projects / GitHub / MotorolaMobilityLLC / kernel-slsi.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
[RAMEN9610-21536]f2fs: avoid wrong decrypted data from disk
[GitHub/MotorolaMobilityLLC/kernel-slsi.git] / fs / f2fs / segment.c
1 /*
2 * fs/f2fs/segment.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/bio.h>
14 #include <linux/blkdev.h>
15 #include <linux/prefetch.h>
16 #include <linux/kthread.h>
17 #include <linux/swap.h>
18 #include <linux/timer.h>
19 #include <linux/freezer.h>
20 #include <linux/sched/signal.h>
21
22 #include "f2fs.h"
23 #include "segment.h"
24 #include "node.h"
25 #include "gc.h"
26 #include "trace.h"
27 #include <trace/events/f2fs.h>
28
29 #define __reverse_ffz(x) __reverse_ffs(~(x))
30
31 static struct kmem_cache *discard_entry_slab;
32 static struct kmem_cache *discard_cmd_slab;
33 static struct kmem_cache *sit_entry_set_slab;
34 static struct kmem_cache *inmem_entry_slab;
35
36 static unsigned long __reverse_ulong(unsigned char *str)
37 {
38 unsigned long tmp = 0;
39 int shift = 24, idx = 0;
40
41 #if BITS_PER_LONG == 64
42 shift = 56;
43 #endif
44 while (shift >= 0) {
45 tmp |= (unsigned long)str[idx++] << shift;
46 shift -= BITS_PER_BYTE;
47 }
48 return tmp;
49 }
50
51 /*
52 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
53 * MSB and LSB are reversed in a byte by f2fs_set_bit.
54 */
55 static inline unsigned long __reverse_ffs(unsigned long word)
56 {
57 int num = 0;
58
59 #if BITS_PER_LONG == 64
60 if ((word & 0xffffffff00000000UL) == 0)
61 num += 32;
62 else
63 word >>= 32;
64 #endif
65 if ((word & 0xffff0000) == 0)
66 num += 16;
67 else
68 word >>= 16;
69
70 if ((word & 0xff00) == 0)
71 num += 8;
72 else
73 word >>= 8;
74
75 if ((word & 0xf0) == 0)
76 num += 4;
77 else
78 word >>= 4;
79
80 if ((word & 0xc) == 0)
81 num += 2;
82 else
83 word >>= 2;
84
85 if ((word & 0x2) == 0)
86 num += 1;
87 return num;
88 }
89
90 /*
91 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
92 * f2fs_set_bit makes MSB and LSB reversed in a byte.
93 * @size must be integral times of unsigned long.
94 * Example:
95 * MSB <--> LSB
96 * f2fs_set_bit(0, bitmap) => 1000 0000
97 * f2fs_set_bit(7, bitmap) => 0000 0001
98 */
99 static unsigned long __find_rev_next_bit(const unsigned long *addr,
100 unsigned long size, unsigned long offset)
101 {
102 const unsigned long *p = addr + BIT_WORD(offset);
103 unsigned long result = size;
104 unsigned long tmp;
105
106 if (offset >= size)
107 return size;
108
109 size -= (offset & ~(BITS_PER_LONG - 1));
110 offset %= BITS_PER_LONG;
111
112 while (1) {
113 if (*p == 0)
114 goto pass;
115
116 tmp = __reverse_ulong((unsigned char *)p);
117
118 tmp &= ~0UL >> offset;
119 if (size < BITS_PER_LONG)
120 tmp &= (~0UL << (BITS_PER_LONG - size));
121 if (tmp)
122 goto found;
123 pass:
124 if (size <= BITS_PER_LONG)
125 break;
126 size -= BITS_PER_LONG;
127 offset = 0;
128 p++;
129 }
130 return result;
131 found:
132 return result - size + __reverse_ffs(tmp);
133 }
134
135 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
136 unsigned long size, unsigned long offset)
137 {
138 const unsigned long *p = addr + BIT_WORD(offset);
139 unsigned long result = size;
140 unsigned long tmp;
141
142 if (offset >= size)
143 return size;
144
145 size -= (offset & ~(BITS_PER_LONG - 1));
146 offset %= BITS_PER_LONG;
147
148 while (1) {
149 if (*p == ~0UL)
150 goto pass;
151
152 tmp = __reverse_ulong((unsigned char *)p);
153
154 if (offset)
155 tmp |= ~0UL << (BITS_PER_LONG - offset);
156 if (size < BITS_PER_LONG)
157 tmp |= ~0UL >> size;
158 if (tmp != ~0UL)
159 goto found;
160 pass:
161 if (size <= BITS_PER_LONG)
162 break;
163 size -= BITS_PER_LONG;
164 offset = 0;
165 p++;
166 }
167 return result;
168 found:
169 return result - size + __reverse_ffz(tmp);
170 }
171
172 bool need_SSR(struct f2fs_sb_info *sbi)
173 {
174 int node_secs = get_blocktype_secs(sbi, F2FS_DIRTY_NODES);
175 int dent_secs = get_blocktype_secs(sbi, F2FS_DIRTY_DENTS);
176 int imeta_secs = get_blocktype_secs(sbi, F2FS_DIRTY_IMETA);
177
178 if (test_opt(sbi, LFS))
179 return false;
180 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
181 return true;
182
183 return free_sections(sbi) <= (node_secs + 2 * dent_secs + imeta_secs +
184 SM_I(sbi)->min_ssr_sections + reserved_sections(sbi));
185 }
186
187 void register_inmem_page(struct inode *inode, struct page *page)
188 {
189 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
190 struct f2fs_inode_info *fi = F2FS_I(inode);
191 struct inmem_pages *new;
192
193 f2fs_trace_pid(page);
194
195 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
196 SetPagePrivate(page);
197
198 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
199
200 /* add atomic page indices to the list */
201 new->page = page;
202 INIT_LIST_HEAD(&new->list);
203
204 /* increase reference count with clean state */
205 mutex_lock(&fi->inmem_lock);
206 get_page(page);
207 list_add_tail(&new->list, &fi->inmem_pages);
208 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
209 if (list_empty(&fi->inmem_ilist))
210 list_add_tail(&fi->inmem_ilist, &sbi->inode_list[ATOMIC_FILE]);
211 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
212 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
213 mutex_unlock(&fi->inmem_lock);
214
215 trace_f2fs_register_inmem_page(page, INMEM);
216 }
217
218 static int __revoke_inmem_pages(struct inode *inode,
219 struct list_head *head, bool drop, bool recover)
220 {
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 = get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
243 if (err) {
244 if (err == -ENOMEM) {
245 congestion_wait(BLK_RW_ASYNC, HZ/50);
246 cond_resched();
247 goto retry;
248 }
249 err = -EAGAIN;
250 goto next;
251 }
252 get_node_info(sbi, dn.nid, &ni);
253 if (cur->old_addr == NEW_ADDR) {
254 invalidate_blocks(sbi, dn.data_blkaddr);
255 f2fs_update_data_blkaddr(&dn, NEW_ADDR);
256 } else
257 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
258 cur->old_addr, ni.version, true, true);
259 f2fs_put_dnode(&dn);
260 }
261 next:
262 /* we don't need to invalidate this in the sccessful status */
263 if (drop || recover)
264 ClearPageUptodate(page);
265 set_page_private(page, 0);
266 ClearPagePrivate(page);
267 f2fs_put_page(page, 1);
268
269 list_del(&cur->list);
270 kmem_cache_free(inmem_entry_slab, cur);
271 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
272 }
273 return err;
274 }
275
276 void drop_inmem_pages_all(struct f2fs_sb_info *sbi)
277 {
278 struct list_head *head = &sbi->inode_list[ATOMIC_FILE];
279 struct inode *inode;
280 struct f2fs_inode_info *fi;
281 next:
282 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
283 if (list_empty(head)) {
284 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
285 return;
286 }
287 fi = list_first_entry(head, struct f2fs_inode_info, inmem_ilist);
288 inode = igrab(&fi->vfs_inode);
289 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
290
291 if (inode) {
292 drop_inmem_pages(inode);
293 iput(inode);
294 }
295 congestion_wait(BLK_RW_ASYNC, HZ/50);
296 cond_resched();
297 goto next;
298 }
299
300 void drop_inmem_pages(struct inode *inode)
301 {
302 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
303 struct f2fs_inode_info *fi = F2FS_I(inode);
304
305 mutex_lock(&fi->inmem_lock);
306 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
307 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
308 if (!list_empty(&fi->inmem_ilist))
309 list_del_init(&fi->inmem_ilist);
310 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
311 mutex_unlock(&fi->inmem_lock);
312
313 clear_inode_flag(inode, FI_ATOMIC_FILE);
314 clear_inode_flag(inode, FI_HOT_DATA);
315 stat_dec_atomic_write(inode);
316 }
317
318 void drop_inmem_page(struct inode *inode, struct page *page)
319 {
320 struct f2fs_inode_info *fi = F2FS_I(inode);
321 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
322 struct list_head *head = &fi->inmem_pages;
323 struct inmem_pages *cur = NULL;
324
325 f2fs_bug_on(sbi, !IS_ATOMIC_WRITTEN_PAGE(page));
326
327 mutex_lock(&fi->inmem_lock);
328 list_for_each_entry(cur, head, list) {
329 if (cur->page == page)
330 break;
331 }
332
333 f2fs_bug_on(sbi, !cur || cur->page != page);
334 list_del(&cur->list);
335 mutex_unlock(&fi->inmem_lock);
336
337 dec_page_count(sbi, F2FS_INMEM_PAGES);
338 kmem_cache_free(inmem_entry_slab, cur);
339
340 ClearPageUptodate(page);
341 set_page_private(page, 0);
342 ClearPagePrivate(page);
343 f2fs_put_page(page, 0);
344
345 trace_f2fs_commit_inmem_page(page, INMEM_INVALIDATE);
346 }
347
348 static int __commit_inmem_pages(struct inode *inode,
349 struct list_head *revoke_list)
350 {
351 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
352 struct f2fs_inode_info *fi = F2FS_I(inode);
353 struct inmem_pages *cur, *tmp;
354 struct f2fs_io_info fio = {
355 .sbi = sbi,
356 .ino = inode->i_ino,
357 .type = DATA,
358 .op = REQ_OP_WRITE,
359 .op_flags = REQ_SYNC | REQ_PRIO,
360 .io_type = FS_DATA_IO,
361 };
362 pgoff_t last_idx = ULONG_MAX;
363 int err = 0;
364
365 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
366 struct page *page = cur->page;
367
368 lock_page(page);
369 if (page->mapping == inode->i_mapping) {
370 trace_f2fs_commit_inmem_page(page, INMEM);
371
372 set_page_dirty(page);
373 f2fs_wait_on_page_writeback(page, DATA, true);
374 if (clear_page_dirty_for_io(page)) {
375 inode_dec_dirty_pages(inode);
376 remove_dirty_inode(inode);
377 }
378 retry:
379 fio.page = page;
380 fio.old_blkaddr = NULL_ADDR;
381 fio.encrypted_page = NULL;
382 fio.need_lock = LOCK_DONE;
383 err = do_write_data_page(&fio);
384 if (err) {
385 if (err == -ENOMEM) {
386 congestion_wait(BLK_RW_ASYNC, HZ/50);
387 cond_resched();
388 goto retry;
389 }
390 unlock_page(page);
391 break;
392 }
393 /* record old blkaddr for revoking */
394 cur->old_addr = fio.old_blkaddr;
395 last_idx = page->index;
396 }
397 unlock_page(page);
398 list_move_tail(&cur->list, revoke_list);
399 }
400
401 if (last_idx != ULONG_MAX)
402 f2fs_submit_merged_write_cond(sbi, inode, 0, last_idx, DATA);
403
404 if (!err)
405 __revoke_inmem_pages(inode, revoke_list, false, false);
406
407 return err;
408 }
409
410 int commit_inmem_pages(struct inode *inode)
411 {
412 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
413 struct f2fs_inode_info *fi = F2FS_I(inode);
414 struct list_head revoke_list;
415 int err;
416
417 INIT_LIST_HEAD(&revoke_list);
418 f2fs_balance_fs(sbi, true);
419 f2fs_lock_op(sbi);
420
421 set_inode_flag(inode, FI_ATOMIC_COMMIT);
422
423 mutex_lock(&fi->inmem_lock);
424 err = __commit_inmem_pages(inode, &revoke_list);
425 if (err) {
426 int ret;
427 /*
428 * try to revoke all committed pages, but still we could fail
429 * due to no memory or other reason, if that happened, EAGAIN
430 * will be returned, which means in such case, transaction is
431 * already not integrity, caller should use journal to do the
432 * recovery or rewrite & commit last transaction. For other
433 * error number, revoking was done by filesystem itself.
434 */
435 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
436 if (ret)
437 err = ret;
438
439 /* drop all uncommitted pages */
440 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
441 }
442 spin_lock(&sbi->inode_lock[ATOMIC_FILE]);
443 if (!list_empty(&fi->inmem_ilist))
444 list_del_init(&fi->inmem_ilist);
445 spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
446 mutex_unlock(&fi->inmem_lock);
447
448 clear_inode_flag(inode, FI_ATOMIC_COMMIT);
449
450 f2fs_unlock_op(sbi);
451 return err;
452 }
453
454 /*
455 * This function balances dirty node and dentry pages.
456 * In addition, it controls garbage collection.
457 */
458 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
459 {
460 #ifdef CONFIG_F2FS_FAULT_INJECTION
461 if (time_to_inject(sbi, FAULT_CHECKPOINT)) {
462 f2fs_show_injection_info(FAULT_CHECKPOINT);
463 f2fs_stop_checkpoint(sbi, false);
464 }
465 #endif
466
467 /* balance_fs_bg is able to be pending */
468 if (need && excess_cached_nats(sbi))
469 f2fs_balance_fs_bg(sbi);
470
471 /*
472 * We should do GC or end up with checkpoint, if there are so many dirty
473 * dir/node pages without enough free segments.
