]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blob - fs/f2fs/segment.c
projects / mirror_ubuntu-eoan-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: fix use-after-free of dicard command entry
[mirror_ubuntu-eoan-kernel.git] / fs / f2fs / segment.c
1 /*
2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/bio.h>
14 #include <linux/blkdev.h>
15 #include <linux/prefetch.h>
16 #include <linux/kthread.h>
17 #include <linux/swap.h>
18 #include <linux/timer.h>
19 #include <linux/freezer.h>
20 #include <linux/sched/signal.h>
21
22 #include "f2fs.h"
23 #include "segment.h"
24 #include "node.h"
25 #include "gc.h"
26 #include "trace.h"
27 #include <trace/events/f2fs.h>
28
29 #define __reverse_ffz(x) __reverse_ffs(~(x))
30
31 static struct kmem_cache *discard_entry_slab;
32 static struct kmem_cache *discard_cmd_slab;
33 static struct kmem_cache *sit_entry_set_slab;
34 static struct kmem_cache *inmem_entry_slab;
35
36 static unsigned long __reverse_ulong(unsigned char *str)
37 {
38 unsigned long tmp = 0;
39 int shift = 24, idx = 0;
40
41 #if BITS_PER_LONG == 64
42 shift = 56;
43 #endif
44 while (shift >= 0) {
45 tmp |= (unsigned long)str[idx++] << shift;
46 shift -= BITS_PER_BYTE;
47 }
48 return tmp;
49 }
50
51 /*
52 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
53 * MSB and LSB are reversed in a byte by f2fs_set_bit.
54 */
55 static inline unsigned long __reverse_ffs(unsigned long word)
56 {
57 int num = 0;
58
59 #if BITS_PER_LONG == 64
60 if ((word & 0xffffffff00000000UL) == 0)
61 num += 32;
62 else
63 word >>= 32;
64 #endif
65 if ((word & 0xffff0000) == 0)
66 num += 16;
67 else
68 word >>= 16;
69
70 if ((word & 0xff00) == 0)
71 num += 8;
72 else
73 word >>= 8;
74
75 if ((word & 0xf0) == 0)
76 num += 4;
77 else
78 word >>= 4;
79
80 if ((word & 0xc) == 0)
81 num += 2;
82 else
83 word >>= 2;
84
85 if ((word & 0x2) == 0)
86 num += 1;
87 return num;
88 }
89
90 /*
91 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
92 * f2fs_set_bit makes MSB and LSB reversed in a byte.
93 * @size must be integral times of unsigned long.
94 * Example:
95 * MSB <--> LSB
96 * f2fs_set_bit(0, bitmap) => 1000 0000
97 * f2fs_set_bit(7, bitmap) => 0000 0001
98 */
99 static unsigned long __find_rev_next_bit(const unsigned long *addr,
100 unsigned long size, unsigned long offset)
101 {
102 const unsigned long *p = addr + BIT_WORD(offset);
103 unsigned long result = size;
104 unsigned long tmp;
105
106 if (offset >= size)
107 return size;
108
109 size -= (offset & ~(BITS_PER_LONG - 1));
110 offset %= BITS_PER_LONG;
111
112 while (1) {
113 if (*p == 0)
114 goto pass;
115
116 tmp = __reverse_ulong((unsigned char *)p);
117
118 tmp &= ~0UL >> offset;
119 if (size < BITS_PER_LONG)
120 tmp &= (~0UL << (BITS_PER_LONG - size));
121 if (tmp)
122 goto found;
123 pass:
124 if (size <= BITS_PER_LONG)
125 break;
126 size -= BITS_PER_LONG;
127 offset = 0;
128 p++;
129 }
130 return result;
131 found:
132 return result - size + __reverse_ffs(tmp);
133 }
134
135 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
136 unsigned long size, unsigned long offset)
137 {
138 const unsigned long *p = addr + BIT_WORD(offset);
139 unsigned long result = size;
140 unsigned long tmp;
141
142 if (offset >= size)
143 return size;
144
145 size -= (offset & ~(BITS_PER_LONG - 1));
146 offset %= BITS_PER_LONG;
147
148 while (1) {
149 if (*p == ~0UL)
150 goto pass;
151
152 tmp = __reverse_ulong((unsigned char *)p);
153
154 if (offset)
155 tmp |= ~0UL << (BITS_PER_LONG - offset);
156 if (size < BITS_PER_LONG)
157 tmp |= ~0UL >> size;
158 if (tmp != ~0UL)
159 goto found;
160 pass:
161 if (size <= BITS_PER_LONG)
162 break;
163 size -= BITS_PER_LONG;
164 offset = 0;
165 p++;
166 }
167 return result;
168 found:
169 return result - size + __reverse_ffz(tmp);
170 }
171
172 bool f2fs_need_SSR(struct f2fs_sb_info *sbi)
173 {
174 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
175 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
176 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
177
178 if (test_opt(sbi, LFS))
179 return false;
180 if (sbi->gc_mode == GC_URGENT)
181 return true;
182
183 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
184 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
185 }
186
187 void f2fs_register_inmem_page(struct inode *inode, struct page *page)
188 {
189 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
190 struct f2fs_inode_info *fi = F2FS_I(inode);
191 struct inmem_pages *new;
192
193 f2fs_trace_pid(page);
194
195 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
196 SetPagePrivate(page);
197
198 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
199
200 /* add atomic page indices to the list */
201 new->page = page;
202 INIT_LIST_HEAD(&new->list);
203
204 /* increase reference count with clean state */
205 mutex_lock(&fi->inmem_lock);
206 get_page(page);
207 list_add_tail(&new->list, &fi->inmem_pages);
208 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
209 if (list_empty(&fi->inmem_ilist))
210 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
211 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
212 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
213 mutex_unlock(&fi->inmem_lock);
214
215 trace_f2fs_register_inmem_page(page, INMEM);
216 }
217
218 static int __revoke_inmem_pages(struct inode *inode,
219 struct list_head *head, bool drop, bool recover)
220 {
221 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
222 struct inmem_pages *cur, *tmp;
223 int err = 0;
224
225 list_for_each_entry_safe(cur, tmp, head, list) {
226 struct page *page = cur->page;
227
228 if (drop)
229 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
230
231 lock_page(page);
232
233 f2fs_wait_on_page_writeback(page, DATA, true);
234
235 if (recover) {
236 struct dnode_of_data dn;
237 struct node_info ni;
238
239 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
240 retry:
241 set_new_dnode(&dn, inode, NULL, NULL, 0);
242 err = f2fs_get_dnode_of_data(&dn, page->index,
243 LOOKUP_NODE);
244 if (err) {
245 if (err == -ENOMEM) {
246 congestion_wait(BLK_RW_ASYNC, HZ/50);
247 cond_resched();
248 goto retry;
249 }
250 err = -EAGAIN;
251 goto next;
252 }
253
254 err = f2fs_get_node_info(sbi, dn.nid, &ni);
255 if (err) {
256 f2fs_put_dnode(&dn);
257 return err;
258 }
259
260 if (cur->old_addr == NEW_ADDR) {
261 f2fs_invalidate_blocks(sbi, dn.data_blkaddr);
262 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
263 } else
264 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
265 cur->old_addr, ni.version, true, true);
266 f2fs_put_dnode(&dn);
267 }
268 next:
269 /* we don't need to invalidate this in the sccessful status */
270 if (drop || recover)
271 ClearPageUptodate(page);
272 set_page_private(page, 0);
273 ClearPagePrivate(page);
274 f2fs_put_page(page, 1);
275
276 list_del(&cur->list);
277 kmem_cache_free(inmem_entry_slab, cur);
278 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
279 }
280 return err;
281 }
282
283 void f2fs_drop_inmem_pages_all(struct f2fs_sb_info *sbi, bool gc_failure)
284 {
285 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
286 struct inode *inode;
287 struct f2fs_inode_info *fi;
288 next:
289 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
290 if (list_empty(head)) {
291 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
292 return;
293 }
294 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
295 inode = igrab(&fi->vfs_inode);
296 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
297
298 if (inode) {
299 if (gc_failure) {
300 if (fi->i_gc_failures[GC_FAILURE_ATOMIC])
301 goto drop;
302 goto skip;
303 }
304 drop:
305 set_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
306 f2fs_drop_inmem_pages(inode);
307 iput(inode);
308 }
309 skip:
310 congestion_wait(BLK_RW_ASYNC, HZ/50);
311 cond_resched();
312 goto next;
313 }
314
315 void f2fs_drop_inmem_pages(struct inode *inode)
316 {
317 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
318 struct f2fs_inode_info *fi = F2FS_I(inode);
319
320 mutex_lock(&fi->inmem_lock);
321 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
322 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
323 if (!list_empty(&fi->inmem_ilist))
324 list_del_init(&fi->inmem_ilist);
325 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
326 mutex_unlock(&fi->inmem_lock);
327
328 clear_inode_flag(inode, FI_ATOMIC_FILE);
329 fi->i_gc_failures[GC_FAILURE_ATOMIC] = 0;
330 stat_dec_atomic_write(inode);
331 }
332
333 void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
334 {
335 struct f2fs_inode_info *fi = F2FS_I(inode);
336 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
337 struct list_head *head = &fi->inmem_pages;
338 struct inmem_pages *cur = NULL;
339
340 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
341
342 mutex_lock(&fi->inmem_lock);
343 list_for_each_entry(cur, head, list) {
344 if (cur->page == page)
345 break;
346 }
347
348 f2fs_bug_on(sbi, list_empty(head) || cur->page != page);
349 list_del(&cur->list);
350 mutex_unlock(&fi->inmem_lock);
351
352 dec_page_count(sbi, F2FS_INMEM_PAGES);
353 kmem_cache_free(inmem_entry_slab, cur);
354
355 ClearPageUptodate(page);
356 set_page_private(page, 0);
357 ClearPagePrivate(page);
358 f2fs_put_page(page, 0);
359
360 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
361 }
362
363 static int __f2fs_commit_inmem_pages(struct inode *inode)
364 {
365 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
366 struct f2fs_inode_info *fi = F2FS_I(inode);
367 struct inmem_pages *cur, *tmp;
368 struct f2fs_io_info fio = {
369 .sbi = sbi,
370 .ino = inode->i_ino,
371 .type = DATA,
372 .op = REQ_OP_WRITE,
373 .op_flags = REQ_SYNC | REQ_PRIO,
374 .io_type = FS_DATA_IO,
375 };
376 struct list_head revoke_list;
377 pgoff_t last_idx = ULONG_MAX;
378 int err = 0;
379
380 INIT_LIST_HEAD(&revoke_list);
381
382 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
383 struct page *page = cur->page;
384
385 lock_page(page);
386 if (page->mapping == inode->i_mapping) {
387 trace_f2fs_commit_inmem_page(page, INMEM);
388
389 set_page_dirty(page);
390 f2fs_wait_on_page_writeback(page, DATA, true);
391 if (clear_page_dirty_for_io(page)) {
392 inode_dec_dirty_pages(inode);
393 f2fs_remove_dirty_inode(inode);
394 }
395 retry:
396 fio.page = page;
397 fio.old_blkaddr = NULL_ADDR;
398 fio.encrypted_page = NULL;
399 fio.need_lock = LOCK_DONE;
400 err = f2fs_do_write_data_page(&fio);
401 if (err) {
402 if (err == -ENOMEM) {
403 congestion_wait(BLK_RW_ASYNC, HZ/50);
404 cond_resched();
405 goto retry;
406 }
407 unlock_page(page);
408 break;
409 }
410 /* record old blkaddr for revoking */
411 cur->old_addr = fio.old_blkaddr;
412 last_idx = page->index;
413 }
414 unlock_page(page);
415 list_move_tail(&cur->list, &revoke_list);
416 }
417
418 if (last_idx != ULONG_MAX)
419 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
420
421 if (err) {
422 /*
423 * try to revoke all committed pages, but still we could fail
424 * due to no memory or other reason, if that happened, EAGAIN
425 * will be returned, which means in such case, transaction is
426 * already not integrity, caller should use journal to do the
427 * recovery or rewrite & commit last transaction. For other
428 * error number, revoking was done by filesystem itself.
429 */
430 err = __revoke_inmem_pages(inode, &revoke_list, false, true);
431
432 /* drop all uncommitted pages */
433 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
434 } else {
435 __revoke_inmem_pages(inode, &revoke_list, false, false);
436 }
437
438 return err;
439 }
440
441 int f2fs_commit_inmem_pages(struct inode *inode)
442 {
443 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
444 struct f2fs_inode_info *fi = F2FS_I(inode);
445 int err;
446
447 f2fs_balance_fs(sbi, true);
448 f2fs_lock_op(sbi);
449
450 set_inode_flag(inode, FI_ATOMIC_COMMIT);
451
452 mutex_lock(&fi->inmem_lock);
453 err = __f2fs_commit_inmem_pages(inode);
454
455 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
456 if (!list_empty(&fi->inmem_ilist))
457 list_del_init(&fi->inmem_ilist);
458 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
459 mutex_unlock(&fi->inmem_lock);
460
461 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
462
463 f2fs_unlock_op(sbi);
464 return err;
465 }
466
467 /*
468 * This function balances dirty node and dentry pages.
469 * In addition, it controls garbage collection.
470 */
471 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
472 {
473 #ifdef CONFIG_F2FS_FAULT_INJECTION
474 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
475 f2fs_show_injection_info(FAULT_CHECKPOINT);
476 f2fs_stop_checkpoint(sbi, false);
477 }
478 #endif
479
480 /* balance_fs_bg is able to be pending */
481 if (need && excess_cached_nats(sbi))
482 f2fs_balance_fs_bg(sbi);
483
484 /*
485 * We should do GC or end up with checkpoint, if there are so many dirty
486 * dir/node pages without enough free segments.
