]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - fs/f2fs/segment.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: mark inode dirty explicitly in recover_inode()
[mirror_ubuntu-jammy-kernel.git] / fs / f2fs / segment.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * fs/f2fs/segment.c
4 *
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
7 */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/bio.h>
11 #include <linux/blkdev.h>
12 #include <linux/prefetch.h>
13 #include <linux/kthread.h>
14 #include <linux/swap.h>
15 #include <linux/timer.h>
16 #include <linux/freezer.h>
17 #include <linux/sched/signal.h>
18
19 #include "f2fs.h"
20 #include "segment.h"
21 #include "node.h"
22 #include "gc.h"
23 #include "trace.h"
24 #include <trace/events/f2fs.h>
25
26 #define __reverse_ffz(x) __reverse_ffs(~(x))
27
28 static struct kmem_cache *discard_entry_slab;
29 static struct kmem_cache *discard_cmd_slab;
30 static struct kmem_cache *sit_entry_set_slab;
31 static struct kmem_cache *inmem_entry_slab;
32
33 static unsigned long __reverse_ulong(unsigned char *str)
34 {
35 unsigned long tmp = 0;
36 int shift = 24, idx = 0;
37
38 #if BITS_PER_LONG == 64
39 shift = 56;
40 #endif
41 while (shift >= 0) {
42 tmp |= (unsigned long)str[idx++] << shift;
43 shift -= BITS_PER_BYTE;
44 }
45 return tmp;
46 }
47
48 /*
49 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
50 * MSB and LSB are reversed in a byte by f2fs_set_bit.
51 */
52 static inline unsigned long __reverse_ffs(unsigned long word)
53 {
54 int num = 0;
55
56 #if BITS_PER_LONG == 64
57 if ((word & 0xffffffff00000000UL) == 0)
58 num += 32;
59 else
60 word >>= 32;
61 #endif
62 if ((word & 0xffff0000) == 0)
63 num += 16;
64 else
65 word >>= 16;
66
67 if ((word & 0xff00) == 0)
68 num += 8;
69 else
70 word >>= 8;
71
72 if ((word & 0xf0) == 0)
73 num += 4;
74 else
75 word >>= 4;
76
77 if ((word & 0xc) == 0)
78 num += 2;
79 else
80 word >>= 2;
81
82 if ((word & 0x2) == 0)
83 num += 1;
84 return num;
85 }
86
87 /*
88 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
89 * f2fs_set_bit makes MSB and LSB reversed in a byte.
90 * @size must be integral times of unsigned long.
91 * Example:
92 * MSB <--> LSB
93 * f2fs_set_bit(0, bitmap) => 1000 0000
94 * f2fs_set_bit(7, bitmap) => 0000 0001
95 */
96 static unsigned long __find_rev_next_bit(const unsigned long *addr,
97 unsigned long size, unsigned long offset)
98 {
99 const unsigned long *p = addr + BIT_WORD(offset);
100 unsigned long result = size;
101 unsigned long tmp;
102
103 if (offset >= size)
104 return size;
105
106 size -= (offset & ~(BITS_PER_LONG - 1));
107 offset %= BITS_PER_LONG;
108
109 while (1) {
110 if (*p == 0)
111 goto pass;
112
113 tmp = __reverse_ulong((unsigned char *)p);
114
115 tmp &= ~0UL >> offset;
116 if (size < BITS_PER_LONG)
117 tmp &= (~0UL << (BITS_PER_LONG - size));
118 if (tmp)
119 goto found;
120 pass:
121 if (size <= BITS_PER_LONG)
122 break;
123 size -= BITS_PER_LONG;
124 offset = 0;
125 p++;
126 }
127 return result;
128 found:
129 return result - size + __reverse_ffs(tmp);
130 }
131
132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
133 unsigned long size, unsigned long offset)
134 {
135 const unsigned long *p = addr + BIT_WORD(offset);
136 unsigned long result = size;
137 unsigned long tmp;
138
139 if (offset >= size)
140 return size;
141
142 size -= (offset & ~(BITS_PER_LONG - 1));
143 offset %= BITS_PER_LONG;
144
145 while (1) {
146 if (*p == ~0UL)
147 goto pass;
148
149 tmp = __reverse_ulong((unsigned char *)p);
150
151 if (offset)
152 tmp |= ~0UL << (BITS_PER_LONG - offset);
153 if (size < BITS_PER_LONG)
154 tmp |= ~0UL >> size;
155 if (tmp != ~0UL)
156 goto found;
157 pass:
158 if (size <= BITS_PER_LONG)
159 break;
160 size -= BITS_PER_LONG;
161 offset = 0;
162 p++;
163 }
164 return result;
165 found:
166 return result - size + __reverse_ffz(tmp);
167 }
168
169 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
170 {
171 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
172 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
173 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
174
175 if (test_opt(sbi, LFS))
176 return false;
177 if (sbi->gc_mode == GC_URGENT)
178 return true;
179
180 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
181 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
182 }
183
184 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
185 {
186 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
187 struct f2fs_inode_info *fi = F2FS_I(inode);
188 struct inmem_pages *new;
189
190 f2fs_trace_pid(page);
191
192 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
193 SetPagePrivate(page);
194
195 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
196
197 /* add atomic page indices to the list */
198 new->page = page;
199 INIT_LIST_HEAD(&new->list);
200
201 /* increase reference count with clean state */
202 mutex_lock(&fi->inmem_lock);
203 get_page(page);
204 list_add_tail(&new->list, &fi->inmem_pages);
205 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
206 if (list_empty(&fi->inmem_ilist))
207 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
208 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
209 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
210 mutex_unlock(&fi->inmem_lock);
211
212 trace_f2fs_register_inmem_page(page, INMEM);
213 }
214
215 static int __revoke_inmem_pages(struct inode *inode,
216 struct list_head *head, bool drop, bool recover)
217 {
218 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
219 struct inmem_pages *cur, *tmp;
220 int err = 0;
221
222 list_for_each_entry_safe(cur, tmp, head, list) {
223 struct page *page = cur->page;
224
225 if (drop)
226 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
227
228 lock_page(page);
229
230 f2fs_wait_on_page_writeback(page, DATA, true);
231
232 if (recover) {
233 struct dnode_of_data dn;
234 struct node_info ni;
235
236 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
237 retry:
238 set_new_dnode(&dn, inode, NULL, NULL, 0);
239 err = f2fs_get_dnode_of_data(&dn, page->index,
240 LOOKUP_NODE);
241 if (err) {
242 if (err == -ENOMEM) {
243 congestion_wait(BLK_RW_ASYNC, HZ/50);
244 cond_resched();
245 goto retry;
246 }
247 err = -EAGAIN;
248 goto next;
249 }
250
251 err = f2fs_get_node_info(sbi, dn.nid, &ni);
252 if (err) {
253 f2fs_put_dnode(&dn);
254 return err;
255 }
256
257 if (cur->old_addr == NEW_ADDR) {
258 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
259 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
260 } else
261 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
262 cur->old_addr, ni.version, true, true);
263 f2fs_put_dnode(&dn);
264 }
265 next:
266 /* we don't need to invalidate this in the sccessful status */
267 if (drop || recover)
268 ClearPageUptodate(page);
269 set_page_private(page, 0);
270 ClearPagePrivate(page);
271 f2fs_put_page(page, 1);
272
273 list_del(&cur->list);
274 kmem_cache_free(inmem_entry_slab, cur);
275 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
276 }
277 return err;
278 }
279
280 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
281 {
282 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
283 struct inode *inode;
284 struct f2fs_inode_info *fi;
285 next:
286 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
287 if (list_empty(head)) {
288 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
289 return;
290 }
291 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
292 inode = igrab(&fi->vfs_inode);
293 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
294
295 if (inode) {
296 if (gc_failure) {
297 if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
298 goto drop;
299 goto skip;
300 }
301 drop:
302 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
303 f2fs_drop_inmem_pages(inode);
304 iput(inode);
305 }
306 skip:
307 congestion_wait(BLK_RW_ASYNC, HZ/50);
308 cond_resched();
309 goto next;
310 }
311
312 void f2fs_drop_inmem_pages(struct inode *inode)
313 {
314 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
315 struct f2fs_inode_info *fi = F2FS_I(inode);
316
317 mutex_lock(&fi->inmem_lock);
318 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
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 mutex_unlock(&fi->inmem_lock);
324
325 clear_inode_flag(inode, FI_ATOMIC_FILE);
326 fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
327 stat_dec_atomic_write(inode);
328 }
329
330 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
331 {
332 struct f2fs_inode_info *fi = F2FS_I(inode);
333 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
334 struct list_head *head = &fi->inmem_pages;
335 struct inmem_pages *cur = NULL;
336
337 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
338
339 mutex_lock(&fi->inmem_lock);
340 list_for_each_entry(cur, head, list) {
341 if (cur->page == page)
342 break;
343 }
344
345 f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
346 list_del(&cur->list);
347 mutex_unlock(&fi->inmem_lock);
348
349 dec_page_count(sbi, F2FS_INMEM_PAGES);
350 kmem_cache_free(inmem_entry_slab, cur);
351
352 ClearPageUptodate(page);
353 set_page_private(page, 0);
354 ClearPagePrivate(page);
355 f2fs_put_page(page, 0);
356
357 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
358 }
359
360 static int __f2fs_commit_inmem_pages(struct inode *inode)
361 {
362 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
363 struct f2fs_inode_info *fi = F2FS_I(inode);
364 struct inmem_pages *cur, *tmp;
365 struct f2fs_io_info fio = {
366 .sbi = sbi,
367 .ino = inode->i_ino,
368 .type = DATA,
369 .op = REQ_OP_WRITE,
370 .op_flags = REQ_SYNC | REQ_PRIO,
371 .io_type = FS_DATA_IO,
372 };
373 struct list_head revoke_list;
374 pgoff_t last_idx = ULONG_MAX;
375 int err = 0;
376
377 INIT_LIST_HEAD(&revoke_list);
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 f2fs_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 = f2fs_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 /*
420 * try to revoke all committed pages, but still we could fail
421 * due to no memory or other reason, if that happened, EAGAIN
422 * will be returned, which means in such case, transaction is
423 * already not integrity, caller should use journal to do the
424 * recovery or rewrite & commit last transaction. For other
425 * error number, revoking was done by filesystem itself.
426 */
427 err = __revoke_inmem_pages(inode, &revoke_list, false, true);
428
429 /* drop all uncommitted pages */
430 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
431 } else {
432 __revoke_inmem_pages(inode, &revoke_list, false, false);
433 }
434
435 return err;
436 }
437
438 int f2fs_commit_inmem_pages(struct inode *inode)
439 {
440 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
441 struct f2fs_inode_info *fi = F2FS_I(inode);
442 int err;
443
444 f2fs_balance_fs(sbi, true);
445
446 down_write(&fi->i_gc_rwsem[WRITE]);
447
448 f2fs_lock_op(sbi);
449 set_inode_flag(inode, FI_ATOMIC_COMMIT);
450
451 mutex_lock(&fi->inmem_lock);
452 err = __f2fs_commit_inmem_pages(inode);
453
454 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
455 if (!list_empty(&fi->inmem_ilist))
456 list_del_init(&fi->inmem_ilist);
457 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
458 mutex_unlock(&fi->inmem_lock);
459
460 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
461
462 f2fs_unlock_op(sbi);
463 up_write(&fi->i_gc_rwsem[WRITE]);
464
465 return err;
466 }
467
468 /*
469 * This function balances dirty node and dentry pages.
470 * In addition, it controls garbage collection.
471 */
472 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
473 {
474 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
475 f2fs_show_injection_info(FAULT_CHECKPOINT);
476 f2fs_stop_checkpoint(sbi, false);
477 }
478
479 /* balance_fs_bg is able to be pending */
480 if (need && excess_cached_nats(sbi))
481 f2fs_balance_fs_bg(sbi);
482
483 /*
484 * We should do GC or end up with checkpoint, if there are so many dirty
485 * dir/node pages without enough free segments.
