]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - fs/f2fs/segment.c
f2fs: fix sparse warnings
[mirror_ubuntu-zesty-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
20 #include "f2fs.h"
21 #include "segment.h"
22 #include "node.h"
23 #include "trace.h"
24 #include <trace/events/f2fs.h>
25
26 #define __reverse_ffz(x) __reverse_ffs(~(x))
27
28 static struct kmem_cache *discard_entry_slab;
29 static struct kmem_cache *bio_entry_slab;
30 static struct kmem_cache *sit_entry_set_slab;
31 static struct kmem_cache *inmem_entry_slab;
32
33 static unsigned long __reverse_ulong(unsigned char *str)
34 {
35 unsigned long tmp = 0;
36 int shift = 24, idx = 0;
37
38 #if BITS_PER_LONG == 64
39 shift = 56;
40 #endif
41 while (shift >= 0) {
42 tmp |= (unsigned long)str[idx++] << shift;
43 shift -= BITS_PER_BYTE;
44 }
45 return tmp;
46 }
47
48 /*
49 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
50 * MSB and LSB are reversed in a byte by f2fs_set_bit.
51 */
52 static inline unsigned long __reverse_ffs(unsigned long word)
53 {
54 int num = 0;
55
56 #if BITS_PER_LONG == 64
57 if ((word & 0xffffffff00000000UL) == 0)
58 num += 32;
59 else
60 word >>= 32;
61 #endif
62 if ((word & 0xffff0000) == 0)
63 num += 16;
64 else
65 word >>= 16;
66
67 if ((word & 0xff00) == 0)
68 num += 8;
69 else
70 word >>= 8;
71
72 if ((word & 0xf0) == 0)
73 num += 4;
74 else
75 word >>= 4;
76
77 if ((word & 0xc) == 0)
78 num += 2;
79 else
80 word >>= 2;
81
82 if ((word & 0x2) == 0)
83 num += 1;
84 return num;
85 }
86
87 /*
88 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
89 * f2fs_set_bit makes MSB and LSB reversed in a byte.
90 * @size must be integral times of unsigned long.
91 * Example:
92 * MSB <--> LSB
93 * f2fs_set_bit(0, bitmap) => 1000 0000
94 * f2fs_set_bit(7, bitmap) => 0000 0001
95 */
96 static unsigned long __find_rev_next_bit(const unsigned long *addr,
97 unsigned long size, unsigned long offset)
98 {
99 const unsigned long *p = addr + BIT_WORD(offset);
100 unsigned long result = size;
101 unsigned long tmp;
102
103 if (offset >= size)
104 return size;
105
106 size -= (offset & ~(BITS_PER_LONG - 1));
107 offset %= BITS_PER_LONG;
108
109 while (1) {
110 if (*p == 0)
111 goto pass;
112
113 tmp = __reverse_ulong((unsigned char *)p);
114
115 tmp &= ~0UL >> offset;
116 if (size < BITS_PER_LONG)
117 tmp &= (~0UL << (BITS_PER_LONG - size));
118 if (tmp)
119 goto found;
120 pass:
121 if (size <= BITS_PER_LONG)
122 break;
123 size -= BITS_PER_LONG;
124 offset = 0;
125 p++;
126 }
127 return result;
128 found:
129 return result - size + __reverse_ffs(tmp);
130 }
131
132 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
133 unsigned long size, unsigned long offset)
134 {
135 const unsigned long *p = addr + BIT_WORD(offset);
136 unsigned long result = size;
137 unsigned long tmp;
138
139 if (offset >= size)
140 return size;
141
142 size -= (offset & ~(BITS_PER_LONG - 1));
143 offset %= BITS_PER_LONG;
144
145 while (1) {
146 if (*p == ~0UL)
147 goto pass;
148
149 tmp = __reverse_ulong((unsigned char *)p);
150
151 if (offset)
152 tmp |= ~0UL << (BITS_PER_LONG - offset);
153 if (size < BITS_PER_LONG)
154 tmp |= ~0UL >> size;
155 if (tmp != ~0UL)
156 goto found;
157 pass:
158 if (size <= BITS_PER_LONG)
159 break;
160 size -= BITS_PER_LONG;
161 offset = 0;
162 p++;
163 }
164 return result;
165 found:
166 return result - size + __reverse_ffz(tmp);
167 }
168
169 void register_inmem_page(struct inode *inode, struct page *page)
170 {
171 struct f2fs_inode_info *fi = F2FS_I(inode);
172 struct inmem_pages *new;
173
174 f2fs_trace_pid(page);
175
176 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
177 SetPagePrivate(page);
178
179 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
180
181 /* add atomic page indices to the list */
182 new->page = page;
183 INIT_LIST_HEAD(&new->list);
184
185 /* increase reference count with clean state */
186 mutex_lock(&fi->inmem_lock);
187 get_page(page);
188 list_add_tail(&new->list, &fi->inmem_pages);
189 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
190 mutex_unlock(&fi->inmem_lock);
191
192 trace_f2fs_register_inmem_page(page, INMEM);
193 }
194
195 static int __revoke_inmem_pages(struct inode *inode,
196 struct list_head *head, bool drop, bool recover)
197 {
198 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
199 struct inmem_pages *cur, *tmp;
200 int err = 0;
201
202 list_for_each_entry_safe(cur, tmp, head, list) {
203 struct page *page = cur->page;
204
205 if (drop)
206 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
207
208 lock_page(page);
209
210 if (recover) {
211 struct dnode_of_data dn;
212 struct node_info ni;
213
214 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
215
216 set_new_dnode(&dn, inode, NULL, NULL, 0);
217 if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
218 err = -EAGAIN;
219 goto next;
220 }
221 get_node_info(sbi, dn.nid, &ni);
222 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
223 cur->old_addr, ni.version, true, true);
224 f2fs_put_dnode(&dn);
225 }
226 next:
227 /* we don't need to invalidate this in the sccessful status */
228 if (drop || recover)
229 ClearPageUptodate(page);
230 set_page_private(page, 0);
231 ClearPagePrivate(page);
232 f2fs_put_page(page, 1);
233
234 list_del(&cur->list);
235 kmem_cache_free(inmem_entry_slab, cur);
236 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
237 }
238 return err;
239 }
240
241 void drop_inmem_pages(struct inode *inode)
242 {
243 struct f2fs_inode_info *fi = F2FS_I(inode);
244
245 clear_inode_flag(inode, FI_ATOMIC_FILE);
246
247 mutex_lock(&fi->inmem_lock);
248 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
249 mutex_unlock(&fi->inmem_lock);
250 }
251
252 static int __commit_inmem_pages(struct inode *inode,
253 struct list_head *revoke_list)
254 {
255 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
256 struct f2fs_inode_info *fi = F2FS_I(inode);
257 struct inmem_pages *cur, *tmp;
258 struct f2fs_io_info fio = {
259 .sbi = sbi,
260 .type = DATA,
261 .op = REQ_OP_WRITE,
262 .op_flags = WRITE_SYNC | REQ_PRIO,
263 .encrypted_page = NULL,
264 };
265 bool submit_bio = false;
266 int err = 0;
267
268 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
269 struct page *page = cur->page;
270
271 lock_page(page);
272 if (page->mapping == inode->i_mapping) {
273 trace_f2fs_commit_inmem_page(page, INMEM);
274
275 set_page_dirty(page);
276 f2fs_wait_on_page_writeback(page, DATA, true);
277 if (clear_page_dirty_for_io(page)) {
278 inode_dec_dirty_pages(inode);
279 remove_dirty_inode(inode);
280 }
281
282 fio.page = page;
283 err = do_write_data_page(&fio);
284 if (err) {
285 unlock_page(page);
286 break;
287 }
288
289 /* record old blkaddr for revoking */
290 cur->old_addr = fio.old_blkaddr;
291
292 clear_cold_data(page);
293 submit_bio = true;
294 }
295 unlock_page(page);
296 list_move_tail(&cur->list, revoke_list);
297 }
298
299 if (submit_bio)
300 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
301
302 if (!err)
303 __revoke_inmem_pages(inode, revoke_list, false, false);
304
305 return err;
306 }
307
308 int commit_inmem_pages(struct inode *inode)
309 {
310 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
311 struct f2fs_inode_info *fi = F2FS_I(inode);
312 struct list_head revoke_list;
313 int err;
314
315 INIT_LIST_HEAD(&revoke_list);
316 f2fs_balance_fs(sbi, true);
317 f2fs_lock_op(sbi);
318
319 mutex_lock(&fi->inmem_lock);
320 err = __commit_inmem_pages(inode, &revoke_list);
321 if (err) {
322 int ret;
323 /*
324 * try to revoke all committed pages, but still we could fail
325 * due to no memory or other reason, if that happened, EAGAIN
326 * will be returned, which means in such case, transaction is
327 * already not integrity, caller should use journal to do the
328 * recovery or rewrite & commit last transaction. For other
329 * error number, revoking was done by filesystem itself.
330 */
331 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
332 if (ret)
333 err = ret;
334
335 /* drop all uncommitted pages */
336 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
337 }
338 mutex_unlock(&fi->inmem_lock);
339
340 f2fs_unlock_op(sbi);
341 return err;
342 }
343
344 /*
345 * This function balances dirty node and dentry pages.
346 * In addition, it controls garbage collection.
347 */
348 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
349 {
350 #ifdef CONFIG_F2FS_FAULT_INJECTION
351 if (time_to_inject(sbi, FAULT_CHECKPOINT))
352 f2fs_stop_checkpoint(sbi, false);
353 #endif
354
355 if (!need)
356 return;
357
358 /* balance_fs_bg is able to be pending */
359 if (excess_cached_nats(sbi))
360 f2fs_balance_fs_bg(sbi);
361
362 /*
363 * We should do GC or end up with checkpoint, if there are so many dirty
364 * dir/node pages without enough free segments.
