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