486 */
487 if (has_not_enough_free_secs(sbi, 0, 0)) {
488 mutex_lock(&sbi->gc_mutex);
489 f2fs_gc(sbi, false, false, NULL_SEGNO);
490 }
491 }
492
493 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
494 {
495 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
496 return;
497
498 /* try to shrink extent cache when there is no enough memory */
499 if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
500 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
501
502 /* check the # of cached NAT entries */
503 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
504 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
505
506 if (!f2fs_available_free_memory(sbi, FREE_NIDS))
507 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
508 else
509 f2fs_build_free_nids(sbi, false, false);
510
511 if (!is_idle(sbi, REQ_TIME) &&
512 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
513 return;
514
515 /* checkpoint is the only way to shrink partial cached entries */
516 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
517 !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
518 excess_prefree_segs(sbi) ||
519 excess_dirty_nats(sbi) ||
520 excess_dirty_nodes(sbi) ||
521 f2fs_time_over(sbi, CP_TIME)) {
522 if (test_opt(sbi, DATA_FLUSH)) {
523 struct blk_plug plug;
524
525 blk_start_plug(&plug);
526 f2fs_sync_dirty_inodes(sbi, FILE_INODE);
527 blk_finish_plug(&plug);
528 }
529 f2fs_sync_fs(sbi->sb, true);
530 stat_inc_bg_cp_count(sbi->stat_info);
531 }
532 }
533
534 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
535 struct block_device *bdev)
536 {
537 struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
538 int ret;
539
540 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
541 bio_set_dev(bio, bdev);
542 ret = submit_bio_wait(bio);
543 bio_put(bio);
544
545 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
546 test_opt(sbi, FLUSH_MERGE), ret);
547 return ret;
548 }
549
550 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
551 {
552 int ret = 0;
553 int i;
554
555 if (!sbi->s_ndevs)
556 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
557
558 for (i = 0; i < sbi->s_ndevs; i++) {
559 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
560 continue;
561 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
562 if (ret)
563 break;
564 }
565 return ret;
566 }
567
568 static int issue_flush_thread(void *data)
569 {
570 struct f2fs_sb_info *sbi = data;
571 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
572 wait_queue_head_t *q = &fcc->flush_wait_queue;
573 repeat:
574 if (kthread_should_stop())
575 return 0;
576
577 sb_start_intwrite(sbi->sb);
578
579 if (!llist_empty(&fcc->issue_list)) {
580 struct flush_cmd *cmd, *next;
581 int ret;
582
583 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
584 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
585
586 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
587
588 ret = submit_flush_wait(sbi, cmd->ino);
589 atomic_inc(&fcc->issued_flush);
590
591 llist_for_each_entry_safe(cmd, next,
592 fcc->dispatch_list, llnode) {
593 cmd->ret = ret;
594 complete(&cmd->wait);
595 }
596 fcc->dispatch_list = NULL;
597 }
598
599 sb_end_intwrite(sbi->sb);
600
601 wait_event_interruptible(*q,
602 kthread_should_stop() || !llist_empty(&fcc->issue_list));
603 goto repeat;
604 }
605
606 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
607 {
608 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
609 struct flush_cmd cmd;
610 int ret;
611
612 if (test_opt(sbi, NOBARRIER))
613 return 0;
614
615 if (!test_opt(sbi, FLUSH_MERGE)) {
616 ret = submit_flush_wait(sbi, ino);
617 atomic_inc(&fcc->issued_flush);
618 return ret;
619 }
620
621 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
622 ret = submit_flush_wait(sbi, ino);
623 atomic_dec(&fcc->issing_flush);
624
625 atomic_inc(&fcc->issued_flush);
626 return ret;
627 }
628
629 cmd.ino = ino;
630 init_completion(&cmd.wait);
631
632 llist_add(&cmd.llnode, &fcc->issue_list);
633
634 /* update issue_list before we wake up issue_flush thread */
635 smp_mb();
636
637 if (waitqueue_active(&fcc->flush_wait_queue))
638 wake_up(&fcc->flush_wait_queue);
639
640 if (fcc->f2fs_issue_flush) {
641 wait_for_completion(&cmd.wait);
642 atomic_dec(&fcc->issing_flush);
643 } else {
644 struct llist_node *list;
645
646 list = llist_del_all(&fcc->issue_list);
647 if (!list) {
648 wait_for_completion(&cmd.wait);
649 atomic_dec(&fcc->issing_flush);
650 } else {
651 struct flush_cmd *tmp, *next;
652
653 ret = submit_flush_wait(sbi, ino);
654
655 llist_for_each_entry_safe(tmp, next, list, llnode) {
656 if (tmp == &cmd) {
657 cmd.ret = ret;
658 atomic_dec(&fcc->issing_flush);
659 continue;
660 }
661 tmp->ret = ret;
662 complete(&tmp->wait);
663 }
664 }
665 }
666
667 return cmd.ret;
668 }
669
670 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
671 {
672 dev_t dev = sbi->sb->s_bdev->bd_dev;
673 struct flush_cmd_control *fcc;
674 int err = 0;
675
676 if (SM_I(sbi)->fcc_info) {
677 fcc = SM_I(sbi)->fcc_info;
678 if (fcc->f2fs_issue_flush)
679 return err;
680 goto init_thread;
681 }
682
683 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
684 if (!fcc)
685 return -ENOMEM;
686 atomic_set(&fcc->issued_flush, 0);
687 atomic_set(&fcc->issing_flush, 0);
688 init_waitqueue_head(&fcc->flush_wait_queue);
689 init_llist_head(&fcc->issue_list);
690 SM_I(sbi)->fcc_info = fcc;
691 if (!test_opt(sbi, FLUSH_MERGE))
692 return err;
693
694 init_thread:
695 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
696 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
697 if (IS_ERR(fcc->f2fs_issue_flush)) {
698 err = PTR_ERR(fcc->f2fs_issue_flush);
699 kfree(fcc);
700 SM_I(sbi)->fcc_info = NULL;
701 return err;
702 }
703
704 return err;
705 }
706
707 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
708 {
709 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
710
711 if (fcc && fcc->f2fs_issue_flush) {
712 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
713
714 fcc->f2fs_issue_flush = NULL;
715 kthread_stop(flush_thread);
716 }
717 if (free) {
718 kfree(fcc);
719 SM_I(sbi)->fcc_info = NULL;
720 }
721 }
722
723 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
724 {
725 int ret = 0, i;
726
727 if (!sbi->s_ndevs)
728 return 0;
729
730 for (i = 1; i < sbi->s_ndevs; i++) {
731 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
732 continue;
733 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
734 if (ret)
735 break;
736
737 spin_lock(&sbi->dev_lock);
738 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
739 spin_unlock(&sbi->dev_lock);
740 }
741
742 return ret;
743 }
744
745 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
746 enum dirty_type dirty_type)
747 {
748 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
749
750 /* need not be added */
751 if (IS_CURSEG(sbi, segno))
752 return;
753
754 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
755 dirty_i->nr_dirty[dirty_type]++;
756
757 if (dirty_type == DIRTY) {
758 struct seg_entry *sentry = get_seg_entry(sbi, segno);
759 enum dirty_type t = sentry->type;
760
761 if (unlikely(t >= DIRTY)) {
762 f2fs_bug_on(sbi, 1);
763 return;
764 }
765 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
766 dirty_i->nr_dirty[t]++;
767 }
768 }
769
770 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
771 enum dirty_type dirty_type)
772 {
773 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
774
775 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
776 dirty_i->nr_dirty[dirty_type]--;
777
778 if (dirty_type == DIRTY) {
779 struct seg_entry *sentry = get_seg_entry(sbi, segno);
780 enum dirty_type t = sentry->type;
781
782 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
783 dirty_i->nr_dirty[t]--;
784
785 if (get_valid_blocks(sbi, segno, true) == 0)
786 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
787 dirty_i->victim_secmap);
788 }
789 }
790
791 /*
792 * Should not occur error such as -ENOMEM.
793 * Adding dirty entry into seglist is not critical operation.
794 * If a given segment is one of current working segments, it won't be added.
795 */
796 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
797 {
798 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
799 unsigned short valid_blocks;
800
801 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
802 return;
803
804 mutex_lock(&dirty_i->seglist_lock);
805
806 valid_blocks = get_valid_blocks(sbi, segno, false);
807
808 if (valid_blocks == 0) {
809 __locate_dirty_segment(sbi, segno, PRE);
810 __remove_dirty_segment(sbi, segno, DIRTY);
811 } else if (valid_blocks < sbi->blocks_per_seg) {
812 __locate_dirty_segment(sbi, segno, DIRTY);
813 } else {
814 /* Recovery routine with SSR needs this */
815 __remove_dirty_segment(sbi, segno, DIRTY);
816 }
817
818 mutex_unlock(&dirty_i->seglist_lock);
819 }
820
821 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
822 struct block_device *bdev, block_t lstart,
823 block_t start, block_t len)
824 {
825 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
826 struct list_head *pend_list;
827 struct discard_cmd *dc;
828
829 f2fs_bug_on(sbi, !len);
830
831 pend_list = &dcc->pend_list[plist_idx(len)];
832
833 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
834 INIT_LIST_HEAD(&dc->list);
835 dc->bdev = bdev;
836 dc->lstart = lstart;
837 dc->start = start;
838 dc->len = len;
839 dc->ref = 0;
840 dc->state = D_PREP;
841 dc->issuing = 0;
842 dc->error = 0;
843 init_completion(&dc->wait);
844 list_add_tail(&dc->list, pend_list);
845 spin_lock_init(&dc->lock);
846 dc->bio_ref = 0;
847 atomic_inc(&dcc->discard_cmd_cnt);
848 dcc->undiscard_blks += len;
849
850 return dc;
851 }
852
853 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
854 struct block_device *bdev, block_t lstart,
855 block_t start, block_t len,
856 struct rb_node *parent, struct rb_node **p)
857 {
858 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
859 struct discard_cmd *dc;
860
861 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
862
863 rb_link_node(&dc->rb_node, parent, p);
864 rb_insert_color(&dc->rb_node, &dcc->root);
865
866 return dc;
867 }
868
869 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
870 struct discard_cmd *dc)
871 {
872 if (dc->state == D_DONE)
873 atomic_sub(dc->issuing, &dcc->issing_discard);
874
875 list_del(&dc->list);
876 rb_erase(&dc->rb_node, &dcc->root);
877 dcc->undiscard_blks -= dc->len;
878
879 kmem_cache_free(discard_cmd_slab, dc);
880
881 atomic_dec(&dcc->discard_cmd_cnt);
882 }
883
884 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
885 struct discard_cmd *dc)
886 {
887 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
888 unsigned long flags;
889
890 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
891
892 spin_lock_irqsave(&dc->lock, flags);
893 if (dc->bio_ref) {
894 spin_unlock_irqrestore(&dc->lock, flags);
895 return;
896 }
897 spin_unlock_irqrestore(&dc->lock, flags);
898
899 f2fs_bug_on(sbi, dc->ref);
900
901 if (dc->error == -EOPNOTSUPP)
902 dc->error = 0;
903
904 if (dc->error)
905 printk_ratelimited(
906 "%sF2FS-fs: Issue discard(%u, %u, %u) failed, ret: %d",
907 KERN_INFO, dc->lstart, dc->start, dc->len, dc->error);
908 __detach_discard_cmd(dcc, dc);
909 }
910
911 static void f2fs_submit_discard_endio(struct bio *bio)
912 {
913 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
914 unsigned long flags;
915
916 dc->error = blk_status_to_errno(bio->bi_status);
917
918 spin_lock_irqsave(&dc->lock, flags);
919 dc->bio_ref--;
920 if (!dc->bio_ref && dc->state == D_SUBMIT) {
921 dc->state = D_DONE;
922 complete_all(&dc->wait);
923 }
924 spin_unlock_irqrestore(&dc->lock, flags);
925 bio_put(bio);
926 }
927
928 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
929 block_t start, block_t end)
930 {
931 #ifdef CONFIG_F2FS_CHECK_FS
932 struct seg_entry *sentry;
933 unsigned int segno;
934 block_t blk = start;
935 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
936 unsigned long *map;
937
938 while (blk < end) {
939 segno = GET_SEGNO(sbi, blk);
940 sentry = get_seg_entry(sbi, segno);
941 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
942
943 if (end < START_BLOCK(sbi, segno + 1))
944 size = GET_BLKOFF_FROM_SEG0(sbi, end);
945 else
946 size = max_blocks;
947 map = (unsigned long *)(sentry->cur_valid_map);
948 offset = __find_rev_next_bit(map, size, offset);
949 f2fs_bug_on(sbi, offset != size);
950 blk = START_BLOCK(sbi, segno + 1);
951 }
952 #endif
953 }
954
955 static void __init_discard_policy(struct f2fs_sb_info *sbi,
956 struct discard_policy *dpolicy,
957 int discard_type, unsigned int granularity)
958 {
959 /* common policy */
960 dpolicy->type = discard_type;
961 dpolicy->sync = true;
962 dpolicy->ordered = false;
963 dpolicy->granularity = granularity;
964
965 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
966 dpolicy->io_aware_gran = MAX_PLIST_NUM;
967
968 if (discard_type == DPOLICY_BG) {
969 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
970 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
971 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
972 dpolicy->io_aware = true;
973 dpolicy->sync = false;
974 dpolicy->ordered = true;
975 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
976 dpolicy->granularity = 1;
977 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
978 }
979 } else if (discard_type == DPOLICY_FORCE) {
980 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
981 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
982 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
983 dpolicy->io_aware = false;
984 } else if (discard_type == DPOLICY_FSTRIM) {
985 dpolicy->io_aware = false;
986 } else if (discard_type == DPOLICY_UMOUNT) {
987 dpolicy->max_requests = UINT_MAX;
988 dpolicy->io_aware = false;
989 }
990 }
991
992 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
993 struct block_device *bdev, block_t lstart,
994 block_t start, block_t len);
995 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
996 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
997 struct discard_policy *dpolicy,
998 struct discard_cmd *dc,
999 unsigned int *issued)
1000 {
1001 struct block_device *bdev = dc->bdev;
1002 struct request_queue *q = bdev_get_queue(bdev);
1003 unsigned int max_discard_blocks =
1004 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1005 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1006 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1007 &(dcc->fstrim_list) : &(dcc->wait_list);
1008 int flag = dpolicy->sync ? REQ_SYNC : 0;
1009 block_t lstart, start, len, total_len;
1010 int err = 0;
1011
1012 if (dc->state != D_PREP)
1013 return 0;
1014
1015 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1016 return 0;
1017
1018 trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1019
1020 lstart = dc->lstart;
1021 start = dc->start;
1022 len = dc->len;
1023 total_len = len;
1024
1025 dc->len = 0;
1026
1027 while (total_len && *issued < dpolicy->max_requests && !err) {
1028 struct bio *bio = NULL;
1029 unsigned long flags;
1030 bool last = true;
1031
1032 if (len > max_discard_blocks) {
1033 len = max_discard_blocks;
1034 last = false;
1035 }
1036
1037 (*issued)++;
1038 if (*issued == dpolicy->max_requests)
1039 last = true;
1040
1041 dc->len += len;
1042
1043 if (time_to_inject(sbi, FAULT_DISCARD)) {
1044 f2fs_show_injection_info(FAULT_DISCARD);
1045 err = -EIO;
1046 goto submit;
1047 }
1048 err = __blkdev_issue_discard(bdev,
1049 SECTOR_FROM_BLOCK(start),
1050 SECTOR_FROM_BLOCK(len),
1051 GFP_NOFS, 0, &bio);
1052 submit:
1053 if (err) {
1054 spin_lock_irqsave(&dc->lock, flags);
1055 if (dc->state == D_PARTIAL)
1056 dc->state = D_SUBMIT;
1057 spin_unlock_irqrestore(&dc->lock, flags);
1058
1059 break;
1060 }
1061
1062 f2fs_bug_on(sbi, !bio);
1063
1064 /*
1065 * should keep before submission to avoid D_DONE
1066 * right away
1067 */
1068 spin_lock_irqsave(&dc->lock, flags);
1069 if (last)
1070 dc->state = D_SUBMIT;
1071 else
1072 dc->state = D_PARTIAL;
1073 dc->bio_ref++;
1074 spin_unlock_irqrestore(&dc->lock, flags);
1075
1076 atomic_inc(&dcc->issing_discard);
1077 dc->issuing++;
1078 list_move_tail(&dc->list, wait_list);
1079
1080 /* sanity check on discard range */
1081 __check_sit_bitmap(sbi, start, start + len);
1082
1083 bio->bi_private = dc;
1084 bio->bi_end_io = f2fs_submit_discard_endio;
1085 bio->bi_opf |= flag;
1086 submit_bio(bio);
1087
1088 atomic_inc(&dcc->issued_discard);
1089
1090 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1091
1092 lstart += len;
1093 start += len;
1094 total_len -= len;
1095 len = total_len;
1096 }
1097
1098 if (!err && len)
1099 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1100 return err;
1101 }
1102
1103 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1104 struct block_device *bdev, block_t lstart,
1105 block_t start, block_t len,
1106 struct rb_node **insert_p,
1107 struct rb_node *insert_parent)
1108 {
1109 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1110 struct rb_node **p;
1111 struct rb_node *parent = NULL;
1112 struct discard_cmd *dc = NULL;
1113
1114 if (insert_p && insert_parent) {
1115 parent = insert_parent;
1116 p = insert_p;
1117 goto do_insert;
1118 }
1119
1120 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
1121 do_insert:
1122 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1123 if (!dc)
1124 return NULL;
1125
1126 return dc;
1127 }
1128
1129 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1130 struct discard_cmd *dc)
1131 {
1132 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1133 }
1134
1135 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1136 struct discard_cmd *dc, block_t blkaddr)
1137 {
1138 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1139 struct discard_info di = dc->di;
1140 bool modified = false;
1141
1142 if (dc->state == D_DONE || dc->len == 1) {
1143 __remove_discard_cmd(sbi, dc);
1144 return;
1145 }
1146
1147 dcc->undiscard_blks -= di.len;
1148
