446 else
2447 cpc.trim_end = min_t(unsigned int,
2448 rounddown(cur_segno +
2449 BATCHED_TRIM_SEGMENTS(sbi),
2450 sbi->segs_per_sec) - 1, end_segno);
2451
2452 mutex_lock(&sbi->gc_mutex);
2453 err = write_checkpoint(sbi, &cpc);
2454 mutex_unlock(&sbi->gc_mutex);
2455 if (err)
2456 break;
2457
2458 schedule();
2459 }
2460
2461 start_block = START_BLOCK(sbi, start_segno);
2462 end_block = START_BLOCK(sbi, min(cur_segno, end_segno) + 1);
2463
2464 init_discard_policy(&dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2465 __issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block);
2466 trimmed = __wait_discard_cmd_range(sbi, &dpolicy,
2467 start_block, end_block);
2468 out:
2469 range->len = F2FS_BLK_TO_BYTES(trimmed);
2470 return err;
2471 }
2472
2473 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2474 {
2475 struct curseg_info *curseg = CURSEG_I(sbi, type);
2476 if (curseg->next_blkoff < sbi->blocks_per_seg)
2477 return true;
2478 return false;
2479 }
2480
2481 int rw_hint_to_seg_type(enum rw_hint hint)
2482 {
2483 switch (hint) {
2484 case WRITE_LIFE_SHORT:
2485 return CURSEG_HOT_DATA;
2486 case WRITE_LIFE_EXTREME:
2487 return CURSEG_COLD_DATA;
2488 default:
2489 return CURSEG_WARM_DATA;
2490 }
2491 }
2492
2493 static int __get_segment_type_2(struct f2fs_io_info *fio)
2494 {
2495 if (fio->type == DATA)
2496 return CURSEG_HOT_DATA;
2497 else
2498 return CURSEG_HOT_NODE;
2499 }
2500
2501 static int __get_segment_type_4(struct f2fs_io_info *fio)
2502 {
2503 if (fio->type == DATA) {
2504 struct inode *inode = fio->page->mapping->host;
2505
2506 if (S_ISDIR(inode->i_mode))
2507 return CURSEG_HOT_DATA;
2508 else
2509 return CURSEG_COLD_DATA;
2510 } else {
2511 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2512 return CURSEG_WARM_NODE;
2513 else
2514 return CURSEG_COLD_NODE;
2515 }
2516 }
2517
2518 static int __get_segment_type_6(struct f2fs_io_info *fio)
2519 {
2520 if (fio->type == DATA) {
2521 struct inode *inode = fio->page->mapping->host;
2522
2523 if (is_cold_data(fio->page) || file_is_cold(inode))
2524 return CURSEG_COLD_DATA;
2525 if (is_inode_flag_set(inode, FI_HOT_DATA))
2526 return CURSEG_HOT_DATA;
2527 return rw_hint_to_seg_type(inode->i_write_hint);
2528 } else {
2529 if (IS_DNODE(fio->page))
2530 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2531 CURSEG_HOT_NODE;
2532 return CURSEG_COLD_NODE;
2533 }
2534 }
2535
2536 static int __get_segment_type(struct f2fs_io_info *fio)
2537 {
2538 int type = 0;
2539
2540 switch (fio->sbi->active_logs) {
2541 case 2:
2542 type = __get_segment_type_2(fio);
2543 break;
2544 case 4:
2545 type = __get_segment_type_4(fio);
2546 break;
2547 case 6:
2548 type = __get_segment_type_6(fio);
2549 break;
2550 default:
2551 f2fs_bug_on(fio->sbi, true);
2552 }
2553
2554 if (IS_HOT(type))
2555 fio->temp = HOT;
2556 else if (IS_WARM(type))
2557 fio->temp = WARM;
2558 else
2559 fio->temp = COLD;
2560 return type;
2561 }
2562
2563 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2564 block_t old_blkaddr, block_t *new_blkaddr,
2565 struct f2fs_summary *sum, int type,
2566 struct f2fs_io_info *fio, bool add_list)
2567 {
2568 struct sit_info *sit_i = SIT_I(sbi);
2569 struct curseg_info *curseg = CURSEG_I(sbi, type);
2570
2571 down_read(&SM_I(sbi)->curseg_lock);
2572
2573 mutex_lock(&curseg->curseg_mutex);
2574 down_write(&sit_i->sentry_lock);
2575
2576 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2577
2578 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2579
2580 /*
2581 * __add_sum_entry should be resided under the curseg_mutex
2582 * because, this function updates a summary entry in the
2583 * current summary block.
2584 */
2585 __add_sum_entry(sbi, type, sum);
2586
2587 __refresh_next_blkoff(sbi, curseg);
2588
2589 stat_inc_block_count(sbi, curseg);
2590
2591 /*
2592 * SIT information should be updated before segment allocation,
2593 * since SSR needs latest valid block information.
2594 */
2595 update_sit_entry(sbi, *new_blkaddr, 1);
2596 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2597 update_sit_entry(sbi, old_blkaddr, -1);
2598
2599 if (!__has_curseg_space(sbi, type))
2600 sit_i->s_ops->allocate_segment(sbi, type, false);
2601
2602 /*
2603 * segment dirty status should be updated after segment allocation,
2604 * so we just need to update status only one time after previous
2605 * segment being closed.
