s390/time: add leap seconds to initial system time
[linux/fpc-iii.git] / fs / f2fs / segment.c
blob2e6f537a0e7df7df4b449df05f222ff4bc5a90b4
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(F2FS_I(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 .rw = WRITE_SYNC | REQ_PRIO,
261 .encrypted_page = NULL,
263 bool submit_bio = false;
264 int err = 0;
266 list_for_each_entry_safe(cur, tmp, &fi->inmem_pages, list) {
267 struct page *page = cur->page;
269 lock_page(page);
270 if (page->mapping == inode->i_mapping) {
271 trace_f2fs_commit_inmem_page(page, INMEM);
273 set_page_dirty(page);
274 f2fs_wait_on_page_writeback(page, DATA, true);
275 if (clear_page_dirty_for_io(page))
276 inode_dec_dirty_pages(inode);
278 fio.page = page;
279 err = do_write_data_page(&fio);
280 if (err) {
281 unlock_page(page);
282 break;
285 /* record old blkaddr for revoking */
286 cur->old_addr = fio.old_blkaddr;
288 clear_cold_data(page);
289 submit_bio = true;
291 unlock_page(page);
292 list_move_tail(&cur->list, revoke_list);
295 if (submit_bio)
296 f2fs_submit_merged_bio_cond(sbi, inode, NULL, 0, DATA, WRITE);
298 if (!err)
299 __revoke_inmem_pages(inode, revoke_list, false, false);
301 return err;
304 int commit_inmem_pages(struct inode *inode)
306 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
307 struct f2fs_inode_info *fi = F2FS_I(inode);
308 struct list_head revoke_list;
309 int err;
311 INIT_LIST_HEAD(&revoke_list);
312 f2fs_balance_fs(sbi, true);
313 f2fs_lock_op(sbi);
315 mutex_lock(&fi->inmem_lock);
316 err = __commit_inmem_pages(inode, &revoke_list);
317 if (err) {
318 int ret;
320 * try to revoke all committed pages, but still we could fail
321 * due to no memory or other reason, if that happened, EAGAIN
322 * will be returned, which means in such case, transaction is
323 * already not integrity, caller should use journal to do the
324 * recovery or rewrite & commit last transaction. For other
325 * error number, revoking was done by filesystem itself.
327 ret = __revoke_inmem_pages(inode, &revoke_list, false, true);
328 if (ret)
329 err = ret;
331 /* drop all uncommitted pages */
332 __revoke_inmem_pages(inode, &fi->inmem_pages, true, false);
334 mutex_unlock(&fi->inmem_lock);
336 f2fs_unlock_op(sbi);
337 return err;
341 * This function balances dirty node and dentry pages.
342 * In addition, it controls garbage collection.
344 void f2fs_balance_fs(struct f2fs_sb_info *sbi, bool need)
346 if (!need)
347 return;
349 * We should do GC or end up with checkpoint, if there are so many dirty
350 * dir/node pages without enough free segments.
352 if (has_not_enough_free_secs(sbi, 0)) {
353 mutex_lock(&sbi->gc_mutex);
354 f2fs_gc(sbi, false);
358 void f2fs_balance_fs_bg(struct f2fs_sb_info *sbi)
360 /* try to shrink extent cache when there is no enough memory */
361 if (!available_free_memory(sbi, EXTENT_CACHE))
362 f2fs_shrink_extent_tree(sbi, EXTENT_CACHE_SHRINK_NUMBER);
364 /* check the # of cached NAT entries */
365 if (!available_free_memory(sbi, NAT_ENTRIES))
366 try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK);
368 if (!available_free_memory(sbi, FREE_NIDS))
369 try_to_free_nids(sbi, NAT_ENTRY_PER_BLOCK * FREE_NID_PAGES);
371 /* checkpoint is the only way to shrink partial cached entries */
372 if (!available_free_memory(sbi, NAT_ENTRIES) ||
373 !available_free_memory(sbi, INO_ENTRIES) ||
374 excess_prefree_segs(sbi) ||
375 excess_dirty_nats(sbi) ||
376 (is_idle(sbi) && f2fs_time_over(sbi, CP_TIME))) {
377 if (test_opt(sbi, DATA_FLUSH)) {
378 struct blk_plug plug;
380 blk_start_plug(&plug);
381 sync_dirty_inodes(sbi, FILE_INODE);
382 blk_finish_plug(&plug);
384 f2fs_sync_fs(sbi->sb, true);
385 stat_inc_bg_cp_count(sbi->stat_info);
389 static int issue_flush_thread(void *data)
391 struct f2fs_sb_info *sbi = data;
392 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
393 wait_queue_head_t *q = &fcc->flush_wait_queue;
394 repeat:
395 if (kthread_should_stop())
396 return 0;
398 if (!llist_empty(&fcc->issue_list)) {
399 struct bio *bio;
400 struct flush_cmd *cmd, *next;
401 int ret;
403 bio = f2fs_bio_alloc(0);
405 fcc->dispatch_list = llist_del_all(&fcc->issue_list);
406 fcc->dispatch_list = llist_reverse_order(fcc->dispatch_list);
408 bio->bi_bdev = sbi->sb->s_bdev;
409 ret = submit_bio_wait(WRITE_FLUSH, bio);
411 llist_for_each_entry_safe(cmd, next,
412 fcc->dispatch_list, llnode) {
413 cmd->ret = ret;
414 complete(&cmd->wait);
416 bio_put(bio);
417 fcc->dispatch_list = NULL;
420 wait_event_interruptible(*q,
421 kthread_should_stop() || !llist_empty(&fcc->issue_list));
422 goto repeat;
425 int f2fs_issue_flush(struct f2fs_sb_info *sbi)
427 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
428 struct flush_cmd cmd;
430 trace_f2fs_issue_flush(sbi->sb, test_opt(sbi, NOBARRIER),
431 test_opt(sbi, FLUSH_MERGE));
433 if (test_opt(sbi, NOBARRIER))
434 return 0;
436 if (!test_opt(sbi, FLUSH_MERGE)) {
437 struct bio *bio = f2fs_bio_alloc(0);
438 int ret;
440 bio->bi_bdev = sbi->sb->s_bdev;
441 ret = submit_bio_wait(WRITE_FLUSH, bio);
442 bio_put(bio);
443 return ret;
446 init_completion(&cmd.wait);
448 llist_add(&cmd.llnode, &fcc->issue_list);
450 if (!fcc->dispatch_list)
451 wake_up(&fcc->flush_wait_queue);
453 wait_for_completion(&cmd.wait);
455 return cmd.ret;
458 int create_flush_cmd_control(struct f2fs_sb_info *sbi)
460 dev_t dev = sbi->sb->s_bdev->bd_dev;
461 struct flush_cmd_control *fcc;
462 int err = 0;
464 fcc = kzalloc(sizeof(struct flush_cmd_control), GFP_KERNEL);
465 if (!fcc)
466 return -ENOMEM;
467 init_waitqueue_head(&fcc->flush_wait_queue);
468 init_llist_head(&fcc->issue_list);
469 SM_I(sbi)->cmd_control_info = fcc;
470 fcc->f2fs_issue_flush = kthread_run(issue_flush_thread, sbi,
471 "f2fs_flush-%u:%u", MAJOR(dev), MINOR(dev));
472 if (IS_ERR(fcc->f2fs_issue_flush)) {
473 err = PTR_ERR(fcc->f2fs_issue_flush);
474 kfree(fcc);
475 SM_I(sbi)->cmd_control_info = NULL;
476 return err;
479 return err;
482 void destroy_flush_cmd_control(struct f2fs_sb_info *sbi)
484 struct flush_cmd_control *fcc = SM_I(sbi)->cmd_control_info;
486 if (fcc && fcc->f2fs_issue_flush)
487 kthread_stop(fcc->f2fs_issue_flush);
488 kfree(fcc);
489 SM_I(sbi)->cmd_control_info = NULL;
492 static void __locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
493 enum dirty_type dirty_type)
495 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
497 /* need not be added */
498 if (IS_CURSEG(sbi, segno))
499 return;
501 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[dirty_type]))
502 dirty_i->nr_dirty[dirty_type]++;
504 if (dirty_type == DIRTY) {
505 struct seg_entry *sentry = get_seg_entry(sbi, segno);
506 enum dirty_type t = sentry->type;
508 if (unlikely(t >= DIRTY)) {
509 f2fs_bug_on(sbi, 1);
510 return;
512 if (!test_and_set_bit(segno, dirty_i->dirty_segmap[t]))
513 dirty_i->nr_dirty[t]++;
517 static void __remove_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno,
518 enum dirty_type dirty_type)
520 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
522 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[dirty_type]))
523 dirty_i->nr_dirty[dirty_type]--;
525 if (dirty_type == DIRTY) {
526 struct seg_entry *sentry = get_seg_entry(sbi, segno);
527 enum dirty_type t = sentry->type;
529 if (test_and_clear_bit(segno, dirty_i->dirty_segmap[t]))
530 dirty_i->nr_dirty[t]--;
532 if (get_valid_blocks(sbi, segno, sbi->segs_per_sec) == 0)
533 clear_bit(GET_SECNO(sbi, segno),
534 dirty_i->victim_secmap);
539 * Should not occur error such as -ENOMEM.
540 * Adding dirty entry into seglist is not critical operation.
541 * If a given segment is one of current working segments, it won't be added.