487 */
488 if (has_not_enough_free_secs(sbi, 0, 0)) {
489 mutex_lock(&sbi->gc_mutex);
490 f2fs_gc(sbi, false, false, NULL_SEGNO);
491 }
492 }
493
494 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
495 {
496 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
497 return;
498
499 /* try to shrink extent cache when there is no enough memory */
500 if (!f2fs_available_free_memory(sbi, EXTENT_CACHE))
501 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
502
503 /* check the # of cached NAT entries */
504 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES))
505 f2fs_try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
506
507 if (!f2fs_available_free_memory(sbi, FREE_NIDS))
508 f2fs_try_to_free_nids(sbi, MAX_FREE_NIDS);
509 else
510 f2fs_build_free_nids(sbi, false, false);
511
512 if (!is_idle(sbi) &&
513 (!excess_dirty_nats(sbi) && !excess_dirty_nodes(sbi)))
514 return;
515
516 /* checkpoint is the only way to shrink partial cached entries */
517 if (!f2fs_available_free_memory(sbi, NAT_ENTRIES) ||
518 !f2fs_available_free_memory(sbi, INO_ENTRIES) ||
519 excess_prefree_segs(sbi) ||
520 excess_dirty_nats(sbi) ||
521 excess_dirty_nodes(sbi) ||
522 f2fs_time_over(sbi, CP_TIME)) {
523 if (test_opt(sbi, DATA_FLUSH)) {
524 struct blk_plug plug;
525
526 blk_start_plug(&plug);
527 f2fs_sync_dirty_inodes(sbi, FILE_INODE);
528 blk_finish_plug(&plug);
529 }
530 f2fs_sync_fs(sbi->sb, true);
531 stat_inc_bg_cp_count(sbi->stat_info);
532 }
533 }
534
535 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
536 struct block_device *bdev)
537 {
538 struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
539 int ret;
540
541 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
542 bio_set_dev(bio, bdev);
543 ret = submit_bio_wait(bio);
544 bio_put(bio);
545
546 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
547 test_opt(sbi, FLUSH_MERGE), ret);
548 return ret;
549 }
550
551 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
552 {
553 int ret = 0;
554 int i;
555
556 if (!sbi->s_ndevs)
557 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
558
559 for (i = 0; i < sbi->s_ndevs; i++) {
560 if (!f2fs_is_dirty_device(sbi, ino, i, FLUSH_INO))
561 continue;
562 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
563 if (ret)
564 break;
565 }
566 return ret;
567 }
568
569 static int issue_flush_thread(void *data)
570 {
571 struct f2fs_sb_info *sbi = data;
572 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
573 wait_queue_head_t *q = &fcc->flush_wait_queue;
574 repeat:
575 if (kthread_should_stop())
576 return 0;
577
578 sb_start_intwrite(sbi->sb);
579
580 if (!llist_empty(&fcc->issue_list)) {
581 struct flush_cmd *cmd, *next;
582 int ret;
583
584 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
585 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
586
587 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
588
589 ret = submit_flush_wait(sbi, cmd->ino);
590 atomic_inc(&fcc->issued_flush);
591
592 llist_for_each_entry_safe(cmd, next,
593 fcc->dispatch_list, llnode) {
594 cmd->ret = ret;
595 complete(&cmd->wait);
596 }
597 fcc->dispatch_list = NULL;
598 }
599
600 sb_end_intwrite(sbi->sb);
601
602 wait_event_interruptible(*q,
603 kthread_should_stop() || !llist_empty(&fcc->issue_list));
604 goto repeat;
605 }
606
607 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
608 {
609 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
610 struct flush_cmd cmd;
611 int ret;
612
613 if (test_opt(sbi, NOBARRIER))
614 return 0;
615
616 if (!test_opt(sbi, FLUSH_MERGE)) {
617 ret = submit_flush_wait(sbi, ino);
618 atomic_inc(&fcc->issued_flush);
619 return ret;
620 }
621
622 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
623 ret = submit_flush_wait(sbi, ino);
624 atomic_dec(&fcc->issing_flush);
625
626 atomic_inc(&fcc->issued_flush);
627 return ret;
628 }
629
630 cmd.ino = ino;
631 init_completion(&cmd.wait);
632
633 llist_add(&cmd.llnode, &fcc->issue_list);
634
635 /* update issue_list before we wake up issue_flush thread */
636 smp_mb();
637
638 if (waitqueue_active(&fcc->flush_wait_queue))
639 wake_up(&fcc->flush_wait_queue);
640
641 if (fcc->f2fs_issue_flush) {
642 wait_for_completion(&cmd.wait);
643 atomic_dec(&fcc->issing_flush);
644 } else {
645 struct llist_node *list;
646
647 list = llist_del_all(&fcc->issue_list);
648 if (!list) {
649 wait_for_completion(&cmd.wait);
650 atomic_dec(&fcc->issing_flush);
651 } else {
652 struct flush_cmd *tmp, *next;
653
654 ret = submit_flush_wait(sbi, ino);
655
656 llist_for_each_entry_safe(tmp, next, list, llnode) {
657 if (tmp == &cmd) {
658 cmd.ret = ret;
659 atomic_dec(&fcc->issing_flush);
660 continue;
661 }
662 tmp->ret = ret;
663 complete(&tmp->wait);
664 }
665 }
666 }
667
668 return cmd.ret;
669 }
670
671 int f2fs_create_flush_cmd_control(struct f2fs_sb_info *sbi)
672 {
673 dev_t dev = sbi->sb->s_bdev->bd_dev;
674 struct flush_cmd_control *fcc;
675 int err = 0;
676
677 if (SM_I(sbi)->fcc_info) {
678 fcc = SM_I(sbi)->fcc_info;
679 if (fcc->f2fs_issue_flush)
680 return err;
681 goto init_thread;
682 }
683
684 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
685 if (!fcc)
686 return -ENOMEM;
687 atomic_set(&fcc->issued_flush, 0);
688 atomic_set(&fcc->issing_flush, 0);
689 init_waitqueue_head(&fcc->flush_wait_queue);
690 init_llist_head(&fcc->issue_list);
691 SM_I(sbi)->fcc_info = fcc;
692 if (!test_opt(sbi, FLUSH_MERGE))
693 return err;
694
695 init_thread:
696 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
697 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
698 if (IS_ERR(fcc->f2fs_issue_flush)) {
699 err = PTR_ERR(fcc->f2fs_issue_flush);
700 kfree(fcc);
701 SM_I(sbi)->fcc_info = NULL;
702 return err;
703 }
704
705 return err;
706 }
707
708 void f2fs_destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
709 {
710 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
711
712 if (fcc && fcc->f2fs_issue_flush) {
713 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
714
715 fcc->f2fs_issue_flush = NULL;
716 kthread_stop(flush_thread);
717 }
718 if (free) {
719 kfree(fcc);
720 SM_I(sbi)->fcc_info = NULL;
721 }
722 }
723
724 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
725 {
726 int ret = 0, i;
727
728 if (!sbi->s_ndevs)
729 return 0;
730
731 for (i = 1; i < sbi->s_ndevs; i++) {
732 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
733 continue;
734 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
735 if (ret)
736 break;
737
738 spin_lock(&sbi->dev_lock);
739 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
740 spin_unlock(&sbi->dev_lock);
741 }
742
743 return ret;
744 }
745
746 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
747 enum dirty_type dirty_type)
748 {
749 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
750
751 /* need not be added */
752 if (IS_CURSEG(sbi, segno))
753 return;
754
755 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
756 dirty_i->nr_dirty[dirty_type]++;
757
758 if (dirty_type == DIRTY) {
759 struct seg_entry *sentry = get_seg_entry(sbi, segno);
760 enum dirty_type t = sentry->type;
761
762 if (unlikely(t >= DIRTY)) {
763 f2fs_bug_on(sbi, 1);
764 return;
765 }
766 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
767 dirty_i->nr_dirty[t]++;
768 }
769 }
770
771 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
772 enum dirty_type dirty_type)
773 {
774 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
775
776 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
777 dirty_i->nr_dirty[dirty_type]--;
778
779 if (dirty_type == DIRTY) {
780 struct seg_entry *sentry = get_seg_entry(sbi, segno);
781 enum dirty_type t = sentry->type;
782
783 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
784 dirty_i->nr_dirty[t]--;
785
786 if (get_valid_blocks(sbi, segno, true) == 0)
787 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
788 dirty_i->victim_secmap);
789 }
790 }
791
792 /*
793 * Should not occur error such as -ENOMEM.
794 * Adding dirty entry into seglist is not critical operation.
795 * If a given segment is one of current working segments, it won't be added.
796 */
797 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
798 {
799 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
800 unsigned short valid_blocks;
801
802 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
803 return;
804
805 mutex_lock(&dirty_i->seglist_lock);
806
807 valid_blocks = get_valid_blocks(sbi, segno, false);
808
809 if (valid_blocks == 0) {
810 __locate_dirty_segment(sbi, segno, PRE);
811 __remove_dirty_segment(sbi, segno, DIRTY);
812 } else if (valid_blocks < sbi->blocks_per_seg) {
813 __locate_dirty_segment(sbi, segno, DIRTY);
814 } else {
815 /* Recovery routine with SSR needs this */
816 __remove_dirty_segment(sbi, segno, DIRTY);
817 }
818
819 mutex_unlock(&dirty_i->seglist_lock);
820 }
821
822 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
823 struct block_device *bdev, block_t lstart,
824 block_t start, block_t len)
825 {
826 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
827 struct list_head *pend_list;
828 struct discard_cmd *dc;
829
830 f2fs_bug_on(sbi, !len);
831
832 pend_list = &dcc->pend_list[plist_idx(len)];
833
834 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
835 INIT_LIST_HEAD(&dc->list);
836 dc->bdev = bdev;
837 dc->lstart = lstart;
838 dc->start = start;
839 dc->len = len;
840 dc->ref = 0;
841 dc->state = D_PREP;
842 dc->issuing = 0;
843 dc->error = 0;
844 init_completion(&dc->wait);
845 list_add_tail(&dc->list, pend_list);
846 spin_lock_init(&dc->lock);
847 dc->bio_ref = 0;
848 atomic_inc(&dcc->discard_cmd_cnt);
849 dcc->undiscard_blks += len;
850
851 return dc;
852 }
853
854 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
855 struct block_device *bdev, block_t lstart,
856 block_t start, block_t len,
857 struct rb_node *parent, struct rb_node **p)
858 {
859 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
860 struct discard_cmd *dc;
861
862 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
863
864 rb_link_node(&dc->rb_node, parent, p);
865 rb_insert_color(&dc->rb_node, &dcc->root);
866
867 return dc;
868 }
869
870 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
871 struct discard_cmd *dc)
872 {
873 if (dc->state == D_DONE)
874 atomic_sub(dc->issuing, &dcc->issing_discard);
875
876 list_del(&dc->list);
877 rb_erase(&dc->rb_node, &dcc->root);
878 dcc->undiscard_blks -= dc->len;
879
880 kmem_cache_free(discard_cmd_slab, dc);
881
882 atomic_dec(&dcc->discard_cmd_cnt);
883 }
884
885 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
886 struct discard_cmd *dc)
887 {
888 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
889 unsigned long flags;
890
891 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
892
893 spin_lock_irqsave(&dc->lock, flags);
894 if (dc->bio_ref) {
895 spin_unlock_irqrestore(&dc->lock, flags);
896 return;
897 }
898 spin_unlock_irqrestore(&dc->lock, flags);
899
900 f2fs_bug_on(sbi, dc->ref);
901
902 if (dc->error == -EOPNOTSUPP)
903 dc->error = 0;
904
905 if (dc->error)
906 f2fs_msg(sbi->sb, KERN_INFO,
907 "Issue discard(%u, %u, %u) failed, ret: %d",
908 dc->lstart, dc->start, dc->len, dc->error);
909 __detach_discard_cmd(dcc, dc);
910 }
911
912 static void f2fs_submit_discard_endio(struct bio *bio)
913 {
914 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
915 unsigned long flags;
916
917 dc->error = blk_status_to_errno(bio->bi_status);
918
919 spin_lock_irqsave(&dc->lock, flags);
920 dc->bio_ref--;
921 if (!dc->bio_ref && dc->state == D_SUBMIT) {
922 dc->state = D_DONE;
923 complete_all(&dc->wait);
924 }
925 spin_unlock_irqrestore(&dc->lock, flags);
926 bio_put(bio);
927 }
928
929 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
930 block_t start, block_t end)
931 {
932 #ifdef CONFIG_F2FS_CHECK_FS
933 struct seg_entry *sentry;
934 unsigned int segno;
935 block_t blk = start;
936 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
937 unsigned long *map;
938
939 while (blk < end) {
940 segno = GET_SEGNO(sbi, blk);
941 sentry = get_seg_entry(sbi, segno);
942 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
943
944 if (end < START_BLOCK(sbi, segno + 1))
945 size = GET_BLKOFF_FROM_SEG0(sbi, end);
946 else
947 size = max_blocks;
948 map = (unsigned long *)(sentry->cur_valid_map);
949 offset = __find_rev_next_bit(map, size, offset);
950 f2fs_bug_on(sbi, offset != size);
951 blk = START_BLOCK(sbi, segno + 1);
952 }
953 #endif
954 }
955
956 static void __init_discard_policy(struct f2fs_sb_info *sbi,
957 struct discard_policy *dpolicy,
958 int discard_type, unsigned int granularity)
959 {
960 /* common policy */
961 dpolicy->type = discard_type;
962 dpolicy->sync = true;
963 dpolicy->ordered = false;
964 dpolicy->granularity = granularity;
965
966 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
967 dpolicy->io_aware_gran = MAX_PLIST_NUM;
968
969 if (discard_type == DPOLICY_BG) {
970 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
971 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
972 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
973 dpolicy->io_aware = true;
974 dpolicy->sync = false;
975 dpolicy->ordered = true;
976 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
977 dpolicy->granularity = 1;
978 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
979 }
980 } else if (discard_type == DPOLICY_FORCE) {
981 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
982 dpolicy->mid_interval = DEF_MID_DISCARD_ISSUE_TIME;
983 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
984 dpolicy->io_aware = false;
985 } else if (discard_type == DPOLICY_FSTRIM) {
986 dpolicy->io_aware = false;
987 } else if (discard_type == DPOLICY_UMOUNT) {
988 dpolicy->max_requests = UINT_MAX;
989 dpolicy->io_aware = false;
990 }
991 }
992
993 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
994 struct block_device *bdev, block_t lstart,
995 block_t start, block_t len);
996 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
997 static int __submit_discard_cmd(struct f2fs_sb_info *sbi,
998 struct discard_policy *dpolicy,
999 struct discard_cmd *dc,
1000 unsigned int *issued)
1001 {
1002 struct block_device *bdev = dc->bdev;
1003 struct request_queue *q = bdev_get_queue(bdev);
1004 unsigned int max_discard_blocks =
1005 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1006 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1007 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1008 &(dcc->fstrim_list) : &(dcc->wait_list);
1009 int flag = dpolicy->sync ? REQ_SYNC : 0;
1010 block_t lstart, start, len, total_len;
1011 int err = 0;
1012
1013 if (dc->state != D_PREP)
1014 return 0;
1015
1016 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1017 return 0;
1018
1019 trace_f2fs_issue_discard(bdev, dc->start, dc->len);
1020
1021 lstart = dc->lstart;
1022 start = dc->start;
1023 len = dc->len;
1024 total_len = len;
1025
1026 dc->len = 0;
1027
1028 while (total_len && *issued < dpolicy->max_requests && !err) {
1029 struct bio *bio = NULL;
1030 unsigned long flags;
1031 bool last = true;
1032
1033 if (len > max_discard_blocks) {
1034 len = max_discard_blocks;
1035 last = false;
1036 }
1037
1038 (*issued)++;
1039 if (*issued == dpolicy->max_requests)
1040 last = true;
1041
1042 dc->len += len;
1043
1044 #ifdef CONFIG_F2FS_FAULT_INJECTION
1045 if (time_to_inject(sbi, FAULT_DISCARD)) {
1046 f2fs_show_injection_info(FAULT_DISCARD);
1047 err = -EIO;
1048 goto submit;
1049 }
1050 #endif
1051 err = __blkdev_issue_discard(bdev,
1052 SECTOR_FROM_BLOCK(start),
1053 SECTOR_FROM_BLOCK(len),