2606 */
2607 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2608 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
2609
2610 up_write(&sit_i->sentry_lock);
2611
2612 if (page && IS_NODESEG(type)) {
2613 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2614
2615 f2fs_inode_chksum_set(sbi, page);
2616 }
2617
2618 if (add_list) {
2619 struct f2fs_bio_info *io;
2620
2621 INIT_LIST_HEAD(&fio->list);
2622 fio->in_list = true;
2623 io = sbi->write_io[fio->type] + fio->temp;
2624 spin_lock(&io->io_lock);
2625 list_add_tail(&fio->list, &io->io_list);
2626 spin_unlock(&io->io_lock);
2627 }
2628
2629 mutex_unlock(&curseg->curseg_mutex);
2630
2631 up_read(&SM_I(sbi)->curseg_lock);
2632 }
2633
2634 static void update_device_state(struct f2fs_io_info *fio)
2635 {
2636 struct f2fs_sb_info *sbi = fio->sbi;
2637 unsigned int devidx;
2638
2639 if (!sbi->s_ndevs)
2640 return;
2641
2642 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2643
2644 /* update device state for fsync */
2645 set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
2646
2647 /* update device state for checkpoint */
2648 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2649 spin_lock(&sbi->dev_lock);
2650 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2651 spin_unlock(&sbi->dev_lock);
2652 }
2653 }
2654
2655 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2656 {
2657 int type = __get_segment_type(fio);
2658 int err;
2659
2660 reallocate:
2661 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2662 &fio->new_blkaddr, sum, type, fio, true);
2663
2664 /* writeout dirty page into bdev */
2665 err = f2fs_submit_page_write(fio);
2666 if (err == -EAGAIN) {
2667 fio->old_blkaddr = fio->new_blkaddr;
2668 goto reallocate;
2669 } else if (!err) {
2670 update_device_state(fio);
2671 }
2672 }
2673
2674 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2675 enum iostat_type io_type)
2676 {
2677 struct f2fs_io_info fio = {
2678 .sbi = sbi,
2679 .type = META,
2680 .op = REQ_OP_WRITE,
2681 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
2682 .old_blkaddr = page->index,
2683 .new_blkaddr = page->index,
2684 .page = page,
2685 .encrypted_page = NULL,
2686 .in_list = false,
2687 };
2688
2689 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
2690 fio.op_flags &= ~REQ_META;
2691
2692 set_page_writeback(page);
2693 f2fs_submit_page_write(&fio);
2694
2695 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
2696 }
2697
2698 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
2699 {
2700 struct f2fs_summary sum;
2701
2702 set_summary(&sum, nid, 0, 0);
2703 do_write_page(&sum, fio);
2704
2705 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2706 }
2707
2708 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
2709 {
2710 struct f2fs_sb_info *sbi = fio->sbi;
2711 struct f2fs_summary sum;
2712 struct node_info ni;
2713
2714 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
2715 get_node_info(sbi, dn->nid, &ni);
2716 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
2717 do_write_page(&sum, fio);
2718 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
2719
2720 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
2721 }
2722
2723 int rewrite_data_page(struct f2fs_io_info *fio)
2724 {
2725 int err;
2726
2727 fio->new_blkaddr = fio->old_blkaddr;
2728 stat_inc_inplace_blocks(fio->sbi);
2729
2730 err = f2fs_submit_page_bio(fio);
2731 if (!err)
2732 update_device_state(fio);
2733
2734 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2735
2736 return err;
2737 }
2738
2739 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
2740 unsigned int segno)
2741 {
2742 int i;
2743
2744 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
2745 if (CURSEG_I(sbi, i)->segno == segno)
2746 break;
2747 }
2748 return i;
2749 }
2750
2751 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
2752 block_t old_blkaddr, block_t new_blkaddr,
2753 bool recover_curseg, bool recover_newaddr)
2754 {
2755 struct sit_info *sit_i = SIT_I(sbi);
2756 struct curseg_info *curseg;
2757 unsigned int segno, old_cursegno;
2758 struct seg_entry *se;
2759 int type;
2760 unsigned short old_blkoff;
2761
2762 segno = GET_SEGNO(sbi, new_blkaddr);
2763 se = get_seg_entry(sbi, segno);
2764 type = se->type;
2765
2766 down_write(&SM_I(sbi)->curseg_lock);
2767
2768 if (!recover_curseg) {
2769 /* for recovery flow */
2770 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2771 if (old_blkaddr == NULL_ADDR)
2772 type = CURSEG_COLD_DATA;
2773 else
2774 type = CURSEG_WARM_DATA;
2775 }
2776 } else {
2777 if (IS_CURSEG(sbi, segno)) {
2778 /* se->type is volatile as SSR allocation */
2779 type = __f2fs_get_curseg(sbi, segno);
2780 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
2781 } else {
2782 type = CURSEG_WARM_DATA;
2783 }
2784 }
2785
2786 curseg = CURSEG_I(sbi, type);
2787
2788 mutex_lock(&curseg->curseg_mutex);
2789 down_write(&sit_i->sentry_lock);
2790
2791 old_cursegno = curseg->segno;
2792 old_blkoff = curseg->next_blkoff;
2793
2794 /* change the current segment */
2795 if (segno != curseg->segno) {
2796 curseg->next_segno = segno;
2797 change_curseg(sbi, type);
2798 }
2799
2800 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
2801 __add_sum_entry(sbi, type, sum);
2802
2803 if (!recover_curseg || recover_newaddr)
2804 update_sit_entry(sbi, new_blkaddr, 1);
2805 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2806 update_sit_entry(sbi, old_blkaddr, -1);
2807
2808 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2809 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2810
2811 locate_dirty_segment(sbi, old_cursegno);
2812
2813 if (recover_curseg) {
2814 if (old_cursegno != curseg->segno) {
2815 curseg->next_segno = old_cursegno;
2816 change_curseg(sbi, type);
2817 }
2818 curseg->next_blkoff = old_blkoff;
2819 }
2820
2821 up_write(&sit_i->sentry_lock);
2822 mutex_unlock(&curseg->curseg_mutex);
2823 up_write(&SM_I(sbi)->curseg_lock);
2824 }
2825
2826 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2827 block_t old_addr, block_t new_addr,
2828 unsigned char version, bool recover_curseg,
2829 bool recover_newaddr)
2830 {
2831 struct f2fs_summary sum;
2832
2833 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2834