1149 if (blkaddr > di.lstart) {
1150 dc->len = blkaddr - dc->lstart;
1151 dcc->undiscard_blks += dc->len;
1152 __relocate_discard_cmd(dcc, dc);
1153 modified = true;
1154 }
1155
1156 if (blkaddr < di.lstart + di.len - 1) {
1157 if (modified) {
1158 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1159 di.start + blkaddr + 1 - di.lstart,
1160 di.lstart + di.len - 1 - blkaddr,
1161 NULL, NULL);
1162 } else {
1163 dc->lstart++;
1164 dc->len--;
1165 dc->start++;
1166 dcc->undiscard_blks += dc->len;
1167 __relocate_discard_cmd(dcc, dc);
1168 }
1169 }
1170 }
1171
1172 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1173 struct block_device *bdev, block_t lstart,
1174 block_t start, block_t len)
1175 {
1176 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1177 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1178 struct discard_cmd *dc;
1179 struct discard_info di = {0};
1180 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1181 struct request_queue *q = bdev_get_queue(bdev);
1182 unsigned int max_discard_blocks =
1183 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1184 block_t end = lstart + len;
1185
1186 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1187 NULL, lstart,
1188 (struct rb_entry **)&prev_dc,
1189 (struct rb_entry **)&next_dc,
1190 &insert_p, &insert_parent, true);
1191 if (dc)
1192 prev_dc = dc;
1193
1194 if (!prev_dc) {
1195 di.lstart = lstart;
1196 di.len = next_dc ? next_dc->lstart - lstart : len;
1197 di.len = min(di.len, len);
1198 di.start = start;
1199 }
1200
1201 while (1) {
1202 struct rb_node *node;
1203 bool merged = false;
1204 struct discard_cmd *tdc = NULL;
1205
1206 if (prev_dc) {
1207 di.lstart = prev_dc->lstart + prev_dc->len;
1208 if (di.lstart < lstart)
1209 di.lstart = lstart;
1210 if (di.lstart >= end)
1211 break;
1212
1213 if (!next_dc || next_dc->lstart > end)
1214 di.len = end - di.lstart;
1215 else
1216 di.len = next_dc->lstart - di.lstart;
1217 di.start = start + di.lstart - lstart;
1218 }
1219
1220 if (!di.len)
1221 goto next;
1222
1223 if (prev_dc && prev_dc->state == D_PREP &&
1224 prev_dc->bdev == bdev &&
1225 __is_discard_back_mergeable(&di, &prev_dc->di,
1226 max_discard_blocks)) {
1227 prev_dc->di.len += di.len;
1228 dcc->undiscard_blks += di.len;
1229 __relocate_discard_cmd(dcc, prev_dc);
1230 di = prev_dc->di;
1231 tdc = prev_dc;
1232 merged = true;
1233 }
1234
1235 if (next_dc && next_dc->state == D_PREP &&
1236 next_dc->bdev == bdev &&
1237 __is_discard_front_mergeable(&di, &next_dc->di,
1238 max_discard_blocks)) {
1239 next_dc->di.lstart = di.lstart;
1240 next_dc->di.len += di.len;
1241 next_dc->di.start = di.start;
1242 dcc->undiscard_blks += di.len;
1243 __relocate_discard_cmd(dcc, next_dc);
1244 if (tdc)
1245 __remove_discard_cmd(sbi, tdc);
1246 merged = true;
1247 }
1248
1249 if (!merged) {
1250 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1251 di.len, NULL, NULL);
1252 }
1253 next:
1254 prev_dc = next_dc;
1255 if (!prev_dc)
1256 break;
1257
1258 node = rb_next(&prev_dc->rb_node);
1259 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1260 }
1261 }
1262
1263 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1264 struct block_device *bdev, block_t blkstart, block_t blklen)
1265 {
1266 block_t lblkstart = blkstart;
1267
1268 trace_f2fs_queue_discard(bdev, blkstart, blklen);
1269
1270 if (sbi->s_ndevs) {
1271 int devi = f2fs_target_device_index(sbi, blkstart);
1272
1273 blkstart -= FDEV(devi).start_blk;
1274 }
1275 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1276 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1277 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1278 return 0;
1279 }
1280
1281 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1282 struct discard_policy *dpolicy)
1283 {
1284 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1285 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1286 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1287 struct discard_cmd *dc;
1288 struct blk_plug plug;
1289 unsigned int pos = dcc->next_pos;
1290 unsigned int issued = 0;
1291 bool io_interrupted = false;
1292
1293 mutex_lock(&dcc->cmd_lock);
1294 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1295 NULL, pos,
1296 (struct rb_entry **)&prev_dc,
1297 (struct rb_entry **)&next_dc,
1298 &insert_p, &insert_parent, true);
1299 if (!dc)
1300 dc = next_dc;
1301
1302 blk_start_plug(&plug);
1303
1304 while (dc) {
1305 struct rb_node *node;
1306 int err = 0;
1307
1308 if (dc->state != D_PREP)
1309 goto next;
1310
1311 if (dpolicy->io_aware && !is_idle(sbi, DISCARD_TIME)) {
1312 io_interrupted = true;
1313 break;
1314 }
1315
1316 dcc->next_pos = dc->lstart + dc->len;
1317 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1318
1319 if (issued >= dpolicy->max_requests)
1320 break;
1321 next:
1322 node = rb_next(&dc->rb_node);
1323 if (err)
1324 __remove_discard_cmd(sbi, dc);
1325 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1326 }
1327
1328 blk_finish_plug(&plug);
1329
1330 if (!dc)
1331 dcc->next_pos = 0;
1332
1333 mutex_unlock(&dcc->cmd_lock);
1334
1335 if (!issued && io_interrupted)
1336 issued = -1;
1337
1338 return issued;
1339 }
1340
1341 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1342 struct discard_policy *dpolicy)
1343 {
1344 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1345 struct list_head *pend_list;
1346 struct discard_cmd *dc, *tmp;
1347 struct blk_plug plug;
1348 int i, issued = 0;
1349 bool io_interrupted = false;
1350
1351 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1352 if (i + 1 < dpolicy->granularity)
1353 break;
1354
1355 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1356 return __issue_discard_cmd_orderly(sbi, dpolicy);
1357
1358 pend_list = &dcc->pend_list[i];
1359
1360 mutex_lock(&dcc->cmd_lock);
1361 if (list_empty(pend_list))
1362 goto next;
1363 if (unlikely(dcc->rbtree_check))
1364 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1365 &dcc->root));
1366 blk_start_plug(&plug);
1367 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1368 f2fs_bug_on(sbi, dc->state != D_PREP);
1369
1370 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1371 !is_idle(sbi, DISCARD_TIME)) {
1372 io_interrupted = true;
1373 break;
1374 }
1375
1376 __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1377
1378 if (issued >= dpolicy->max_requests)
1379 break;
1380 }
1381 blk_finish_plug(&plug);
1382 next:
1383 mutex_unlock(&dcc->cmd_lock);
1384
1385 if (issued >= dpolicy->max_requests || io_interrupted)
1386 break;
1387 }
1388
1389 if (!issued && io_interrupted)
1390 issued = -1;
1391
1392 return issued;
1393 }
1394
1395 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1396 {
1397 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1398 struct list_head *pend_list;
1399 struct discard_cmd *dc, *tmp;
1400 int i;
1401 bool dropped = false;
1402
1403 mutex_lock(&dcc->cmd_lock);
1404 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1405 pend_list = &dcc->pend_list[i];
1406 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1407 f2fs_bug_on(sbi, dc->state != D_PREP);
1408 __remove_discard_cmd(sbi, dc);
1409 dropped = true;
1410 }
1411 }
1412 mutex_unlock(&dcc->cmd_lock);
1413
1414 return dropped;
1415 }
1416
1417 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1418 {
1419 __drop_discard_cmd(sbi);
1420 }
1421
1422 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1423 struct discard_cmd *dc)
1424 {
1425 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1426 unsigned int len = 0;
1427
1428 wait_for_completion_io(&dc->wait);
1429 mutex_lock(&dcc->cmd_lock);
1430 f2fs_bug_on(sbi, dc->state != D_DONE);
1431 dc->ref--;
1432 if (!dc->ref) {
1433 if (!dc->error)
1434 len = dc->len;
1435 __remove_discard_cmd(sbi, dc);
1436 }
1437 mutex_unlock(&dcc->cmd_lock);
1438
1439 return len;
1440 }
1441
1442 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1443 struct discard_policy *dpolicy,
1444 block_t start, block_t end)
1445 {
1446 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1447 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1448 &(dcc->fstrim_list) : &(dcc->wait_list);
1449 struct discard_cmd *dc, *tmp;
1450 bool need_wait;
1451 unsigned int trimmed = 0;
1452
1453 next:
1454 need_wait = false;
1455
1456 mutex_lock(&dcc->cmd_lock);
1457 list_for_each_entry_safe(dc, tmp, wait_list, list) {
1458 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1459 continue;
1460 if (dc->len < dpolicy->granularity)
1461 continue;
1462 if (dc->state == D_DONE && !dc->ref) {
1463 wait_for_completion_io(&dc->wait);
1464 if (!dc->error)
1465 trimmed += dc->len;
1466 __remove_discard_cmd(sbi, dc);
1467 } else {
1468 dc->ref++;
1469 need_wait = true;
1470 break;
1471 }
1472 }
1473 mutex_unlock(&dcc->cmd_lock);
1474
1475 if (need_wait) {
1476 trimmed += __wait_one_discard_bio(sbi, dc);
1477 goto next;
1478 }
1479
1480 return trimmed;
1481 }
1482
1483 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1484 struct discard_policy *dpolicy)
1485 {
1486 struct discard_policy dp;
1487 unsigned int discard_blks;
1488
1489 if (dpolicy)
1490 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1491
1492 /* wait all */
1493 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1494 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1495 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1496 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1497
1498 return discard_blks;
1499 }
1500
1501 /* This should be covered by global mutex, &sit_i->sentry_lock */
1502 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1503 {
1504 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1505 struct discard_cmd *dc;
1506 bool need_wait = false;
1507
1508 mutex_lock(&dcc->cmd_lock);
1509 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1510 NULL, blkaddr);
1511 if (dc) {
1512 if (dc->state == D_PREP) {
1513 __punch_discard_cmd(sbi, dc, blkaddr);
1514 } else {
1515 dc->ref++;
1516 need_wait = true;
1517 }
1518 }
1519 mutex_unlock(&dcc->cmd_lock);
1520
1521 if (need_wait)
1522 __wait_one_discard_bio(sbi, dc);
1523 }
1524
1525 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1526 {
1527 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1528
1529 if (dcc && dcc->f2fs_issue_discard) {
1530 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1531
1532 dcc->f2fs_issue_discard = NULL;
1533 kthread_stop(discard_thread);
1534 }
1535 }
1536
1537 /* This comes from f2fs_put_super */
1538 bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1539 {
1540 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1541 struct discard_policy dpolicy;
1542 bool dropped;
1543
1544 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1545 dcc->discard_granularity);
1546 __issue_discard_cmd(sbi, &dpolicy);
1547 dropped = __drop_discard_cmd(sbi);
1548
1549 /* just to make sure there is no pending discard commands */
1550 __wait_all_discard_cmd(sbi, NULL);
1551
1552 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1553 return dropped;
1554 }
1555
1556 static int issue_discard_thread(void *data)
1557 {
1558 struct f2fs_sb_info *sbi = data;
1559 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1560 wait_queue_head_t *q = &dcc->discard_wait_queue;
1561 struct discard_policy dpolicy;
1562 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1563 int issued;
1564
1565 set_freezable();
1566
1567 do {
1568 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1569 dcc->discard_granularity);
1570
1571 wait_event_interruptible_timeout(*q,
1572 kthread_should_stop() || freezing(current) ||
1573 dcc->discard_wake,
1574 msecs_to_jiffies(wait_ms));
1575
1576 if (dcc->discard_wake)
1577 dcc->discard_wake = 0;
1578
1579 if (try_to_freeze())
1580 continue;
1581 if (f2fs_readonly(sbi->sb))
1582 continue;
1583 if (kthread_should_stop())
1584 return 0;
1585 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1586 wait_ms = dpolicy.max_interval;
1587 continue;
1588 }
1589
1590 if (sbi->gc_mode == GC_URGENT)
1591 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1592
1593 sb_start_intwrite(sbi->sb);
1594
1595 issued = __issue_discard_cmd(sbi, &dpolicy);
1596 if (issued > 0) {
1597 __wait_all_discard_cmd(sbi, &dpolicy);
1598 wait_ms = dpolicy.min_interval;
1599 } else if (issued == -1){
1600 wait_ms = f2fs_time_to_wait(sbi, DISCARD_TIME);
1601 if (!wait_ms)
1602 wait_ms = dpolicy.mid_interval;
1603 } else {
1604 wait_ms = dpolicy.max_interval;
1605 }
1606
1607 sb_end_intwrite(sbi->sb);
1608
1609 } while (!kthread_should_stop());
1610 return 0;
1611 }
1612
1613 #ifdef CONFIG_BLK_DEV_ZONED
1614 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1615 struct block_device *bdev, block_t blkstart, block_t blklen)
1616 {
1617 sector_t sector, nr_sects;
1618 block_t lblkstart = blkstart;
1619 int devi = 0;
1620
1621 if (sbi->s_ndevs) {
1622 devi = f2fs_target_device_index(sbi, blkstart);
1623 blkstart -= FDEV(devi).start_blk;
1624 }
1625
1626 /*
1627 * We need to know the type of the zone: for conventional zones,
1628 * use regular discard if the drive supports it. For sequential
1629 * zones, reset the zone write pointer.
1630 */
1631 switch (get_blkz_type(sbi, bdev, blkstart)) {
1632
1633 case BLK_ZONE_TYPE_CONVENTIONAL:
1634 if (!blk_queue_discard(bdev_get_queue(bdev)))
1635 return 0;
1636 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1637 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1638 case BLK_ZONE_TYPE_SEQWRITE_PREF:
1639 sector = SECTOR_FROM_BLOCK(blkstart);
1640 nr_sects = SECTOR_FROM_BLOCK(blklen);
1641
1642 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1643 nr_sects != bdev_zone_sectors(bdev)) {
1644 f2fs_msg(sbi->sb, KERN_INFO,
1645 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1646 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1647 blkstart, blklen);
1648 return -EIO;
1649 }
1650 trace_f2fs_issue_reset_zone(bdev, blkstart);
1651 return blkdev_reset_zones(bdev, sector,
1652 nr_sects, GFP_NOFS);
1653 default:
1654 /* Unknown zone type: broken device ? */
1655 return -EIO;
1656 }
1657 }
1658 #endif
1659
1660 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1661 struct block_device *bdev, block_t blkstart, block_t blklen)
1662 {
1663 #ifdef CONFIG_BLK_DEV_ZONED
1664 if (f2fs_sb_has_blkzoned(sbi->sb) &&
1665 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1666 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1667 #endif
1668 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1669 }
1670
1671 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1672 block_t blkstart, block_t blklen)
1673 {
1674 sector_t start = blkstart, len = 0;
1675 struct block_device *bdev;
1676 struct seg_entry *se;
1677 unsigned int offset;
1678 block_t i;
1679 int err = 0;
1680
1681 bdev = f2fs_target_device(sbi, blkstart, NULL);
1682
1683 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1684 if (i != start) {
1685 struct block_device *bdev2 =
1686 f2fs_target_device(sbi, i, NULL);
1687
1688 if (bdev2 != bdev) {
1689 err = __issue_discard_async(sbi, bdev,
1690 start, len);
1691 if (err)
1692 return err;
1693 bdev = bdev2;
1694 start = i;
1695 len = 0;
1696 }
1697 }
1698
1699 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1700 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1701
1702 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1703 sbi->discard_blks--;
1704 }
1705
1706 if (len)
1707 err = __issue_discard_async(sbi, bdev, start, len);
1708 return err;
1709 }
1710
1711 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1712 bool check_only)
1713 {
1714 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1715 int max_blocks = sbi->blocks_per_seg;
1716 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1717 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1718 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1719 unsigned long *discard_map = (unsigned long *)se->discard_map;
1720 unsigned long *dmap = SIT_I(sbi)->tmp_map;
1721 unsigned int start = 0, end = -1;
1722 bool force = (cpc->reason & CP_DISCARD);
1723 struct discard_entry *de = NULL;
1724 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1725 int i;
1726
1727 if (se->valid_blocks == max_blocks || !f2fs_hw_support_discard(sbi))
1728 return false;
1729
1730 if (!force) {
1731 if (!f2fs_realtime_discard_enable(sbi) || !se->valid_blocks ||
1732 SM_I(sbi)->dcc_info->nr_discards >=
1733 SM_I(sbi)->dcc_info->max_discards)
1734 return false;
1735 }
1736
1737 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1738 for (i = 0; i < entries; i++)
1739 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1740 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1741
1742 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1743 SM_I(sbi)->dcc_info->max_discards) {
1744 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1745 if (start >= max_blocks)
1746 break;
1747
1748 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1749 if (force && start && end != max_blocks
1750 && (end - start) < cpc->trim_minlen)
1751 continue;
1752
1753 if (check_only)
1754 return true;
1755
1756 if (!de) {
1757 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1758 GFP_F2FS_ZERO);
1759 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1760 list_add_tail(&de->list, head);
1761 }
1762
1763 for (i = start; i < end; i++)
1764 __set_bit_le(i, (void *)de->discard_map);
1765
1766 SM_I(sbi)->dcc_info->nr_discards += end - start;
1767 }
1768 return false;
1769 }
1770
1771 static void release_discard_addr(struct discard_entry *entry)
1772 {
1773 list_del(&entry->list);
1774 kmem_cache_free(discard_entry_slab, entry);
1775 }
1776
1777 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1778 {
1779 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1780 struct discard_entry *entry, *this;
1781
1782 /* drop caches */
1783 list_for_each_entry_safe(entry, this, head, list)
1784 release_discard_addr(entry);
1785 }
1786
1787 /*
1788 * Should call f2fs_clear_prefree_segments after checkpoint is done.
