Linux 4.8.3
[linux/fpc-iii.git] / fs / f2fs / segment.c
bloba46296f57b026e650f2a82950ae7f63a02fd2fb0
1 /*
2 * fs/f2fs/segment.c
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
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.
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>
20 #include "f2fs.h"
21 #include "segment.h"
22 #include "node.h"
23 #include "trace.h"
24 #include <trace/events/f2fs.h>
26 #define __reverse_ffz(x) __reverse_ffs(~(x))
28 static struct kmem_cache *discard_entry_slab;
29 static struct kmem_cache *sit_entry_set_slab;
30 static struct kmem_cache *inmem_entry_slab;
32 static unsigned long __reverse_ulong(unsigned char *str)
34 unsigned long tmp = 0;
35 int shift = 24, idx = 0;
37 #if BITS_PER_LONG == 64
38 shift = 56;
39 #endif
40 while (shift >= 0) {
41 tmp |= (unsigned long)str[idx++] << shift;
42 shift -= BITS_PER_BYTE;
44 return tmp;
48 * __reverse_ffs is copied from include/asm-generic/bitops/__ffs.h since
49 * MSB and LSB are reversed in a byte by f2fs_set_bit.
51 static inline unsigned long __reverse_ffs(unsigned long word)
53 int num = 0;
55 #if BITS_PER_LONG == 64
56 if ((word & 0xffffffff00000000UL) == 0)
57 num += 32;
58 else
59 word >>= 32;
60 #endif
61 if ((word & 0xffff0000) == 0)
62 num += 16;
63 else
64 word >>= 16;
66 if ((word & 0xff00) == 0)
67 num += 8;
68 else
69 word >>= 8;
71 if ((word & 0xf0) == 0)
72 num += 4;
73 else
74 word >>= 4;
76 if ((word & 0xc) == 0)
77 num += 2;
78 else
79 word >>= 2;
81 if ((word & 0x2) == 0)
82 num += 1;
83 return num;
87 * __find_rev_next(_zero)_bit is copied from lib/find_next_bit.c because
88 * f2fs_set_bit makes MSB and LSB reversed in a byte.
89 * @size must be integral times of unsigned long.
90 * Example:
91 * MSB <--> LSB
92 * f2fs_set_bit(0, bitmap) => 1000 0000
93 * f2fs_set_bit(7, bitmap) => 0000 0001
95 static unsigned long __find_rev_next_bit(const unsigned long *addr,
96 unsigned long size, unsigned long offset)
98 const unsigned long *p = addr + BIT_WORD(offset);
99 unsigned long result = size;
100 unsigned long tmp;
102 if (offset >= size)
103 return size;
105 size -= (offset & ~(BITS_PER_LONG - 1));
106 offset %= BITS_PER_LONG;
108 while (1) {
109 if (*p == 0)
110 goto pass;
112 tmp = __reverse_ulong((unsigned char *)p);
114 tmp &= ~0UL >> offset;
115 if (size < BITS_PER_LONG)
116 tmp &= (~0UL << (BITS_PER_LONG - size));
117 if (tmp)
118 goto found;
119 pass:
120 if (size <= BITS_PER_LONG)
121 break;
122 size -= BITS_PER_LONG;
123 offset = 0;
124 p++;
126 return result;
127 found:
128 return result - size + __reverse_ffs(tmp);
131 static unsigned long __find_rev_next_zero_bit(const unsigned long *addr,
132 unsigned long size, unsigned long offset)
134 const unsigned long *p = addr + BIT_WORD(offset);
135 unsigned long result = size;
136 unsigned long tmp;
138 if (offset >= size)
139 return size;
141 size -= (offset & ~(BITS_PER_LONG - 1));
142 offset %= BITS_PER_LONG;
144 while (1) {
145 if (*p == ~0UL)
146 goto pass;
148 tmp = __reverse_ulong((unsigned char *)p);
150 if (offset)
151 tmp |= ~0UL << (BITS_PER_LONG - offset);
152 if (size < BITS_PER_LONG)
153 tmp |= ~0UL >> size;
154 if (tmp != ~0UL)
155 goto found;
156 pass:
157 if (size <= BITS_PER_LONG)
158 break;
159 size -= BITS_PER_LONG;
160 offset = 0;
161 p++;
163 return result;
164 found:
165 return result - size + __reverse_ffz(tmp);
168 void register_inmem_page(struct inode *inode, struct page *page)
170 struct f2fs_inode_info *fi = F2FS_I(inode);
171 struct inmem_pages *new;
173 f2fs_trace_pid(page);
175 set_page_private(page, (unsigned long)ATOMIC_WRITTEN_PAGE);
176 SetPagePrivate(page);
178 new = f2fs_kmem_cache_alloc(inmem_entry_slab, GFP_NOFS);
180 /* add atomic page indices to the list */
181 new->page = page;
182 INIT_LIST_HEAD(&new->list);
184 /* increase reference count with clean state */
185 mutex_lock(&fi->inmem_lock);
186 get_page(page);
187 list_add_tail(&new->list, &fi->inmem_pages);
188 inc_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
189 mutex_unlock(&fi->inmem_lock);
191 trace_f2fs_register_inmem_page(page, INMEM);
194 static int __revoke_inmem_pages(struct inode *inode,
195 struct list_head *head, bool drop, bool recover)
197 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
198 struct inmem_pages *cur, *tmp;
199 int err = 0;
201 list_for_each_entry_safe(cur, tmp, head, list) {
202 struct page *page = cur->page;
204 if (drop)
205 trace_f2fs_commit_inmem_page(page, INMEM_DROP);
207 lock_page(page);
209 if (recover) {
210 struct dnode_of_data dn;
211 struct node_info ni;
213 trace_f2fs_commit_inmem_page(page, INMEM_REVOKE);
215 set_new_dnode(&dn, inode, NULL, NULL, 0);
216 if (get_dnode_of_data(&dn, page->index, LOOKUP_NODE)) {
217 err = -EAGAIN;
218 goto next;
220 get_node_info(sbi, dn.nid, &ni);
221 f2fs_replace_block(sbi, &dn, dn.data_blkaddr,
222 cur->old_addr, ni.version, true, true);
223 f2fs_put_dnode(&dn);
225 next:
226 /* we don't need to invalidate this in the sccessful status */
227 if (drop || recover)
228 ClearPageUptodate(page);
229 set_page_private(page, 0);
230 ClearPagePrivate(page);
231 f2fs_put_page(page, 1);
233 list_del(&cur->list);
234 kmem_cache_free(inmem_entry_slab, cur);
235 dec_page_count(F2FS_I_SB(inode), F2FS_INMEM_PAGES);
237 return err;
240 void drop_inmem_pages(struct inode *inode)
242 struct f2fs_inode_info *fi = F2FS_I(inode);
244 clear_inode_flag(inode, FI_ATOMIC_FILE);
246 mutex_lock(&fi->inmem_lock);
247 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
248 mutex_unlock(&fi->inmem_lock);
251 static int __commit_inmem_pages(struct inode *inode,
252 struct list_head *revoke_list)
254 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
255 struct f2fs_inode_info *fi = F2FS_I(inode);
256 struct inmem_pages *cur, *tmp;
257 struct f2fs_io_info fio = {
258 .sbi = sbi,
259 .type = DATA,
260 .op = REQ_OP_WRITE,
261 .op_flags = WRITE_SYNC | REQ_PRIO,
262 .encrypted_page = NULL,
264 bool submit_bio = false;
265 int err = 0;
267 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
268 struct page *page = cur->page;
270 lock_page(page);
271 if (page->mapping == inode->i_mapping) {
272 trace_f2fs_commit_inmem_page(page, INMEM);
274 set_page_dirty(page);
275 f2fs_wait_on_page_writeback(page, DATA, true);
276 if (clear_page_dirty_for_io(page))
277 inode_dec_dirty_pages(inode);
279 fio.page = page;
280 err = do_write_data_page(&fio);
281 if (err) {
282 unlock_page(page);
283 break;
286 /* record old blkaddr for revoking */
287 cur->old_addr = fio.old_blkaddr;
289 clear_cold_data(page);
290 submit_bio = true;
292 unlock_page(page);
293 list_move_tail(&cur->list, revoke_list);
296 if (submit_bio)
297 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
299 if (!err)
300 __revoke_inmem_pages(inode, revoke_list, false, false);
302 return err;
305 int commit_inmem_pages(struct inode *inode)
307 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
308 struct f2fs_inode_info *fi = F2FS_I(inode);
309 struct list_head revoke_list;
310 int err;
312 INIT_LIST_HEAD(&revoke_list);
313 f2fs_balance_fs(sbi, true);
314 f2fs_lock_op(sbi);
316 mutex_lock(&fi->inmem_lock);
317 err = __commit_inmem_pages(inode, &revoke_list);
318 if (err) {
319 int ret;
321 * try to revoke all committed pages, but still we could fail
322 * due to no memory or other reason, if that happened, EAGAIN
323 * will be returned, which means in such case, transaction is
324 * already not integrity, caller should use journal to do the
325 * recovery or rewrite & commit last transaction. For other
326 * error number, revoking was done by filesystem itself.