1054 GFP_NOFS, 0, &bio);
1055 submit:
1056 if (err) {
1057 spin_lock_irqsave(&dc->lock, flags);
1058 if (dc->state == D_PARTIAL)
1059 dc->state = D_SUBMIT;
1060 spin_unlock_irqrestore(&dc->lock, flags);
1061
1062 break;
1063 }
1064
1065 f2fs_bug_on(sbi, !bio);
1066
1067 /*
1068 * should keep before submission to avoid D_DONE
1069 * right away
1070 */
1071 spin_lock_irqsave(&dc->lock, flags);
1072 if (last)
1073 dc->state = D_SUBMIT;
1074 else
1075 dc->state = D_PARTIAL;
1076 dc->bio_ref++;
1077 spin_unlock_irqrestore(&dc->lock, flags);
1078
1079 atomic_inc(&dcc->issing_discard);
1080 dc->issuing++;
1081 list_move_tail(&dc->list, wait_list);
1082
1083 /* sanity check on discard range */
1084 __check_sit_bitmap(sbi, start, start + len);
1085
1086 bio->bi_private = dc;
1087 bio->bi_end_io = f2fs_submit_discard_endio;
1088 bio->bi_opf |= flag;
1089 submit_bio(bio);
1090
1091 atomic_inc(&dcc->issued_discard);
1092
1093 f2fs_update_iostat(sbi, FS_DISCARD, 1);
1094
1095 lstart += len;
1096 start += len;
1097 total_len -= len;
1098 len = total_len;
1099 }
1100
1101 if (!err && len)
1102 __update_discard_tree_range(sbi, bdev, lstart, start, len);
1103 return err;
1104 }
1105
1106 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
1107 struct block_device *bdev, block_t lstart,
1108 block_t start, block_t len,
1109 struct rb_node **insert_p,
1110 struct rb_node *insert_parent)
1111 {
1112 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1113 struct rb_node **p;
1114 struct rb_node *parent = NULL;
1115 struct discard_cmd *dc = NULL;
1116
1117 if (insert_p && insert_parent) {
1118 parent = insert_parent;
1119 p = insert_p;
1120 goto do_insert;
1121 }
1122
1123 p = f2fs_lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
1124 do_insert:
1125 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1126 if (!dc)
1127 return NULL;
1128
1129 return dc;
1130 }
1131
1132 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1133 struct discard_cmd *dc)
1134 {
1135 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1136 }
1137
1138 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1139 struct discard_cmd *dc, block_t blkaddr)
1140 {
1141 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1142 struct discard_info di = dc->di;
1143 bool modified = false;
1144
1145 if (dc->state == D_DONE || dc->len == 1) {
1146 __remove_discard_cmd(sbi, dc);
1147 return;
1148 }
1149
1150 dcc->undiscard_blks -= di.len;
1151
1152 if (blkaddr > di.lstart) {
1153 dc->len = blkaddr - dc->lstart;
1154 dcc->undiscard_blks += dc->len;
1155 __relocate_discard_cmd(dcc, dc);
1156 modified = true;
1157 }
1158
1159 if (blkaddr < di.lstart + di.len - 1) {
1160 if (modified) {
1161 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1162 di.start + blkaddr + 1 - di.lstart,
1163 di.lstart + di.len - 1 - blkaddr,
1164 NULL, NULL);
1165 } else {
1166 dc->lstart++;
1167 dc->len--;
1168 dc->start++;
1169 dcc->undiscard_blks += dc->len;
1170 __relocate_discard_cmd(dcc, dc);
1171 }
1172 }
1173 }
1174
1175 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1176 struct block_device *bdev, block_t lstart,
1177 block_t start, block_t len)
1178 {
1179 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1180 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1181 struct discard_cmd *dc;
1182 struct discard_info di = {0};
1183 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1184 struct request_queue *q = bdev_get_queue(bdev);
1185 unsigned int max_discard_blocks =
1186 SECTOR_TO_BLOCK(q->limits.max_discard_sectors);
1187 block_t end = lstart + len;
1188
1189 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1190 NULL, lstart,
1191 (struct rb_entry **)&prev_dc,
1192 (struct rb_entry **)&next_dc,
1193 &insert_p, &insert_parent, true);
1194 if (dc)
1195 prev_dc = dc;
1196
1197 if (!prev_dc) {
1198 di.lstart = lstart;
1199 di.len = next_dc ? next_dc->lstart - lstart : len;
1200 di.len = min(di.len, len);
1201 di.start = start;
1202 }
1203
1204 while (1) {
1205 struct rb_node *node;
1206 bool merged = false;
1207 struct discard_cmd *tdc = NULL;
1208
1209 if (prev_dc) {
1210 di.lstart = prev_dc->lstart + prev_dc->len;
1211 if (di.lstart < lstart)
1212 di.lstart = lstart;
1213 if (di.lstart >= end)
1214 break;
1215
1216 if (!next_dc || next_dc->lstart > end)
1217 di.len = end - di.lstart;
1218 else
1219 di.len = next_dc->lstart - di.lstart;
1220 di.start = start + di.lstart - lstart;
1221 }
1222
1223 if (!di.len)
1224 goto next;
1225
1226 if (prev_dc && prev_dc->state == D_PREP &&
1227 prev_dc->bdev == bdev &&
1228 __is_discard_back_mergeable(&di, &prev_dc->di,
1229 max_discard_blocks)) {
1230 prev_dc->di.len += di.len;
1231 dcc->undiscard_blks += di.len;
1232 __relocate_discard_cmd(dcc, prev_dc);
1233 di = prev_dc->di;
1234 tdc = prev_dc;
1235 merged = true;
1236 }
1237
1238 if (next_dc && next_dc->state == D_PREP &&
1239 next_dc->bdev == bdev &&
1240 __is_discard_front_mergeable(&di, &next_dc->di,
1241 max_discard_blocks)) {
1242 next_dc->di.lstart = di.lstart;
1243 next_dc->di.len += di.len;
1244 next_dc->di.start = di.start;
1245 dcc->undiscard_blks += di.len;
1246 __relocate_discard_cmd(dcc, next_dc);
1247 if (tdc)
1248 __remove_discard_cmd(sbi, tdc);
1249 merged = true;
1250 }
1251
1252 if (!merged) {
1253 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1254 di.len, NULL, NULL);
1255 }
1256 next:
1257 prev_dc = next_dc;
1258 if (!prev_dc)
1259 break;
1260
1261 node = rb_next(&prev_dc->rb_node);
1262 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1263 }
1264 }
1265
1266 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1267 struct block_device *bdev, block_t blkstart, block_t blklen)
1268 {
1269 block_t lblkstart = blkstart;
1270
1271 trace_f2fs_queue_discard(bdev, blkstart, blklen);
1272
1273 if (sbi->s_ndevs) {
1274 int devi = f2fs_target_device_index(sbi, blkstart);
1275
1276 blkstart -= FDEV(devi).start_blk;
1277 }
1278 mutex_lock(&SM_I(sbi)->dcc_info->cmd_lock);
1279 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1280 mutex_unlock(&SM_I(sbi)->dcc_info->cmd_lock);
1281 return 0;
1282 }
1283
1284 static unsigned int __issue_discard_cmd_orderly(struct f2fs_sb_info *sbi,
1285 struct discard_policy *dpolicy)
1286 {
1287 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1288 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1289 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1290 struct discard_cmd *dc;
1291 struct blk_plug plug;
1292 unsigned int pos = dcc->next_pos;
1293 unsigned int issued = 0;
1294 bool io_interrupted = false;
1295
1296 mutex_lock(&dcc->cmd_lock);
1297 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
1298 NULL, pos,
1299 (struct rb_entry **)&prev_dc,
1300 (struct rb_entry **)&next_dc,
1301 &insert_p, &insert_parent, true);
1302 if (!dc)
1303 dc = next_dc;
1304
1305 blk_start_plug(&plug);
1306
1307 while (dc) {
1308 struct rb_node *node;
1309 int err = 0;
1310
1311 if (dc->state != D_PREP)
1312 goto next;
1313
1314 if (dpolicy->io_aware && !is_idle(sbi)) {
1315 io_interrupted = true;
1316 break;
1317 }
1318
1319 dcc->next_pos = dc->lstart + dc->len;
1320 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1321
1322 if (issued >= dpolicy->max_requests)
1323 break;
1324 next:
1325 node = rb_next(&dc->rb_node);
1326 if (err)
1327 __remove_discard_cmd(sbi, dc);
1328 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1329 }
1330
1331 blk_finish_plug(&plug);
1332
1333 if (!dc)
1334 dcc->next_pos = 0;
1335
1336 mutex_unlock(&dcc->cmd_lock);
1337
1338 if (!issued && io_interrupted)
1339 issued = -1;
1340
1341 return issued;
1342 }
1343
1344 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1345 struct discard_policy *dpolicy)
1346 {
1347 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1348 struct list_head *pend_list;
1349 struct discard_cmd *dc, *tmp;
1350 struct blk_plug plug;
1351 int i, issued = 0;
1352 bool io_interrupted = false;
1353
1354 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1355 if (i + 1 < dpolicy->granularity)
1356 break;
1357
1358 if (i < DEFAULT_DISCARD_GRANULARITY && dpolicy->ordered)
1359 return __issue_discard_cmd_orderly(sbi, dpolicy);
1360
1361 pend_list = &dcc->pend_list[i];
1362
1363 mutex_lock(&dcc->cmd_lock);
1364 if (list_empty(pend_list))
1365 goto next;
1366 if (unlikely(dcc->rbtree_check))
1367 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
1368 &dcc->root));
1369 blk_start_plug(&plug);
1370 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1371 f2fs_bug_on(sbi, dc->state != D_PREP);
1372
1373 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1374 !is_idle(sbi)) {
1375 io_interrupted = true;
1376 break;
1377 }
1378
1379 __submit_discard_cmd(sbi, dpolicy, dc, &issued);
1380
1381 if (issued >= dpolicy->max_requests)
1382 break;
1383 }
1384 blk_finish_plug(&plug);
1385 next:
1386 mutex_unlock(&dcc->cmd_lock);
1387
1388 if (issued >= dpolicy->max_requests || io_interrupted)
1389 break;
1390 }
1391
1392 if (!issued && io_interrupted)
1393 issued = -1;
1394
1395 return issued;
1396 }
1397
1398 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1399 {
1400 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1401 struct list_head *pend_list;
1402 struct discard_cmd *dc, *tmp;
1403 int i;
1404 bool dropped = false;
1405
1406 mutex_lock(&dcc->cmd_lock);
1407 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1408 pend_list = &dcc->pend_list[i];
1409 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1410 f2fs_bug_on(sbi, dc->state != D_PREP);
1411 __remove_discard_cmd(sbi, dc);
1412 dropped = true;
1413 }
1414 }
1415 mutex_unlock(&dcc->cmd_lock);
1416
1417 return dropped;
1418 }
1419
1420 void f2fs_drop_discard_cmd(struct f2fs_sb_info *sbi)
1421 {
1422 __drop_discard_cmd(sbi);
1423 }
1424
1425 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1426 struct discard_cmd *dc)
1427 {
1428 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1429 unsigned int len = 0;
1430
1431 wait_for_completion_io(&dc->wait);
1432 mutex_lock(&dcc->cmd_lock);
1433 f2fs_bug_on(sbi, dc->state != D_DONE);
1434 dc->ref--;
1435 if (!dc->ref) {
1436 if (!dc->error)
1437 len = dc->len;
1438 __remove_discard_cmd(sbi, dc);
1439 }
1440 mutex_unlock(&dcc->cmd_lock);
1441
1442 return len;
1443 }
1444
1445 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1446 struct discard_policy *dpolicy,
1447 block_t start, block_t end)
1448 {
1449 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1450 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1451 &(dcc->fstrim_list) : &(dcc->wait_list);
1452 struct discard_cmd *dc, *tmp;
1453 bool need_wait;
1454 unsigned int trimmed = 0;
1455
1456 next:
1457 need_wait = false;
1458
1459 mutex_lock(&dcc->cmd_lock);
1460 list_for_each_entry_safe(dc, tmp, wait_list, list) {
1461 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1462 continue;
1463 if (dc->len < dpolicy->granularity)
1464 continue;
1465 if (dc->state == D_DONE && !dc->ref) {
1466 wait_for_completion_io(&dc->wait);
1467 if (!dc->error)
1468 trimmed += dc->len;
1469 __remove_discard_cmd(sbi, dc);
1470 } else {
1471 dc->ref++;
1472 need_wait = true;
1473 break;
1474 }
1475 }
1476 mutex_unlock(&dcc->cmd_lock);
1477
1478 if (need_wait) {
1479 trimmed += __wait_one_discard_bio(sbi, dc);
1480 goto next;
1481 }
1482
1483 return trimmed;
1484 }
1485
1486 static unsigned int __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1487 struct discard_policy *dpolicy)
1488 {
1489 struct discard_policy dp;
1490 unsigned int discard_blks;
1491
1492 if (dpolicy)
1493 return __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1494
1495 /* wait all */
1496 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1497 discard_blks = __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1498 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1499 discard_blks += __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1500
1501 return discard_blks;
1502 }
1503
1504 /* This should be covered by global mutex, &sit_i->sentry_lock */
1505 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1506 {
1507 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1508 struct discard_cmd *dc;
1509 bool need_wait = false;
1510
1511 mutex_lock(&dcc->cmd_lock);
1512 dc = (struct discard_cmd *)f2fs_lookup_rb_tree(&dcc->root,
1513 NULL, blkaddr);
1514 if (dc) {
1515 if (dc->state == D_PREP) {
1516 __punch_discard_cmd(sbi, dc, blkaddr);
1517 } else {
1518 dc->ref++;
1519 need_wait = true;
1520 }
1521 }
1522 mutex_unlock(&dcc->cmd_lock);
1523
1524 if (need_wait)
1525 __wait_one_discard_bio(sbi, dc);
1526 }
1527
1528 void f2fs_stop_discard_thread(struct f2fs_sb_info *sbi)
1529 {
1530 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1531
1532 if (dcc && dcc->f2fs_issue_discard) {
1533 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1534
1535 dcc->f2fs_issue_discard = NULL;
1536 kthread_stop(discard_thread);
1537 }
1538 }
1539
1540 /* This comes from f2fs_put_super */
1541 bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1542 {
1543 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1544 struct discard_policy dpolicy;
1545 bool dropped;
1546
1547 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1548 dcc->discard_granularity);
1549 __issue_discard_cmd(sbi, &dpolicy);
1550 dropped = __drop_discard_cmd(sbi);
1551
1552 /* just to make sure there is no pending discard commands */
1553 __wait_all_discard_cmd(sbi, NULL);
1554
1555 f2fs_bug_on(sbi, atomic_read(&dcc->discard_cmd_cnt));
1556 return dropped;
1557 }
1558
1559 static int issue_discard_thread(void *data)
1560 {
1561 struct f2fs_sb_info *sbi = data;
1562 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1563 wait_queue_head_t *q = &dcc->discard_wait_queue;
1564 struct discard_policy dpolicy;
1565 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1566 int issued;
1567
1568 set_freezable();
1569
1570 do {
1571 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1572 dcc->discard_granularity);
1573
1574 wait_event_interruptible_timeout(*q,
1575 kthread_should_stop() || freezing(current) ||
1576 dcc->discard_wake,
1577 msecs_to_jiffies(wait_ms));
1578
1579 if (dcc->discard_wake)
1580 dcc->discard_wake = 0;
1581
1582 if (try_to_freeze())
1583 continue;
1584 if (f2fs_readonly(sbi->sb))
1585 continue;
1586 if (kthread_should_stop())
1587 return 0;
1588 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1589 wait_ms = dpolicy.max_interval;
1590 continue;
1591 }
1592
1593 if (sbi->gc_mode == GC_URGENT)
1594 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1595
1596 sb_start_intwrite(sbi->sb);
1597
1598 issued = __issue_discard_cmd(sbi, &dpolicy);
1599 if (issued > 0) {
1600 __wait_all_discard_cmd(sbi, &dpolicy);
1601 wait_ms = dpolicy.min_interval;
1602 } else if (issued == -1){
1603 wait_ms = dpolicy.mid_interval;
1604 } else {
1605 wait_ms = dpolicy.max_interval;
1606 }
1607
1608 sb_end_intwrite(sbi->sb);
1609
1610 } while (!kthread_should_stop());
1611 return 0;
1612 }
1613
1614 #ifdef CONFIG_BLK_DEV_ZONED
1615 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1616 struct block_device *bdev, block_t blkstart, block_t blklen)
1617 {
1618 sector_t sector, nr_sects;
1619 block_t lblkstart = blkstart;
1620 int devi = 0;
1621
1622 if (sbi->s_ndevs) {
1623 devi = f2fs_target_device_index(sbi, blkstart);
1624 blkstart -= FDEV(devi).start_blk;
1625 }
1626
1627 /*
1628 * We need to know the type of the zone: for conventional zones,
1629 * use regular discard if the drive supports it. For sequential
1630 * zones, reset the zone write pointer.