1789 */
1790 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1791 {
1792 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1793 unsigned int segno;
1794
1795 mutex_lock(&dirty_i->seglist_lock);
1796 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1797 __set_test_and_free(sbi, segno);
1798 mutex_unlock(&dirty_i->seglist_lock);
1799 }
1800
1801 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1802 struct cp_control *cpc)
1803 {
1804 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1805 struct list_head *head = &dcc->entry_list;
1806 struct discard_entry *entry, *this;
1807 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1808 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1809 unsigned int start = 0, end = -1;
1810 unsigned int secno, start_segno;
1811 bool force = (cpc->reason & CP_DISCARD);
1812 bool need_align = test_opt(sbi, LFS) && sbi->segs_per_sec > 1;
1813
1814 mutex_lock(&dirty_i->seglist_lock);
1815
1816 while (1) {
1817 int i;
1818
1819 if (need_align && end != -1)
1820 end--;
1821 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1822 if (start >= MAIN_SEGS(sbi))
1823 break;
1824 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1825 start + 1);
1826
1827 if (need_align) {
1828 start = rounddown(start, sbi->segs_per_sec);
1829 end = roundup(end, sbi->segs_per_sec);
1830 }
1831
1832 for (i = start; i < end; i++) {
1833 if (test_and_clear_bit(i, prefree_map))
1834 dirty_i->nr_dirty[PRE]--;
1835 }
1836
1837 if (!f2fs_realtime_discard_enable(sbi))
1838 continue;
1839
1840 if (force && start >= cpc->trim_start &&
1841 (end - 1) <= cpc->trim_end)
1842 continue;
1843
1844 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1845 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1846 (end - start) << sbi->log_blocks_per_seg);
1847 continue;
1848 }
1849 next:
1850 secno = GET_SEC_FROM_SEG(sbi, start);
1851 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1852 if (!IS_CURSEC(sbi, secno) &&
1853 !get_valid_blocks(sbi, start, true))
1854 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1855 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1856
1857 start = start_segno + sbi->segs_per_sec;
1858 if (start < end)
1859 goto next;
1860 else
1861 end = start - 1;
1862 }
1863 mutex_unlock(&dirty_i->seglist_lock);
1864
1865 /* send small discards */
1866 list_for_each_entry_safe(entry, this, head, list) {
1867 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1868 bool is_valid = test_bit_le(0, entry->discard_map);
1869
1870 find_next:
1871 if (is_valid) {
1872 next_pos = find_next_zero_bit_le(entry->discard_map,
1873 sbi->blocks_per_seg, cur_pos);
1874 len = next_pos - cur_pos;
1875
1876 if (f2fs_sb_has_blkzoned(sbi->sb) ||
1877 (force && len < cpc->trim_minlen))
1878 goto skip;
1879
1880 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1881 len);
1882 total_len += len;
1883 } else {
1884 next_pos = find_next_bit_le(entry->discard_map,
1885 sbi->blocks_per_seg, cur_pos);
1886 }
1887 skip:
1888 cur_pos = next_pos;
1889 is_valid = !is_valid;
1890
1891 if (cur_pos < sbi->blocks_per_seg)
1892 goto find_next;
1893
1894 release_discard_addr(entry);
1895 dcc->nr_discards -= total_len;
1896 }
1897
1898 wake_up_discard_thread(sbi, false);
1899 }
1900
1901 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1902 {
1903 dev_t dev = sbi->sb->s_bdev->bd_dev;
1904 struct discard_cmd_control *dcc;
1905 int err = 0, i;
1906
1907 if (SM_I(sbi)->dcc_info) {
1908 dcc = SM_I(sbi)->dcc_info;
1909 goto init_thread;
1910 }
1911
1912 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
1913 if (!dcc)
1914 return -ENOMEM;
1915
1916 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
1917 INIT_LIST_HEAD(&dcc->entry_list);
1918 for (i = 0; i < MAX_PLIST_NUM; i++)
1919 INIT_LIST_HEAD(&dcc->pend_list[i]);
1920 INIT_LIST_HEAD(&dcc->wait_list);
1921 INIT_LIST_HEAD(&dcc->fstrim_list);
1922 mutex_init(&dcc->cmd_lock);
1923 atomic_set(&dcc->issued_discard, 0);
1924 atomic_set(&dcc->issing_discard, 0);
1925 atomic_set(&dcc->discard_cmd_cnt, 0);
1926 dcc->nr_discards = 0;
1927 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1928 dcc->undiscard_blks = 0;
1929 dcc->next_pos = 0;
1930 dcc->root = RB_ROOT;
1931 dcc->rbtree_check = false;
1932
1933 init_waitqueue_head(&dcc->discard_wait_queue);
1934 SM_I(sbi)->dcc_info = dcc;
1935 init_thread:
1936 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1937 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1938 if (IS_ERR(dcc->f2fs_issue_discard)) {
1939 err = PTR_ERR(dcc->f2fs_issue_discard);
1940 kfree(dcc);
1941 SM_I(sbi)->dcc_info = NULL;
1942 return err;
1943 }
1944
1945 return err;
1946 }
1947
1948 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1949 {
1950 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1951
1952 if (!dcc)
1953 return;
1954
1955 f2fs_stop_discard_thread(sbi);
1956
1957 kfree(dcc);
1958 SM_I(sbi)->dcc_info = NULL;
1959 }
1960
1961 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1962 {
1963 struct sit_info *sit_i = SIT_I(sbi);
1964
1965 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1966 sit_i->dirty_sentries++;
1967 return false;
1968 }
1969
1970 return true;
1971 }
1972
1973 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1974 unsigned int segno, int modified)
1975 {
1976 struct seg_entry *se = get_seg_entry(sbi, segno);
1977 se->type = type;
1978 if (modified)
1979 __mark_sit_entry_dirty(sbi, segno);
1980 }
1981
1982 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1983 {
1984 struct seg_entry *se;
1985 unsigned int segno, offset;
1986 long int new_vblocks;
1987 bool exist;
1988 #ifdef CONFIG_F2FS_CHECK_FS
1989 bool mir_exist;
1990 #endif
1991
1992 segno = GET_SEGNO(sbi, blkaddr);
1993
1994 se = get_seg_entry(sbi, segno);
1995 new_vblocks = se->valid_blocks + del;
1996 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1997
1998 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1999 (new_vblocks > sbi->blocks_per_seg)));
2000
2001 se->valid_blocks = new_vblocks;
2002 se->mtime = get_mtime(sbi, false);
2003 if (se->mtime > SIT_I(sbi)->max_mtime)
2004 SIT_I(sbi)->max_mtime = se->mtime;
2005
2006 /* Update valid block bitmap */
2007 if (del > 0) {
2008 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2009 #ifdef CONFIG_F2FS_CHECK_FS
2010 mir_exist = f2fs_test_and_set_bit(offset,
2011 se->cur_valid_map_mir);
2012 if (unlikely(exist != mir_exist)) {
2013 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2014 "when setting bitmap, blk:%u, old bit:%d",
2015 blkaddr, exist);
2016 f2fs_bug_on(sbi, 1);
2017 }
2018 #endif
2019 if (unlikely(exist)) {
2020 f2fs_msg(sbi->sb, KERN_ERR,
2021 "Bitmap was wrongly set, blk:%u", blkaddr);
2022 f2fs_bug_on(sbi, 1);
2023 se->valid_blocks--;
2024 del = 0;
2025 }
2026
2027 if (!f2fs_test_and_set_bit(offset, se->discard_map))
2028 sbi->discard_blks--;
2029
2030 /* don't overwrite by SSR to keep node chain */
2031 if (IS_NODESEG(se->type)) {
2032 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2033 se->ckpt_valid_blocks++;
2034 }
2035 } else {
2036 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2037 #ifdef CONFIG_F2FS_CHECK_FS
2038 mir_exist = f2fs_test_and_clear_bit(offset,
2039 se->cur_valid_map_mir);
2040 if (unlikely(exist != mir_exist)) {
2041 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2042 "when clearing bitmap, blk:%u, old bit:%d",
2043 blkaddr, exist);
2044 f2fs_bug_on(sbi, 1);
2045 }
2046 #endif
2047 if (unlikely(!exist)) {
2048 f2fs_msg(sbi->sb, KERN_ERR,
2049 "Bitmap was wrongly cleared, blk:%u", blkaddr);
2050 f2fs_bug_on(sbi, 1);
2051 se->valid_blocks++;
2052 del = 0;
2053 }
2054
2055 if (f2fs_test_and_clear_bit(offset, se->discard_map))
2056 sbi->discard_blks++;
2057 }
2058 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2059 se->ckpt_valid_blocks += del;
2060
2061 __mark_sit_entry_dirty(sbi, segno);
2062
2063 /* update total number of valid blocks to be written in ckpt area */
2064 SIT_I(sbi)->written_valid_blocks += del;
2065
2066 if (sbi->segs_per_sec > 1)
2067 get_sec_entry(sbi, segno)->valid_blocks += del;
2068 }
2069
2070 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2071 {
2072 unsigned int segno = GET_SEGNO(sbi, addr);
2073 struct sit_info *sit_i = SIT_I(sbi);
2074
2075 f2fs_bug_on(sbi, addr == NULL_ADDR);
2076 if (addr == NEW_ADDR)
2077 return;
2078
2079 invalidate_mapping_pages(META_MAPPING(sbi), addr, addr);
2080
2081 /* add it into sit main buffer */
2082 down_write(&sit_i->sentry_lock);
2083
2084 update_sit_entry(sbi, addr, -1);
2085
2086 /* add it into dirty seglist */
2087 locate_dirty_segment(sbi, segno);
2088
2089 up_write(&sit_i->sentry_lock);
2090 }
2091
2092 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2093 {
2094 struct sit_info *sit_i = SIT_I(sbi);
2095 unsigned int segno, offset;
2096 struct seg_entry *se;
2097 bool is_cp = false;
2098
2099 if (!is_valid_data_blkaddr(sbi, blkaddr))
2100 return true;
2101
2102 down_read(&sit_i->sentry_lock);
2103
2104 segno = GET_SEGNO(sbi, blkaddr);
2105 se = get_seg_entry(sbi, segno);
2106 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2107
2108 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2109 is_cp = true;
2110
2111 up_read(&sit_i->sentry_lock);
2112
2113 return is_cp;
2114 }
2115
2116 /*
2117 * This function should be resided under the curseg_mutex lock
2118 */
2119 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2120 struct f2fs_summary *sum)
2121 {
2122 struct curseg_info *curseg = CURSEG_I(sbi, type);
2123 void *addr = curseg->sum_blk;
2124 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2125 memcpy(addr, sum, sizeof(struct f2fs_summary));
2126 }
2127
2128 /*
2129 * Calculate the number of current summary pages for writing
2130 */
2131 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2132 {
2133 int valid_sum_count = 0;
2134 int i, sum_in_page;
2135
2136 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2137 if (sbi->ckpt->alloc_type[i] == SSR)
2138 valid_sum_count += sbi->blocks_per_seg;
2139 else {
2140 if (for_ra)
2141 valid_sum_count += le16_to_cpu(
2142 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2143 else
2144 valid_sum_count += curseg_blkoff(sbi, i);
2145 }
2146 }
2147
2148 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2149 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2150 if (valid_sum_count <= sum_in_page)
2151 return 1;
2152 else if ((valid_sum_count - sum_in_page) <=
2153 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2154 return 2;
2155 return 3;
2156 }
2157
2158 /*
2159 * Caller should put this summary page
2160 */
2161 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2162 {
2163 return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2164 }
2165
2166 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2167 void *src, block_t blk_addr)
2168 {
2169 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2170
2171 memcpy(page_address(page), src, PAGE_SIZE);
2172 set_page_dirty(page);
2173 f2fs_put_page(page, 1);
2174 }
2175
2176 static void write_sum_page(struct f2fs_sb_info *sbi,
2177 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2178 {
2179 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2180 }
2181
2182 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2183 int type, block_t blk_addr)
2184 {
2185 struct curseg_info *curseg = CURSEG_I(sbi, type);
2186 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2187 struct f2fs_summary_block *src = curseg->sum_blk;
2188 struct f2fs_summary_block *dst;
2189
2190 dst = (struct f2fs_summary_block *)page_address(page);
2191 memset(dst, 0, PAGE_SIZE);
2192
2193 mutex_lock(&curseg->curseg_mutex);
2194
2195 down_read(&curseg->journal_rwsem);
2196 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2197 up_read(&curseg->journal_rwsem);
2198
2199 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2200 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2201
2202 mutex_unlock(&curseg->curseg_mutex);
2203
2204 set_page_dirty(page);
2205 f2fs_put_page(page, 1);
2206 }
2207
2208 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2209 {
2210 struct curseg_info *curseg = CURSEG_I(sbi, type);
2211 unsigned int segno = curseg->segno + 1;
2212 struct free_segmap_info *free_i = FREE_I(sbi);
2213
2214 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2215 return !test_bit(segno, free_i->free_segmap);
2216 return 0;
2217 }
2218
2219 /*
2220 * Find a new segment from the free segments bitmap to right order
2221 * This function should be returned with success, otherwise BUG
2222 */
2223 static void get_new_segment(struct f2fs_sb_info *sbi,
2224 unsigned int *newseg, bool new_sec, int dir)
2225 {
2226 struct free_segmap_info *free_i = FREE_I(sbi);
2227 unsigned int segno, secno, zoneno;
2228 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2229 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2230 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2231 unsigned int left_start = hint;
2232 bool init = true;
2233 int go_left = 0;
2234 int i;
2235
2236 spin_lock(&free_i->segmap_lock);
2237
2238 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2239 segno = find_next_zero_bit(free_i->free_segmap,
2240 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2241 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2242 goto got_it;
2243 }
2244 find_other_zone:
2245 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2246 if (secno >= MAIN_SECS(sbi)) {
2247 if (dir == ALLOC_RIGHT) {
2248 secno = find_next_zero_bit(free_i->free_secmap,
2249 MAIN_SECS(sbi), 0);
2250 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2251 } else {
2252 go_left = 1;
2253 left_start = hint - 1;
2254 }
2255 }
2256 if (go_left == 0)
2257 goto skip_left;
2258
2259 while (test_bit(left_start, free_i->free_secmap)) {
2260 if (left_start > 0) {
2261 left_start--;
2262 continue;
2263 }
2264 left_start = find_next_zero_bit(free_i->free_secmap,
2265 MAIN_SECS(sbi), 0);
2266 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2267 break;
2268 }
2269 secno = left_start;
2270 skip_left:
2271 segno = GET_SEG_FROM_SEC(sbi, secno);
2272 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2273
2274 /* give up on finding another zone */
2275 if (!init)
2276 goto got_it;
2277 if (sbi->secs_per_zone == 1)
2278 goto got_it;
2279 if (zoneno == old_zoneno)
2280 goto got_it;
2281 if (dir == ALLOC_LEFT) {
2282 if (!go_left && zoneno + 1 >= total_zones)
2283 goto got_it;
2284 if (go_left && zoneno == 0)
2285 goto got_it;
2286 }
2287 for (i = 0; i < NR_CURSEG_TYPE; i++)
2288 if (CURSEG_I(sbi, i)->zone == zoneno)
2289 break;
2290
2291 if (i < NR_CURSEG_TYPE) {
2292 /* zone is in user, try another */
2293 if (go_left)
2294 hint = zoneno * sbi->secs_per_zone - 1;
2295 else if (zoneno + 1 >= total_zones)
2296 hint = 0;
2297 else
2298 hint = (zoneno + 1) * sbi->secs_per_zone;
2299 init = false;
2300 goto find_other_zone;
2301 }
2302 got_it:
2303 /* set it as dirty segment in free segmap */
2304 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2305 __set_inuse(sbi, segno);
2306 *newseg = segno;
2307 spin_unlock(&free_i->segmap_lock);
2308 }
2309
2310 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2311 {
2312 struct curseg_info *curseg = CURSEG_I(sbi, type);
2313 struct summary_footer *sum_footer;
2314
2315 curseg->segno = curseg->next_segno;
2316 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2317 curseg->next_blkoff = 0;
2318 curseg->next_segno = NULL_SEGNO;
2319
2320 sum_footer = &(curseg->sum_blk->footer);
2321 memset(sum_footer, 0, sizeof(struct summary_footer));
2322 if (IS_DATASEG(type))
2323 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2324 if (IS_NODESEG(type))
2325 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2326 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2327 }
2328
2329 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2330 {
2331 /* if segs_per_sec is large than 1, we need to keep original policy. */
2332 if (sbi->segs_per_sec != 1)
2333 return CURSEG_I(sbi, type)->segno;
2334
2335 if (test_opt(sbi, NOHEAP) &&
2336 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2337 return 0;
2338
2339 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2340 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2341
2342 /* find segments from 0 to reuse freed segments */
2343 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2344 return 0;
2345
2346 return CURSEG_I(sbi, type)->segno;
2347 }
2348
2349 /*
2350 * Allocate a current working segment.