365 */
366 if (has_not_enough_free_secs(sbi, 0, 0)) {
367 mutex_lock(&sbi->gc_mutex);
368 f2fs_gc(sbi, false);
369 }
370 }
371
372 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
373 {
374 /* try to shrink extent cache when there is no enough memory */
375 if (!available_free_memory(sbi, EXTENT_CACHE))
376 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
377
378 /* check the # of cached NAT entries */
379 if (!available_free_memory(sbi, NAT_ENTRIES))
380 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
381
382 if (!available_free_memory(sbi, FREE_NIDS))
383 try_to_free_nids(sbi, MAX_FREE_NIDS);
384 else
385 build_free_nids(sbi);
386
387 /* checkpoint is the only way to shrink partial cached entries */
388 if (!available_free_memory(sbi, NAT_ENTRIES) ||
389 !available_free_memory(sbi, INO_ENTRIES) ||
390 excess_prefree_segs(sbi) ||
391 excess_dirty_nats(sbi) ||
392 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
393 if (test_opt(sbi, DATA_FLUSH)) {
394 struct blk_plug plug;
395
396 blk_start_plug(&plug);
397 sync_dirty_inodes(sbi, FILE_INODE);
398 blk_finish_plug(&plug);
399 }
400 f2fs_sync_fs(sbi->sb, true);
401 stat_inc_bg_cp_count(sbi->stat_info);
402 }
403 }
404
405 static int issue_flush_thread(void *data)
406 {
407 struct f2fs_sb_info *sbi = data;
408 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
409 wait_queue_head_t *q = &fcc->flush_wait_queue;
410 repeat:
411 if (kthread_should_stop())
412 return 0;
413
414 if (!llist_empty(&fcc->issue_list)) {
415 struct bio *bio;
416 struct flush_cmd *cmd, *next;
417 int ret;
418
419 bio = f2fs_bio_alloc(0);
420
421 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
422 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
423
424 bio->bi_bdev = sbi->sb->s_bdev;
425 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
426 ret = submit_bio_wait(bio);
427
428 llist_for_each_entry_safe(cmd, next,
429 fcc->dispatch_list, llnode) {
430 cmd->ret = ret;
431 complete(&cmd->wait);
432 }
433 bio_put(bio);
434 fcc->dispatch_list = NULL;
435 }
436
437 wait_event_interruptible(*q,
438 kthread_should_stop() || !llist_empty(&fcc->issue_list));
439 goto repeat;
440 }
441
442 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
443 {
444 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
445 struct flush_cmd cmd;
446
447 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
448 test_opt(sbi, FLUSH_MERGE));
449
450 if (test_opt(sbi, NOBARRIER))
451 return 0;
452
453 if (!test_opt(sbi, FLUSH_MERGE) || !atomic_read(&fcc->submit_flush)) {
454 struct bio *bio = f2fs_bio_alloc(0);
455 int ret;
456
457 atomic_inc(&fcc->submit_flush);
458 bio->bi_bdev = sbi->sb->s_bdev;
459 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
460 ret = submit_bio_wait(bio);
461 atomic_dec(&fcc->submit_flush);
462 bio_put(bio);
463 return ret;
464 }
465
466 init_completion(&cmd.wait);
467
468 atomic_inc(&fcc->submit_flush);
469 llist_add(&cmd.llnode, &fcc->issue_list);
470
471 if (!fcc->dispatch_list)
472 wake_up(&fcc->flush_wait_queue);
473
474 wait_for_completion(&cmd.wait);
475 atomic_dec(&fcc->submit_flush);
476
477 return cmd.ret;
478 }
479
480 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
481 {
482 dev_t dev = sbi->sb->s_bdev->bd_dev;
483 struct flush_cmd_control *fcc;
484 int err = 0;
485
486 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
487 if (!fcc)
488 return -ENOMEM;
489 atomic_set(&fcc->submit_flush, 0);
490 init_waitqueue_head(&fcc->flush_wait_queue);
491 init_llist_head(&fcc->issue_list);
492 SM_I(sbi)->cmd_control_info = fcc;
493 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
494 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
495 if (IS_ERR(fcc->f2fs_issue_flush)) {
496 err = PTR_ERR(fcc->f2fs_issue_flush);
497 kfree(fcc);
498 SM_I(sbi)->cmd_control_info = NULL;
499 return err;
500 }
501
502 return err;
503 }
504
505 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
506 {
507 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
508
509 if (fcc && fcc->f2fs_issue_flush)
510 kthread_stop(fcc->f2fs_issue_flush);
511 kfree(fcc);
512 SM_I(sbi)->cmd_control_info = NULL;
513 }
514
515 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
516 enum dirty_type dirty_type)
517 {
518 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
519
520 /* need not be added */
521 if (IS_CURSEG(sbi, segno))
522 return;
523
524 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
525 dirty_i->nr_dirty[dirty_type]++;
526
527 if (dirty_type == DIRTY) {
528 struct seg_entry *sentry = get_seg_entry(sbi, segno);
529 enum dirty_type t = sentry->type;
530
531 if (unlikely(t >= DIRTY)) {
532 f2fs_bug_on(sbi, 1);
533 return;
534 }
535 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
536 dirty_i->nr_dirty[t]++;
537 }
538 }
539
540 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
541 enum dirty_type dirty_type)
542 {
543 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
544
545 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
546 dirty_i->nr_dirty[dirty_type]--;
547
548 if (dirty_type == DIRTY) {
549 struct seg_entry *sentry = get_seg_entry(sbi, segno);
550 enum dirty_type t = sentry->type;
551
552 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
553 dirty_i->nr_dirty[t]--;
554
555 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
556 clear_bit(GET_SECNO(sbi, segno),
557 dirty_i->victim_secmap);
558 }
559 }
560
561 /*
562 * Should not occur error such as -ENOMEM.
563 * Adding dirty entry into seglist is not critical operation.
564 * If a given segment is one of current working segments, it won't be added.
565 */
566 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
567 {
568 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
569 unsigned short valid_blocks;
570
571 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
572 return;
573
574 mutex_lock(&dirty_i->seglist_lock);
575
576 valid_blocks = get_valid_blocks(sbi, segno, 0);
577
578 if (valid_blocks == 0) {
579 __locate_dirty_segment(sbi, segno, PRE);
580 __remove_dirty_segment(sbi, segno, DIRTY);
581 } else if (valid_blocks < sbi->blocks_per_seg) {
582 __locate_dirty_segment(sbi, segno, DIRTY);
583 } else {
584 /* Recovery routine with SSR needs this */
585 __remove_dirty_segment(sbi, segno, DIRTY);
586 }
587
588 mutex_unlock(&dirty_i->seglist_lock);
589 }
590
591 static struct bio_entry *__add_bio_entry(struct f2fs_sb_info *sbi,
592 struct bio *bio)
593 {
594 struct list_head *wait_list = &(SM_I(sbi)->wait_list);
595 struct bio_entry *be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS);
596
597 INIT_LIST_HEAD(&be->list);
598 be->bio = bio;
599 init_completion(&be->event);
600 list_add_tail(&be->list, wait_list);
601
602 return be;
603 }
604
605 void f2fs_wait_all_discard_bio(struct f2fs_sb_info *sbi)
606 {
607 struct list_head *wait_list = &(SM_I(sbi)->wait_list);
608 struct bio_entry *be, *tmp;
609
610 list_for_each_entry_safe(be, tmp, wait_list, list) {
611 struct bio *bio = be->bio;
612 int err;
613
614 wait_for_completion_io(&be->event);
615 err = be->error;
616 if (err == -EOPNOTSUPP)
617 err = 0;
618
619 if (err)
620 f2fs_msg(sbi->sb, KERN_INFO,
621 "Issue discard failed, ret: %d", err);
622
623 bio_put(bio);
624 list_del(&be->list);
625 kmem_cache_free(bio_entry_slab, be);
626 }
627 }
628
629 static void f2fs_submit_bio_wait_endio(struct bio *bio)
630 {
631 struct bio_entry *be = (struct bio_entry *)bio->bi_private;
632
633 be->error = bio->bi_error;
634 complete(&be->event);
635 }
636
637 /* this function is copied from blkdev_issue_discard from block/blk-lib.c */
638 static int __f2fs_issue_discard_async(struct f2fs_sb_info *sbi, sector_t sector,
639 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags)
640 {
641 struct block_device *bdev = sbi->sb->s_bdev;
642 struct bio *bio = NULL;
643 int err;
644
645 err = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, flags,
646 &bio);
647 if (!err && bio) {
648 struct bio_entry *be = __add_bio_entry(sbi, bio);
649
650 bio->bi_private = be;
651 bio->bi_end_io = f2fs_submit_bio_wait_endio;
652 bio->bi_opf |= REQ_SYNC;
653 submit_bio(bio);
654 }
655
656 return err;
657 }
658
659 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
660 block_t blkstart, block_t blklen)
661 {
662 sector_t start = SECTOR_FROM_BLOCK(blkstart);
663 sector_t len = SECTOR_FROM_BLOCK(blklen);
664 struct seg_entry *se;
665 unsigned int offset;
666 block_t i;
667
668 for (i = blkstart; i < blkstart + blklen; i++) {
669 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
670 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
671
672 if (!f2fs_test_and_set_bit(offset, se->discard_map))
673 sbi->discard_blks--;
674 }
675 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
676 return __f2fs_issue_discard_async(sbi, start, len, GFP_NOFS, 0);
677 }
678
679 static void __add_discard_entry(struct f2fs_sb_info *sbi,
680 struct cp_control *cpc, struct seg_entry *se,
681 unsigned int start, unsigned int end)
682 {
683 struct list_head *head = &SM_I(sbi)->discard_list;
684 struct discard_entry *new, *last;
685
686 if (!list_empty(head)) {
687 last = list_last_entry(head, struct discard_entry, list);
688 if (START_BLOCK(sbi, cpc->trim_start) + start ==
689 last->blkaddr + last->len) {
690 last->len += end - start;
691 goto done;
692 }
693 }
694
695 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
696 INIT_LIST_HEAD(&new->list);
697 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
698 new->len = end - start;
699 list_add_tail(&new->list, head);
700 done:
701 SM_I(sbi)->nr_discards += end - start;
702 }
703
704 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
705 {
706 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
707 int max_blocks = sbi->blocks_per_seg;
708 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
709 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
710 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
711 unsigned long *discard_map = (unsigned long *)se->discard_map;
712 unsigned long *dmap = SIT_I(sbi)->tmp_map;
713 unsigned int start = 0, end = -1;
714 bool force = (cpc->reason == CP_DISCARD);
715 int i;
716
717 if (se->valid_blocks == max_blocks || !f2fs_discard_en(sbi))
718 return;
719
720 if (!force) {
721 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
722 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
723 return;
724 }
725
726 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
727 for (i = 0; i < entries; i++)
728 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
729 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
730
731 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
732 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
733 if (start >= max_blocks)
734 break;
735
736 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
737 if (force && start && end != max_blocks
738 && (end - start) < cpc->trim_minlen)
739 continue;
740
741 __add_discard_entry(sbi, cpc, se, start, end);
742 }
743 }
744
745 void release_discard_addrs(struct f2fs_sb_info *sbi)
746 {
747 struct list_head *head = &(SM_I(sbi)->discard_list);
748 struct discard_entry *entry, *this;
749
750 /* drop caches */
751 list_for_each_entry_safe(entry, this, head, list) {
752 list_del(&entry->list);
753 kmem_cache_free(discard_entry_slab, entry);
754 }
755 }
756
757 /*
758 * Should call clear_prefree_segments after checkpoint is done.