2835 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2836 recover_curseg, recover_newaddr);
2837
2838 f2fs_update_data_blkaddr(dn, new_addr);
2839 }
2840
2841 void f2fs_wait_on_page_writeback(struct page *page,
2842 enum page_type type, bool ordered)
2843 {
2844 if (PageWriteback(page)) {
2845 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2846
2847 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2848 0, page->index, type);
2849 if (ordered)
2850 wait_on_page_writeback(page);
2851 else
2852 wait_for_stable_page(page);
2853 }
2854 }
2855
2856 void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
2857 {
2858 struct page *cpage;
2859
2860 if (!is_valid_data_blkaddr(sbi, blkaddr))
2861 return;
2862
2863 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2864 if (cpage) {
2865 f2fs_wait_on_page_writeback(cpage, DATA, true);
2866 f2fs_put_page(cpage, 1);
2867 }
2868 }
2869
2870 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
2871 {
2872 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2873 struct curseg_info *seg_i;
2874 unsigned char *kaddr;
2875 struct page *page;
2876 block_t start;
2877 int i, j, offset;
2878
2879 start = start_sum_block(sbi);
2880
2881 page = get_meta_page(sbi, start++);
2882 kaddr = (unsigned char *)page_address(page);
2883
2884 /* Step 1: restore nat cache */
2885 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2886 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
2887
2888 /* Step 2: restore sit cache */
2889 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2890 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
2891 offset = 2 * SUM_JOURNAL_SIZE;
2892
2893 /* Step 3: restore summary entries */
2894 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2895 unsigned short blk_off;
2896 unsigned int segno;
2897
2898 seg_i = CURSEG_I(sbi, i);
2899 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
2900 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
2901 seg_i->next_segno = segno;
2902 reset_curseg(sbi, i, 0);
2903 seg_i->alloc_type = ckpt->alloc_type[i];
2904 seg_i->next_blkoff = blk_off;
2905
2906 if (seg_i->alloc_type == SSR)
2907 blk_off = sbi->blocks_per_seg;
2908
2909 for (j = 0; j < blk_off; j++) {
2910 struct f2fs_summary *s;
2911 s = (struct f2fs_summary *)(kaddr + offset);
2912 seg_i->sum_blk->entries[j] = *s;
2913 offset += SUMMARY_SIZE;
2914 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
2915 SUM_FOOTER_SIZE)
2916 continue;
2917
2918 f2fs_put_page(page, 1);
2919 page = NULL;
2920
2921 page = get_meta_page(sbi, start++);
2922 kaddr = (unsigned char *)page_address(page);
2923 offset = 0;
2924 }
2925 }
2926 f2fs_put_page(page, 1);
2927 return 0;
2928 }
2929
2930 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
2931 {
2932 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2933 struct f2fs_summary_block *sum;
2934 struct curseg_info *curseg;
2935 struct page *new;
2936 unsigned short blk_off;
2937 unsigned int segno = 0;
2938 block_t blk_addr = 0;
2939
2940 /* get segment number and block addr */
2941 if (IS_DATASEG(type)) {
2942 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
2943 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
2944 CURSEG_HOT_DATA]);
2945 if (__exist_node_summaries(sbi))
2946 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
2947 else
2948 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
2949 } else {
2950 segno = le32_to_cpu(ckpt->cur_node_segno[type -
2951 CURSEG_HOT_NODE]);
2952 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
2953 CURSEG_HOT_NODE]);
2954 if (__exist_node_summaries(sbi))
2955 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
2956 type - CURSEG_HOT_NODE);
2957 else
2958 blk_addr = GET_SUM_BLOCK(sbi, segno);
2959 }
2960
2961 new = get_meta_page(sbi, blk_addr);
2962 sum = (struct f2fs_summary_block *)page_address(new);
2963
2964 if (IS_NODESEG(type)) {
2965 if (__exist_node_summaries(sbi)) {
2966 struct f2fs_summary *ns = &sum->entries[0];
2967 int i;
2968 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
2969 ns->version = 0;
2970 ns->ofs_in_node = 0;
2971 }
2972 } else {
2973 int err;
2974
2975 err = restore_node_summary(sbi, segno, sum);
2976 if (err) {
2977 f2fs_put_page(new, 1);
2978 return err;
2979 }
2980 }
2981 }
2982
2983 /* set uncompleted segment to curseg */
2984 curseg = CURSEG_I(sbi, type);
2985 mutex_lock(&curseg->curseg_mutex);
2986
2987 /* update journal info */
2988 down_write(&curseg->journal_rwsem);
2989 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
2990 up_write(&curseg->journal_rwsem);
2991
2992 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
2993 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
2994 curseg->next_segno = segno;
2995 reset_curseg(sbi, type, 0);
2996 curseg->alloc_type = ckpt->alloc_type[type];
2997 curseg->next_blkoff = blk_off;
2998 mutex_unlock(&curseg->curseg_mutex);
2999 f2fs_put_page(new, 1);
3000 return 0;
3001 }
3002
3003 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3004 {
3005 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3006 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3007 int type = CURSEG_HOT_DATA;
3008 int err;
3009
3010 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3011 int npages = npages_for_summary_flush(sbi, true);
3012
3013 if (npages >= 2)
3014 ra_meta_pages(sbi, start_sum_block(sbi), npages,
3015 META_CP, true);
3016
3017 /* restore for compacted data summary */
3018 if (read_compacted_summaries(sbi))
3019 return -EINVAL;
3020 type = CURSEG_HOT_NODE;
3021 }
3022
3023 if (__exist_node_summaries(sbi))
3024 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3025 NR_CURSEG_TYPE - type, META_CP, true);
3026
3027 for (; type <= CURSEG_COLD_NODE; type++) {
3028 err = read_normal_summaries(sbi, type);
3029 if (err)
3030 return err;
3031 }
3032
3033 /* sanity check for summary blocks */
3034 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3035 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3036 return -EINVAL;
3037
3038 return 0;
3039 }
3040
3041 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3042 {
3043 struct page *page;
3044 unsigned char *kaddr;
3045 struct f2fs_summary *summary;
3046 struct curseg_info *seg_i;
3047 int written_size = 0;
3048 int i, j;
3049