2351 * This function always allocates a free segment in LFS manner.
2352 */
2353 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2354 {
2355 struct curseg_info *curseg = CURSEG_I(sbi, type);
2356 unsigned int segno = curseg->segno;
2357 int dir = ALLOC_LEFT;
2358
2359 write_sum_page(sbi, curseg->sum_blk,
2360 GET_SUM_BLOCK(sbi, segno));
2361 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2362 dir = ALLOC_RIGHT;
2363
2364 if (test_opt(sbi, NOHEAP))
2365 dir = ALLOC_RIGHT;
2366
2367 segno = __get_next_segno(sbi, type);
2368 get_new_segment(sbi, &segno, new_sec, dir);
2369 curseg->next_segno = segno;
2370 reset_curseg(sbi, type, 1);
2371 curseg->alloc_type = LFS;
2372 }
2373
2374 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2375 struct curseg_info *seg, block_t start)
2376 {
2377 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2378 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2379 unsigned long *target_map = SIT_I(sbi)->tmp_map;
2380 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2381 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2382 int i, pos;
2383
2384 for (i = 0; i < entries; i++)
2385 target_map[i] = ckpt_map[i] | cur_map[i];
2386
2387 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2388
2389 seg->next_blkoff = pos;
2390 }
2391
2392 /*
2393 * If a segment is written by LFS manner, next block offset is just obtained
2394 * by increasing the current block offset. However, if a segment is written by
2395 * SSR manner, next block offset obtained by calling __next_free_blkoff
2396 */
2397 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2398 struct curseg_info *seg)
2399 {
2400 if (seg->alloc_type == SSR)
2401 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2402 else
2403 seg->next_blkoff++;
2404 }
2405
2406 /*
2407 * This function always allocates a used segment(from dirty seglist) by SSR
2408 * manner, so it should recover the existing segment information of valid blocks
2409 */
2410 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2411 {
2412 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2413 struct curseg_info *curseg = CURSEG_I(sbi, type);
2414 unsigned int new_segno = curseg->next_segno;
2415 struct f2fs_summary_block *sum_node;
2416 struct page *sum_page;
2417
2418 write_sum_page(sbi, curseg->sum_blk,
2419 GET_SUM_BLOCK(sbi, curseg->segno));
2420 __set_test_and_inuse(sbi, new_segno);
2421
2422 mutex_lock(&dirty_i->seglist_lock);
2423 __remove_dirty_segment(sbi, new_segno, PRE);
2424 __remove_dirty_segment(sbi, new_segno, DIRTY);
2425 mutex_unlock(&dirty_i->seglist_lock);
2426
2427 reset_curseg(sbi, type, 1);
2428 curseg->alloc_type = SSR;
2429 __next_free_blkoff(sbi, curseg, 0);
2430
2431 sum_page = f2fs_get_sum_page(sbi, new_segno);
2432 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2433 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2434 f2fs_put_page(sum_page, 1);
2435 }
2436
2437 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2438 {
2439 struct curseg_info *curseg = CURSEG_I(sbi, type);
2440 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2441 unsigned segno = NULL_SEGNO;
2442 int i, cnt;
2443 bool reversed = false;
2444
2445 /* f2fs_need_SSR() already forces to do this */
2446 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2447 curseg->next_segno = segno;
2448 return 1;
2449 }
2450
2451 /* For node segments, let's do SSR more intensively */
2452 if (IS_NODESEG(type)) {
2453 if (type >= CURSEG_WARM_NODE) {
2454 reversed = true;
2455 i = CURSEG_COLD_NODE;
2456 } else {
2457 i = CURSEG_HOT_NODE;
2458 }
2459 cnt = NR_CURSEG_NODE_TYPE;
2460 } else {
2461 if (type >= CURSEG_WARM_DATA) {
2462 reversed = true;
2463 i = CURSEG_COLD_DATA;
2464 } else {
2465 i = CURSEG_HOT_DATA;
2466 }
2467 cnt = NR_CURSEG_DATA_TYPE;
2468 }
2469
2470 for (; cnt-- > 0; reversed ? i-- : i++) {
2471 if (i == type)
2472 continue;
2473 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2474 curseg->next_segno = segno;
2475 return 1;
2476 }
2477 }
2478 return 0;
2479 }
2480
2481 /*
2482 * flush out current segment and replace it with new segment
2483 * This function should be returned with success, otherwise BUG
2484 */
2485 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2486 int type, bool force)
2487 {
2488 struct curseg_info *curseg = CURSEG_I(sbi, type);
2489
2490 if (force)
2491 new_curseg(sbi, type, true);
2492 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2493 type == CURSEG_WARM_NODE)
2494 new_curseg(sbi, type, false);
2495 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2496 new_curseg(sbi, type, false);
2497 else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2498 change_curseg(sbi, type);
2499 else
2500 new_curseg(sbi, type, false);
2501
2502 stat_inc_seg_type(sbi, curseg);
2503 }
2504
2505 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2506 {
2507 struct curseg_info *curseg;
2508 unsigned int old_segno;
2509 int i;
2510
2511 down_write(&SIT_I(sbi)->sentry_lock);
2512
2513 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2514 curseg = CURSEG_I(sbi, i);
2515 old_segno = curseg->segno;
2516 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2517 locate_dirty_segment(sbi, old_segno);
2518 }
2519
2520 up_write(&SIT_I(sbi)->sentry_lock);
2521 }
2522
2523 static const struct segment_allocation default_salloc_ops = {
2524 .allocate_segment = allocate_segment_by_default,
2525 };
2526
2527 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2528 struct cp_control *cpc)
2529 {
2530 __u64 trim_start = cpc->trim_start;
2531 bool has_candidate = false;
2532
2533 down_write(&SIT_I(sbi)->sentry_lock);
2534 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2535 if (add_discard_addrs(sbi, cpc, true)) {
2536 has_candidate = true;
2537 break;
2538 }
2539 }
2540 up_write(&SIT_I(sbi)->sentry_lock);
2541
2542 cpc->trim_start = trim_start;
2543 return has_candidate;
2544 }
2545
2546 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2547 struct discard_policy *dpolicy,
2548 unsigned int start, unsigned int end)
2549 {
2550 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2551 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2552 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2553 struct discard_cmd *dc;
2554 struct blk_plug plug;
2555 int issued;
2556 unsigned int trimmed = 0;
2557
2558 next:
2559 issued = 0;
2560
2561 mutex_lock(&dcc->cmd_lock);
2562 if (unlikely(dcc->rbtree_check))
2563 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2564 &dcc->root));
2565
2566 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2567 NULL, start,
2568 (struct rb_entry **)&prev_dc,
2569 (struct rb_entry **)&next_dc,
2570 &insert_p, &insert_parent, true);
2571 if (!dc)
2572 dc = next_dc;
2573
2574 blk_start_plug(&plug);
2575
2576 while (dc && dc->lstart <= end) {
2577 struct rb_node *node;
2578 int err = 0;
2579
2580 if (dc->len < dpolicy->granularity)
2581 goto skip;
2582
2583 if (dc->state != D_PREP) {
2584 list_move_tail(&dc->list, &dcc->fstrim_list);
2585 goto skip;
2586 }
2587
2588 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2589
2590 if (issued >= dpolicy->max_requests) {
2591 start = dc->lstart + dc->len;
2592
2593 if (err)
2594 __remove_discard_cmd(sbi, dc);
2595
2596 blk_finish_plug(&plug);
2597 mutex_unlock(&dcc->cmd_lock);
2598 trimmed += __wait_all_discard_cmd(sbi, NULL);
2599 congestion_wait(BLK_RW_ASYNC, HZ/50);
2600 goto next;
2601 }
2602 skip:
2603 node = rb_next(&dc->rb_node);
2604 if (err)
2605 __remove_discard_cmd(sbi, dc);
2606 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2607
2608 if (fatal_signal_pending(current))
2609 break;
2610 }
2611
2612 blk_finish_plug(&plug);
2613 mutex_unlock(&dcc->cmd_lock);
2614
2615 return trimmed;
2616 }
2617
2618 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2619 {
2620 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2621 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2622 unsigned int start_segno, end_segno;
2623 block_t start_block, end_block;
2624 struct cp_control cpc;
2625 struct discard_policy dpolicy;
2626 unsigned long long trimmed = 0;
2627 int err = 0;
2628 bool need_align = test_opt(sbi, LFS) && sbi->segs_per_sec > 1;
2629
2630 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2631 return -EINVAL;
2632
2633 if (end < MAIN_BLKADDR(sbi))
2634 goto out;
2635
2636 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2637 f2fs_msg(sbi->sb, KERN_WARNING,
2638 "Found FS corruption, run fsck to fix.");
2639 return -EIO;
2640 }
2641
2642 /* start/end segment number in main_area */
2643 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2644 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2645 GET_SEGNO(sbi, end);
2646 if (need_align) {
2647 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2648 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2649 }
2650
2651 cpc.reason = CP_DISCARD;
2652 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2653 cpc.trim_start = start_segno;
2654 cpc.trim_end = end_segno;
2655
2656 if (sbi->discard_blks == 0)
2657 goto out;
2658
2659 mutex_lock(&sbi->gc_mutex);
2660 err = f2fs_write_checkpoint(sbi, &cpc);
2661 mutex_unlock(&sbi->gc_mutex);
2662 if (err)
2663 goto out;
2664
2665 /*
2666 * We filed discard candidates, but actually we don't need to wait for
2667 * all of them, since they'll be issued in idle time along with runtime
2668 * discard option. User configuration looks like using runtime discard
2669 * or periodic fstrim instead of it.
2670 */
2671 if (f2fs_realtime_discard_enable(sbi))
2672 goto out;
2673
2674 start_block = START_BLOCK(sbi, start_segno);
2675 end_block = START_BLOCK(sbi, end_segno + 1);
2676
2677 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2678 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2679 start_block, end_block);
2680
2681 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2682 start_block, end_block);
2683 out:
2684 if (!err)
2685 range->len = F2FS_BLK_TO_BYTES(trimmed);
2686 return err;
2687 }
2688
2689 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2690 {
2691 struct curseg_info *curseg = CURSEG_I(sbi, type);
2692 if (curseg->next_blkoff < sbi->blocks_per_seg)
2693 return true;
2694 return false;
2695 }
2696
2697 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2698 {
2699 switch (hint) {
2700 case WRITE_LIFE_SHORT:
2701 return CURSEG_HOT_DATA;
2702 case WRITE_LIFE_EXTREME:
2703 return CURSEG_COLD_DATA;
2704 default:
2705 return CURSEG_WARM_DATA;
2706 }
2707 }
2708
2709 /* This returns write hints for each segment type. This hints will be
2710 * passed down to block layer. There are mapping tables which depend on
2711 * the mount option 'whint_mode'.