328 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
329 if (ret)
330 err = ret;
332 /* drop all uncommitted pages */
333 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
335 mutex_unlock(&fi->inmem_lock);
337 f2fs_unlock_op(sbi);
338 return err;
342 * This function balances dirty node and dentry pages.
343 * In addition, it controls garbage collection.
345 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
347 if (!need)
348 return;
350 /* balance_fs_bg is able to be pending */
351 if (excess_cached_nats(sbi))
352 f2fs_balance_fs_bg(sbi);
355 * We should do GC or end up with checkpoint, if there are so many dirty
356 * dir/node pages without enough free segments.
358 if (has_not_enough_free_secs(sbi, 0)) {
359 mutex_lock(&sbi->gc_mutex);
360 f2fs_gc(sbi, false);
364 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
366 /* try to shrink extent cache when there is no enough memory */
367 if (!available_free_memory(sbi, EXTENT_CACHE))
368 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
370 /* check the # of cached NAT entries */
371 if (!available_free_memory(sbi, NAT_ENTRIES))
372 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
374 if (!available_free_memory(sbi, FREE_NIDS))
375 try_to_free_nids(sbi, MAX_FREE_NIDS);
376 else
377 build_free_nids(sbi);
379 /* checkpoint is the only way to shrink partial cached entries */
380 if (!available_free_memory(sbi, NAT_ENTRIES) ||
381 !available_free_memory(sbi, INO_ENTRIES) ||
382 excess_prefree_segs(sbi) ||
383 excess_dirty_nats(sbi) ||
384 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
385 if (test_opt(sbi, DATA_FLUSH)) {
386 struct blk_plug plug;
388 blk_start_plug(&plug);
389 sync_dirty_inodes(sbi, FILE_INODE);
390 blk_finish_plug(&plug);
392 f2fs_sync_fs(sbi->sb, true);
393 stat_inc_bg_cp_count(sbi->stat_info);
397 static int issue_flush_thread(void *data)
399 struct f2fs_sb_info *sbi = data;
400 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
401 wait_queue_head_t *q = &fcc->flush_wait_queue;
402 repeat:
403 if (kthread_should_stop())
404 return 0;
406 if (!llist_empty(&fcc->issue_list)) {
407 struct bio *bio;
408 struct flush_cmd *cmd, *next;
409 int ret;
411 bio = f2fs_bio_alloc(0);
413 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
414 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
416 bio->bi_bdev = sbi->sb->s_bdev;
417 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
418 ret = submit_bio_wait(bio);
420 llist_for_each_entry_safe(cmd, next,
421 fcc->dispatch_list, llnode) {
422 cmd->ret = ret;
423 complete(&cmd->wait);
425 bio_put(bio);
426 fcc->dispatch_list = NULL;
429 wait_event_interruptible(*q,
430 kthread_should_stop() || !llist_empty(&fcc->issue_list));
431 goto repeat;
434 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
436 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
437 struct flush_cmd cmd;
439 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
440 test_opt(sbi, FLUSH_MERGE));
442 if (test_opt(sbi, NOBARRIER))
443 return 0;
445 if (!test_opt(sbi, FLUSH_MERGE) || !atomic_read(&fcc->submit_flush)) {
446 struct bio *bio = f2fs_bio_alloc(0);
447 int ret;
449 atomic_inc(&fcc->submit_flush);
450 bio->bi_bdev = sbi->sb->s_bdev;
451 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH);
452 ret = submit_bio_wait(bio);
453 atomic_dec(&fcc->submit_flush);
454 bio_put(bio);
455 return ret;
458 init_completion(&cmd.wait);
460 atomic_inc(&fcc->submit_flush);
461 llist_add(&cmd.llnode, &fcc->issue_list);
463 if (!fcc->dispatch_list)
464 wake_up(&fcc->flush_wait_queue);
466 wait_for_completion(&cmd.wait);
467 atomic_dec(&fcc->submit_flush);
469 return cmd.ret;
472 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
474 dev_t dev = sbi->sb->s_bdev->bd_dev;
475 struct flush_cmd_control *fcc;
476 int err = 0;
478 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
479 if (!fcc)
480 return -ENOMEM;
481 atomic_set(&fcc->submit_flush, 0);
482 init_waitqueue_head(&fcc->flush_wait_queue);
483 init_llist_head(&fcc->issue_list);
484 SM_I(sbi)->cmd_control_info = fcc;
485 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
486 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
487 if (IS_ERR(fcc->f2fs_issue_flush)) {
488 err = PTR_ERR(fcc->f2fs_issue_flush);
489 kfree(fcc);
490 SM_I(sbi)->cmd_control_info = NULL;
491 return err;
494 return err;
497 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
499 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
501 if (fcc && fcc->f2fs_issue_flush)
502 kthread_stop(fcc->f2fs_issue_flush);
503 kfree(fcc);
504 SM_I(sbi)->cmd_control_info = NULL;
507 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
508 enum dirty_type dirty_type)
510 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
512 /* need not be added */
513 if (IS_CURSEG(sbi, segno))
514 return;
516 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
517 dirty_i->nr_dirty[dirty_type]++;
519 if (dirty_type == DIRTY) {
520 struct seg_entry *sentry = get_seg_entry(sbi, segno);
521 enum dirty_type t = sentry->type;
523 if (unlikely(t >= DIRTY)) {
524 f2fs_bug_on(sbi, 1);
525 return;
527 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
528 dirty_i->nr_dirty[t]++;
532 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
533 enum dirty_type dirty_type)
535 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
537 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
538 dirty_i->nr_dirty[dirty_type]--;
540 if (dirty_type == DIRTY) {
541 struct seg_entry *sentry = get_seg_entry(sbi, segno);
542 enum dirty_type t = sentry->type;
544 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
545 dirty_i->nr_dirty[t]--;
547 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
548 clear_bit(GET_SECNO(sbi, segno),
549 dirty_i->victim_secmap);
554 * Should not occur error such as -ENOMEM.
555 * Adding dirty entry into seglist is not critical operation.
556 * If a given segment is one of current working segments, it won't be added.
558 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
560 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
561 unsigned short valid_blocks;
563 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
564 return;
566 mutex_lock(&dirty_i->seglist_lock);
568 valid_blocks = get_valid_blocks(sbi, segno, 0);
570 if (valid_blocks == 0) {
571 __locate_dirty_segment(sbi, segno, PRE);
572 __remove_dirty_segment(sbi, segno, DIRTY);
573 } else if (valid_blocks < sbi->blocks_per_seg) {
574 __locate_dirty_segment(sbi, segno, DIRTY);
575 } else {
576 /* Recovery routine with SSR needs this */
577 __remove_dirty_segment(sbi, segno, DIRTY);
580 mutex_unlock(&dirty_i->seglist_lock);
583 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
584 block_t blkstart, block_t blklen)
586 sector_t start = SECTOR_FROM_BLOCK(blkstart);
587 sector_t len = SECTOR_FROM_BLOCK(blklen);
588 struct seg_entry *se;
589 unsigned int offset;
590 block_t i;
592 for (i = blkstart; i < blkstart + blklen; i++) {
593 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
594 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
596 if (!f2fs_test_and_set_bit(offset, se->discard_map))
597 sbi->discard_blks--;
599 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
600 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
603 bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
605 int err = -EOPNOTSUPP;
607 if (test_opt(sbi, DISCARD)) {
608 struct seg_entry *se = get_seg_entry(sbi,
609 GET_SEGNO(sbi, blkaddr));
610 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
612 if (f2fs_test_bit(offset, se->discard_map))
613 return false;
615 err = f2fs_issue_discard(sbi, blkaddr, 1);
618 if (err) {
619 update_meta_page(sbi, NULL, blkaddr);
620 return true;
622 return false;
625 static void __add_discard_entry(struct f2fs_sb_info *sbi,
626 struct cp_control *cpc, struct seg_entry *se,
627 unsigned int start, unsigned int end)
629 struct list_head *head = &SM_I(sbi)->discard_list;
630 struct discard_entry *new, *last;
632 if (!list_empty(head)) {
633 last = list_last_entry(head, struct discard_entry, list);
634 if (START_BLOCK(sbi, cpc->trim_start) + start ==
635 last->blkaddr + last->len) {
636 last->len += end - start;
637 goto done;
641 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
642 INIT_LIST_HEAD(&new->list);
643 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
644 new->len = end - start;
645 list_add_tail(&new->list, head);
646 done:
647 SM_I(sbi)->nr_discards += end - start;
650 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
652 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
653 int max_blocks = sbi->blocks_per_seg;
654 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
655 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
656 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
657 unsigned long *discard_map = (unsigned long *)se->discard_map;
658 unsigned long *dmap = SIT_I(sbi)->tmp_map;
659 unsigned int start = 0, end = -1;
660 bool force = (cpc->reason == CP_DISCARD);
661 int i;
663 if (se->valid_blocks == max_blocks)
664 return;
666 if (!force) {
667 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
668 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
669 return;
672 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
673 for (i = 0; i < entries; i++)
674 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
675 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
677 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
678 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
679 if (start >= max_blocks)
680 break;
682 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
683 if (force && start && end != max_blocks
684 && (end - start) < cpc->trim_minlen)
685 continue;
687 __add_discard_entry(sbi, cpc, se, start, end);
691 void release_discard_addrs(struct f2fs_sb_info *sbi)
693 struct list_head *head = &(SM_I(sbi)->discard_list);
694 struct discard_entry *entry, *this;
696 /* drop caches */
697 list_for_each_entry_safe(entry, this, head, list) {
698 list_del(&entry->list);
699 kmem_cache_free(discard_entry_slab, entry);
704 * Should call clear_prefree_segments after checkpoint is done.