3050 page = grab_meta_page(sbi, blkaddr++);
3051 kaddr = (unsigned char *)page_address(page);
3052 memset(kaddr, 0, PAGE_SIZE);
3053
3054 /* Step 1: write nat cache */
3055 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3056 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3057 written_size += SUM_JOURNAL_SIZE;
3058
3059 /* Step 2: write sit cache */
3060 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3061 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3062 written_size += SUM_JOURNAL_SIZE;
3063
3064 /* Step 3: write summary entries */
3065 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3066 unsigned short blkoff;
3067 seg_i = CURSEG_I(sbi, i);
3068 if (sbi->ckpt->alloc_type[i] == SSR)
3069 blkoff = sbi->blocks_per_seg;
3070 else
3071 blkoff = curseg_blkoff(sbi, i);
3072
3073 for (j = 0; j < blkoff; j++) {
3074 if (!page) {
3075 page = grab_meta_page(sbi, blkaddr++);
3076 kaddr = (unsigned char *)page_address(page);
3077 memset(kaddr, 0, PAGE_SIZE);
3078 written_size = 0;
3079 }
3080 summary = (struct f2fs_summary *)(kaddr + written_size);
3081 *summary = seg_i->sum_blk->entries[j];
3082 written_size += SUMMARY_SIZE;
3083
3084 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3085 SUM_FOOTER_SIZE)
3086 continue;
3087
3088 set_page_dirty(page);
3089 f2fs_put_page(page, 1);
3090 page = NULL;
3091 }
3092 }
3093 if (page) {
3094 set_page_dirty(page);
3095 f2fs_put_page(page, 1);
3096 }
3097 }
3098
3099 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3100 block_t blkaddr, int type)
3101 {
3102 int i, end;
3103 if (IS_DATASEG(type))
3104 end = type + NR_CURSEG_DATA_TYPE;
3105 else
3106 end = type + NR_CURSEG_NODE_TYPE;
3107
3108 for (i = type; i < end; i++)
3109 write_current_sum_page(sbi, i, blkaddr + (i - type));
3110 }
3111
3112 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3113 {
3114 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3115 write_compacted_summaries(sbi, start_blk);
3116 else
3117 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3118 }
3119
3120 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3121 {
3122 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3123 }
3124
3125 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3126 unsigned int val, int alloc)
3127 {
3128 int i;
3129
3130 if (type == NAT_JOURNAL) {
3131 for (i = 0; i < nats_in_cursum(journal); i++) {
3132 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3133 return i;
3134 }
3135 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3136 return update_nats_in_cursum(journal, 1);
3137 } else if (type == SIT_JOURNAL) {
3138 for (i = 0; i < sits_in_cursum(journal); i++)
3139 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3140 return i;
3141 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3142 return update_sits_in_cursum(journal, 1);
3143 }
3144 return -1;
3145 }
3146
3147 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3148 unsigned int segno)
3149 {
3150 return get_meta_page(sbi, current_sit_addr(sbi, segno));
3151 }
3152
3153 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3154 unsigned int start)
3155 {
3156 struct sit_info *sit_i = SIT_I(sbi);
3157 struct page *src_page, *dst_page;
3158 pgoff_t src_off, dst_off;
3159 void *src_addr, *dst_addr;
3160
3161 src_off = current_sit_addr(sbi, start);
3162 dst_off = next_sit_addr(sbi, src_off);
3163
3164 /* get current sit block page without lock */
3165 src_page = get_meta_page(sbi, src_off);
3166 dst_page = grab_meta_page(sbi, dst_off);
3167 f2fs_bug_on(sbi, PageDirty(src_page));
3168
3169 src_addr = page_address(src_page);
3170 dst_addr = page_address(dst_page);
3171 memcpy(dst_addr, src_addr, PAGE_SIZE);
3172
3173 set_page_dirty(dst_page);
3174 f2fs_put_page(src_page, 1);
3175
3176 set_to_next_sit(sit_i, start);
3177
3178 return dst_page;
3179 }
3180
3181 static struct sit_entry_set *grab_sit_entry_set(void)
3182 {
3183 struct sit_entry_set *ses =
3184 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3185
3186 ses->entry_cnt = 0;
3187 INIT_LIST_HEAD(&ses->set_list);
3188 return ses;
3189 }
3190
3191 static void release_sit_entry_set(struct sit_entry_set *ses)
3192 {
3193 list_del(&ses->set_list);
3194 kmem_cache_free(sit_entry_set_slab, ses);
3195 }
3196
3197 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3198 struct list_head *head)
3199 {
3200 struct sit_entry_set *next = ses;
3201
3202 if (list_is_last(&ses->set_list, head))
3203 return;
3204
3205 list_for_each_entry_continue(next, head, set_list)
3206 if (ses->entry_cnt <= next->entry_cnt)
3207 break;
3208
3209 list_move_tail(&ses->set_list, &next->set_list);
3210 }
3211
3212 static void add_sit_entry(unsigned int segno, struct list_head *head)
3213 {
3214 struct sit_entry_set *ses;
3215 unsigned int start_segno = START_SEGNO(segno);
3216
3217 list_for_each_entry(ses, head, set_list) {
3218 if (ses->start_segno == start_segno) {
3219 ses->entry_cnt++;
3220 adjust_sit_entry_set(ses, head);
3221 return;
3222 }
3223 }
3224
3225 ses = grab_sit_entry_set();
3226
3227 ses->start_segno = start_segno;
3228 ses->entry_cnt++;
3229 list_add(&ses->set_list, head);
3230 }
3231
3232 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3233 {
3234 struct f2fs_sm_info *sm_info = SM_I(sbi);
3235 struct list_head *set_list = &sm_info->sit_entry_set;
3236 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3237 unsigned int segno;
3238
3239 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3240 add_sit_entry(segno, set_list);
3241 }
3242
3243 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3244 {
3245 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3246 struct f2fs_journal *journal = curseg->journal;
3247 int i;
3248
3249 down_write(&curseg->journal_rwsem);
3250 for (i = 0; i < sits_in_cursum(journal); i++) {
3251 unsigned int segno;
3252 bool dirtied;
3253
3254 segno = le32_to_cpu(segno_in_journal(journal, i));
3255 dirtied = __mark_sit_entry_dirty(sbi, segno);
3256
3257 if (!dirtied)
3258 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3259 }
3260 update_sits_in_cursum(journal, -i);
3261 up_write(&curseg->journal_rwsem);
3262 }
3263
3264 /*
3265 * CP calls this function, which flushes SIT entries including sit_journal,
3266 * and moves prefree segs to free segs.