2712 *
2713 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2714 *
2715 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2716 *
2717 * User F2FS Block
2718 * ---- ---- -----
2719 * META WRITE_LIFE_NOT_SET
2720 * HOT_NODE "
2721 * WARM_NODE "
2722 * COLD_NODE "
2723 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2724 * extension list " "
2725 *
2726 * -- buffered io
2727 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2728 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2729 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2730 * WRITE_LIFE_NONE " "
2731 * WRITE_LIFE_MEDIUM " "
2732 * WRITE_LIFE_LONG " "
2733 *
2734 * -- direct io
2735 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2736 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2737 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2738 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2739 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2740 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2741 *
2742 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2743 *
2744 * User F2FS Block
2745 * ---- ---- -----
2746 * META WRITE_LIFE_MEDIUM;
2747 * HOT_NODE WRITE_LIFE_NOT_SET
2748 * WARM_NODE "
2749 * COLD_NODE WRITE_LIFE_NONE
2750 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2751 * extension list " "
2752 *
2753 * -- buffered io
2754 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2755 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2756 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
2757 * WRITE_LIFE_NONE " "
2758 * WRITE_LIFE_MEDIUM " "
2759 * WRITE_LIFE_LONG " "
2760 *
2761 * -- direct io
2762 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2763 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2764 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2765 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2766 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2767 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2768 */
2769
2770 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2771 enum page_type type, enum temp_type temp)
2772 {
2773 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2774 if (type == DATA) {
2775 if (temp == WARM)
2776 return WRITE_LIFE_NOT_SET;
2777 else if (temp == HOT)
2778 return WRITE_LIFE_SHORT;
2779 else if (temp == COLD)
2780 return WRITE_LIFE_EXTREME;
2781 } else {
2782 return WRITE_LIFE_NOT_SET;
2783 }
2784 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2785 if (type == DATA) {
2786 if (temp == WARM)
2787 return WRITE_LIFE_LONG;
2788 else if (temp == HOT)
2789 return WRITE_LIFE_SHORT;
2790 else if (temp == COLD)
2791 return WRITE_LIFE_EXTREME;
2792 } else if (type == NODE) {
2793 if (temp == WARM || temp == HOT)
2794 return WRITE_LIFE_NOT_SET;
2795 else if (temp == COLD)
2796 return WRITE_LIFE_NONE;
2797 } else if (type == META) {
2798 return WRITE_LIFE_MEDIUM;
2799 }
2800 }
2801 return WRITE_LIFE_NOT_SET;
2802 }
2803
2804 static int __get_segment_type_2(struct f2fs_io_info *fio)
2805 {
2806 if (fio->type == DATA)
2807 return CURSEG_HOT_DATA;
2808 else
2809 return CURSEG_HOT_NODE;
2810 }
2811
2812 static int __get_segment_type_4(struct f2fs_io_info *fio)
2813 {
2814 if (fio->type == DATA) {
2815 struct inode *inode = fio->page->mapping->host;
2816
2817 if (S_ISDIR(inode->i_mode))
2818 return CURSEG_HOT_DATA;
2819 else
2820 return CURSEG_COLD_DATA;
2821 } else {
2822 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2823 return CURSEG_WARM_NODE;
2824 else
2825 return CURSEG_COLD_NODE;
2826 }
2827 }
2828
2829 static int __get_segment_type_6(struct f2fs_io_info *fio)
2830 {
2831 if (fio->type == DATA) {
2832 struct inode *inode = fio->page->mapping->host;
2833
2834 if (is_cold_data(fio->page) || file_is_cold(inode))
2835 return CURSEG_COLD_DATA;
2836 if (file_is_hot(inode) ||
2837 is_inode_flag_set(inode, FI_HOT_DATA) ||
2838 f2fs_is_atomic_file(inode) ||
2839 f2fs_is_volatile_file(inode))
2840 return CURSEG_HOT_DATA;
2841 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
2842 } else {
2843 if (IS_DNODE(fio->page))
2844 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2845 CURSEG_HOT_NODE;
2846 return CURSEG_COLD_NODE;
2847 }
2848 }
2849
2850 static int __get_segment_type(struct f2fs_io_info *fio)
2851 {
2852 int type = 0;
2853
2854 switch (F2FS_OPTION(fio->sbi).active_logs) {
2855 case 2:
2856 type = __get_segment_type_2(fio);
2857 break;
2858 case 4:
2859 type = __get_segment_type_4(fio);
2860 break;
2861 case 6:
2862 type = __get_segment_type_6(fio);
2863 break;
2864 default:
2865 f2fs_bug_on(fio->sbi, true);
2866 }
2867
2868 if (IS_HOT(type))
2869 fio->temp = HOT;
2870 else if (IS_WARM(type))
2871 fio->temp = WARM;
2872 else
2873 fio->temp = COLD;
2874 return type;
2875 }
2876
2877 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2878 block_t old_blkaddr, block_t *new_blkaddr,
2879 struct f2fs_summary *sum, int type,
2880 struct f2fs_io_info *fio, bool add_list)
2881 {
2882 struct sit_info *sit_i = SIT_I(sbi);
2883 struct curseg_info *curseg = CURSEG_I(sbi, type);
2884
2885 down_read(&SM_I(sbi)->curseg_lock);
2886
2887 mutex_lock(&curseg->curseg_mutex);
2888 down_write(&sit_i->sentry_lock);
2889
2890 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2891
2892 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2893
2894 /*
2895 * __add_sum_entry should be resided under the curseg_mutex
2896 * because, this function updates a summary entry in the
2897 * current summary block.
2898 */
2899 __add_sum_entry(sbi, type, sum);
2900
2901 __refresh_next_blkoff(sbi, curseg);
2902
2903 stat_inc_block_count(sbi, curseg);
2904
2905 /*
2906 * SIT information should be updated before segment allocation,
2907 * since SSR needs latest valid block information.
2908 */
2909 update_sit_entry(sbi, *new_blkaddr, 1);
2910 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2911 update_sit_entry(sbi, old_blkaddr, -1);
2912
2913 if (!__has_curseg_space(sbi, type))
2914 sit_i->s_ops->allocate_segment(sbi, type, false);
2915
2916 /*
2917 * segment dirty status should be updated after segment allocation,
2918 * so we just need to update status only one time after previous
2919 * segment being closed.
2920 */
2921 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2922 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
2923
2924 up_write(&sit_i->sentry_lock);
2925
2926 if (page && IS_NODESEG(type)) {
2927 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2928
2929 f2fs_inode_chksum_set(sbi, page);
2930 }
2931
2932 if (add_list) {
2933 struct f2fs_bio_info *io;
2934
2935 INIT_LIST_HEAD(&fio->list);
2936 fio->in_list = true;
2937 fio->retry = false;
2938 io = sbi->write_io[fio->type] + fio->temp;
2939 spin_lock(&io->io_lock);
2940 list_add_tail(&fio->list, &io->io_list);
2941 spin_unlock(&io->io_lock);
2942 }
2943
2944 mutex_unlock(&curseg->curseg_mutex);
2945
2946 up_read(&SM_I(sbi)->curseg_lock);
2947 }
2948
2949 static void update_device_state(struct f2fs_io_info *fio)
2950 {
2951 struct f2fs_sb_info *sbi = fio->sbi;
2952 unsigned int devidx;
2953
2954 if (!sbi->s_ndevs)
2955 return;
2956
2957 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2958
2959 /* update device state for fsync */
2960 f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
2961
2962 /* update device state for checkpoint */
2963 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2964 spin_lock(&sbi->dev_lock);
2965 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2966 spin_unlock(&sbi->dev_lock);
2967 }
2968 }
2969
2970 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2971 {
2972 int type = __get_segment_type(fio);
2973 bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
2974
2975 if (keep_order)
2976 down_read(&fio->sbi->io_order_lock);
2977 reallocate:
2978 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2979 &fio->new_blkaddr, sum, type, fio, true);
2980 if (GET_SEGNO(fio->sbi, fio->old_blkaddr) != NULL_SEGNO)
2981 invalidate_mapping_pages(META_MAPPING(fio->sbi),
2982 fio->old_blkaddr, fio->old_blkaddr);
2983
2984 /* writeout dirty page into bdev */
2985 f2fs_submit_page_write(fio);
2986 if (fio->retry) {
2987 fio->old_blkaddr = fio->new_blkaddr;
2988 goto reallocate;
2989 }
2990
2991 update_device_state(fio);
2992
2993 if (keep_order)
2994 up_read(&fio->sbi->io_order_lock);
2995 }
2996
2997 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2998 enum iostat_type io_type)
2999 {
3000 struct f2fs_io_info fio = {
3001 .sbi = sbi,
3002 .type = META,
3003 .temp = HOT,
3004 .op = REQ_OP_WRITE,
3005 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3006 .old_blkaddr = page->index,
3007 .new_blkaddr = page->index,
3008 .page = page,
3009 .encrypted_page = NULL,
3010 .in_list = false,
3011 };
3012
3013 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3014 fio.op_flags &= ~REQ_META;
3015
3016 set_page_writeback(page);
3017 ClearPageError(page);
3018 f2fs_submit_page_write(&fio);
3019
3020 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3021 }
3022
3023 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3024 {
3025 struct f2fs_summary sum;
3026
3027 set_summary(&sum, nid, 0, 0);
3028 do_write_page(&sum, fio);
3029
3030 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3031 }
3032
3033 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3034 struct f2fs_io_info *fio)
3035 {
3036 struct f2fs_sb_info *sbi = fio->sbi;
3037 struct f2fs_summary sum;
3038
3039 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3040 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3041 do_write_page(&sum, fio);
3042 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3043
3044 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3045 }
3046
3047 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3048 {
3049 int err;
3050 struct f2fs_sb_info *sbi = fio->sbi;
3051
3052 fio->new_blkaddr = fio->old_blkaddr;
3053 /* i/o temperature is needed for passing down write hints */
3054 __get_segment_type(fio);
3055
3056 f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
3057 GET_SEGNO(sbi, fio->new_blkaddr))->type));
3058
3059 stat_inc_inplace_blocks(fio->sbi);
3060
3061 err = f2fs_submit_page_bio(fio);
3062 if (!err)
3063 update_device_state(fio);
3064
3065 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3066
3067 return err;
3068 }
3069
3070 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3071 unsigned int segno)
3072 {
3073 int i;
3074
3075 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3076 if (CURSEG_I(sbi, i)->segno == segno)
3077 break;
3078 }
3079 return i;
3080 }
3081
3082 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3083 block_t old_blkaddr, block_t new_blkaddr,
3084 bool recover_curseg, bool recover_newaddr)
3085 {
3086 struct sit_info *sit_i = SIT_I(sbi);
3087 struct curseg_info *curseg;
3088 unsigned int segno, old_cursegno;
3089 struct seg_entry *se;
3090 int type;
3091 unsigned short old_blkoff;
3092
3093 segno = GET_SEGNO(sbi, new_blkaddr);
3094 se = get_seg_entry(sbi, segno);
3095 type = se->type;
3096
3097 down_write(&SM_I(sbi)->curseg_lock);
3098
3099 if (!recover_curseg) {
3100 /* for recovery flow */
3101 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3102 if (old_blkaddr == NULL_ADDR)
3103 type = CURSEG_COLD_DATA;
3104 else
3105 type = CURSEG_WARM_DATA;
3106 }
3107 } else {
3108 if (IS_CURSEG(sbi, segno)) {
3109 /* se->type is volatile as SSR allocation */
3110 type = __f2fs_get_curseg(sbi, segno);
3111 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3112 } else {
3113 type = CURSEG_WARM_DATA;
3114 }
3115 }
3116
3117 f2fs_bug_on(sbi, !IS_DATASEG(type));
3118 curseg = CURSEG_I(sbi, type);
3119
3120 mutex_lock(&curseg->curseg_mutex);
3121 down_write(&sit_i->sentry_lock);
3122
3123 old_cursegno = curseg->segno;
3124 old_blkoff = curseg->next_blkoff;
3125
3126 /* change the current segment */
3127 if (segno != curseg->segno) {
3128 curseg->next_segno = segno;
3129 change_curseg(sbi, type);
3130 }
3131
3132 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3133 __add_sum_entry(sbi, type, sum);
3134
3135 if (!recover_curseg || recover_newaddr)
3136 update_sit_entry(sbi, new_blkaddr, 1);
3137 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO) {
3138 invalidate_mapping_pages(META_MAPPING(sbi),
3139 old_blkaddr, old_blkaddr);
3140 update_sit_entry(sbi, old_blkaddr, -1);
3141 }
3142
3143 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3144 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3145
3146 locate_dirty_segment(sbi, old_cursegno);
3147
3148 if (recover_curseg) {
3149 if (old_cursegno != curseg->segno) {
3150 curseg->next_segno = old_cursegno;
3151 change_curseg(sbi, type);
3152 }
3153 curseg->next_blkoff = old_blkoff;
3154 }
3155
3156 up_write(&sit_i->sentry_lock);
3157 mutex_unlock(&curseg->curseg_mutex);
3158 up_write(&SM_I(sbi)->curseg_lock);
3159 }
3160
3161 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3162 block_t old_addr, block_t new_addr,
3163 unsigned char version, bool recover_curseg,
3164 bool recover_newaddr)
3165 {
3166 struct f2fs_summary sum;
3167
3168 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3169
3170 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3171 recover_curseg, recover_newaddr);
3172
3173 f2fs_update_data_blkaddr(dn, new_addr);
3174 }
3175
3176 void f2fs_wait_on_page_writeback(struct page *page,
3177 enum page_type type, bool ordered)
3178 {
3179 if (PageWriteback(page)) {
3180 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3181
3182 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
3183 0, page->index, type);
3184 if (ordered)
3185 wait_on_page_writeback(page);
3186 else
3187 wait_for_stable_page(page);
3188 }
3189 }
3190
3191 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
3192 {
3193 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3194 struct page *cpage;
3195
3196 if (!f2fs_post_read_required(inode))
3197 return;
3198
3199 if (!is_valid_data_blkaddr(sbi, blkaddr))
3200 return;
3201
3202 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3203 if (cpage) {
3204 f2fs_wait_on_page_writeback(cpage, DATA, true);
3205 f2fs_put_page(cpage, 1);
3206 }
3207 }
3208
3209 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3210 {
3211 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3212 struct curseg_info *seg_i;
3213 unsigned char *kaddr;
3214 struct page *page;
3215 block_t start;
3216 int i, j, offset;
3217
3218 start = start_sum_block(sbi);
3219
3220 page = f2fs_get_meta_page(sbi, start++);
3221 if (IS_ERR(page))
3222 return PTR_ERR(page);
3223 kaddr = (unsigned char *)page_address(page);
3224
3225 /* Step 1: restore nat cache */
3226 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3227 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3228
3229 /* Step 2: restore sit cache */
3230 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3231 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3232 offset = 2 * SUM_JOURNAL_SIZE;
3233
3234 /* Step 3: restore summary entries */
3235 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3236 unsigned short blk_off;
3237 unsigned int segno;
3238
3239 seg_i = CURSEG_I(sbi, i);
3240 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3241 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3242 seg_i->next_segno = segno;
3243 reset_curseg(sbi, i, 0);
3244 seg_i->alloc_type = ckpt->alloc_type[i];
3245 seg_i->next_blkoff = blk_off;
3246
3247 if (seg_i->alloc_type == SSR)
3248 blk_off = sbi->blocks_per_seg;
3249
3250 for (j = 0; j < blk_off; j++) {
3251 struct f2fs_summary *s;
3252 s = (struct f2fs_summary *)(kaddr + offset);
3253 seg_i->sum_blk->entries[j] = *s;