706 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
708 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
709 unsigned int segno;
711 mutex_lock(&dirty_i->seglist_lock);
712 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
713 __set_test_and_free(sbi, segno);
714 mutex_unlock(&dirty_i->seglist_lock);
717 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
719 struct list_head *head = &(SM_I(sbi)->discard_list);
720 struct discard_entry *entry, *this;
721 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
722 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
723 unsigned int start = 0, end = -1;
724 unsigned int secno, start_segno;
725 bool force = (cpc->reason == CP_DISCARD);
727 mutex_lock(&dirty_i->seglist_lock);
729 while (1) {
730 int i;
731 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
732 if (start >= MAIN_SEGS(sbi))
733 break;
734 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
735 start + 1);
737 for (i = start; i < end; i++)
738 clear_bit(i, prefree_map);
740 dirty_i->nr_dirty[PRE] -= end - start;
742 if (force || !test_opt(sbi, DISCARD))
743 continue;
745 if (!test_opt(sbi, LFS) || sbi->segs_per_sec == 1) {
746 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
747 (end - start) << sbi->log_blocks_per_seg);
748 continue;
750 next:
751 secno = GET_SECNO(sbi, start);
752 start_segno = secno * sbi->segs_per_sec;
753 if (!IS_CURSEC(sbi, secno) &&
754 !get_valid_blocks(sbi, start, sbi->segs_per_sec))
755 f2fs_issue_discard(sbi, START_BLOCK(sbi, start_segno),
756 sbi->segs_per_sec << sbi->log_blocks_per_seg);
758 start = start_segno + sbi->segs_per_sec;
759 if (start < end)
760 goto next;
762 mutex_unlock(&dirty_i->seglist_lock);
764 /* send small discards */
765 list_for_each_entry_safe(entry, this, head, list) {
766 if (force && entry->len < cpc->trim_minlen)
767 goto skip;
768 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
769 cpc->trimmed += entry->len;
770 skip:
771 list_del(&entry->list);
772 SM_I(sbi)->nr_discards -= entry->len;
773 kmem_cache_free(discard_entry_slab, entry);
777 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
779 struct sit_info *sit_i = SIT_I(sbi);
781 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
782 sit_i->dirty_sentries++;
783 return false;
786 return true;
789 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
790 unsigned int segno, int modified)
792 struct seg_entry *se = get_seg_entry(sbi, segno);
793 se->type = type;
794 if (modified)
795 __mark_sit_entry_dirty(sbi, segno);
798 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
800 struct seg_entry *se;
801 unsigned int segno, offset;
802 long int new_vblocks;
804 segno = GET_SEGNO(sbi, blkaddr);
806 se = get_seg_entry(sbi, segno);
807 new_vblocks = se->valid_blocks + del;
808 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
810 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
811 (new_vblocks > sbi->blocks_per_seg)));
813 se->valid_blocks = new_vblocks;
814 se->mtime = get_mtime(sbi);
815 SIT_I(sbi)->max_mtime = se->mtime;
817 /* Update valid block bitmap */
818 if (del > 0) {
819 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
820 f2fs_bug_on(sbi, 1);
821 if (!f2fs_test_and_set_bit(offset, se->discard_map))
822 sbi->discard_blks--;
823 } else {
824 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
825 f2fs_bug_on(sbi, 1);
826 if (f2fs_test_and_clear_bit(offset, se->discard_map))
827 sbi->discard_blks++;
829 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
830 se->ckpt_valid_blocks += del;
832 __mark_sit_entry_dirty(sbi, segno);
834 /* update total number of valid blocks to be written in ckpt area */
835 SIT_I(sbi)->written_valid_blocks += del;
837 if (sbi->segs_per_sec > 1)
838 get_sec_entry(sbi, segno)->valid_blocks += del;
841 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
843 update_sit_entry(sbi, new, 1);
844 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
845 update_sit_entry(sbi, old, -1);
847 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
848 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
851 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
853 unsigned int segno = GET_SEGNO(sbi, addr);
854 struct sit_info *sit_i = SIT_I(sbi);
856 f2fs_bug_on(sbi, addr == NULL_ADDR);
857 if (addr == NEW_ADDR)
858 return;
860 /* add it into sit main buffer */
861 mutex_lock(&sit_i->sentry_lock);
863 update_sit_entry(sbi, addr, -1);
865 /* add it into dirty seglist */
866 locate_dirty_segment(sbi, segno);
868 mutex_unlock(&sit_i->sentry_lock);
871 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
873 struct sit_info *sit_i = SIT_I(sbi);
874 unsigned int segno, offset;
875 struct seg_entry *se;
876 bool is_cp = false;
878 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
879 return true;
881 mutex_lock(&sit_i->sentry_lock);
883 segno = GET_SEGNO(sbi, blkaddr);
884 se = get_seg_entry(sbi, segno);
885 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
887 if (f2fs_test_bit(offset, se->ckpt_valid_map))
888 is_cp = true;
890 mutex_unlock(&sit_i->sentry_lock);
892 return is_cp;
896 * This function should be resided under the curseg_mutex lock
898 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
899 struct f2fs_summary *sum)
901 struct curseg_info *curseg = CURSEG_I(sbi, type);
902 void *addr = curseg->sum_blk;
903 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
904 memcpy(addr, sum, sizeof(struct f2fs_summary));
908 * Calculate the number of current summary pages for writing
910 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
912 int valid_sum_count = 0;
913 int i, sum_in_page;
915 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
916 if (sbi->ckpt->alloc_type[i] == SSR)
917 valid_sum_count += sbi->blocks_per_seg;
918 else {
919 if (for_ra)
920 valid_sum_count += le16_to_cpu(
921 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
922 else
923 valid_sum_count += curseg_blkoff(sbi, i);
927 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
928 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
929 if (valid_sum_count <= sum_in_page)
930 return 1;
931 else if ((valid_sum_count - sum_in_page) <=
932 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
933 return 2;
934 return 3;
938 * Caller should put this summary page
940 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
942 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
945 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
947 struct page *page = grab_meta_page(sbi, blk_addr);
948 void *dst = page_address(page);
950 if (src)
951 memcpy(dst, src, PAGE_SIZE);
952 else
953 memset(dst, 0, PAGE_SIZE);
954 set_page_dirty(page);
955 f2fs_put_page(page, 1);
958 static void write_sum_page(struct f2fs_sb_info *sbi,
959 struct f2fs_summary_block *sum_blk, block_t blk_addr)
961 update_meta_page(sbi, (void *)sum_blk, blk_addr);
964 static void write_current_sum_page(struct f2fs_sb_info *sbi,
965 int type, block_t blk_addr)
967 struct curseg_info *curseg = CURSEG_I(sbi, type);
968 struct page *page = grab_meta_page(sbi, blk_addr);
969 struct f2fs_summary_block *src = curseg->sum_blk;
970 struct f2fs_summary_block *dst;
972 dst = (struct f2fs_summary_block *)page_address(page);
974 mutex_lock(&curseg->curseg_mutex);
976 down_read(&curseg->journal_rwsem);
977 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
978 up_read(&curseg->journal_rwsem);
980 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
981 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
983 mutex_unlock(&curseg->curseg_mutex);
985 set_page_dirty(page);
986 f2fs_put_page(page, 1);
989 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
991 struct curseg_info *curseg = CURSEG_I(sbi, type);
992 unsigned int segno = curseg->segno + 1;
993 struct free_segmap_info *free_i = FREE_I(sbi);
995 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
996 return !test_bit(segno, free_i->free_segmap);
997 return 0;
1001 * Find a new segment from the free segments bitmap to right order
1002 * This function should be returned with success, otherwise BUG
1004 static void get_new_segment(struct f2fs_sb_info *sbi,
1005 unsigned int *newseg, bool new_sec, int dir)
1007 struct free_segmap_info *free_i = FREE_I(sbi);
1008 unsigned int segno, secno, zoneno;
1009 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
1010 unsigned int hint = *newseg / sbi->segs_per_sec;
1011 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
1012 unsigned int left_start = hint;
1013 bool init = true;
1014 int go_left = 0;
1015 int i;
1017 spin_lock(&free_i->segmap_lock);
1019 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
1020 segno = find_next_zero_bit(free_i->free_segmap,
1021 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
1022 if (segno < (hint + 1) * sbi->segs_per_sec)
1023 goto got_it;
1025 find_other_zone:
1026 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
1027 if (secno >= MAIN_SECS(sbi)) {
1028 if (dir == ALLOC_RIGHT) {
1029 secno = find_next_zero_bit(free_i->free_secmap,
1030 MAIN_SECS(sbi), 0);
1031 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
1032 } else {
1033 go_left = 1;
1034 left_start = hint - 1;
1037 if (go_left == 0)
1038 goto skip_left;
1040 while (test_bit(left_start, free_i->free_secmap)) {
1041 if (left_start > 0) {
1042 left_start--;
1043 continue;
1045 left_start = find_next_zero_bit(free_i->free_secmap,
1046 MAIN_SECS(sbi), 0);
1047 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1048 break;
1050 secno = left_start;
1051 skip_left:
1052 hint = secno;
1053 segno = secno * sbi->segs_per_sec;
1054 zoneno = secno / sbi->secs_per_zone;
1056 /* give up on finding another zone */
1057 if (!init)
1058 goto got_it;
1059 if (sbi->secs_per_zone == 1)
1060 goto got_it;
1061 if (zoneno == old_zoneno)
1062 goto got_it;
1063 if (dir == ALLOC_LEFT) {
1064 if (!go_left && zoneno + 1 >= total_zones)
1065 goto got_it;
1066 if (go_left && zoneno == 0)
1067 goto got_it;
1069 for (i = 0; i < NR_CURSEG_TYPE; i++)
1070 if (CURSEG_I(sbi, i)->zone == zoneno)
1071 break;
1073 if (i < NR_CURSEG_TYPE) {
1074 /* zone is in user, try another */
1075 if (go_left)
1076 hint = zoneno * sbi->secs_per_zone - 1;
1077 else if (zoneno + 1 >= total_zones)
1078 hint = 0;
1079 else
1080 hint = (zoneno + 1) * sbi->secs_per_zone;
1081 init = false;
1082 goto find_other_zone;
1084 got_it:
1085 /* set it as dirty segment in free segmap */
1086 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1087 __set_inuse(sbi, segno);
1088 *newseg = segno;
1089 spin_unlock(&free_i->segmap_lock);
1092 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1094 struct curseg_info *curseg = CURSEG_I(sbi, type);
1095 struct summary_footer *sum_footer;
1097 curseg->segno = curseg->next_segno;
1098 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1099 curseg->next_blkoff = 0;
1100 curseg->next_segno = NULL_SEGNO;
1102 sum_footer = &(curseg->sum_blk->footer);
1103 memset(sum_footer, 0, sizeof(struct summary_footer));
1104 if (IS_DATASEG(type))
1105 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1106 if (IS_NODESEG(type))
1107 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1108 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1112 * Allocate a current working segment.
1113 * This function always allocates a free segment in LFS manner.
1115 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1117 struct curseg_info *curseg = CURSEG_I(sbi, type);
1118 unsigned int segno = curseg->segno;
1119 int dir = ALLOC_LEFT;
1121 write_sum_page(sbi, curseg->sum_blk,
1122 GET_SUM_BLOCK(sbi, segno));
1123 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1124 dir = ALLOC_RIGHT;
1126 if (test_opt(sbi, NOHEAP))
1127 dir = ALLOC_RIGHT;
1129 get_new_segment(sbi, &segno, new_sec, dir);
1130 curseg->next_segno = segno;
1131 reset_curseg(sbi, type, 1);
1132 curseg->alloc_type = LFS;
1135 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1136 struct curseg_info *seg, block_t start)
1138 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1139 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1140 unsigned long *target_map = SIT_I(sbi)->tmp_map;
1141 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1142 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1143 int i, pos;
1145 for (i = 0; i < entries; i++)
1146 target_map[i] = ckpt_map[i] | cur_map[i];
1148 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1150 seg->next_blkoff = pos;
1154 * If a segment is written by LFS manner, next block offset is just obtained
1155 * by increasing the current block offset. However, if a segment is written by
1156 * SSR manner, next block offset obtained by calling __next_free_blkoff
1158 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1159 struct curseg_info *seg)
1161 if (seg->alloc_type == SSR)
1162 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1163 else
1164 seg->next_blkoff++;
1168 * This function always allocates a used segment(from dirty seglist) by SSR
1169 * manner, so it should recover the existing segment information of valid blocks
1171 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1173 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1174 struct curseg_info *curseg = CURSEG_I(sbi, type);
1175 unsigned int new_segno = curseg->next_segno;
1176 struct f2fs_summary_block *sum_node;
1177 struct page *sum_page;
1179 write_sum_page(sbi, curseg->sum_blk,
1180 GET_SUM_BLOCK(sbi, curseg->segno));
1181 __set_test_and_inuse(sbi, new_segno);
1183 mutex_lock(&dirty_i->seglist_lock);
1184 __remove_dirty_segment(sbi, new_segno, PRE);
1185 __remove_dirty_segment(sbi, new_segno, DIRTY);
1186 mutex_unlock(&dirty_i->seglist_lock);
1188 reset_curseg(sbi, type, 1);
1189 curseg->alloc_type = SSR;
1190 __next_free_blkoff(sbi, curseg, 0);
1192 if (reuse) {
1193 sum_page = get_sum_page(sbi, new_segno);
1194 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1195 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1196 f2fs_put_page(sum_page, 1);
1200 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1202 struct curseg_info *curseg = CURSEG_I(sbi, type);
1203 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1205 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1206 return v_ops->get_victim(sbi,
1207 &(curseg)->next_segno, BG_GC, type, SSR);
1209 /* For data segments, let's do SSR more intensively */
1210 for (; type >= CURSEG_HOT_DATA; type--)
1211 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1212 BG_GC, type, SSR))
1213 return 1;
1214 return 0;
1218 * flush out current segment and replace it with new segment
1219 * This function should be returned with success, otherwise BUG
1221 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1222 int type, bool force)
1224 struct curseg_info *curseg = CURSEG_I(sbi, type);
1226 if (force)
1227 new_curseg(sbi, type, true);
1228 else if (type == CURSEG_WARM_NODE)
1229 new_curseg(sbi, type, false);
1230 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1231 new_curseg(sbi, type, false);
1232 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1233 change_curseg(sbi, type, true);
1234 else
1235 new_curseg(sbi, type, false);
1237 stat_inc_seg_type(sbi, curseg);
1240 static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1242 struct curseg_info *curseg = CURSEG_I(sbi, type);
1243 unsigned int old_segno;
1245 old_segno = curseg->segno;
1246 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1247 locate_dirty_segment(sbi, old_segno);
1250 void allocate_new_segments(struct f2fs_sb_info *sbi)
1252 int i;
1254 if (test_opt(sbi, LFS))
1255 return;
1257 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1258 __allocate_new_segments(sbi, i);
1261 static const struct segment_allocation default_salloc_ops = {
1262 .allocate_segment = allocate_segment_by_default,
1265 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1267 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1268 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1269 unsigned int start_segno, end_segno;
1270 struct cp_control cpc;
1271 int err = 0;
1273 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1274 return -EINVAL;
1276 cpc.trimmed = 0;
1277 if (end <= MAIN_BLKADDR(sbi))
1278 goto out;
1280 /* start/end segment number in main_area */
1281 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1282 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1283 GET_SEGNO(sbi, end);
1284 cpc.reason = CP_DISCARD;
1285 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1287 /* do checkpoint to issue discard commands safely */
1288 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1289 cpc.trim_start = start_segno;
1291 if (sbi->discard_blks == 0)
1292 break;
1293 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1294 cpc.trim_end = end_segno;
1295 else
1296 cpc.trim_end = min_t(unsigned int,
1297 rounddown(start_segno +
1298 BATCHED_TRIM_SEGMENTS(sbi),
1299 sbi->segs_per_sec) - 1, end_segno);
1301 mutex_lock(&sbi->gc_mutex);
1302 err = write_checkpoint(sbi, &cpc);
1303 mutex_unlock(&sbi->gc_mutex);
1305 out:
1306 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1307 return err;
1310 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1312 struct curseg_info *curseg = CURSEG_I(sbi, type);
1313 if (curseg->next_blkoff < sbi->blocks_per_seg)
1314 return true;
1315 return false;
1318 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1320 if (p_type == DATA)
1321 return CURSEG_HOT_DATA;
1322 else
1323 return CURSEG_HOT_NODE;
1326 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1328 if (p_type == DATA) {
1329 struct inode *inode = page->mapping->host;
1331 if (S_ISDIR(inode->i_mode))
1332 return CURSEG_HOT_DATA;
1333 else