1631 */
1632 switch (get_blkz_type(sbi, bdev, blkstart)) {
1633
1634 case BLK_ZONE_TYPE_CONVENTIONAL:
1635 if (!blk_queue_discard(bdev_get_queue(bdev)))
1636 return 0;
1637 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1638 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1639 case BLK_ZONE_TYPE_SEQWRITE_PREF:
1640 sector = SECTOR_FROM_BLOCK(blkstart);
1641 nr_sects = SECTOR_FROM_BLOCK(blklen);
1642
1643 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1644 nr_sects != bdev_zone_sectors(bdev)) {
1645 f2fs_msg(sbi->sb, KERN_INFO,
1646 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1647 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1648 blkstart, blklen);
1649 return -EIO;
1650 }
1651 trace_f2fs_issue_reset_zone(bdev, blkstart);
1652 return blkdev_reset_zones(bdev, sector,
1653 nr_sects, GFP_NOFS);
1654 default:
1655 /* Unknown zone type: broken device ? */
1656 return -EIO;
1657 }
1658 }
1659 #endif
1660
1661 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1662 struct block_device *bdev, block_t blkstart, block_t blklen)
1663 {
1664 #ifdef CONFIG_BLK_DEV_ZONED
1665 if (f2fs_sb_has_blkzoned(sbi->sb) &&
1666 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1667 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1668 #endif
1669 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1670 }
1671
1672 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1673 block_t blkstart, block_t blklen)
1674 {
1675 sector_t start = blkstart, len = 0;
1676 struct block_device *bdev;
1677 struct seg_entry *se;
1678 unsigned int offset;
1679 block_t i;
1680 int err = 0;
1681
1682 bdev = f2fs_target_device(sbi, blkstart, NULL);
1683
1684 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1685 if (i != start) {
1686 struct block_device *bdev2 =
1687 f2fs_target_device(sbi, i, NULL);
1688
1689 if (bdev2 != bdev) {
1690 err = __issue_discard_async(sbi, bdev,
1691 start, len);
1692 if (err)
1693 return err;
1694 bdev = bdev2;
1695 start = i;
1696 len = 0;
1697 }
1698 }
1699
1700 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1701 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1702
1703 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1704 sbi->discard_blks--;
1705 }
1706
1707 if (len)
1708 err = __issue_discard_async(sbi, bdev, start, len);
1709 return err;
1710 }
1711
1712 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1713 bool check_only)
1714 {
1715 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1716 int max_blocks = sbi->blocks_per_seg;
1717 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1718 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1719 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1720 unsigned long *discard_map = (unsigned long *)se->discard_map;
1721 unsigned long *dmap = SIT_I(sbi)->tmp_map;
1722 unsigned int start = 0, end = -1;
1723 bool force = (cpc->reason & CP_DISCARD);
1724 struct discard_entry *de = NULL;
1725 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1726 int i;
1727
1728 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1729 return false;
1730
1731 if (!force) {
1732 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1733 SM_I(sbi)->dcc_info->nr_discards >=
1734 SM_I(sbi)->dcc_info->max_discards)
1735 return false;
1736 }
1737
1738 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1739 for (i = 0; i < entries; i++)
1740 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1741 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1742
1743 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1744 SM_I(sbi)->dcc_info->max_discards) {
1745 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1746 if (start >= max_blocks)
1747 break;
1748
1749 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1750 if (force && start && end != max_blocks
1751 && (end - start) < cpc->trim_minlen)
1752 continue;
1753
1754 if (check_only)
1755 return true;
1756
1757 if (!de) {
1758 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1759 GFP_F2FS_ZERO);
1760 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1761 list_add_tail(&de->list, head);
1762 }
1763
1764 for (i = start; i < end; i++)
1765 __set_bit_le(i, (void *)de->discard_map);
1766
1767 SM_I(sbi)->dcc_info->nr_discards += end - start;
1768 }
1769 return false;
1770 }
1771
1772 static void release_discard_addr(struct discard_entry *entry)
1773 {
1774 list_del(&entry->list);
1775 kmem_cache_free(discard_entry_slab, entry);
1776 }
1777
1778 void f2fs_release_discard_addrs(struct f2fs_sb_info *sbi)
1779 {
1780 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1781 struct discard_entry *entry, *this;
1782
1783 /* drop caches */
1784 list_for_each_entry_safe(entry, this, head, list)
1785 release_discard_addr(entry);
1786 }
1787
1788 /*
1789 * Should call f2fs_clear_prefree_segments after checkpoint is done.
1790 */
1791 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1792 {
1793 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1794 unsigned int segno;
1795
1796 mutex_lock(&dirty_i->seglist_lock);
1797 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1798 __set_test_and_free(sbi, segno);
1799 mutex_unlock(&dirty_i->seglist_lock);
1800 }
1801
1802 void f2fs_clear_prefree_segments(struct f2fs_sb_info *sbi,
1803 struct cp_control *cpc)
1804 {
1805 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1806 struct list_head *head = &dcc->entry_list;
1807 struct discard_entry *entry, *this;
1808 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1809 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1810 unsigned int start = 0, end = -1;
1811 unsigned int secno, start_segno;
1812 bool force = (cpc->reason & CP_DISCARD);
1813 bool need_align = test_opt(sbi, LFS) && sbi->segs_per_sec > 1;
1814
1815 mutex_lock(&dirty_i->seglist_lock);
1816
1817 while (1) {
1818 int i;
1819
1820 if (need_align && end != -1)
1821 end--;
1822 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1823 if (start >= MAIN_SEGS(sbi))
1824 break;
1825 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1826 start + 1);
1827
1828 if (need_align) {
1829 start = rounddown(start, sbi->segs_per_sec);
1830 end = roundup(end, sbi->segs_per_sec);
1831 }
1832
1833 for (i = start; i < end; i++) {
1834 if (test_and_clear_bit(i, prefree_map))
1835 dirty_i->nr_dirty[PRE]--;
1836 }
1837
1838 if (!test_opt(sbi, DISCARD))
1839 continue;
1840
1841 if (force && start >= cpc->trim_start &&
1842 (end - 1) <= cpc->trim_end)
1843 continue;
1844
1845 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1846 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1847 (end - start) << sbi->log_blocks_per_seg);
1848 continue;
1849 }
1850 next:
1851 secno = GET_SEC_FROM_SEG(sbi, start);
1852 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1853 if (!IS_CURSEC(sbi, secno) &&
1854 !get_valid_blocks(sbi, start, true))
1855 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1856 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1857
1858 start = start_segno + sbi->segs_per_sec;
1859 if (start < end)
1860 goto next;
1861 else
1862 end = start - 1;
1863 }
1864 mutex_unlock(&dirty_i->seglist_lock);
1865
1866 /* send small discards */
1867 list_for_each_entry_safe(entry, this, head, list) {
1868 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1869 bool is_valid = test_bit_le(0, entry->discard_map);
1870
1871 find_next:
1872 if (is_valid) {
1873 next_pos = find_next_zero_bit_le(entry->discard_map,
1874 sbi->blocks_per_seg, cur_pos);
1875 len = next_pos - cur_pos;
1876
1877 if (f2fs_sb_has_blkzoned(sbi->sb) ||
1878 (force && len < cpc->trim_minlen))
1879 goto skip;
1880
1881 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1882 len);
1883 total_len += len;
1884 } else {
1885 next_pos = find_next_bit_le(entry->discard_map,
1886 sbi->blocks_per_seg, cur_pos);
1887 }
1888 skip:
1889 cur_pos = next_pos;
1890 is_valid = !is_valid;
1891
1892 if (cur_pos < sbi->blocks_per_seg)
1893 goto find_next;
1894
1895 release_discard_addr(entry);
1896 dcc->nr_discards -= total_len;
1897 }
1898
1899 wake_up_discard_thread(sbi, false);
1900 }
1901
1902 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1903 {
1904 dev_t dev = sbi->sb->s_bdev->bd_dev;
1905 struct discard_cmd_control *dcc;
1906 int err = 0, i;
1907
1908 if (SM_I(sbi)->dcc_info) {
1909 dcc = SM_I(sbi)->dcc_info;
1910 goto init_thread;
1911 }
1912
1913 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
1914 if (!dcc)
1915 return -ENOMEM;
1916
1917 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
1918 INIT_LIST_HEAD(&dcc->entry_list);
1919 for (i = 0; i < MAX_PLIST_NUM; i++)
1920 INIT_LIST_HEAD(&dcc->pend_list[i]);
1921 INIT_LIST_HEAD(&dcc->wait_list);
1922 INIT_LIST_HEAD(&dcc->fstrim_list);
1923 mutex_init(&dcc->cmd_lock);
1924 atomic_set(&dcc->issued_discard, 0);
1925 atomic_set(&dcc->issing_discard, 0);
1926 atomic_set(&dcc->discard_cmd_cnt, 0);
1927 dcc->nr_discards = 0;
1928 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1929 dcc->undiscard_blks = 0;
1930 dcc->next_pos = 0;
1931 dcc->root = RB_ROOT;
1932 dcc->rbtree_check = false;
1933
1934 init_waitqueue_head(&dcc->discard_wait_queue);
1935 SM_I(sbi)->dcc_info = dcc;
1936 init_thread:
1937 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1938 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1939 if (IS_ERR(dcc->f2fs_issue_discard)) {
1940 err = PTR_ERR(dcc->f2fs_issue_discard);
1941 kfree(dcc);
1942 SM_I(sbi)->dcc_info = NULL;
1943 return err;
1944 }
1945
1946 return err;
1947 }
1948
1949 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1950 {
1951 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1952
1953 if (!dcc)
1954 return;
1955
1956 f2fs_stop_discard_thread(sbi);
1957
1958 kfree(dcc);
1959 SM_I(sbi)->dcc_info = NULL;
1960 }
1961
1962 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1963 {
1964 struct sit_info *sit_i = SIT_I(sbi);
1965
1966 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1967 sit_i->dirty_sentries++;
1968 return false;
1969 }
1970
1971 return true;
1972 }
1973
1974 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1975 unsigned int segno, int modified)
1976 {
1977 struct seg_entry *se = get_seg_entry(sbi, segno);
1978 se->type = type;
1979 if (modified)
1980 __mark_sit_entry_dirty(sbi, segno);
1981 }
1982
1983 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1984 {
1985 struct seg_entry *se;
1986 unsigned int segno, offset;
1987 long int new_vblocks;
1988 bool exist;
1989 #ifdef CONFIG_F2FS_CHECK_FS
1990 bool mir_exist;
1991 #endif
1992
1993 segno = GET_SEGNO(sbi, blkaddr);
1994
1995 se = get_seg_entry(sbi, segno);
1996 new_vblocks = se->valid_blocks + del;
1997 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1998
1999 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
2000 (new_vblocks > sbi->blocks_per_seg)));
2001
2002 se->valid_blocks = new_vblocks;
2003 se->mtime = get_mtime(sbi, false);
2004 if (se->mtime > SIT_I(sbi)->max_mtime)
2005 SIT_I(sbi)->max_mtime = se->mtime;
2006
2007 /* Update valid block bitmap */
2008 if (del > 0) {
2009 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
2010 #ifdef CONFIG_F2FS_CHECK_FS
2011 mir_exist = f2fs_test_and_set_bit(offset,
2012 se->cur_valid_map_mir);
2013 if (unlikely(exist != mir_exist)) {
2014 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2015 "when setting bitmap, blk:%u, old bit:%d",
2016 blkaddr, exist);
2017 f2fs_bug_on(sbi, 1);
2018 }
2019 #endif
2020 if (unlikely(exist)) {
2021 f2fs_msg(sbi->sb, KERN_ERR,
2022 "Bitmap was wrongly set, blk:%u", blkaddr);
2023 f2fs_bug_on(sbi, 1);
2024 se->valid_blocks--;
2025 del = 0;
2026 }
2027
2028 if (f2fs_discard_en(sbi) &&
2029 !f2fs_test_and_set_bit(offset, se->discard_map))
2030 sbi->discard_blks--;
2031
2032 /* don't overwrite by SSR to keep node chain */
2033 if (IS_NODESEG(se->type)) {
2034 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
2035 se->ckpt_valid_blocks++;
2036 }
2037 } else {
2038 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
2039 #ifdef CONFIG_F2FS_CHECK_FS
2040 mir_exist = f2fs_test_and_clear_bit(offset,
2041 se->cur_valid_map_mir);
2042 if (unlikely(exist != mir_exist)) {
2043 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
2044 "when clearing bitmap, blk:%u, old bit:%d",
2045 blkaddr, exist);
2046 f2fs_bug_on(sbi, 1);
2047 }
2048 #endif
2049 if (unlikely(!exist)) {
2050 f2fs_msg(sbi->sb, KERN_ERR,
2051 "Bitmap was wrongly cleared, blk:%u", blkaddr);
2052 f2fs_bug_on(sbi, 1);
2053 se->valid_blocks++;
2054 del = 0;
2055 }
2056
2057 if (f2fs_discard_en(sbi) &&
2058 f2fs_test_and_clear_bit(offset, se->discard_map))
2059 sbi->discard_blks++;
2060 }
2061 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
2062 se->ckpt_valid_blocks += del;
2063
2064 __mark_sit_entry_dirty(sbi, segno);
2065
2066 /* update total number of valid blocks to be written in ckpt area */
2067 SIT_I(sbi)->written_valid_blocks += del;
2068
2069 if (sbi->segs_per_sec > 1)
2070 get_sec_entry(sbi, segno)->valid_blocks += del;
2071 }
2072
2073 void f2fs_invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
2074 {
2075 unsigned int segno = GET_SEGNO(sbi, addr);
2076 struct sit_info *sit_i = SIT_I(sbi);
2077
2078 f2fs_bug_on(sbi, addr == NULL_ADDR);
2079 if (addr == NEW_ADDR)
2080 return;
2081
2082 /* add it into sit main buffer */
2083 down_write(&sit_i->sentry_lock);
2084
2085 update_sit_entry(sbi, addr, -1);
2086
2087 /* add it into dirty seglist */
2088 locate_dirty_segment(sbi, segno);
2089
2090 up_write(&sit_i->sentry_lock);
2091 }
2092
2093 bool f2fs_is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
2094 {
2095 struct sit_info *sit_i = SIT_I(sbi);
2096 unsigned int segno, offset;
2097 struct seg_entry *se;
2098 bool is_cp = false;
2099
2100 if (!is_valid_data_blkaddr(sbi, blkaddr))
2101 return true;
2102
2103 down_read(&sit_i->sentry_lock);
2104
2105 segno = GET_SEGNO(sbi, blkaddr);
2106 se = get_seg_entry(sbi, segno);
2107 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
2108
2109 if (f2fs_test_bit(offset, se->ckpt_valid_map))
2110 is_cp = true;
2111
2112 up_read(&sit_i->sentry_lock);
2113
2114 return is_cp;
2115 }
2116
2117 /*
2118 * This function should be resided under the curseg_mutex lock
2119 */
2120 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
2121 struct f2fs_summary *sum)
2122 {
2123 struct curseg_info *curseg = CURSEG_I(sbi, type);
2124 void *addr = curseg->sum_blk;
2125 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
2126 memcpy(addr, sum, sizeof(struct f2fs_summary));
2127 }
2128
2129 /*
2130 * Calculate the number of current summary pages for writing
2131 */
2132 int f2fs_npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
2133 {
2134 int valid_sum_count = 0;
2135 int i, sum_in_page;
2136
2137 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2138 if (sbi->ckpt->alloc_type[i] == SSR)
2139 valid_sum_count += sbi->blocks_per_seg;
2140 else {
2141 if (for_ra)
2142 valid_sum_count += le16_to_cpu(
2143 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
2144 else
2145 valid_sum_count += curseg_blkoff(sbi, i);
2146 }
2147 }
2148
2149 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
2150 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
2151 if (valid_sum_count <= sum_in_page)
2152 return 1;
2153 else if ((valid_sum_count - sum_in_page) <=
2154 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
2155 return 2;
2156 return 3;
2157 }
2158
2159 /*
2160 * Caller should put this summary page
2161 */
2162 struct page *f2fs_get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
2163 {
2164 return f2fs_get_meta_page_nofail(sbi, GET_SUM_BLOCK(sbi, segno));
2165 }
2166
2167 void f2fs_update_meta_page(struct f2fs_sb_info *sbi,
2168 void *src, block_t blk_addr)
2169 {
2170 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2171
2172 memcpy(page_address(page), src, PAGE_SIZE);
2173 set_page_dirty(page);
2174 f2fs_put_page(page, 1);
2175 }
2176
2177 static void write_sum_page(struct f2fs_sb_info *sbi,
2178 struct f2fs_summary_block *sum_blk, block_t blk_addr)
2179 {
2180 f2fs_update_meta_page(sbi, (void *)sum_blk, blk_addr);
2181 }
2182
2183 static void write_current_sum_page(struct f2fs_sb_info *sbi,
2184 int type, block_t blk_addr)
2185 {
2186 struct curseg_info *curseg = CURSEG_I(sbi, type);
2187 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
2188 struct f2fs_summary_block *src = curseg->sum_blk;
2189 struct f2fs_summary_block *dst;
2190
2191 dst = (struct f2fs_summary_block *)page_address(page);
2192 memset(dst, 0, PAGE_SIZE);
2193
2194 mutex_lock(&curseg->curseg_mutex);
2195
2196 down_read(&curseg->journal_rwsem);
2197 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
2198 up_read(&curseg->journal_rwsem);
2199
2200 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
2201 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
2202
2203 mutex_unlock(&curseg->curseg_mutex);
2204
2205 set_page_dirty(page);
2206 f2fs_put_page(page, 1);
2207 }
2208
2209 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2210 {
2211 struct curseg_info *curseg = CURSEG_I(sbi, type);
2212 unsigned int segno = curseg->segno + 1;
2213 struct free_segmap_info *free_i = FREE_I(sbi);
2214
2215 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2216 return !test_bit(segno, free_i->free_segmap);
2217 return 0;
2218 }
2219
2220 /*
2221 * Find a new segment from the free segments bitmap to right order
2222 * This function should be returned with success, otherwise BUG
2223 */
2224 static void get_new_segment(struct f2fs_sb_info *sbi,
2225 unsigned int *newseg, bool new_sec, int dir)
2226 {
2227 struct free_segmap_info *free_i = FREE_I(sbi);
2228 unsigned int segno, secno, zoneno;
2229 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2230 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2231 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2232 unsigned int left_start = hint;
2233 bool init = true;
2234 int go_left = 0;
2235 int i;
2236
2237 spin_lock(&free_i->segmap_lock);
2238
2239 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2240 segno = find_next_zero_bit(free_i->free_segmap,
2241 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2242 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2243 goto got_it;
2244 }
2245 find_other_zone:
2246 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2247 if (secno >= MAIN_SECS(sbi)) {
2248 if (dir == ALLOC_RIGHT) {
2249 secno = find_next_zero_bit(free_i->free_secmap,
2250 MAIN_SECS(sbi), 0);
2251 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2252 } else {
2253 go_left = 1;
2254 left_start = hint - 1;
2255 }
2256 }
2257 if (go_left == 0)
2258 goto skip_left;
2259
2260 while (test_bit(left_start, free_i->free_secmap)) {
2261 if (left_start > 0) {
2262 left_start--;
2263 continue;
2264 }
2265 left_start = find_next_zero_bit(free_i->free_secmap,
2266 MAIN_SECS(sbi), 0);
2267 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2268 break;
2269 }
2270 secno = left_start;
2271 skip_left:
2272 segno = GET_SEG_FROM_SEC(sbi, secno);
2273 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2274
2275 /* give up on finding another zone */
2276 if (!init)
2277 goto got_it;
2278 if (sbi->secs_per_zone == 1)
2279 goto got_it;
2280 if (zoneno == old_zoneno)
2281 goto got_it;
2282 if (dir == ALLOC_LEFT) {
2283 if (!go_left && zoneno + 1 >= total_zones)
2284 goto got_it;
2285 if (go_left && zoneno == 0)
2286 goto got_it;
2287 }
2288 for (i = 0; i < NR_CURSEG_TYPE; i++)
2289 if (CURSEG_I(sbi, i)->zone == zoneno)
2290 break;
2291
2292 if (i < NR_CURSEG_TYPE) {
2293 /* zone is in user, try another */
2294 if (go_left)
2295 hint = zoneno * sbi->secs_per_zone - 1;
2296 else if (zoneno + 1 >= total_zones)
2297 hint = 0;
2298 else
2299 hint = (zoneno + 1) * sbi->secs_per_zone;
2300 init = false;
2301 goto find_other_zone;
2302 }
2303 got_it:
2304 /* set it as dirty segment in free segmap */
2305 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2306 __set_inuse(sbi, segno);
2307 *newseg = segno;
2308 spin_unlock(&free_i->segmap_lock);
2309 }
2310
2311 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2312 {
2313 struct curseg_info *curseg = CURSEG_I(sbi, type);
2314 struct summary_footer *sum_footer;
2315
2316 curseg->segno = curseg->next_segno;
2317 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2318 curseg->next_blkoff = 0;
2319 curseg->next_segno = NULL_SEGNO;
2320
2321 sum_footer = &(curseg->sum_blk->footer);
2322 memset(sum_footer, 0, sizeof(struct summary_footer));
2323 if (IS_DATASEG(type))
2324 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2325 if (IS_NODESEG(type))
2326 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2327 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2328 }
2329
2330 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2331 {
2332 /* if segs_per_sec is large than 1, we need to keep original policy. */
2333 if (sbi->segs_per_sec != 1)
2334 return CURSEG_I(sbi, type)->segno;
2335
2336 if (test_opt(sbi, NOHEAP) &&
2337 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2338 return 0;
2339
2340 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2341 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2342
2343 /* find segments from 0 to reuse freed segments */
2344 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2345 return 0;
2346
2347 return CURSEG_I(sbi, type)->segno;
2348 }
2349
2350 /*
2351 * Allocate a current working segment.