3254 offset += SUMMARY_SIZE;
3255 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3256 SUM_FOOTER_SIZE)
3257 continue;
3258
3259 f2fs_put_page(page, 1);
3260 page = NULL;
3261
3262 page = f2fs_get_meta_page(sbi, start++);
3263 if (IS_ERR(page))
3264 return PTR_ERR(page);
3265 kaddr = (unsigned char *)page_address(page);
3266 offset = 0;
3267 }
3268 }
3269 f2fs_put_page(page, 1);
3270 return 0;
3271 }
3272
3273 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3274 {
3275 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3276 struct f2fs_summary_block *sum;
3277 struct curseg_info *curseg;
3278 struct page *new;
3279 unsigned short blk_off;
3280 unsigned int segno = 0;
3281 block_t blk_addr = 0;
3282 int err = 0;
3283
3284 /* get segment number and block addr */
3285 if (IS_DATASEG(type)) {
3286 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3287 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3288 CURSEG_HOT_DATA]);
3289 if (__exist_node_summaries(sbi))
3290 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3291 else
3292 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3293 } else {
3294 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3295 CURSEG_HOT_NODE]);
3296 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3297 CURSEG_HOT_NODE]);
3298 if (__exist_node_summaries(sbi))
3299 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3300 type - CURSEG_HOT_NODE);
3301 else
3302 blk_addr = GET_SUM_BLOCK(sbi, segno);
3303 }
3304
3305 new = f2fs_get_meta_page(sbi, blk_addr);
3306 if (IS_ERR(new))
3307 return PTR_ERR(new);
3308 sum = (struct f2fs_summary_block *)page_address(new);
3309
3310 if (IS_NODESEG(type)) {
3311 if (__exist_node_summaries(sbi)) {
3312 struct f2fs_summary *ns = &sum->entries[0];
3313 int i;
3314 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3315 ns->version = 0;
3316 ns->ofs_in_node = 0;
3317 }
3318 } else {
3319 err = f2fs_restore_node_summary(sbi, segno, sum);
3320 if (err)
3321 goto out;
3322 }
3323 }
3324
3325 /* set uncompleted segment to curseg */
3326 curseg = CURSEG_I(sbi, type);
3327 mutex_lock(&curseg->curseg_mutex);
3328
3329 /* update journal info */
3330 down_write(&curseg->journal_rwsem);
3331 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3332 up_write(&curseg->journal_rwsem);
3333
3334 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3335 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3336 curseg->next_segno = segno;
3337 reset_curseg(sbi, type, 0);
3338 curseg->alloc_type = ckpt->alloc_type[type];
3339 curseg->next_blkoff = blk_off;
3340 mutex_unlock(&curseg->curseg_mutex);
3341 out:
3342 f2fs_put_page(new, 1);
3343 return err;
3344 }
3345
3346 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3347 {
3348 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3349 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3350 int type = CURSEG_HOT_DATA;
3351 int err;
3352
3353 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3354 int npages = f2fs_npages_for_summary_flush(sbi, true);
3355
3356 if (npages >= 2)
3357 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3358 META_CP, true);
3359
3360 /* restore for compacted data summary */
3361 err = read_compacted_summaries(sbi);
3362 if (err)
3363 return err;
3364 type = CURSEG_HOT_NODE;
3365 }
3366
3367 if (__exist_node_summaries(sbi))
3368 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3369 NR_CURSEG_TYPE - type, META_CP, true);
3370
3371 for (; type <= CURSEG_COLD_NODE; type++) {
3372 err = read_normal_summaries(sbi, type);
3373 if (err)
3374 return err;
3375 }
3376
3377 /* sanity check for summary blocks */
3378 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3379 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3380 return -EINVAL;
3381
3382 return 0;
3383 }
3384
3385 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3386 {
3387 struct page *page;
3388 unsigned char *kaddr;
3389 struct f2fs_summary *summary;
3390 struct curseg_info *seg_i;
3391 int written_size = 0;
3392 int i, j;
3393
3394 page = f2fs_grab_meta_page(sbi, blkaddr++);
3395 kaddr = (unsigned char *)page_address(page);
3396 memset(kaddr, 0, PAGE_SIZE);
3397
3398 /* Step 1: write nat cache */
3399 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3400 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3401 written_size += SUM_JOURNAL_SIZE;
3402
3403 /* Step 2: write sit cache */
3404 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3405 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3406 written_size += SUM_JOURNAL_SIZE;
3407
3408 /* Step 3: write summary entries */
3409 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3410 unsigned short blkoff;
3411 seg_i = CURSEG_I(sbi, i);
3412 if (sbi->ckpt->alloc_type[i] == SSR)
3413 blkoff = sbi->blocks_per_seg;
3414 else
3415 blkoff = curseg_blkoff(sbi, i);
3416
3417 for (j = 0; j < blkoff; j++) {
3418 if (!page) {
3419 page = f2fs_grab_meta_page(sbi, blkaddr++);
3420 kaddr = (unsigned char *)page_address(page);
3421 memset(kaddr, 0, PAGE_SIZE);
3422 written_size = 0;
3423 }
3424 summary = (struct f2fs_summary *)(kaddr + written_size);
3425 *summary = seg_i->sum_blk->entries[j];
3426 written_size += SUMMARY_SIZE;
3427
3428 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3429 SUM_FOOTER_SIZE)
3430 continue;
3431
3432 set_page_dirty(page);
3433 f2fs_put_page(page, 1);
3434 page = NULL;
3435 }
3436 }
3437 if (page) {
3438 set_page_dirty(page);
3439 f2fs_put_page(page, 1);
3440 }
3441 }
3442
3443 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3444 block_t blkaddr, int type)
3445 {
3446 int i, end;
3447 if (IS_DATASEG(type))
3448 end = type + NR_CURSEG_DATA_TYPE;
3449 else
3450 end = type + NR_CURSEG_NODE_TYPE;
3451
3452 for (i = type; i < end; i++)
3453 write_current_sum_page(sbi, i, blkaddr + (i - type));
3454 }
3455
3456 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3457 {
3458 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3459 write_compacted_summaries(sbi, start_blk);
3460 else
3461 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3462 }
3463
3464 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3465 {
3466 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3467 }
3468
3469 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3470 unsigned int val, int alloc)
3471 {
3472 int i;
3473
3474 if (type == NAT_JOURNAL) {
3475 for (i = 0; i < nats_in_cursum(journal); i++) {
3476 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3477 return i;
3478 }
3479 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3480 return update_nats_in_cursum(journal, 1);
3481 } else if (type == SIT_JOURNAL) {
3482 for (i = 0; i < sits_in_cursum(journal); i++)
3483 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3484 return i;
3485 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3486 return update_sits_in_cursum(journal, 1);
3487 }
3488 return -1;
3489 }
3490
3491 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3492 unsigned int segno)
3493 {
3494 return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3495 }
3496
3497 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3498 unsigned int start)
3499 {
3500 struct sit_info *sit_i = SIT_I(sbi);
3501 struct page *page;
3502 pgoff_t src_off, dst_off;
3503
3504 src_off = current_sit_addr(sbi, start);
3505 dst_off = next_sit_addr(sbi, src_off);
3506
3507 page = f2fs_grab_meta_page(sbi, dst_off);
3508 seg_info_to_sit_page(sbi, page, start);
3509
3510 set_page_dirty(page);
3511 set_to_next_sit(sit_i, start);
3512
3513 return page;
3514 }
3515
3516 static struct sit_entry_set *grab_sit_entry_set(void)
3517 {
3518 struct sit_entry_set *ses =
3519 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3520
3521 ses->entry_cnt = 0;
3522 INIT_LIST_HEAD(&ses->set_list);
3523 return ses;
3524 }
3525
3526 static void release_sit_entry_set(struct sit_entry_set *ses)
3527 {
3528 list_del(&ses->set_list);
3529 kmem_cache_free(sit_entry_set_slab, ses);
3530 }
3531
3532 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3533 struct list_head *head)
3534 {
3535 struct sit_entry_set *next = ses;
3536
3537 if (list_is_last(&ses->set_list, head))
3538 return;
3539
3540 list_for_each_entry_continue(next, head, set_list)
3541 if (ses->entry_cnt <= next->entry_cnt)
3542 break;
3543
3544 list_move_tail(&ses->set_list, &next->set_list);
3545 }
3546
3547 static void add_sit_entry(unsigned int segno, struct list_head *head)
3548 {
3549 struct sit_entry_set *ses;
3550 unsigned int start_segno = START_SEGNO(segno);
3551
3552 list_for_each_entry(ses, head, set_list) {
3553 if (ses->start_segno == start_segno) {
3554 ses->entry_cnt++;
3555 adjust_sit_entry_set(ses, head);
3556 return;
3557 }
3558 }
3559
3560 ses = grab_sit_entry_set();
3561
3562 ses->start_segno = start_segno;
3563 ses->entry_cnt++;
3564 list_add(&ses->set_list, head);
3565 }
3566
3567 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3568 {
3569 struct f2fs_sm_info *sm_info = SM_I(sbi);
3570 struct list_head *set_list = &sm_info->sit_entry_set;
3571 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3572 unsigned int segno;
3573
3574 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3575 add_sit_entry(segno, set_list);
3576 }
3577
3578 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3579 {
3580 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3581 struct f2fs_journal *journal = curseg->journal;
3582 int i;
3583
3584 down_write(&curseg->journal_rwsem);
3585 for (i = 0; i < sits_in_cursum(journal); i++) {
3586 unsigned int segno;
3587 bool dirtied;
3588
3589 segno = le32_to_cpu(segno_in_journal(journal, i));
3590 dirtied = __mark_sit_entry_dirty(sbi, segno);
3591
3592 if (!dirtied)
3593 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3594 }
3595 update_sits_in_cursum(journal, -i);
3596 up_write(&curseg->journal_rwsem);
3597 }
3598
3599 /*
3600 * CP calls this function, which flushes SIT entries including sit_journal,
3601 * and moves prefree segs to free segs.
3602 */
3603 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3604 {
3605 struct sit_info *sit_i = SIT_I(sbi);
3606 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3607 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3608 struct f2fs_journal *journal = curseg->journal;
3609 struct sit_entry_set *ses, *tmp;
3610 struct list_head *head = &SM_I(sbi)->sit_entry_set;
3611 bool to_journal = true;
3612 struct seg_entry *se;
3613
3614 down_write(&sit_i->sentry_lock);
3615
3616 if (!sit_i->dirty_sentries)
3617 goto out;
3618
3619 /*
3620 * add and account sit entries of dirty bitmap in sit entry
3621 * set temporarily
3622 */
3623 add_sits_in_set(sbi);
3624
3625 /*
3626 * if there are no enough space in journal to store dirty sit
3627 * entries, remove all entries from journal and add and account
3628 * them in sit entry set.
3629 */
3630 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3631 remove_sits_in_journal(sbi);
3632
3633 /*
3634 * there are two steps to flush sit entries:
3635 * #1, flush sit entries to journal in current cold data summary block.
3636 * #2, flush sit entries to sit page.
3637 */
3638 list_for_each_entry_safe(ses, tmp, head, set_list) {
3639 struct page *page = NULL;
3640 struct f2fs_sit_block *raw_sit = NULL;
3641 unsigned int start_segno = ses->start_segno;
3642 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3643 (unsigned long)MAIN_SEGS(sbi));
3644 unsigned int segno = start_segno;
3645
3646 if (to_journal &&
3647 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3648 to_journal = false;
3649
3650 if (to_journal) {
3651 down_write(&curseg->journal_rwsem);
3652 } else {
3653 page = get_next_sit_page(sbi, start_segno);
3654 raw_sit = page_address(page);
3655 }
3656
3657 /* flush dirty sit entries in region of current sit set */
3658 for_each_set_bit_from(segno, bitmap, end) {
3659 int offset, sit_offset;
3660
3661 se = get_seg_entry(sbi, segno);
3662 #ifdef CONFIG_F2FS_CHECK_FS
3663 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3664 SIT_VBLOCK_MAP_SIZE))
3665 f2fs_bug_on(sbi, 1);
3666 #endif
3667
3668 /* add discard candidates */
3669 if (!(cpc->reason & CP_DISCARD)) {
3670 cpc->trim_start = segno;
3671 add_discard_addrs(sbi, cpc, false);
3672 }
3673
3674 if (to_journal) {
3675 offset = f2fs_lookup_journal_in_cursum(journal,
3676 SIT_JOURNAL, segno, 1);
3677 f2fs_bug_on(sbi, offset < 0);
3678 segno_in_journal(journal, offset) =
3679 cpu_to_le32(segno);
3680 seg_info_to_raw_sit(se,
3681 &sit_in_journal(journal, offset));
3682 check_block_count(sbi, segno,
3683 &sit_in_journal(journal, offset));
3684 } else {
3685 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3686 seg_info_to_raw_sit(se,
3687 &raw_sit->entries[sit_offset]);
3688 check_block_count(sbi, segno,
3689 &raw_sit->entries[sit_offset]);
3690 }
3691
3692 __clear_bit(segno, bitmap);
3693 sit_i->dirty_sentries--;
3694 ses->entry_cnt--;
3695 }
3696
3697 if (to_journal)
3698 up_write(&curseg->journal_rwsem);
3699 else
3700 f2fs_put_page(page, 1);
3701
3702 f2fs_bug_on(sbi, ses->entry_cnt);
3703 release_sit_entry_set(ses);
3704 }
3705
3706 f2fs_bug_on(sbi, !list_empty(head));
3707 f2fs_bug_on(sbi, sit_i->dirty_sentries);
3708 out:
3709 if (cpc->reason & CP_DISCARD) {
3710 __u64 trim_start = cpc->trim_start;
3711
3712 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3713 add_discard_addrs(sbi, cpc, false);
3714
3715 cpc->trim_start = trim_start;
3716 }
3717 up_write(&sit_i->sentry_lock);
3718
3719 set_prefree_as_free_segments(sbi);
3720 }
3721
3722 static int build_sit_info(struct f2fs_sb_info *sbi)
3723 {
3724 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3725 struct sit_info *sit_i;
3726 unsigned int sit_segs, start;
3727 char *src_bitmap;
3728 unsigned int bitmap_size;
3729
3730 /* allocate memory for SIT information */
3731 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3732 if (!sit_i)
3733 return -ENOMEM;
3734
3735 SM_I(sbi)->sit_info = sit_i;
3736
3737 sit_i->sentries =
3738 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3739 MAIN_SEGS(sbi)),
3740 GFP_KERNEL);
3741 if (!sit_i->sentries)
3742 return -ENOMEM;
3743
3744 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3745 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3746 GFP_KERNEL);
3747 if (!sit_i->dirty_sentries_bitmap)
3748 return -ENOMEM;
3749
3750 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3751 sit_i->sentries[start].cur_valid_map
3752 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3753 sit_i->sentries[start].ckpt_valid_map
3754 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3755 if (!sit_i->sentries[start].cur_valid_map ||
3756 !sit_i->sentries[start].ckpt_valid_map)
3757 return -ENOMEM;
3758
3759 #ifdef CONFIG_F2FS_CHECK_FS
3760 sit_i->sentries[start].cur_valid_map_mir
3761 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3762 if (!sit_i->sentries[start].cur_valid_map_mir)
3763 return -ENOMEM;
3764 #endif
3765
3766 sit_i->sentries[start].discard_map
3767 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3768 GFP_KERNEL);
3769 if (!sit_i->sentries[start].discard_map)
3770 return -ENOMEM;
3771 }
3772
3773 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3774 if (!sit_i->tmp_map)
3775 return -ENOMEM;
3776
3777 if (sbi->segs_per_sec > 1) {
3778 sit_i->sec_entries =
3779 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3780 MAIN_SECS(sbi)),
3781 GFP_KERNEL);
3782 if (!sit_i->sec_entries)
3783 return -ENOMEM;
3784 }
3785
3786 /* get information related with SIT */
3787 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3788
3789 /* setup SIT bitmap from ckeckpoint pack */