1334 return CURSEG_COLD_DATA;
1335 } else {
1336 if (IS_DNODE(page) && is_cold_node(page))
1337 return CURSEG_WARM_NODE;
1338 else
1339 return CURSEG_COLD_NODE;
1343 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1345 if (p_type == DATA) {
1346 struct inode *inode = page->mapping->host;
1348 if (S_ISDIR(inode->i_mode))
1349 return CURSEG_HOT_DATA;
1350 else if (is_cold_data(page) || file_is_cold(inode))
1351 return CURSEG_COLD_DATA;
1352 else
1353 return CURSEG_WARM_DATA;
1354 } else {
1355 if (IS_DNODE(page))
1356 return is_cold_node(page) ? CURSEG_WARM_NODE :
1357 CURSEG_HOT_NODE;
1358 else
1359 return CURSEG_COLD_NODE;
1363 static int __get_segment_type(struct page *page, enum page_type p_type)
1365 switch (F2FS_P_SB(page)->active_logs) {
1366 case 2:
1367 return __get_segment_type_2(page, p_type);
1368 case 4:
1369 return __get_segment_type_4(page, p_type);
1371 /* NR_CURSEG_TYPE(6) logs by default */
1372 f2fs_bug_on(F2FS_P_SB(page),
1373 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1374 return __get_segment_type_6(page, p_type);
1377 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1378 block_t old_blkaddr, block_t *new_blkaddr,
1379 struct f2fs_summary *sum, int type)
1381 struct sit_info *sit_i = SIT_I(sbi);
1382 struct curseg_info *curseg;
1383 bool direct_io = (type == CURSEG_DIRECT_IO);
1385 type = direct_io ? CURSEG_WARM_DATA : type;
1387 curseg = CURSEG_I(sbi, type);
1389 mutex_lock(&curseg->curseg_mutex);
1390 mutex_lock(&sit_i->sentry_lock);
1392 /* direct_io'ed data is aligned to the segment for better performance */
1393 if (direct_io && curseg->next_blkoff &&
1394 !has_not_enough_free_secs(sbi, 0))
1395 __allocate_new_segments(sbi, type);
1397 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1400 * __add_sum_entry should be resided under the curseg_mutex
1401 * because, this function updates a summary entry in the
1402 * current summary block.
1404 __add_sum_entry(sbi, type, sum);
1406 __refresh_next_blkoff(sbi, curseg);
1408 stat_inc_block_count(sbi, curseg);
1410 if (!__has_curseg_space(sbi, type))
1411 sit_i->s_ops->allocate_segment(sbi, type, false);
1413 * SIT information should be updated before segment allocation,
1414 * since SSR needs latest valid block information.
1416 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1418 mutex_unlock(&sit_i->sentry_lock);
1420 if (page && IS_NODESEG(type))
1421 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1423 mutex_unlock(&curseg->curseg_mutex);
1426 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1428 int type = __get_segment_type(fio->page, fio->type);
1430 if (fio->type == NODE || fio->type == DATA)
1431 mutex_lock(&fio->sbi->wio_mutex[fio->type]);
1433 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1434 &fio->new_blkaddr, sum, type);
1436 /* writeout dirty page into bdev */
1437 f2fs_submit_page_mbio(fio);
1439 if (fio->type == NODE || fio->type == DATA)
1440 mutex_unlock(&fio->sbi->wio_mutex[fio->type]);
1443 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1445 struct f2fs_io_info fio = {
1446 .sbi = sbi,
1447 .type = META,
1448 .op = REQ_OP_WRITE,
1449 .op_flags = WRITE_SYNC | REQ_META | REQ_PRIO,
1450 .old_blkaddr = page->index,
1451 .new_blkaddr = page->index,
1452 .page = page,
1453 .encrypted_page = NULL,
1456 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1457 fio.op_flags &= ~REQ_META;
1459 set_page_writeback(page);
1460 f2fs_submit_page_mbio(&fio);
1463 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1465 struct f2fs_summary sum;
1467 set_summary(&sum, nid, 0, 0);
1468 do_write_page(&sum, fio);
1471 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1473 struct f2fs_sb_info *sbi = fio->sbi;
1474 struct f2fs_summary sum;
1475 struct node_info ni;
1477 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1478 get_node_info(sbi, dn->nid, &ni);
1479 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1480 do_write_page(&sum, fio);
1481 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1484 void rewrite_data_page(struct f2fs_io_info *fio)
1486 fio->new_blkaddr = fio->old_blkaddr;
1487 stat_inc_inplace_blocks(fio->sbi);
1488 f2fs_submit_page_mbio(fio);
1491 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1492 block_t old_blkaddr, block_t new_blkaddr,
1493 bool recover_curseg, bool recover_newaddr)
1495 struct sit_info *sit_i = SIT_I(sbi);
1496 struct curseg_info *curseg;
1497 unsigned int segno, old_cursegno;
1498 struct seg_entry *se;
1499 int type;
1500 unsigned short old_blkoff;
1502 segno = GET_SEGNO(sbi, new_blkaddr);
1503 se = get_seg_entry(sbi, segno);
1504 type = se->type;
1506 if (!recover_curseg) {
1507 /* for recovery flow */
1508 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1509 if (old_blkaddr == NULL_ADDR)
1510 type = CURSEG_COLD_DATA;
1511 else
1512 type = CURSEG_WARM_DATA;
1514 } else {
1515 if (!IS_CURSEG(sbi, segno))
1516 type = CURSEG_WARM_DATA;
1519 curseg = CURSEG_I(sbi, type);
1521 mutex_lock(&curseg->curseg_mutex);
1522 mutex_lock(&sit_i->sentry_lock);
1524 old_cursegno = curseg->segno;
1525 old_blkoff = curseg->next_blkoff;
1527 /* change the current segment */
1528 if (segno != curseg->segno) {
1529 curseg->next_segno = segno;
1530 change_curseg(sbi, type, true);
1533 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1534 __add_sum_entry(sbi, type, sum);
1536 if (!recover_curseg || recover_newaddr)
1537 update_sit_entry(sbi, new_blkaddr, 1);
1538 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1539 update_sit_entry(sbi, old_blkaddr, -1);
1541 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1542 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1544 locate_dirty_segment(sbi, old_cursegno);
1546 if (recover_curseg) {
1547 if (old_cursegno != curseg->segno) {
1548 curseg->next_segno = old_cursegno;
1549 change_curseg(sbi, type, true);
1551 curseg->next_blkoff = old_blkoff;
1554 mutex_unlock(&sit_i->sentry_lock);
1555 mutex_unlock(&curseg->curseg_mutex);
1558 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1559 block_t old_addr, block_t new_addr,
1560 unsigned char version, bool recover_curseg,
1561 bool recover_newaddr)
1563 struct f2fs_summary sum;
1565 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1567 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1568 recover_curseg, recover_newaddr);
1570 f2fs_update_data_blkaddr(dn, new_addr);
1573 void f2fs_wait_on_page_writeback(struct page *page,
1574 enum page_type type, bool ordered)
1576 if (PageWriteback(page)) {
1577 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1579 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
1580 if (ordered)
1581 wait_on_page_writeback(page);
1582 else
1583 wait_for_stable_page(page);
1587 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1588 block_t blkaddr)
1590 struct page *cpage;
1592 if (blkaddr == NEW_ADDR)
1593 return;
1595 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1597 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1598 if (cpage) {
1599 f2fs_wait_on_page_writeback(cpage, DATA, true);
1600 f2fs_put_page(cpage, 1);
1604 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1606 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1607 struct curseg_info *seg_i;
1608 unsigned char *kaddr;
1609 struct page *page;
1610 block_t start;
1611 int i, j, offset;
1613 start = start_sum_block(sbi);
1615 page = get_meta_page(sbi, start++);
1616 kaddr = (unsigned char *)page_address(page);
1618 /* Step 1: restore nat cache */
1619 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1620 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
1622 /* Step 2: restore sit cache */
1623 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1624 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
1625 offset = 2 * SUM_JOURNAL_SIZE;
1627 /* Step 3: restore summary entries */
1628 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1629 unsigned short blk_off;
1630 unsigned int segno;
1632 seg_i = CURSEG_I(sbi, i);
1633 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1634 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1635 seg_i->next_segno = segno;
1636 reset_curseg(sbi, i, 0);
1637 seg_i->alloc_type = ckpt->alloc_type[i];
1638 seg_i->next_blkoff = blk_off;
1640 if (seg_i->alloc_type == SSR)
1641 blk_off = sbi->blocks_per_seg;
1643 for (j = 0; j < blk_off; j++) {
1644 struct f2fs_summary *s;
1645 s = (struct f2fs_summary *)(kaddr + offset);
1646 seg_i->sum_blk->entries[j] = *s;
1647 offset += SUMMARY_SIZE;
1648 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
1649 SUM_FOOTER_SIZE)
1650 continue;
1652 f2fs_put_page(page, 1);
1653 page = NULL;
1655 page = get_meta_page(sbi, start++);
1656 kaddr = (unsigned char *)page_address(page);
1657 offset = 0;
1660 f2fs_put_page(page, 1);
1661 return 0;