2352 * This function always allocates a free segment in LFS manner.
2353 */
2354 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2355 {
2356 struct curseg_info *curseg = CURSEG_I(sbi, type);
2357 unsigned int segno = curseg->segno;
2358 int dir = ALLOC_LEFT;
2359
2360 write_sum_page(sbi, curseg->sum_blk,
2361 GET_SUM_BLOCK(sbi, segno));
2362 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2363 dir = ALLOC_RIGHT;
2364
2365 if (test_opt(sbi, NOHEAP))
2366 dir = ALLOC_RIGHT;
2367
2368 segno = __get_next_segno(sbi, type);
2369 get_new_segment(sbi, &segno, new_sec, dir);
2370 curseg->next_segno = segno;
2371 reset_curseg(sbi, type, 1);
2372 curseg->alloc_type = LFS;
2373 }
2374
2375 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2376 struct curseg_info *seg, block_t start)
2377 {
2378 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2379 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2380 unsigned long *target_map = SIT_I(sbi)->tmp_map;
2381 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2382 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2383 int i, pos;
2384
2385 for (i = 0; i < entries; i++)
2386 target_map[i] = ckpt_map[i] | cur_map[i];
2387
2388 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2389
2390 seg->next_blkoff = pos;
2391 }
2392
2393 /*
2394 * If a segment is written by LFS manner, next block offset is just obtained
2395 * by increasing the current block offset. However, if a segment is written by
2396 * SSR manner, next block offset obtained by calling __next_free_blkoff
2397 */
2398 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2399 struct curseg_info *seg)
2400 {
2401 if (seg->alloc_type == SSR)
2402 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2403 else
2404 seg->next_blkoff++;
2405 }
2406
2407 /*
2408 * This function always allocates a used segment(from dirty seglist) by SSR
2409 * manner, so it should recover the existing segment information of valid blocks
2410 */
2411 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2412 {
2413 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2414 struct curseg_info *curseg = CURSEG_I(sbi, type);
2415 unsigned int new_segno = curseg->next_segno;
2416 struct f2fs_summary_block *sum_node;
2417 struct page *sum_page;
2418
2419 write_sum_page(sbi, curseg->sum_blk,
2420 GET_SUM_BLOCK(sbi, curseg->segno));
2421 __set_test_and_inuse(sbi, new_segno);
2422
2423 mutex_lock(&dirty_i->seglist_lock);
2424 __remove_dirty_segment(sbi, new_segno, PRE);
2425 __remove_dirty_segment(sbi, new_segno, DIRTY);
2426 mutex_unlock(&dirty_i->seglist_lock);
2427
2428 reset_curseg(sbi, type, 1);
2429 curseg->alloc_type = SSR;
2430 __next_free_blkoff(sbi, curseg, 0);
2431
2432 sum_page = f2fs_get_sum_page(sbi, new_segno);
2433 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2434 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2435 f2fs_put_page(sum_page, 1);
2436 }
2437
2438 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2439 {
2440 struct curseg_info *curseg = CURSEG_I(sbi, type);
2441 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2442 unsigned segno = NULL_SEGNO;
2443 int i, cnt;
2444 bool reversed = false;
2445
2446 /* f2fs_need_SSR() already forces to do this */
2447 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2448 curseg->next_segno = segno;
2449 return 1;
2450 }
2451
2452 /* For node segments, let's do SSR more intensively */
2453 if (IS_NODESEG(type)) {
2454 if (type >= CURSEG_WARM_NODE) {
2455 reversed = true;
2456 i = CURSEG_COLD_NODE;
2457 } else {
2458 i = CURSEG_HOT_NODE;
2459 }
2460 cnt = NR_CURSEG_NODE_TYPE;
2461 } else {
2462 if (type >= CURSEG_WARM_DATA) {
2463 reversed = true;
2464 i = CURSEG_COLD_DATA;
2465 } else {
2466 i = CURSEG_HOT_DATA;
2467 }
2468 cnt = NR_CURSEG_DATA_TYPE;
2469 }
2470
2471 for (; cnt-- > 0; reversed ? i-- : i++) {
2472 if (i == type)
2473 continue;
2474 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2475 curseg->next_segno = segno;
2476 return 1;
2477 }
2478 }
2479 return 0;
2480 }
2481
2482 /*
2483 * flush out current segment and replace it with new segment
2484 * This function should be returned with success, otherwise BUG
2485 */
2486 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2487 int type, bool force)
2488 {
2489 struct curseg_info *curseg = CURSEG_I(sbi, type);
2490
2491 if (force)
2492 new_curseg(sbi, type, true);
2493 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2494 type == CURSEG_WARM_NODE)
2495 new_curseg(sbi, type, false);
2496 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2497 new_curseg(sbi, type, false);
2498 else if (f2fs_need_SSR(sbi) && get_ssr_segment(sbi, type))
2499 change_curseg(sbi, type);
2500 else
2501 new_curseg(sbi, type, false);
2502
2503 stat_inc_seg_type(sbi, curseg);
2504 }
2505
2506 void f2fs_allocate_new_segments(struct f2fs_sb_info *sbi)
2507 {
2508 struct curseg_info *curseg;
2509 unsigned int old_segno;
2510 int i;
2511
2512 down_write(&SIT_I(sbi)->sentry_lock);
2513
2514 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2515 curseg = CURSEG_I(sbi, i);
2516 old_segno = curseg->segno;
2517 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2518 locate_dirty_segment(sbi, old_segno);
2519 }
2520
2521 up_write(&SIT_I(sbi)->sentry_lock);
2522 }
2523
2524 static const struct segment_allocation default_salloc_ops = {
2525 .allocate_segment = allocate_segment_by_default,
2526 };
2527
2528 bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
2529 struct cp_control *cpc)
2530 {
2531 __u64 trim_start = cpc->trim_start;
2532 bool has_candidate = false;
2533
2534 down_write(&SIT_I(sbi)->sentry_lock);
2535 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2536 if (add_discard_addrs(sbi, cpc, true)) {
2537 has_candidate = true;
2538 break;
2539 }
2540 }
2541 up_write(&SIT_I(sbi)->sentry_lock);
2542
2543 cpc->trim_start = trim_start;
2544 return has_candidate;
2545 }
2546
2547 static unsigned int __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2548 struct discard_policy *dpolicy,
2549 unsigned int start, unsigned int end)
2550 {
2551 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2552 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2553 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2554 struct discard_cmd *dc;
2555 struct blk_plug plug;
2556 int issued;
2557 unsigned int trimmed = 0;
2558
2559 next:
2560 issued = 0;
2561
2562 mutex_lock(&dcc->cmd_lock);
2563 if (unlikely(dcc->rbtree_check))
2564 f2fs_bug_on(sbi, !f2fs_check_rb_tree_consistence(sbi,
2565 &dcc->root));
2566
2567 dc = (struct discard_cmd *)f2fs_lookup_rb_tree_ret(&dcc->root,
2568 NULL, start,
2569 (struct rb_entry **)&prev_dc,
2570 (struct rb_entry **)&next_dc,
2571 &insert_p, &insert_parent, true);
2572 if (!dc)
2573 dc = next_dc;
2574
2575 blk_start_plug(&plug);
2576
2577 while (dc && dc->lstart <= end) {
2578 struct rb_node *node;
2579 int err = 0;
2580
2581 if (dc->len < dpolicy->granularity)
2582 goto skip;
2583
2584 if (dc->state != D_PREP) {
2585 list_move_tail(&dc->list, &dcc->fstrim_list);
2586 goto skip;
2587 }
2588
2589 err = __submit_discard_cmd(sbi, dpolicy, dc, &issued);
2590
2591 if (issued >= dpolicy->max_requests) {
2592 start = dc->lstart + dc->len;
2593
2594 if (err)
2595 __remove_discard_cmd(sbi, dc);
2596
2597 blk_finish_plug(&plug);
2598 mutex_unlock(&dcc->cmd_lock);
2599 trimmed += __wait_all_discard_cmd(sbi, NULL);
2600 congestion_wait(BLK_RW_ASYNC, HZ/50);
2601 goto next;
2602 }
2603 skip:
2604 node = rb_next(&dc->rb_node);
2605 if (err)
2606 __remove_discard_cmd(sbi, dc);
2607 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2608
2609 if (fatal_signal_pending(current))
2610 break;
2611 }
2612
2613 blk_finish_plug(&plug);
2614 mutex_unlock(&dcc->cmd_lock);
2615
2616 return trimmed;
2617 }
2618
2619 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2620 {
2621 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2622 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2623 unsigned int start_segno, end_segno;
2624 block_t start_block, end_block;
2625 struct cp_control cpc;
2626 struct discard_policy dpolicy;
2627 unsigned long long trimmed = 0;
2628 int err = 0;
2629 bool need_align = test_opt(sbi, LFS) && sbi->segs_per_sec > 1;
2630
2631 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2632 return -EINVAL;
2633
2634 if (end <= MAIN_BLKADDR(sbi))
2635 return -EINVAL;
2636
2637 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2638 f2fs_msg(sbi->sb, KERN_WARNING,
2639 "Found FS corruption, run fsck to fix.");
2640 return -EIO;
2641 }
2642
2643 /* start/end segment number in main_area */
2644 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2645 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2646 GET_SEGNO(sbi, end);
2647 if (need_align) {
2648 start_segno = rounddown(start_segno, sbi->segs_per_sec);
2649 end_segno = roundup(end_segno + 1, sbi->segs_per_sec) - 1;
2650 }
2651
2652 cpc.reason = CP_DISCARD;
2653 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2654 cpc.trim_start = start_segno;
2655 cpc.trim_end = end_segno;
2656
2657 if (sbi->discard_blks == 0)
2658 goto out;
2659
2660 mutex_lock(&sbi->gc_mutex);
2661 err = f2fs_write_checkpoint(sbi, &cpc);
2662 mutex_unlock(&sbi->gc_mutex);
2663 if (err)
2664 goto out;
2665
2666 /*
2667 * We filed discard candidates, but actually we don't need to wait for
2668 * all of them, since they'll be issued in idle time along with runtime
2669 * discard option. User configuration looks like using runtime discard
2670 * or periodic fstrim instead of it.
2671 */
2672 if (test_opt(sbi, DISCARD))
2673 goto out;
2674
2675 start_block = START_BLOCK(sbi, start_segno);
2676 end_block = START_BLOCK(sbi, end_segno + 1);
2677
2678 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2679 trimmed = __issue_discard_cmd_range(sbi, &dpolicy,
2680 start_block, end_block);
2681
2682 trimmed += __wait_discard_cmd_range(sbi, &dpolicy,
2683 start_block, end_block);
2684 out:
2685 if (!err)
2686 range->len = F2FS_BLK_TO_BYTES(trimmed);
2687 return err;
2688 }
2689
2690 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2691 {
2692 struct curseg_info *curseg = CURSEG_I(sbi, type);
2693 if (curseg->next_blkoff < sbi->blocks_per_seg)
2694 return true;
2695 return false;
2696 }
2697
2698 int f2fs_rw_hint_to_seg_type(enum rw_hint hint)
2699 {
2700 switch (hint) {
2701 case WRITE_LIFE_SHORT:
2702 return CURSEG_HOT_DATA;
2703 case WRITE_LIFE_EXTREME:
2704 return CURSEG_COLD_DATA;
2705 default:
2706 return CURSEG_WARM_DATA;
2707 }
2708 }
2709
2710 /* This returns write hints for each segment type. This hints will be
2711 * passed down to block layer. There are mapping tables which depend on
2712 * the mount option 'whint_mode'.