543 static void locate_dirty_segment(struct f2fs_sb_info *sbi, unsigned int segno)
545 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
546 unsigned short valid_blocks;
548 if (segno == NULL_SEGNO || IS_CURSEG(sbi, segno))
549 return;
551 mutex_lock(&dirty_i->seglist_lock);
553 valid_blocks = get_valid_blocks(sbi, segno, 0);
555 if (valid_blocks == 0) {
556 __locate_dirty_segment(sbi, segno, PRE);
557 __remove_dirty_segment(sbi, segno, DIRTY);
558 } else if (valid_blocks < sbi->blocks_per_seg) {
559 __locate_dirty_segment(sbi, segno, DIRTY);
560 } else {
561 /* Recovery routine with SSR needs this */
562 __remove_dirty_segment(sbi, segno, DIRTY);
565 mutex_unlock(&dirty_i->seglist_lock);
568 static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
569 block_t blkstart, block_t blklen)
571 sector_t start = SECTOR_FROM_BLOCK(blkstart);
572 sector_t len = SECTOR_FROM_BLOCK(blklen);
573 struct seg_entry *se;
574 unsigned int offset;
575 block_t i;
577 for (i = blkstart; i < blkstart + blklen; i++) {
578 se = get_seg_entry(sbi, GET_SEGNO(sbi, i));
579 offset = GET_BLKOFF_FROM_SEG0(sbi, i);
581 if (!f2fs_test_and_set_bit(offset, se->discard_map))
582 sbi->discard_blks--;
584 trace_f2fs_issue_discard(sbi->sb, blkstart, blklen);
585 return blkdev_issue_discard(sbi->sb->s_bdev, start, len, GFP_NOFS, 0);
588 bool discard_next_dnode(struct f2fs_sb_info *sbi, block_t blkaddr)
590 int err = -EOPNOTSUPP;
592 if (test_opt(sbi, DISCARD)) {
593 struct seg_entry *se = get_seg_entry(sbi,
594 GET_SEGNO(sbi, blkaddr));
595 unsigned int offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
597 if (f2fs_test_bit(offset, se->discard_map))
598 return false;
600 err = f2fs_issue_discard(sbi, blkaddr, 1);
603 if (err) {
604 update_meta_page(sbi, NULL, blkaddr);
605 return true;
607 return false;
610 static void __add_discard_entry(struct f2fs_sb_info *sbi,
611 struct cp_control *cpc, struct seg_entry *se,
612 unsigned int start, unsigned int end)
614 struct list_head *head = &SM_I(sbi)->discard_list;
615 struct discard_entry *new, *last;
617 if (!list_empty(head)) {
618 last = list_last_entry(head, struct discard_entry, list);
619 if (START_BLOCK(sbi, cpc->trim_start) + start ==
620 last->blkaddr + last->len) {
621 last->len += end - start;
622 goto done;
626 new = f2fs_kmem_cache_alloc(discard_entry_slab, GFP_NOFS);
627 INIT_LIST_HEAD(&new->list);
628 new->blkaddr = START_BLOCK(sbi, cpc->trim_start) + start;
629 new->len = end - start;
630 list_add_tail(&new->list, head);
631 done:
632 SM_I(sbi)->nr_discards += end - start;
635 static void add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc)
637 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
638 int max_blocks = sbi->blocks_per_seg;
639 struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
640 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
641 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
642 unsigned long *discard_map = (unsigned long *)se->discard_map;
643 unsigned long *dmap = SIT_I(sbi)->tmp_map;
644 unsigned int start = 0, end = -1;
645 bool force = (cpc->reason == CP_DISCARD);
646 int i;
648 if (se->valid_blocks == max_blocks)
649 return;
651 if (!force) {
652 if (!test_opt(sbi, DISCARD) || !se->valid_blocks ||
653 SM_I(sbi)->nr_discards >= SM_I(sbi)->max_discards)
654 return;
657 /* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
658 for (i = 0; i < entries; i++)
659 dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
660 (cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
662 while (force || SM_I(sbi)->nr_discards <= SM_I(sbi)->max_discards) {
663 start = __find_rev_next_bit(dmap, max_blocks, end + 1);
664 if (start >= max_blocks)
665 break;
667 end = __find_rev_next_zero_bit(dmap, max_blocks, start + 1);
668 __add_discard_entry(sbi, cpc, se, start, end);
672 void release_discard_addrs(struct f2fs_sb_info *sbi)
674 struct list_head *head = &(SM_I(sbi)->discard_list);
675 struct discard_entry *entry, *this;
677 /* drop caches */
678 list_for_each_entry_safe(entry, this, head, list) {
679 list_del(&entry->list);
680 kmem_cache_free(discard_entry_slab, entry);
685 * Should call clear_prefree_segments after checkpoint is done.
687 static void set_prefree_as_free_segments(struct f2fs_sb_info *sbi)
689 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
690 unsigned int segno;
692 mutex_lock(&dirty_i->seglist_lock);
693 for_each_set_bit(segno, dirty_i->dirty_segmap[PRE], MAIN_SEGS(sbi))
694 __set_test_and_free(sbi, segno);
695 mutex_unlock(&dirty_i->seglist_lock);
698 void clear_prefree_segments(struct f2fs_sb_info *sbi, struct cp_control *cpc)
700 struct list_head *head = &(SM_I(sbi)->discard_list);
701 struct discard_entry *entry, *this;
702 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
703 unsigned long *prefree_map = dirty_i->dirty_segmap[PRE];
704 unsigned int start = 0, end = -1;
706 mutex_lock(&dirty_i->seglist_lock);
708 while (1) {
709 int i;
710 start = find_next_bit(prefree_map, MAIN_SEGS(sbi), end + 1);
711 if (start >= MAIN_SEGS(sbi))
712 break;
713 end = find_next_zero_bit(prefree_map, MAIN_SEGS(sbi),
714 start + 1);
716 for (i = start; i < end; i++)
717 clear_bit(i, prefree_map);
719 dirty_i->nr_dirty[PRE] -= end - start;
721 if (!test_opt(sbi, DISCARD))
722 continue;
724 f2fs_issue_discard(sbi, START_BLOCK(sbi, start),
725 (end - start) << sbi->log_blocks_per_seg);
727 mutex_unlock(&dirty_i->seglist_lock);
729 /* send small discards */
730 list_for_each_entry_safe(entry, this, head, list) {
731 if (cpc->reason == CP_DISCARD && entry->len < cpc->trim_minlen)
732 goto skip;
733 f2fs_issue_discard(sbi, entry->blkaddr, entry->len);
734 cpc->trimmed += entry->len;
735 skip:
736 list_del(&entry->list);
737 SM_I(sbi)->nr_discards -= entry->len;
738 kmem_cache_free(discard_entry_slab, entry);
742 static bool __mark_sit_entry_dirty(struct f2fs_sb_info *sbi, unsigned int segno)
744 struct sit_info *sit_i = SIT_I(sbi);
746 if (!__test_and_set_bit(segno, sit_i->dirty_sentries_bitmap)) {
747 sit_i->dirty_sentries++;
748 return false;
751 return true;
754 static void __set_sit_entry_type(struct f2fs_sb_info *sbi, int type,
755 unsigned int segno, int modified)
757 struct seg_entry *se = get_seg_entry(sbi, segno);
758 se->type = type;
759 if (modified)
760 __mark_sit_entry_dirty(sbi, segno);
763 static void update_sit_entry(struct f2fs_sb_info *sbi, block_t blkaddr, int del)
765 struct seg_entry *se;
766 unsigned int segno, offset;
767 long int new_vblocks;
769 segno = GET_SEGNO(sbi, blkaddr);
771 se = get_seg_entry(sbi, segno);
772 new_vblocks = se->valid_blocks + del;
773 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
775 f2fs_bug_on(sbi, (new_vblocks >> (sizeof(unsigned short) << 3) ||
776 (new_vblocks > sbi->blocks_per_seg)));
778 se->valid_blocks = new_vblocks;
779 se->mtime = get_mtime(sbi);
780 SIT_I(sbi)->max_mtime = se->mtime;
782 /* Update valid block bitmap */
783 if (del > 0) {
784 if (f2fs_test_and_set_bit(offset, se->cur_valid_map))
785 f2fs_bug_on(sbi, 1);
786 if (!f2fs_test_and_set_bit(offset, se->discard_map))
787 sbi->discard_blks--;
788 } else {
789 if (!f2fs_test_and_clear_bit(offset, se->cur_valid_map))
790 f2fs_bug_on(sbi, 1);
791 if (f2fs_test_and_clear_bit(offset, se->discard_map))
792 sbi->discard_blks++;
794 if (!f2fs_test_bit(offset, se->ckpt_valid_map))
795 se->ckpt_valid_blocks += del;
797 __mark_sit_entry_dirty(sbi, segno);
799 /* update total number of valid blocks to be written in ckpt area */
800 SIT_I(sbi)->written_valid_blocks += del;
802 if (sbi->segs_per_sec > 1)
803 get_sec_entry(sbi, segno)->valid_blocks += del;