3267 */
3268 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3269 {
3270 struct sit_info *sit_i = SIT_I(sbi);
3271 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3272 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3273 struct f2fs_journal *journal = curseg->journal;
3274 struct sit_entry_set *ses, *tmp;
3275 struct list_head *head = &SM_I(sbi)->sit_entry_set;
3276 bool to_journal = true;
3277 struct seg_entry *se;
3278
3279 down_write(&sit_i->sentry_lock);
3280
3281 if (!sit_i->dirty_sentries)
3282 goto out;
3283
3284 /*
3285 * add and account sit entries of dirty bitmap in sit entry
3286 * set temporarily
3287 */
3288 add_sits_in_set(sbi);
3289
3290 /*
3291 * if there are no enough space in journal to store dirty sit
3292 * entries, remove all entries from journal and add and account
3293 * them in sit entry set.
3294 */
3295 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3296 remove_sits_in_journal(sbi);
3297
3298 /*
3299 * there are two steps to flush sit entries:
3300 * #1, flush sit entries to journal in current cold data summary block.
3301 * #2, flush sit entries to sit page.
3302 */
3303 list_for_each_entry_safe(ses, tmp, head, set_list) {
3304 struct page *page = NULL;
3305 struct f2fs_sit_block *raw_sit = NULL;
3306 unsigned int start_segno = ses->start_segno;
3307 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3308 (unsigned long)MAIN_SEGS(sbi));
3309 unsigned int segno = start_segno;
3310
3311 if (to_journal &&
3312 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3313 to_journal = false;
3314
3315 if (to_journal) {
3316 down_write(&curseg->journal_rwsem);
3317 } else {
3318 page = get_next_sit_page(sbi, start_segno);
3319 raw_sit = page_address(page);
3320 }
3321
3322 /* flush dirty sit entries in region of current sit set */
3323 for_each_set_bit_from(segno, bitmap, end) {
3324 int offset, sit_offset;
3325
3326 se = get_seg_entry(sbi, segno);
3327
3328 /* add discard candidates */
3329 if (!(cpc->reason & CP_DISCARD)) {
3330 cpc->trim_start = segno;
3331 add_discard_addrs(sbi, cpc, false);
3332 }
3333
3334 if (to_journal) {
3335 offset = lookup_journal_in_cursum(journal,
3336 SIT_JOURNAL, segno, 1);
3337 f2fs_bug_on(sbi, offset < 0);
3338 segno_in_journal(journal, offset) =
3339 cpu_to_le32(segno);
3340 seg_info_to_raw_sit(se,
3341 &sit_in_journal(journal, offset));
3342 } else {
3343 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3344 seg_info_to_raw_sit(se,
3345 &raw_sit->entries[sit_offset]);
3346 }
3347
3348 __clear_bit(segno, bitmap);
3349 sit_i->dirty_sentries--;
3350 ses->entry_cnt--;
3351 }
3352
3353 if (to_journal)
3354 up_write(&curseg->journal_rwsem);
3355 else
3356 f2fs_put_page(page, 1);
3357
3358 f2fs_bug_on(sbi, ses->entry_cnt);
3359 release_sit_entry_set(ses);
3360 }
3361
3362 f2fs_bug_on(sbi, !list_empty(head));
3363 f2fs_bug_on(sbi, sit_i->dirty_sentries);
3364 out:
3365 if (cpc->reason & CP_DISCARD) {
3366 __u64 trim_start = cpc->trim_start;
3367
3368 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3369 add_discard_addrs(sbi, cpc, false);
3370
3371 cpc->trim_start = trim_start;
3372 }
3373 up_write(&sit_i->sentry_lock);
3374
3375 set_prefree_as_free_segments(sbi);
3376 }
3377
3378 static int build_sit_info(struct f2fs_sb_info *sbi)
3379 {
3380 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3381 struct sit_info *sit_i;
3382 unsigned int sit_segs, start;
3383 char *src_bitmap;
3384 unsigned int bitmap_size;
3385
3386 /* allocate memory for SIT information */
3387 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
3388 if (!sit_i)
3389 return -ENOMEM;
3390
3391 SM_I(sbi)->sit_info = sit_i;
3392
3393 sit_i->sentries = kvzalloc(MAIN_SEGS(sbi) *
3394 sizeof(struct seg_entry), GFP_KERNEL);
3395 if (!sit_i->sentries)
3396 return -ENOMEM;
3397
3398 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3399 sit_i->dirty_sentries_bitmap = kvzalloc(bitmap_size, GFP_KERNEL);
3400 if (!sit_i->dirty_sentries_bitmap)
3401 return -ENOMEM;
3402
3403 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3404 sit_i->sentries[start].cur_valid_map
3405 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3406 sit_i->sentries[start].ckpt_valid_map
3407 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3408 if (!sit_i->sentries[start].cur_valid_map ||
3409 !sit_i->sentries[start].ckpt_valid_map)
3410 return -ENOMEM;
3411
3412 #ifdef CONFIG_F2FS_CHECK_FS
3413 sit_i->sentries[start].cur_valid_map_mir
3414 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3415 if (!sit_i->sentries[start].cur_valid_map_mir)
3416 return -ENOMEM;
3417 #endif
3418
3419 if (f2fs_discard_en(sbi)) {
3420 sit_i->sentries[start].discard_map
3421 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3422 if (!sit_i->sentries[start].discard_map)
3423 return -ENOMEM;
3424 }
3425 }
3426
3427 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3428 if (!sit_i->tmp_map)
3429 return -ENOMEM;
3430
3431 if (sbi->segs_per_sec > 1) {
3432 sit_i->sec_entries = kvzalloc(MAIN_SECS(sbi) *
3433 sizeof(struct sec_entry), GFP_KERNEL);
3434 if (!sit_i->sec_entries)
3435 return -ENOMEM;
3436 }
3437
3438 /* get information related with SIT */
3439 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3440
3441 /* setup SIT bitmap from ckeckpoint pack */
3442 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3443 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3444
3445 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3446 if (!sit_i->sit_bitmap)
3447 return -ENOMEM;
3448