474 */
475 if (has_not_enough_free_secs(sbi, 0, 0)) {
476 mutex_lock(&sbi->gc_mutex);
477 f2fs_gc(sbi, false, false, NULL_SEGNO);
478 }
479 }
480
481 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
482 {
483 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
484 return;
485
486 /* try to shrink extent cache when there is no enough memory */
487 if (!available_free_memory(sbi, EXTENT_CACHE))
488 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
489
490 /* check the # of cached NAT entries */
491 if (!available_free_memory(sbi, NAT_ENTRIES))
492 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
493
494 if (!available_free_memory(sbi, FREE_NIDS))
495 try_to_free_nids(sbi, MAX_FREE_NIDS);
496 else
497 build_free_nids(sbi, false, false);
498
499 if (!is_idle(sbi) && !excess_dirty_nats(sbi))
500 return;
501
502 /* checkpoint is the only way to shrink partial cached entries */
503 if (!available_free_memory(sbi, NAT_ENTRIES) ||
504 !available_free_memory(sbi, INO_ENTRIES) ||
505 excess_prefree_segs(sbi) ||
506 excess_dirty_nats(sbi) ||
507 f2fs_time_over(sbi, CP_TIME)) {
508 if (test_opt(sbi, DATA_FLUSH)) {
509 struct blk_plug plug;
510
511 blk_start_plug(&plug);
512 sync_dirty_inodes(sbi, FILE_INODE);
513 blk_finish_plug(&plug);
514 }
515 f2fs_sync_fs(sbi->sb, true);
516 stat_inc_bg_cp_count(sbi->stat_info);
517 }
518 }
519
520 static int __submit_flush_wait(struct f2fs_sb_info *sbi,
521 struct block_device *bdev)
522 {
523 struct bio *bio = f2fs_bio_alloc(sbi, 0, true);
524 int ret;
525
526 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH;
527 bio_set_dev(bio, bdev);
528 ret = submit_bio_wait(bio);
529 bio_put(bio);
530
531 trace_f2fs_issue_flush(bdev, test_opt(sbi, NOBARRIER),
532 test_opt(sbi, FLUSH_MERGE), ret);
533 return ret;
534 }
535
536 static int submit_flush_wait(struct f2fs_sb_info *sbi, nid_t ino)
537 {
538 int ret = 0;
539 int i;
540
541 if (!sbi->s_ndevs)
542 return __submit_flush_wait(sbi, sbi->sb->s_bdev);
543
544 for (i = 0; i < sbi->s_ndevs; i++) {
545 if (!is_dirty_device(sbi, ino, i, FLUSH_INO))
546 continue;
547 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
548 if (ret)
549 break;
550 }
551 return ret;
552 }
553
554 static int issue_flush_thread(void *data)
555 {
556 struct f2fs_sb_info *sbi = data;
557 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
558 wait_queue_head_t *q = &fcc->flush_wait_queue;
559 repeat:
560 if (kthread_should_stop())
561 return 0;
562
563 sb_start_intwrite(sbi->sb);
564
565 if (!llist_empty(&fcc->issue_list)) {
566 struct flush_cmd *cmd, *next;
567 int ret;
568
569 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
570 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
571
572 cmd = llist_entry(fcc->dispatch_list, struct flush_cmd, llnode);
573
574 ret = submit_flush_wait(sbi, cmd->ino);
575 atomic_inc(&fcc->issued_flush);
576
577 llist_for_each_entry_safe(cmd, next,
578 fcc->dispatch_list, llnode) {
579 cmd->ret = ret;
580 complete(&cmd->wait);
581 }
582 fcc->dispatch_list = NULL;
583 }
584
585 sb_end_intwrite(sbi->sb);
586
587 wait_event_interruptible(*q,
588 kthread_should_stop() || !llist_empty(&fcc->issue_list));
589 goto repeat;
590 }
591
592 int f2fs_issue_flush(struct f2fs_sb_info *sbi, nid_t ino)
593 {
594 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
595 struct flush_cmd cmd;
596 int ret;
597
598 if (test_opt(sbi, NOBARRIER))
599 return 0;
600
601 if (!test_opt(sbi, FLUSH_MERGE)) {
602 ret = submit_flush_wait(sbi, ino);
603 atomic_inc(&fcc->issued_flush);
604 return ret;
605 }
606
607 if (atomic_inc_return(&fcc->issing_flush) == 1 || sbi->s_ndevs > 1) {
608 ret = submit_flush_wait(sbi, ino);
609 atomic_dec(&fcc->issing_flush);
610
611 atomic_inc(&fcc->issued_flush);
612 return ret;
613 }
614
615 cmd.ino = ino;
616 init_completion(&cmd.wait);
617
618 llist_add(&cmd.llnode, &fcc->issue_list);
619
620 /* update issue_list before we wake up issue_flush thread */
621 smp_mb();
622
623 if (waitqueue_active(&fcc->flush_wait_queue))
624 wake_up(&fcc->flush_wait_queue);
625
626 if (fcc->f2fs_issue_flush) {
627 wait_for_completion(&cmd.wait);
628 atomic_dec(&fcc->issing_flush);
629 } else {
630 struct llist_node *list;
631
632 list = llist_del_all(&fcc->issue_list);
633 if (!list) {
634 wait_for_completion(&cmd.wait);
635 atomic_dec(&fcc->issing_flush);
636 } else {
637 struct flush_cmd *tmp, *next;
638
639 ret = submit_flush_wait(sbi, ino);
640
641 llist_for_each_entry_safe(tmp, next, list, llnode) {
642 if (tmp == &cmd) {
643 cmd.ret = ret;
644 atomic_dec(&fcc->issing_flush);
645 continue;
646 }
647 tmp->ret = ret;
648 complete(&tmp->wait);
649 }
650 }
651 }
652
653 return cmd.ret;
654 }
655
656 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
657 {
658 dev_t dev = sbi->sb->s_bdev->bd_dev;
659 struct flush_cmd_control *fcc;
660 int err = 0;
661
662 if (SM_I(sbi)->fcc_info) {
663 fcc = SM_I(sbi)->fcc_info;
664 if (fcc->f2fs_issue_flush)
665 return err;
666 goto init_thread;
667 }
668
669 fcc = f2fs_kzalloc(sbi, sizeof(struct flush_cmd_control), GFP_KERNEL);
670 if (!fcc)
671 return -ENOMEM;
672 atomic_set(&fcc->issued_flush, 0);
673 atomic_set(&fcc->issing_flush, 0);
674 init_waitqueue_head(&fcc->flush_wait_queue);
675 init_llist_head(&fcc->issue_list);
676 SM_I(sbi)->fcc_info = fcc;
677 if (!test_opt(sbi, FLUSH_MERGE))
678 return err;
679
680 init_thread:
681 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
682 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
683 if (IS_ERR(fcc->f2fs_issue_flush)) {
684 err = PTR_ERR(fcc->f2fs_issue_flush);
685 kfree(fcc);
686 SM_I(sbi)->fcc_info = NULL;
687 return err;
688 }
689
690 return err;
691 }
692
693 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi, bool free)
694 {
695 struct flush_cmd_control *fcc = SM_I(sbi)->fcc_info;
696
697 if (fcc && fcc->f2fs_issue_flush) {
698 struct task_struct *flush_thread = fcc->f2fs_issue_flush;
699
700 fcc->f2fs_issue_flush = NULL;
701 kthread_stop(flush_thread);
702 }
703 if (free) {
704 kfree(fcc);
705 SM_I(sbi)->fcc_info = NULL;
706 }
707 }
708
709 int f2fs_flush_device_cache(struct f2fs_sb_info *sbi)
710 {
711 int ret = 0, i;
712
713 if (!sbi->s_ndevs)
714 return 0;
715
716 for (i = 1; i < sbi->s_ndevs; i++) {
717 if (!f2fs_test_bit(i, (char *)&sbi->dirty_device))
718 continue;
719 ret = __submit_flush_wait(sbi, FDEV(i).bdev);
720 if (ret)
721 break;
722
723 spin_lock(&sbi->dev_lock);
724 f2fs_clear_bit(i, (char *)&sbi->dirty_device);
725 spin_unlock(&sbi->dev_lock);
726 }
727
728 return ret;
729 }
730
731 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
732 enum dirty_type dirty_type)
733 {
734 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
735
736 /* need not be added */
737 if (IS_CURSEG(sbi, segno))
738 return;
739
740 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
741 dirty_i->nr_dirty[dirty_type]++;
742
743 if (dirty_type == DIRTY) {
744 struct seg_entry *sentry = get_seg_entry(sbi, segno);
745 enum dirty_type t = sentry->type;
746
747 if (unlikely(t >= DIRTY)) {
748 f2fs_bug_on(sbi, 1);
749 return;
750 }
751 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
752 dirty_i->nr_dirty[t]++;
753 }
754 }
755
756 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
757 enum dirty_type dirty_type)
758 {
759 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
760
761 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
762 dirty_i->nr_dirty[dirty_type]--;
763
764 if (dirty_type == DIRTY) {
765 struct seg_entry *sentry = get_seg_entry(sbi, segno);
766 enum dirty_type t = sentry->type;
767
768 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
769 dirty_i->nr_dirty[t]--;
770
771 if (get_valid_blocks(sbi, segno, true) == 0)
772 clear_bit(GET_SEC_FROM_SEG(sbi, segno),
773 dirty_i->victim_secmap);
774 }
775 }
776
777 /*
778 * Should not occur error such as -ENOMEM.
779 * Adding dirty entry into seglist is not critical operation.
780 * If a given segment is one of current working segments, it won't be added.
781 */
782 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
783 {
784 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
785 unsigned short valid_blocks;
786
787 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
788 return;
789
790 mutex_lock(&dirty_i->seglist_lock);
791
792 valid_blocks = get_valid_blocks(sbi, segno, false);
793
794 if (valid_blocks == 0) {
795 __locate_dirty_segment(sbi, segno, PRE);
796 __remove_dirty_segment(sbi, segno, DIRTY);
797 } else if (valid_blocks < sbi->blocks_per_seg) {
798 __locate_dirty_segment(sbi, segno, DIRTY);
799 } else {
800 /* Recovery routine with SSR needs this */
801 __remove_dirty_segment(sbi, segno, DIRTY);
802 }
803
804 mutex_unlock(&dirty_i->seglist_lock);
805 }
806
807 static struct discard_cmd *__create_discard_cmd(struct f2fs_sb_info *sbi,
808 struct block_device *bdev, block_t lstart,
809 block_t start, block_t len)
810 {
811 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
812 struct list_head *pend_list;
813 struct discard_cmd *dc;
814
815 f2fs_bug_on(sbi, !len);
816
817 pend_list = &dcc->pend_list[plist_idx(len)];
818
819 dc = f2fs_kmem_cache_alloc(discard_cmd_slab, GFP_NOFS);
820 INIT_LIST_HEAD(&dc->list);
821 dc->bdev = bdev;
822 dc->lstart = lstart;
823 dc->start = start;
824 dc->len = len;
825 dc->ref = 0;
826 dc->state = D_PREP;
827 dc->error = 0;
828 init_completion(&dc->wait);
829 list_add_tail(&dc->list, pend_list);
830 atomic_inc(&dcc->discard_cmd_cnt);
831 dcc->undiscard_blks += len;
832
833 return dc;
834 }
835
836 static struct discard_cmd *__attach_discard_cmd(struct f2fs_sb_info *sbi,
837 struct block_device *bdev, block_t lstart,
838 block_t start, block_t len,
839 struct rb_node *parent, struct rb_node **p)
840 {
841 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
842 struct discard_cmd *dc;
843
844 dc = __create_discard_cmd(sbi, bdev, lstart, start, len);
845
846 rb_link_node(&dc->rb_node, parent, p);
847 rb_insert_color(&dc->rb_node, &dcc->root);
848
849 return dc;
850 }
851
852 static void __detach_discard_cmd(struct discard_cmd_control *dcc,
853 struct discard_cmd *dc)
854 {
855 if (dc->state == D_DONE)
856 atomic_dec(&dcc->issing_discard);
857
858 list_del(&dc->list);
859 rb_erase(&dc->rb_node, &dcc->root);
860 dcc->undiscard_blks -= dc->len;
861
862 kmem_cache_free(discard_cmd_slab, dc);
863
864 atomic_dec(&dcc->discard_cmd_cnt);
865 }
866
867 static void __remove_discard_cmd(struct f2fs_sb_info *sbi,
868 struct discard_cmd *dc)
869 {
870 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
871
872 trace_f2fs_remove_discard(dc->bdev, dc->start, dc->len);
873
874 f2fs_bug_on(sbi, dc->ref);
875
876 if (dc->error == -EOPNOTSUPP)
877 dc->error = 0;
878
879 if (dc->error)
880 f2fs_msg(sbi->sb, KERN_INFO,
881 "Issue discard(%u, %u, %u) failed, ret: %d",
882 dc->lstart, dc->start, dc->len, dc->error);
883 __detach_discard_cmd(dcc, dc);
884 }
885
886 static void f2fs_submit_discard_endio(struct bio *bio)
887 {
888 struct discard_cmd *dc = (struct discard_cmd *)bio->bi_private;
889
890 dc->error = blk_status_to_errno(bio->bi_status);
891 dc->state = D_DONE;
892 complete_all(&dc->wait);
893 bio_put(bio);
894 }
895
896 static void __check_sit_bitmap(struct f2fs_sb_info *sbi,
897 block_t start, block_t end)
898 {
899 #ifdef CONFIG_F2FS_CHECK_FS
900 struct seg_entry *sentry;
901 unsigned int segno;
902 block_t blk = start;
903 unsigned long offset, size, max_blocks = sbi->blocks_per_seg;
904 unsigned long *map;
905
906 while (blk < end) {
907 segno = GET_SEGNO(sbi, blk);
908 sentry = get_seg_entry(sbi, segno);
909 offset = GET_BLKOFF_FROM_SEG0(sbi, blk);
910
911 if (end < START_BLOCK(sbi, segno + 1))
912 size = GET_BLKOFF_FROM_SEG0(sbi, end);
913 else
914 size = max_blocks;
915 map = (unsigned long *)(sentry->cur_valid_map);
916 offset = __find_rev_next_bit(map, size, offset);
917 f2fs_bug_on(sbi, offset != size);
918 blk = START_BLOCK(sbi, segno + 1);
919 }
920 #endif
921 }
922
923 static void __init_discard_policy(struct f2fs_sb_info *sbi,
924 struct discard_policy *dpolicy,
925 int discard_type, unsigned int granularity)
926 {
927 /* common policy */
928 dpolicy->type = discard_type;
929 dpolicy->sync = true;
930 dpolicy->granularity = granularity;
931
932 dpolicy->max_requests = DEF_MAX_DISCARD_REQUEST;
933 dpolicy->io_aware_gran = MAX_PLIST_NUM;
934
935 if (discard_type == DPOLICY_BG) {
936 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
937 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
938 dpolicy->io_aware = true;
939 dpolicy->sync = false;
940 if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
941 dpolicy->granularity = 1;
942 dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
943 }
944 } else if (discard_type == DPOLICY_FORCE) {
945 dpolicy->min_interval = DEF_MIN_DISCARD_ISSUE_TIME;
946 dpolicy->max_interval = DEF_MAX_DISCARD_ISSUE_TIME;
947 dpolicy->io_aware = false;
948 } else if (discard_type == DPOLICY_FSTRIM) {
949 dpolicy->io_aware = false;
950 } else if (discard_type == DPOLICY_UMOUNT) {
951 dpolicy->io_aware = false;
952 }
953 }
954
955
956 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
957 static void __submit_discard_cmd(struct f2fs_sb_info *sbi,
958 struct discard_policy *dpolicy,
959 struct discard_cmd *dc)
960 {
961 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
962 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
963 &(dcc->fstrim_list) : &(dcc->wait_list);
964 struct bio *bio = NULL;
965 int flag = dpolicy->sync ? REQ_SYNC : 0;
966
967 if (dc->state != D_PREP)
968 return;
969
970 trace_f2fs_issue_discard(dc->bdev, dc->start, dc->len);
971
972 dc->error = __blkdev_issue_discard(dc->bdev,
973 SECTOR_FROM_BLOCK(dc->start),
974 SECTOR_FROM_BLOCK(dc->len),
975 GFP_NOFS, 0, &bio);
976 if (!dc->error) {
977 /* should keep before submission to avoid D_DONE right away */
978 dc->state = D_SUBMIT;
979 atomic_inc(&dcc->issued_discard);
980 atomic_inc(&dcc->issing_discard);
981 if (bio) {
982 bio->bi_private = dc;
983 bio->bi_end_io = f2fs_submit_discard_endio;
984 bio->bi_opf |= flag;
985 submit_bio(bio);
986 list_move_tail(&dc->list, wait_list);
987 __check_sit_bitmap(sbi, dc->start, dc->start + dc->len);
988
989 f2fs_update_iostat(sbi, FS_DISCARD, 1);
990 }
991 } else {
992 __remove_discard_cmd(sbi, dc);
993 }
994 }
995
996 static struct discard_cmd *__insert_discard_tree(struct f2fs_sb_info *sbi,
997 struct block_device *bdev, block_t lstart,
998 block_t start, block_t len,
999 struct rb_node **insert_p,
1000 struct rb_node *insert_parent)
1001 {
1002 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1003 struct rb_node **p;
1004 struct rb_node *parent = NULL;
1005 struct discard_cmd *dc = NULL;
1006
1007 if (insert_p && insert_parent) {
1008 parent = insert_parent;
1009 p = insert_p;
1010 goto do_insert;
1011 }
1012
1013 p = __lookup_rb_tree_for_insert(sbi, &dcc->root, &parent, lstart);
1014 do_insert:
1015 dc = __attach_discard_cmd(sbi, bdev, lstart, start, len, parent, p);
1016 if (!dc)
1017 return NULL;
1018
1019 return dc;
1020 }
1021
1022 static void __relocate_discard_cmd(struct discard_cmd_control *dcc,
1023 struct discard_cmd *dc)
1024 {
1025 list_move_tail(&dc->list, &dcc->pend_list[plist_idx(dc->len)]);
1026 }
1027
1028 static void __punch_discard_cmd(struct f2fs_sb_info *sbi,
1029 struct discard_cmd *dc, block_t blkaddr)
1030 {
1031 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1032 struct discard_info di = dc->di;
1033 bool modified = false;
1034
1035 if (dc->state == D_DONE || dc->len == 1) {
1036 __remove_discard_cmd(sbi, dc);
1037 return;
1038 }
1039
1040 dcc->undiscard_blks -= di.len;
1041
1042 if (blkaddr > di.lstart) {
1043 dc->len = blkaddr - dc->lstart;