806 void refresh_sit_entry(struct f2fs_sb_info *sbi, block_t old, block_t new)
808 update_sit_entry(sbi, new, 1);
809 if (GET_SEGNO(sbi, old) != NULL_SEGNO)
810 update_sit_entry(sbi, old, -1);
812 locate_dirty_segment(sbi, GET_SEGNO(sbi, old));
813 locate_dirty_segment(sbi, GET_SEGNO(sbi, new));
816 void invalidate_blocks(struct f2fs_sb_info *sbi, block_t addr)
818 unsigned int segno = GET_SEGNO(sbi, addr);
819 struct sit_info *sit_i = SIT_I(sbi);
821 f2fs_bug_on(sbi, addr == NULL_ADDR);
822 if (addr == NEW_ADDR)
823 return;
825 /* add it into sit main buffer */
826 mutex_lock(&sit_i->sentry_lock);
828 update_sit_entry(sbi, addr, -1);
830 /* add it into dirty seglist */
831 locate_dirty_segment(sbi, segno);
833 mutex_unlock(&sit_i->sentry_lock);
836 bool is_checkpointed_data(struct f2fs_sb_info *sbi, block_t blkaddr)
838 struct sit_info *sit_i = SIT_I(sbi);
839 unsigned int segno, offset;
840 struct seg_entry *se;
841 bool is_cp = false;
843 if (blkaddr == NEW_ADDR || blkaddr == NULL_ADDR)
844 return true;
846 mutex_lock(&sit_i->sentry_lock);
848 segno = GET_SEGNO(sbi, blkaddr);
849 se = get_seg_entry(sbi, segno);
850 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
852 if (f2fs_test_bit(offset, se->ckpt_valid_map))
853 is_cp = true;
855 mutex_unlock(&sit_i->sentry_lock);
857 return is_cp;
861 * This function should be resided under the curseg_mutex lock
863 static void __add_sum_entry(struct f2fs_sb_info *sbi, int type,
864 struct f2fs_summary *sum)
866 struct curseg_info *curseg = CURSEG_I(sbi, type);
867 void *addr = curseg->sum_blk;
868 addr += curseg->next_blkoff * sizeof(struct f2fs_summary);
869 memcpy(addr, sum, sizeof(struct f2fs_summary));
873 * Calculate the number of current summary pages for writing
875 int npages_for_summary_flush(struct f2fs_sb_info *sbi, bool for_ra)
877 int valid_sum_count = 0;
878 int i, sum_in_page;
880 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
881 if (sbi->ckpt->alloc_type[i] == SSR)
882 valid_sum_count += sbi->blocks_per_seg;
883 else {
884 if (for_ra)
885 valid_sum_count += le16_to_cpu(
886 F2FS_CKPT(sbi)->cur_data_blkoff[i]);
887 else
888 valid_sum_count += curseg_blkoff(sbi, i);
892 sum_in_page = (PAGE_SIZE - 2 * SUM_JOURNAL_SIZE -
893 SUM_FOOTER_SIZE) / SUMMARY_SIZE;
894 if (valid_sum_count <= sum_in_page)
895 return 1;
896 else if ((valid_sum_count - sum_in_page) <=
897 (PAGE_SIZE - SUM_FOOTER_SIZE) / SUMMARY_SIZE)
898 return 2;
899 return 3;
903 * Caller should put this summary page
905 struct page *get_sum_page(struct f2fs_sb_info *sbi, unsigned int segno)
907 return get_meta_page(sbi, GET_SUM_BLOCK(sbi, segno));
910 void update_meta_page(struct f2fs_sb_info *sbi, void *src, block_t blk_addr)
912 struct page *page = grab_meta_page(sbi, blk_addr);
913 void *dst = page_address(page);
915 if (src)
916 memcpy(dst, src, PAGE_SIZE);
917 else
918 memset(dst, 0, PAGE_SIZE);
919 set_page_dirty(page);
920 f2fs_put_page(page, 1);
923 static void write_sum_page(struct f2fs_sb_info *sbi,
924 struct f2fs_summary_block *sum_blk, block_t blk_addr)
926 update_meta_page(sbi, (void *)sum_blk, blk_addr);
929 static void write_current_sum_page(struct f2fs_sb_info *sbi,
930 int type, block_t blk_addr)
932 struct curseg_info *curseg = CURSEG_I(sbi, type);
933 struct page *page = grab_meta_page(sbi, blk_addr);
934 struct f2fs_summary_block *src = curseg->sum_blk;
935 struct f2fs_summary_block *dst;
937 dst = (struct f2fs_summary_block *)page_address(page);
939 mutex_lock(&curseg->curseg_mutex);
941 down_read(&curseg->journal_rwsem);
942 memcpy(&dst->journal, curseg->journal, SUM_JOURNAL_SIZE);
943 up_read(&curseg->journal_rwsem);
945 memcpy(dst->entries, src->entries, SUM_ENTRY_SIZE);
946 memcpy(&dst->footer, &src->footer, SUM_FOOTER_SIZE);
948 mutex_unlock(&curseg->curseg_mutex);
950 set_page_dirty(page);
951 f2fs_put_page(page, 1);
954 static int is_next_segment_free(struct f2fs_sb_info *sbi, int type)
956 struct curseg_info *curseg = CURSEG_I(sbi, type);
957 unsigned int segno = curseg->segno + 1;
958 struct free_segmap_info *free_i = FREE_I(sbi);
960 if (segno < MAIN_SEGS(sbi) && segno % sbi->segs_per_sec)
961 return !test_bit(segno, free_i->free_segmap);
962 return 0;
966 * Find a new segment from the free segments bitmap to right order
967 * This function should be returned with success, otherwise BUG
969 static void get_new_segment(struct f2fs_sb_info *sbi,
970 unsigned int *newseg, bool new_sec, int dir)
972 struct free_segmap_info *free_i = FREE_I(sbi);
973 unsigned int segno, secno, zoneno;
974 unsigned int total_zones = MAIN_SECS(sbi) / sbi->secs_per_zone;
975 unsigned int hint = *newseg / sbi->segs_per_sec;
976 unsigned int old_zoneno = GET_ZONENO_FROM_SEGNO(sbi, *newseg);
977 unsigned int left_start = hint;
978 bool init = true;
979 int go_left = 0;
980 int i;
982 spin_lock(&free_i->segmap_lock);
984 if (!new_sec && ((*newseg + 1) % sbi->segs_per_sec)) {
985 segno = find_next_zero_bit(free_i->free_segmap,
986 (hint + 1) * sbi->segs_per_sec, *newseg + 1);
987 if (segno < (hint + 1) * sbi->segs_per_sec)
988 goto got_it;
990 find_other_zone:
991 secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
992 if (secno >= MAIN_SECS(sbi)) {
993 if (dir == ALLOC_RIGHT) {
994 secno = find_next_zero_bit(free_i->free_secmap,
995 MAIN_SECS(sbi), 0);
996 f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
997 } else {
998 go_left = 1;
999 left_start = hint - 1;
1002 if (go_left == 0)
1003 goto skip_left;
1005 while (test_bit(left_start, free_i->free_secmap)) {
1006 if (left_start > 0) {
1007 left_start--;
1008 continue;
1010 left_start = find_next_zero_bit(free_i->free_secmap,
1011 MAIN_SECS(sbi), 0);
1012 f2fs_bug_on(sbi, left_start >= MAIN_SECS(sbi));
1013 break;
1015 secno = left_start;
1016 skip_left:
1017 hint = secno;
1018 segno = secno * sbi->segs_per_sec;
1019 zoneno = secno / sbi->secs_per_zone;
1021 /* give up on finding another zone */
1022 if (!init)
1023 goto got_it;
1024 if (sbi->secs_per_zone == 1)
1025 goto got_it;
1026 if (zoneno == old_zoneno)
1027 goto got_it;
1028 if (dir == ALLOC_LEFT) {
1029 if (!go_left && zoneno + 1 >= total_zones)
1030 goto got_it;
1031 if (go_left && zoneno == 0)
1032 goto got_it;
1034 for (i = 0; i < NR_CURSEG_TYPE; i++)
1035 if (CURSEG_I(sbi, i)->zone == zoneno)
1036 break;
1038 if (i < NR_CURSEG_TYPE) {
1039 /* zone is in user, try another */
1040 if (go_left)
1041 hint = zoneno * sbi->secs_per_zone - 1;
1042 else if (zoneno + 1 >= total_zones)
1043 hint = 0;
1044 else
1045 hint = (zoneno + 1) * sbi->secs_per_zone;
1046 init = false;
1047 goto find_other_zone;
1049 got_it:
1050 /* set it as dirty segment in free segmap */
1051 f2fs_bug_on(sbi, test_bit(segno, free_i->free_segmap));
1052 __set_inuse(sbi, segno);
1053 *newseg = segno;
1054 spin_unlock(&free_i->segmap_lock);
1057 static void reset_curseg(struct f2fs_sb_info *sbi, int type, int modified)
1059 struct curseg_info *curseg = CURSEG_I(sbi, type);
1060 struct summary_footer *sum_footer;
1062 curseg->segno = curseg->next_segno;
1063 curseg->zone = GET_ZONENO_FROM_SEGNO(sbi, curseg->segno);
1064 curseg->next_blkoff = 0;
1065 curseg->next_segno = NULL_SEGNO;
1067 sum_footer = &(curseg->sum_blk->footer);
1068 memset(sum_footer, 0, sizeof(struct summary_footer));
1069 if (IS_DATASEG(type))
1070 SET_SUM_TYPE(sum_footer, SUM_TYPE_DATA);
1071 if (IS_NODESEG(type))
1072 SET_SUM_TYPE(sum_footer, SUM_TYPE_NODE);
1073 __set_sit_entry_type(sbi, type, curseg->segno, modified);
1077 * Allocate a current working segment.
1078 * This function always allocates a free segment in LFS manner.