3449 #ifdef CONFIG_F2FS_CHECK_FS
3450 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3451 if (!sit_i->sit_bitmap_mir)
3452 return -ENOMEM;
3453 #endif
3454
3455 /* init SIT information */
3456 sit_i->s_ops = &default_salloc_ops;
3457
3458 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3459 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3460 sit_i->written_valid_blocks = 0;
3461 sit_i->bitmap_size = bitmap_size;
3462 sit_i->dirty_sentries = 0;
3463 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3464 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3465 sit_i->mounted_time = ktime_get_real_seconds();
3466 init_rwsem(&sit_i->sentry_lock);
3467 return 0;
3468 }
3469
3470 static int build_free_segmap(struct f2fs_sb_info *sbi)
3471 {
3472 struct free_segmap_info *free_i;
3473 unsigned int bitmap_size, sec_bitmap_size;
3474
3475 /* allocate memory for free segmap information */
3476 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
3477 if (!free_i)
3478 return -ENOMEM;
3479
3480 SM_I(sbi)->free_info = free_i;
3481
3482 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3483 free_i->free_segmap = kvmalloc(bitmap_size, GFP_KERNEL);
3484 if (!free_i->free_segmap)
3485 return -ENOMEM;
3486
3487 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3488 free_i->free_secmap = kvmalloc(sec_bitmap_size, GFP_KERNEL);
3489 if (!free_i->free_secmap)
3490 return -ENOMEM;
3491
3492 /* set all segments as dirty temporarily */
3493 memset(free_i->free_segmap, 0xff, bitmap_size);
3494 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3495
3496 /* init free segmap information */
3497 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3498 free_i->free_segments = 0;
3499 free_i->free_sections = 0;
3500 spin_lock_init(&free_i->segmap_lock);
3501 return 0;
3502 }
3503
3504 static int build_curseg(struct f2fs_sb_info *sbi)
3505 {
3506 struct curseg_info *array;
3507 int i;
3508
3509 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
3510 if (!array)
3511 return -ENOMEM;
3512
3513 SM_I(sbi)->curseg_array = array;
3514
3515 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3516 mutex_init(&array[i].curseg_mutex);
3517 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
3518 if (!array[i].sum_blk)
3519 return -ENOMEM;
3520 init_rwsem(&array[i].journal_rwsem);
3521 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
3522 GFP_KERNEL);
3523 if (!array[i].journal)
3524 return -ENOMEM;
3525 array[i].segno = NULL_SEGNO;
3526 array[i].next_blkoff = 0;
3527 }
3528 return restore_curseg_summaries(sbi);
3529 }
3530
3531 static int build_sit_entries(struct f2fs_sb_info *sbi)
3532 {
3533 struct sit_info *sit_i = SIT_I(sbi);
3534 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3535 struct f2fs_journal *journal = curseg->journal;
3536 struct seg_entry *se;
3537 struct f2fs_sit_entry sit;
3538 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3539 unsigned int i, start, end;
3540 unsigned int readed, start_blk = 0;
3541 int err = 0;
3542 block_t total_node_blocks = 0;
3543
3544 do {
3545 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3546 META_SIT, true);
3547
3548 start = start_blk * sit_i->sents_per_block;
3549 end = (start_blk + readed) * sit_i->sents_per_block;
3550
3551 for (; start < end && start < MAIN_SEGS(sbi); start++) {
3552 struct f2fs_sit_block *sit_blk;
3553 struct page *page;
3554
3555 se = &sit_i->sentries[start];
3556 page = get_current_sit_page(sbi, start);
3557 sit_blk = (struct f2fs_sit_block *)page_address(page);
3558 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3559 f2fs_put_page(page, 1);
3560
3561 err = check_block_count(sbi, start, &sit);
3562 if (err)
3563 return err;
3564 seg_info_from_raw_sit(se, &sit);
3565 if (IS_NODESEG(se->type))
3566 total_node_blocks += se->valid_blocks;
3567
3568 /* build discard map only one time */
3569 if (f2fs_discard_en(sbi)) {
3570 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3571 memset(se->discard_map, 0xff,
3572 SIT_VBLOCK_MAP_SIZE);
3573 } else {
3574 memcpy(se->discard_map,
3575 se->cur_valid_map,
3576 SIT_VBLOCK_MAP_SIZE);
3577 sbi->discard_blks +=
3578 sbi->blocks_per_seg -
3579 se->valid_blocks;
3580 }
3581 }
3582
3583 if (sbi->segs_per_sec > 1)
3584 get_sec_entry(sbi, start)->valid_blocks +=
3585 se->valid_blocks;
3586 }
3587 start_blk += readed;
3588 } while (start_blk < sit_blk_cnt);
3589
3590 down_read(&curseg->journal_rwsem);
3591 for (i = 0; i < sits_in_cursum(journal); i++) {
3592 unsigned int old_valid_blocks;
3593
3594 start = le32_to_cpu(segno_in_journal(journal, i));
3595 if (start >= MAIN_SEGS(sbi)) {
3596 f2fs_msg(sbi->sb, KERN_ERR,
3597 "Wrong journal entry on segno %u",
3598 start);
3599 set_sbi_flag(sbi, SBI_NEED_FSCK);
3600 err = -EFSCORRUPTED;
3601 break;
3602 }
3603
3604 se = &sit_i->sentries[start];
3605 sit = sit_in_journal(journal, i);
3606
3607 old_valid_blocks = se->valid_blocks;
3608 if (IS_NODESEG(se->type))
3609 total_node_blocks -= old_valid_blocks;
3610
3611 err = check_block_count(sbi, start, &sit);
3612 if (err)
3613 break;
3614 seg_info_from_raw_sit(se, &sit);
3615 if (IS_NODESEG(se->type))
3616 total_node_blocks += se->valid_blocks;
3617
3618 if (f2fs_discard_en(sbi)) {
3619 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3620 memset(se->discard_map, 0xff,
3621 SIT_VBLOCK_MAP_SIZE);
3622 } else {
3623 memcpy(se->discard_map, se->cur_valid_map,
3624 SIT_VBLOCK_MAP_SIZE);
3625 sbi->discard_blks += old_valid_blocks -
3626 se->valid_blocks;
3627 }
3628 }
3629
3630 if (sbi->segs_per_sec > 1)