1044 dcc->undiscard_blks += dc->len;
1045 __relocate_discard_cmd(dcc, dc);
1046 modified = true;
1047 }
1048
1049 if (blkaddr < di.lstart + di.len - 1) {
1050 if (modified) {
1051 __insert_discard_tree(sbi, dc->bdev, blkaddr + 1,
1052 di.start + blkaddr + 1 - di.lstart,
1053 di.lstart + di.len - 1 - blkaddr,
1054 NULL, NULL);
1055 } else {
1056 dc->lstart++;
1057 dc->len--;
1058 dc->start++;
1059 dcc->undiscard_blks += dc->len;
1060 __relocate_discard_cmd(dcc, dc);
1061 }
1062 }
1063 }
1064
1065 static void __update_discard_tree_range(struct f2fs_sb_info *sbi,
1066 struct block_device *bdev, block_t lstart,
1067 block_t start, block_t len)
1068 {
1069 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1070 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
1071 struct discard_cmd *dc;
1072 struct discard_info di = {0};
1073 struct rb_node **insert_p = NULL, *insert_parent = NULL;
1074 block_t end = lstart + len;
1075
1076 mutex_lock(&dcc->cmd_lock);
1077
1078 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
1079 NULL, lstart,
1080 (struct rb_entry **)&prev_dc,
1081 (struct rb_entry **)&next_dc,
1082 &insert_p, &insert_parent, true);
1083 if (dc)
1084 prev_dc = dc;
1085
1086 if (!prev_dc) {
1087 di.lstart = lstart;
1088 di.len = next_dc ? next_dc->lstart - lstart : len;
1089 di.len = min(di.len, len);
1090 di.start = start;
1091 }
1092
1093 while (1) {
1094 struct rb_node *node;
1095 bool merged = false;
1096 struct discard_cmd *tdc = NULL;
1097
1098 if (prev_dc) {
1099 di.lstart = prev_dc->lstart + prev_dc->len;
1100 if (di.lstart < lstart)
1101 di.lstart = lstart;
1102 if (di.lstart >= end)
1103 break;
1104
1105 if (!next_dc || next_dc->lstart > end)
1106 di.len = end - di.lstart;
1107 else
1108 di.len = next_dc->lstart - di.lstart;
1109 di.start = start + di.lstart - lstart;
1110 }
1111
1112 if (!di.len)
1113 goto next;
1114
1115 if (prev_dc && prev_dc->state == D_PREP &&
1116 prev_dc->bdev == bdev &&
1117 __is_discard_back_mergeable(&di, &prev_dc->di)) {
1118 prev_dc->di.len += di.len;
1119 dcc->undiscard_blks += di.len;
1120 __relocate_discard_cmd(dcc, prev_dc);
1121 di = prev_dc->di;
1122 tdc = prev_dc;
1123 merged = true;
1124 }
1125
1126 if (next_dc && next_dc->state == D_PREP &&
1127 next_dc->bdev == bdev &&
1128 __is_discard_front_mergeable(&di, &next_dc->di)) {
1129 next_dc->di.lstart = di.lstart;
1130 next_dc->di.len += di.len;
1131 next_dc->di.start = di.start;
1132 dcc->undiscard_blks += di.len;
1133 __relocate_discard_cmd(dcc, next_dc);
1134 if (tdc)
1135 __remove_discard_cmd(sbi, tdc);
1136 merged = true;
1137 }
1138
1139 if (!merged) {
1140 __insert_discard_tree(sbi, bdev, di.lstart, di.start,
1141 di.len, NULL, NULL);
1142 }
1143 next:
1144 prev_dc = next_dc;
1145 if (!prev_dc)
1146 break;
1147
1148 node = rb_next(&prev_dc->rb_node);
1149 next_dc = rb_entry_safe(node, struct discard_cmd, rb_node);
1150 }
1151
1152 mutex_unlock(&dcc->cmd_lock);
1153 }
1154
1155 static int __queue_discard_cmd(struct f2fs_sb_info *sbi,
1156 struct block_device *bdev, block_t blkstart, block_t blklen)
1157 {
1158 block_t lblkstart = blkstart;
1159
1160 trace_f2fs_queue_discard(bdev, blkstart, blklen);
1161
1162 if (sbi->s_ndevs) {
1163 int devi = f2fs_target_device_index(sbi, blkstart);
1164
1165 blkstart -= FDEV(devi).start_blk;
1166 }
1167 __update_discard_tree_range(sbi, bdev, lblkstart, blkstart, blklen);
1168 return 0;
1169 }
1170
1171 static int __issue_discard_cmd(struct f2fs_sb_info *sbi,
1172 struct discard_policy *dpolicy)
1173 {
1174 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1175 struct list_head *pend_list;
1176 struct discard_cmd *dc, *tmp;
1177 struct blk_plug plug;
1178 int i, iter = 0, issued = 0;
1179 bool io_interrupted = false;
1180
1181 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1182 if (i + 1 < dpolicy->granularity)
1183 break;
1184 pend_list = &dcc->pend_list[i];
1185
1186 mutex_lock(&dcc->cmd_lock);
1187 if (list_empty(pend_list))
1188 goto next;
1189 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
1190 blk_start_plug(&plug);
1191 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1192 f2fs_bug_on(sbi, dc->state != D_PREP);
1193
1194 if (dpolicy->io_aware && i < dpolicy->io_aware_gran &&
1195 !is_idle(sbi)) {
1196 io_interrupted = true;
1197 goto skip;
1198 }
1199
1200 __submit_discard_cmd(sbi, dpolicy, dc);
1201 issued++;
1202 skip:
1203 if (++iter >= dpolicy->max_requests)
1204 break;
1205 }
1206 blk_finish_plug(&plug);
1207 next:
1208 mutex_unlock(&dcc->cmd_lock);
1209
1210 if (iter >= dpolicy->max_requests)
1211 break;
1212 }
1213
1214 if (!issued && io_interrupted)
1215 issued = -1;
1216
1217 return issued;
1218 }
1219
1220 static bool __drop_discard_cmd(struct f2fs_sb_info *sbi)
1221 {
1222 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1223 struct list_head *pend_list;
1224 struct discard_cmd *dc, *tmp;
1225 int i;
1226 bool dropped = false;
1227
1228 mutex_lock(&dcc->cmd_lock);
1229 for (i = MAX_PLIST_NUM - 1; i >= 0; i--) {
1230 pend_list = &dcc->pend_list[i];
1231 list_for_each_entry_safe(dc, tmp, pend_list, list) {
1232 f2fs_bug_on(sbi, dc->state != D_PREP);
1233 __remove_discard_cmd(sbi, dc);
1234 dropped = true;
1235 }
1236 }
1237 mutex_unlock(&dcc->cmd_lock);
1238
1239 return dropped;
1240 }
1241
1242 void drop_discard_cmd(struct f2fs_sb_info *sbi)
1243 {
1244 __drop_discard_cmd(sbi);
1245 }
1246
1247 static unsigned int __wait_one_discard_bio(struct f2fs_sb_info *sbi,
1248 struct discard_cmd *dc)
1249 {
1250 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1251 unsigned int len = 0;
1252
1253 wait_for_completion_io(&dc->wait);
1254 mutex_lock(&dcc->cmd_lock);
1255 f2fs_bug_on(sbi, dc->state != D_DONE);
1256 dc->ref--;
1257 if (!dc->ref) {
1258 if (!dc->error)
1259 len = dc->len;
1260 __remove_discard_cmd(sbi, dc);
1261 }
1262 mutex_unlock(&dcc->cmd_lock);
1263
1264 return len;
1265 }
1266
1267 static unsigned int __wait_discard_cmd_range(struct f2fs_sb_info *sbi,
1268 struct discard_policy *dpolicy,
1269 block_t start, block_t end)
1270 {
1271 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1272 struct list_head *wait_list = (dpolicy->type == DPOLICY_FSTRIM) ?
1273 &(dcc->fstrim_list) : &(dcc->wait_list);
1274 struct discard_cmd *dc, *tmp;
1275 bool need_wait;
1276 unsigned int trimmed = 0;
1277
1278 next:
1279 need_wait = false;
1280
1281 mutex_lock(&dcc->cmd_lock);
1282 list_for_each_entry_safe(dc, tmp, wait_list, list) {
1283 if (dc->lstart + dc->len <= start || end <= dc->lstart)
1284 continue;
1285 if (dc->len < dpolicy->granularity)
1286 continue;
1287 if (dc->state == D_DONE && !dc->ref) {
1288 wait_for_completion_io(&dc->wait);
1289 if (!dc->error)
1290 trimmed += dc->len;
1291 __remove_discard_cmd(sbi, dc);
1292 } else {
1293 dc->ref++;
1294 need_wait = true;
1295 break;
1296 }
1297 }
1298 mutex_unlock(&dcc->cmd_lock);
1299
1300 if (need_wait) {
1301 trimmed += __wait_one_discard_bio(sbi, dc);
1302 goto next;
1303 }
1304
1305 return trimmed;
1306 }
1307
1308 static void __wait_all_discard_cmd(struct f2fs_sb_info *sbi,
1309 struct discard_policy *dpolicy)
1310 {
1311 struct discard_policy dp;
1312
1313 if (dpolicy) {
1314 __wait_discard_cmd_range(sbi, dpolicy, 0, UINT_MAX);
1315 return;
1316 }
1317
1318 /* wait all */
1319 __init_discard_policy(sbi, &dp, DPOLICY_FSTRIM, 1);
1320 __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1321 __init_discard_policy(sbi, &dp, DPOLICY_UMOUNT, 1);
1322 __wait_discard_cmd_range(sbi, &dp, 0, UINT_MAX);
1323 }
1324
1325 /* This should be covered by global mutex, &sit_i->sentry_lock */
1326 static void f2fs_wait_discard_bio(struct f2fs_sb_info *sbi, block_t blkaddr)
1327 {
1328 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1329 struct discard_cmd *dc;
1330 bool need_wait = false;
1331
1332 mutex_lock(&dcc->cmd_lock);
1333 dc = (struct discard_cmd *)__lookup_rb_tree(&dcc->root, NULL, blkaddr);
1334 if (dc) {
1335 if (dc->state == D_PREP) {
1336 __punch_discard_cmd(sbi, dc, blkaddr);
1337 } else {
1338 dc->ref++;
1339 need_wait = true;
1340 }
1341 }
1342 mutex_unlock(&dcc->cmd_lock);
1343
1344 if (need_wait)
1345 __wait_one_discard_bio(sbi, dc);
1346 }
1347
1348 void stop_discard_thread(struct f2fs_sb_info *sbi)
1349 {
1350 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1351
1352 if (dcc && dcc->f2fs_issue_discard) {
1353 struct task_struct *discard_thread = dcc->f2fs_issue_discard;
1354
1355 dcc->f2fs_issue_discard = NULL;
1356 kthread_stop(discard_thread);
1357 }
1358 }
1359
1360 /* This comes from f2fs_put_super */
1361 bool f2fs_wait_discard_bios(struct f2fs_sb_info *sbi)
1362 {
1363 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1364 struct discard_policy dpolicy;
1365 bool dropped;
1366
1367 __init_discard_policy(sbi, &dpolicy, DPOLICY_UMOUNT,
1368 dcc->discard_granularity);
1369 __issue_discard_cmd(sbi, &dpolicy);
1370 dropped = __drop_discard_cmd(sbi);
1371
1372 /* just to make sure there is no pending discard commands */
1373 __wait_all_discard_cmd(sbi, NULL);
1374 return dropped;
1375 }
1376
1377 static int issue_discard_thread(void *data)
1378 {
1379 struct f2fs_sb_info *sbi = data;
1380 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1381 wait_queue_head_t *q = &dcc->discard_wait_queue;
1382 struct discard_policy dpolicy;
1383 unsigned int wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;
1384 int issued;
1385
1386 set_freezable();
1387
1388 do {
1389 __init_discard_policy(sbi, &dpolicy, DPOLICY_BG,
1390 dcc->discard_granularity);
1391
1392 wait_event_interruptible_timeout(*q,
1393 kthread_should_stop() || freezing(current) ||
1394 dcc->discard_wake,
1395 msecs_to_jiffies(wait_ms));
1396 if (try_to_freeze())
1397 continue;
1398 if (f2fs_readonly(sbi->sb))
1399 continue;
1400 if (kthread_should_stop())
1401 return 0;
1402
1403 if (dcc->discard_wake)
1404 dcc->discard_wake = 0;
1405
1406 if (sbi->gc_thread && sbi->gc_thread->gc_urgent)
1407 __init_discard_policy(sbi, &dpolicy, DPOLICY_FORCE, 1);
1408
1409 sb_start_intwrite(sbi->sb);
1410
1411 issued = __issue_discard_cmd(sbi, &dpolicy);
1412 if (issued) {
1413 __wait_all_discard_cmd(sbi, &dpolicy);
1414 wait_ms = dpolicy.min_interval;
1415 } else {
1416 wait_ms = dpolicy.max_interval;
1417 }
1418
1419 sb_end_intwrite(sbi->sb);
1420
1421 } while (!kthread_should_stop());
1422 return 0;
1423 }
1424
1425 #ifdef CONFIG_BLK_DEV_ZONED
1426 static int __f2fs_issue_discard_zone(struct f2fs_sb_info *sbi,
1427 struct block_device *bdev, block_t blkstart, block_t blklen)
1428 {
1429 sector_t sector, nr_sects;
1430 block_t lblkstart = blkstart;
1431 int devi = 0;
1432
1433 if (sbi->s_ndevs) {
1434 devi = f2fs_target_device_index(sbi, blkstart);
1435 blkstart -= FDEV(devi).start_blk;
1436 }
1437
1438 /*
1439 * We need to know the type of the zone: for conventional zones,
1440 * use regular discard if the drive supports it. For sequential
1441 * zones, reset the zone write pointer.
1442 */
1443 switch (get_blkz_type(sbi, bdev, blkstart)) {
1444
1445 case BLK_ZONE_TYPE_CONVENTIONAL:
1446 if (!blk_queue_discard(bdev_get_queue(bdev)))
1447 return 0;
1448 return __queue_discard_cmd(sbi, bdev, lblkstart, blklen);
1449 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1450 case BLK_ZONE_TYPE_SEQWRITE_PREF:
1451 sector = SECTOR_FROM_BLOCK(blkstart);
1452 nr_sects = SECTOR_FROM_BLOCK(blklen);
1453
1454 if (sector & (bdev_zone_sectors(bdev) - 1) ||
1455 nr_sects != bdev_zone_sectors(bdev)) {
1456 f2fs_msg(sbi->sb, KERN_INFO,
1457 "(%d) %s: Unaligned discard attempted (block %x + %x)",
1458 devi, sbi->s_ndevs ? FDEV(devi).path: "",
1459 blkstart, blklen);
1460 return -EIO;
1461 }
1462 trace_f2fs_issue_reset_zone(bdev, blkstart);
1463 return blkdev_reset_zones(bdev, sector,
1464 nr_sects, GFP_NOFS);
1465 default:
1466 /* Unknown zone type: broken device ? */
1467 return -EIO;
1468 }
1469 }
1470 #endif
1471
1472 static int __issue_discard_async(struct f2fs_sb_info *sbi,
1473 struct block_device *bdev, block_t blkstart, block_t blklen)
1474 {
1475 #ifdef CONFIG_BLK_DEV_ZONED
1476 if (f2fs_sb_has_blkzoned(sbi->sb) &&
1477 bdev_zoned_model(bdev) != BLK_ZONED_NONE)
1478 return __f2fs_issue_discard_zone(sbi, bdev, blkstart, blklen);
1479 #endif
1480 return __queue_discard_cmd(sbi, bdev, blkstart, blklen);
1481 }
1482
1483 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
1484 block_t blkstart, block_t blklen)
1485 {
1486 sector_t start = blkstart, len = 0;
1487 struct block_device *bdev;
1488 struct seg_entry *se;
1489 unsigned int offset;
1490 block_t i;
1491 int err = 0;
1492
1493 bdev = f2fs_target_device(sbi, blkstart, NULL);
1494
1495 for (i = blkstart; i < blkstart + blklen; i++, len++) {
1496 if (i != start) {
1497 struct block_device *bdev2 =
1498 f2fs_target_device(sbi, i, NULL);
1499
1500 if (bdev2 != bdev) {
1501 err = __issue_discard_async(sbi, bdev,
1502 start, len);
1503 if (err)
1504 return err;
1505 bdev = bdev2;
1506 start = i;
1507 len = 0;
1508 }
1509 }
1510
1511 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
1512 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
1513
1514 if (!f2fs_test_and_set_bit(offset, se->discard_map))
1515 sbi->discard_blks--;
1516 }
1517
1518 if (len)
1519 err = __issue_discard_async(sbi, bdev, start, len);
1520 return err;
1521 }
1522
1523 static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
1524 bool check_only)
1525 {
1526 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1527 int max_blocks = sbi->blocks_per_seg;
1528 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
1529 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1530 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1531 unsigned long *discard_map = (unsigned long *)se->discard_map;
1532 unsigned long *dmap = SIT_I(sbi)->tmp_map;
1533 unsigned int start = 0, end = -1;
1534 bool force = (cpc->reason & CP_DISCARD);
1535 struct discard_entry *de = NULL;
1536 struct list_head *head = &SM_I(sbi)->dcc_info->entry_list;
1537 int i;
1538
1539 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
1540 return false;
1541
1542 if (!force) {
1543 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
1544 SM_I(sbi)->dcc_info->nr_discards >=
1545 SM_I(sbi)->dcc_info->max_discards)
1546 return false;
1547 }
1548
1549 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
1550 for (i = 0; i < entries; i++)
1551 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
1552 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
1553
1554 while (force || SM_I(sbi)->dcc_info->nr_discards <=
1555 SM_I(sbi)->dcc_info->max_discards) {
1556 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
1557 if (start >= max_blocks)
1558 break;
1559
1560 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
1561 if (force && start && end != max_blocks
1562 && (end - start) < cpc->trim_minlen)
1563 continue;
1564
1565 if (check_only)
1566 return true;
1567
1568 if (!de) {
1569 de = f2fs_kmem_cache_alloc(discard_entry_slab,
1570 GFP_F2FS_ZERO);
1571 de->start_blkaddr = START_BLOCK(sbi, cpc->trim_start);
1572 list_add_tail(&de->list, head);
1573 }
1574
1575 for (i = start; i < end; i++)
1576 __set_bit_le(i, (void *)de->discard_map);
1577
1578 SM_I(sbi)->dcc_info->nr_discards += end - start;
1579 }
1580 return false;
1581 }
1582
1583 void release_discard_addrs(struct f2fs_sb_info *sbi)
1584 {
1585 struct list_head *head = &(SM_I(sbi)->dcc_info->entry_list);
1586 struct discard_entry *entry, *this;
1587
1588 /* drop caches */
1589 list_for_each_entry_safe(entry, this, head, list) {
1590 list_del(&entry->list);
1591 kmem_cache_free(discard_entry_slab, entry);
1592 }
1593 }
1594
1595 /*
1596 * Should call clear_prefree_segments after checkpoint is done.