1080 static void new_curseg(struct f2fs_sb_info *sbi, int type, bool new_sec)
1082 struct curseg_info *curseg = CURSEG_I(sbi, type);
1083 unsigned int segno = curseg->segno;
1084 int dir = ALLOC_LEFT;
1086 write_sum_page(sbi, curseg->sum_blk,
1087 GET_SUM_BLOCK(sbi, segno));
1088 if (type == CURSEG_WARM_DATA || type == CURSEG_COLD_DATA)
1089 dir = ALLOC_RIGHT;
1091 if (test_opt(sbi, NOHEAP))
1092 dir = ALLOC_RIGHT;
1094 get_new_segment(sbi, &segno, new_sec, dir);
1095 curseg->next_segno = segno;
1096 reset_curseg(sbi, type, 1);
1097 curseg->alloc_type = LFS;
1100 static void __next_free_blkoff(struct f2fs_sb_info *sbi,
1101 struct curseg_info *seg, block_t start)
1103 struct seg_entry *se = get_seg_entry(sbi, seg->segno);
1104 int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
1105 unsigned long *target_map = SIT_I(sbi)->tmp_map;
1106 unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
1107 unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
1108 int i, pos;
1110 for (i = 0; i < entries; i++)
1111 target_map[i] = ckpt_map[i] | cur_map[i];
1113 pos = __find_rev_next_zero_bit(target_map, sbi->blocks_per_seg, start);
1115 seg->next_blkoff = pos;
1119 * If a segment is written by LFS manner, next block offset is just obtained
1120 * by increasing the current block offset. However, if a segment is written by
1121 * SSR manner, next block offset obtained by calling __next_free_blkoff
1123 static void __refresh_next_blkoff(struct f2fs_sb_info *sbi,
1124 struct curseg_info *seg)
1126 if (seg->alloc_type == SSR)
1127 __next_free_blkoff(sbi, seg, seg->next_blkoff + 1);
1128 else
1129 seg->next_blkoff++;
1133 * This function always allocates a used segment(from dirty seglist) by SSR
1134 * manner, so it should recover the existing segment information of valid blocks
1136 static void change_curseg(struct f2fs_sb_info *sbi, int type, bool reuse)
1138 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
1139 struct curseg_info *curseg = CURSEG_I(sbi, type);
1140 unsigned int new_segno = curseg->next_segno;
1141 struct f2fs_summary_block *sum_node;
1142 struct page *sum_page;
1144 write_sum_page(sbi, curseg->sum_blk,
1145 GET_SUM_BLOCK(sbi, curseg->segno));
1146 __set_test_and_inuse(sbi, new_segno);
1148 mutex_lock(&dirty_i->seglist_lock);
1149 __remove_dirty_segment(sbi, new_segno, PRE);
1150 __remove_dirty_segment(sbi, new_segno, DIRTY);
1151 mutex_unlock(&dirty_i->seglist_lock);
1153 reset_curseg(sbi, type, 1);
1154 curseg->alloc_type = SSR;
1155 __next_free_blkoff(sbi, curseg, 0);
1157 if (reuse) {
1158 sum_page = get_sum_page(sbi, new_segno);
1159 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
1160 memcpy(curseg->sum_blk, sum_node, SUM_ENTRY_SIZE);
1161 f2fs_put_page(sum_page, 1);
1165 static int get_ssr_segment(struct f2fs_sb_info *sbi, int type)
1167 struct curseg_info *curseg = CURSEG_I(sbi, type);
1168 const struct victim_selection *v_ops = DIRTY_I(sbi)->v_ops;
1170 if (IS_NODESEG(type) || !has_not_enough_free_secs(sbi, 0))
1171 return v_ops->get_victim(sbi,
1172 &(curseg)->next_segno, BG_GC, type, SSR);
1174 /* For data segments, let's do SSR more intensively */
1175 for (; type >= CURSEG_HOT_DATA; type--)
1176 if (v_ops->get_victim(sbi, &(curseg)->next_segno,
1177 BG_GC, type, SSR))
1178 return 1;
1179 return 0;
1183 * flush out current segment and replace it with new segment
1184 * This function should be returned with success, otherwise BUG
1186 static void allocate_segment_by_default(struct f2fs_sb_info *sbi,
1187 int type, bool force)
1189 struct curseg_info *curseg = CURSEG_I(sbi, type);
1191 if (force)
1192 new_curseg(sbi, type, true);
1193 else if (type == CURSEG_WARM_NODE)
1194 new_curseg(sbi, type, false);
1195 else if (curseg->alloc_type == LFS && is_next_segment_free(sbi, type))
1196 new_curseg(sbi, type, false);
1197 else if (need_SSR(sbi) && get_ssr_segment(sbi, type))
1198 change_curseg(sbi, type, true);
1199 else
1200 new_curseg(sbi, type, false);
1202 stat_inc_seg_type(sbi, curseg);
1205 static void __allocate_new_segments(struct f2fs_sb_info *sbi, int type)
1207 struct curseg_info *curseg = CURSEG_I(sbi, type);
1208 unsigned int old_segno;
1210 old_segno = curseg->segno;
1211 SIT_I(sbi)->s_ops->allocate_segment(sbi, type, true);
1212 locate_dirty_segment(sbi, old_segno);
1215 void allocate_new_segments(struct f2fs_sb_info *sbi)
1217 int i;
1219 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++)
1220 __allocate_new_segments(sbi, i);
1223 static const struct segment_allocation default_salloc_ops = {
1224 .allocate_segment = allocate_segment_by_default,
1227 int f2fs_trim_fs(struct f2fs_sb_info *sbi, struct fstrim_range *range)
1229 __u64 start = F2FS_BYTES_TO_BLK(range->start);
1230 __u64 end = start + F2FS_BYTES_TO_BLK(range->len) - 1;
1231 unsigned int start_segno, end_segno;
1232 struct cp_control cpc;
1233 int err = 0;
1235 if (start >= MAX_BLKADDR(sbi) || range->len < sbi->blocksize)
1236 return -EINVAL;
1238 cpc.trimmed = 0;
1239 if (end <= MAIN_BLKADDR(sbi))
1240 goto out;
1242 /* start/end segment number in main_area */
1243 start_segno = (start <= MAIN_BLKADDR(sbi)) ? 0 : GET_SEGNO(sbi, start);
1244 end_segno = (end >= MAX_BLKADDR(sbi)) ? MAIN_SEGS(sbi) - 1 :
1245 GET_SEGNO(sbi, end);
1246 cpc.reason = CP_DISCARD;
1247 cpc.trim_minlen = max_t(__u64, 1, F2FS_BYTES_TO_BLK(range->minlen));
1249 /* do checkpoint to issue discard commands safely */
1250 for (; start_segno <= end_segno; start_segno = cpc.trim_end + 1) {
1251 cpc.trim_start = start_segno;
1253 if (sbi->discard_blks == 0)
1254 break;
1255 else if (sbi->discard_blks < BATCHED_TRIM_BLOCKS(sbi))
1256 cpc.trim_end = end_segno;
1257 else
1258 cpc.trim_end = min_t(unsigned int,
1259 rounddown(start_segno +
1260 BATCHED_TRIM_SEGMENTS(sbi),
1261 sbi->segs_per_sec) - 1, end_segno);
1263 mutex_lock(&sbi->gc_mutex);
1264 err = write_checkpoint(sbi, &cpc);
1265 mutex_unlock(&sbi->gc_mutex);
1267 out:
1268 range->len = F2FS_BLK_TO_BYTES(cpc.trimmed);
1269 return err;
1272 static bool __has_curseg_space(struct f2fs_sb_info *sbi, int type)
1274 struct curseg_info *curseg = CURSEG_I(sbi, type);
1275 if (curseg->next_blkoff < sbi->blocks_per_seg)
1276 return true;
1277 return false;
1280 static int __get_segment_type_2(struct page *page, enum page_type p_type)
1282 if (p_type == DATA)
1283 return CURSEG_HOT_DATA;
1284 else
1285 return CURSEG_HOT_NODE;
1288 static int __get_segment_type_4(struct page *page, enum page_type p_type)
1290 if (p_type == DATA) {
1291 struct inode *inode = page->mapping->host;
1293 if (S_ISDIR(inode->i_mode))
1294 return CURSEG_HOT_DATA;
1295 else
1296 return CURSEG_COLD_DATA;
1297 } else {
1298 if (IS_DNODE(page) && is_cold_node(page))
1299 return CURSEG_WARM_NODE;
1300 else
1301 return CURSEG_COLD_NODE;
1305 static int __get_segment_type_6(struct page *page, enum page_type p_type)
1307 if (p_type == DATA) {
1308 struct inode *inode = page->mapping->host;
1310 if (S_ISDIR(inode->i_mode))
1311 return CURSEG_HOT_DATA;
1312 else if (is_cold_data(page) || file_is_cold(inode))
1313 return CURSEG_COLD_DATA;
1314 else
1315 return CURSEG_WARM_DATA;
1316 } else {
1317 if (IS_DNODE(page))
1318 return is_cold_node(page) ? CURSEG_WARM_NODE :
1319 CURSEG_HOT_NODE;
1320 else
1321 return CURSEG_COLD_NODE;
1325 static int __get_segment_type(struct page *page, enum page_type p_type)
1327 switch (F2FS_P_SB(page)->active_logs) {
1328 case 2:
1329 return __get_segment_type_2(page, p_type);
1330 case 4:
1331 return __get_segment_type_4(page, p_type);
1333 /* NR_CURSEG_TYPE(6) logs by default */
1334 f2fs_bug_on(F2FS_P_SB(page),
1335 F2FS_P_SB(page)->active_logs != NR_CURSEG_TYPE);
1336 return __get_segment_type_6(page, p_type);
1339 void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page,
1340 block_t old_blkaddr, block_t *new_blkaddr,
1341 struct f2fs_summary *sum, int type)
1343 struct sit_info *sit_i = SIT_I(sbi);
1344 struct curseg_info *curseg;
1345 bool direct_io = (type == CURSEG_DIRECT_IO);
1347 type = direct_io ? CURSEG_WARM_DATA : type;
1349 curseg = CURSEG_I(sbi, type);
1351 mutex_lock(&curseg->curseg_mutex);
1352 mutex_lock(&sit_i->sentry_lock);
1354 /* direct_io'ed data is aligned to the segment for better performance */
1355 if (direct_io && curseg->next_blkoff &&
1356 !has_not_enough_free_secs(sbi, 0))
1357 __allocate_new_segments(sbi, type);
1359 *new_blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
1362 * __add_sum_entry should be resided under the curseg_mutex
1363 * because, this function updates a summary entry in the
1364 * current summary block.
1366 __add_sum_entry(sbi, type, sum);
1368 __refresh_next_blkoff(sbi, curseg);
1370 stat_inc_block_count(sbi, curseg);
1372 if (!__has_curseg_space(sbi, type))
1373 sit_i->s_ops->allocate_segment(sbi, type, false);
1375 * SIT information should be updated before segment allocation,
1376 * since SSR needs latest valid block information.