2713 *
2714 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2715 *
2716 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2717 *
2718 * User F2FS Block
2719 * ---- ---- -----
2720 * META WRITE_LIFE_NOT_SET
2721 * HOT_NODE "
2722 * WARM_NODE "
2723 * COLD_NODE "
2724 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2725 * extension list " "
2726 *
2727 * -- buffered io
2728 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2729 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2730 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2731 * WRITE_LIFE_NONE " "
2732 * WRITE_LIFE_MEDIUM " "
2733 * WRITE_LIFE_LONG " "
2734 *
2735 * -- direct io
2736 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2737 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2738 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2739 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2740 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2741 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2742 *
2743 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2744 *
2745 * User F2FS Block
2746 * ---- ---- -----
2747 * META WRITE_LIFE_MEDIUM;
2748 * HOT_NODE WRITE_LIFE_NOT_SET
2749 * WARM_NODE "
2750 * COLD_NODE WRITE_LIFE_NONE
2751 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2752 * extension list " "
2753 *
2754 * -- buffered io
2755 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2756 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2757 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
2758 * WRITE_LIFE_NONE " "
2759 * WRITE_LIFE_MEDIUM " "
2760 * WRITE_LIFE_LONG " "
2761 *
2762 * -- direct io
2763 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2764 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2765 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2766 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2767 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2768 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2769 */
2770
2771 enum rw_hint f2fs_io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2772 enum page_type type, enum temp_type temp)
2773 {
2774 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2775 if (type == DATA) {
2776 if (temp == WARM)
2777 return WRITE_LIFE_NOT_SET;
2778 else if (temp == HOT)
2779 return WRITE_LIFE_SHORT;
2780 else if (temp == COLD)
2781 return WRITE_LIFE_EXTREME;
2782 } else {
2783 return WRITE_LIFE_NOT_SET;
2784 }
2785 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2786 if (type == DATA) {
2787 if (temp == WARM)
2788 return WRITE_LIFE_LONG;
2789 else if (temp == HOT)
2790 return WRITE_LIFE_SHORT;
2791 else if (temp == COLD)
2792 return WRITE_LIFE_EXTREME;
2793 } else if (type == NODE) {
2794 if (temp == WARM || temp == HOT)
2795 return WRITE_LIFE_NOT_SET;
2796 else if (temp == COLD)
2797 return WRITE_LIFE_NONE;
2798 } else if (type == META) {
2799 return WRITE_LIFE_MEDIUM;
2800 }
2801 }
2802 return WRITE_LIFE_NOT_SET;
2803 }
2804
2805 static int __get_segment_type_2(struct f2fs_io_info *fio)
2806 {
2807 if (fio->type == DATA)
2808 return CURSEG_HOT_DATA;
2809 else
2810 return CURSEG_HOT_NODE;
2811 }
2812
2813 static int __get_segment_type_4(struct f2fs_io_info *fio)
2814 {
2815 if (fio->type == DATA) {
2816 struct inode *inode = fio->page->mapping->host;
2817
2818 if (S_ISDIR(inode->i_mode))
2819 return CURSEG_HOT_DATA;
2820 else
2821 return CURSEG_COLD_DATA;
2822 } else {
2823 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2824 return CURSEG_WARM_NODE;
2825 else
2826 return CURSEG_COLD_NODE;
2827 }
2828 }
2829
2830 static int __get_segment_type_6(struct f2fs_io_info *fio)
2831 {
2832 if (fio->type == DATA) {
2833 struct inode *inode = fio->page->mapping->host;
2834
2835 if (is_cold_data(fio->page) || file_is_cold(inode))
2836 return CURSEG_COLD_DATA;
2837 if (file_is_hot(inode) ||
2838 is_inode_flag_set(inode, FI_HOT_DATA) ||
2839 f2fs_is_atomic_file(inode) ||
2840 f2fs_is_volatile_file(inode))
2841 return CURSEG_HOT_DATA;
2842 return f2fs_rw_hint_to_seg_type(inode->i_write_hint);
2843 } else {
2844 if (IS_DNODE(fio->page))
2845 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2846 CURSEG_HOT_NODE;
2847 return CURSEG_COLD_NODE;
2848 }
2849 }
2850
2851 static int __get_segment_type(struct f2fs_io_info *fio)
2852 {
2853 int type = 0;
2854
2855 switch (F2FS_OPTION(fio->sbi).active_logs) {
2856 case 2:
2857 type = __get_segment_type_2(fio);
2858 break;
2859 case 4:
2860 type = __get_segment_type_4(fio);
2861 break;
2862 case 6:
2863 type = __get_segment_type_6(fio);
2864 break;
2865 default:
2866 f2fs_bug_on(fio->sbi, true);
2867 }
2868
2869 if (IS_HOT(type))
2870 fio->temp = HOT;
2871 else if (IS_WARM(type))
2872 fio->temp = WARM;
2873 else
2874 fio->temp = COLD;
2875 return type;
2876 }
2877
2878 void f2fs_allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2879 block_t old_blkaddr, block_t *new_blkaddr,
2880 struct f2fs_summary *sum, int type,
2881 struct f2fs_io_info *fio, bool add_list)
2882 {
2883 struct sit_info *sit_i = SIT_I(sbi);
2884 struct curseg_info *curseg = CURSEG_I(sbi, type);
2885
2886 down_read(&SM_I(sbi)->curseg_lock);
2887
2888 mutex_lock(&curseg->curseg_mutex);
2889 down_write(&sit_i->sentry_lock);
2890
2891 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2892
2893 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2894
2895 /*
2896 * __add_sum_entry should be resided under the curseg_mutex
2897 * because, this function updates a summary entry in the
2898 * current summary block.
2899 */
2900 __add_sum_entry(sbi, type, sum);
2901
2902 __refresh_next_blkoff(sbi, curseg);
2903
2904 stat_inc_block_count(sbi, curseg);
2905
2906 /*
2907 * SIT information should be updated before segment allocation,
2908 * since SSR needs latest valid block information.
2909 */
2910 update_sit_entry(sbi, *new_blkaddr, 1);
2911 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2912 update_sit_entry(sbi, old_blkaddr, -1);
2913
2914 if (!__has_curseg_space(sbi, type))
2915 sit_i->s_ops->allocate_segment(sbi, type, false);
2916
2917 /*
2918 * segment dirty status should be updated after segment allocation,
2919 * so we just need to update status only one time after previous
2920 * segment being closed.
2921 */
2922 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2923 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
2924
2925 up_write(&sit_i->sentry_lock);
2926
2927 if (page && IS_NODESEG(type)) {
2928 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2929
2930 f2fs_inode_chksum_set(sbi, page);
2931 }
2932
2933 if (add_list) {
2934 struct f2fs_bio_info *io;
2935
2936 INIT_LIST_HEAD(&fio->list);
2937 fio->in_list = true;
2938 fio->retry = false;
2939 io = sbi->write_io[fio->type] + fio->temp;
2940 spin_lock(&io->io_lock);
2941 list_add_tail(&fio->list, &io->io_list);
2942 spin_unlock(&io->io_lock);
2943 }
2944
2945 mutex_unlock(&curseg->curseg_mutex);
2946
2947 up_read(&SM_I(sbi)->curseg_lock);
2948 }
2949
2950 static void update_device_state(struct f2fs_io_info *fio)
2951 {
2952 struct f2fs_sb_info *sbi = fio->sbi;
2953 unsigned int devidx;
2954
2955 if (!sbi->s_ndevs)
2956 return;
2957
2958 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2959
2960 /* update device state for fsync */
2961 f2fs_set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
2962
2963 /* update device state for checkpoint */
2964 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2965 spin_lock(&sbi->dev_lock);
2966 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2967 spin_unlock(&sbi->dev_lock);
2968 }
2969 }
2970
2971 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2972 {
2973 int type = __get_segment_type(fio);
2974 bool keep_order = (test_opt(fio->sbi, LFS) && type == CURSEG_COLD_DATA);
2975
2976 if (keep_order)
2977 down_read(&fio->sbi->io_order_lock);
2978 reallocate:
2979 f2fs_allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2980 &fio->new_blkaddr, sum, type, fio, true);
2981
2982 /* writeout dirty page into bdev */
2983 f2fs_submit_page_write(fio);
2984 if (fio->retry) {
2985 fio->old_blkaddr = fio->new_blkaddr;
2986 goto reallocate;
2987 }
2988
2989 update_device_state(fio);
2990
2991 if (keep_order)
2992 up_read(&fio->sbi->io_order_lock);
2993 }
2994
2995 void f2fs_do_write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2996 enum iostat_type io_type)
2997 {
2998 struct f2fs_io_info fio = {
2999 .sbi = sbi,
3000 .type = META,
3001 .temp = HOT,
3002 .op = REQ_OP_WRITE,
3003 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
3004 .old_blkaddr = page->index,
3005 .new_blkaddr = page->index,
3006 .page = page,
3007 .encrypted_page = NULL,
3008 .in_list = false,
3009 };
3010
3011 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
3012 fio.op_flags &= ~REQ_META;
3013
3014 set_page_writeback(page);
3015 ClearPageError(page);
3016 f2fs_submit_page_write(&fio);
3017
3018 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
3019 }
3020
3021 void f2fs_do_write_node_page(unsigned int nid, struct f2fs_io_info *fio)
3022 {
3023 struct f2fs_summary sum;
3024
3025 set_summary(&sum, nid, 0, 0);
3026 do_write_page(&sum, fio);
3027
3028 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3029 }
3030
3031 void f2fs_outplace_write_data(struct dnode_of_data *dn,
3032 struct f2fs_io_info *fio)
3033 {
3034 struct f2fs_sb_info *sbi = fio->sbi;
3035 struct f2fs_summary sum;
3036
3037 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
3038 set_summary(&sum, dn->nid, dn->ofs_in_node, fio->version);
3039 do_write_page(&sum, fio);
3040 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
3041
3042 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
3043 }
3044
3045 int f2fs_inplace_write_data(struct f2fs_io_info *fio)
3046 {
3047 int err;
3048 struct f2fs_sb_info *sbi = fio->sbi;
3049
3050 fio->new_blkaddr = fio->old_blkaddr;
3051 /* i/o temperature is needed for passing down write hints */
3052 __get_segment_type(fio);
3053
3054 f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
3055 GET_SEGNO(sbi, fio->new_blkaddr))->type));
3056
3057 stat_inc_inplace_blocks(fio->sbi);
3058
3059 err = f2fs_submit_page_bio(fio);
3060 if (!err)
3061 update_device_state(fio);
3062
3063 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
3064
3065 return err;
3066 }
3067
3068 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
3069 unsigned int segno)
3070 {
3071 int i;
3072
3073 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
3074 if (CURSEG_I(sbi, i)->segno == segno)
3075 break;
3076 }
3077 return i;
3078 }
3079
3080 void f2fs_do_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
3081 block_t old_blkaddr, block_t new_blkaddr,
3082 bool recover_curseg, bool recover_newaddr)
3083 {
3084 struct sit_info *sit_i = SIT_I(sbi);
3085 struct curseg_info *curseg;
3086 unsigned int segno, old_cursegno;
3087 struct seg_entry *se;
3088 int type;
3089 unsigned short old_blkoff;
3090
3091 segno = GET_SEGNO(sbi, new_blkaddr);
3092 se = get_seg_entry(sbi, segno);
3093 type = se->type;
3094
3095 down_write(&SM_I(sbi)->curseg_lock);
3096
3097 if (!recover_curseg) {
3098 /* for recovery flow */
3099 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
3100 if (old_blkaddr == NULL_ADDR)
3101 type = CURSEG_COLD_DATA;
3102 else
3103 type = CURSEG_WARM_DATA;
3104 }
3105 } else {
3106 if (IS_CURSEG(sbi, segno)) {
3107 /* se->type is volatile as SSR allocation */
3108 type = __f2fs_get_curseg(sbi, segno);
3109 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
3110 } else {
3111 type = CURSEG_WARM_DATA;
3112 }
3113 }
3114
3115 f2fs_bug_on(sbi, !IS_DATASEG(type));
3116 curseg = CURSEG_I(sbi, type);
3117
3118 mutex_lock(&curseg->curseg_mutex);
3119 down_write(&sit_i->sentry_lock);
3120
3121 old_cursegno = curseg->segno;
3122 old_blkoff = curseg->next_blkoff;
3123
3124 /* change the current segment */
3125 if (segno != curseg->segno) {
3126 curseg->next_segno = segno;
3127 change_curseg(sbi, type);
3128 }
3129
3130 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
3131 __add_sum_entry(sbi, type, sum);
3132
3133 if (!recover_curseg || recover_newaddr)
3134 update_sit_entry(sbi, new_blkaddr, 1);
3135 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
3136 update_sit_entry(sbi, old_blkaddr, -1);
3137
3138 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
3139 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
3140
3141 locate_dirty_segment(sbi, old_cursegno);
3142
3143 if (recover_curseg) {
3144 if (old_cursegno != curseg->segno) {
3145 curseg->next_segno = old_cursegno;
3146 change_curseg(sbi, type);
3147 }
3148 curseg->next_blkoff = old_blkoff;
3149 }
3150
3151 up_write(&sit_i->sentry_lock);
3152 mutex_unlock(&curseg->curseg_mutex);
3153 up_write(&SM_I(sbi)->curseg_lock);
3154 }
3155
3156 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
3157 block_t old_addr, block_t new_addr,
3158 unsigned char version, bool recover_curseg,
3159 bool recover_newaddr)
3160 {
3161 struct f2fs_summary sum;
3162
3163 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
3164
3165 f2fs_do_replace_block(sbi, &sum, old_addr, new_addr,
3166 recover_curseg, recover_newaddr);
3167
3168 f2fs_update_data_blkaddr(dn, new_addr);
3169 }
3170
3171 void f2fs_wait_on_page_writeback(struct page *page,
3172 enum page_type type, bool ordered)
3173 {
3174 if (PageWriteback(page)) {
3175 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
3176
3177 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
3178 0, page->index, type);
3179 if (ordered)
3180 wait_on_page_writeback(page);
3181 else
3182 wait_for_stable_page(page);
3183 }
3184 }
3185
3186 void f2fs_wait_on_block_writeback(struct f2fs_sb_info *sbi, block_t blkaddr)
3187 {
3188 struct page *cpage;
3189
3190 if (!is_valid_data_blkaddr(sbi, blkaddr))
3191 return;
3192
3193 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
3194 if (cpage) {
3195 f2fs_wait_on_page_writeback(cpage, DATA, true);
3196 f2fs_put_page(cpage, 1);
3197 }
3198 }
3199
3200 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
3201 {
3202 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3203 struct curseg_info *seg_i;
3204 unsigned char *kaddr;
3205 struct page *page;
3206 block_t start;
3207 int i, j, offset;
3208
3209 start = start_sum_block(sbi);
3210
3211 page = f2fs_get_meta_page(sbi, start++);
3212 if (IS_ERR(page))
3213 return PTR_ERR(page);
3214 kaddr = (unsigned char *)page_address(page);
3215
3216 /* Step 1: restore nat cache */
3217 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3218 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
3219
3220 /* Step 2: restore sit cache */
3221 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3222 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
3223 offset = 2 * SUM_JOURNAL_SIZE;
3224
3225 /* Step 3: restore summary entries */
3226 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3227 unsigned short blk_off;
3228 unsigned int segno;
3229
3230 seg_i = CURSEG_I(sbi, i);
3231 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3232 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3233 seg_i->next_segno = segno;
3234 reset_curseg(sbi, i, 0);
3235 seg_i->alloc_type = ckpt->alloc_type[i];
3236 seg_i->next_blkoff = blk_off;
3237
3238 if (seg_i->alloc_type == SSR)
3239 blk_off = sbi->blocks_per_seg;
3240
3241 for (j = 0; j < blk_off; j++) {
3242 struct f2fs_summary *s;
3243 s = (struct f2fs_summary *)(kaddr + offset);
3244 seg_i->sum_blk->entries[j] = *s;
3245 offset += SUMMARY_SIZE;
3246 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3247 SUM_FOOTER_SIZE)
3248 continue;
3249
3250 f2fs_put_page(page, 1);
3251 page = NULL;
3252