1597 */
1598 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
1599 {
1600 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1601 unsigned int segno;
1602
1603 mutex_lock(&dirty_i->seglist_lock);
1604 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
1605 __set_test_and_free(sbi, segno);
1606 mutex_unlock(&dirty_i->seglist_lock);
1607 }
1608
1609 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1610 {
1611 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1612 struct list_head *head = &dcc->entry_list;
1613 struct discard_entry *entry, *this;
1614 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1615 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
1616 unsigned int start = 0, end = -1;
1617 unsigned int secno, start_segno;
1618 bool force = (cpc->reason & CP_DISCARD);
1619
1620 mutex_lock(&dirty_i->seglist_lock);
1621
1622 while (1) {
1623 int i;
1624 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
1625 if (start >= MAIN_SEGS(sbi))
1626 break;
1627 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
1628 start + 1);
1629
1630 for (i = start; i < end; i++)
1631 clear_bit(i, prefree_map);
1632
1633 dirty_i->nr_dirty[PRE] -= end - start;
1634
1635 if (!test_opt(sbi, DISCARD))
1636 continue;
1637
1638 if (force && start >= cpc->trim_start &&
1639 (end - 1) <= cpc->trim_end)
1640 continue;
1641
1642 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
1643 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
1644 (end - start) << sbi->log_blocks_per_seg);
1645 continue;
1646 }
1647 next:
1648 secno = GET_SEC_FROM_SEG(sbi, start);
1649 start_segno = GET_SEG_FROM_SEC(sbi, secno);
1650 if (!IS_CURSEC(sbi, secno) &&
1651 !get_valid_blocks(sbi, start, true))
1652 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
1653 sbi->segs_per_sec << sbi->log_blocks_per_seg);
1654
1655 start = start_segno + sbi->segs_per_sec;
1656 if (start < end)
1657 goto next;
1658 else
1659 end = start - 1;
1660 }
1661 mutex_unlock(&dirty_i->seglist_lock);
1662
1663 /* send small discards */
1664 list_for_each_entry_safe(entry, this, head, list) {
1665 unsigned int cur_pos = 0, next_pos, len, total_len = 0;
1666 bool is_valid = test_bit_le(0, entry->discard_map);
1667
1668 find_next:
1669 if (is_valid) {
1670 next_pos = find_next_zero_bit_le(entry->discard_map,
1671 sbi->blocks_per_seg, cur_pos);
1672 len = next_pos - cur_pos;
1673
1674 if (f2fs_sb_has_blkzoned(sbi->sb) ||
1675 (force && len < cpc->trim_minlen))
1676 goto skip;
1677
1678 f2fs_issue_discard(sbi, entry->start_blkaddr + cur_pos,
1679 len);
1680 total_len += len;
1681 } else {
1682 next_pos = find_next_bit_le(entry->discard_map,
1683 sbi->blocks_per_seg, cur_pos);
1684 }
1685 skip:
1686 cur_pos = next_pos;
1687 is_valid = !is_valid;
1688
1689 if (cur_pos < sbi->blocks_per_seg)
1690 goto find_next;
1691
1692 list_del(&entry->list);
1693 dcc->nr_discards -= total_len;
1694 kmem_cache_free(discard_entry_slab, entry);
1695 }
1696
1697 wake_up_discard_thread(sbi, false);
1698 }
1699
1700 static int create_discard_cmd_control(struct f2fs_sb_info *sbi)
1701 {
1702 dev_t dev = sbi->sb->s_bdev->bd_dev;
1703 struct discard_cmd_control *dcc;
1704 int err = 0, i;
1705
1706 if (SM_I(sbi)->dcc_info) {
1707 dcc = SM_I(sbi)->dcc_info;
1708 goto init_thread;
1709 }
1710
1711 dcc = f2fs_kzalloc(sbi, sizeof(struct discard_cmd_control), GFP_KERNEL);
1712 if (!dcc)
1713 return -ENOMEM;
1714
1715 dcc->discard_granularity = DEFAULT_DISCARD_GRANULARITY;
1716 INIT_LIST_HEAD(&dcc->entry_list);
1717 for (i = 0; i < MAX_PLIST_NUM; i++)
1718 INIT_LIST_HEAD(&dcc->pend_list[i]);
1719 INIT_LIST_HEAD(&dcc->wait_list);
1720 INIT_LIST_HEAD(&dcc->fstrim_list);
1721 mutex_init(&dcc->cmd_lock);
1722 atomic_set(&dcc->issued_discard, 0);
1723 atomic_set(&dcc->issing_discard, 0);
1724 atomic_set(&dcc->discard_cmd_cnt, 0);
1725 dcc->nr_discards = 0;
1726 dcc->max_discards = MAIN_SEGS(sbi) << sbi->log_blocks_per_seg;
1727 dcc->undiscard_blks = 0;
1728 dcc->root = RB_ROOT;
1729
1730 init_waitqueue_head(&dcc->discard_wait_queue);
1731 SM_I(sbi)->dcc_info = dcc;
1732 init_thread:
1733 dcc->f2fs_issue_discard = kthread_run(issue_discard_thread, sbi,
1734 "f2fs_discard-%u:%u", MAJOR(dev), MINOR(dev));
1735 if (IS_ERR(dcc->f2fs_issue_discard)) {
1736 err = PTR_ERR(dcc->f2fs_issue_discard);
1737 kfree(dcc);
1738 SM_I(sbi)->dcc_info = NULL;
1739 return err;
1740 }
1741
1742 return err;
1743 }
1744
1745 static void destroy_discard_cmd_control(struct f2fs_sb_info *sbi)
1746 {
1747 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
1748
1749 if (!dcc)
1750 return;
1751
1752 stop_discard_thread(sbi);
1753
1754 kfree(dcc);
1755 SM_I(sbi)->dcc_info = NULL;
1756 }
1757
1758 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
1759 {
1760 struct sit_info *sit_i = SIT_I(sbi);
1761
1762 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
1763 sit_i->dirty_sentries++;
1764 return false;
1765 }
1766
1767 return true;
1768 }
1769
1770 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
1771 unsigned int segno, int modified)
1772 {
1773 struct seg_entry *se = get_seg_entry(sbi, segno);
1774 se->type = type;
1775 if (modified)
1776 __mark_sit_entry_dirty(sbi, segno);
1777 }
1778
1779 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
1780 {
1781 struct seg_entry *se;
1782 unsigned int segno, offset;
1783 long int new_vblocks;
1784 bool exist;
1785 #ifdef CONFIG_F2FS_CHECK_FS
1786 bool mir_exist;
1787 #endif
1788
1789 segno = GET_SEGNO(sbi, blkaddr);
1790
1791 se = get_seg_entry(sbi, segno);
1792 new_vblocks = se->valid_blocks + del;
1793 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1794
1795 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
1796 (new_vblocks > sbi->blocks_per_seg)));
1797
1798 se->valid_blocks = new_vblocks;
1799 se->mtime = get_mtime(sbi);
1800 SIT_I(sbi)->max_mtime = se->mtime;
1801
1802 /* Update valid block bitmap */
1803 if (del > 0) {
1804 exist = f2fs_test_and_set_bit(offset, se->cur_valid_map);
1805 #ifdef CONFIG_F2FS_CHECK_FS
1806 mir_exist = f2fs_test_and_set_bit(offset,
1807 se->cur_valid_map_mir);
1808 if (unlikely(exist != mir_exist)) {
1809 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1810 "when setting bitmap, blk:%u, old bit:%d",
1811 blkaddr, exist);
1812 f2fs_bug_on(sbi, 1);
1813 }
1814 #endif
1815 if (unlikely(exist)) {
1816 f2fs_msg(sbi->sb, KERN_ERR,
1817 "Bitmap was wrongly set, blk:%u", blkaddr);
1818 f2fs_bug_on(sbi, 1);
1819 se->valid_blocks--;
1820 del = 0;
1821 }
1822
1823 if (f2fs_discard_en(sbi) &&
1824 !f2fs_test_and_set_bit(offset, se->discard_map))
1825 sbi->discard_blks--;
1826
1827 /* don't overwrite by SSR to keep node chain */
1828 if (IS_NODESEG(se->type)) {
1829 if (!f2fs_test_and_set_bit(offset, se->ckpt_valid_map))
1830 se->ckpt_valid_blocks++;
1831 }
1832 } else {
1833 exist = f2fs_test_and_clear_bit(offset, se->cur_valid_map);
1834 #ifdef CONFIG_F2FS_CHECK_FS
1835 mir_exist = f2fs_test_and_clear_bit(offset,
1836 se->cur_valid_map_mir);
1837 if (unlikely(exist != mir_exist)) {
1838 f2fs_msg(sbi->sb, KERN_ERR, "Inconsistent error "
1839 "when clearing bitmap, blk:%u, old bit:%d",
1840 blkaddr, exist);
1841 f2fs_bug_on(sbi, 1);
1842 }
1843 #endif
1844 if (unlikely(!exist)) {
1845 f2fs_msg(sbi->sb, KERN_ERR,
1846 "Bitmap was wrongly cleared, blk:%u", blkaddr);
1847 f2fs_bug_on(sbi, 1);
1848 se->valid_blocks++;
1849 del = 0;
1850 }
1851
1852 if (f2fs_discard_en(sbi) &&
1853 f2fs_test_and_clear_bit(offset, se->discard_map))
1854 sbi->discard_blks++;
1855 }
1856 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
1857 se->ckpt_valid_blocks += del;
1858
1859 __mark_sit_entry_dirty(sbi, segno);
1860
1861 /* update total number of valid blocks to be written in ckpt area */
1862 SIT_I(sbi)->written_valid_blocks += del;
1863
1864 if (sbi->segs_per_sec > 1)
1865 get_sec_entry(sbi, segno)->valid_blocks += del;
1866 }
1867
1868 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
1869 {
1870 unsigned int segno = GET_SEGNO(sbi, addr);
1871 struct sit_info *sit_i = SIT_I(sbi);
1872
1873 f2fs_bug_on(sbi, addr == NULL_ADDR);
1874 if (addr == NEW_ADDR)
1875 return;
1876
1877 /* add it into sit main buffer */
1878 down_write(&sit_i->sentry_lock);
1879
1880 update_sit_entry(sbi, addr, -1);
1881
1882 /* add it into dirty seglist */
1883 locate_dirty_segment(sbi, segno);
1884
1885 up_write(&sit_i->sentry_lock);
1886 }
1887
1888 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
1889 {
1890 struct sit_info *sit_i = SIT_I(sbi);
1891 unsigned int segno, offset;
1892 struct seg_entry *se;
1893 bool is_cp = false;
1894
1895 if (!is_valid_data_blkaddr(sbi, blkaddr))
1896 return true;
1897
1898 down_read(&sit_i->sentry_lock);
1899
1900 segno = GET_SEGNO(sbi, blkaddr);
1901 se = get_seg_entry(sbi, segno);
1902 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
1903
1904 if (f2fs_test_bit(offset, se->ckpt_valid_map))
1905 is_cp = true;
1906
1907 up_read(&sit_i->sentry_lock);
1908
1909 return is_cp;
1910 }
1911
1912 /*
1913 * This function should be resided under the curseg_mutex lock
1914 */
1915 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
1916 struct f2fs_summary *sum)
1917 {
1918 struct curseg_info *curseg = CURSEG_I(sbi, type);
1919 void *addr = curseg->sum_blk;
1920 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
1921 memcpy(addr, sum, sizeof(struct f2fs_summary));
1922 }
1923
1924 /*
1925 * Calculate the number of current summary pages for writing
1926 */
1927 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
1928 {
1929 int valid_sum_count = 0;
1930 int i, sum_in_page;
1931
1932 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1933 if (sbi->ckpt->alloc_type[i] == SSR)
1934 valid_sum_count += sbi->blocks_per_seg;
1935 else {
1936 if (for_ra)
1937 valid_sum_count += le16_to_cpu(
1938 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
1939 else
1940 valid_sum_count += curseg_blkoff(sbi, i);
1941 }
1942 }
1943
1944 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
1945 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
1946 if (valid_sum_count <= sum_in_page)
1947 return 1;
1948 else if ((valid_sum_count - sum_in_page) <=
1949 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
1950 return 2;
1951 return 3;
1952 }
1953
1954 /*
1955 * Caller should put this summary page
1956 */
1957 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1958 {
1959 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1960 }
1961
1962 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1963 {
1964 struct page *page = grab_meta_page(sbi, blk_addr);
1965
1966 memcpy(page_address(page), src, PAGE_SIZE);
1967 set_page_dirty(page);
1968 f2fs_put_page(page, 1);
1969 }
1970
1971 static void write_sum_page(struct f2fs_sb_info *sbi,
1972 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1973 {
1974 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1975 }
1976
1977 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1978 int type, block_t blk_addr)
1979 {
1980 struct curseg_info *curseg = CURSEG_I(sbi, type);
1981 struct page *page = grab_meta_page(sbi, blk_addr);
1982 struct f2fs_summary_block *src = curseg->sum_blk;
1983 struct f2fs_summary_block *dst;
1984
1985 dst = (struct f2fs_summary_block *)page_address(page);
1986
1987 mutex_lock(&curseg->curseg_mutex);
1988
1989 down_read(&curseg->journal_rwsem);
1990 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1991 up_read(&curseg->journal_rwsem);
1992
1993 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1994 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1995
1996 mutex_unlock(&curseg->curseg_mutex);
1997
1998 set_page_dirty(page);
1999 f2fs_put_page(page, 1);
2000 }
2001
2002 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
2003 {
2004 struct curseg_info *curseg = CURSEG_I(sbi, type);
2005 unsigned int segno = curseg->segno + 1;
2006 struct free_segmap_info *free_i = FREE_I(sbi);
2007
2008 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
2009 return !test_bit(segno, free_i->free_segmap);
2010 return 0;
2011 }
2012
2013 /*
2014 * Find a new segment from the free segments bitmap to right order
2015 * This function should be returned with success, otherwise BUG
2016 */
2017 static void get_new_segment(struct f2fs_sb_info *sbi,
2018 unsigned int *newseg, bool new_sec, int dir)
2019 {
2020 struct free_segmap_info *free_i = FREE_I(sbi);
2021 unsigned int segno, secno, zoneno;
2022 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
2023 unsigned int hint = GET_SEC_FROM_SEG(sbi, *newseg);
2024 unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
2025 unsigned int left_start = hint;
2026 bool init = true;
2027 int go_left = 0;
2028 int i;
2029
2030 spin_lock(&free_i->segmap_lock);
2031
2032 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
2033 segno = find_next_zero_bit(free_i->free_segmap,
2034 GET_SEG_FROM_SEC(sbi, hint + 1), *newseg + 1);
2035 if (segno < GET_SEG_FROM_SEC(sbi, hint + 1))
2036 goto got_it;
2037 }
2038 find_other_zone:
2039 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
2040 if (secno >= MAIN_SECS(sbi)) {
2041 if (dir == ALLOC_RIGHT) {
2042 secno = find_next_zero_bit(free_i->free_secmap,
2043 MAIN_SECS(sbi), 0);
2044 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
2045 } else {
2046 go_left = 1;
2047 left_start = hint - 1;
2048 }
2049 }
2050 if (go_left == 0)
2051 goto skip_left;
2052
2053 while (test_bit(left_start, free_i->free_secmap)) {
2054 if (left_start > 0) {
2055 left_start--;
2056 continue;
2057 }
2058 left_start = find_next_zero_bit(free_i->free_secmap,
2059 MAIN_SECS(sbi), 0);
2060 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
2061 break;
2062 }
2063 secno = left_start;
2064 skip_left:
2065 segno = GET_SEG_FROM_SEC(sbi, secno);
2066 zoneno = GET_ZONE_FROM_SEC(sbi, secno);
2067
2068 /* give up on finding another zone */
2069 if (!init)
2070 goto got_it;
2071 if (sbi->secs_per_zone == 1)
2072 goto got_it;
2073 if (zoneno == old_zoneno)
2074 goto got_it;
2075 if (dir == ALLOC_LEFT) {
2076 if (!go_left && zoneno + 1 >= total_zones)
2077 goto got_it;
2078 if (go_left && zoneno == 0)
2079 goto got_it;
2080 }
2081 for (i = 0; i < NR_CURSEG_TYPE; i++)
2082 if (CURSEG_I(sbi, i)->zone == zoneno)
2083 break;
2084
2085 if (i < NR_CURSEG_TYPE) {
2086 /* zone is in user, try another */
2087 if (go_left)
2088 hint = zoneno * sbi->secs_per_zone - 1;
2089 else if (zoneno + 1 >= total_zones)
2090 hint = 0;
2091 else
2092 hint = (zoneno + 1) * sbi->secs_per_zone;
2093 init = false;
2094 goto find_other_zone;
2095 }
2096 got_it:
2097 /* set it as dirty segment in free segmap */
2098 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
2099 __set_inuse(sbi, segno);
2100 *newseg = segno;
2101 spin_unlock(&free_i->segmap_lock);
2102 }
2103
2104 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
2105 {
2106 struct curseg_info *curseg = CURSEG_I(sbi, type);
2107 struct summary_footer *sum_footer;
2108
2109 curseg->segno = curseg->next_segno;
2110 curseg->zone = GET_ZONE_FROM_SEG(sbi, curseg->segno);
2111 curseg->next_blkoff = 0;
2112 curseg->next_segno = NULL_SEGNO;
2113
2114 sum_footer = &(curseg->sum_blk->footer);
2115 memset(sum_footer, 0, sizeof(struct summary_footer));
2116 if (IS_DATASEG(type))
2117 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
2118 if (IS_NODESEG(type))
2119 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
2120 __set_sit_entry_type(sbi, type, curseg->segno, modified);
2121 }
2122
2123 static unsigned int __get_next_segno(struct f2fs_sb_info *sbi, int type)
2124 {
2125 /* if segs_per_sec is large than 1, we need to keep original policy. */
2126 if (sbi->segs_per_sec != 1)
2127 return CURSEG_I(sbi, type)->segno;
2128
2129 if (test_opt(sbi, NOHEAP) &&
2130 (type == CURSEG_HOT_DATA || IS_NODESEG(type)))
2131 return 0;
2132
2133 if (SIT_I(sbi)->last_victim[ALLOC_NEXT])
2134 return SIT_I(sbi)->last_victim[ALLOC_NEXT];
2135
2136 /* find segments from 0 to reuse freed segments */
2137 if (F2FS_OPTION(sbi).alloc_mode == ALLOC_MODE_REUSE)
2138 return 0;
2139
2140 return CURSEG_I(sbi, type)->segno;
2141 }
2142
2143 /*
2144 * Allocate a current working segment.