3790 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3791 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3792
3793 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3794 if (!sit_i->sit_bitmap)
3795 return -ENOMEM;
3796
3797 #ifdef CONFIG_F2FS_CHECK_FS
3798 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3799 if (!sit_i->sit_bitmap_mir)
3800 return -ENOMEM;
3801 #endif
3802
3803 /* init SIT information */
3804 sit_i->s_ops = &default_salloc_ops;
3805
3806 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3807 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3808 sit_i->written_valid_blocks = 0;
3809 sit_i->bitmap_size = bitmap_size;
3810 sit_i->dirty_sentries = 0;
3811 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3812 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3813 sit_i->mounted_time = ktime_get_real_seconds();
3814 init_rwsem(&sit_i->sentry_lock);
3815 return 0;
3816 }
3817
3818 static int build_free_segmap(struct f2fs_sb_info *sbi)
3819 {
3820 struct free_segmap_info *free_i;
3821 unsigned int bitmap_size, sec_bitmap_size;
3822
3823 /* allocate memory for free segmap information */
3824 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
3825 if (!free_i)
3826 return -ENOMEM;
3827
3828 SM_I(sbi)->free_info = free_i;
3829
3830 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3831 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
3832 if (!free_i->free_segmap)
3833 return -ENOMEM;
3834
3835 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3836 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
3837 if (!free_i->free_secmap)
3838 return -ENOMEM;
3839
3840 /* set all segments as dirty temporarily */
3841 memset(free_i->free_segmap, 0xff, bitmap_size);
3842 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3843
3844 /* init free segmap information */
3845 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3846 free_i->free_segments = 0;
3847 free_i->free_sections = 0;
3848 spin_lock_init(&free_i->segmap_lock);
3849 return 0;
3850 }
3851
3852 static int build_curseg(struct f2fs_sb_info *sbi)
3853 {
3854 struct curseg_info *array;
3855 int i;
3856
3857 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
3858 GFP_KERNEL);
3859 if (!array)
3860 return -ENOMEM;
3861
3862 SM_I(sbi)->curseg_array = array;
3863
3864 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3865 mutex_init(&array[i].curseg_mutex);
3866 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
3867 if (!array[i].sum_blk)
3868 return -ENOMEM;
3869 init_rwsem(&array[i].journal_rwsem);
3870 array[i].journal = f2fs_kzalloc(sbi,
3871 sizeof(struct f2fs_journal), GFP_KERNEL);
3872 if (!array[i].journal)
3873 return -ENOMEM;
3874 array[i].segno = NULL_SEGNO;
3875 array[i].next_blkoff = 0;
3876 }
3877 return restore_curseg_summaries(sbi);
3878 }
3879
3880 static int build_sit_entries(struct f2fs_sb_info *sbi)
3881 {
3882 struct sit_info *sit_i = SIT_I(sbi);
3883 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3884 struct f2fs_journal *journal = curseg->journal;
3885 struct seg_entry *se;
3886 struct f2fs_sit_entry sit;
3887 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3888 unsigned int i, start, end;
3889 unsigned int readed, start_blk = 0;
3890 int err = 0;
3891 block_t total_node_blocks = 0;
3892
3893 do {
3894 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3895 META_SIT, true);
3896
3897 start = start_blk * sit_i->sents_per_block;
3898 end = (start_blk + readed) * sit_i->sents_per_block;
3899
3900 for (; start < end && start < MAIN_SEGS(sbi); start++) {
3901 struct f2fs_sit_block *sit_blk;
3902 struct page *page;
3903
3904 se = &sit_i->sentries[start];
3905 page = get_current_sit_page(sbi, start);
3906 sit_blk = (struct f2fs_sit_block *)page_address(page);
3907 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3908 f2fs_put_page(page, 1);
3909
3910 err = check_block_count(sbi, start, &sit);
3911 if (err)
3912 return err;
3913 seg_info_from_raw_sit(se, &sit);
3914 if (IS_NODESEG(se->type))
3915 total_node_blocks += se->valid_blocks;
3916
3917 /* build discard map only one time */
3918 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3919 memset(se->discard_map, 0xff,
3920 SIT_VBLOCK_MAP_SIZE);
3921 } else {
3922 memcpy(se->discard_map,
3923 se->cur_valid_map,
3924 SIT_VBLOCK_MAP_SIZE);
3925 sbi->discard_blks +=
3926 sbi->blocks_per_seg -
3927 se->valid_blocks;
3928 }
3929
3930 if (sbi->segs_per_sec > 1)
3931 get_sec_entry(sbi, start)->valid_blocks +=
3932 se->valid_blocks;
3933 }
3934 start_blk += readed;
3935 } while (start_blk < sit_blk_cnt);
3936
3937 down_read(&curseg->journal_rwsem);
3938 for (i = 0; i < sits_in_cursum(journal); i++) {
3939 unsigned int old_valid_blocks;
3940
3941 start = le32_to_cpu(segno_in_journal(journal, i));
3942 if (start >= MAIN_SEGS(sbi)) {
3943 f2fs_msg(sbi->sb, KERN_ERR,
3944 "Wrong journal entry on segno %u",
3945 start);
3946 set_sbi_flag(sbi, SBI_NEED_FSCK);
3947 err = -EINVAL;
3948 break;
3949 }
3950
3951 se = &sit_i->sentries[start];
3952 sit = sit_in_journal(journal, i);
3953
3954 old_valid_blocks = se->valid_blocks;
3955 if (IS_NODESEG(se->type))
3956 total_node_blocks -= old_valid_blocks;
3957
3958 err = check_block_count(sbi, start, &sit);
3959 if (err)
3960 break;
3961 seg_info_from_raw_sit(se, &sit);
3962 if (IS_NODESEG(se->type))
3963 total_node_blocks += se->valid_blocks;
3964
3965 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3966 memset(se->discard_map, 0xff, SIT_VBLOCK_MAP_SIZE);
3967 } else {
3968 memcpy(se->discard_map, se->cur_valid_map,
3969 SIT_VBLOCK_MAP_SIZE);
3970 sbi->discard_blks += old_valid_blocks;
3971 sbi->discard_blks -= se->valid_blocks;
3972 }
3973
3974 if (sbi->segs_per_sec > 1) {
3975 get_sec_entry(sbi, start)->valid_blocks +=
3976 se->valid_blocks;
3977 get_sec_entry(sbi, start)->valid_blocks -=
3978 old_valid_blocks;
3979 }
3980 }
3981 up_read(&curseg->journal_rwsem);
3982
3983 if (!err && total_node_blocks != valid_node_count(sbi)) {
3984 f2fs_msg(sbi->sb, KERN_ERR,
3985 "SIT is corrupted node# %u vs %u",
3986 total_node_blocks, valid_node_count(sbi));
3987 set_sbi_flag(sbi, SBI_NEED_FSCK);
3988 err = -EINVAL;
3989 }
3990
3991 return err;
3992 }
3993
3994 static void init_free_segmap(struct f2fs_sb_info *sbi)
3995 {
3996 unsigned int start;
3997 int type;
3998
3999 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4000 struct seg_entry *sentry = get_seg_entry(sbi, start);
4001 if (!sentry->valid_blocks)
4002 __set_free(sbi, start);
4003 else
4004 SIT_I(sbi)->written_valid_blocks +=
4005 sentry->valid_blocks;
4006 }
4007
4008 /* set use the current segments */
4009 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4010 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4011 __set_test_and_inuse(sbi, curseg_t->segno);
4012 }
4013 }
4014
4015 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4016 {
4017 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4018 struct free_segmap_info *free_i = FREE_I(sbi);
4019 unsigned int segno = 0, offset = 0;
4020 unsigned short valid_blocks;
4021
4022 while (1) {
4023 /* find dirty segment based on free segmap */
4024 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4025 if (segno >= MAIN_SEGS(sbi))
4026 break;
4027 offset = segno + 1;
4028 valid_blocks = get_valid_blocks(sbi, segno, false);
4029 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4030 continue;
4031 if (valid_blocks > sbi->blocks_per_seg) {
4032 f2fs_bug_on(sbi, 1);
4033 continue;
4034 }
4035 mutex_lock(&dirty_i->seglist_lock);
4036 __locate_dirty_segment(sbi, segno, DIRTY);
4037 mutex_unlock(&dirty_i->seglist_lock);
4038 }
4039 }
4040
4041 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4042 {
4043 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4044 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4045
4046 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4047 if (!dirty_i->victim_secmap)
4048 return -ENOMEM;
4049 return 0;
4050 }
4051
4052 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4053 {
4054 struct dirty_seglist_info *dirty_i;
4055 unsigned int bitmap_size, i;
4056
4057 /* allocate memory for dirty segments list information */
4058 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4059 GFP_KERNEL);
4060 if (!dirty_i)
4061 return -ENOMEM;
4062
4063 SM_I(sbi)->dirty_info = dirty_i;
4064 mutex_init(&dirty_i->seglist_lock);
4065
4066 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4067
4068 for (i = 0; i < NR_DIRTY_TYPE; i++) {
4069 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4070 GFP_KERNEL);
4071 if (!dirty_i->dirty_segmap[i])
4072 return -ENOMEM;
4073 }
4074
4075 init_dirty_segmap(sbi);
4076 return init_victim_secmap(sbi);
4077 }
4078
4079 /*
4080 * Update min, max modified time for cost-benefit GC algorithm
4081 */
4082 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4083 {
4084 struct sit_info *sit_i = SIT_I(sbi);
4085 unsigned int segno;
4086
4087 down_write(&sit_i->sentry_lock);
4088
4089 sit_i->min_mtime = ULLONG_MAX;
4090
4091 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4092 unsigned int i;
4093 unsigned long long mtime = 0;
4094
4095 for (i = 0; i < sbi->segs_per_sec; i++)
4096 mtime += get_seg_entry(sbi, segno + i)->mtime;
4097
4098 mtime = div_u64(mtime, sbi->segs_per_sec);
4099
4100 if (sit_i->min_mtime > mtime)
4101 sit_i->min_mtime = mtime;
4102 }
4103 sit_i->max_mtime = get_mtime(sbi, false);
4104 up_write(&sit_i->sentry_lock);
4105 }
4106
4107 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4108 {
4109 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4110 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4111 struct f2fs_sm_info *sm_info;
4112 int err;
4113
4114 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4115 if (!sm_info)
4116 return -ENOMEM;
4117
4118 /* init sm info */
4119 sbi->sm_info = sm_info;
4120 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4121 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4122 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4123 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4124 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4125 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4126 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4127 sm_info->rec_prefree_segments = sm_info->main_segments *
4128 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4129 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4130 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4131
4132 if (!test_opt(sbi, LFS))
4133 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4134 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4135 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4136 sm_info->min_seq_blocks = sbi->blocks_per_seg * sbi->segs_per_sec;
4137 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4138 sm_info->min_ssr_sections = reserved_sections(sbi);
4139
4140 INIT_LIST_HEAD(&sm_info->sit_entry_set);
4141
4142 init_rwsem(&sm_info->curseg_lock);
4143
4144 if (!f2fs_readonly(sbi->sb)) {
4145 err = f2fs_create_flush_cmd_control(sbi);
4146 if (err)
4147 return err;
4148 }
4149
4150 err = create_discard_cmd_control(sbi);
4151 if (err)
4152 return err;
4153
4154 err = build_sit_info(sbi);
4155 if (err)
4156 return err;
4157 err = build_free_segmap(sbi);
4158 if (err)
4159 return err;
4160 err = build_curseg(sbi);
4161 if (err)
4162 return err;
4163
4164 /* reinit free segmap based on SIT */
4165 err = build_sit_entries(sbi);
4166 if (err)
4167 return err;
4168
4169 init_free_segmap(sbi);
4170 err = build_dirty_segmap(sbi);
4171 if (err)
4172 return err;
4173
4174 init_min_max_mtime(sbi);
4175 return 0;
4176 }
4177
4178 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4179 enum dirty_type dirty_type)
4180 {
4181 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4182
4183 mutex_lock(&dirty_i->seglist_lock);
4184 kvfree(dirty_i->dirty_segmap[dirty_type]);
4185 dirty_i->nr_dirty[dirty_type] = 0;
4186 mutex_unlock(&dirty_i->seglist_lock);
4187 }
4188
4189 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4190 {
4191 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4192 kvfree(dirty_i->victim_secmap);
4193 }
4194
4195 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4196 {
4197 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4198 int i;
4199
4200 if (!dirty_i)
4201 return;
4202
4203 /* discard pre-free/dirty segments list */
4204 for (i = 0; i < NR_DIRTY_TYPE; i++)
4205 discard_dirty_segmap(sbi, i);
4206
4207 destroy_victim_secmap(sbi);
4208 SM_I(sbi)->dirty_info = NULL;
4209 kfree(dirty_i);
4210 }
4211
4212 static void destroy_curseg(struct f2fs_sb_info *sbi)
4213 {
4214 struct curseg_info *array = SM_I(sbi)->curseg_array;
4215 int i;
4216
4217 if (!array)
4218 return;
4219 SM_I(sbi)->curseg_array = NULL;
4220 for (i = 0; i < NR_CURSEG_TYPE; i++) {
4221 kfree(array[i].sum_blk);
4222 kfree(array[i].journal);
4223 }
4224 kfree(array);
4225 }
4226
4227 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4228 {
4229 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4230 if (!free_i)
4231 return;
4232 SM_I(sbi)->free_info = NULL;
4233 kvfree(free_i->free_segmap);
4234 kvfree(free_i->free_secmap);
4235 kfree(free_i);
4236 }
4237
4238 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4239 {
4240 struct sit_info *sit_i = SIT_I(sbi);
4241 unsigned int start;
4242
4243 if (!sit_i)
4244 return;
4245
4246 if (sit_i->sentries) {
4247 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4248 kfree(sit_i->sentries[start].cur_valid_map);
4249 #ifdef CONFIG_F2FS_CHECK_FS
4250 kfree(sit_i->sentries[start].cur_valid_map_mir);
4251 #endif
4252 kfree(sit_i->sentries[start].ckpt_valid_map);
4253 kfree(sit_i->sentries[start].discard_map);
4254 }
4255 }
4256 kfree(sit_i->tmp_map);
4257
4258 kvfree(sit_i->sentries);
4259 kvfree(sit_i->sec_entries);
4260 kvfree(sit_i->dirty_sentries_bitmap);
4261
4262 SM_I(sbi)->sit_info = NULL;
4263 kfree(sit_i->sit_bitmap);
4264 #ifdef CONFIG_F2FS_CHECK_FS
4265 kfree(sit_i->sit_bitmap_mir);
4266 #endif
4267 kfree(sit_i);
4268 }
4269
4270 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4271 {
4272 struct f2fs_sm_info *sm_info = SM_I(sbi);
4273
4274 if (!sm_info)
4275 return;
4276 f2fs_destroy_flush_cmd_control(sbi, true);
4277 destroy_discard_cmd_control(sbi);
4278 destroy_dirty_segmap(sbi);
4279 destroy_curseg(sbi);
4280 destroy_free_segmap(sbi);
4281 destroy_sit_info(sbi);
4282 sbi->sm_info = NULL;
4283 kfree(sm_info);
4284 }
4285
4286 int __init f2fs_create_segment_manager_caches(void)
4287 {
4288 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4289 sizeof(struct discard_entry));
4290 if (!discard_entry_slab)
4291 goto fail;
4292
4293 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4294 sizeof(struct discard_cmd));
4295 if (!discard_cmd_slab)
4296 goto destroy_discard_entry;
4297
4298 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4299 sizeof(struct sit_entry_set));
4300 if (!sit_entry_set_slab)
4301 goto destroy_discard_cmd;
4302
4303 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4304 sizeof(struct inmem_pages));
4305 if (!inmem_entry_slab)
4306 goto destroy_sit_entry_set;
4307 return 0;
4308
4309 destroy_sit_entry_set:
4310 kmem_cache_destroy(sit_entry_set_slab);
4311 destroy_discard_cmd:
4312 kmem_cache_destroy(discard_cmd_slab);
4313 destroy_discard_entry:
4314 kmem_cache_destroy(discard_entry_slab);
4315 fail:
4316 return -ENOMEM;
4317 }
4318
4319 void f2fs_destroy_segment_manager_caches(void)
4320 {
4321 kmem_cache_destroy(sit_entry_set_slab);
4322 kmem_cache_destroy(discard_cmd_slab);
4323 kmem_cache_destroy(discard_entry_slab);
4324 kmem_cache_destroy(inmem_entry_slab);
4325 }
Ubuntu Jammy Linux kernel mirror