3631 get_sec_entry(sbi, start)->valid_blocks +=
3632 se->valid_blocks - old_valid_blocks;
3633 }
3634 up_read(&curseg->journal_rwsem);
3635
3636 if (!err && total_node_blocks != valid_node_count(sbi)) {
3637 f2fs_msg(sbi->sb, KERN_ERR,
3638 "SIT is corrupted node# %u vs %u",
3639 total_node_blocks, valid_node_count(sbi));
3640 set_sbi_flag(sbi, SBI_NEED_FSCK);
3641 err = -EFSCORRUPTED;
3642 }
3643
3644 return err;
3645 }
3646
3647 static void init_free_segmap(struct f2fs_sb_info *sbi)
3648 {
3649 unsigned int start;
3650 int type;
3651
3652 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3653 struct seg_entry *sentry = get_seg_entry(sbi, start);
3654 if (!sentry->valid_blocks)
3655 __set_free(sbi, start);
3656 else
3657 SIT_I(sbi)->written_valid_blocks +=
3658 sentry->valid_blocks;
3659 }
3660
3661 /* set use the current segments */
3662 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3663 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3664 __set_test_and_inuse(sbi, curseg_t->segno);
3665 }
3666 }
3667
3668 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3669 {
3670 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3671 struct free_segmap_info *free_i = FREE_I(sbi);
3672 unsigned int segno = 0, offset = 0;
3673 unsigned short valid_blocks;
3674
3675 while (1) {
3676 /* find dirty segment based on free segmap */
3677 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3678 if (segno >= MAIN_SEGS(sbi))
3679 break;
3680 offset = segno + 1;
3681 valid_blocks = get_valid_blocks(sbi, segno, false);
3682 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
3683 continue;
3684 if (valid_blocks > sbi->blocks_per_seg) {
3685 f2fs_bug_on(sbi, 1);
3686 continue;
3687 }
3688 mutex_lock(&dirty_i->seglist_lock);
3689 __locate_dirty_segment(sbi, segno, DIRTY);
3690 mutex_unlock(&dirty_i->seglist_lock);
3691 }
3692 }
3693
3694 static int init_victim_secmap(struct f2fs_sb_info *sbi)
3695 {
3696 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3697 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3698
3699 dirty_i->victim_secmap = kvzalloc(bitmap_size, GFP_KERNEL);
3700 if (!dirty_i->victim_secmap)
3701 return -ENOMEM;
3702 return 0;
3703 }
3704
3705 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3706 {
3707 struct dirty_seglist_info *dirty_i;
3708 unsigned int bitmap_size, i;
3709
3710 /* allocate memory for dirty segments list information */
3711 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
3712 if (!dirty_i)
3713 return -ENOMEM;
3714
3715 SM_I(sbi)->dirty_info = dirty_i;
3716 mutex_init(&dirty_i->seglist_lock);
3717
3718 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3719
3720 for (i = 0; i < NR_DIRTY_TYPE; i++) {
3721 dirty_i->dirty_segmap[i] = kvzalloc(bitmap_size, GFP_KERNEL);
3722 if (!dirty_i->dirty_segmap[i])
3723 return -ENOMEM;
3724 }
3725
3726 init_dirty_segmap(sbi);
3727 return init_victim_secmap(sbi);
3728 }
3729
3730 static int sanity_check_curseg(struct f2fs_sb_info *sbi)
3731 {
3732 int i;
3733
3734 /*
3735 * In LFS/SSR curseg, .next_blkoff should point to an unused blkaddr;
3736 * In LFS curseg, all blkaddr after .next_blkoff should be unused.
3737 */
3738 for (i = 0; i < NO_CHECK_TYPE; i++) {
3739 struct curseg_info *curseg = CURSEG_I(sbi, i);
3740 struct seg_entry *se = get_seg_entry(sbi, curseg->segno);
3741 unsigned int blkofs = curseg->next_blkoff;
3742
3743 if (f2fs_test_bit(blkofs, se->cur_valid_map))
3744 goto out;
3745
3746 if (curseg->alloc_type == SSR)
3747 continue;
3748
3749 for (blkofs += 1; blkofs < sbi->blocks_per_seg; blkofs++) {
3750 if (!f2fs_test_bit(blkofs, se->cur_valid_map))
3751 continue;
3752 out:
3753 f2fs_msg(sbi->sb, KERN_ERR,
3754 "Current segment's next free block offset is "
3755 "inconsistent with bitmap, logtype:%u, "
3756 "segno:%u, type:%u, next_blkoff:%u, blkofs:%u",
3757 i, curseg->segno, curseg->alloc_type,
3758 curseg->next_blkoff, blkofs);
3759 return -EFSCORRUPTED;
3760 }
3761 }
3762 return 0;
3763 }
3764
3765 /*
3766 * Update min, max modified time for cost-benefit GC algorithm
3767 */
3768 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3769 {
3770 struct sit_info *sit_i = SIT_I(sbi);
3771 unsigned int segno;
3772
3773 down_write(&sit_i->sentry_lock);
3774
3775 sit_i->min_mtime = LLONG_MAX;
3776
3777 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
3778 unsigned int i;
3779 unsigned long long mtime = 0;
3780
3781 for (i = 0; i < sbi->segs_per_sec; i++)
3782 mtime += get_seg_entry(sbi, segno + i)->mtime;
3783
3784 mtime = div_u64(mtime, sbi->segs_per_sec);
3785
3786 if (sit_i->min_mtime > mtime)
3787 sit_i->min_mtime = mtime;
3788 }
3789 sit_i->max_mtime = get_mtime(sbi);
3790 up_write(&sit_i->sentry_lock);
3791 }
3792
3793 int build_segment_manager(struct f2fs_sb_info *sbi)
3794 {
3795 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3796 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3797 struct f2fs_sm_info *sm_info;
3798 int err;
3799
3800 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
3801 if (!sm_info)
3802 return -ENOMEM;
3803
3804 /* init sm info */
3805 sbi->sm_info = sm_info;
3806 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3807 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3808 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3809 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3810 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3811 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3812 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