1378 refresh_sit_entry(sbi, old_blkaddr, *new_blkaddr);
1380 mutex_unlock(&sit_i->sentry_lock);
1382 if (page && IS_NODESEG(type))
1383 fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg));
1385 mutex_unlock(&curseg->curseg_mutex);
1388 static void do_write_page(struct f2fs_summary *sum, struct f2fs_io_info *fio)
1390 int type = __get_segment_type(fio->page, fio->type);
1392 allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr,
1393 &fio->new_blkaddr, sum, type);
1395 /* writeout dirty page into bdev */
1396 f2fs_submit_page_mbio(fio);
1399 void write_meta_page(struct f2fs_sb_info *sbi, struct page *page)
1401 struct f2fs_io_info fio = {
1402 .sbi = sbi,
1403 .type = META,
1404 .rw = WRITE_SYNC | REQ_META | REQ_PRIO,
1405 .old_blkaddr = page->index,
1406 .new_blkaddr = page->index,
1407 .page = page,
1408 .encrypted_page = NULL,
1411 if (unlikely(page->index >= MAIN_BLKADDR(sbi)))
1412 fio.rw &= ~REQ_META;
1414 set_page_writeback(page);
1415 f2fs_submit_page_mbio(&fio);
1418 void write_node_page(unsigned int nid, struct f2fs_io_info *fio)
1420 struct f2fs_summary sum;
1422 set_summary(&sum, nid, 0, 0);
1423 do_write_page(&sum, fio);
1426 void write_data_page(struct dnode_of_data *dn, struct f2fs_io_info *fio)
1428 struct f2fs_sb_info *sbi = fio->sbi;
1429 struct f2fs_summary sum;
1430 struct node_info ni;
1432 f2fs_bug_on(sbi, dn->data_blkaddr == NULL_ADDR);
1433 get_node_info(sbi, dn->nid, &ni);
1434 set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1435 do_write_page(&sum, fio);
1436 f2fs_update_data_blkaddr(dn, fio->new_blkaddr);
1439 void rewrite_data_page(struct f2fs_io_info *fio)
1441 fio->new_blkaddr = fio->old_blkaddr;
1442 stat_inc_inplace_blocks(fio->sbi);
1443 f2fs_submit_page_mbio(fio);
1446 void __f2fs_replace_block(struct f2fs_sb_info *sbi, struct f2fs_summary *sum,
1447 block_t old_blkaddr, block_t new_blkaddr,
1448 bool recover_curseg, bool recover_newaddr)
1450 struct sit_info *sit_i = SIT_I(sbi);
1451 struct curseg_info *curseg;
1452 unsigned int segno, old_cursegno;
1453 struct seg_entry *se;
1454 int type;
1455 unsigned short old_blkoff;
1457 segno = GET_SEGNO(sbi, new_blkaddr);
1458 se = get_seg_entry(sbi, segno);
1459 type = se->type;
1461 if (!recover_curseg) {
1462 /* for recovery flow */
1463 if (se->valid_blocks == 0 && !IS_CURSEG(sbi, segno)) {
1464 if (old_blkaddr == NULL_ADDR)
1465 type = CURSEG_COLD_DATA;
1466 else
1467 type = CURSEG_WARM_DATA;
1469 } else {
1470 if (!IS_CURSEG(sbi, segno))
1471 type = CURSEG_WARM_DATA;
1474 curseg = CURSEG_I(sbi, type);
1476 mutex_lock(&curseg->curseg_mutex);
1477 mutex_lock(&sit_i->sentry_lock);
1479 old_cursegno = curseg->segno;
1480 old_blkoff = curseg->next_blkoff;
1482 /* change the current segment */
1483 if (segno != curseg->segno) {
1484 curseg->next_segno = segno;
1485 change_curseg(sbi, type, true);
1488 curseg->next_blkoff = GET_BLKOFF_FROM_SEG0(sbi, new_blkaddr);
1489 __add_sum_entry(sbi, type, sum);
1491 if (!recover_curseg || recover_newaddr)
1492 update_sit_entry(sbi, new_blkaddr, 1);
1493 if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1494 update_sit_entry(sbi, old_blkaddr, -1);
1496 locate_dirty_segment(sbi, GET_SEGNO(sbi, old_blkaddr));
1497 locate_dirty_segment(sbi, GET_SEGNO(sbi, new_blkaddr));
1499 locate_dirty_segment(sbi, old_cursegno);
1501 if (recover_curseg) {
1502 if (old_cursegno != curseg->segno) {
1503 curseg->next_segno = old_cursegno;
1504 change_curseg(sbi, type, true);
1506 curseg->next_blkoff = old_blkoff;
1509 mutex_unlock(&sit_i->sentry_lock);
1510 mutex_unlock(&curseg->curseg_mutex);
1513 void f2fs_replace_block(struct f2fs_sb_info *sbi, struct dnode_of_data *dn,
1514 block_t old_addr, block_t new_addr,
1515 unsigned char version, bool recover_curseg,
1516 bool recover_newaddr)
1518 struct f2fs_summary sum;
1520 set_summary(&sum, dn->nid, dn->ofs_in_node, version);
1522 __f2fs_replace_block(sbi, &sum, old_addr, new_addr,
1523 recover_curseg, recover_newaddr);
1525 f2fs_update_data_blkaddr(dn, new_addr);
1528 void f2fs_wait_on_page_writeback(struct page *page,
1529 enum page_type type, bool ordered)
1531 if (PageWriteback(page)) {
1532 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1534 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, type, WRITE);
1535 if (ordered)
1536 wait_on_page_writeback(page);
1537 else
1538 wait_for_stable_page(page);
1542 void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi,
1543 block_t blkaddr)
1545 struct page *cpage;
1547 if (blkaddr == NEW_ADDR)
1548 return;
1550 f2fs_bug_on(sbi, blkaddr == NULL_ADDR);
1552 cpage = find_lock_page(META_MAPPING(sbi), blkaddr);
1553 if (cpage) {
1554 f2fs_wait_on_page_writeback(cpage, DATA, true);
1555 f2fs_put_page(cpage, 1);
1559 static int read_compacted_summaries(struct f2fs_sb_info *sbi)
1561 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1562 struct curseg_info *seg_i;
1563 unsigned char *kaddr;
1564 struct page *page;
1565 block_t start;
1566 int i, j, offset;
1568 start = start_sum_block(sbi);
1570 page = get_meta_page(sbi, start++);
1571 kaddr = (unsigned char *)page_address(page);
1573 /* Step 1: restore nat cache */
1574 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1575 memcpy(seg_i->journal, kaddr, SUM_JOURNAL_SIZE);
1577 /* Step 2: restore sit cache */
1578 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1579 memcpy(seg_i->journal, kaddr + SUM_JOURNAL_SIZE, SUM_JOURNAL_SIZE);
1580 offset = 2 * SUM_JOURNAL_SIZE;
1582 /* Step 3: restore summary entries */
1583 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1584 unsigned short blk_off;
1585 unsigned int segno;
1587 seg_i = CURSEG_I(sbi, i);
1588 segno = le32_to_cpu(ckpt->cur_data_segno[i]);
1589 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[i]);
1590 seg_i->next_segno = segno;
1591 reset_curseg(sbi, i, 0);
1592 seg_i->alloc_type = ckpt->alloc_type[i];
1593 seg_i->next_blkoff = blk_off;
1595 if (seg_i->alloc_type == SSR)
1596 blk_off = sbi->blocks_per_seg;
1598 for (j = 0; j < blk_off; j++) {
1599 struct f2fs_summary *s;
1600 s = (struct f2fs_summary *)(kaddr + offset);
1601 seg_i->sum_blk->entries[j] = *s;
1602 offset += SUMMARY_SIZE;
1603 if (offset + SUMMARY_SIZE <= PAGE_SIZE -
1604 SUM_FOOTER_SIZE)
1605 continue;
1607 f2fs_put_page(page, 1);
1608 page = NULL;
1610 page = get_meta_page(sbi, start++);
1611 kaddr = (unsigned char *)page_address(page);
1612 offset = 0;
1615 f2fs_put_page(page, 1);
1616 return 0;
1619 static int read_normal_summaries(struct f2fs_sb_info *sbi, int type)
1621 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1622 struct f2fs_summary_block *sum;
1623 struct curseg_info *curseg;
1624 struct page *new;
1625 unsigned short blk_off;
1626 unsigned int segno = 0;
1627 block_t blk_addr = 0;
1629 /* get segment number and block addr */
1630 if (IS_DATASEG(type)) {
1631 segno = le32_to_cpu(ckpt->cur_data_segno[type]);
1632 blk_off = le16_to_cpu(ckpt->cur_data_blkoff[type -
1633 CURSEG_HOT_DATA]);
1634 if (__exist_node_summaries(sbi))
1635 blk_addr = sum_blk_addr(sbi, NR_CURSEG_TYPE, type);
1636 else
1637 blk_addr = sum_blk_addr(sbi, NR_CURSEG_DATA_TYPE, type);
1638 } else {
1639 segno = le32_to_cpu(ckpt->cur_node_segno[type -
1640 CURSEG_HOT_NODE]);
1641 blk_off = le16_to_cpu(ckpt->cur_node_blkoff[type -
1642 CURSEG_HOT_NODE]);
1643 if (__exist_node_summaries(sbi))
1644 blk_addr = sum_blk_addr(sbi, NR_CURSEG_NODE_TYPE,
1645 type - CURSEG_HOT_NODE);
1646 else
1647 blk_addr = GET_SUM_BLOCK(sbi, segno);
1650 new = get_meta_page(sbi, blk_addr);
1651 sum = (struct f2fs_summary_block *)page_address(new);
1653 if (IS_NODESEG(type)) {
1654 if (__exist_node_summaries(sbi)) {
1655 struct f2fs_summary *ns = &sum->entries[0];
1656 int i;
1657 for (i = 0; i < sbi->blocks_per_seg; i++, ns++) {
1658 ns->version = 0;
1659 ns->ofs_in_node = 0;
1661 } else {
1662 int err;
1664 err = restore_node_summary(sbi, segno, sum);
1665 if (err) {
1666 f2fs_put_page(new, 1);
1667 return err;
1672 /* set uncompleted segment to curseg */
1673 curseg = CURSEG_I(sbi, type);
1674 mutex_lock(&curseg->curseg_mutex);
1676 /* update journal info */
1677 down_write(&curseg->journal_rwsem);
1678 memcpy(curseg->journal, &sum->journal, SUM_JOURNAL_SIZE);
1679 up_write(&curseg->journal_rwsem);