759 */
760 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
761 {
762 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
763 unsigned int segno;
764
765 mutex_lock(&dirty_i->seglist_lock);
766 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
767 __set_test_and_free(sbi, segno);
768 mutex_unlock(&dirty_i->seglist_lock);
769 }
770
771 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
772 {
773 struct list_head *head = &(SM_I(sbi)->discard_list);
774 struct discard_entry *entry, *this;
775 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
776 struct blk_plug plug;
777 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
778 unsigned int start = 0, end = -1;
779 unsigned int secno, start_segno;
780 bool force = (cpc->reason == CP_DISCARD);
781
782 blk_start_plug(&plug);
783
784 mutex_lock(&dirty_i->seglist_lock);
785
786 while (1) {
787 int i;
788 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
789 if (start >= MAIN_SEGS(sbi))
790 break;
791 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
792 start + 1);
793
794 for (i = start; i < end; i++)
795 clear_bit(i, prefree_map);
796
797 dirty_i->nr_dirty[PRE] -= end - start;
798
799 if (force || !test_opt(sbi, DISCARD))
800 continue;
801
802 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
803 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
804 (end - start) << sbi->log_blocks_per_seg);
805 continue;
806 }
807 next:
808 secno = GET_SECNO(sbi, start);
809 start_segno = secno * sbi->segs_per_sec;
810 if (!IS_CURSEC(sbi, secno) &&
811 !get_valid_blocks(sbi, start, sbi->segs_per_sec))
812 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
813 sbi->segs_per_sec << sbi->log_blocks_per_seg);
814
815 start = start_segno + sbi->segs_per_sec;
816 if (start < end)
817 goto next;
818 }
819 mutex_unlock(&dirty_i->seglist_lock);
820
821 /* send small discards */
822 list_for_each_entry_safe(entry, this, head, list) {
823 if (force && entry->len < cpc->trim_minlen)
824 goto skip;
825 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
826 cpc->trimmed += entry->len;
827 skip:
828 list_del(&entry->list);
829 SM_I(sbi)->nr_discards -= entry->len;
830 kmem_cache_free(discard_entry_slab, entry);
831 }
832
833 blk_finish_plug(&plug);
834 }
835
836 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
837 {
838 struct sit_info *sit_i = SIT_I(sbi);
839
840 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
841 sit_i->dirty_sentries++;
842 return false;
843 }
844
845 return true;
846 }
847
848 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
849 unsigned int segno, int modified)
850 {
851 struct seg_entry *se = get_seg_entry(sbi, segno);
852 se->type = type;
853 if (modified)
854 __mark_sit_entry_dirty(sbi, segno);
855 }
856
857 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
858 {
859 struct seg_entry *se;
860 unsigned int segno, offset;
861 long int new_vblocks;
862
863 segno = GET_SEGNO(sbi, blkaddr);
864
865 se = get_seg_entry(sbi, segno);
866 new_vblocks = se->valid_blocks + del;
867 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
868
869 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
870 (new_vblocks > sbi->blocks_per_seg)));
871
872 se->valid_blocks = new_vblocks;
873 se->mtime = get_mtime(sbi);
874 SIT_I(sbi)->max_mtime = se->mtime;
875
876 /* Update valid block bitmap */
877 if (del > 0) {
878 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
879 f2fs_bug_on(sbi, 1);
880 if (f2fs_discard_en(sbi) &&
881 !f2fs_test_and_set_bit(offset, se->discard_map))
882 sbi->discard_blks--;
883 } else {
884 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
885 f2fs_bug_on(sbi, 1);
886 if (f2fs_discard_en(sbi) &&
887 f2fs_test_and_clear_bit(offset, se->discard_map))
888 sbi->discard_blks++;
889 }
890 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
891 se->ckpt_valid_blocks += del;
892
893 __mark_sit_entry_dirty(sbi, segno);
894
895 /* update total number of valid blocks to be written in ckpt area */
896 SIT_I(sbi)->written_valid_blocks += del;
897
898 if (sbi->segs_per_sec > 1)
899 get_sec_entry(sbi, segno)->valid_blocks += del;
900 }
901
902 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
903 {
904 update_sit_entry(sbi, new, 1);
905 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
906 update_sit_entry(sbi, old, -1);
907
908 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
909 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
910 }
911
912 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
913 {
914 unsigned int segno = GET_SEGNO(sbi, addr);
915 struct sit_info *sit_i = SIT_I(sbi);
916
917 f2fs_bug_on(sbi, addr == NULL_ADDR);
918 if (addr == NEW_ADDR)
919 return;
920
921 /* add it into sit main buffer */
922 mutex_lock(&sit_i->sentry_lock);
923
924 update_sit_entry(sbi, addr, -1);
925
926 /* add it into dirty seglist */
927 locate_dirty_segment(sbi, segno);
928
929 mutex_unlock(&sit_i->sentry_lock);
930 }
931
932 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
933 {
934 struct sit_info *sit_i = SIT_I(sbi);
935 unsigned int segno, offset;
936 struct seg_entry *se;
937 bool is_cp = false;
938
939 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
940 return true;
941
942 mutex_lock(&sit_i->sentry_lock);
943
944 segno = GET_SEGNO(sbi, blkaddr);
945 se = get_seg_entry(sbi, segno);
946 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
947
948 if (f2fs_test_bit(offset, se->ckpt_valid_map))
949 is_cp = true;
950
951 mutex_unlock(&sit_i->sentry_lock);
952
953 return is_cp;
954 }
955
956 /*
957 * This function should be resided under the curseg_mutex lock
958 */
959 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
960 struct f2fs_summary *sum)
961 {
962 struct curseg_info *curseg = CURSEG_I(sbi, type);
963 void *addr = curseg->sum_blk;
964 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
965 memcpy(addr, sum, sizeof(struct f2fs_summary));
966 }
967
968 /*
969 * Calculate the number of current summary pages for writing
970 */
971 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
972 {
973 int valid_sum_count = 0;
974 int i, sum_in_page;
975
976 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
977 if (sbi->ckpt->alloc_type[i] == SSR)
978 valid_sum_count += sbi->blocks_per_seg;
979 else {
980 if (for_ra)
981 valid_sum_count += le16_to_cpu(
982 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
983 else
984 valid_sum_count += curseg_blkoff(sbi, i);
985 }
986 }
987
988 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
989 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
990 if (valid_sum_count <= sum_in_page)
991 return 1;
992 else if ((valid_sum_count - sum_in_page) <=
993 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
994 return 2;
995 return 3;
996 }
997
998 /*
999 * Caller should put this summary page
1000 */
1001 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
1002 {
1003 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
1004 }
1005
1006 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
1007 {
1008 struct page *page = grab_meta_page(sbi, blk_addr);
1009 void *dst = page_address(page);
1010
1011 if (src)
1012 memcpy(dst, src, PAGE_SIZE);
1013 else
1014 memset(dst, 0, PAGE_SIZE);
1015 set_page_dirty(page);
1016 f2fs_put_page(page, 1);
1017 }
1018
1019 static void write_sum_page(struct f2fs_sb_info *sbi,
1020 struct f2fs_summary_block *sum_blk, block_t blk_addr)
1021 {
1022 update_meta_page(sbi, (void *)sum_blk, blk_addr);
1023 }
1024
1025 static void write_current_sum_page(struct f2fs_sb_info *sbi,
1026 int type, block_t blk_addr)
1027 {
1028 struct curseg_info *curseg = CURSEG_I(sbi, type);
1029 struct page *page = grab_meta_page(sbi, blk_addr);
1030 struct f2fs_summary_block *src = curseg->sum_blk;
1031 struct f2fs_summary_block *dst;
1032
1033 dst = (struct f2fs_summary_block *)page_address(page);
1034
1035 mutex_lock(&curseg->curseg_mutex);
1036
1037 down_read(&curseg->journal_rwsem);
1038 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
1039 up_read(&curseg->journal_rwsem);
1040
1041 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
1042 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
1043
1044 mutex_unlock(&curseg->curseg_mutex);
1045
1046 set_page_dirty(page);
1047 f2fs_put_page(page, 1);
1048 }
1049
1050 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
1051 {
1052 struct curseg_info *curseg = CURSEG_I(sbi, type);
1053 unsigned int segno = curseg->segno + 1;
1054 struct free_segmap_info *free_i = FREE_I(sbi);
1055
1056 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
1057 return !test_bit(segno, free_i->free_segmap);
1058 return 0;
1059 }
1060
1061 /*
1062 * Find a new segment from the free segments bitmap to right order
1063 * This function should be returned with success, otherwise BUG
1064 */
1065 static void get_new_segment(struct f2fs_sb_info *sbi,
1066 unsigned int *newseg, bool new_sec, int dir)
1067 {
1068 struct free_segmap_info *free_i = FREE_I(sbi);
1069 unsigned int segno, secno, zoneno;
1070 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1071 unsigned int hint = *newseg / sbi->segs_per_sec;
1072 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1073 unsigned int left_start = hint;
1074 bool init = true;
1075 int go_left = 0;
1076 int i;
1077
1078 spin_lock(&free_i->segmap_lock);
1079
1080 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1081 segno = find_next_zero_bit(free_i->free_segmap,
1082 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1083 if (segno < (hint + 1) * sbi->segs_per_sec)