3253 page = f2fs_get_meta_page(sbi, start++);
3254 if (IS_ERR(page))
3255 return PTR_ERR(page);
3256 kaddr = (unsigned char *)page_address(page);
3257 offset = 0;
3258 }
3259 }
3260 f2fs_put_page(page, 1);
3261 return 0;
3262 }
3263
3264 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3265 {
3266 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3267 struct f2fs_summary_block *sum;
3268 struct curseg_info *curseg;
3269 struct page *new;
3270 unsigned short blk_off;
3271 unsigned int segno = 0;
3272 block_t blk_addr = 0;
3273 int err = 0;
3274
3275 /* get segment number and block addr */
3276 if (IS_DATASEG(type)) {
3277 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3278 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3279 CURSEG_HOT_DATA]);
3280 if (__exist_node_summaries(sbi))
3281 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3282 else
3283 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3284 } else {
3285 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3286 CURSEG_HOT_NODE]);
3287 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3288 CURSEG_HOT_NODE]);
3289 if (__exist_node_summaries(sbi))
3290 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3291 type - CURSEG_HOT_NODE);
3292 else
3293 blk_addr = GET_SUM_BLOCK(sbi, segno);
3294 }
3295
3296 new = f2fs_get_meta_page(sbi, blk_addr);
3297 if (IS_ERR(new))
3298 return PTR_ERR(new);
3299 sum = (struct f2fs_summary_block *)page_address(new);
3300
3301 if (IS_NODESEG(type)) {
3302 if (__exist_node_summaries(sbi)) {
3303 struct f2fs_summary *ns = &sum->entries[0];
3304 int i;
3305 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3306 ns->version = 0;
3307 ns->ofs_in_node = 0;
3308 }
3309 } else {
3310 err = f2fs_restore_node_summary(sbi, segno, sum);
3311 if (err)
3312 goto out;
3313 }
3314 }
3315
3316 /* set uncompleted segment to curseg */
3317 curseg = CURSEG_I(sbi, type);
3318 mutex_lock(&curseg->curseg_mutex);
3319
3320 /* update journal info */
3321 down_write(&curseg->journal_rwsem);
3322 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3323 up_write(&curseg->journal_rwsem);
3324
3325 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3326 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3327 curseg->next_segno = segno;
3328 reset_curseg(sbi, type, 0);
3329 curseg->alloc_type = ckpt->alloc_type[type];
3330 curseg->next_blkoff = blk_off;
3331 mutex_unlock(&curseg->curseg_mutex);
3332 out:
3333 f2fs_put_page(new, 1);
3334 return err;
3335 }
3336
3337 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3338 {
3339 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3340 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3341 int type = CURSEG_HOT_DATA;
3342 int err;
3343
3344 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3345 int npages = f2fs_npages_for_summary_flush(sbi, true);
3346
3347 if (npages >= 2)
3348 f2fs_ra_meta_pages(sbi, start_sum_block(sbi), npages,
3349 META_CP, true);
3350
3351 /* restore for compacted data summary */
3352 err = read_compacted_summaries(sbi);
3353 if (err)
3354 return err;
3355 type = CURSEG_HOT_NODE;
3356 }
3357
3358 if (__exist_node_summaries(sbi))
3359 f2fs_ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3360 NR_CURSEG_TYPE - type, META_CP, true);
3361
3362 for (; type <= CURSEG_COLD_NODE; type++) {
3363 err = read_normal_summaries(sbi, type);
3364 if (err)
3365 return err;
3366 }
3367
3368 /* sanity check for summary blocks */
3369 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3370 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3371 return -EINVAL;
3372
3373 return 0;
3374 }
3375
3376 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3377 {
3378 struct page *page;
3379 unsigned char *kaddr;
3380 struct f2fs_summary *summary;
3381 struct curseg_info *seg_i;
3382 int written_size = 0;
3383 int i, j;
3384
3385 page = f2fs_grab_meta_page(sbi, blkaddr++);
3386 kaddr = (unsigned char *)page_address(page);
3387 memset(kaddr, 0, PAGE_SIZE);
3388
3389 /* Step 1: write nat cache */
3390 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3391 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3392 written_size += SUM_JOURNAL_SIZE;
3393
3394 /* Step 2: write sit cache */
3395 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3396 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3397 written_size += SUM_JOURNAL_SIZE;
3398
3399 /* Step 3: write summary entries */
3400 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3401 unsigned short blkoff;
3402 seg_i = CURSEG_I(sbi, i);
3403 if (sbi->ckpt->alloc_type[i] == SSR)
3404 blkoff = sbi->blocks_per_seg;
3405 else
3406 blkoff = curseg_blkoff(sbi, i);
3407
3408 for (j = 0; j < blkoff; j++) {
3409 if (!page) {
3410 page = f2fs_grab_meta_page(sbi, blkaddr++);
3411 kaddr = (unsigned char *)page_address(page);
3412 memset(kaddr, 0, PAGE_SIZE);
3413 written_size = 0;
3414 }
3415 summary = (struct f2fs_summary *)(kaddr + written_size);
3416 *summary = seg_i->sum_blk->entries[j];
3417 written_size += SUMMARY_SIZE;
3418
3419 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3420 SUM_FOOTER_SIZE)
3421 continue;
3422
3423 set_page_dirty(page);
3424 f2fs_put_page(page, 1);
3425 page = NULL;
3426 }
3427 }
3428 if (page) {
3429 set_page_dirty(page);
3430 f2fs_put_page(page, 1);
3431 }
3432 }
3433
3434 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3435 block_t blkaddr, int type)
3436 {
3437 int i, end;
3438 if (IS_DATASEG(type))
3439 end = type + NR_CURSEG_DATA_TYPE;
3440 else
3441 end = type + NR_CURSEG_NODE_TYPE;
3442
3443 for (i = type; i < end; i++)
3444 write_current_sum_page(sbi, i, blkaddr + (i - type));
3445 }
3446
3447 void f2fs_write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3448 {
3449 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3450 write_compacted_summaries(sbi, start_blk);
3451 else
3452 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3453 }
3454
3455 void f2fs_write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3456 {
3457 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3458 }
3459
3460 int f2fs_lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3461 unsigned int val, int alloc)
3462 {
3463 int i;
3464
3465 if (type == NAT_JOURNAL) {
3466 for (i = 0; i < nats_in_cursum(journal); i++) {
3467 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3468 return i;
3469 }
3470 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3471 return update_nats_in_cursum(journal, 1);
3472 } else if (type == SIT_JOURNAL) {
3473 for (i = 0; i < sits_in_cursum(journal); i++)
3474 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3475 return i;
3476 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3477 return update_sits_in_cursum(journal, 1);
3478 }
3479 return -1;
3480 }
3481
3482 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3483 unsigned int segno)
3484 {
3485 return f2fs_get_meta_page_nofail(sbi, current_sit_addr(sbi, segno));
3486 }
3487
3488 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3489 unsigned int start)
3490 {
3491 struct sit_info *sit_i = SIT_I(sbi);
3492 struct page *page;
3493 pgoff_t src_off, dst_off;
3494
3495 src_off = current_sit_addr(sbi, start);
3496 dst_off = next_sit_addr(sbi, src_off);
3497
3498 page = f2fs_grab_meta_page(sbi, dst_off);
3499 seg_info_to_sit_page(sbi, page, start);
3500
3501 set_page_dirty(page);
3502 set_to_next_sit(sit_i, start);
3503
3504 return page;
3505 }
3506
3507 static struct sit_entry_set *grab_sit_entry_set(void)
3508 {
3509 struct sit_entry_set *ses =
3510 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3511
3512 ses->entry_cnt = 0;
3513 INIT_LIST_HEAD(&ses->set_list);
3514 return ses;
3515 }
3516
3517 static void release_sit_entry_set(struct sit_entry_set *ses)
3518 {
3519 list_del(&ses->set_list);
3520 kmem_cache_free(sit_entry_set_slab, ses);
3521 }
3522
3523 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3524 struct list_head *head)
3525 {
3526 struct sit_entry_set *next = ses;
3527
3528 if (list_is_last(&ses->set_list, head))
3529 return;
3530
3531 list_for_each_entry_continue(next, head, set_list)
3532 if (ses->entry_cnt <= next->entry_cnt)
3533 break;
3534
3535 list_move_tail(&ses->set_list, &next->set_list);
3536 }
3537
3538 static void add_sit_entry(unsigned int segno, struct list_head *head)
3539 {
3540 struct sit_entry_set *ses;
3541 unsigned int start_segno = START_SEGNO(segno);
3542
3543 list_for_each_entry(ses, head, set_list) {
3544 if (ses->start_segno == start_segno) {
3545 ses->entry_cnt++;
3546 adjust_sit_entry_set(ses, head);
3547 return;
3548 }
3549 }
3550
3551 ses = grab_sit_entry_set();
3552
3553 ses->start_segno = start_segno;
3554 ses->entry_cnt++;
3555 list_add(&ses->set_list, head);
3556 }
3557
3558 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3559 {
3560 struct f2fs_sm_info *sm_info = SM_I(sbi);
3561 struct list_head *set_list = &sm_info->sit_entry_set;
3562 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3563 unsigned int segno;
3564
3565 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3566 add_sit_entry(segno, set_list);
3567 }
3568
3569 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3570 {
3571 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3572 struct f2fs_journal *journal = curseg->journal;
3573 int i;
3574
3575 down_write(&curseg->journal_rwsem);
3576 for (i = 0; i < sits_in_cursum(journal); i++) {
3577 unsigned int segno;
3578 bool dirtied;
3579
3580 segno = le32_to_cpu(segno_in_journal(journal, i));
3581 dirtied = __mark_sit_entry_dirty(sbi, segno);
3582
3583 if (!dirtied)
3584 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3585 }
3586 update_sits_in_cursum(journal, -i);
3587 up_write(&curseg->journal_rwsem);
3588 }
3589
3590 /*
3591 * CP calls this function, which flushes SIT entries including sit_journal,
3592 * and moves prefree segs to free segs.
3593 */
3594 void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3595 {
3596 struct sit_info *sit_i = SIT_I(sbi);
3597 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3598 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3599 struct f2fs_journal *journal = curseg->journal;
3600 struct sit_entry_set *ses, *tmp;
3601 struct list_head *head = &SM_I(sbi)->sit_entry_set;
3602 bool to_journal = true;
3603 struct seg_entry *se;
3604
3605 down_write(&sit_i->sentry_lock);
3606
3607 if (!sit_i->dirty_sentries)
3608 goto out;
3609
3610 /*
3611 * add and account sit entries of dirty bitmap in sit entry
3612 * set temporarily
3613 */
3614 add_sits_in_set(sbi);
3615
3616 /*
3617 * if there are no enough space in journal to store dirty sit
3618 * entries, remove all entries from journal and add and account
3619 * them in sit entry set.
3620 */
3621 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3622 remove_sits_in_journal(sbi);
3623
3624 /*
3625 * there are two steps to flush sit entries:
3626 * #1, flush sit entries to journal in current cold data summary block.
3627 * #2, flush sit entries to sit page.
3628 */
3629 list_for_each_entry_safe(ses, tmp, head, set_list) {
3630 struct page *page = NULL;
3631 struct f2fs_sit_block *raw_sit = NULL;
3632 unsigned int start_segno = ses->start_segno;
3633 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3634 (unsigned long)MAIN_SEGS(sbi));
3635 unsigned int segno = start_segno;
3636
3637 if (to_journal &&
3638 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3639 to_journal = false;
3640
3641 if (to_journal) {
3642 down_write(&curseg->journal_rwsem);
3643 } else {
3644 page = get_next_sit_page(sbi, start_segno);
3645 raw_sit = page_address(page);
3646 }
3647
3648 /* flush dirty sit entries in region of current sit set */
3649 for_each_set_bit_from(segno, bitmap, end) {
3650 int offset, sit_offset;
3651
3652 se = get_seg_entry(sbi, segno);
3653 #ifdef CONFIG_F2FS_CHECK_FS
3654 if (memcmp(se->cur_valid_map, se->cur_valid_map_mir,
3655 SIT_VBLOCK_MAP_SIZE))
3656 f2fs_bug_on(sbi, 1);
3657 #endif
3658
3659 /* add discard candidates */
3660 if (!(cpc->reason & CP_DISCARD)) {
3661 cpc->trim_start = segno;
3662 add_discard_addrs(sbi, cpc, false);
3663 }
3664
3665 if (to_journal) {
3666 offset = f2fs_lookup_journal_in_cursum(journal,
3667 SIT_JOURNAL, segno, 1);
3668 f2fs_bug_on(sbi, offset < 0);
3669 segno_in_journal(journal, offset) =
3670 cpu_to_le32(segno);
3671 seg_info_to_raw_sit(se,
3672 &sit_in_journal(journal, offset));
3673 check_block_count(sbi, segno,
3674 &sit_in_journal(journal, offset));
3675 } else {
3676 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3677 seg_info_to_raw_sit(se,
3678 &raw_sit->entries[sit_offset]);
3679 check_block_count(sbi, segno,
3680 &raw_sit->entries[sit_offset]);
3681 }
3682
3683 __clear_bit(segno, bitmap);
3684 sit_i->dirty_sentries--;
3685 ses->entry_cnt--;
3686 }
3687
3688 if (to_journal)
3689 up_write(&curseg->journal_rwsem);
3690 else
3691 f2fs_put_page(page, 1);
3692
3693 f2fs_bug_on(sbi, ses->entry_cnt);
3694 release_sit_entry_set(ses);
3695 }
3696
3697 f2fs_bug_on(sbi, !list_empty(head));
3698 f2fs_bug_on(sbi, sit_i->dirty_sentries);
3699 out:
3700 if (cpc->reason & CP_DISCARD) {
3701 __u64 trim_start = cpc->trim_start;
3702
3703 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3704 add_discard_addrs(sbi, cpc, false);
3705
3706 cpc->trim_start = trim_start;
3707 }
3708 up_write(&sit_i->sentry_lock);
3709
3710 set_prefree_as_free_segments(sbi);
3711 }
3712
3713 static int build_sit_info(struct f2fs_sb_info *sbi)
3714 {
3715 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3716 struct sit_info *sit_i;
3717 unsigned int sit_segs, start;
3718 char *src_bitmap;
3719 unsigned int bitmap_size;
3720
3721 /* allocate memory for SIT information */
3722 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3723 if (!sit_i)
3724 return -ENOMEM;
3725
3726 SM_I(sbi)->sit_info = sit_i;
3727
3728 sit_i->sentries =
3729 f2fs_kvzalloc(sbi, array_size(sizeof(struct seg_entry),
3730 MAIN_SEGS(sbi)),
3731 GFP_KERNEL);
3732 if (!sit_i->sentries)
3733 return -ENOMEM;
3734
3735 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3736 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3737 GFP_KERNEL);
3738 if (!sit_i->dirty_sentries_bitmap)
3739 return -ENOMEM;
3740
3741 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3742 sit_i->sentries[start].cur_valid_map
3743 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3744 sit_i->sentries[start].ckpt_valid_map
3745 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3746 if (!sit_i->sentries[start].cur_valid_map ||
3747 !sit_i->sentries[start].ckpt_valid_map)
3748 return -ENOMEM;
3749
3750 #ifdef CONFIG_F2FS_CHECK_FS
3751 sit_i->sentries[start].cur_valid_map_mir
3752 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3753 if (!sit_i->sentries[start].cur_valid_map_mir)
3754 return -ENOMEM;
3755 #endif
3756
3757 if (f2fs_discard_en(sbi)) {
3758 sit_i->sentries[start].discard_map
3759 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3760 GFP_KERNEL);
3761 if (!sit_i->sentries[start].discard_map)
3762 return -ENOMEM;
3763 }
3764 }
3765
3766 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3767 if (!sit_i->tmp_map)
3768 return -ENOMEM;
3769
3770 if (sbi->segs_per_sec > 1) {
3771 sit_i->sec_entries =
3772 f2fs_kvzalloc(sbi, array_size(sizeof(struct sec_entry),
3773 MAIN_SECS(sbi)),
3774 GFP_KERNEL);
3775 if (!sit_i->sec_entries)
3776 return -ENOMEM;
3777 }
3778
3779 /* get information related with SIT */
3780 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3781
3782 /* setup SIT bitmap from ckeckpoint pack */
3783 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3784 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3785