2145 * This function always allocates a free segment in LFS manner.
2146 */
2147 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
2148 {
2149 struct curseg_info *curseg = CURSEG_I(sbi, type);
2150 unsigned int segno = curseg->segno;
2151 int dir = ALLOC_LEFT;
2152
2153 write_sum_page(sbi, curseg->sum_blk,
2154 GET_SUM_BLOCK(sbi, segno));
2155 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
2156 dir = ALLOC_RIGHT;
2157
2158 if (test_opt(sbi, NOHEAP))
2159 dir = ALLOC_RIGHT;
2160
2161 segno = __get_next_segno(sbi, type);
2162 get_new_segment(sbi, &segno, new_sec, dir);
2163 curseg->next_segno = segno;
2164 reset_curseg(sbi, type, 1);
2165 curseg->alloc_type = LFS;
2166 }
2167
2168 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
2169 struct curseg_info *seg, block_t start)
2170 {
2171 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
2172 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
2173 unsigned long *target_map = SIT_I(sbi)->tmp_map;
2174 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
2175 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
2176 int i, pos;
2177
2178 for (i = 0; i < entries; i++)
2179 target_map[i] = ckpt_map[i] | cur_map[i];
2180
2181 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
2182
2183 seg->next_blkoff = pos;
2184 }
2185
2186 /*
2187 * If a segment is written by LFS manner, next block offset is just obtained
2188 * by increasing the current block offset. However, if a segment is written by
2189 * SSR manner, next block offset obtained by calling __next_free_blkoff
2190 */
2191 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
2192 struct curseg_info *seg)
2193 {
2194 if (seg->alloc_type == SSR)
2195 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
2196 else
2197 seg->next_blkoff++;
2198 }
2199
2200 /*
2201 * This function always allocates a used segment(from dirty seglist) by SSR
2202 * manner, so it should recover the existing segment information of valid blocks
2203 */
2204 static void change_curseg(struct f2fs_sb_info *sbi, int type)
2205 {
2206 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2207 struct curseg_info *curseg = CURSEG_I(sbi, type);
2208 unsigned int new_segno = curseg->next_segno;
2209 struct f2fs_summary_block *sum_node;
2210 struct page *sum_page;
2211
2212 write_sum_page(sbi, curseg->sum_blk,
2213 GET_SUM_BLOCK(sbi, curseg->segno));
2214 __set_test_and_inuse(sbi, new_segno);
2215
2216 mutex_lock(&dirty_i->seglist_lock);
2217 __remove_dirty_segment(sbi, new_segno, PRE);
2218 __remove_dirty_segment(sbi, new_segno, DIRTY);
2219 mutex_unlock(&dirty_i->seglist_lock);
2220
2221 reset_curseg(sbi, type, 1);
2222 curseg->alloc_type = SSR;
2223 __next_free_blkoff(sbi, curseg, 0);
2224
2225 sum_page = get_sum_page(sbi, new_segno);
2226 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
2227 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
2228 f2fs_put_page(sum_page, 1);
2229 }
2230
2231 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
2232 {
2233 struct curseg_info *curseg = CURSEG_I(sbi, type);
2234 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
2235 unsigned segno = NULL_SEGNO;
2236 int i, cnt;
2237 bool reversed = false;
2238
2239 /* need_SSR() already forces to do this */
2240 if (v_ops->get_victim(sbi, &segno, BG_GC, type, SSR)) {
2241 curseg->next_segno = segno;
2242 return 1;
2243 }
2244
2245 /* For node segments, let's do SSR more intensively */
2246 if (IS_NODESEG(type)) {
2247 if (type >= CURSEG_WARM_NODE) {
2248 reversed = true;
2249 i = CURSEG_COLD_NODE;
2250 } else {
2251 i = CURSEG_HOT_NODE;
2252 }
2253 cnt = NR_CURSEG_NODE_TYPE;
2254 } else {
2255 if (type >= CURSEG_WARM_DATA) {
2256 reversed = true;
2257 i = CURSEG_COLD_DATA;
2258 } else {
2259 i = CURSEG_HOT_DATA;
2260 }
2261 cnt = NR_CURSEG_DATA_TYPE;
2262 }
2263
2264 for (; cnt-- > 0; reversed ? i-- : i++) {
2265 if (i == type)
2266 continue;
2267 if (v_ops->get_victim(sbi, &segno, BG_GC, i, SSR)) {
2268 curseg->next_segno = segno;
2269 return 1;
2270 }
2271 }
2272 return 0;
2273 }
2274
2275 /*
2276 * flush out current segment and replace it with new segment
2277 * This function should be returned with success, otherwise BUG
2278 */
2279 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
2280 int type, bool force)
2281 {
2282 struct curseg_info *curseg = CURSEG_I(sbi, type);
2283
2284 if (force)
2285 new_curseg(sbi, type, true);
2286 else if (!is_set_ckpt_flags(sbi, CP_CRC_RECOVERY_FLAG) &&
2287 type == CURSEG_WARM_NODE)
2288 new_curseg(sbi, type, false);
2289 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
2290 new_curseg(sbi, type, false);
2291 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
2292 change_curseg(sbi, type);
2293 else
2294 new_curseg(sbi, type, false);
2295
2296 stat_inc_seg_type(sbi, curseg);
2297 }
2298
2299 void allocate_new_segments(struct f2fs_sb_info *sbi)
2300 {
2301 struct curseg_info *curseg;
2302 unsigned int old_segno;
2303 int i;
2304
2305 down_write(&SIT_I(sbi)->sentry_lock);
2306
2307 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2308 curseg = CURSEG_I(sbi, i);
2309 old_segno = curseg->segno;
2310 SIT_I(sbi)->s_ops->allocate_segment(sbi, i, true);
2311 locate_dirty_segment(sbi, old_segno);
2312 }
2313
2314 up_write(&SIT_I(sbi)->sentry_lock);
2315 }
2316
2317 static const struct segment_allocation default_salloc_ops = {
2318 .allocate_segment = allocate_segment_by_default,
2319 };
2320
2321 bool exist_trim_candidates(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2322 {
2323 __u64 trim_start = cpc->trim_start;
2324 bool has_candidate = false;
2325
2326 down_write(&SIT_I(sbi)->sentry_lock);
2327 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
2328 if (add_discard_addrs(sbi, cpc, true)) {
2329 has_candidate = true;
2330 break;
2331 }
2332 }
2333 up_write(&SIT_I(sbi)->sentry_lock);
2334
2335 cpc->trim_start = trim_start;
2336 return has_candidate;
2337 }
2338
2339 static void __issue_discard_cmd_range(struct f2fs_sb_info *sbi,
2340 struct discard_policy *dpolicy,
2341 unsigned int start, unsigned int end)
2342 {
2343 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
2344 struct discard_cmd *prev_dc = NULL, *next_dc = NULL;
2345 struct rb_node **insert_p = NULL, *insert_parent = NULL;
2346 struct discard_cmd *dc;
2347 struct blk_plug plug;
2348 int issued;
2349
2350 next:
2351 issued = 0;
2352
2353 mutex_lock(&dcc->cmd_lock);
2354 f2fs_bug_on(sbi, !__check_rb_tree_consistence(sbi, &dcc->root));
2355
2356 dc = (struct discard_cmd *)__lookup_rb_tree_ret(&dcc->root,
2357 NULL, start,
2358 (struct rb_entry **)&prev_dc,
2359 (struct rb_entry **)&next_dc,
2360 &insert_p, &insert_parent, true);
2361 if (!dc)
2362 dc = next_dc;
2363
2364 blk_start_plug(&plug);
2365
2366 while (dc && dc->lstart <= end) {
2367 struct rb_node *node;
2368
2369 if (dc->len < dpolicy->granularity)
2370 goto skip;
2371
2372 if (dc->state != D_PREP) {
2373 list_move_tail(&dc->list, &dcc->fstrim_list);
2374 goto skip;
2375 }
2376
2377 __submit_discard_cmd(sbi, dpolicy, dc);
2378
2379 if (++issued >= dpolicy->max_requests) {
2380 start = dc->lstart + dc->len;
2381
2382 blk_finish_plug(&plug);
2383 mutex_unlock(&dcc->cmd_lock);
2384 __wait_all_discard_cmd(sbi, NULL);
2385 congestion_wait(BLK_RW_ASYNC, HZ/50);
2386 goto next;
2387 }
2388 skip:
2389 node = rb_next(&dc->rb_node);
2390 dc = rb_entry_safe(node, struct discard_cmd, rb_node);
2391
2392 if (fatal_signal_pending(current))
2393 break;
2394 }
2395
2396 blk_finish_plug(&plug);
2397 mutex_unlock(&dcc->cmd_lock);
2398 }
2399
2400 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
2401 {
2402 __u64 start = F2FS_BYTES_TO_BLK(range->start);
2403 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
2404 unsigned int start_segno, end_segno;
2405 block_t start_block, end_block;
2406 struct cp_control cpc;
2407 struct discard_policy dpolicy;
2408 unsigned long long trimmed = 0;
2409 int err = 0;
2410
2411 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
2412 return -EINVAL;
2413
2414 if (end <= MAIN_BLKADDR(sbi))
2415 goto out;
2416
2417 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
2418 f2fs_msg(sbi->sb, KERN_WARNING,
2419 "Found FS corruption, run fsck to fix.");
2420 goto out;
2421 }
2422
2423 /* start/end segment number in main_area */
2424 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
2425 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
2426 GET_SEGNO(sbi, end);
2427
2428 cpc.reason = CP_DISCARD;
2429 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
2430 cpc.trim_start = start_segno;
2431 cpc.trim_end = end_segno;
2432
2433 if (sbi->discard_blks == 0)
2434 goto out;
2435
2436 mutex_lock(&sbi->gc_mutex);
2437 err = write_checkpoint(sbi, &cpc);
2438 mutex_unlock(&sbi->gc_mutex);
2439 if (err)
2440 goto out;
2441
2442 start_block = START_BLOCK(sbi, start_segno);
2443 end_block = START_BLOCK(sbi, end_segno + 1);
2444
2445 __init_discard_policy(sbi, &dpolicy, DPOLICY_FSTRIM, cpc.trim_minlen);
2446 __issue_discard_cmd_range(sbi, &dpolicy, start_block, end_block);
2447
2448 /*
2449 * We filed discard candidates, but actually we don't need to wait for
2450 * all of them, since they'll be issued in idle time along with runtime
2451 * discard option. User configuration looks like using runtime discard
2452 * or periodic fstrim instead of it.
2453 */
2454 if (!test_opt(sbi, DISCARD)) {
2455 trimmed = __wait_discard_cmd_range(sbi, &dpolicy,
2456 start_block, end_block);
2457 range->len = F2FS_BLK_TO_BYTES(trimmed);
2458 }
2459 out:
2460 return err;
2461 }
2462
2463 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
2464 {
2465 struct curseg_info *curseg = CURSEG_I(sbi, type);
2466 if (curseg->next_blkoff < sbi->blocks_per_seg)
2467 return true;
2468 return false;
2469 }
2470
2471 int rw_hint_to_seg_type(enum rw_hint hint)
2472 {
2473 switch (hint) {
2474 case WRITE_LIFE_SHORT:
2475 return CURSEG_HOT_DATA;
2476 case WRITE_LIFE_EXTREME:
2477 return CURSEG_COLD_DATA;
2478 default:
2479 return CURSEG_WARM_DATA;
2480 }
2481 }
2482
2483 /* This returns write hints for each segment type. This hints will be
2484 * passed down to block layer. There are mapping tables which depend on
2485 * the mount option 'whint_mode'.
2486 *
2487 * 1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET.
2488 *
2489 * 2) whint_mode=user-based. F2FS tries to pass down hints given by users.
2490 *
2491 * User F2FS Block
2492 * ---- ---- -----
2493 * META WRITE_LIFE_NOT_SET
2494 * HOT_NODE "
2495 * WARM_NODE "
2496 * COLD_NODE "
2497 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2498 * extension list " "
2499 *
2500 * -- buffered io
2501 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2502 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2503 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2504 * WRITE_LIFE_NONE " "
2505 * WRITE_LIFE_MEDIUM " "
2506 * WRITE_LIFE_LONG " "
2507 *
2508 * -- direct io
2509 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2510 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2511 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2512 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2513 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2514 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2515 *
2516 * 3) whint_mode=fs-based. F2FS passes down hints with its policy.