1084 goto got_it;
1085 }
1086 find_other_zone:
1087 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1088 if (secno >= MAIN_SECS(sbi)) {
1089 if (dir == ALLOC_RIGHT) {
1090 secno = find_next_zero_bit(free_i->free_secmap,
1091 MAIN_SECS(sbi), 0);
1092 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1093 } else {
1094 go_left = 1;
1095 left_start = hint - 1;
1096 }
1097 }
1098 if (go_left == 0)
1099 goto skip_left;
1100
1101 while (test_bit(left_start, free_i->free_secmap)) {
1102 if (left_start > 0) {
1103 left_start--;
1104 continue;
1105 }
1106 left_start = find_next_zero_bit(free_i->free_secmap,
1107 MAIN_SECS(sbi), 0);
1108 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1109 break;
1110 }
1111 secno = left_start;
1112 skip_left:
1113 hint = secno;
1114 segno = secno * sbi->segs_per_sec;
1115 zoneno = secno / sbi->secs_per_zone;
1116
1117 /* give up on finding another zone */
1118 if (!init)
1119 goto got_it;
1120 if (sbi->secs_per_zone == 1)
1121 goto got_it;
1122 if (zoneno == old_zoneno)
1123 goto got_it;
1124 if (dir == ALLOC_LEFT) {
1125 if (!go_left && zoneno + 1 >= total_zones)
1126 goto got_it;
1127 if (go_left && zoneno == 0)
1128 goto got_it;
1129 }
1130 for (i = 0; i < NR_CURSEG_TYPE; i++)
1131 if (CURSEG_I(sbi, i)->zone == zoneno)
1132 break;
1133
1134 if (i < NR_CURSEG_TYPE) {
1135 /* zone is in user, try another */
1136 if (go_left)
1137 hint = zoneno * sbi->secs_per_zone - 1;
1138 else if (zoneno + 1 >= total_zones)
1139 hint = 0;
1140 else
1141 hint = (zoneno + 1) * sbi->secs_per_zone;
1142 init = false;
1143 goto find_other_zone;
1144 }
1145 got_it:
1146 /* set it as dirty segment in free segmap */
1147 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1148 __set_inuse(sbi, segno);
1149 *newseg = segno;
1150 spin_unlock(&free_i->segmap_lock);
1151 }
1152
1153 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1154 {
1155 struct curseg_info *curseg = CURSEG_I(sbi, type);
1156 struct summary_footer *sum_footer;
1157
1158 curseg->segno = curseg->next_segno;
1159 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1160 curseg->next_blkoff = 0;
1161 curseg->next_segno = NULL_SEGNO;
1162
1163 sum_footer = &(curseg->sum_blk->footer);
1164 memset(sum_footer, 0, sizeof(struct summary_footer));
1165 if (IS_DATASEG(type))
1166 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1167 if (IS_NODESEG(type))
1168 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1169 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1170 }
1171
1172 /*
1173 * Allocate a current working segment.
1174 * This function always allocates a free segment in LFS manner.
1175 */
1176 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1177 {
1178 struct curseg_info *curseg = CURSEG_I(sbi, type);
1179 unsigned int segno = curseg->segno;
1180 int dir = ALLOC_LEFT;
1181
1182 write_sum_page(sbi, curseg->sum_blk,
1183 GET_SUM_BLOCK(sbi, segno));
1184 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1185 dir = ALLOC_RIGHT;
1186
1187 if (test_opt(sbi, NOHEAP))
1188 dir = ALLOC_RIGHT;
1189
1190 get_new_segment(sbi, &segno, new_sec, dir);
1191 curseg->next_segno = segno;
1192 reset_curseg(sbi, type, 1);
1193 curseg->alloc_type = LFS;
1194 }
1195
1196 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1197 struct curseg_info *seg, block_t start)
1198 {
1199 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1200 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1201 unsigned long *target_map = SIT_I(sbi)->tmp_map;
1202 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1203 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1204 int i, pos;
1205
1206 for (i = 0; i < entries; i++)
1207 target_map[i] = ckpt_map[i] | cur_map[i];
1208
1209 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1210
1211 seg->next_blkoff = pos;
1212 }
1213
1214 /*
1215 * If a segment is written by LFS manner, next block offset is just obtained
1216 * by increasing the current block offset. However, if a segment is written by
1217 * SSR manner, next block offset obtained by calling __next_free_blkoff
1218 */
1219 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1220 struct curseg_info *seg)
1221 {
1222 if (seg->alloc_type == SSR)
1223 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1224 else
1225 seg->next_blkoff++;
1226 }
1227
1228 /*
1229 * This function always allocates a used segment(from dirty seglist) by SSR
1230 * manner, so it should recover the existing segment information of valid blocks
1231 */
1232 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1233 {
1234 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1235 struct curseg_info *curseg = CURSEG_I(sbi, type);
1236 unsigned int new_segno = curseg->next_segno;
1237 struct f2fs_summary_block *sum_node;
1238 struct page *sum_page;
1239
1240 write_sum_page(sbi, curseg->sum_blk,
1241 GET_SUM_BLOCK(sbi, curseg->segno));
1242 __set_test_and_inuse(sbi, new_segno);
1243
1244 mutex_lock(&dirty_i->seglist_lock);
1245 __remove_dirty_segment(sbi, new_segno, PRE);
1246 __remove_dirty_segment(sbi, new_segno, DIRTY);
1247 mutex_unlock(&dirty_i->seglist_lock);
1248
1249 reset_curseg(sbi, type, 1);
1250 curseg->alloc_type = SSR;
1251 __next_free_blkoff(sbi, curseg, 0);
1252
1253 if (reuse) {
1254 sum_page = get_sum_page(sbi, new_segno);
1255 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1256 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1257 f2fs_put_page(sum_page, 1);
1258 }
1259 }
1260
1261 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1262 {
1263 struct curseg_info *curseg = CURSEG_I(sbi, type);
1264 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1265
1266 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0, 0))
1267 return v_ops->get_victim(sbi,
1268 &(curseg)->next_segno, BG_GC, type, SSR);
1269
1270 /* For data segments, let's do SSR more intensively */
1271 for (; type >= CURSEG_HOT_DATA; type--)
1272 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1273 BG_GC, type, SSR))
1274 return 1;
1275 return 0;
1276 }
1277
1278 /*
1279 * flush out current segment and replace it with new segment
1280 * This function should be returned with success, otherwise BUG
1281 */
1282 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1283 int type, bool force)
1284 {
1285 struct curseg_info *curseg = CURSEG_I(sbi, type);
1286
1287 if (force)
1288 new_curseg(sbi, type, true);
1289 else if (type == CURSEG_WARM_NODE)
1290 new_curseg(sbi, type, false);
1291 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1292 new_curseg(sbi, type, false);
1293 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1294 change_curseg(sbi, type, true);
1295 else
1296 new_curseg(sbi, type, false);
1297
1298 stat_inc_seg_type(sbi, curseg);
1299 }
1300
1301 static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1302 {
1303 struct curseg_info *curseg = CURSEG_I(sbi, type);
1304 unsigned int old_segno;
1305
1306 old_segno = curseg->segno;
1307 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1308 locate_dirty_segment(sbi, old_segno);
1309 }
1310
1311 void allocate_new_segments(struct f2fs_sb_info *sbi)
1312 {
1313 int i;
1314
1315 if (test_opt(sbi, LFS))
1316 return;
1317
1318 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1319 __allocate_new_segments(sbi, i);
1320 }
1321
1322 static const struct segment_allocation default_salloc_ops = {
1323 .allocate_segment = allocate_segment_by_default,
1324 };
1325
1326 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1327 {
1328 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1329 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1330 unsigned int start_segno, end_segno;
1331 struct cp_control cpc;
1332 int err = 0;
1333
1334 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1335 return -EINVAL;
1336
1337 cpc.trimmed = 0;
1338 if (end <= MAIN_BLKADDR(sbi))
1339 goto out;
1340
1341 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK)) {
1342 f2fs_msg(sbi->sb, KERN_WARNING,
1343 "Found FS corruption, run fsck to fix.");
1344 goto out;
1345 }
1346
1347 /* start/end segment number in main_area */
1348 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1349 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1350 GET_SEGNO(sbi, end);
1351 cpc.reason = CP_DISCARD;
1352 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1353
1354 /* do checkpoint to issue discard commands safely */
1355 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1356 cpc.trim_start = start_segno;
1357
1358 if (sbi->discard_blks == 0)
1359 break;
1360 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1361 cpc.trim_end = end_segno;
1362 else
1363 cpc.trim_end = min_t(unsigned int,
1364 rounddown(start_segno +
1365 BATCHED_TRIM_SEGMENTS(sbi),
1366 sbi->segs_per_sec) - 1, end_segno);
1367
1368 mutex_lock(&sbi->gc_mutex);
1369 err = write_checkpoint(sbi, &cpc);
1370 mutex_unlock(&sbi->gc_mutex);
1371 if (err)
1372 break;
1373
1374 schedule();
1375 }
1376 out:
1377 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1378 return err;
1379 }
1380
1381 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1382 {
1383 struct curseg_info *curseg = CURSEG_I(sbi, type);
1384 if (curseg->next_blkoff < sbi->blocks_per_seg)
1385 return true;
1386 return false;
1387 }
1388
1389 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1390 {
1391 if (p_type == DATA)
1392 return CURSEG_HOT_DATA;
1393 else
1394 return CURSEG_HOT_NODE;
1395 }
1396
1397 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1398 {