3786 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3787 if (!sit_i->sit_bitmap)
3788 return -ENOMEM;
3789
3790 #ifdef CONFIG_F2FS_CHECK_FS
3791 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3792 if (!sit_i->sit_bitmap_mir)
3793 return -ENOMEM;
3794 #endif
3795
3796 /* init SIT information */
3797 sit_i->s_ops = &default_salloc_ops;
3798
3799 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3800 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3801 sit_i->written_valid_blocks = 0;
3802 sit_i->bitmap_size = bitmap_size;
3803 sit_i->dirty_sentries = 0;
3804 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3805 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3806 sit_i->mounted_time = ktime_get_real_seconds();
3807 init_rwsem(&sit_i->sentry_lock);
3808 return 0;
3809 }
3810
3811 static int build_free_segmap(struct f2fs_sb_info *sbi)
3812 {
3813 struct free_segmap_info *free_i;
3814 unsigned int bitmap_size, sec_bitmap_size;
3815
3816 /* allocate memory for free segmap information */
3817 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
3818 if (!free_i)
3819 return -ENOMEM;
3820
3821 SM_I(sbi)->free_info = free_i;
3822
3823 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3824 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
3825 if (!free_i->free_segmap)
3826 return -ENOMEM;
3827
3828 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3829 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
3830 if (!free_i->free_secmap)
3831 return -ENOMEM;
3832
3833 /* set all segments as dirty temporarily */
3834 memset(free_i->free_segmap, 0xff, bitmap_size);
3835 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3836
3837 /* init free segmap information */
3838 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3839 free_i->free_segments = 0;
3840 free_i->free_sections = 0;
3841 spin_lock_init(&free_i->segmap_lock);
3842 return 0;
3843 }
3844
3845 static int build_curseg(struct f2fs_sb_info *sbi)
3846 {
3847 struct curseg_info *array;
3848 int i;
3849
3850 array = f2fs_kzalloc(sbi, array_size(NR_CURSEG_TYPE, sizeof(*array)),
3851 GFP_KERNEL);
3852 if (!array)
3853 return -ENOMEM;
3854
3855 SM_I(sbi)->curseg_array = array;
3856
3857 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3858 mutex_init(&array[i].curseg_mutex);
3859 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
3860 if (!array[i].sum_blk)
3861 return -ENOMEM;
3862 init_rwsem(&array[i].journal_rwsem);
3863 array[i].journal = f2fs_kzalloc(sbi,
3864 sizeof(struct f2fs_journal), GFP_KERNEL);
3865 if (!array[i].journal)
3866 return -ENOMEM;
3867 array[i].segno = NULL_SEGNO;
3868 array[i].next_blkoff = 0;
3869 }
3870 return restore_curseg_summaries(sbi);
3871 }
3872
3873 static int build_sit_entries(struct f2fs_sb_info *sbi)
3874 {
3875 struct sit_info *sit_i = SIT_I(sbi);
3876 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3877 struct f2fs_journal *journal = curseg->journal;
3878 struct seg_entry *se;
3879 struct f2fs_sit_entry sit;
3880 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3881 unsigned int i, start, end;
3882 unsigned int readed, start_blk = 0;
3883 int err = 0;
3884 block_t total_node_blocks = 0;
3885
3886 do {
3887 readed = f2fs_ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3888 META_SIT, true);
3889
3890 start = start_blk * sit_i->sents_per_block;
3891 end = (start_blk + readed) * sit_i->sents_per_block;
3892
3893 for (; start < end && start < MAIN_SEGS(sbi); start++) {
3894 struct f2fs_sit_block *sit_blk;
3895 struct page *page;
3896
3897 se = &sit_i->sentries[start];
3898 page = get_current_sit_page(sbi, start);
3899 sit_blk = (struct f2fs_sit_block *)page_address(page);
3900 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3901 f2fs_put_page(page, 1);
3902
3903 err = check_block_count(sbi, start, &sit);
3904 if (err)
3905 return err;
3906 seg_info_from_raw_sit(se, &sit);
3907 if (IS_NODESEG(se->type))
3908 total_node_blocks += se->valid_blocks;
3909
3910 /* build discard map only one time */
3911 if (f2fs_discard_en(sbi)) {
3912 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3913 memset(se->discard_map, 0xff,
3914 SIT_VBLOCK_MAP_SIZE);
3915 } else {
3916 memcpy(se->discard_map,
3917 se->cur_valid_map,
3918 SIT_VBLOCK_MAP_SIZE);
3919 sbi->discard_blks +=
3920 sbi->blocks_per_seg -
3921 se->valid_blocks;
3922 }
3923 }
3924
3925 if (sbi->segs_per_sec > 1)
3926 get_sec_entry(sbi, start)->valid_blocks +=
3927 se->valid_blocks;
3928 }
3929 start_blk += readed;
3930 } while (start_blk < sit_blk_cnt);
3931
3932 down_read(&curseg->journal_rwsem);
3933 for (i = 0; i < sits_in_cursum(journal); i++) {
3934 unsigned int old_valid_blocks;
3935
3936 start = le32_to_cpu(segno_in_journal(journal, i));
3937 if (start >= MAIN_SEGS(sbi)) {
3938 f2fs_msg(sbi->sb, KERN_ERR,
3939 "Wrong journal entry on segno %u",
3940 start);
3941 set_sbi_flag(sbi, SBI_NEED_FSCK);
3942 err = -EINVAL;
3943 break;
3944 }
3945
3946 se = &sit_i->sentries[start];
3947 sit = sit_in_journal(journal, i);
3948
3949 old_valid_blocks = se->valid_blocks;
3950 if (IS_NODESEG(se->type))
3951 total_node_blocks -= old_valid_blocks;
3952
3953 err = check_block_count(sbi, start, &sit);
3954 if (err)
3955 break;
3956 seg_info_from_raw_sit(se, &sit);
3957 if (IS_NODESEG(se->type))
3958 total_node_blocks += se->valid_blocks;
3959
3960 if (f2fs_discard_en(sbi)) {
3961 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3962 memset(se->discard_map, 0xff,
3963 SIT_VBLOCK_MAP_SIZE);
3964 } else {
3965 memcpy(se->discard_map, se->cur_valid_map,
3966 SIT_VBLOCK_MAP_SIZE);
3967 sbi->discard_blks += old_valid_blocks;
3968 sbi->discard_blks -= se->valid_blocks;
3969 }
3970 }
3971
3972 if (sbi->segs_per_sec > 1) {
3973 get_sec_entry(sbi, start)->valid_blocks +=
3974 se->valid_blocks;
3975 get_sec_entry(sbi, start)->valid_blocks -=
3976 old_valid_blocks;
3977 }
3978 }
3979 up_read(&curseg->journal_rwsem);
3980
3981 if (!err && total_node_blocks != valid_node_count(sbi)) {
3982 f2fs_msg(sbi->sb, KERN_ERR,
3983 "SIT is corrupted node# %u vs %u",
3984 total_node_blocks, valid_node_count(sbi));
3985 set_sbi_flag(sbi, SBI_NEED_FSCK);
3986 err = -EINVAL;
3987 }
3988
3989 return err;
3990 }
3991
3992 static void init_free_segmap(struct f2fs_sb_info *sbi)
3993 {
3994 unsigned int start;
3995 int type;
3996
3997 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3998 struct seg_entry *sentry = get_seg_entry(sbi, start);
3999 if (!sentry->valid_blocks)
4000 __set_free(sbi, start);
4001 else
4002 SIT_I(sbi)->written_valid_blocks +=
4003 sentry->valid_blocks;
4004 }
4005
4006 /* set use the current segments */
4007 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
4008 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
4009 __set_test_and_inuse(sbi, curseg_t->segno);
4010 }
4011 }
4012
4013 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
4014 {
4015 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4016 struct free_segmap_info *free_i = FREE_I(sbi);
4017 unsigned int segno = 0, offset = 0;
4018 unsigned short valid_blocks;
4019
4020 while (1) {
4021 /* find dirty segment based on free segmap */
4022 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
4023 if (segno >= MAIN_SEGS(sbi))
4024 break;
4025 offset = segno + 1;
4026 valid_blocks = get_valid_blocks(sbi, segno, false);
4027 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
4028 continue;
4029 if (valid_blocks > sbi->blocks_per_seg) {
4030 f2fs_bug_on(sbi, 1);
4031 continue;
4032 }
4033 mutex_lock(&dirty_i->seglist_lock);
4034 __locate_dirty_segment(sbi, segno, DIRTY);
4035 mutex_unlock(&dirty_i->seglist_lock);
4036 }
4037 }
4038
4039 static int init_victim_secmap(struct f2fs_sb_info *sbi)
4040 {
4041 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4042 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
4043
4044 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
4045 if (!dirty_i->victim_secmap)
4046 return -ENOMEM;
4047 return 0;
4048 }
4049
4050 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
4051 {
4052 struct dirty_seglist_info *dirty_i;
4053 unsigned int bitmap_size, i;
4054
4055 /* allocate memory for dirty segments list information */
4056 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
4057 GFP_KERNEL);
4058 if (!dirty_i)
4059 return -ENOMEM;
4060
4061 SM_I(sbi)->dirty_info = dirty_i;
4062 mutex_init(&dirty_i->seglist_lock);
4063
4064 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
4065
4066 for (i = 0; i < NR_DIRTY_TYPE; i++) {
4067 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
4068 GFP_KERNEL);
4069 if (!dirty_i->dirty_segmap[i])
4070 return -ENOMEM;
4071 }
4072
4073 init_dirty_segmap(sbi);
4074 return init_victim_secmap(sbi);
4075 }
4076
4077 /*
4078 * Update min, max modified time for cost-benefit GC algorithm
4079 */
4080 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
4081 {
4082 struct sit_info *sit_i = SIT_I(sbi);
4083 unsigned int segno;
4084
4085 down_write(&sit_i->sentry_lock);
4086
4087 sit_i->min_mtime = ULLONG_MAX;
4088
4089 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
4090 unsigned int i;
4091 unsigned long long mtime = 0;
4092
4093 for (i = 0; i < sbi->segs_per_sec; i++)
4094 mtime += get_seg_entry(sbi, segno + i)->mtime;
4095
4096 mtime = div_u64(mtime, sbi->segs_per_sec);
4097
4098 if (sit_i->min_mtime > mtime)
4099 sit_i->min_mtime = mtime;
4100 }
4101 sit_i->max_mtime = get_mtime(sbi, false);
4102 up_write(&sit_i->sentry_lock);
4103 }
4104
4105 int f2fs_build_segment_manager(struct f2fs_sb_info *sbi)
4106 {
4107 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
4108 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
4109 struct f2fs_sm_info *sm_info;
4110 int err;
4111
4112 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
4113 if (!sm_info)
4114 return -ENOMEM;
4115
4116 /* init sm info */
4117 sbi->sm_info = sm_info;
4118 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
4119 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
4120 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
4121 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
4122 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
4123 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
4124 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
4125 sm_info->rec_prefree_segments = sm_info->main_segments *
4126 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
4127 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
4128 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
4129
4130 if (!test_opt(sbi, LFS))
4131 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
4132 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
4133 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
4134 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
4135 sm_info->min_ssr_sections = reserved_sections(sbi);
4136
4137 INIT_LIST_HEAD(&sm_info->sit_entry_set);
4138
4139 init_rwsem(&sm_info->curseg_lock);
4140
4141 if (!f2fs_readonly(sbi->sb)) {
4142 err = f2fs_create_flush_cmd_control(sbi);
4143 if (err)
4144 return err;
4145 }
4146
4147 err = create_discard_cmd_control(sbi);
4148 if (err)
4149 return err;
4150
4151 err = build_sit_info(sbi);
4152 if (err)
4153 return err;
4154 err = build_free_segmap(sbi);
4155 if (err)
4156 return err;
4157 err = build_curseg(sbi);
4158 if (err)
4159 return err;
4160
4161 /* reinit free segmap based on SIT */
4162 err = build_sit_entries(sbi);
4163 if (err)
4164 return err;
4165
4166 init_free_segmap(sbi);
4167 err = build_dirty_segmap(sbi);
4168 if (err)
4169 return err;
4170
4171 init_min_max_mtime(sbi);
4172 return 0;
4173 }
4174
4175 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
4176 enum dirty_type dirty_type)
4177 {
4178 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4179
4180 mutex_lock(&dirty_i->seglist_lock);
4181 kvfree(dirty_i->dirty_segmap[dirty_type]);
4182 dirty_i->nr_dirty[dirty_type] = 0;
4183 mutex_unlock(&dirty_i->seglist_lock);
4184 }
4185
4186 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
4187 {
4188 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4189 kvfree(dirty_i->victim_secmap);
4190 }
4191
4192 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
4193 {
4194 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
4195 int i;
4196
4197 if (!dirty_i)
4198 return;
4199
4200 /* discard pre-free/dirty segments list */
4201 for (i = 0; i < NR_DIRTY_TYPE; i++)
4202 discard_dirty_segmap(sbi, i);
4203
4204 destroy_victim_secmap(sbi);
4205 SM_I(sbi)->dirty_info = NULL;
4206 kfree(dirty_i);
4207 }
4208
4209 static void destroy_curseg(struct f2fs_sb_info *sbi)
4210 {
4211 struct curseg_info *array = SM_I(sbi)->curseg_array;
4212 int i;
4213
4214 if (!array)
4215 return;
4216 SM_I(sbi)->curseg_array = NULL;
4217 for (i = 0; i < NR_CURSEG_TYPE; i++) {
4218 kfree(array[i].sum_blk);
4219 kfree(array[i].journal);
4220 }
4221 kfree(array);
4222 }
4223
4224 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
4225 {
4226 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
4227 if (!free_i)
4228 return;
4229 SM_I(sbi)->free_info = NULL;
4230 kvfree(free_i->free_segmap);
4231 kvfree(free_i->free_secmap);
4232 kfree(free_i);
4233 }
4234
4235 static void destroy_sit_info(struct f2fs_sb_info *sbi)
4236 {
4237 struct sit_info *sit_i = SIT_I(sbi);
4238 unsigned int start;
4239
4240 if (!sit_i)
4241 return;
4242
4243 if (sit_i->sentries) {
4244 for (start = 0; start < MAIN_SEGS(sbi); start++) {
4245 kfree(sit_i->sentries[start].cur_valid_map);
4246 #ifdef CONFIG_F2FS_CHECK_FS
4247 kfree(sit_i->sentries[start].cur_valid_map_mir);
4248 #endif
4249 kfree(sit_i->sentries[start].ckpt_valid_map);
4250 kfree(sit_i->sentries[start].discard_map);
4251 }
4252 }
4253 kfree(sit_i->tmp_map);
4254
4255 kvfree(sit_i->sentries);
4256 kvfree(sit_i->sec_entries);
4257 kvfree(sit_i->dirty_sentries_bitmap);
4258
4259 SM_I(sbi)->sit_info = NULL;
4260 kfree(sit_i->sit_bitmap);
4261 #ifdef CONFIG_F2FS_CHECK_FS
4262 kfree(sit_i->sit_bitmap_mir);
4263 #endif
4264 kfree(sit_i);
4265 }
4266
4267 void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi)
4268 {
4269 struct f2fs_sm_info *sm_info = SM_I(sbi);
4270
4271 if (!sm_info)
4272 return;
4273 f2fs_destroy_flush_cmd_control(sbi, true);
4274 destroy_discard_cmd_control(sbi);
4275 destroy_dirty_segmap(sbi);
4276 destroy_curseg(sbi);
4277 destroy_free_segmap(sbi);
4278 destroy_sit_info(sbi);
4279 sbi->sm_info = NULL;
4280 kfree(sm_info);
4281 }
4282
4283 int __init f2fs_create_segment_manager_caches(void)
4284 {
4285 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4286 sizeof(struct discard_entry));
4287 if (!discard_entry_slab)
4288 goto fail;
4289
4290 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4291 sizeof(struct discard_cmd));
4292 if (!discard_cmd_slab)
4293 goto destroy_discard_entry;
4294
4295 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4296 sizeof(struct sit_entry_set));
4297 if (!sit_entry_set_slab)
4298 goto destroy_discard_cmd;
4299
4300 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4301 sizeof(struct inmem_pages));
4302 if (!inmem_entry_slab)
4303 goto destroy_sit_entry_set;
4304 return 0;
4305
4306 destroy_sit_entry_set:
4307 kmem_cache_destroy(sit_entry_set_slab);
4308 destroy_discard_cmd:
4309 kmem_cache_destroy(discard_cmd_slab);
4310 destroy_discard_entry:
4311 kmem_cache_destroy(discard_entry_slab);
4312 fail:
4313 return -ENOMEM;
4314 }
4315
4316 void f2fs_destroy_segment_manager_caches(void)
4317 {
4318 kmem_cache_destroy(sit_entry_set_slab);
4319 kmem_cache_destroy(discard_cmd_slab);
4320 kmem_cache_destroy(discard_entry_slab);
4321 kmem_cache_destroy(inmem_entry_slab);
4322 }
Ubuntu Eoan Linux kernel mirror