2517 *
2518 * User F2FS Block
2519 * ---- ---- -----
2520 * META WRITE_LIFE_MEDIUM;
2521 * HOT_NODE WRITE_LIFE_NOT_SET
2522 * WARM_NODE "
2523 * COLD_NODE WRITE_LIFE_NONE
2524 * ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME
2525 * extension list " "
2526 *
2527 * -- buffered io
2528 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2529 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2530 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG
2531 * WRITE_LIFE_NONE " "
2532 * WRITE_LIFE_MEDIUM " "
2533 * WRITE_LIFE_LONG " "
2534 *
2535 * -- direct io
2536 * WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME
2537 * WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT
2538 * WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET
2539 * WRITE_LIFE_NONE " WRITE_LIFE_NONE
2540 * WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM
2541 * WRITE_LIFE_LONG " WRITE_LIFE_LONG
2542 */
2543
2544 enum rw_hint io_type_to_rw_hint(struct f2fs_sb_info *sbi,
2545 enum page_type type, enum temp_type temp)
2546 {
2547 if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_USER) {
2548 if (type == DATA) {
2549 if (temp == WARM)
2550 return WRITE_LIFE_NOT_SET;
2551 else if (temp == HOT)
2552 return WRITE_LIFE_SHORT;
2553 else if (temp == COLD)
2554 return WRITE_LIFE_EXTREME;
2555 } else {
2556 return WRITE_LIFE_NOT_SET;
2557 }
2558 } else if (F2FS_OPTION(sbi).whint_mode == WHINT_MODE_FS) {
2559 if (type == DATA) {
2560 if (temp == WARM)
2561 return WRITE_LIFE_LONG;
2562 else if (temp == HOT)
2563 return WRITE_LIFE_SHORT;
2564 else if (temp == COLD)
2565 return WRITE_LIFE_EXTREME;
2566 } else if (type == NODE) {
2567 if (temp == WARM || temp == HOT)
2568 return WRITE_LIFE_NOT_SET;
2569 else if (temp == COLD)
2570 return WRITE_LIFE_NONE;
2571 } else if (type == META) {
2572 return WRITE_LIFE_MEDIUM;
2573 }
2574 }
2575 return WRITE_LIFE_NOT_SET;
2576 }
2577
2578 static int __get_segment_type_2(struct f2fs_io_info *fio)
2579 {
2580 if (fio->type == DATA)
2581 return CURSEG_HOT_DATA;
2582 else
2583 return CURSEG_HOT_NODE;
2584 }
2585
2586 static int __get_segment_type_4(struct f2fs_io_info *fio)
2587 {
2588 if (fio->type == DATA) {
2589 struct inode *inode = fio->page->mapping->host;
2590
2591 if (S_ISDIR(inode->i_mode))
2592 return CURSEG_HOT_DATA;
2593 else
2594 return CURSEG_COLD_DATA;
2595 } else {
2596 if (IS_DNODE(fio->page) && is_cold_node(fio->page))
2597 return CURSEG_WARM_NODE;
2598 else
2599 return CURSEG_COLD_NODE;
2600 }
2601 }
2602
2603 static int __get_segment_type_6(struct f2fs_io_info *fio)
2604 {
2605 if (fio->type == DATA) {
2606 struct inode *inode = fio->page->mapping->host;
2607
2608 if (is_cold_data(fio->page) || file_is_cold(inode))
2609 return CURSEG_COLD_DATA;
2610 if (file_is_hot(inode) ||
2611 is_inode_flag_set(inode, FI_HOT_DATA))
2612 return CURSEG_HOT_DATA;
2613 return rw_hint_to_seg_type(inode->i_write_hint);
2614 } else {
2615 if (IS_DNODE(fio->page))
2616 return is_cold_node(fio->page) ? CURSEG_WARM_NODE :
2617 CURSEG_HOT_NODE;
2618 return CURSEG_COLD_NODE;
2619 }
2620 }
2621
2622 static int __get_segment_type(struct f2fs_io_info *fio)
2623 {
2624 int type = 0;
2625
2626 switch (F2FS_OPTION(fio->sbi).active_logs) {
2627 case 2:
2628 type = __get_segment_type_2(fio);
2629 break;
2630 case 4:
2631 type = __get_segment_type_4(fio);
2632 break;
2633 case 6:
2634 type = __get_segment_type_6(fio);
2635 break;
2636 default:
2637 f2fs_bug_on(fio->sbi, true);
2638 }
2639
2640 if (IS_HOT(type))
2641 fio->temp = HOT;
2642 else if (IS_WARM(type))
2643 fio->temp = WARM;
2644 else
2645 fio->temp = COLD;
2646 return type;
2647 }
2648
2649 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
2650 block_t old_blkaddr, block_t *new_blkaddr,
2651 struct f2fs_summary *sum, int type,
2652 struct f2fs_io_info *fio, bool add_list)
2653 {
2654 struct sit_info *sit_i = SIT_I(sbi);
2655 struct curseg_info *curseg = CURSEG_I(sbi, type);
2656
2657 down_read(&SM_I(sbi)->curseg_lock);
2658
2659 mutex_lock(&curseg->curseg_mutex);
2660 down_write(&sit_i->sentry_lock);
2661
2662 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
2663
2664 f2fs_wait_discard_bio(sbi, *new_blkaddr);
2665
2666 /*
2667 * __add_sum_entry should be resided under the curseg_mutex
2668 * because, this function updates a summary entry in the
2669 * current summary block.
2670 */
2671 __add_sum_entry(sbi, type, sum);
2672
2673 __refresh_next_blkoff(sbi, curseg);
2674
2675 stat_inc_block_count(sbi, curseg);
2676
2677 /*
2678 * SIT information should be updated before segment allocation,
2679 * since SSR needs latest valid block information.
2680 */
2681 update_sit_entry(sbi, *new_blkaddr, 1);
2682 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2683 update_sit_entry(sbi, old_blkaddr, -1);
2684
2685 if (!__has_curseg_space(sbi, type))
2686 sit_i->s_ops->allocate_segment(sbi, type, false);
2687
2688 /*
2689 * segment dirty status should be updated after segment allocation,
2690 * so we just need to update status only one time after previous
2691 * segment being closed.
2692 */
2693 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2694 locate_dirty_segment(sbi, GET_SEGNO(sbi, *new_blkaddr));
2695
2696 up_write(&sit_i->sentry_lock);
2697
2698 if (page && IS_NODESEG(type)) {
2699 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
2700
2701 f2fs_inode_chksum_set(sbi, page);
2702 }
2703
2704 if (add_list) {
2705 struct f2fs_bio_info *io;
2706
2707 INIT_LIST_HEAD(&fio->list);
2708 fio->in_list = true;
2709 io = sbi->write_io[fio->type] + fio->temp;
2710 spin_lock(&io->io_lock);
2711 list_add_tail(&fio->list, &io->io_list);
2712 spin_unlock(&io->io_lock);
2713 }
2714
2715 mutex_unlock(&curseg->curseg_mutex);
2716
2717 up_read(&SM_I(sbi)->curseg_lock);
2718 }
2719
2720 static void update_device_state(struct f2fs_io_info *fio)
2721 {
2722 struct f2fs_sb_info *sbi = fio->sbi;
2723 unsigned int devidx;
2724
2725 if (!sbi->s_ndevs)
2726 return;
2727
2728 devidx = f2fs_target_device_index(sbi, fio->new_blkaddr);
2729
2730 /* update device state for fsync */
2731 set_dirty_device(sbi, fio->ino, devidx, FLUSH_INO);
2732
2733 /* update device state for checkpoint */
2734 if (!f2fs_test_bit(devidx, (char *)&sbi->dirty_device)) {
2735 spin_lock(&sbi->dev_lock);
2736 f2fs_set_bit(devidx, (char *)&sbi->dirty_device);
2737 spin_unlock(&sbi->dev_lock);
2738 }
2739 }
2740
2741 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
2742 {
2743 int type = __get_segment_type(fio);
2744 int err;
2745
2746 reallocate:
2747 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
2748 &fio->new_blkaddr, sum, type, fio, true);
2749
2750 /* writeout dirty page into bdev */
2751 err = f2fs_submit_page_write(fio);
2752 if (err == -EAGAIN) {
2753 fio->old_blkaddr = fio->new_blkaddr;
2754 goto reallocate;
2755 } else if (!err) {
2756 update_device_state(fio);
2757 }
2758 }
2759
2760 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page,
2761 enum iostat_type io_type)
2762 {
2763 struct f2fs_io_info fio = {
2764 .sbi = sbi,
2765 .type = META,
2766 .temp = HOT,
2767 .op = REQ_OP_WRITE,
2768 .op_flags = REQ_SYNC | REQ_META | REQ_PRIO,
2769 .old_blkaddr = page->index,
2770 .new_blkaddr = page->index,
2771 .page = page,
2772 .encrypted_page = NULL,
2773 .in_list = false,
2774 };
2775
2776 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
2777 fio.op_flags &= ~REQ_META;
2778
2779 set_page_writeback(page);
2780 ClearPageError(page);
2781 f2fs_submit_page_write(&fio);
2782
2783 f2fs_update_iostat(sbi, io_type, F2FS_BLKSIZE);
2784 }
2785
2786 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
2787 {
2788 struct f2fs_summary sum;
2789
2790 set_summary(&sum, nid, 0, 0);
2791 do_write_page(&sum, fio);
2792
2793 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2794 }
2795
2796 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
2797 {
2798 struct f2fs_sb_info *sbi = fio->sbi;
2799 struct f2fs_summary sum;
2800 struct node_info ni;
2801
2802 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
2803 get_node_info(sbi, dn->nid, &ni);
2804 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
2805 do_write_page(&sum, fio);
2806 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
2807
2808 f2fs_update_iostat(sbi, fio->io_type, F2FS_BLKSIZE);
2809 }
2810
2811 int rewrite_data_page(struct f2fs_io_info *fio)
2812 {
2813 int err;
2814 struct f2fs_sb_info *sbi = fio->sbi;
2815
2816 fio->new_blkaddr = fio->old_blkaddr;
2817 /* i/o temperature is needed for passing down write hints */
2818 __get_segment_type(fio);
2819
2820 f2fs_bug_on(sbi, !IS_DATASEG(get_seg_entry(sbi,
2821 GET_SEGNO(sbi, fio->new_blkaddr))->type));
2822
2823 stat_inc_inplace_blocks(fio->sbi);
2824
2825 err = f2fs_submit_page_bio(fio);
2826 if (!err)
2827 update_device_state(fio);
2828
2829 f2fs_update_iostat(fio->sbi, fio->io_type, F2FS_BLKSIZE);
2830
2831 return err;
2832 }
2833
2834 static inline int __f2fs_get_curseg(struct f2fs_sb_info *sbi,
2835 unsigned int segno)
2836 {
2837 int i;
2838
2839 for (i = CURSEG_HOT_DATA; i < NO_CHECK_TYPE; i++) {
2840 if (CURSEG_I(sbi, i)->segno == segno)
2841 break;
2842 }
2843 return i;
2844 }
2845
2846 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
2847 block_t old_blkaddr, block_t new_blkaddr,
2848 bool recover_curseg, bool recover_newaddr)
2849 {
2850 struct sit_info *sit_i = SIT_I(sbi);
2851 struct curseg_info *curseg;
2852 unsigned int segno, old_cursegno;
2853 struct seg_entry *se;
2854 int type;
2855 unsigned short old_blkoff;
2856
2857 segno = GET_SEGNO(sbi, new_blkaddr);
2858 se = get_seg_entry(sbi, segno);
2859 type = se->type;
2860
2861 down_write(&SM_I(sbi)->curseg_lock);
2862
2863 if (!recover_curseg) {
2864 /* for recovery flow */
2865 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
2866 if (old_blkaddr == NULL_ADDR)
2867 type = CURSEG_COLD_DATA;
2868 else
2869 type = CURSEG_WARM_DATA;
2870 }
2871 } else {
2872 if (IS_CURSEG(sbi, segno)) {
2873 /* se->type is volatile as SSR allocation */
2874 type = __f2fs_get_curseg(sbi, segno);
2875 f2fs_bug_on(sbi, type == NO_CHECK_TYPE);
2876 } else {
2877 type = CURSEG_WARM_DATA;
2878 }
2879 }
2880
2881 f2fs_bug_on(sbi, !IS_DATASEG(type));
2882 curseg = CURSEG_I(sbi, type);
2883
2884 mutex_lock(&curseg->curseg_mutex);
2885 down_write(&sit_i->sentry_lock);
2886
2887 old_cursegno = curseg->segno;
2888 old_blkoff = curseg->next_blkoff;
2889
2890 /* change the current segment */
2891 if (segno != curseg->segno) {
2892 curseg->next_segno = segno;
2893 change_curseg(sbi, type);
2894 }
2895
2896 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
2897 __add_sum_entry(sbi, type, sum);
2898
2899 if (!recover_curseg || recover_newaddr)
2900 update_sit_entry(sbi, new_blkaddr, 1);
2901 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
2902 update_sit_entry(sbi, old_blkaddr, -1);
2903
2904 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
2905 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
2906
2907 locate_dirty_segment(sbi, old_cursegno);
2908
2909 if (recover_curseg) {
2910 if (old_cursegno != curseg->segno) {
2911 curseg->next_segno = old_cursegno;
2912 change_curseg(sbi, type);
2913 }
2914 curseg->next_blkoff = old_blkoff;
2915 }
2916
2917 up_write(&sit_i->sentry_lock);
2918 mutex_unlock(&curseg->curseg_mutex);
2919 up_write(&SM_I(sbi)->curseg_lock);
2920 }
2921
2922 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
2923 block_t old_addr, block_t new_addr,
2924 unsigned char version, bool recover_curseg,
2925 bool recover_newaddr)
2926 {
2927 struct f2fs_summary sum;
2928
2929 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
2930
2931 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
2932 recover_curseg, recover_newaddr);
2933
2934 f2fs_update_data_blkaddr(dn, new_addr);
2935 }
2936
2937 void f2fs_wait_on_page_writeback(struct page *page,
2938 enum page_type type, bool ordered)
2939 {
2940 if (PageWriteback(page)) {
2941 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
2942
2943 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
2944 0, page->index, type);
2945 if (ordered)
2946 wait_on_page_writeback(page);
2947 else
2948 wait_for_stable_page(page);
2949 }
2950 }
2951
2952 void f2fs_wait_on_block_writeback(struct inode *inode, block_t blkaddr)
2953 {
2954 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2955 struct page *cpage;
2956
2957 if (!f2fs_post_read_required(inode))
2958 return;
2959
2960 if (!is_valid_data_blkaddr(sbi, blkaddr))
2961 return;
2962
2963 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
2964 if (cpage) {
2965 f2fs_wait_on_page_writeback(cpage, DATA, true);
2966 f2fs_put_page(cpage, 1);
2967 }
2968 }
2969
2970 static void read_compacted_summaries(struct f2fs_sb_info *sbi)
2971 {
2972 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2973 struct curseg_info *seg_i;
2974 unsigned char *kaddr;
2975 struct page *page;
2976 block_t start;
2977 int i, j, offset;
2978
2979 start = start_sum_block(sbi);
2980
2981 page = get_meta_page(sbi, start++);
2982 kaddr = (unsigned char *)page_address(page);
2983
2984 /* Step 1: restore nat cache */
2985 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
2986 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
2987
2988 /* Step 2: restore sit cache */
2989 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
2990 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
2991 offset = 2 * SUM_JOURNAL_SIZE;
2992
2993 /* Step 3: restore summary entries */
2994 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
2995 unsigned short blk_off;
2996 unsigned int segno;
2997
2998 seg_i = CURSEG_I(sbi, i);
2999 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
3000 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
3001 seg_i->next_segno = segno;
3002 reset_curseg(sbi, i, 0);
3003 seg_i->alloc_type = ckpt->alloc_type[i];
3004 seg_i->next_blkoff = blk_off;
3005
3006 if (seg_i->alloc_type == SSR)
3007 blk_off = sbi->blocks_per_seg;
3008
3009 for (j = 0; j < blk_off; j++) {
3010 struct f2fs_summary *s;
3011 s = (struct f2fs_summary *)(kaddr + offset);
3012 seg_i->sum_blk->entries[j] = *s;
3013 offset += SUMMARY_SIZE;
3014 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
3015 SUM_FOOTER_SIZE)
3016 continue;
3017
3018 f2fs_put_page(page, 1);
3019 page = NULL;
3020
3021 page = get_meta_page(sbi, start++);
3022 kaddr = (unsigned char *)page_address(page);
3023 offset = 0;
3024 }
3025 }
3026 f2fs_put_page(page, 1);
3027 }
3028
3029 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
3030 {
3031 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3032 struct f2fs_summary_block *sum;
3033 struct curseg_info *curseg;
3034 struct page *new;
3035 unsigned short blk_off;
3036 unsigned int segno = 0;
3037 block_t blk_addr = 0;
3038
3039 /* get segment number and block addr */
3040 if (IS_DATASEG(type)) {
3041 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
3042 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
3043 CURSEG_HOT_DATA]);
3044 if (__exist_node_summaries(sbi))