1399 if (p_type == DATA) {
1400 struct inode *inode = page->mapping->host;
1401
1402 if (S_ISDIR(inode->i_mode))
1403 return CURSEG_HOT_DATA;
1404 else
1405 return CURSEG_COLD_DATA;
1406 } else {
1407 if (IS_DNODE(page) && is_cold_node(page))
1408 return CURSEG_WARM_NODE;
1409 else
1410 return CURSEG_COLD_NODE;
1411 }
1412 }
1413
1414 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1415 {
1416 if (p_type == DATA) {
1417 struct inode *inode = page->mapping->host;
1418
1419 if (S_ISDIR(inode->i_mode))
1420 return CURSEG_HOT_DATA;
1421 else if (is_cold_data(page) || file_is_cold(inode))
1422 return CURSEG_COLD_DATA;
1423 else
1424 return CURSEG_WARM_DATA;
1425 } else {
1426 if (IS_DNODE(page))
1427 return is_cold_node(page) ? CURSEG_WARM_NODE :
1428 CURSEG_HOT_NODE;
1429 else
1430 return CURSEG_COLD_NODE;
1431 }
1432 }
1433
1434 static int __get_segment_type(struct page *page, enum page_type p_type)
1435 {
1436 switch (F2FS_P_SB(page)->active_logs) {
1437 case 2:
1438 return __get_segment_type_2(page, p_type);
1439 case 4:
1440 return __get_segment_type_4(page, p_type);
1441 }
1442 /* NR_CURSEG_TYPE(6) logs by default */
1443 f2fs_bug_on(F2FS_P_SB(page),
1444 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1445 return __get_segment_type_6(page, p_type);
1446 }
1447
1448 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1449 block_t old_blkaddr, block_t *new_blkaddr,
1450 struct f2fs_summary *sum, int type)
1451 {
1452 struct sit_info *sit_i = SIT_I(sbi);
1453 struct curseg_info *curseg;
1454 bool direct_io = (type == CURSEG_DIRECT_IO);
1455
1456 type = direct_io ? CURSEG_WARM_DATA : type;
1457
1458 curseg = CURSEG_I(sbi, type);
1459
1460 mutex_lock(&curseg->curseg_mutex);
1461 mutex_lock(&sit_i->sentry_lock);
1462
1463 /* direct_io'ed data is aligned to the segment for better performance */
1464 if (direct_io && curseg->next_blkoff &&
1465 !has_not_enough_free_secs(sbi, 0, 0))
1466 __allocate_new_segments(sbi, type);
1467
1468 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1469
1470 /*
1471 * __add_sum_entry should be resided under the curseg_mutex
1472 * because, this function updates a summary entry in the
1473 * current summary block.
1474 */
1475 __add_sum_entry(sbi, type, sum);
1476
1477 __refresh_next_blkoff(sbi, curseg);
1478
1479 stat_inc_block_count(sbi, curseg);
1480
1481 if (!__has_curseg_space(sbi, type))
1482 sit_i->s_ops->allocate_segment(sbi, type, false);
1483 /*
1484 * SIT information should be updated before segment allocation,
1485 * since SSR needs latest valid block information.
1486 */
1487 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1488
1489 mutex_unlock(&sit_i->sentry_lock);
1490
1491 if (page && IS_NODESEG(type))
1492 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1493
1494 mutex_unlock(&curseg->curseg_mutex);
1495 }
1496
1497 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1498 {
1499 int type = __get_segment_type(fio->page, fio->type);
1500
1501 if (fio->type == NODE || fio->type == DATA)
1502 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1503
1504 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1505 &fio->new_blkaddr, sum, type);
1506
1507 /* writeout dirty page into bdev */
1508 f2fs_submit_page_mbio(fio);
1509
1510 if (fio->type == NODE || fio->type == DATA)
1511 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
1512 }
1513
1514 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1515 {
1516 struct f2fs_io_info fio = {
1517 .sbi = sbi,
1518 .type = META,
1519 .op = REQ_OP_WRITE,
1520 .op_flags = WRITE_SYNC | REQ_META | REQ_PRIO,
1521 .old_blkaddr = page->index,
1522 .new_blkaddr = page->index,
1523 .page = page,
1524 .encrypted_page = NULL,
1525 };
1526
1527 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1528 fio.op_flags &= ~REQ_META;
1529
1530 set_page_writeback(page);
1531 f2fs_submit_page_mbio(&fio);
1532 }
1533
1534 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1535 {
1536 struct f2fs_summary sum;
1537
1538 set_summary(&sum, nid, 0, 0);
1539 do_write_page(&sum, fio);
1540 }
1541
1542 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1543 {
1544 struct f2fs_sb_info *sbi = fio->sbi;
1545 struct f2fs_summary sum;
1546 struct node_info ni;
1547
1548 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1549 get_node_info(sbi, dn->nid, &ni);
1550 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1551 do_write_page(&sum, fio);
1552 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1553 }
1554
1555 void rewrite_data_page(struct f2fs_io_info *fio)
1556 {
1557 fio->new_blkaddr = fio->old_blkaddr;
1558 stat_inc_inplace_blocks(fio->sbi);
1559 f2fs_submit_page_mbio(fio);
1560 }
1561
1562 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1563 block_t old_blkaddr, block_t new_blkaddr,
1564 bool recover_curseg, bool recover_newaddr)
1565 {
1566 struct sit_info *sit_i = SIT_I(sbi);
1567 struct curseg_info *curseg;
1568 unsigned int segno, old_cursegno;
1569 struct seg_entry *se;
1570 int type;
1571 unsigned short old_blkoff;
1572
1573 segno = GET_SEGNO(sbi, new_blkaddr);
1574 se = get_seg_entry(sbi, segno);
1575 type = se->type;
1576
1577 if (!recover_curseg) {
1578 /* for recovery flow */
1579 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1580 if (old_blkaddr == NULL_ADDR)
1581 type = CURSEG_COLD_DATA;
1582 else
1583 type = CURSEG_WARM_DATA;
1584 }
1585 } else {
1586 if (!IS_CURSEG(sbi, segno))
1587 type = CURSEG_WARM_DATA;
1588 }
1589
1590 curseg = CURSEG_I(sbi, type);
1591
1592 mutex_lock(&curseg->curseg_mutex);
1593 mutex_lock(&sit_i->sentry_lock);
1594
1595 old_cursegno = curseg->segno;
1596 old_blkoff = curseg->next_blkoff;
1597
1598 /* change the current segment */
1599 if (segno != curseg->segno) {
1600 curseg->next_segno = segno;
1601 change_curseg(sbi, type, true);
1602 }
1603
1604 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1605 __add_sum_entry(sbi, type, sum);
1606
1607 if (!recover_curseg || recover_newaddr)
1608 update_sit_entry(sbi, new_blkaddr, 1);
1609 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1610 update_sit_entry(sbi, old_blkaddr, -1);
1611
1612 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1613 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1614
1615 locate_dirty_segment(sbi, old_cursegno);
1616
1617 if (recover_curseg) {
1618 if (old_cursegno != curseg->segno) {
1619 curseg->next_segno = old_cursegno;
1620 change_curseg(sbi, type, true);
1621 }
1622 curseg->next_blkoff = old_blkoff;
1623 }
1624
1625 mutex_unlock(&sit_i->sentry_lock);
1626 mutex_unlock(&curseg->curseg_mutex);
1627 }
1628
1629 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1630 block_t old_addr, block_t new_addr,
1631 unsigned char version, bool recover_curseg,
1632 bool recover_newaddr)
1633 {
1634 struct f2fs_summary sum;
1635
1636 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1637
1638 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1639 recover_curseg, recover_newaddr);
1640
1641 f2fs_update_data_blkaddr(dn, new_addr);
1642 }
1643
1644 void f2fs_wait_on_page_writeback(struct page *page,
1645 enum page_type type, bool ordered)
1646 {
1647 if (PageWriteback(page)) {
1648 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1649
1650 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
1651 if (ordered)
1652 wait_on_page_writeback(page);
1653 else
1654 wait_for_stable_page(page);
1655 }
1656 }
1657
1658 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1659 block_t blkaddr)
1660 {
1661 struct page *cpage;
1662
1663 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
1664 return;
1665
1666 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1667 if (cpage) {
1668 f2fs_wait_on_page_writeback(cpage, DATA, true);
1669 f2fs_put_page(cpage, 1);
1670 }
1671 }
1672
1673 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1674 {
1675 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1676 struct curseg_info *seg_i;
1677 unsigned char *kaddr;
1678 struct page *page;
1679 block_t start;
1680 int i, j, offset;
1681
1682 start = start_sum_block(sbi);
1683
1684 page = get_meta_page(sbi, start++);
1685 kaddr = (unsigned char *)page_address(page);
1686
1687 /* Step 1: restore nat cache */
1688 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1689 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
1690
1691 /* Step 2: restore sit cache */
1692 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1693 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
1694 offset = 2 * SUM_JOURNAL_SIZE;
1695
1696 /* Step 3: restore summary entries */
1697 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1698 unsigned short blk_off;
1699 unsigned int segno;
1700
1701 seg_i = CURSEG_I(sbi, i);
1702 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1703 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1704 seg_i->next_segno = segno;
1705 reset_curseg(sbi, i, 0);
1706 seg_i->alloc_type = ckpt->alloc_type[i];
1707 seg_i->next_blkoff = blk_off;
1708
1709 if (seg_i->alloc_type == SSR)
1710 blk_off = sbi->blocks_per_seg;
1711
1712 for (j = 0; j < blk_off; j++) {
1713 struct f2fs_summary *s;
1714 s = (struct f2fs_summary *)(kaddr + offset);
1715 seg_i->sum_blk->entries[j] = *s;
1716 offset += SUMMARY_SIZE;
1717 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
1718 SUM_FOOTER_SIZE)
1719 continue;
1720
1721 f2fs_put_page(page, 1);
1722 page = NULL;
1723
1724 page = get_meta_page(sbi, start++);
1725 kaddr = (unsigned char *)page_address(page);
1726 offset = 0;
1727 }
1728 }
1729 f2fs_put_page(page, 1);