3813 sm_info->rec_prefree_segments = sm_info->main_segments *
3814 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
3815 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3816 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3817
3818 if (!test_opt(sbi, LFS))
3819 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
3820 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
3821 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
3822 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
3823 sm_info->min_ssr_sections = reserved_sections(sbi);
3824
3825 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
3826
3827 INIT_LIST_HEAD(&sm_info->sit_entry_set);
3828
3829 init_rwsem(&sm_info->curseg_lock);
3830
3831 if (!f2fs_readonly(sbi->sb)) {
3832 err = create_flush_cmd_control(sbi);
3833 if (err)
3834 return err;
3835 }
3836
3837 err = create_discard_cmd_control(sbi);
3838 if (err)
3839 return err;
3840
3841 err = build_sit_info(sbi);
3842 if (err)
3843 return err;
3844 err = build_free_segmap(sbi);
3845 if (err)
3846 return err;
3847 err = build_curseg(sbi);
3848 if (err)
3849 return err;
3850
3851 /* reinit free segmap based on SIT */
3852 err = build_sit_entries(sbi);
3853 if (err)
3854 return err;
3855
3856 init_free_segmap(sbi);
3857 err = build_dirty_segmap(sbi);
3858 if (err)
3859 return err;
3860
3861 err = sanity_check_curseg(sbi);
3862 if (err)
3863 return err;
3864
3865 init_min_max_mtime(sbi);
3866 return 0;
3867 }
3868
3869 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3870 enum dirty_type dirty_type)
3871 {
3872 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3873
3874 mutex_lock(&dirty_i->seglist_lock);
3875 kvfree(dirty_i->dirty_segmap[dirty_type]);
3876 dirty_i->nr_dirty[dirty_type] = 0;
3877 mutex_unlock(&dirty_i->seglist_lock);
3878 }
3879
3880 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
3881 {
3882 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3883 kvfree(dirty_i->victim_secmap);
3884 }
3885
3886 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3887 {
3888 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3889 int i;
3890
3891 if (!dirty_i)
3892 return;
3893
3894 /* discard pre-free/dirty segments list */
3895 for (i = 0; i < NR_DIRTY_TYPE; i++)
3896 discard_dirty_segmap(sbi, i);
3897
3898 destroy_victim_secmap(sbi);
3899 SM_I(sbi)->dirty_info = NULL;
3900 kfree(dirty_i);
3901 }
3902
3903 static void destroy_curseg(struct f2fs_sb_info *sbi)
3904 {
3905 struct curseg_info *array = SM_I(sbi)->curseg_array;
3906 int i;
3907
3908 if (!array)
3909 return;
3910 SM_I(sbi)->curseg_array = NULL;
3911 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3912 kfree(array[i].sum_blk);
3913 kfree(array[i].journal);
3914 }
3915 kfree(array);
3916 }
3917
3918 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3919 {
3920 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3921 if (!free_i)
3922 return;
3923 SM_I(sbi)->free_info = NULL;
3924 kvfree(free_i->free_segmap);
3925 kvfree(free_i->free_secmap);
3926 kfree(free_i);
3927 }
3928
3929 static void destroy_sit_info(struct f2fs_sb_info *sbi)
3930 {
3931 struct sit_info *sit_i = SIT_I(sbi);
3932 unsigned int start;
3933
3934 if (!sit_i)
3935 return;
3936
3937 if (sit_i->sentries) {
3938 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3939 kfree(sit_i->sentries[start].cur_valid_map);
3940 #ifdef CONFIG_F2FS_CHECK_FS
3941 kfree(sit_i->sentries[start].cur_valid_map_mir);
3942 #endif
3943 kfree(sit_i->sentries[start].ckpt_valid_map);
3944 kfree(sit_i->sentries[start].discard_map);
3945 }
3946 }
3947 kfree(sit_i->tmp_map);
3948
3949 kvfree(sit_i->sentries);
3950 kvfree(sit_i->sec_entries);
3951 kvfree(sit_i->dirty_sentries_bitmap);
3952
3953 SM_I(sbi)->sit_info = NULL;
3954 kfree(sit_i->sit_bitmap);
3955 #ifdef CONFIG_F2FS_CHECK_FS
3956 kfree(sit_i->sit_bitmap_mir);
3957 #endif
3958 kfree(sit_i);
3959 }
3960
3961 void destroy_segment_manager(struct f2fs_sb_info *sbi)
3962 {
3963 struct f2fs_sm_info *sm_info = SM_I(sbi);
3964
3965 if (!sm_info)
3966 return;
3967 destroy_flush_cmd_control(sbi, true);
3968 destroy_discard_cmd_control(sbi);
3969 destroy_dirty_segmap(sbi);
3970 destroy_curseg(sbi);
3971 destroy_free_segmap(sbi);
3972 destroy_sit_info(sbi);
3973 sbi->sm_info = NULL;
3974 kfree(sm_info);
3975 }
3976
3977 int __init create_segment_manager_caches(void)
3978 {
3979 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
3980 sizeof(struct discard_entry));
3981 if (!discard_entry_slab)
3982 goto fail;
3983
3984 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
3985 sizeof(struct discard_cmd));
3986 if (!discard_cmd_slab)
3987 goto destroy_discard_entry;
3988
3989 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
3990 sizeof(struct sit_entry_set));
3991 if (!sit_entry_set_slab)
3992 goto destroy_discard_cmd;
3993
3994 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
3995 sizeof(struct inmem_pages));
3996 if (!inmem_entry_slab)
3997 goto destroy_sit_entry_set;
3998 return 0;
3999
4000 destroy_sit_entry_set:
4001 kmem_cache_destroy(sit_entry_set_slab);
4002 destroy_discard_cmd:
4003 kmem_cache_destroy(discard_cmd_slab);
4004 destroy_discard_entry:
4005 kmem_cache_destroy(discard_entry_slab);
4006 fail:
4007 return -ENOMEM;
4008 }
4009
4010 void destroy_segment_manager_caches(void)
4011 {
4012 kmem_cache_destroy(sit_entry_set_slab);
4013 kmem_cache_destroy(discard_cmd_slab);
4014 kmem_cache_destroy(discard_entry_slab);
4015 kmem_cache_destroy(inmem_entry_slab);
4016 }
ubuntu-bionic-kernel mirror