3045 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
3046 else
3047 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
3048 } else {
3049 segno = le32_to_cpu(ckpt->cur_node_segno[type -
3050 CURSEG_HOT_NODE]);
3051 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
3052 CURSEG_HOT_NODE]);
3053 if (__exist_node_summaries(sbi))
3054 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
3055 type - CURSEG_HOT_NODE);
3056 else
3057 blk_addr = GET_SUM_BLOCK(sbi, segno);
3058 }
3059
3060 new = get_meta_page(sbi, blk_addr);
3061 sum = (struct f2fs_summary_block *)page_address(new);
3062
3063 if (IS_NODESEG(type)) {
3064 if (__exist_node_summaries(sbi)) {
3065 struct f2fs_summary *ns = &sum->entries[0];
3066 int i;
3067 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
3068 ns->version = 0;
3069 ns->ofs_in_node = 0;
3070 }
3071 } else {
3072 restore_node_summary(sbi, segno, sum);
3073 }
3074 }
3075
3076 /* set uncompleted segment to curseg */
3077 curseg = CURSEG_I(sbi, type);
3078 mutex_lock(&curseg->curseg_mutex);
3079
3080 /* update journal info */
3081 down_write(&curseg->journal_rwsem);
3082 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
3083 up_write(&curseg->journal_rwsem);
3084
3085 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
3086 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
3087 curseg->next_segno = segno;
3088 reset_curseg(sbi, type, 0);
3089 curseg->alloc_type = ckpt->alloc_type[type];
3090 curseg->next_blkoff = blk_off;
3091 mutex_unlock(&curseg->curseg_mutex);
3092 f2fs_put_page(new, 1);
3093 return 0;
3094 }
3095
3096 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
3097 {
3098 struct f2fs_journal *sit_j = CURSEG_I(sbi, CURSEG_COLD_DATA)->journal;
3099 struct f2fs_journal *nat_j = CURSEG_I(sbi, CURSEG_HOT_DATA)->journal;
3100 int type = CURSEG_HOT_DATA;
3101 int err;
3102
3103 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
3104 int npages = npages_for_summary_flush(sbi, true);
3105
3106 if (npages >= 2)
3107 ra_meta_pages(sbi, start_sum_block(sbi), npages,
3108 META_CP, true);
3109
3110 /* restore for compacted data summary */
3111 read_compacted_summaries(sbi);
3112 type = CURSEG_HOT_NODE;
3113 }
3114
3115 if (__exist_node_summaries(sbi))
3116 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
3117 NR_CURSEG_TYPE - type, META_CP, true);
3118
3119 for (; type <= CURSEG_COLD_NODE; type++) {
3120 err = read_normal_summaries(sbi, type);
3121 if (err)
3122 return err;
3123 }
3124
3125 /* sanity check for summary blocks */
3126 if (nats_in_cursum(nat_j) > NAT_JOURNAL_ENTRIES ||
3127 sits_in_cursum(sit_j) > SIT_JOURNAL_ENTRIES)
3128 return -EINVAL;
3129
3130 return 0;
3131 }
3132
3133 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
3134 {
3135 struct page *page;
3136 unsigned char *kaddr;
3137 struct f2fs_summary *summary;
3138 struct curseg_info *seg_i;
3139 int written_size = 0;
3140 int i, j;
3141
3142 page = grab_meta_page(sbi, blkaddr++);
3143 kaddr = (unsigned char *)page_address(page);
3144
3145 /* Step 1: write nat cache */
3146 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
3147 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
3148 written_size += SUM_JOURNAL_SIZE;
3149
3150 /* Step 2: write sit cache */
3151 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
3152 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
3153 written_size += SUM_JOURNAL_SIZE;
3154
3155 /* Step 3: write summary entries */
3156 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
3157 unsigned short blkoff;
3158 seg_i = CURSEG_I(sbi, i);
3159 if (sbi->ckpt->alloc_type[i] == SSR)
3160 blkoff = sbi->blocks_per_seg;
3161 else
3162 blkoff = curseg_blkoff(sbi, i);
3163
3164 for (j = 0; j < blkoff; j++) {
3165 if (!page) {
3166 page = grab_meta_page(sbi, blkaddr++);
3167 kaddr = (unsigned char *)page_address(page);
3168 written_size = 0;
3169 }
3170 summary = (struct f2fs_summary *)(kaddr + written_size);
3171 *summary = seg_i->sum_blk->entries[j];
3172 written_size += SUMMARY_SIZE;
3173
3174 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
3175 SUM_FOOTER_SIZE)
3176 continue;
3177
3178 set_page_dirty(page);
3179 f2fs_put_page(page, 1);
3180 page = NULL;
3181 }
3182 }
3183 if (page) {
3184 set_page_dirty(page);
3185 f2fs_put_page(page, 1);
3186 }
3187 }
3188
3189 static void write_normal_summaries(struct f2fs_sb_info *sbi,
3190 block_t blkaddr, int type)
3191 {
3192 int i, end;
3193 if (IS_DATASEG(type))
3194 end = type + NR_CURSEG_DATA_TYPE;
3195 else
3196 end = type + NR_CURSEG_NODE_TYPE;
3197
3198 for (i = type; i < end; i++)
3199 write_current_sum_page(sbi, i, blkaddr + (i - type));
3200 }
3201
3202 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3203 {
3204 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
3205 write_compacted_summaries(sbi, start_blk);
3206 else
3207 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
3208 }
3209
3210 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
3211 {
3212 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
3213 }
3214
3215 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
3216 unsigned int val, int alloc)
3217 {
3218 int i;
3219
3220 if (type == NAT_JOURNAL) {
3221 for (i = 0; i < nats_in_cursum(journal); i++) {
3222 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
3223 return i;
3224 }
3225 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
3226 return update_nats_in_cursum(journal, 1);
3227 } else if (type == SIT_JOURNAL) {
3228 for (i = 0; i < sits_in_cursum(journal); i++)
3229 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
3230 return i;
3231 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
3232 return update_sits_in_cursum(journal, 1);
3233 }
3234 return -1;
3235 }
3236
3237 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
3238 unsigned int segno)
3239 {
3240 return get_meta_page(sbi, current_sit_addr(sbi, segno));
3241 }
3242
3243 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
3244 unsigned int start)
3245 {
3246 struct sit_info *sit_i = SIT_I(sbi);
3247 struct page *page;
3248 pgoff_t src_off, dst_off;
3249
3250 src_off = current_sit_addr(sbi, start);
3251 dst_off = next_sit_addr(sbi, src_off);
3252
3253 page = grab_meta_page(sbi, dst_off);
3254 seg_info_to_sit_page(sbi, page, start);
3255
3256 set_page_dirty(page);
3257 set_to_next_sit(sit_i, start);
3258
3259 return page;
3260 }
3261
3262 static struct sit_entry_set *grab_sit_entry_set(void)
3263 {
3264 struct sit_entry_set *ses =
3265 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
3266
3267 ses->entry_cnt = 0;
3268 INIT_LIST_HEAD(&ses->set_list);
3269 return ses;
3270 }
3271
3272 static void release_sit_entry_set(struct sit_entry_set *ses)
3273 {
3274 list_del(&ses->set_list);
3275 kmem_cache_free(sit_entry_set_slab, ses);
3276 }
3277
3278 static void adjust_sit_entry_set(struct sit_entry_set *ses,
3279 struct list_head *head)
3280 {
3281 struct sit_entry_set *next = ses;
3282
3283 if (list_is_last(&ses->set_list, head))
3284 return;
3285
3286 list_for_each_entry_continue(next, head, set_list)
3287 if (ses->entry_cnt <= next->entry_cnt)
3288 break;
3289
3290 list_move_tail(&ses->set_list, &next->set_list);
3291 }
3292
3293 static void add_sit_entry(unsigned int segno, struct list_head *head)
3294 {
3295 struct sit_entry_set *ses;
3296 unsigned int start_segno = START_SEGNO(segno);
3297
3298 list_for_each_entry(ses, head, set_list) {
3299 if (ses->start_segno == start_segno) {
3300 ses->entry_cnt++;
3301 adjust_sit_entry_set(ses, head);
3302 return;
3303 }
3304 }
3305
3306 ses = grab_sit_entry_set();
3307
3308 ses->start_segno = start_segno;
3309 ses->entry_cnt++;
3310 list_add(&ses->set_list, head);
3311 }
3312
3313 static void add_sits_in_set(struct f2fs_sb_info *sbi)
3314 {
3315 struct f2fs_sm_info *sm_info = SM_I(sbi);
3316 struct list_head *set_list = &sm_info->sit_entry_set;
3317 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
3318 unsigned int segno;
3319
3320 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
3321 add_sit_entry(segno, set_list);
3322 }
3323
3324 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
3325 {
3326 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3327 struct f2fs_journal *journal = curseg->journal;
3328 int i;
3329
3330 down_write(&curseg->journal_rwsem);
3331 for (i = 0; i < sits_in_cursum(journal); i++) {
3332 unsigned int segno;
3333 bool dirtied;
3334
3335 segno = le32_to_cpu(segno_in_journal(journal, i));
3336 dirtied = __mark_sit_entry_dirty(sbi, segno);
3337
3338 if (!dirtied)
3339 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
3340 }
3341 update_sits_in_cursum(journal, -i);
3342 up_write(&curseg->journal_rwsem);
3343 }
3344
3345 /*
3346 * CP calls this function, which flushes SIT entries including sit_journal,
3347 * and moves prefree segs to free segs.
3348 */
3349 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3350 {
3351 struct sit_info *sit_i = SIT_I(sbi);
3352 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
3353 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3354 struct f2fs_journal *journal = curseg->journal;
3355 struct sit_entry_set *ses, *tmp;
3356 struct list_head *head = &SM_I(sbi)->sit_entry_set;
3357 bool to_journal = true;
3358 struct seg_entry *se;
3359
3360 down_write(&sit_i->sentry_lock);
3361
3362 if (!sit_i->dirty_sentries)
3363 goto out;
3364
3365 /*
3366 * add and account sit entries of dirty bitmap in sit entry
3367 * set temporarily
3368 */
3369 add_sits_in_set(sbi);
3370
3371 /*
3372 * if there are no enough space in journal to store dirty sit
3373 * entries, remove all entries from journal and add and account
3374 * them in sit entry set.
3375 */
3376 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
3377 remove_sits_in_journal(sbi);
3378
3379 /*
3380 * there are two steps to flush sit entries:
3381 * #1, flush sit entries to journal in current cold data summary block.
3382 * #2, flush sit entries to sit page.
3383 */
3384 list_for_each_entry_safe(ses, tmp, head, set_list) {
3385 struct page *page = NULL;
3386 struct f2fs_sit_block *raw_sit = NULL;
3387 unsigned int start_segno = ses->start_segno;
3388 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
3389 (unsigned long)MAIN_SEGS(sbi));
3390 unsigned int segno = start_segno;
3391
3392 if (to_journal &&
3393 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
3394 to_journal = false;
3395
3396 if (to_journal) {
3397 down_write(&curseg->journal_rwsem);
3398 } else {
3399 page = get_next_sit_page(sbi, start_segno);
3400 raw_sit = page_address(page);
3401 }
3402
3403 /* flush dirty sit entries in region of current sit set */
3404 for_each_set_bit_from(segno, bitmap, end) {
3405 int offset, sit_offset;
3406
3407 se = get_seg_entry(sbi, segno);
3408
3409 /* add discard candidates */
3410 if (!(cpc->reason & CP_DISCARD)) {
3411 cpc->trim_start = segno;
3412 add_discard_addrs(sbi, cpc, false);
3413 }
3414
3415 if (to_journal) {
3416 offset = lookup_journal_in_cursum(journal,
3417 SIT_JOURNAL, segno, 1);
3418 f2fs_bug_on(sbi, offset < 0);
3419 segno_in_journal(journal, offset) =
3420 cpu_to_le32(segno);
3421 seg_info_to_raw_sit(se,
3422 &sit_in_journal(journal, offset));
3423 } else {
3424 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
3425 seg_info_to_raw_sit(se,
3426 &raw_sit->entries[sit_offset]);
3427 }
3428
3429 __clear_bit(segno, bitmap);
3430 sit_i->dirty_sentries--;
3431 ses->entry_cnt--;
3432 }
3433
3434 if (to_journal)
3435 up_write(&curseg->journal_rwsem);
3436 else
3437 f2fs_put_page(page, 1);
3438
3439 f2fs_bug_on(sbi, ses->entry_cnt);
3440 release_sit_entry_set(ses);
3441 }
3442
3443 f2fs_bug_on(sbi, !list_empty(head));
3444 f2fs_bug_on(sbi, sit_i->dirty_sentries);
3445 out:
3446 if (cpc->reason & CP_DISCARD) {
3447 __u64 trim_start = cpc->trim_start;
3448
3449 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
3450 add_discard_addrs(sbi, cpc, false);
3451
3452 cpc->trim_start = trim_start;
3453 }
3454 up_write(&sit_i->sentry_lock);
3455
3456 set_prefree_as_free_segments(sbi);
3457 }
3458
3459 static int build_sit_info(struct f2fs_sb_info *sbi)
3460 {
3461 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3462 struct sit_info *sit_i;
3463 unsigned int sit_segs, start;
3464 char *src_bitmap;
3465 unsigned int bitmap_size;
3466
3467 /* allocate memory for SIT information */
3468 sit_i = f2fs_kzalloc(sbi, sizeof(struct sit_info), GFP_KERNEL);
3469 if (!sit_i)
3470 return -ENOMEM;
3471
3472 SM_I(sbi)->sit_info = sit_i;
3473
3474 sit_i->sentries = f2fs_kvzalloc(sbi, MAIN_SEGS(sbi) *
3475 sizeof(struct seg_entry), GFP_KERNEL);
3476 if (!sit_i->sentries)
3477 return -ENOMEM;
3478
3479 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3480 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(sbi, bitmap_size,
3481 GFP_KERNEL);
3482 if (!sit_i->dirty_sentries_bitmap)
3483 return -ENOMEM;
3484
3485 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3486 sit_i->sentries[start].cur_valid_map
3487 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3488 sit_i->sentries[start].ckpt_valid_map
3489 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3490 if (!sit_i->sentries[start].cur_valid_map ||
3491 !sit_i->sentries[start].ckpt_valid_map)
3492 return -ENOMEM;
3493
3494 #ifdef CONFIG_F2FS_CHECK_FS
3495 sit_i->sentries[start].cur_valid_map_mir
3496 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3497 if (!sit_i->sentries[start].cur_valid_map_mir)
3498 return -ENOMEM;
3499 #endif
3500
3501 if (f2fs_discard_en(sbi)) {
3502 sit_i->sentries[start].discard_map
3503 = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE,
3504 GFP_KERNEL);
3505 if (!sit_i->sentries[start].discard_map)
3506 return -ENOMEM;
3507 }
3508 }
3509
3510 sit_i->tmp_map = f2fs_kzalloc(sbi, SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
3511 if (!sit_i->tmp_map)
3512 return -ENOMEM;
3513
3514 if (sbi->segs_per_sec > 1) {
3515 sit_i->sec_entries = f2fs_kvzalloc(sbi, MAIN_SECS(sbi) *
3516 sizeof(struct sec_entry), GFP_KERNEL);
3517 if (!sit_i->sec_entries)
3518 return -ENOMEM;
3519 }
3520
3521 /* get information related with SIT */
3522 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
3523
3524 /* setup SIT bitmap from ckeckpoint pack */
3525 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
3526 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
3527
3528 sit_i->sit_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3529 if (!sit_i->sit_bitmap)
3530 return -ENOMEM;
3531
3532 #ifdef CONFIG_F2FS_CHECK_FS
3533 sit_i->sit_bitmap_mir = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
3534 if (!sit_i->sit_bitmap_mir)
3535 return -ENOMEM;
3536 #endif
3537
3538 /* init SIT information */
3539 sit_i->s_ops = &default_salloc_ops;
3540
3541 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
3542 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
3543 sit_i->written_valid_blocks = 0;
3544 sit_i->bitmap_size = bitmap_size;
3545 sit_i->dirty_sentries = 0;
3546 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