1681 memcpy(curseg->sum_blk->entries, sum->entries, SUM_ENTRY_SIZE);
1682 memcpy(&curseg->sum_blk->footer, &sum->footer, SUM_FOOTER_SIZE);
1683 curseg->next_segno = segno;
1684 reset_curseg(sbi, type, 0);
1685 curseg->alloc_type = ckpt->alloc_type[type];
1686 curseg->next_blkoff = blk_off;
1687 mutex_unlock(&curseg->curseg_mutex);
1688 f2fs_put_page(new, 1);
1689 return 0;
1692 static int restore_curseg_summaries(struct f2fs_sb_info *sbi)
1694 int type = CURSEG_HOT_DATA;
1695 int err;
1697 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG)) {
1698 int npages = npages_for_summary_flush(sbi, true);
1700 if (npages >= 2)
1701 ra_meta_pages(sbi, start_sum_block(sbi), npages,
1702 META_CP, true);
1704 /* restore for compacted data summary */
1705 if (read_compacted_summaries(sbi))
1706 return -EINVAL;
1707 type = CURSEG_HOT_NODE;
1710 if (__exist_node_summaries(sbi))
1711 ra_meta_pages(sbi, sum_blk_addr(sbi, NR_CURSEG_TYPE, type),
1712 NR_CURSEG_TYPE - type, META_CP, true);
1714 for (; type <= CURSEG_COLD_NODE; type++) {
1715 err = read_normal_summaries(sbi, type);
1716 if (err)
1717 return err;
1720 return 0;
1723 static void write_compacted_summaries(struct f2fs_sb_info *sbi, block_t blkaddr)
1725 struct page *page;
1726 unsigned char *kaddr;
1727 struct f2fs_summary *summary;
1728 struct curseg_info *seg_i;
1729 int written_size = 0;
1730 int i, j;
1732 page = grab_meta_page(sbi, blkaddr++);
1733 kaddr = (unsigned char *)page_address(page);
1735 /* Step 1: write nat cache */
1736 seg_i = CURSEG_I(sbi, CURSEG_HOT_DATA);
1737 memcpy(kaddr, seg_i->journal, SUM_JOURNAL_SIZE);
1738 written_size += SUM_JOURNAL_SIZE;
1740 /* Step 2: write sit cache */
1741 seg_i = CURSEG_I(sbi, CURSEG_COLD_DATA);
1742 memcpy(kaddr + written_size, seg_i->journal, SUM_JOURNAL_SIZE);
1743 written_size += SUM_JOURNAL_SIZE;
1745 /* Step 3: write summary entries */
1746 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
1747 unsigned short blkoff;
1748 seg_i = CURSEG_I(sbi, i);
1749 if (sbi->ckpt->alloc_type[i] == SSR)
1750 blkoff = sbi->blocks_per_seg;
1751 else
1752 blkoff = curseg_blkoff(sbi, i);
1754 for (j = 0; j < blkoff; j++) {
1755 if (!page) {
1756 page = grab_meta_page(sbi, blkaddr++);
1757 kaddr = (unsigned char *)page_address(page);
1758 written_size = 0;
1760 summary = (struct f2fs_summary *)(kaddr + written_size);
1761 *summary = seg_i->sum_blk->entries[j];
1762 written_size += SUMMARY_SIZE;
1764 if (written_size + SUMMARY_SIZE <= PAGE_SIZE -
1765 SUM_FOOTER_SIZE)
1766 continue;
1768 set_page_dirty(page);
1769 f2fs_put_page(page, 1);
1770 page = NULL;
1773 if (page) {
1774 set_page_dirty(page);
1775 f2fs_put_page(page, 1);
1779 static void write_normal_summaries(struct f2fs_sb_info *sbi,
1780 block_t blkaddr, int type)
1782 int i, end;
1783 if (IS_DATASEG(type))
1784 end = type + NR_CURSEG_DATA_TYPE;
1785 else
1786 end = type + NR_CURSEG_NODE_TYPE;
1788 for (i = type; i < end; i++)
1789 write_current_sum_page(sbi, i, blkaddr + (i - type));
1792 void write_data_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1794 if (is_set_ckpt_flags(F2FS_CKPT(sbi), CP_COMPACT_SUM_FLAG))
1795 write_compacted_summaries(sbi, start_blk);
1796 else
1797 write_normal_summaries(sbi, start_blk, CURSEG_HOT_DATA);
1800 void write_node_summaries(struct f2fs_sb_info *sbi, block_t start_blk)
1802 write_normal_summaries(sbi, start_blk, CURSEG_HOT_NODE);
1805 int lookup_journal_in_cursum(struct f2fs_journal *journal, int type,
1806 unsigned int val, int alloc)
1808 int i;
1810 if (type == NAT_JOURNAL) {
1811 for (i = 0; i < nats_in_cursum(journal); i++) {
1812 if (le32_to_cpu(nid_in_journal(journal, i)) == val)
1813 return i;
1815 if (alloc && __has_cursum_space(journal, 1, NAT_JOURNAL))
1816 return update_nats_in_cursum(journal, 1);
1817 } else if (type == SIT_JOURNAL) {
1818 for (i = 0; i < sits_in_cursum(journal); i++)
1819 if (le32_to_cpu(segno_in_journal(journal, i)) == val)
1820 return i;
1821 if (alloc && __has_cursum_space(journal, 1, SIT_JOURNAL))
1822 return update_sits_in_cursum(journal, 1);
1824 return -1;
1827 static struct page *get_current_sit_page(struct f2fs_sb_info *sbi,
1828 unsigned int segno)
1830 return get_meta_page(sbi, current_sit_addr(sbi, segno));
1833 static struct page *get_next_sit_page(struct f2fs_sb_info *sbi,
1834 unsigned int start)
1836 struct sit_info *sit_i = SIT_I(sbi);
1837 struct page *src_page, *dst_page;
1838 pgoff_t src_off, dst_off;
1839 void *src_addr, *dst_addr;
1841 src_off = current_sit_addr(sbi, start);
1842 dst_off = next_sit_addr(sbi, src_off);
1844 /* get current sit block page without lock */
1845 src_page = get_meta_page(sbi, src_off);
1846 dst_page = grab_meta_page(sbi, dst_off);
1847 f2fs_bug_on(sbi, PageDirty(src_page));
1849 src_addr = page_address(src_page);
1850 dst_addr = page_address(dst_page);
1851 memcpy(dst_addr, src_addr, PAGE_SIZE);
1853 set_page_dirty(dst_page);
1854 f2fs_put_page(src_page, 1);
1856 set_to_next_sit(sit_i, start);
1858 return dst_page;
1861 static struct sit_entry_set *grab_sit_entry_set(void)
1863 struct sit_entry_set *ses =
1864 f2fs_kmem_cache_alloc(sit_entry_set_slab, GFP_NOFS);
1866 ses->entry_cnt = 0;
1867 INIT_LIST_HEAD(&ses->set_list);
1868 return ses;
1871 static void release_sit_entry_set(struct sit_entry_set *ses)
1873 list_del(&ses->set_list);
1874 kmem_cache_free(sit_entry_set_slab, ses);
1877 static void adjust_sit_entry_set(struct sit_entry_set *ses,
1878 struct list_head *head)
1880 struct sit_entry_set *next = ses;
1882 if (list_is_last(&ses->set_list, head))
1883 return;
1885 list_for_each_entry_continue(next, head, set_list)
1886 if (ses->entry_cnt <= next->entry_cnt)
1887 break;
1889 list_move_tail(&ses->set_list, &next->set_list);
1892 static void add_sit_entry(unsigned int segno, struct list_head *head)
1894 struct sit_entry_set *ses;
1895 unsigned int start_segno = START_SEGNO(segno);
1897 list_for_each_entry(ses, head, set_list) {
1898 if (ses->start_segno == start_segno) {
1899 ses->entry_cnt++;
1900 adjust_sit_entry_set(ses, head);
1901 return;
1905 ses = grab_sit_entry_set();
1907 ses->start_segno = start_segno;
1908 ses->entry_cnt++;
1909 list_add(&ses->set_list, head);
1912 static void add_sits_in_set(struct f2fs_sb_info *sbi)
1914 struct f2fs_sm_info *sm_info = SM_I(sbi);
1915 struct list_head *set_list = &sm_info->sit_entry_set;
1916 unsigned long *bitmap = SIT_I(sbi)->dirty_sentries_bitmap;
1917 unsigned int segno;
1919 for_each_set_bit(segno, bitmap, MAIN_SEGS(sbi))
1920 add_sit_entry(segno, set_list);
1923 static void remove_sits_in_journal(struct f2fs_sb_info *sbi)
1925 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1926 struct f2fs_journal *journal = curseg->journal;
1927 int i;
1929 down_write(&curseg->journal_rwsem);
1930 for (i = 0; i < sits_in_cursum(journal); i++) {
1931 unsigned int segno;
1932 bool dirtied;
1934 segno = le32_to_cpu(segno_in_journal(journal, i));
1935 dirtied = __mark_sit_entry_dirty(sbi, segno);
1937 if (!dirtied)
1938 add_sit_entry(segno, &SM_I(sbi)->sit_entry_set);
1940 update_sits_in_cursum(journal, -i);
1941 up_write(&curseg->journal_rwsem);
1945 * CP calls this function, which flushes SIT entries including sit_journal,
1946 * and moves prefree segs to free segs.
1948 void flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1950 struct sit_info *sit_i = SIT_I(sbi);
1951 unsigned long *bitmap = sit_i->dirty_sentries_bitmap;
1952 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
1953 struct f2fs_journal *journal = curseg->journal;
1954 struct sit_entry_set *ses, *tmp;
1955 struct list_head *head = &SM_I(sbi)->sit_entry_set;
1956 bool to_journal = true;
1957 struct seg_entry *se;
1959 mutex_lock(&sit_i->sentry_lock);
1961 if (!sit_i->dirty_sentries)
1962 goto out;
1965 * add and account sit entries of dirty bitmap in sit entry
1966 * set temporarily
1968 add_sits_in_set(sbi);
1971 * if there are no enough space in journal to store dirty sit
1972 * entries, remove all entries from journal and add and account
1973 * them in sit entry set.
1975 if (!__has_cursum_space(journal, sit_i->dirty_sentries, SIT_JOURNAL))
1976 remove_sits_in_journal(sbi);
1979 * there are two steps to flush sit entries:
1980 * #1, flush sit entries to journal in current cold data summary block.
1981 * #2, flush sit entries to sit page.