1730 return 0;
1731 }
1732
1733 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1734 {
1735 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1736 struct f2fs_summary_block *sum;
1737 struct curseg_info *curseg;
1738 struct page *new;
1739 unsigned short blk_off;
1740 unsigned int segno = 0;
1741 block_t blk_addr = 0;
1742
1743 /* get segment number and block addr */
1744 if (IS_DATASEG(type)) {
1745 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1746 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1747 CURSEG_HOT_DATA]);
1748 if (__exist_node_summaries(sbi))
1749 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1750 else
1751 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1752 } else {
1753 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1754 CURSEG_HOT_NODE]);
1755 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1756 CURSEG_HOT_NODE]);
1757 if (__exist_node_summaries(sbi))
1758 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1759 type - CURSEG_HOT_NODE);
1760 else
1761 blk_addr = GET_SUM_BLOCK(sbi, segno);
1762 }
1763
1764 new = get_meta_page(sbi, blk_addr);
1765 sum = (struct f2fs_summary_block *)page_address(new);
1766
1767 if (IS_NODESEG(type)) {
1768 if (__exist_node_summaries(sbi)) {
1769 struct f2fs_summary *ns = &sum->entries[0];
1770 int i;
1771 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1772 ns->version = 0;
1773 ns->ofs_in_node = 0;
1774 }
1775 } else {
1776 int err;
1777
1778 err = restore_node_summary(sbi, segno, sum);
1779 if (err) {
1780 f2fs_put_page(new, 1);
1781 return err;
1782 }
1783 }
1784 }
1785
1786 /* set uncompleted segment to curseg */
1787 curseg = CURSEG_I(sbi, type);
1788 mutex_lock(&curseg->curseg_mutex);
1789
1790 /* update journal info */
1791 down_write(&curseg->journal_rwsem);
1792 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1793 up_write(&curseg->journal_rwsem);
1794
1795 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1796 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
1797 curseg->next_segno = segno;
1798 reset_curseg(sbi, type, 0);
1799 curseg->alloc_type = ckpt->alloc_type[type];
1800 curseg->next_blkoff = blk_off;
1801 mutex_unlock(&curseg->curseg_mutex);
1802 f2fs_put_page(new, 1);
1803 return 0;
1804 }
1805
1806 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1807 {
1808 int type = CURSEG_HOT_DATA;
1809 int err;
1810
1811 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG)) {
1812 int npages = npages_for_summary_flush(sbi, true);
1813
1814 if (npages >= 2)
1815 ra_meta_pages(sbi, start_sum_block(sbi), npages,
1816 META_CP, true);
1817
1818 /* restore for compacted data summary */
1819 if (read_compacted_summaries(sbi))
1820 return -EINVAL;
1821 type = CURSEG_HOT_NODE;
1822 }
1823
1824 if (__exist_node_summaries(sbi))
1825 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1826 NR_CURSEG_TYPE - type, META_CP, true);
1827
1828 for (; type <= CURSEG_COLD_NODE; type++) {
1829 err = read_normal_summaries(sbi, type);
1830 if (err)
1831 return err;
1832 }
1833
1834 return 0;
1835 }
1836
1837 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1838 {
1839 struct page *page;
1840 unsigned char *kaddr;
1841 struct f2fs_summary *summary;
1842 struct curseg_info *seg_i;
1843 int written_size = 0;
1844 int i, j;
1845
1846 page = grab_meta_page(sbi, blkaddr++);
1847 kaddr = (unsigned char *)page_address(page);
1848
1849 /* Step 1: write nat cache */
1850 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1851 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
1852 written_size += SUM_JOURNAL_SIZE;
1853
1854 /* Step 2: write sit cache */
1855 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1856 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
1857 written_size += SUM_JOURNAL_SIZE;
1858
1859 /* Step 3: write summary entries */
1860 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1861 unsigned short blkoff;
1862 seg_i = CURSEG_I(sbi, i);
1863 if (sbi->ckpt->alloc_type[i] == SSR)
1864 blkoff = sbi->blocks_per_seg;
1865 else
1866 blkoff = curseg_blkoff(sbi, i);
1867
1868 for (j = 0; j < blkoff; j++) {
1869 if (!page) {
1870 page = grab_meta_page(sbi, blkaddr++);
1871 kaddr = (unsigned char *)page_address(page);
1872 written_size = 0;
1873 }
1874 summary = (struct f2fs_summary *)(kaddr + written_size);
1875 *summary = seg_i->sum_blk->entries[j];
1876 written_size += SUMMARY_SIZE;
1877
1878 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
1879 SUM_FOOTER_SIZE)
1880 continue;
1881
1882 set_page_dirty(page);
1883 f2fs_put_page(page, 1);
1884 page = NULL;
1885 }
1886 }
1887 if (page) {
1888 set_page_dirty(page);
1889 f2fs_put_page(page, 1);
1890 }
1891 }
1892
1893 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1894 block_t blkaddr, int type)
1895 {
1896 int i, end;
1897 if (IS_DATASEG(type))
1898 end = type + NR_CURSEG_DATA_TYPE;
1899 else
1900 end = type + NR_CURSEG_NODE_TYPE;
1901
1902 for (i = type; i < end; i++)
1903 write_current_sum_page(sbi, i, blkaddr + (i - type));
1904 }
1905
1906 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1907 {
1908 if (is_set_ckpt_flags(sbi, CP_COMPACT_SUM_FLAG))
1909 write_compacted_summaries(sbi, start_blk);
1910 else
1911 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1912 }
1913
1914 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1915 {
1916 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1917 }
1918
1919 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
1920 unsigned int val, int alloc)
1921 {
1922 int i;
1923
1924 if (type == NAT_JOURNAL) {
1925 for (i = 0; i < nats_in_cursum(journal); i++) {
1926 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
1927 return i;
1928 }
1929 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1930 return update_nats_in_cursum(journal, 1);
1931 } else if (type == SIT_JOURNAL) {
1932 for (i = 0; i < sits_in_cursum(journal); i++)
1933 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
1934 return i;
1935 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1936 return update_sits_in_cursum(journal, 1);
1937 }
1938 return -1;
1939 }
1940
1941 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1942 unsigned int segno)
1943 {
1944 return get_meta_page(sbi, current_sit_addr(sbi, segno));
1945 }
1946
1947 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1948 unsigned int start)
1949 {
1950 struct sit_info *sit_i = SIT_I(sbi);
1951 struct page *src_page, *dst_page;
1952 pgoff_t src_off, dst_off;
1953 void *src_addr, *dst_addr;
1954
1955 src_off = current_sit_addr(sbi, start);
1956 dst_off = next_sit_addr(sbi, src_off);
1957
1958 /* get current sit block page without lock */
1959 src_page = get_meta_page(sbi, src_off);
1960 dst_page = grab_meta_page(sbi, dst_off);
1961 f2fs_bug_on(sbi, PageDirty(src_page));
1962
1963 src_addr = page_address(src_page);
1964 dst_addr = page_address(dst_page);
1965 memcpy(dst_addr, src_addr, PAGE_SIZE);
1966
1967 set_page_dirty(dst_page);
1968 f2fs_put_page(src_page, 1);
1969
1970 set_to_next_sit(sit_i, start);
1971
1972 return dst_page;
1973 }
1974
1975 static struct sit_entry_set *grab_sit_entry_set(void)
1976 {
1977 struct sit_entry_set *ses =
1978 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
1979
1980 ses->entry_cnt = 0;
1981 INIT_LIST_HEAD(&ses->set_list);
1982 return ses;
1983 }
1984
1985 static void release_sit_entry_set(struct sit_entry_set *ses)
1986 {
1987 list_del(&ses->set_list);
1988 kmem_cache_free(sit_entry_set_slab, ses);
1989 }
1990
1991 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1992 struct list_head *head)
1993 {
1994 struct sit_entry_set *next = ses;
1995
1996 if (list_is_last(&ses->set_list, head))
1997 return;
1998
1999 list_for_each_entry_continue(next, head, set_list)
2000 if (ses->entry_cnt <= next->entry_cnt)
2001 break;
2002
2003 list_move_tail(&ses->set_list, &next->set_list);
2004 }
2005
2006 static void add_sit_entry(unsigned int segno, struct list_head *head)
2007 {
2008 struct sit_entry_set *ses;
2009 unsigned int start_segno = START_SEGNO(segno);
2010
2011 list_for_each_entry(ses, head, set_list) {
2012 if (ses->start_segno == start_segno) {
2013 ses->entry_cnt++;
2014 adjust_sit_entry_set(ses, head);
2015 return;
2016 }
2017 }
2018
2019 ses = grab_sit_entry_set();
2020
2021 ses->start_segno = start_segno;
2022 ses->entry_cnt++;
2023 list_add(&ses->set_list, head);
2024 }
2025
2026 static void add_sits_in_set(struct f2fs_sb_info *sbi)
2027 {
2028 struct f2fs_sm_info *sm_info = SM_I(sbi);
2029 struct list_head *set_list = &sm_info->sit_entry_set;
2030 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
2031 unsigned int segno;
2032
2033 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
2034 add_sit_entry(segno, set_list);
2035 }
2036
2037 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
2038 {
2039 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2040 struct f2fs_journal *journal = curseg->journal;
2041 int i;
2042
2043 down_write(&curseg->journal_rwsem);
2044 for (i = 0; i < sits_in_cursum(journal); i++) {
2045 unsigned int segno;
2046 bool dirtied;
2047
2048 segno = le32_to_cpu(segno_in_journal(journal, i));
2049 dirtied = __mark_sit_entry_dirty(sbi, segno);
2050
2051 if (!dirtied)
2052 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
2053 }
2054 update_sits_in_cursum(journal, -i);
2055 up_write(&curseg->journal_rwsem);
2056 }
2057
2058 /*
2059 * CP calls this function, which flushes SIT entries including sit_journal,
2060 * and moves prefree segs to free segs.