1664 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1666 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1667 struct f2fs_summary_block *sum;
1668 struct curseg_info *curseg;
1669 struct page *new;
1670 unsigned short blk_off;
1671 unsigned int segno = 0;
1672 block_t blk_addr = 0;
1674 /* get segment number and block addr */
1675 if (IS_DATASEG(type)) {
1676 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1677 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1678 CURSEG_HOT_DATA]);
1679 if (__exist_node_summaries(sbi))
1680 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1681 else
1682 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1683 } else {
1684 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1685 CURSEG_HOT_NODE]);
1686 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1687 CURSEG_HOT_NODE]);
1688 if (__exist_node_summaries(sbi))
1689 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1690 type - CURSEG_HOT_NODE);
1691 else
1692 blk_addr = GET_SUM_BLOCK(sbi, segno);
1695 new = get_meta_page(sbi, blk_addr);
1696 sum = (struct f2fs_summary_block *)page_address(new);
1698 if (IS_NODESEG(type)) {
1699 if (__exist_node_summaries(sbi)) {
1700 struct f2fs_summary *ns = &sum->entries[0];
1701 int i;
1702 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1703 ns->version = 0;
1704 ns->ofs_in_node = 0;
1706 } else {
1707 int err;
1709 err = restore_node_summary(sbi, segno, sum);
1710 if (err) {
1711 f2fs_put_page(new, 1);
1712 return err;
1717 /* set uncompleted segment to curseg */
1718 curseg = CURSEG_I(sbi, type);
1719 mutex_lock(&curseg->curseg_mutex);
1721 /* update journal info */
1722 down_write(&curseg->journal_rwsem);
1723 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1724 up_write(&curseg->journal_rwsem);
1726 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1727 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
1728 curseg->next_segno = segno;
1729 reset_curseg(sbi, type, 0);
1730 curseg->alloc_type = ckpt->alloc_type[type];
1731 curseg->next_blkoff = blk_off;
1732 mutex_unlock(&curseg->curseg_mutex);
1733 f2fs_put_page(new, 1);
1734 return 0;
1737 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1739 int type = CURSEG_HOT_DATA;
1740 int err;
1742 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1743 int npages = npages_for_summary_flush(sbi, true);
1745 if (npages >= 2)
1746 ra_meta_pages(sbi, start_sum_block(sbi), npages,
1747 META_CP, true);
1749 /* restore for compacted data summary */
1750 if (read_compacted_summaries(sbi))
1751 return -EINVAL;
1752 type = CURSEG_HOT_NODE;
1755 if (__exist_node_summaries(sbi))
1756 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1757 NR_CURSEG_TYPE - type, META_CP, true);
1759 for (; type <= CURSEG_COLD_NODE; type++) {
1760 err = read_normal_summaries(sbi, type);
1761 if (err)
1762 return err;
1765 return 0;
1768 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1770 struct page *page;
1771 unsigned char *kaddr;
1772 struct f2fs_summary *summary;
1773 struct curseg_info *seg_i;
1774 int written_size = 0;
1775 int i, j;
1777 page = grab_meta_page(sbi, blkaddr++);
1778 kaddr = (unsigned char *)page_address(page);
1780 /* Step 1: write nat cache */
1781 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1782 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
1783 written_size += SUM_JOURNAL_SIZE;
1785 /* Step 2: write sit cache */
1786 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1787 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
1788 written_size += SUM_JOURNAL_SIZE;
1790 /* Step 3: write summary entries */
1791 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1792 unsigned short blkoff;
1793 seg_i = CURSEG_I(sbi, i);
1794 if (sbi->ckpt->alloc_type[i] == SSR)
1795 blkoff = sbi->blocks_per_seg;
1796 else
1797 blkoff = curseg_blkoff(sbi, i);
1799 for (j = 0; j < blkoff; j++) {
1800 if (!page) {
1801 page = grab_meta_page(sbi, blkaddr++);
1802 kaddr = (unsigned char *)page_address(page);
1803 written_size = 0;
1805 summary = (struct f2fs_summary *)(kaddr + written_size);
1806 *summary = seg_i->sum_blk->entries[j];
1807 written_size += SUMMARY_SIZE;
1809 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
1810 SUM_FOOTER_SIZE)
1811 continue;
1813 set_page_dirty(page);
1814 f2fs_put_page(page, 1);
1815 page = NULL;
1818 if (page) {
1819 set_page_dirty(page);
1820 f2fs_put_page(page, 1);
1824 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1825 block_t blkaddr, int type)
1827 int i, end;
1828 if (IS_DATASEG(type))
1829 end = type + NR_CURSEG_DATA_TYPE;
1830 else
1831 end = type + NR_CURSEG_NODE_TYPE;
1833 for (i = type; i < end; i++)
1834 write_current_sum_page(sbi, i, blkaddr + (i - type));
1837 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1839 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1840 write_compacted_summaries(sbi, start_blk);
1841 else
1842 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1845 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1847 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1850 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
1851 unsigned int val, int alloc)
1853 int i;
1855 if (type == NAT_JOURNAL) {
1856 for (i = 0; i < nats_in_cursum(journal); i++) {
1857 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
1858 return i;
1860 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1861 return update_nats_in_cursum(journal, 1);
1862 } else if (type == SIT_JOURNAL) {
1863 for (i = 0; i < sits_in_cursum(journal); i++)
1864 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
1865 return i;
1866 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1867 return update_sits_in_cursum(journal, 1);
1869 return -1;
1872 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1873 unsigned int segno)
1875 return get_meta_page(sbi, current_sit_addr(sbi, segno));
1878 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1879 unsigned int start)
1881 struct sit_info *sit_i = SIT_I(sbi);
1882 struct page *src_page, *dst_page;
1883 pgoff_t src_off, dst_off;
1884 void *src_addr, *dst_addr;
1886 src_off = current_sit_addr(sbi, start);
1887 dst_off = next_sit_addr(sbi, src_off);
1889 /* get current sit block page without lock */
1890 src_page = get_meta_page(sbi, src_off);
1891 dst_page = grab_meta_page(sbi, dst_off);
1892 f2fs_bug_on(sbi, PageDirty(src_page));
1894 src_addr = page_address(src_page);
1895 dst_addr = page_address(dst_page);
1896 memcpy(dst_addr, src_addr, PAGE_SIZE);
1898 set_page_dirty(dst_page);
1899 f2fs_put_page(src_page, 1);
1901 set_to_next_sit(sit_i, start);
1903 return dst_page;
1906 static struct sit_entry_set *grab_sit_entry_set(void)
1908 struct sit_entry_set *ses =
1909 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
1911 ses->entry_cnt = 0;
1912 INIT_LIST_HEAD(&ses->set_list);
1913 return ses;
1916 static void release_sit_entry_set(struct sit_entry_set *ses)
1918 list_del(&ses->set_list);
1919 kmem_cache_free(sit_entry_set_slab, ses);
1922 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1923 struct list_head *head)
1925 struct sit_entry_set *next = ses;
1927 if (list_is_last(&ses->set_list, head))
1928 return;
1930 list_for_each_entry_continue(next, head, set_list)
1931 if (ses->entry_cnt <= next->entry_cnt)
1932 break;
1934 list_move_tail(&ses->set_list, &next->set_list);
1937 static void add_sit_entry(unsigned int segno, struct list_head *head)
1939 struct sit_entry_set *ses;
1940 unsigned int start_segno = START_SEGNO(segno);
1942 list_for_each_entry(ses, head, set_list) {
1943 if (ses->start_segno == start_segno) {
1944 ses->entry_cnt++;
1945 adjust_sit_entry_set(ses, head);
1946 return;
1950 ses = grab_sit_entry_set();
1952 ses->start_segno = start_segno;
1953 ses->entry_cnt++;
1954 list_add(&ses->set_list, head);
1957 static void add_sits_in_set(struct f2fs_sb_info *sbi)
1959 struct f2fs_sm_info *sm_info = SM_I(sbi);
1960 struct list_head *set_list = &sm_info->sit_entry_set;
1961 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
1962 unsigned int segno;
1964 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
1965 add_sit_entry(segno, set_list);
1968 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
1970 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1971 struct f2fs_journal *journal = curseg->journal;
1972 int i;
1974 down_write(&curseg->journal_rwsem);
1975 for (i = 0; i < sits_in_cursum(journal); i++) {
1976 unsigned int segno;
1977 bool dirtied;
1979 segno = le32_to_cpu(segno_in_journal(journal, i));
1980 dirtied = __mark_sit_entry_dirty(sbi, segno);
1982 if (!dirtied)
1983 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
1985 update_sits_in_cursum(journal, -i);
1986 up_write(&curseg->journal_rwsem);
1990 * CP calls this function, which flushes SIT entries including sit_journal,
1991 * and moves prefree segs to free segs.