1983 list_for_each_entry_safe(ses, tmp, head, set_list) {
1984 struct page *page = NULL;
1985 struct f2fs_sit_block *raw_sit = NULL;
1986 unsigned int start_segno = ses->start_segno;
1987 unsigned int end = min(start_segno + SIT_ENTRY_PER_BLOCK,
1988 (unsigned long)MAIN_SEGS(sbi));
1989 unsigned int segno = start_segno;
1991 if (to_journal &&
1992 !__has_cursum_space(journal, ses->entry_cnt, SIT_JOURNAL))
1993 to_journal = false;
1995 if (to_journal) {
1996 down_write(&curseg->journal_rwsem);
1997 } else {
1998 page = get_next_sit_page(sbi, start_segno);
1999 raw_sit = page_address(page);
2002 /* flush dirty sit entries in region of current sit set */
2003 for_each_set_bit_from(segno, bitmap, end) {
2004 int offset, sit_offset;
2006 se = get_seg_entry(sbi, segno);
2008 /* add discard candidates */
2009 if (cpc->reason != CP_DISCARD) {
2010 cpc->trim_start = segno;
2011 add_discard_addrs(sbi, cpc);
2014 if (to_journal) {
2015 offset = lookup_journal_in_cursum(journal,
2016 SIT_JOURNAL, segno, 1);
2017 f2fs_bug_on(sbi, offset < 0);
2018 segno_in_journal(journal, offset) =
2019 cpu_to_le32(segno);
2020 seg_info_to_raw_sit(se,
2021 &sit_in_journal(journal, offset));
2022 } else {
2023 sit_offset = SIT_ENTRY_OFFSET(sit_i, segno);
2024 seg_info_to_raw_sit(se,
2025 &raw_sit->entries[sit_offset]);
2028 __clear_bit(segno, bitmap);
2029 sit_i->dirty_sentries--;
2030 ses->entry_cnt--;
2033 if (to_journal)
2034 up_write(&curseg->journal_rwsem);
2035 else
2036 f2fs_put_page(page, 1);
2038 f2fs_bug_on(sbi, ses->entry_cnt);
2039 release_sit_entry_set(ses);
2042 f2fs_bug_on(sbi, !list_empty(head));
2043 f2fs_bug_on(sbi, sit_i->dirty_sentries);
2044 out:
2045 if (cpc->reason == CP_DISCARD) {
2046 for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
2047 add_discard_addrs(sbi, cpc);
2049 mutex_unlock(&sit_i->sentry_lock);
2051 set_prefree_as_free_segments(sbi);
2054 static int build_sit_info(struct f2fs_sb_info *sbi)
2056 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2057 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2058 struct sit_info *sit_i;
2059 unsigned int sit_segs, start;
2060 char *src_bitmap, *dst_bitmap;
2061 unsigned int bitmap_size;
2063 /* allocate memory for SIT information */
2064 sit_i = kzalloc(sizeof(struct sit_info), GFP_KERNEL);
2065 if (!sit_i)
2066 return -ENOMEM;
2068 SM_I(sbi)->sit_info = sit_i;
2070 sit_i->sentries = f2fs_kvzalloc(MAIN_SEGS(sbi) *
2071 sizeof(struct seg_entry), GFP_KERNEL);
2072 if (!sit_i->sentries)
2073 return -ENOMEM;
2075 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2076 sit_i->dirty_sentries_bitmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2077 if (!sit_i->dirty_sentries_bitmap)
2078 return -ENOMEM;
2080 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2081 sit_i->sentries[start].cur_valid_map
2082 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2083 sit_i->sentries[start].ckpt_valid_map
2084 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2085 sit_i->sentries[start].discard_map
2086 = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2087 if (!sit_i->sentries[start].cur_valid_map ||
2088 !sit_i->sentries[start].ckpt_valid_map ||
2089 !sit_i->sentries[start].discard_map)
2090 return -ENOMEM;
2093 sit_i->tmp_map = kzalloc(SIT_VBLOCK_MAP_SIZE, GFP_KERNEL);
2094 if (!sit_i->tmp_map)
2095 return -ENOMEM;
2097 if (sbi->segs_per_sec > 1) {
2098 sit_i->sec_entries = f2fs_kvzalloc(MAIN_SECS(sbi) *
2099 sizeof(struct sec_entry), GFP_KERNEL);
2100 if (!sit_i->sec_entries)
2101 return -ENOMEM;
2104 /* get information related with SIT */
2105 sit_segs = le32_to_cpu(raw_super->segment_count_sit) >> 1;
2107 /* setup SIT bitmap from ckeckpoint pack */
2108 bitmap_size = __bitmap_size(sbi, SIT_BITMAP);
2109 src_bitmap = __bitmap_ptr(sbi, SIT_BITMAP);
2111 dst_bitmap = kmemdup(src_bitmap, bitmap_size, GFP_KERNEL);
2112 if (!dst_bitmap)
2113 return -ENOMEM;
2115 /* init SIT information */
2116 sit_i->s_ops = &default_salloc_ops;
2118 sit_i->sit_base_addr = le32_to_cpu(raw_super->sit_blkaddr);
2119 sit_i->sit_blocks = sit_segs << sbi->log_blocks_per_seg;
2120 sit_i->written_valid_blocks = le64_to_cpu(ckpt->valid_block_count);
2121 sit_i->sit_bitmap = dst_bitmap;
2122 sit_i->bitmap_size = bitmap_size;
2123 sit_i->dirty_sentries = 0;
2124 sit_i->sents_per_block = SIT_ENTRY_PER_BLOCK;
2125 sit_i->elapsed_time = le64_to_cpu(sbi->ckpt->elapsed_time);
2126 sit_i->mounted_time = CURRENT_TIME_SEC.tv_sec;
2127 mutex_init(&sit_i->sentry_lock);
2128 return 0;
2131 static int build_free_segmap(struct f2fs_sb_info *sbi)
2133 struct free_segmap_info *free_i;
2134 unsigned int bitmap_size, sec_bitmap_size;
2136 /* allocate memory for free segmap information */
2137 free_i = kzalloc(sizeof(struct free_segmap_info), GFP_KERNEL);
2138 if (!free_i)
2139 return -ENOMEM;
2141 SM_I(sbi)->free_info = free_i;
2143 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2144 free_i->free_segmap = f2fs_kvmalloc(bitmap_size, GFP_KERNEL);
2145 if (!free_i->free_segmap)
2146 return -ENOMEM;
2148 sec_bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2149 free_i->free_secmap = f2fs_kvmalloc(sec_bitmap_size, GFP_KERNEL);
2150 if (!free_i->free_secmap)
2151 return -ENOMEM;
2153 /* set all segments as dirty temporarily */
2154 memset(free_i->free_segmap, 0xff, bitmap_size);
2155 memset(free_i->free_secmap, 0xff, sec_bitmap_size);
2157 /* init free segmap information */
2158 free_i->start_segno = GET_SEGNO_FROM_SEG0(sbi, MAIN_BLKADDR(sbi));
2159 free_i->free_segments = 0;
2160 free_i->free_sections = 0;
2161 spin_lock_init(&free_i->segmap_lock);
2162 return 0;
2165 static int build_curseg(struct f2fs_sb_info *sbi)
2167 struct curseg_info *array;
2168 int i;
2170 array = kcalloc(NR_CURSEG_TYPE, sizeof(*array), GFP_KERNEL);
2171 if (!array)
2172 return -ENOMEM;
2174 SM_I(sbi)->curseg_array = array;
2176 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2177 mutex_init(&array[i].curseg_mutex);
2178 array[i].sum_blk = kzalloc(PAGE_SIZE, GFP_KERNEL);
2179 if (!array[i].sum_blk)
2180 return -ENOMEM;
2181 init_rwsem(&array[i].journal_rwsem);
2182 array[i].journal = kzalloc(sizeof(struct f2fs_journal),
2183 GFP_KERNEL);
2184 if (!array[i].journal)
2185 return -ENOMEM;
2186 array[i].segno = NULL_SEGNO;
2187 array[i].next_blkoff = 0;
2189 return restore_curseg_summaries(sbi);
2192 static void build_sit_entries(struct f2fs_sb_info *sbi)
2194 struct sit_info *sit_i = SIT_I(sbi);
2195 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_COLD_DATA);
2196 struct f2fs_journal *journal = curseg->journal;
2197 int sit_blk_cnt = SIT_BLK_CNT(sbi);
2198 unsigned int i, start, end;
2199 unsigned int readed, start_blk = 0;
2200 int nrpages = MAX_BIO_BLOCKS(sbi) * 8;
2202 do {
2203 readed = ra_meta_pages(sbi, start_blk, nrpages, META_SIT, true);
2205 start = start_blk * sit_i->sents_per_block;
2206 end = (start_blk + readed) * sit_i->sents_per_block;
2208 for (; start < end && start < MAIN_SEGS(sbi); start++) {
2209 struct seg_entry *se = &sit_i->sentries[start];
2210 struct f2fs_sit_block *sit_blk;
2211 struct f2fs_sit_entry sit;
2212 struct page *page;
2214 down_read(&curseg->journal_rwsem);
2215 for (i = 0; i < sits_in_cursum(journal); i++) {
2216 if (le32_to_cpu(segno_in_journal(journal, i))
2217 == start) {
2218 sit = sit_in_journal(journal, i);
2219 up_read(&curseg->journal_rwsem);
2220 goto got_it;
2223 up_read(&curseg->journal_rwsem);
2225 page = get_current_sit_page(sbi, start);
2226 sit_blk = (struct f2fs_sit_block *)page_address(page);
2227 sit = sit_blk->entries[SIT_ENTRY_OFFSET(sit_i, start)];
2228 f2fs_put_page(page, 1);
2229 got_it:
2230 check_block_count(sbi, start, &sit);
2231 seg_info_from_raw_sit(se, &sit);
2233 /* build discard map only one time */
2234 memcpy(se->discard_map, se->cur_valid_map, SIT_VBLOCK_MAP_SIZE);
2235 sbi->discard_blks += sbi->blocks_per_seg - se->valid_blocks;
2237 if (sbi->segs_per_sec > 1) {
2238 struct sec_entry *e = get_sec_entry(sbi, start);
2239 e->valid_blocks += se->valid_blocks;
2242 start_blk += readed;
2243 } while (start_blk < sit_blk_cnt);
2246 static void init_free_segmap(struct f2fs_sb_info *sbi)