2061 */
2062 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2063 {
2064 struct sit_info *sit_i = SIT_I(sbi);
2065 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
2066 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2067 struct f2fs_journal *journal = curseg->journal;
2068 struct sit_entry_set *ses, *tmp;
2069 struct list_head *head = &SM_I(sbi)->sit_entry_set;
2070 bool to_journal = true;
2071 struct seg_entry *se;
2072
2073 mutex_lock(&sit_i->sentry_lock);
2074
2075 if (!sit_i->dirty_sentries)
2076 goto out;
2077
2078 /*
2079 * add and account sit entries of dirty bitmap in sit entry
2080 * set temporarily
2081 */
2082 add_sits_in_set(sbi);
2083
2084 /*
2085 * if there are no enough space in journal to store dirty sit
2086 * entries, remove all entries from journal and add and account
2087 * them in sit entry set.
2088 */
2089 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2090 remove_sits_in_journal(sbi);
2091
2092 /*
2093 * there are two steps to flush sit entries:
2094 * #1, flush sit entries to journal in current cold data summary block.
2095 * #2, flush sit entries to sit page.
2096 */
2097 list_for_each_entry_safe(ses, tmp, head, set_list) {
2098 struct page *page = NULL;
2099 struct f2fs_sit_block *raw_sit = NULL;
2100 unsigned int start_segno = ses->start_segno;
2101 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2102 (unsigned long)MAIN_SEGS(sbi));
2103 unsigned int segno = start_segno;
2104
2105 if (to_journal &&
2106 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2107 to_journal = false;
2108
2109 if (to_journal) {
2110 down_write(&curseg->journal_rwsem);
2111 } else {
2112 page = get_next_sit_page(sbi, start_segno);
2113 raw_sit = page_address(page);
2114 }
2115
2116 /* flush dirty sit entries in region of current sit set */
2117 for_each_set_bit_from(segno, bitmap, end) {
2118 int offset, sit_offset;
2119
2120 se = get_seg_entry(sbi, segno);
2121
2122 /* add discard candidates */
2123 if (cpc->reason != CP_DISCARD) {
2124 cpc->trim_start = segno;
2125 add_discard_addrs(sbi, cpc);
2126 }
2127
2128 if (to_journal) {
2129 offset = lookup_journal_in_cursum(journal,
2130 SIT_JOURNAL, segno, 1);
2131 f2fs_bug_on(sbi, offset < 0);
2132 segno_in_journal(journal, offset) =
2133 cpu_to_le32(segno);
2134 seg_info_to_raw_sit(se,
2135 &sit_in_journal(journal, offset));
2136 } else {
2137 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2138 seg_info_to_raw_sit(se,
2139 &raw_sit->entries[sit_offset]);
2140 }
2141
2142 __clear_bit(segno, bitmap);
2143 sit_i->dirty_sentries--;
2144 ses->entry_cnt--;
2145 }
2146
2147 if (to_journal)
2148 up_write(&curseg->journal_rwsem);
2149 else
2150 f2fs_put_page(page, 1);
2151
2152 f2fs_bug_on(sbi, ses->entry_cnt);
2153 release_sit_entry_set(ses);
2154 }
2155
2156 f2fs_bug_on(sbi, !list_empty(head));
2157 f2fs_bug_on(sbi, sit_i->dirty_sentries);
2158 out:
2159 if (cpc->reason == CP_DISCARD) {
2160 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2161 add_discard_addrs(sbi, cpc);
2162 }
2163 mutex_unlock(&sit_i->sentry_lock);
2164
2165 set_prefree_as_free_segments(sbi);
2166 }
2167
2168 static int build_sit_info(struct f2fs_sb_info *sbi)
2169 {
2170 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2171 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2172 struct sit_info *sit_i;
2173 unsigned int sit_segs, start;
2174 char *src_bitmap, *dst_bitmap;
2175 unsigned int bitmap_size;
2176
2177 /* allocate memory for SIT information */
2178 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2179 if (!sit_i)
2180 return -ENOMEM;
2181
2182 SM_I(sbi)->sit_info = sit_i;
2183
2184 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2185 sizeof(struct seg_entry), GFP_KERNEL);
2186 if (!sit_i->sentries)
2187 return -ENOMEM;
2188
2189 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2190 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2191 if (!sit_i->dirty_sentries_bitmap)
2192 return -ENOMEM;
2193
2194 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2195 sit_i->sentries[start].cur_valid_map
2196 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2197 sit_i->sentries[start].ckpt_valid_map
2198 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2199 if (!sit_i->sentries[start].cur_valid_map ||
2200 !sit_i->sentries[start].ckpt_valid_map)
2201 return -ENOMEM;
2202
2203 if (f2fs_discard_en(sbi)) {
2204 sit_i->sentries[start].discard_map
2205 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2206 if (!sit_i->sentries[start].discard_map)
2207 return -ENOMEM;
2208 }
2209 }
2210
2211 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2212 if (!sit_i->tmp_map)
2213 return -ENOMEM;
2214
2215 if (sbi->segs_per_sec > 1) {
2216 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2217 sizeof(struct sec_entry), GFP_KERNEL);
2218 if (!sit_i->sec_entries)
2219 return -ENOMEM;
2220 }
2221
2222 /* get information related with SIT */
2223 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2224
2225 /* setup SIT bitmap from ckeckpoint pack */
2226 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2227 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2228
2229 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2230 if (!dst_bitmap)
2231 return -ENOMEM;
2232
2233 /* init SIT information */
2234 sit_i->s_ops = &default_salloc_ops;
2235
2236 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2237 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2238 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2239 sit_i->sit_bitmap = dst_bitmap;
2240 sit_i->bitmap_size = bitmap_size;
2241 sit_i->dirty_sentries = 0;
2242 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2243 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2244 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2245 mutex_init(&sit_i->sentry_lock);
2246 return 0;
2247 }
2248
2249 static int build_free_segmap(struct f2fs_sb_info *sbi)
2250 {
2251 struct free_segmap_info *free_i;
2252 unsigned int bitmap_size, sec_bitmap_size;
2253
2254 /* allocate memory for free segmap information */
2255 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2256 if (!free_i)
2257 return -ENOMEM;
2258
2259 SM_I(sbi)->free_info = free_i;
2260
2261 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2262 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2263 if (!free_i->free_segmap)
2264 return -ENOMEM;
2265
2266 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2267 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2268 if (!free_i->free_secmap)
2269 return -ENOMEM;
2270
2271 /* set all segments as dirty temporarily */
2272 memset(free_i->free_segmap, 0xff, bitmap_size);
2273 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2274
2275 /* init free segmap information */
2276 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2277 free_i->free_segments = 0;
2278 free_i->free_sections = 0;
2279 spin_lock_init(&free_i->segmap_lock);
2280 return 0;
2281 }
2282
2283 static int build_curseg(struct f2fs_sb_info *sbi)
2284 {
2285 struct curseg_info *array;
2286 int i;
2287
2288 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2289 if (!array)
2290 return -ENOMEM;
2291
2292 SM_I(sbi)->curseg_array = array;
2293
2294 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2295 mutex_init(&array[i].curseg_mutex);
2296 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2297 if (!array[i].sum_blk)
2298 return -ENOMEM;
2299 init_rwsem(&array[i].journal_rwsem);
2300 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2301 GFP_KERNEL);
2302 if (!array[i].journal)
2303 return -ENOMEM;
2304 array[i].segno = NULL_SEGNO;
2305 array[i].next_blkoff = 0;
2306 }
2307 return restore_curseg_summaries(sbi);
2308 }
2309
2310 static void build_sit_entries(struct f2fs_sb_info *sbi)
2311 {
2312 struct sit_info *sit_i = SIT_I(sbi);
2313 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2314 struct f2fs_journal *journal = curseg->journal;
2315 struct seg_entry *se;
2316 struct f2fs_sit_entry sit;
2317 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2318 unsigned int i, start, end;
2319 unsigned int readed, start_blk = 0;
2320 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
2321
2322 do {
2323 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
2324
2325 start = start_blk * sit_i->sents_per_block;
2326 end = (start_blk + readed) * sit_i->sents_per_block;
2327
2328 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2329 struct f2fs_sit_block *sit_blk;
2330 struct page *page;
2331
2332 se = &sit_i->sentries[start];
2333 page = get_current_sit_page(sbi, start);
2334 sit_blk = (struct f2fs_sit_block *)page_address(page);
2335 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2336 f2fs_put_page(page, 1);
2337
2338 check_block_count(sbi, start, &sit);
2339 seg_info_from_raw_sit(se, &sit);
2340
2341 /* build discard map only one time */
2342 if (f2fs_discard_en(sbi)) {
2343 memcpy(se->discard_map, se->cur_valid_map,
2344 SIT_VBLOCK_MAP_SIZE);
2345 sbi->discard_blks += sbi->blocks_per_seg -
2346 se->valid_blocks;
2347 }
2348
2349 if (sbi->segs_per_sec > 1)
2350 get_sec_entry(sbi, start)->valid_blocks +=
2351 se->valid_blocks;
2352 }
2353 start_blk += readed;
2354 } while (start_blk < sit_blk_cnt);
2355
2356 down_read(&curseg->journal_rwsem);
2357 for (i = 0; i < sits_in_cursum(journal); i++) {
2358 unsigned int old_valid_blocks;
2359
2360 start = le32_to_cpu(segno_in_journal(journal, i));
2361 se = &sit_i->sentries[start];
2362 sit = sit_in_journal(journal, i);
2363