1993 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1995 struct sit_info *sit_i = SIT_I(sbi);
1996 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1997 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1998 struct f2fs_journal *journal = curseg->journal;
1999 struct sit_entry_set *ses, *tmp;
2000 struct list_head *head = &SM_I(sbi)->sit_entry_set;
2001 bool to_journal = true;
2002 struct seg_entry *se;
2004 mutex_lock(&sit_i->sentry_lock);
2006 if (!sit_i->dirty_sentries)
2007 goto out;
2010 * add and account sit entries of dirty bitmap in sit entry
2011 * set temporarily
2013 add_sits_in_set(sbi);
2016 * if there are no enough space in journal to store dirty sit
2017 * entries, remove all entries from journal and add and account
2018 * them in sit entry set.
2020 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
2021 remove_sits_in_journal(sbi);
2024 * there are two steps to flush sit entries:
2025 * #1, flush sit entries to journal in current cold data summary block.
2026 * #2, flush sit entries to sit page.
2028 list_for_each_entry_safe(ses, tmp, head, set_list) {
2029 struct page *page = NULL;
2030 struct f2fs_sit_block *raw_sit = NULL;
2031 unsigned int start_segno = ses->start_segno;
2032 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
2033 (unsigned long)MAIN_SEGS(sbi));
2034 unsigned int segno = start_segno;
2036 if (to_journal &&
2037 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
2038 to_journal = false;
2040 if (to_journal) {
2041 down_write(&curseg->journal_rwsem);
2042 } else {
2043 page = get_next_sit_page(sbi, start_segno);
2044 raw_sit = page_address(page);
2047 /* flush dirty sit entries in region of current sit set */
2048 for_each_set_bit_from(segno, bitmap, end) {
2049 int offset, sit_offset;
2051 se = get_seg_entry(sbi, segno);
2053 /* add discard candidates */
2054 if (cpc->reason != CP_DISCARD) {
2055 cpc->trim_start = segno;
2056 add_discard_addrs(sbi, cpc);
2059 if (to_journal) {
2060 offset = lookup_journal_in_cursum(journal,
2061 SIT_JOURNAL, segno, 1);
2062 f2fs_bug_on(sbi, offset < 0);
2063 segno_in_journal(journal, offset) =
2064 cpu_to_le32(segno);
2065 seg_info_to_raw_sit(se,
2066 &sit_in_journal(journal, offset));
2067 } else {
2068 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2069 seg_info_to_raw_sit(se,
2070 &raw_sit->entries[sit_offset]);
2073 __clear_bit(segno, bitmap);
2074 sit_i->dirty_sentries--;
2075 ses->entry_cnt--;
2078 if (to_journal)
2079 up_write(&curseg->journal_rwsem);
2080 else
2081 f2fs_put_page(page, 1);
2083 f2fs_bug_on(sbi, ses->entry_cnt);
2084 release_sit_entry_set(ses);
2087 f2fs_bug_on(sbi, !list_empty(head));
2088 f2fs_bug_on(sbi, sit_i->dirty_sentries);
2089 out:
2090 if (cpc->reason == CP_DISCARD) {
2091 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2092 add_discard_addrs(sbi, cpc);
2094 mutex_unlock(&sit_i->sentry_lock);
2096 set_prefree_as_free_segments(sbi);
2099 static int build_sit_info(struct f2fs_sb_info *sbi)
2101 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2102 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2103 struct sit_info *sit_i;
2104 unsigned int sit_segs, start;
2105 char *src_bitmap, *dst_bitmap;
2106 unsigned int bitmap_size;
2108 /* allocate memory for SIT information */
2109 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2110 if (!sit_i)
2111 return -ENOMEM;
2113 SM_I(sbi)->sit_info = sit_i;
2115 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2116 sizeof(struct seg_entry), GFP_KERNEL);
2117 if (!sit_i->sentries)
2118 return -ENOMEM;
2120 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2121 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2122 if (!sit_i->dirty_sentries_bitmap)
2123 return -ENOMEM;
2125 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2126 sit_i->sentries[start].cur_valid_map
2127 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2128 sit_i->sentries[start].ckpt_valid_map
2129 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2130 sit_i->sentries[start].discard_map
2131 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2132 if (!sit_i->sentries[start].cur_valid_map ||
2133 !sit_i->sentries[start].ckpt_valid_map ||
2134 !sit_i->sentries[start].discard_map)
2135 return -ENOMEM;
2138 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2139 if (!sit_i->tmp_map)
2140 return -ENOMEM;
2142 if (sbi->segs_per_sec > 1) {
2143 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2144 sizeof(struct sec_entry), GFP_KERNEL);
2145 if (!sit_i->sec_entries)
2146 return -ENOMEM;
2149 /* get information related with SIT */
2150 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2152 /* setup SIT bitmap from ckeckpoint pack */
2153 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2154 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2156 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2157 if (!dst_bitmap)
2158 return -ENOMEM;
2160 /* init SIT information */
2161 sit_i->s_ops = &default_salloc_ops;
2163 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2164 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2165 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2166 sit_i->sit_bitmap = dst_bitmap;
2167 sit_i->bitmap_size = bitmap_size;
2168 sit_i->dirty_sentries = 0;
2169 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2170 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2171 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2172 mutex_init(&sit_i->sentry_lock);
2173 return 0;
2176 static int build_free_segmap(struct f2fs_sb_info *sbi)
2178 struct free_segmap_info *free_i;
2179 unsigned int bitmap_size, sec_bitmap_size;
2181 /* allocate memory for free segmap information */
2182 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2183 if (!free_i)
2184 return -ENOMEM;
2186 SM_I(sbi)->free_info = free_i;
2188 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2189 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2190 if (!free_i->free_segmap)
2191 return -ENOMEM;
2193 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2194 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2195 if (!free_i->free_secmap)
2196 return -ENOMEM;
2198 /* set all segments as dirty temporarily */
2199 memset(free_i->free_segmap, 0xff, bitmap_size);
2200 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2202 /* init free segmap information */
2203 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2204 free_i->free_segments = 0;
2205 free_i->free_sections = 0;
2206 spin_lock_init(&free_i->segmap_lock);
2207 return 0;
2210 static int build_curseg(struct f2fs_sb_info *sbi)
2212 struct curseg_info *array;
2213 int i;
2215 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2216 if (!array)
2217 return -ENOMEM;
2219 SM_I(sbi)->curseg_array = array;
2221 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2222 mutex_init(&array[i].curseg_mutex);
2223 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2224 if (!array[i].sum_blk)
2225 return -ENOMEM;
2226 init_rwsem(&array[i].journal_rwsem);
2227 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2228 GFP_KERNEL);
2229 if (!array[i].journal)
2230 return -ENOMEM;
2231 array[i].segno = NULL_SEGNO;
2232 array[i].next_blkoff = 0;
2234 return restore_curseg_summaries(sbi);
2237 static void build_sit_entries(struct f2fs_sb_info *sbi)
2239 struct sit_info *sit_i = SIT_I(sbi);
2240 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2241 struct f2fs_journal *journal = curseg->journal;
2242 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2243 unsigned int i, start, end;
2244 unsigned int readed, start_blk = 0;
2245 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
2247 do {
2248 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
2250 start = start_blk * sit_i->sents_per_block;
2251 end = (start_blk + readed) * sit_i->sents_per_block;
2253 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2254 struct seg_entry *se = &sit_i->sentries[start];
2255 struct f2fs_sit_block *sit_blk;
2256 struct f2fs_sit_entry sit;
2257 struct page *page;
2259 down_read(&curseg->journal_rwsem);
2260 for (i = 0; i < sits_in_cursum(journal); i++) {
2261 if (le32_to_cpu(segno_in_journal(journal, i))
2262 == start) {
2263 sit = sit_in_journal(journal, i);
2264 up_read(&curseg->journal_rwsem);
2265 goto got_it;
2268 up_read(&curseg->journal_rwsem);
2270 page = get_current_sit_page(sbi, start);
2271 sit_blk = (struct f2fs_sit_block *)page_address(page);
2272 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2273 f2fs_put_page(page, 1);
2274 got_it:
2275 check_block_count(sbi, start, &sit);