3547 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
3548 sit_i->mounted_time = ktime_get_real_seconds();
3549 init_rwsem(&sit_i->sentry_lock);
3550 return 0;
3551 }
3552
3553 static int build_free_segmap(struct f2fs_sb_info *sbi)
3554 {
3555 struct free_segmap_info *free_i;
3556 unsigned int bitmap_size, sec_bitmap_size;
3557
3558 /* allocate memory for free segmap information */
3559 free_i = f2fs_kzalloc(sbi, sizeof(struct free_segmap_info), GFP_KERNEL);
3560 if (!free_i)
3561 return -ENOMEM;
3562
3563 SM_I(sbi)->free_info = free_i;
3564
3565 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3566 free_i->free_segmap = f2fs_kvmalloc(sbi, bitmap_size, GFP_KERNEL);
3567 if (!free_i->free_segmap)
3568 return -ENOMEM;
3569
3570 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3571 free_i->free_secmap = f2fs_kvmalloc(sbi, sec_bitmap_size, GFP_KERNEL);
3572 if (!free_i->free_secmap)
3573 return -ENOMEM;
3574
3575 /* set all segments as dirty temporarily */
3576 memset(free_i->free_segmap, 0xff, bitmap_size);
3577 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
3578
3579 /* init free segmap information */
3580 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
3581 free_i->free_segments = 0;
3582 free_i->free_sections = 0;
3583 spin_lock_init(&free_i->segmap_lock);
3584 return 0;
3585 }
3586
3587 static int build_curseg(struct f2fs_sb_info *sbi)
3588 {
3589 struct curseg_info *array;
3590 int i;
3591
3592 array = f2fs_kzalloc(sbi, sizeof(*array) * NR_CURSEG_TYPE, GFP_KERNEL);
3593 if (!array)
3594 return -ENOMEM;
3595
3596 SM_I(sbi)->curseg_array = array;
3597
3598 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3599 mutex_init(&array[i].curseg_mutex);
3600 array[i].sum_blk = f2fs_kzalloc(sbi, PAGE_SIZE, GFP_KERNEL);
3601 if (!array[i].sum_blk)
3602 return -ENOMEM;
3603 init_rwsem(&array[i].journal_rwsem);
3604 array[i].journal = f2fs_kzalloc(sbi,
3605 sizeof(struct f2fs_journal), GFP_KERNEL);
3606 if (!array[i].journal)
3607 return -ENOMEM;
3608 array[i].segno = NULL_SEGNO;
3609 array[i].next_blkoff = 0;
3610 }
3611 return restore_curseg_summaries(sbi);
3612 }
3613
3614 static int build_sit_entries(struct f2fs_sb_info *sbi)
3615 {
3616 struct sit_info *sit_i = SIT_I(sbi);
3617 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
3618 struct f2fs_journal *journal = curseg->journal;
3619 struct seg_entry *se;
3620 struct f2fs_sit_entry sit;
3621 int sit_blk_cnt = SIT_BLK_CNT(sbi);
3622 unsigned int i, start, end;
3623 unsigned int readed, start_blk = 0;
3624 int err = 0;
3625 block_t total_node_blocks = 0;
3626
3627 do {
3628 readed = ra_meta_pages(sbi, start_blk, BIO_MAX_PAGES,
3629 META_SIT, true);
3630
3631 start = start_blk * sit_i->sents_per_block;
3632 end = (start_blk + readed) * sit_i->sents_per_block;
3633
3634 for (; start < end && start < MAIN_SEGS(sbi); start++) {
3635 struct f2fs_sit_block *sit_blk;
3636 struct page *page;
3637
3638 se = &sit_i->sentries[start];
3639 page = get_current_sit_page(sbi, start);
3640 sit_blk = (struct f2fs_sit_block *)page_address(page);
3641 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
3642 f2fs_put_page(page, 1);
3643
3644 err = check_block_count(sbi, start, &sit);
3645 if (err)
3646 return err;
3647 seg_info_from_raw_sit(se, &sit);
3648 if (IS_NODESEG(se->type))
3649 total_node_blocks += se->valid_blocks;
3650
3651 /* build discard map only one time */
3652 if (f2fs_discard_en(sbi)) {
3653 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3654 memset(se->discard_map, 0xff,
3655 SIT_VBLOCK_MAP_SIZE);
3656 } else {
3657 memcpy(se->discard_map,
3658 se->cur_valid_map,
3659 SIT_VBLOCK_MAP_SIZE);
3660 sbi->discard_blks +=
3661 sbi->blocks_per_seg -
3662 se->valid_blocks;
3663 }
3664 }
3665
3666 if (sbi->segs_per_sec > 1)
3667 get_sec_entry(sbi, start)->valid_blocks +=
3668 se->valid_blocks;
3669 }
3670 start_blk += readed;
3671 } while (start_blk < sit_blk_cnt);
3672
3673 down_read(&curseg->journal_rwsem);
3674 for (i = 0; i < sits_in_cursum(journal); i++) {
3675 unsigned int old_valid_blocks;
3676
3677 start = le32_to_cpu(segno_in_journal(journal, i));
3678 if (start >= MAIN_SEGS(sbi)) {
3679 f2fs_msg(sbi->sb, KERN_ERR,
3680 "Wrong journal entry on segno %u",
3681 start);
3682 set_sbi_flag(sbi, SBI_NEED_FSCK);
3683 err = -EINVAL;
3684 break;
3685 }
3686
3687 se = &sit_i->sentries[start];
3688 sit = sit_in_journal(journal, i);
3689
3690 old_valid_blocks = se->valid_blocks;
3691 if (IS_NODESEG(se->type))
3692 total_node_blocks -= old_valid_blocks;
3693
3694 err = check_block_count(sbi, start, &sit);
3695 if (err)
3696 break;
3697 seg_info_from_raw_sit(se, &sit);
3698 if (IS_NODESEG(se->type))
3699 total_node_blocks += se->valid_blocks;
3700
3701 if (f2fs_discard_en(sbi)) {
3702 if (is_set_ckpt_flags(sbi, CP_TRIMMED_FLAG)) {
3703 memset(se->discard_map, 0xff,
3704 SIT_VBLOCK_MAP_SIZE);
3705 } else {
3706 memcpy(se->discard_map, se->cur_valid_map,
3707 SIT_VBLOCK_MAP_SIZE);
3708 sbi->discard_blks += old_valid_blocks -
3709 se->valid_blocks;
3710 }
3711 }
3712
3713 if (sbi->segs_per_sec > 1)
3714 get_sec_entry(sbi, start)->valid_blocks +=
3715 se->valid_blocks - old_valid_blocks;
3716 }
3717 up_read(&curseg->journal_rwsem);
3718
3719 if (!err && total_node_blocks != valid_node_count(sbi)) {
3720 f2fs_msg(sbi->sb, KERN_ERR,
3721 "SIT is corrupted node# %u vs %u",
3722 total_node_blocks, valid_node_count(sbi));
3723 set_sbi_flag(sbi, SBI_NEED_FSCK);
3724 err = -EINVAL;
3725 }
3726
3727 return err;
3728 }
3729
3730 static void init_free_segmap(struct f2fs_sb_info *sbi)
3731 {
3732 unsigned int start;
3733 int type;
3734
3735 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3736 struct seg_entry *sentry = get_seg_entry(sbi, start);
3737 if (!sentry->valid_blocks)
3738 __set_free(sbi, start);
3739 else
3740 SIT_I(sbi)->written_valid_blocks +=
3741 sentry->valid_blocks;
3742 }
3743
3744 /* set use the current segments */
3745 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
3746 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
3747 __set_test_and_inuse(sbi, curseg_t->segno);
3748 }
3749 }
3750
3751 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
3752 {
3753 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3754 struct free_segmap_info *free_i = FREE_I(sbi);
3755 unsigned int segno = 0, offset = 0;
3756 unsigned short valid_blocks;
3757
3758 while (1) {
3759 /* find dirty segment based on free segmap */
3760 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
3761 if (segno >= MAIN_SEGS(sbi))
3762 break;
3763 offset = segno + 1;
3764 valid_blocks = get_valid_blocks(sbi, segno, false);
3765 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
3766 continue;
3767 if (valid_blocks > sbi->blocks_per_seg) {
3768 f2fs_bug_on(sbi, 1);
3769 continue;
3770 }
3771 mutex_lock(&dirty_i->seglist_lock);
3772 __locate_dirty_segment(sbi, segno, DIRTY);
3773 mutex_unlock(&dirty_i->seglist_lock);
3774 }
3775 }
3776
3777 static int init_victim_secmap(struct f2fs_sb_info *sbi)
3778 {
3779 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3780 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
3781
3782 dirty_i->victim_secmap = f2fs_kvzalloc(sbi, bitmap_size, GFP_KERNEL);
3783 if (!dirty_i->victim_secmap)
3784 return -ENOMEM;
3785 return 0;
3786 }
3787
3788 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
3789 {
3790 struct dirty_seglist_info *dirty_i;
3791 unsigned int bitmap_size, i;
3792
3793 /* allocate memory for dirty segments list information */
3794 dirty_i = f2fs_kzalloc(sbi, sizeof(struct dirty_seglist_info),
3795 GFP_KERNEL);
3796 if (!dirty_i)
3797 return -ENOMEM;
3798
3799 SM_I(sbi)->dirty_info = dirty_i;
3800 mutex_init(&dirty_i->seglist_lock);
3801
3802 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
3803
3804 for (i = 0; i < NR_DIRTY_TYPE; i++) {
3805 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(sbi, bitmap_size,
3806 GFP_KERNEL);
3807 if (!dirty_i->dirty_segmap[i])
3808 return -ENOMEM;
3809 }
3810
3811 init_dirty_segmap(sbi);
3812 return init_victim_secmap(sbi);
3813 }
3814
3815 /*
3816 * Update min, max modified time for cost-benefit GC algorithm
3817 */
3818 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
3819 {
3820 struct sit_info *sit_i = SIT_I(sbi);
3821 unsigned int segno;
3822
3823 down_write(&sit_i->sentry_lock);
3824
3825 sit_i->min_mtime = LLONG_MAX;
3826
3827 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
3828 unsigned int i;
3829 unsigned long long mtime = 0;
3830
3831 for (i = 0; i < sbi->segs_per_sec; i++)
3832 mtime += get_seg_entry(sbi, segno + i)->mtime;
3833
3834 mtime = div_u64(mtime, sbi->segs_per_sec);
3835
3836 if (sit_i->min_mtime > mtime)
3837 sit_i->min_mtime = mtime;
3838 }
3839 sit_i->max_mtime = get_mtime(sbi);
3840 up_write(&sit_i->sentry_lock);
3841 }
3842
3843 int build_segment_manager(struct f2fs_sb_info *sbi)
3844 {
3845 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
3846 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3847 struct f2fs_sm_info *sm_info;
3848 int err;
3849
3850 sm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_sm_info), GFP_KERNEL);
3851 if (!sm_info)
3852 return -ENOMEM;
3853
3854 /* init sm info */
3855 sbi->sm_info = sm_info;
3856 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
3857 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
3858 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
3859 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
3860 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
3861 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
3862 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
3863 sm_info->rec_prefree_segments = sm_info->main_segments *
3864 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
3865 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
3866 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
3867
3868 if (!test_opt(sbi, LFS))
3869 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
3870 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
3871 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
3872 sm_info->min_hot_blocks = DEF_MIN_HOT_BLOCKS;
3873 sm_info->min_ssr_sections = reserved_sections(sbi);
3874
3875 INIT_LIST_HEAD(&sm_info->sit_entry_set);
3876
3877 init_rwsem(&sm_info->curseg_lock);
3878
3879 if (!f2fs_readonly(sbi->sb)) {
3880 err = create_flush_cmd_control(sbi);
3881 if (err)
3882 return err;
3883 }
3884
3885 err = create_discard_cmd_control(sbi);
3886 if (err)
3887 return err;
3888
3889 err = build_sit_info(sbi);
3890 if (err)
3891 return err;
3892 err = build_free_segmap(sbi);
3893 if (err)
3894 return err;
3895 err = build_curseg(sbi);
3896 if (err)
3897 return err;
3898
3899 /* reinit free segmap based on SIT */
3900 err = build_sit_entries(sbi);
3901 if (err)
3902 return err;
3903
3904 init_free_segmap(sbi);
3905 err = build_dirty_segmap(sbi);
3906 if (err)
3907 return err;
3908
3909 init_min_max_mtime(sbi);
3910 return 0;
3911 }
3912
3913 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
3914 enum dirty_type dirty_type)
3915 {
3916 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3917
3918 mutex_lock(&dirty_i->seglist_lock);
3919 kvfree(dirty_i->dirty_segmap[dirty_type]);
3920 dirty_i->nr_dirty[dirty_type] = 0;
3921 mutex_unlock(&dirty_i->seglist_lock);
3922 }
3923
3924 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
3925 {
3926 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3927 kvfree(dirty_i->victim_secmap);
3928 }
3929
3930 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
3931 {
3932 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
3933 int i;
3934
3935 if (!dirty_i)
3936 return;
3937
3938 /* discard pre-free/dirty segments list */
3939 for (i = 0; i < NR_DIRTY_TYPE; i++)
3940 discard_dirty_segmap(sbi, i);
3941
3942 destroy_victim_secmap(sbi);
3943 SM_I(sbi)->dirty_info = NULL;
3944 kfree(dirty_i);
3945 }
3946
3947 static void destroy_curseg(struct f2fs_sb_info *sbi)
3948 {
3949 struct curseg_info *array = SM_I(sbi)->curseg_array;
3950 int i;
3951
3952 if (!array)
3953 return;
3954 SM_I(sbi)->curseg_array = NULL;
3955 for (i = 0; i < NR_CURSEG_TYPE; i++) {
3956 kfree(array[i].sum_blk);
3957 kfree(array[i].journal);
3958 }
3959 kfree(array);
3960 }
3961
3962 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
3963 {
3964 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
3965 if (!free_i)
3966 return;
3967 SM_I(sbi)->free_info = NULL;
3968 kvfree(free_i->free_segmap);
3969 kvfree(free_i->free_secmap);
3970 kfree(free_i);
3971 }
3972
3973 static void destroy_sit_info(struct f2fs_sb_info *sbi)
3974 {
3975 struct sit_info *sit_i = SIT_I(sbi);
3976 unsigned int start;
3977
3978 if (!sit_i)
3979 return;
3980
3981 if (sit_i->sentries) {
3982 for (start = 0; start < MAIN_SEGS(sbi); start++) {
3983 kfree(sit_i->sentries[start].cur_valid_map);
3984 #ifdef CONFIG_F2FS_CHECK_FS
3985 kfree(sit_i->sentries[start].cur_valid_map_mir);
3986 #endif
3987 kfree(sit_i->sentries[start].ckpt_valid_map);
3988 kfree(sit_i->sentries[start].discard_map);
3989 }
3990 }
3991 kfree(sit_i->tmp_map);
3992
3993 kvfree(sit_i->sentries);
3994 kvfree(sit_i->sec_entries);
3995 kvfree(sit_i->dirty_sentries_bitmap);
3996
3997 SM_I(sbi)->sit_info = NULL;
3998 kfree(sit_i->sit_bitmap);
3999 #ifdef CONFIG_F2FS_CHECK_FS
4000 kfree(sit_i->sit_bitmap_mir);
4001 #endif
4002 kfree(sit_i);
4003 }
4004
4005 void destroy_segment_manager(struct f2fs_sb_info *sbi)
4006 {
4007 struct f2fs_sm_info *sm_info = SM_I(sbi);
4008
4009 if (!sm_info)
4010 return;
4011 destroy_flush_cmd_control(sbi, true);
4012 destroy_discard_cmd_control(sbi);
4013 destroy_dirty_segmap(sbi);
4014 destroy_curseg(sbi);
4015 destroy_free_segmap(sbi);
4016 destroy_sit_info(sbi);
4017 sbi->sm_info = NULL;
4018 kfree(sm_info);
4019 }
4020
4021 int __init create_segment_manager_caches(void)
4022 {
4023 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
4024 sizeof(struct discard_entry));
4025 if (!discard_entry_slab)
4026 goto fail;
4027
4028 discard_cmd_slab = f2fs_kmem_cache_create("discard_cmd",
4029 sizeof(struct discard_cmd));
4030 if (!discard_cmd_slab)
4031 goto destroy_discard_entry;
4032
4033 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
4034 sizeof(struct sit_entry_set));
4035 if (!sit_entry_set_slab)
4036 goto destroy_discard_cmd;
4037
4038 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
4039 sizeof(struct inmem_pages));
4040 if (!inmem_entry_slab)
4041 goto destroy_sit_entry_set;
4042 return 0;
4043
4044 destroy_sit_entry_set:
4045 kmem_cache_destroy(sit_entry_set_slab);
4046 destroy_discard_cmd:
4047 kmem_cache_destroy(discard_cmd_slab);
4048 destroy_discard_entry:
4049 kmem_cache_destroy(discard_entry_slab);
4050 fail:
4051 return -ENOMEM;
4052 }
4053
4054 void destroy_segment_manager_caches(void)
4055 {
4056 kmem_cache_destroy(sit_entry_set_slab);
4057 kmem_cache_destroy(discard_cmd_slab);
4058 kmem_cache_destroy(discard_entry_slab);
4059 kmem_cache_destroy(inmem_entry_slab);
4060 }
Motorola kernel-slsi