2364 old_valid_blocks = se->valid_blocks;
2365
2366 check_block_count(sbi, start, &sit);
2367 seg_info_from_raw_sit(se, &sit);
2368
2369 if (f2fs_discard_en(sbi)) {
2370 memcpy(se->discard_map, se->cur_valid_map,
2371 SIT_VBLOCK_MAP_SIZE);
2372 sbi->discard_blks += old_valid_blocks -
2373 se->valid_blocks;
2374 }
2375
2376 if (sbi->segs_per_sec > 1)
2377 get_sec_entry(sbi, start)->valid_blocks +=
2378 se->valid_blocks - old_valid_blocks;
2379 }
2380 up_read(&curseg->journal_rwsem);
2381 }
2382
2383 static void init_free_segmap(struct f2fs_sb_info *sbi)
2384 {
2385 unsigned int start;
2386 int type;
2387
2388 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2389 struct seg_entry *sentry = get_seg_entry(sbi, start);
2390 if (!sentry->valid_blocks)
2391 __set_free(sbi, start);
2392 }
2393
2394 /* set use the current segments */
2395 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2396 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2397 __set_test_and_inuse(sbi, curseg_t->segno);
2398 }
2399 }
2400
2401 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2402 {
2403 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2404 struct free_segmap_info *free_i = FREE_I(sbi);
2405 unsigned int segno = 0, offset = 0;
2406 unsigned short valid_blocks;
2407
2408 while (1) {
2409 /* find dirty segment based on free segmap */
2410 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2411 if (segno >= MAIN_SEGS(sbi))
2412 break;
2413 offset = segno + 1;
2414 valid_blocks = get_valid_blocks(sbi, segno, 0);
2415 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2416 continue;
2417 if (valid_blocks > sbi->blocks_per_seg) {
2418 f2fs_bug_on(sbi, 1);
2419 continue;
2420 }
2421 mutex_lock(&dirty_i->seglist_lock);
2422 __locate_dirty_segment(sbi, segno, DIRTY);
2423 mutex_unlock(&dirty_i->seglist_lock);
2424 }
2425 }
2426
2427 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2428 {
2429 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2430 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2431
2432 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2433 if (!dirty_i->victim_secmap)
2434 return -ENOMEM;
2435 return 0;
2436 }
2437
2438 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2439 {
2440 struct dirty_seglist_info *dirty_i;
2441 unsigned int bitmap_size, i;
2442
2443 /* allocate memory for dirty segments list information */
2444 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2445 if (!dirty_i)
2446 return -ENOMEM;
2447
2448 SM_I(sbi)->dirty_info = dirty_i;
2449 mutex_init(&dirty_i->seglist_lock);
2450
2451 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2452
2453 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2454 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2455 if (!dirty_i->dirty_segmap[i])
2456 return -ENOMEM;
2457 }
2458
2459 init_dirty_segmap(sbi);
2460 return init_victim_secmap(sbi);
2461 }
2462
2463 /*
2464 * Update min, max modified time for cost-benefit GC algorithm
2465 */
2466 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2467 {
2468 struct sit_info *sit_i = SIT_I(sbi);
2469 unsigned int segno;
2470
2471 mutex_lock(&sit_i->sentry_lock);
2472
2473 sit_i->min_mtime = LLONG_MAX;
2474
2475 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2476 unsigned int i;
2477 unsigned long long mtime = 0;
2478
2479 for (i = 0; i < sbi->segs_per_sec; i++)
2480 mtime += get_seg_entry(sbi, segno + i)->mtime;
2481
2482 mtime = div_u64(mtime, sbi->segs_per_sec);
2483
2484 if (sit_i->min_mtime > mtime)
2485 sit_i->min_mtime = mtime;
2486 }
2487 sit_i->max_mtime = get_mtime(sbi);
2488 mutex_unlock(&sit_i->sentry_lock);
2489 }
2490
2491 int build_segment_manager(struct f2fs_sb_info *sbi)
2492 {
2493 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2494 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2495 struct f2fs_sm_info *sm_info;
2496 int err;
2497
2498 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2499 if (!sm_info)
2500 return -ENOMEM;
2501
2502 /* init sm info */
2503 sbi->sm_info = sm_info;
2504 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2505 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2506 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2507 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2508 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2509 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2510 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2511 sm_info->rec_prefree_segments = sm_info->main_segments *
2512 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2513 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2514 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2515
2516 if (!test_opt(sbi, LFS))
2517 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2518 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2519 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2520
2521 INIT_LIST_HEAD(&sm_info->discard_list);
2522 INIT_LIST_HEAD(&sm_info->wait_list);
2523 sm_info->nr_discards = 0;
2524 sm_info->max_discards = 0;
2525
2526 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2527
2528 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2529
2530 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2531 err = create_flush_cmd_control(sbi);
2532 if (err)
2533 return err;
2534 }
2535
2536 err = build_sit_info(sbi);
2537 if (err)
2538 return err;
2539 err = build_free_segmap(sbi);
2540 if (err)
2541 return err;
2542 err = build_curseg(sbi);
2543 if (err)
2544 return err;
2545
2546 /* reinit free segmap based on SIT */
2547 build_sit_entries(sbi);
2548
2549 init_free_segmap(sbi);
2550 err = build_dirty_segmap(sbi);
2551 if (err)
2552 return err;
2553
2554 init_min_max_mtime(sbi);
2555 return 0;
2556 }
2557
2558 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2559 enum dirty_type dirty_type)
2560 {
2561 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2562
2563 mutex_lock(&dirty_i->seglist_lock);
2564 kvfree(dirty_i->dirty_segmap[dirty_type]);
2565 dirty_i->nr_dirty[dirty_type] = 0;
2566 mutex_unlock(&dirty_i->seglist_lock);
2567 }
2568
2569 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2570 {
2571 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2572 kvfree(dirty_i->victim_secmap);
2573 }
2574
2575 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2576 {
2577 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2578 int i;
2579
2580 if (!dirty_i)
2581 return;
2582
2583 /* discard pre-free/dirty segments list */
2584 for (i = 0; i < NR_DIRTY_TYPE; i++)
2585 discard_dirty_segmap(sbi, i);
2586
2587 destroy_victim_secmap(sbi);
2588 SM_I(sbi)->dirty_info = NULL;
2589 kfree(dirty_i);
2590 }
2591
2592 static void destroy_curseg(struct f2fs_sb_info *sbi)
2593 {
2594 struct curseg_info *array = SM_I(sbi)->curseg_array;
2595 int i;
2596
2597 if (!array)
2598 return;
2599 SM_I(sbi)->curseg_array = NULL;
2600 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2601 kfree(array[i].sum_blk);
2602 kfree(array[i].journal);
2603 }
2604 kfree(array);
2605 }
2606
2607 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2608 {
2609 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2610 if (!free_i)
2611 return;
2612 SM_I(sbi)->free_info = NULL;
2613 kvfree(free_i->free_segmap);
2614 kvfree(free_i->free_secmap);
2615 kfree(free_i);
2616 }
2617
2618 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2619 {
2620 struct sit_info *sit_i = SIT_I(sbi);
2621 unsigned int start;
2622
2623 if (!sit_i)
2624 return;
2625
2626 if (sit_i->sentries) {
2627 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2628 kfree(sit_i->sentries[start].cur_valid_map);
2629 kfree(sit_i->sentries[start].ckpt_valid_map);
2630 kfree(sit_i->sentries[start].discard_map);
2631 }
2632 }
2633 kfree(sit_i->tmp_map);
2634
2635 kvfree(sit_i->sentries);
2636 kvfree(sit_i->sec_entries);
2637 kvfree(sit_i->dirty_sentries_bitmap);
2638
2639 SM_I(sbi)->sit_info = NULL;
2640 kfree(sit_i->sit_bitmap);
2641 kfree(sit_i);
2642 }
2643
2644 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2645 {
2646 struct f2fs_sm_info *sm_info = SM_I(sbi);
2647
2648 if (!sm_info)
2649 return;
2650 destroy_flush_cmd_control(sbi);
2651 destroy_dirty_segmap(sbi);
2652 destroy_curseg(sbi);
2653 destroy_free_segmap(sbi);
2654 destroy_sit_info(sbi);
2655 sbi->sm_info = NULL;
2656 kfree(sm_info);
2657 }
2658
2659 int __init create_segment_manager_caches(void)
2660 {
2661 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2662 sizeof(struct discard_entry));
2663 if (!discard_entry_slab)
2664 goto fail;
2665
2666 bio_entry_slab = f2fs_kmem_cache_create("bio_entry",
2667 sizeof(struct bio_entry));
2668 if (!bio_entry_slab)
2669 goto destroy_discard_entry;
2670
2671 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2672 sizeof(struct sit_entry_set));
2673 if (!sit_entry_set_slab)
2674 goto destroy_bio_entry;
2675
2676 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2677 sizeof(struct inmem_pages));
2678 if (!inmem_entry_slab)
2679 goto destroy_sit_entry_set;
2680 return 0;
2681
2682 destroy_sit_entry_set:
2683 kmem_cache_destroy(sit_entry_set_slab);
2684 destroy_bio_entry:
2685 kmem_cache_destroy(bio_entry_slab);
2686 destroy_discard_entry:
2687 kmem_cache_destroy(discard_entry_slab);
2688 fail:
2689 return -ENOMEM;
2690 }
2691
2692 void destroy_segment_manager_caches(void)
2693 {
2694 kmem_cache_destroy(sit_entry_set_slab);
2695 kmem_cache_destroy(bio_entry_slab);
2696 kmem_cache_destroy(discard_entry_slab);
2697 kmem_cache_destroy(inmem_entry_slab);
2698 }