2276 seg_info_from_raw_sit(se, &sit);
2278 /* build discard map only one time */
2279 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2280 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2282 if (sbi->segs_per_sec > 1) {
2283 struct sec_entry *e = get_sec_entry(sbi, start);
2284 e->valid_blocks += se->valid_blocks;
2287 start_blk += readed;
2288 } while (start_blk < sit_blk_cnt);
2291 static void init_free_segmap(struct f2fs_sb_info *sbi)
2293 unsigned int start;
2294 int type;
2296 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2297 struct seg_entry *sentry = get_seg_entry(sbi, start);
2298 if (!sentry->valid_blocks)
2299 __set_free(sbi, start);
2302 /* set use the current segments */
2303 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2304 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2305 __set_test_and_inuse(sbi, curseg_t->segno);
2309 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2311 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2312 struct free_segmap_info *free_i = FREE_I(sbi);
2313 unsigned int segno = 0, offset = 0;
2314 unsigned short valid_blocks;
2316 while (1) {
2317 /* find dirty segment based on free segmap */
2318 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2319 if (segno >= MAIN_SEGS(sbi))
2320 break;
2321 offset = segno + 1;
2322 valid_blocks = get_valid_blocks(sbi, segno, 0);
2323 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2324 continue;
2325 if (valid_blocks > sbi->blocks_per_seg) {
2326 f2fs_bug_on(sbi, 1);
2327 continue;
2329 mutex_lock(&dirty_i->seglist_lock);
2330 __locate_dirty_segment(sbi, segno, DIRTY);
2331 mutex_unlock(&dirty_i->seglist_lock);
2335 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2337 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2338 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2340 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2341 if (!dirty_i->victim_secmap)
2342 return -ENOMEM;
2343 return 0;
2346 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2348 struct dirty_seglist_info *dirty_i;
2349 unsigned int bitmap_size, i;
2351 /* allocate memory for dirty segments list information */
2352 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2353 if (!dirty_i)
2354 return -ENOMEM;
2356 SM_I(sbi)->dirty_info = dirty_i;
2357 mutex_init(&dirty_i->seglist_lock);
2359 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2361 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2362 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2363 if (!dirty_i->dirty_segmap[i])
2364 return -ENOMEM;
2367 init_dirty_segmap(sbi);
2368 return init_victim_secmap(sbi);
2372 * Update min, max modified time for cost-benefit GC algorithm
2374 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2376 struct sit_info *sit_i = SIT_I(sbi);
2377 unsigned int segno;
2379 mutex_lock(&sit_i->sentry_lock);
2381 sit_i->min_mtime = LLONG_MAX;
2383 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2384 unsigned int i;
2385 unsigned long long mtime = 0;
2387 for (i = 0; i < sbi->segs_per_sec; i++)
2388 mtime += get_seg_entry(sbi, segno + i)->mtime;
2390 mtime = div_u64(mtime, sbi->segs_per_sec);
2392 if (sit_i->min_mtime > mtime)
2393 sit_i->min_mtime = mtime;
2395 sit_i->max_mtime = get_mtime(sbi);
2396 mutex_unlock(&sit_i->sentry_lock);
2399 int build_segment_manager(struct f2fs_sb_info *sbi)
2401 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2402 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2403 struct f2fs_sm_info *sm_info;
2404 int err;
2406 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2407 if (!sm_info)
2408 return -ENOMEM;
2410 /* init sm info */
2411 sbi->sm_info = sm_info;
2412 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2413 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2414 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2415 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2416 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2417 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2418 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2419 sm_info->rec_prefree_segments = sm_info->main_segments *
2420 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2421 if (sm_info->rec_prefree_segments > DEF_MAX_RECLAIM_PREFREE_SEGMENTS)
2422 sm_info->rec_prefree_segments = DEF_MAX_RECLAIM_PREFREE_SEGMENTS;
2424 if (!test_opt(sbi, LFS))
2425 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2426 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2427 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2429 INIT_LIST_HEAD(&sm_info->discard_list);
2430 sm_info->nr_discards = 0;
2431 sm_info->max_discards = 0;
2433 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2435 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2437 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2438 err = create_flush_cmd_control(sbi);
2439 if (err)
2440 return err;
2443 err = build_sit_info(sbi);
2444 if (err)
2445 return err;
2446 err = build_free_segmap(sbi);
2447 if (err)
2448 return err;
2449 err = build_curseg(sbi);
2450 if (err)
2451 return err;
2453 /* reinit free segmap based on SIT */
2454 build_sit_entries(sbi);
2456 init_free_segmap(sbi);
2457 err = build_dirty_segmap(sbi);
2458 if (err)
2459 return err;
2461 init_min_max_mtime(sbi);
2462 return 0;
2465 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2466 enum dirty_type dirty_type)
2468 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2470 mutex_lock(&dirty_i->seglist_lock);
2471 kvfree(dirty_i->dirty_segmap[dirty_type]);
2472 dirty_i->nr_dirty[dirty_type] = 0;
2473 mutex_unlock(&dirty_i->seglist_lock);
2476 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2478 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2479 kvfree(dirty_i->victim_secmap);
2482 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2484 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2485 int i;
2487 if (!dirty_i)
2488 return;
2490 /* discard pre-free/dirty segments list */
2491 for (i = 0; i < NR_DIRTY_TYPE; i++)
2492 discard_dirty_segmap(sbi, i);
2494 destroy_victim_secmap(sbi);
2495 SM_I(sbi)->dirty_info = NULL;
2496 kfree(dirty_i);
2499 static void destroy_curseg(struct f2fs_sb_info *sbi)
2501 struct curseg_info *array = SM_I(sbi)->curseg_array;
2502 int i;
2504 if (!array)
2505 return;
2506 SM_I(sbi)->curseg_array = NULL;
2507 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2508 kfree(array[i].sum_blk);
2509 kfree(array[i].journal);
2511 kfree(array);
2514 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2516 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2517 if (!free_i)
2518 return;
2519 SM_I(sbi)->free_info = NULL;
2520 kvfree(free_i->free_segmap);
2521 kvfree(free_i->free_secmap);
2522 kfree(free_i);
2525 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2527 struct sit_info *sit_i = SIT_I(sbi);
2528 unsigned int start;
2530 if (!sit_i)
2531 return;
2533 if (sit_i->sentries) {
2534 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2535 kfree(sit_i->sentries[start].cur_valid_map);
2536 kfree(sit_i->sentries[start].ckpt_valid_map);
2537 kfree(sit_i->sentries[start].discard_map);
2540 kfree(sit_i->tmp_map);
2542 kvfree(sit_i->sentries);
2543 kvfree(sit_i->sec_entries);
2544 kvfree(sit_i->dirty_sentries_bitmap);
2546 SM_I(sbi)->sit_info = NULL;
2547 kfree(sit_i->sit_bitmap);
2548 kfree(sit_i);
2551 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2553 struct f2fs_sm_info *sm_info = SM_I(sbi);
2555 if (!sm_info)
2556 return;
2557 destroy_flush_cmd_control(sbi);
2558 destroy_dirty_segmap(sbi);
2559 destroy_curseg(sbi);
2560 destroy_free_segmap(sbi);
2561 destroy_sit_info(sbi);
2562 sbi->sm_info = NULL;
2563 kfree(sm_info);
2566 int __init create_segment_manager_caches(void)
2568 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2569 sizeof(struct discard_entry));
2570 if (!discard_entry_slab)
2571 goto fail;
2573 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2574 sizeof(struct sit_entry_set));
2575 if (!sit_entry_set_slab)
2576 goto destory_discard_entry;
2578 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2579 sizeof(struct inmem_pages));
2580 if (!inmem_entry_slab)
2581 goto destroy_sit_entry_set;
2582 return 0;
2584 destroy_sit_entry_set:
2585 kmem_cache_destroy(sit_entry_set_slab);
2586 destory_discard_entry:
2587 kmem_cache_destroy(discard_entry_slab);
2588 fail:
2589 return -ENOMEM;
2592 void destroy_segment_manager_caches(void)
2594 kmem_cache_destroy(sit_entry_set_slab);
2595 kmem_cache_destroy(discard_entry_slab);
2596 kmem_cache_destroy(inmem_entry_slab);