2248 unsigned int start;
2249 int type;
2251 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2252 struct seg_entry *sentry = get_seg_entry(sbi, start);
2253 if (!sentry->valid_blocks)
2254 __set_free(sbi, start);
2257 /* set use the current segments */
2258 for (type = CURSEG_HOT_DATA; type <= CURSEG_COLD_NODE; type++) {
2259 struct curseg_info *curseg_t = CURSEG_I(sbi, type);
2260 __set_test_and_inuse(sbi, curseg_t->segno);
2264 static void init_dirty_segmap(struct f2fs_sb_info *sbi)
2266 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2267 struct free_segmap_info *free_i = FREE_I(sbi);
2268 unsigned int segno = 0, offset = 0;
2269 unsigned short valid_blocks;
2271 while (1) {
2272 /* find dirty segment based on free segmap */
2273 segno = find_next_inuse(free_i, MAIN_SEGS(sbi), offset);
2274 if (segno >= MAIN_SEGS(sbi))
2275 break;
2276 offset = segno + 1;
2277 valid_blocks = get_valid_blocks(sbi, segno, 0);
2278 if (valid_blocks == sbi->blocks_per_seg || !valid_blocks)
2279 continue;
2280 if (valid_blocks > sbi->blocks_per_seg) {
2281 f2fs_bug_on(sbi, 1);
2282 continue;
2284 mutex_lock(&dirty_i->seglist_lock);
2285 __locate_dirty_segment(sbi, segno, DIRTY);
2286 mutex_unlock(&dirty_i->seglist_lock);
2290 static int init_victim_secmap(struct f2fs_sb_info *sbi)
2292 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2293 unsigned int bitmap_size = f2fs_bitmap_size(MAIN_SECS(sbi));
2295 dirty_i->victim_secmap = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2296 if (!dirty_i->victim_secmap)
2297 return -ENOMEM;
2298 return 0;
2301 static int build_dirty_segmap(struct f2fs_sb_info *sbi)
2303 struct dirty_seglist_info *dirty_i;
2304 unsigned int bitmap_size, i;
2306 /* allocate memory for dirty segments list information */
2307 dirty_i = kzalloc(sizeof(struct dirty_seglist_info), GFP_KERNEL);
2308 if (!dirty_i)
2309 return -ENOMEM;
2311 SM_I(sbi)->dirty_info = dirty_i;
2312 mutex_init(&dirty_i->seglist_lock);
2314 bitmap_size = f2fs_bitmap_size(MAIN_SEGS(sbi));
2316 for (i = 0; i < NR_DIRTY_TYPE; i++) {
2317 dirty_i->dirty_segmap[i] = f2fs_kvzalloc(bitmap_size, GFP_KERNEL);
2318 if (!dirty_i->dirty_segmap[i])
2319 return -ENOMEM;
2322 init_dirty_segmap(sbi);
2323 return init_victim_secmap(sbi);
2327 * Update min, max modified time for cost-benefit GC algorithm
2329 static void init_min_max_mtime(struct f2fs_sb_info *sbi)
2331 struct sit_info *sit_i = SIT_I(sbi);
2332 unsigned int segno;
2334 mutex_lock(&sit_i->sentry_lock);
2336 sit_i->min_mtime = LLONG_MAX;
2338 for (segno = 0; segno < MAIN_SEGS(sbi); segno += sbi->segs_per_sec) {
2339 unsigned int i;
2340 unsigned long long mtime = 0;
2342 for (i = 0; i < sbi->segs_per_sec; i++)
2343 mtime += get_seg_entry(sbi, segno + i)->mtime;
2345 mtime = div_u64(mtime, sbi->segs_per_sec);
2347 if (sit_i->min_mtime > mtime)
2348 sit_i->min_mtime = mtime;
2350 sit_i->max_mtime = get_mtime(sbi);
2351 mutex_unlock(&sit_i->sentry_lock);
2354 int build_segment_manager(struct f2fs_sb_info *sbi)
2356 struct f2fs_super_block *raw_super = F2FS_RAW_SUPER(sbi);
2357 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2358 struct f2fs_sm_info *sm_info;
2359 int err;
2361 sm_info = kzalloc(sizeof(struct f2fs_sm_info), GFP_KERNEL);
2362 if (!sm_info)
2363 return -ENOMEM;
2365 /* init sm info */
2366 sbi->sm_info = sm_info;
2367 sm_info->seg0_blkaddr = le32_to_cpu(raw_super->segment0_blkaddr);
2368 sm_info->main_blkaddr = le32_to_cpu(raw_super->main_blkaddr);
2369 sm_info->segment_count = le32_to_cpu(raw_super->segment_count);
2370 sm_info->reserved_segments = le32_to_cpu(ckpt->rsvd_segment_count);
2371 sm_info->ovp_segments = le32_to_cpu(ckpt->overprov_segment_count);
2372 sm_info->main_segments = le32_to_cpu(raw_super->segment_count_main);
2373 sm_info->ssa_blkaddr = le32_to_cpu(raw_super->ssa_blkaddr);
2374 sm_info->rec_prefree_segments = sm_info->main_segments *
2375 DEF_RECLAIM_PREFREE_SEGMENTS / 100;
2376 sm_info->ipu_policy = 1 << F2FS_IPU_FSYNC;
2377 sm_info->min_ipu_util = DEF_MIN_IPU_UTIL;
2378 sm_info->min_fsync_blocks = DEF_MIN_FSYNC_BLOCKS;
2380 INIT_LIST_HEAD(&sm_info->discard_list);
2381 sm_info->nr_discards = 0;
2382 sm_info->max_discards = 0;
2384 sm_info->trim_sections = DEF_BATCHED_TRIM_SECTIONS;
2386 INIT_LIST_HEAD(&sm_info->sit_entry_set);
2388 if (test_opt(sbi, FLUSH_MERGE) && !f2fs_readonly(sbi->sb)) {
2389 err = create_flush_cmd_control(sbi);
2390 if (err)
2391 return err;
2394 err = build_sit_info(sbi);
2395 if (err)
2396 return err;
2397 err = build_free_segmap(sbi);
2398 if (err)
2399 return err;
2400 err = build_curseg(sbi);
2401 if (err)
2402 return err;
2404 /* reinit free segmap based on SIT */
2405 build_sit_entries(sbi);
2407 init_free_segmap(sbi);
2408 err = build_dirty_segmap(sbi);
2409 if (err)
2410 return err;
2412 init_min_max_mtime(sbi);
2413 return 0;
2416 static void discard_dirty_segmap(struct f2fs_sb_info *sbi,
2417 enum dirty_type dirty_type)
2419 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2421 mutex_lock(&dirty_i->seglist_lock);
2422 kvfree(dirty_i->dirty_segmap[dirty_type]);
2423 dirty_i->nr_dirty[dirty_type] = 0;
2424 mutex_unlock(&dirty_i->seglist_lock);
2427 static void destroy_victim_secmap(struct f2fs_sb_info *sbi)
2429 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2430 kvfree(dirty_i->victim_secmap);
2433 static void destroy_dirty_segmap(struct f2fs_sb_info *sbi)
2435 struct dirty_seglist_info *dirty_i = DIRTY_I(sbi);
2436 int i;
2438 if (!dirty_i)
2439 return;
2441 /* discard pre-free/dirty segments list */
2442 for (i = 0; i < NR_DIRTY_TYPE; i++)
2443 discard_dirty_segmap(sbi, i);
2445 destroy_victim_secmap(sbi);
2446 SM_I(sbi)->dirty_info = NULL;
2447 kfree(dirty_i);
2450 static void destroy_curseg(struct f2fs_sb_info *sbi)
2452 struct curseg_info *array = SM_I(sbi)->curseg_array;
2453 int i;
2455 if (!array)
2456 return;
2457 SM_I(sbi)->curseg_array = NULL;
2458 for (i = 0; i < NR_CURSEG_TYPE; i++) {
2459 kfree(array[i].sum_blk);
2460 kfree(array[i].journal);
2462 kfree(array);
2465 static void destroy_free_segmap(struct f2fs_sb_info *sbi)
2467 struct free_segmap_info *free_i = SM_I(sbi)->free_info;
2468 if (!free_i)
2469 return;
2470 SM_I(sbi)->free_info = NULL;
2471 kvfree(free_i->free_segmap);
2472 kvfree(free_i->free_secmap);
2473 kfree(free_i);
2476 static void destroy_sit_info(struct f2fs_sb_info *sbi)
2478 struct sit_info *sit_i = SIT_I(sbi);
2479 unsigned int start;
2481 if (!sit_i)
2482 return;
2484 if (sit_i->sentries) {
2485 for (start = 0; start < MAIN_SEGS(sbi); start++) {
2486 kfree(sit_i->sentries[start].cur_valid_map);
2487 kfree(sit_i->sentries[start].ckpt_valid_map);
2488 kfree(sit_i->sentries[start].discard_map);
2491 kfree(sit_i->tmp_map);
2493 kvfree(sit_i->sentries);
2494 kvfree(sit_i->sec_entries);
2495 kvfree(sit_i->dirty_sentries_bitmap);
2497 SM_I(sbi)->sit_info = NULL;
2498 kfree(sit_i->sit_bitmap);
2499 kfree(sit_i);
2502 void destroy_segment_manager(struct f2fs_sb_info *sbi)
2504 struct f2fs_sm_info *sm_info = SM_I(sbi);
2506 if (!sm_info)
2507 return;
2508 destroy_flush_cmd_control(sbi);
2509 destroy_dirty_segmap(sbi);
2510 destroy_curseg(sbi);
2511 destroy_free_segmap(sbi);
2512 destroy_sit_info(sbi);
2513 sbi->sm_info = NULL;
2514 kfree(sm_info);
2517 int __init create_segment_manager_caches(void)
2519 discard_entry_slab = f2fs_kmem_cache_create("discard_entry",
2520 sizeof(struct discard_entry));
2521 if (!discard_entry_slab)
2522 goto fail;
2524 sit_entry_set_slab = f2fs_kmem_cache_create("sit_entry_set",
2525 sizeof(struct sit_entry_set));
2526 if (!sit_entry_set_slab)
2527 goto destory_discard_entry;
2529 inmem_entry_slab = f2fs_kmem_cache_create("inmem_page_entry",
2530 sizeof(struct inmem_pages));
2531 if (!inmem_entry_slab)
2532 goto destroy_sit_entry_set;
2533 return 0;
2535 destroy_sit_entry_set:
2536 kmem_cache_destroy(sit_entry_set_slab);
2537 destory_discard_entry:
2538 kmem_cache_destroy(discard_entry_slab);
2539 fail:
2540 return -ENOMEM;
2543 void destroy_segment_manager_caches(void)
2545 kmem_cache_destroy(sit_entry_set_slab);
2546 kmem_cache_destroy(discard_entry_slab);
2547 kmem_cache_destroy(inmem_entry_slab);